This book gives an introduction to fiber spaces and differential operators on smooth manifolds. Over the last 20 years,

*164*
*37*
*7MB*

*English*
*Pages 433
[441]*
*Year 2020*

- Author / Uploaded
- Jet Nestruev

*Table of contents : ForewordBook_BookNotesTitle_1PrefaceContents1 Introduction2 Cutoff and Other Special Smooth Functions on mathbbRn3 Algebras and Points4 Smooth Manifolds (Algebraic Definition)5 Charts and Atlases6 Smooth Maps7 Equivalence of Coordinate and Algebraic Definitions8 Points, Spectra, and Ghosts9 Differential Calculus as Part of Commutative Algebra10 Symbols and the Hamiltonian Formalism11 Smooth Bundles12 Vector Bundles and Projective Modules13 Localization14 Differential 1-forms and Jets15 Functors of the Differential Calculus and their Representations16 Cosymbols, Tensors, and Smoothness17 Spencer Complexes and Differential Forms18 The (Co)Chain Complexes Coming from the Spencer Sequence19 Differential Forms: Classical and Algebraic Approach20 Cohomology21 Differential Operators over Graded AlgebrasAfterwordAppendix A.M. Vinogradov Observability Principle, Set Theory and the ``Foundations of Mathematics''ReferencesIndex*

Graduate Texts in Mathematics

Jet Nestruev

Smooth Manifolds and Observables Second Edition

Graduate Texts in Mathematics

220

Graduate Texts in Mathematics Series Editors Sheldon Axler San Francisco State University, San Francisco, CA, USA Kenneth Ribet University of California, Berkeley, CA, USA Advisory Board Alejandro Adem, University of British Columbia David Eisenbud, MSRI & University of California, Berkeley Brian C. Hall, University of Notre Dame Patricia Hersh, University of Oregon J.F. Jardine, University of Western Ontario Jeffrey C. Lagarias, University of Michigan Eugenia Malinnikova, Stanford University Ken Ono, University of Virginia Jeremy Quastel, University of Toronto Barry Simon, California Institute of Technology Ravi Vakil, Stanford University Steven H. Weintraub, Lehigh University Melanie Matchett Wood, University of California, Berkeley

Graduate Texts in Mathematics bridge the gap between passive study and creative understanding, offering graduate-level introductions to advanced topics in mathematics. The volumes are carefully written as teaching aids and highlight characteristic features of the theory. Although these books are frequently used as textbooks in graduate courses, they are also suitable for individual study.

More information about this series at http://www.springer.com/series/136

Jet Nestruev

Smooth Manifolds and Observables Second Edition

123

Jet Nestruev Lizzano in Belvedere, Italy Moscow, Russia

ISSN 0072-5285 ISSN 2197-5612 (electronic) Graduate Texts in Mathematics ISBN 978-3-030-45649-8 ISBN 978-3-030-45650-4 (eBook) https://doi.org/10.1007/978-3-030-45650-4 Mathematics Subject Classiﬁcation: 58-01, 58Jxx, 55R05, 13Nxx, 08-01, 14F40, 58Axx, 17A70, 13B30, 37J05 1st edition: © 2003 Springer-Verlag New York, Inc. 2nd edition: © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional afﬁliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Foreword

The author and his team undertook the revision of the book “Smooth Manifolds and Observables”, published in 2002 by Springer Verlag, in 2016 at the suggestion of the Editors of Springer. It immediately became clear that the 2002 book should not only be revised, but must also be significantly expanded, since the main underlying idea of the book—passing from the differential geometry of smooth manifolds to the differential calculus in commutative algebras—has been successfully developed in the last 20 years in many new areas of mathematics and physics. As a result, the 11 original chapters have been expanded to 21, the number of pages has almost doubled (from 222 to 431), and the Preface has become much longer. Nevertheless, the main principles of the first edition, in particular the “observability principle”, which determines the relationship of mathematics to physics, have been preserved. We have tried to make the book as self-contained as possible, the only prerequisites being a general mathematical culture on the level of two years of graduate school. In this edition, we have also included the Afterword, as well as the Appendix to the first edition, written by A. M. Vinogradov, without any changes. Unlike the famous French general, Jet Nestruev is a civilian, and the names of the members of his team are not veiled in secrecy, as are those of the Bourbaki team. For the first edition, they were: Alexander Astashov, Alexey Bocharov, Sergei Duzhin, Alexandre Vinogradov, Mikhail Vinogradov, and Alexey Sossinsky. For the present second edition, they are Alexander Astashov, Alexandre Vinogradov, Mikhail Vinogradov, and Alexey Sossinsky. The figures for both editions were expertly prepared by Alexander Astashov. The author ackowledges the invaluable help of Joseph Krasil’shchik, who read the entire first draft of this edition and noticed, besides misprints and other minor snafus, numerous parts of the text that required essential changes; these changes have considerably improved the exposition. A. Sossinsky acknowledges the partial support of a J. Simons Foundation grant. Jet Nestruev Lizzano in Belvedere, Italy and Moscow, Russia December 2019 v

Alexandre Vinogradov Александр Михайлович Виноградов 1938–2019 Alexandre Vinogradov, an extraordinary human being and outstanding mathematician, deserves a more thorough and detailed life story than can fit into this page2, so here we will limit ourselves to a quote from a letter from his daughter Katya, who was at his bedside to the very end. “A few days ago Father was asleep in his bed and I was sitting next to him. He was no longer able to swallow, even water. He slept all the time. Suddenly he opened his eyes and tried to say something. Gently I said: ‘Daddy, do you want me to wet your lips?’ Unexpectedly he spoke, clearly and angrily: ‘Katya, damn, are you out of your mind? Lips, what lips? Don’t you understand anything? I have finally figured out quantization! and here you are wanting to wet my lips, stupid girl!’ Then I fully understood, once and for all, what my father is.”

2

A necrologue with a short biography of A. M. Vinogradov and a sketch of his contribution to mathematics appears in the second issue for 2020 of Russian Mathematical Surveys.

vii

Preface

The limits of my language are the limits of my world. —L. Wittgenstein

This book is a self-contained introduction to smooth manifolds, fiber spaces, and differential operators on them. It is not only accessible to graduate students specializing in mathematics and physics, but also intended for readers who are already familiar with the subject. Since there are many excellent textbooks in manifold theory, the first question that should be answered is: Why another book on manifolds? The main reason is that the good old differential calculus is actually a particular case of a much more general construction, which may be described as the differential calculus over commutative algebras. And this calculus, in its entirety, is just the consequence of properties of arithmetical operations. This fact, remarkable in itself, has numerous applications, ranging from delicate questions of algebraic geometry to the theory of elementary particles. Our book explains in detail why the differential calculus on manifolds is simply an aspect of commutative algebra. In the standard approach to smooth manifold theory, the subject is developed along the following lines. First one defines the notion of smooth manifold, say M. Then one defines the algebra F M of smooth functions on M, and so on. In this book this sequence is reversed: we begin with a certain commutative R-algebra1 F , and then define the manifold M ¼ MF

1 Here and below R stands for the real number ﬁeld. Nevertheless, and this is very important, nothing prevents us from replacing it by an arbitrary ﬁeld (or even a ring) if this is appropriate for the problem under consideration.

ix

x

Preface

as the R-spectrum of this algebra. (Of course, in order that MF deserve the title of a smooth manifold, the algebra F must satisfy certain conditions; these conditions appear in Chapter 3, where the main definitions mentioned here are presented in detail.) This approach is by no means new: it is used, say, in algebraic geometry. One of its advantages is that from the outset it is not related to the choice of a specific coordinate system, so that (in contrast to the standard analytical approach) there is no need to constantly check that various notions or properties are independent of this choice. This explains the popularity of the coordinate-free algebraic viewpoint among mathematicians attracted by sophisticated algebra, but its level of abstraction discourages the more pragmatically inclined applied mathematicians and physicists. But what is really new in the book is the physical motivation of the algebraic approach. It is based, in particular, on the (physical) notion of observable and the related observability principle, which asserts

what exists is only that which can be observed As we all know, in developing scientific theories, homo sapiens gather the necessary data from observation and experiment. At the beginning, this data comes from our sensory organs, eventually replaced by accurate measuring devices. An appropriate collection of these devices constitutes a physical laboratory that studies some physical system. Let us formalize this notion, having in mind that we are not interested in how the devices function, only in what data from the physical system under study they register. Measuring devices provide us with numerical data, i.e., their readings are real numbers. The obtained numbers are then processed, mainly by using the four arithmetical operations. Now if two devices P1 and P2 give two readings, instead of adding these numbers by hand, we can construct (or imagine) a device, P1 þ P2 that gives us the sum of the two readings. Since the choice of the scale of the readings is subjective, for the sake of objectivity, instead of a measuring device P, we must use (virtual) devices ‚P, ‚ 2 R. Further, in order to fix the unit in our scales, we need a special device, I, whose only possible reading is 1R . Indeed, as compared to P, the scale of the device P þ ‚1R is shifted by ‚. If P1 ; . . .; Pn are real measuring devices in the laboratory, then, continuing this process, we can associate to any polynomial pðx1 ; . . .; xn Þ the device pðP1 ; . . .; Pn Þ, whose reading is pða1 ; . . .; an Þ when a1 ; . . .; an are the readings of the devices P1 ; . . .; Pn , respectively. Thus the operations of addition and multiplication by real numbers defined above transform the set of all virtual measuring devices into an algebra isomorphic to the polynomial algebra R½x1 ; . . .; xn . But we can go even further and introduce virtual devices of the form f ðP1 ; . . .; Pn Þ, where f ðx1 ; . . .; xn Þ is a smooth (i.e., infinitely differentiable) function, as it was done above for polynomial functions. Then the laboratory

Preface

xi

consisting of all such virtual devices will be, from the mathematical point of view, an algebra isomorphic to the algebra C1 ðRn Þ of smooth functions in Euclidean space Rn . The device I will be the unit of this algebra. To summarize the above for future reference, let us agree that

the mathematical formalization of a physical laboratory is an appropriate commutative algebra with unit over R. In classical physics, we assess the state of a given physical system S only from the readings of all the measuring devices. In that sense “state” and “observation” are identical notions. Mathematically, an observation of the state of the system S may be understood as a map h : A ! R of the algebra A of all measuring devices, which to any device P assigns its reading hðPÞ in that state. On the other hand, from the definition of arithmetical operations on measuring devices, it obviously follows that h is a homomorphism of unitary algebras, in particular, we have hðIÞ ¼ 1R . It is important to understand that not every such homomorphism is an observation of the system S. To clarify this, let us consider the following construction. Let the ideal I A be the intersection of the ideals KerðhÞ for all obserdef vations from S, and AS A=I S . Then any such a homomorphism h may be represented in the form of the composition pr

~h

A!AS !R; where pr is the natural projection and ~ h ða mod I S Þ ¼ hðaÞ . Conversely, under certain restrictions on the algebra A (which we omit at this point), any homomorphism AS ! R composed with the projection pr is an observation of the system S. Thus the family of all states of the system S can be identified with the set of unitary algebra homomorphisms from AS to R. This set is denoted by SpecR AS and is called the R-spectrum of the algebra AS . Summarizing the above, let us agree that

the algebra AS is the mathematical model of the physical system S, while SpecR AS is the set of its states The algebra AS is sometimes called the algebra of observables, while its elements are observable quantities or simply observables. If we agree, as we did above, that A ¼ C 1 ðRn Þ, then SpecR A=I , where I C 1 ðRn Þ is the ideal naturally identified with some closed subset of Rn . On the other hand, we know that the space of states of mechanical and other physical systems with a finite number of degrees of freedom is a smooth manifold. From this we can conclude that the algebra of observables of such a system is none other than AS ¼ C 1 ðMÞ, the algebra of smooth functions on a manifold M.

xii

Preface

Since Hilbert and Bourbaki, it is generally considered that set theory is the foundation of all mathematics. This popular point of view, however, has many drawbacks. In particular, working with sets supplied with a certain structure, mathematicians are unable to distinguish concrete elements of those sets within the framework of set theory, and are forced to label them in some external way. A good example is the introduction of Cartesian coordinates in geometry, which considerably widened the possibilities of the axiomatic geometry of Euclid. Note that Descartes’ coordinate method is precisely equivalent to the interpretation of n-dimensional Euclidean space as the spectrum of the polynomial algebra SpecR R½x1 ; . . .; xn .

* * * A few words about how the book is organized. It has 21 chapters. Chapter 1 consists of a series of examples of smooth manifolds taken from various branches of mathematics, mechanics, and physics. Analyzing these examples, we give a preliminary idea of the notion of smooth manifold from the classical differential-geometrical viewpoint (the corresponding formal definition appears only in Chapter 5) as well as from our (algebraic) point of view (which is presented in Chapters 3 and 4). The equivalence of the two approaches is proved in Chapter 7. In the framework of our algebraic approach, smooth (i.e., differentiable) manifolds appear as R-spectra of a certain class of R-algebras (the latter are therefore called smooth), and their elements turn out to be the smooth functions defined on the corresponding spectra. Here the R-spectrum of some R-algebra A is the set of all its unital homomorphisms into the R-algebra R, i.e., the set that is “visible” by means of this algebra. Thus smooth manifolds are “worlds” whose observation can be carried out by means of smooth algebras. Because of the algebraic universality of the approach described above, “nonsmooth” algebras will allow us to observe “nonsmooth worlds” and study their singularities by using the differential calculus. But this differential calculus is not the naive calculus studied in introductory (or even “advanced”) university courses; it is a much more sophisticated construction. It is to the foundations of this calculus that the part of this book beginning with Chapter 9 is devoted. In it we “discover” the notion of differential operator over a commutative algebra and carefully analyze the main notion of the classical differential calculus, that of the derivative (or more precisely, that of the tangent vector). Moreover, in that chapter, we deal with the other simplest constructions of the differential calculus from the new point of view, e.g., with tangent and cotangent bundles, as well as jet bundles. The latter are used to prove the equivalence of the algebraic and the standard analytic definitions of differential operators for the case in which the basic algebra is the algebra of smooth functions on a smooth manifold.

Preface

xiii

Chapter 10 is devoted to the classical notion of symbol of a differential operator. In it we show how this notion can be carried over to the general situation of commutative algebras. Moreover, as an illustration of the “algebraic differential calculus,” we present the construction of the Hamiltonian formalism over an arbitrary commutative algebra. In Chapters 11 and 12, we study fiber bundles and vector bundles from the algebraic point of view. In particular, we establish the equivalence of the category of vector bundles over a manifold M, and the category of finitely generated projective modules over the algebra C 1 ðM Þ. Chapter 11 is concluded by a study of jet modules of an arbitrary vector bundle and an explanation of the universal role played by these modules in the theory of differential operators. The present edition of this book contains ten new chapters that further develop the differential calculus over commutative algebras, show that this calculus is a generalization of the differential calculus on smooth manifolds, and indicate some of the various branches of the mathematical sciences to which it can be (or is being) applied. In the process, several additional examples of the basic functors (that we call diffunctors) of the differential calculus are presented and discussed. Chapter 13 shows how the localization of various operators (which is practically automatic in the classical theory) can be carried out in the general context of commutative algebras, thereby showing that the algebraic theory generalizes the classical one in that respect as well. In Chapter 14, we study differential 1-forms and jet spaces, first in the classical context, and then in their general algebraic guise. It is stressed in the chapter that jet spaces possess natural smooth manifold structures and are the usual domain of definition of differential operators. Chapter 15 is a general introduction to the functors of the differential calculus (both in classical and in general algebraic form), to their representing objects and to the universal maps that arise in that context. In Chapters 16 and 17, the study of examples of diffunctors is continued along the lines of Chapter 15. Specifically, these chapters deal with cosymbols, tensors, Spencer complexes, and differential forms. Chapter 18 is a systematic study of the fundamental notion of differential form—the basis of such concepts as de Rham cohomology, symplectic and contact manifolds, infinitesimal symmetries, Jordan algebras, Killing forms, etc. The chapter indicates numerous directions in which the classical and general algebraic theory can be applied to different specific branches of mathematics, mechanics, and physics. The last three chapters, which deal with topics of recent (or even ongoing) research, are written in a style closer to that of a research monograph than to that of an expository one, they are not as self-contained as the previous ones, and involve several references to recent publications. The extremely important Chapter 19 (written by Alexandre Vinogardov) deals with differential n-forms and their applications, including symplectic, Poisson, contact manifolds, Monge–Ampère equations, Lie derivatives, infinitesimal symmetries,

xiv

Preface

Killing fields, Jacobi brackets, while Chapter 20 (written jointly by Alexandre and Mikhail Vinogardov) and Chapter 21 (written by Mikhail Vinogardov) are devoted to such topics as de Rham cohomology, cohomological integration theory, Poisson brackets, graded commutative algebras, Batalin–Vilkovisky operators. Looking ahead beyond the framework of this book, let us note that the mechanism of “quantum observability” is in principle of cohomological nature and is an appropriate specification of those natural observation methods of solutions to (nonlinear) partial differential equations that have appeared in the secondary differential calculus (see [17]) and in the fairly new branch of mathematical physics known as cohomological physics. * * * In 1969 Alexandre Vinogradov, one of the authors of this book, started a seminar aimed at understanding the mathematics underlying quantum field theory. Its participants were his mathematics students, and several young physicists, the most assiduous of whom were Dmitry Popov, Vladimir Kholopov, and Vladimir Andreev. In a couple of years, it became apparent that the difficulties of quantum field theory come from the fact that physicists express their ideas in an inadequate language, and that an adequate language simply does not exist (see the quotation preceding the Preface). If we analyze, for example, what physicists call the covariance principle, it becomes clear that its elaboration requires a correct definition of differential operators, differential equations, and, say, second-order differential forms. For this reason in 1971 a mathematical seminar split out from the physical one, and began studying the structure of the differential calculus and searching for an analog of algebraic geometry for systems of (nonlinear) partial differential equations. At the same time, the abovementioned author began systematically lecturing on the subject. At first, the participants of the seminar and the listeners of the lectures had to manage with some very schematic summaries of the lectures and their own lecture notes. But after ten years or so, it became obvious that all these materials should be systematically written down and edited. Thus Jet Nestruev was born, and he began writing a series of books entitled Elements of the Differential Calculus. Detailed contents of the first installments of the series appeared, and the first one was written. It contained, basically, the first eight chapters of the first edition of the present book. Then, after an interruption of nearly fifteen years, due to a series of objective and subjective circumstances, work on the project was resumed, and the second installment was written. Amalgamated with the first one, it constituted the first edition of “Smooth Manifolds and Observables”, first published in Russian by MCCME Publishers, then in English by Springer Verlag. The circumstances of the writing (directly in English) of the second edition are explained in the Foreword.

Preface

xv

The present second edition of the book is self-contained, but open-ended. The reader who wishes to have a look ahead without delay can consult the references appearing on page 423 or the more complete bibliography in [1] and [21]. But we hope for much more: that he/she will go fearlessly into terra incognita, choosing one of the numerous branches of mathematics or physics to which our approach can be applied, and armed with the methods developed in the book, will discover the appropriate (and hopefully beautiful!) mathematical explanations of real-life phenomena that seem mysterious at present. Jet Nestruev Lizzano in Belvedere, Italy and Moscow, Russia December 2019

Contents

Foreword Preface

v ix

1

Introduction

1

2

Cutoff and Other Special Smooth Functions on Rn

13

3

Algebras and Points

21

4

Smooth Manifolds (Algebraic Definition)

37

5

Charts and Atlases

53

6

Smooth Maps

65

7

Equivalence of Coordinate and Algebraic Def initions

77

8

Points, Spectra, and Ghosts

85

9

Differential Calculus as Part of Commutative Algebra

101

10 Symbols and the Hamiltonian Formalism

141

11 Smooth Bundles

151

xvii

xviii

Contents

12 Vector Bundles and Projective Modules

171

13 Localization

203

14 Differential 1-forms and Jets

223

15 Functors of the Differential Calculus and their Representations

243

16 Cosymbols, Tensors, and Smoothness

273

17 Spencer Complexes and Differential Forms

293

18 The (Co)Chain Complexes Coming from the Spencer Sequence

323

19 Differential Forms: Classical and Algebraic Approach

341

20 Cohomology

363

21 Differential Operators over Graded Algebras

387

Afterword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415 Appendix A. M. Vinogradov Observability Principle, Set Theory and the “Foundations of Mathematics” . . . . . . . . . . . . . . . . . . . . . . . . . 417

References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425

1 Introduction

1.0. This chapter is a preliminary discussion of ﬁnite-dimensional smooth (inﬁnitely diﬀerentiable) real manifolds, the main protagonists of this book. Why are smooth manifolds important? Well, we live in a manifold (a four-dimensional one, according to Einstein) and on a manifold (the Earth’s surface, whose model is the sphere S 2 ). We are surrounded by manifolds: The surface of a coﬀee cup is a manifold (namely, the torus S 1 × S 1 , more often described as the surface of a doughnut or an anchor ring, or as the tube of an automobile tire); a shirt is a two-dimensional manifold with boundary. Processes taking place in nature are often adequately modeled by points moving on a manifold, especially if they involve no discontinuities or catastrophes. (Incidentally, catastrophes—in nature or on the stock market—as studied in “catastrophe theory” may not be manifolds, but then they are smooth maps of manifolds.) What is more important from the point of view of this book is that manifolds arise quite naturally in various branches of mathematics (in algebra and analysis as well as in geometry) and its applications (especially mechanics). Before trying to explain what smooth manifolds are, we give some examples. 1.1. The configuration space Rot(3) of a rotating solid in space. Consider a solid body in space ﬁxed by a hinge O that allows it to rotate in any direction (Figure 1.1). We want to describe the set of positions of the body, or, as it is called in classical mechanics, its conﬁguration space. © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 1

1

2

Chapter 1

z

O x A(xA, yA, zA)

y

B(xB, yB , zB ) Figure 1.1. Rotating solid.

One way of going about it is to choose a coordinate system Oxyz and determine the body’s position by the coordinates (xA , yA , zA ), (xB , yB , zB ) of two of its points A, B. But this is obviously not an economical choice of parameters: It is intuitively clear that only three real parameters are required, at least when the solid is not displaced too greatly from its initial position OA0 B0 . Indeed, two parameters determine the direction of OA (e.g., xA , yA ; see Figure 1.1), and one more is needed to show how the solid is turned about the OA axis (e.g., the angle ϕB = B0 OB, where AB0 is parallel to A0 B0 ). It should be noted that these are not ordinary Euclidean coordinates; the positions of the solid do not correspond bijectively in any natural way to ordinary three-dimensional space R3 . Indeed, if we rotate AB through the angle ϕ = 2π, the solid does not acquire a new position; it returns to the position OAB; besides, two positions of OA correspond to the coordinates (xA , yA ): For the second one, A is below the Oxy plane. However, locally, say near the initial position OA0 B0 , there is a bijective correspondence between the position of the solid and a neighborhood of the origin in 3-space R3 , given by the map OAB → (xA , yA , ϕB ). Thus the conﬁguration space Rot(3) of a rotating solid is an object that can be described locally by three Euclidean coordinates, but globally has a more complicated structure. 1.2. An algebraic surface V . In nine-dimensional Euclidean space R9 consider the set of points satisfying the following system of six algebraic

Introduction

equations:

⎧ ⎪ ⎨

x21 + x22 + x23 = 1; x24 + x25 + x26 = 1; ⎪ ⎩ x27 + x28 + x29 = 1;

3

x1 x4 + x2 x5 + x3 x6 = 0; x1 x7 + x2 x8 + x3 x9 = 0; x4 x7 + x5 x8 + x6 x9 = 0.

This happens to be a nice three-dimensional surface in R9 (3 = 9 − 6). It is not diﬃcult (try!) to describe a bijective map of a neighborhood of any point (say (1, 0, 0, 0, 1, 0, 0, 0, 1)) of the surface onto a neighborhood of the origin of Euclidean 3-space. But this map cannot be extended to cover the entire surface, which is compact (why?). Thus again we have an example of an object V locally like 3-space, but with a diﬀerent global structure. It should perhaps be pointed out that the solution set of six algebraic equations with nine unknowns chosen at random will not always have such a simple local structure; it may have self-intersections and other singularities. (This is one of the reasons why algebraic geometry, which studies such algebraic varieties, as they are called, is not a part of smooth manifold theory.) 1.3. Three-dimensional projective space RP 3 . In four-dimensional Euclidean space R4 consider the set of all straight lines passing through the origin. We want to view this set as a “space” whose “points” are the lines. Each “point” of this space—called projective space RP 3 by nineteenth century geometers—is determined by the line’s directing vec 2 ai = 0, i.e., a quadruple of real numbers. Since tor (a1 , a2 , a3 , a4 ), proportional quadruples deﬁne the same line, each point of RP 3 is an equivalence class of proportional quadruples of numbers, which are denoted by P = (a1 : a2 : a3 : a4 ), where (a1 , a2 , a3 , a4 ) is any representative of the class. In the vicinity point, RP 3 is like R3 . Indeed, if we are given 0 0of each 0 0 a04 = 0, it can be written in the a point P0 = a1 : a2 : a3 : a4 for which 0 0 0 0 0 0 form P0 = a1 /a4 : a2 /a4 : a3 /a4 : 1 and the three ratios viewed as its three coordinates. If we consider all the points P for which a4 = 0 and take x1 = a1 /a4 , x2 = a2 /a4 , and x3 = a3 /a4 to be their coordinates, we obtain a bijection of a neighborhood of P0 onto R3 . This neighborhood, together with three similar neighborhoods (for a1 = 0, a2 = 0, a3 = 0), covers all the points of RP 3 . But points belonging to more than one neighborhood are assigned to diﬀerent triples of coordinates (e.g., the point (6 : 12 : 2 : 3) will have the coordinates 2, 4, 23 in one system of coordinates and 3, 6, 32 in another). Thus the overall structure of RP 3 is not that of R3 . 1.4. The special orthogonal group SO(3). Consider the group SO(3) of orientation-preserving isometries of R3 . In a ﬁxed orthonormal basis, each element A ∈ SO(3) is deﬁned by an orthogonal positive deﬁnite matrix, thus by nine real numbers (9 = 3 × 3). But of course, fewer than 9 numbers

4

Chapter 1

are needed to determine A. In canonical form, the matrix of A will be ⎛ ⎞ 1 0 0 ⎝0 cos ϕ sin ϕ ⎠ , 0 − sin ϕ cos ϕ and A is deﬁned if we know ϕ and are given the eigenvector corresponding to the eigenvalue λ = 1 (two real coordinates a, b are needed for that, since eigenvectors are deﬁned up to a scalar multiplier). Thus again three coordinates (ϕ, a, b) determine elements of SO(3), and they are Euclidean coordinates only locally.

ϕ 1 ϕ 2

Figure 1.2. Billiards on a disk.

1.5. The phase space of billiards on a disk B(D2 ). A tiny billiard ball P moves with unit velocity in a closed disk D2 , bouncing oﬀ its circular boundary C in the natural way (angle of incidence = angle of reﬂection). We want to describe the phase space B(D2 ) of this mechanical system, whose “points” are all the possible states of the system (each state being deﬁned by the position of P and the direction of its velocity vector). Since each state is determined by three coordinates (x, y; ϕ) (Figure 1.2), it would seem that as a set, B(D2 ) is D2 ×S 1 , where S 1 is the unit circle (i.e., S 1 = R mod 2π). But this is not the case, because at the moment of collision with the boundary, say at (x0 , y0 ), the direction of the velocity vector jumps from ϕ1 to ϕ2 (see Figure 1.2), so that we must identify the states (x0 , y0 , ϕ1 ) ≡ (x0 , y0 , ϕ2 ).

(1.1)

Introduction

5

Thus B(D2 ) = (D2 × S 1 )/ ∼, where / ∼ denotes the factorization deﬁned by the equivalence relation of all the identiﬁcations (1.1) due to all possible collisions with the boundary C. Since the identiﬁcations take place only on C, all the points of B 0 (D2 ) = Int D2 × S 1 = (Int D2 × S 1 )/ ∼ , where Int D2 = D2 C is the interior of D2 , have neighborhoods with a structure like that of open sets in R3 (with coordinates (x, y; ϕ)). It is a rather nice fact (not obvious to the beginner) that after identiﬁcations the “boundary states” (x, y; ϕ), (x, y) ∈ C, also have such neighborhoods, so that again B(D2 ) is locally like R3 , but not like R3 globally (as we shall later show). As a more sophisticated example, the advanced reader might try to describe the phase √ space of billiards in a right triangle with an acute angle of (a) π/6, (b) 2π/4. 1.6. The ﬁve examples of three-dimensional manifolds described above all come from diﬀerent sources: classical mechanics 1.1, algebraic geometry 1.2, classical geometry 1.3, linear algebra 1.4, and mechanics 1.5. The advanced reader has not failed to notice that 1.1–1.4 are actually examples of one and the same manifold (appearing in diﬀerent garb): Rot(3) = V = RP 3 = SO(3). To be more precise, the ﬁrst four manifolds are all “diﬀeomorphic,” i.e., equivalent as smooth manifolds (the deﬁnition is given in Section 6.7). As for Example 1.5, B(D2 ) diﬀers from (i.e., is not diﬀeomorphic to) the other manifolds, because it happens to be diﬀeomorphic to the three-dimensional sphere S 3 (the beginner should not be discouraged if he/she fails to see this; it is not obvious). What is the moral of the story? The history of mathematics teaches us that if the same object appears in diﬀerent guises in various branches of mathematics and its applications, and plays an important role there, then it should be studied intrinsically, as a separate concept. That was what happened to such fundamental concepts as group and linear space, and is true of the no less important concept of smooth manifold. 1.7. The examples show us that a manifold M is a point set locally like Euclidean space Rn with global structure not necessarily that of Rn . How does one go about studying such an object? Since there are Euclidean coordinates near each point, we can try to cover M with coordinate neighborhoods (or charts, or local coordinate systems, as they are also called). A family of charts covering M is called an atlas. The term is evocative; indeed, a geographical atlas is a set of charts or maps of the manifold S 2 (the Earth’s surface) in that sense. In order to use the separate charts to study the overall structure of M , we must know how to move from one chart to the next, thus “gluing

6

Chapter 1

Figure 1.3. Charts for the surface of the Earth.

together” the charts along their common parts, so as to recover M (see Figure 1.3). In less intuitive language, we must be in possession of coordinate transformations, expressing the coordinates of points of any chart in terms of those of a neighboring chart. (The industrious reader might proﬁt by actually writing out these transformations for the case of the four-chart atlas of RP 3 described in 1.3.) If we wish to obtain a smooth manifold in this way, we must require that the coordinate transformations be “nice” functions (in a certain sense). We then arrive at the coordinate or classical approach to smooth manifolds. It is developed in detail in Chapter 5. 1.8. Perhaps more important is the algebraic approach to the study of manifolds. In it we forget about charts and coordinate transformations and work only with the R-algebra FM of smooth functions f : M → R on the manifold M . It turns out that FM entirely determines M and is a convenient object to work with. An attempt to give the reader an intuitive understanding of the natural philosophy underlying the algebraic approach is undertaken in the next sections. 1.9. In the description of a classical physical system or process, the key notion is the state of the system. Thus, in classical mechanics, the state of a moving point is described by its position and velocity at the given moment of time. The state of a given gas from the point of view of thermodynamics is described by its temperature, volume, pressure, etc. In order to actually assess the state of a given system, the experimentalist must use various measuring devices whose readings describe the state. Suppose M is the set of all states of the classical physical system S. Then to each measuring device D there corresponds a function fD on the

Introduction

7

set M , assigning to each state s ∈ M the reading fD (s) (a real number) that the device D yields in that state. From the physical point of view, we are interested only in those characteristics of each state that can be measured in principle, so that the set M of all states is described by the collection ΦS of all functions fD , where the D’s are measuring devices (possibly imaginary ones, since it is not necessary—nor indeed practically possible—to construct all possible measuring devices). Thus, theoretically, a physical system S is nothing more that the collection ΦS of all functions determined by adequate measuring devices (real or imagined ) on S. 1.10. Now, if the functions f1 , . . . , fk correspond to the measuring devices D1 , . . . , Dk of the physical system S, and ϕ(x1 , . . . , xk ) is any “nice” realvalued function in k real variables, then in principle it is possible to construct a device D such that the corresponding function fD is the composite function ϕ(f1 , . . . , fk ). Indeed, such a device may be obtained by constructing an auxiliary device, synthesizing the value ϕ(x1 , . . . , xk ) from input entries x1 , . . . , xk (this can always be done if ϕ is nice enough), and then “plugging in” the outputs (f1 , . . . , fk ) of the devices D1 , . . . , Dk into the inputs (x1 , . . . , xk ) of the auxiliary device. Let us denote this device D by ϕ(D1 , . . . , Dk ). In particular, if we take ϕ(x1 , x2 ) = x1 + x2 (or ϕ(x) = λx, λ ∈ R, or ϕ(x1 , x2 ) = x1 x2 ), we can construct the devices D1 +D2 (or λDi , or D1 D2 ) from any given devices D1 , D2 . In other words, if fi = fDi ∈ ΦS , then the functions f1 + f2 , λfi , f1 f2 also belong to ΦS . Thus the set ΦS of all functions f = fD describing the system S has the structure of an algebra over R (or R-algebra). 1.11. Actually, the set ΦS of all functions fD : MS → R is much too large and cumbersome for most classical problems. Systems (and processes) described in classical physics are usually continuous or smooth in some sense. Discontinuous functions fD are irrelevant to their description; only “smoothly working” measuring devices D are needed. Moreover, the problems of classical physics are usually set in terms of diﬀerential equations, so that we must be able to take derivatives of the relevant functions from ΦS as many times as we wish. Thus we are led to consider, rather than ΦS , the smaller set FS of smooth functions fD : MS → R. The set FS inherits an R-algebra structure from the inclusion FS ⊂ ΦS , but from now on we shall forget about Φ, since the smooth R-algebra FS will be our main object of study. 1.12. Let us describe in more detail what the algebra FS might be like in classical situations. For example, from the point of view of classical mechanics, a system S of N points in space is adequately described by the positions and velocities of the points, so that we need 6N measuring devices Di to record them. Then the algebra FS consists of all elements of the form ϕ(f1 , . . . , f6N ), where the fi are the “basic functions” determined by the devices Di , while ϕ : R6N → R is any nice (smooth) function.

8

Chapter 1

In more complicated situations, certain relations among the basis functions fi may arise. For example, if we are studying a system of two mass points joined by a rigid rod of negligible mass, we have the relation 3

(fi − fi+3 )2 = r2 , i=1

where r is the length of the rod and the functions fi (respectively fi+3 ) measure the ith coordinate of the ﬁrst (respectively second) mass point. (There is another relation for the velocity components, which the reader might want to write out explicitly.) Generalizing, we can say that there usually exists a basis system of devices D1 , . . . , Dk adequately describing the system S (from the chosen point of view). Then the R-algebra FS consists of all elements of the form ϕ(f1 , . . . , fk ), where ϕ : Rk → R is a nice function and the fi = fDi are the relevant measurements (given by the devices Di ) that may be involved in relations of the form F (f1 , . . . , fk ) ≡ 0. Then FS may be described as follows. Let Rk be Euclidean space with coordinates f1 , . . . , fk and U = {(f1 , . . . , fk ) | ai < fi < bi }, where the open intervals ]ai , bi [ contain all the possible readings given by the devices Di . The relations Fj (f1 , . . . , fk ) = 0 between the basis variables f1 , . . . , fk determine a surface M in U . Then FS is the R-algebra of all smooth functions on the surface M . 1.13. Example (thermodynamics of an ideal gas). Consider a certain volume of ideal gas. From the point of view of thermodynamics, we are interested in the following measurements: the volume V , the pressure p, and the absolute temperature T of the gas. These parameters, as is well known, satisfy the relation pV = cT , where c is a certain constant. Since 0 < p < ∞, 0 < V < ∞, and 0 < T < ∞, the domain U is the ﬁrst octant in the space R3 (V, p, T ), and the hypersurface M in this domain is given by the equation pV = cT . The relevant R-algebra F consists of all smooth functions on M .

Figure 1.4. Hinge mechanisms (5; 2, 2, 2), (1; 4, 1, 4), (1; 1, 1, 1), (2; 1, 2, 1), (5; 3, 3, 1).

1.14. Example: plane hinge mechanisms (also known as plane linkages). Such a mechanism consists of n > 3 ideal rods in the plane of lengths, say, (l1 ; l2 , . . . , ln ) (see Figure 1.4 for n = 4); the rods are joined in cyclic order

Introduction

9

to each other by ideal hinges at their end points; the hinges of the ﬁrst rod (and hence the rod itself) are ﬁxed to the plane; the other hinges and rods move freely (insofar as the conﬁguration allows them to); the rods can sweep freely over (“through”) each other. Obviously, the conﬁguration space of a hinge mechanism is determined completely by the sequence of lengths of its rods. So, one can refer to a concrete mechanism just by indicating the corresponding sequence, for instance, (5; 2, 3, 2). The reader is invited to solve the following problems in the process of reading the book. The ﬁrst of them she/he can attack even now. Exercise. Describe the conﬁguration spaces of the following hinge mechanisms: 1. Quadrilaterals: (5; 2, 2, 2); (1; 4, 1, 4); (1; 1, 1, 1); (2; 1, 2, 1); (5; 3, 3, 1). 2. Pentagons: (3.9; 1, 1, 1, 1); (1; 4, 1, 1, 4); (6; 6, 2, 2, 6); (1; 1, 1, 1, 1). The reader will enjoy discovering that the conﬁguration space of (1; 1, 1, 1, 1) is the sphere with four handles. Exercise. Show that the conﬁguration space of a pentagon depends only on the set of lengths of the rods and not on the order in which the rods are joined to each other. Exercise. Show that the conﬁguration space of the hinge mechanism (n − α; 1, . . . , 1) consisting of n + 1 rods is 1. The sphere S n−2 if α = 12 . 2. The (n − 2)-dimensional torus T n−2 = S 1 × · · · × S 1 if α = 32 . 1.15. So far we have not said anything to explain what a state s ∈ MS of our physical system S really is, relying on the reader’s physical intuition. But once the set of relevant functions FS has been speciﬁed, this can easily be done in a mathematically rigorous and physically meaningful way. The methodological basis of physical considerations is measurement. Therefore, two states of our system must be considered identical if and only if all the relevant measuring devices yield the same readings. Hence each state s ∈ MS is entirely determined by the readings in this state on all the relevant measuring devices, i.e., by the correspondence FS → R assigning to each fD ∈ FS its reading (in the state s) fD (s) ∈ R. This assignment will clearly be an R-algebra homomorphism. Thus we can say, by deﬁnition, that any state s of our system is simply an R-algebra homomorphism s : FS → R. The set of all R-algebra homomorphisms FS → R will be denoted by |FS |; it should coincide with the set MS of all states of the system. 1.16. Summarizing Sections 1.9–1.15, we can say that any classical physical system is described by an appropriate collection of measuring devices, each

10

Chapter 1

state of the system being the collection of readings that this state determines on the measuring devices. The sentence in italics may be translated into mathematical language by means of the following dictionary: • physical system = manifold, M ; • state of the system = point of the manifold, x ∈ M ; • measuring device = function on M , f ∈ F; • adequate collection of measuring devices = smooth R-algebra, F; • reading on a device = value of the function, f (x); • collection of readings in the given state = R-algebra homomorphism x : F → R,

f → f (x).

The resulting translation reads: Any manifold M is determined by the smooth R-algebra F of functions on it, each point x on M being the R-algebra homomorphism F → R that assigns to every function f ∈ F its value f (x) at the point x. 1.17. Mathematically, the crucial idea in the previous sentence is the identiﬁcation of points x ∈ M of a manifold and R-algebra homomorphisms x : F → R of its R-algebra of functions F, governed by the formula x(f ) = f (x). (1.2) This formula, read from left to right, deﬁnes the homomorphism x : F → R when the functions f ∈ F are given. Read from right to left, it deﬁnes the functions f : M → R, when the homomorphisms x ∈ M are known. Thus formula (1.2) is right in the middle of the important duality relationship existing between points of a manifold and functions on it, a duality similar to, but much more delicate than, the one between vectors and covectors in linear algebra. 1.18. In the general mathematical situation, the identiﬁcation M ↔ |F| between the set M on which the functions f ∈ F are deﬁned and the family of all R-algebra homomorphisms F → R cannot be correctly carried out. This is because, ﬁrst of all, |F| may turn out to be “much smaller” than M (an example is given in Section 3.6) or “bigger” than M , as we can see from the following example. Example. Suppose M is the set N of natural numbers and F is the set of all functions on N (i.e., sequences {a(k)}) such that the limit limk→∞ a(k) exists and is ﬁnite. Then the homomorphism α : F → R,

{a(k)} → lim a(k), k→∞

does not correspond to any point of M = N.

Introduction

11

Indeed, if α did correspond to some point n ∈ N, we would have by (1.2) n(a(·)) = a(n), so that lim a(k) = α(a(·)) = n(a(·)) = a(n)

k→∞

for any sequence {a(k)}. But this is not the case, say, for the sequence ai = 0, i n, ai = 1, i > n. Thus |F| is bigger than M , at least by the homomorphism α. However, we can always add to N the “point at inﬁnity” ∞ and extend the sequences (elements of F) by putting a(∞) = limk→∞ a(k), thus viewing the sequences in F as functions on N ∪ {∞}. Then obviously the homomorphism above corresponds to the “point” ∞. This trick of adding points at inﬁnity (or imaginary points, improper points, points of the absolute, etc.) is extremely useful and will be exploited to great advantage in Chapter 8. 1.19. In our mathematical development of the algebraic approach (Chapter 3), we shall start from an R-algebra F of abstract elements called “functions.” Of course, F will not be just any algebra; it must meet certain “smoothness” requirements. Roughly speaking, the algebra F must be smooth in the sense that locally (the meaning of that word must be deﬁned in abstract algebraic terms!) it is like the R-algebra C ∞ (Rn ) of inﬁnitely diﬀerentiable functions in Rn . This will be the algebraic way of saying that the manifold M is locally like Rn ; it will be explained rigorously and in detail in Chapter 3. When the smoothness requirements are met, it will turn out that F entirely determines the manifold M as the set |F| of all R-algebra homomorphisms of F into R, and F can be identiﬁed with the R-algebra of smooth functions on M . The algebraic deﬁnition of smooth manifold appears in the ﬁrst section of Chapter 4. 1.20. Smoothness requirements are also needed in the classical coordinate approach, developed in detail below (see Chapter 5). In particular, coordinate transformations must be inﬁnitely diﬀerentiable. The rigorous coordinate deﬁnition of a smooth manifold appears in Section 5.8. 1.21. The two deﬁnitions of smooth manifold (in which the algebraic approach and the coordinate approach result) are of course equivalent. This is proved in Chapter 7 below. Essentially, this book is a detailed exposition of these two approaches to the notion of smooth manifold and their equivalence, involving many examples, including a more rigorous treatment of the examples given in Sections 1.1–1.5 above.

2 Cutoﬀ and Other Special Smooth Functions on Rn

2.1. This chapter is an auxiliary one and can be omitted on ﬁrst reading. In it we show how to construct certain speciﬁc inﬁnitely diﬀerentiable functions on Rn (the R-algebra of all such functions is denoted by C ∞ (Rn )) that vanish (or do not vanish) on subsets of Rn of special form. These functions will be useful further on in the proof of many statements, especially in the very important Chapter 3. 2.2 Proposition. There exists a function f ∈ C ∞ (R) that vanishes for all negative values of the variable and is strictly positive for its positive values. We claim that such is the function 0 f (x) = e−1/x

for x 0, for x > 0

(2.1)

(see Figure 2.1 in the background). The only thing that must be checked is that f is smooth, i.e., f ∈ C ∞ (R). By induction over n, we shall show that the nth derivative of f is of the form 0 for x 0, (n) (2.2) f (x) = −1/x −2n e Pn (x)x for x > 0, where Pn (x) is a polynomial, and that f (n) is continuous. For n = 0 this is obvious, since limx→+0 e−1/x = 0.

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 2

13

14

Chapter 2

1

1

0

1

2

4

Figure 2.1. Special functions for Proposition 2.2 and Corollary 2.3.

If (2.2) is established for some n 0, then obviously f (n+1) (x) = 0 when x < 0, while if x > 0, we have f (n+1) (x) = e−1/x Pn (x) + x2 Pn (x) − 2nxPn (x) x−2n−2 , which shows that f (n+1) is of the form (2.2). To show that it is continuous, note that limx→+0 e−1/x xα = 0 by L’Hospital’s rule for any real α. Hence limx→+0 f (n+1) (x) = 0 and (again by L’Hospital’s rule) f (n) (x) − f (n) (0) f (n+1) (x) − 0 = lim , x→0 x→0 x 1

f (n+1) (0) = lim

so that f (n+1) equals 0 for x 0 and is continuous for all x.

Exercise. Let f be the function deﬁned in (2.1) and let ck = max f (k) . 1. Prove that ck < ∞ for all k. 2. Investigate the behavior of the sequence {ck } when k → ∞. 2.3 Corollary. For any r > 0 and a ∈ Rn there exists a function g ∈ C ∞ (Rn ) that vanishes for all x ∈ Rn satisfying x − a r and is positive for all other x ∈ Rn . Such is, for example, the function g(x) = f r2 − x − a2 , where f is the function (2.1) from Section 2.2 (see Figure 2.1).

2.4 Proposition. For any open set U ⊂ Rn there exists a function f ∈ C ∞ (Rn ) such that f (x) = 0, if x ∈ / U, f (x) > 0, if x ∈ U.

Special Smooth Functions on Rn

15

If U =Rn , take f ≡ 1; if U = ∅, take f ≡ 0. Now suppose U = Rn , U = ∅, and let {Uk } be a covering of U by a countable collection of open balls (e.g., all the balls of rational radius centered at the points with rational coordinates and contained in U ). By Corollary 2.3, there exist smooth functions fk ∈ C ∞ (Rn ) such that fk (x) > 0 if x ∈ Uk and fk (x) = 0 if x ∈ / Uk . Put ∂ p fk . sup (x) Mk = p p ∂ 1 x1 · · · ∂ n xn 0pk p1 +···+pn =p x∈Rn

Note that Mk < ∞, since outside the compact set U k (the bar denotes closure) the function fk and all its derivatives vanish. Further, the series ∞ fk k=1

2k Mk

converges to a smooth function f , since for all p1 , . . . , pn the series ∞ k=1

fk ∂ p1 +···+pn fk 2k Mk ∂xp11 · · · ∂xpnn

converges uniformly (because whenever k p1 +· · ·+pn , the absolute value of the kth term is no greater than 2−k ). Clearly, the function f possesses the required properties. 2.5 Corollary. For any two nonintersecting closed sets A, B ⊂ Rn there exists a function f ∈ C ∞ (Rn ) such that ⎧ ⎪ when x ∈ A; ⎨f (x) = 0, f (x) = 1, when x ∈ B; ⎪ ⎩ 0 < f (x) < 1, for all other x ∈ Rn . Using Proposition 2.4, choose a function fA that vanishes on A and is positive outside A and a similar function fB for B. Then for f we can take the function fA f= fA + fB (see Figure 2.2).

2.6 Corollary. Suppose U ⊂ Rn is an open set and f ∈ C ∞ (U ). Then for any point x ∈ U there exists a neighborhood V ⊂ U and a function g ∈ C ∞ (Rn ) such that f V ≡ g V . Suppose W is an open ball centered at x whose closure is contained in U . Let V be a smaller concentric ball. The required function g can be deﬁned

16

Chapter 2

B

A

Figure 2.2. Smooth function separating two sets.

as

h(y) · f (y), g(y) = 0,

when y ∈ U, when y ∈ Rn \ U,

where the function h ∈ C ∞ (Rn ) is obtained from Corollary 2.3 and satisﬁes hV ≡ 1, hRn \W ≡ 0. 2.7 Proposition. On any nonempty open set U ⊂ Rn there exists a smooth function with compact level surfaces, i.e.,, a function f ∈ C ∞ (U ) such that for any λ ∈ R the set f −1 (λ) is compact. Denote by Ak the set of points x ∈ U satisfying both of the following conditions: (i) x k, (ii) the distance from x to the boundary of U is not less than 1/k (if U = Rn , then condition (ii) can be omitted). Obviously, all points of Ak are interior points of Ak+1 . Hence Ak and the complement in Rn to the interior of the set Ak+1 are two closed nonintersecting sets. By Corollary 2.5 there exists a function fk ∈ C ∞ (Rn ) such that ⎧ ⎪fk (x) = 0 if x ∈ Ak , ⎨ if x ∈ / Ak+1 , fk (x) = 1 ⎪ ⎩ 0 < fk (x) < 1 otherwise. Since any point x ∈ U belongs to the interior of the set Ak for all suﬃciently large k, the sum ∞ fk f= k=1

Special Smooth Functions on Rn

17

is well deﬁned, and f is smooth on U (locally it is a ﬁnite sum of smooth functions). Consider a point x ∈ U \ Ak . Since all the functions fi are nonnegative, and for i < k we have fi (x) = 1, it follows that f (x) k − 1. Hence, for any λ ∈ R the set f −1 (λ) is a closed subset of the compact set Ak , where k is an integer such that λ < k − 1. A closed subset of a compact set is always compact, so that f is the required function. Let us ﬁx a coordinate system x1 , . . . , xn in a neighborhood U of a point z. Recall that a domain U is called starlike with respect to z if together with any point y ∈ U it contains the whole interval (z, y). 2.8 Hadamard’s lemma. Any smooth function f in a starlike neighborhood of a point z is representable in the form f (x) = f (z) +

n

(xi − zi )gi (x),

(2.3)

i=1

where gi are smooth functions. In fact, consider the function ϕ(t) = f (z + (x − z)t). Then ϕ(0) = f (z) and ϕ(1) = f (x), and by the Newton–Leibniz formula, 1 1 n dϕ ∂f ϕ(1) − ϕ(0) = dt = (z + (xi − zi )t)(xi − zi )dt 0 dt 0 i=1 ∂xi 1 n ∂f = (xi − zi ) (z + (xi − zi )t)dt. 0 ∂xi i=1 Since the functions

gi (x) =

0

1

∂f (z + (xi − zi )t)dt ∂xi

are smooth, this concludes the proof of Hadamard’s lemma. Let, as before, x1 , . . . , xn be a ﬁxed coordinate system of a point z in a neighborhood U and let τ = (i1 , . . . , in ) be a multiindex. Set |τ | = i1 + · · · + in ,

∂ |τ | ∂ |τ | = , i τ ∂x ∂x11 · · · ∂xinn

(x − z)τ = (x1 − z1 )i1 · · · (xn − zn )in ,

τ ! = i1 ! · · · in !.

2.9 Corollary. (Taylor expansion in Hadamard’s form.) Any smooth function f in a starlike neighborhood U of a point z is representable in the form n 1 ∂ |τ | f (x − z)τ (z) + f (x) = τ! ∂xτ |τ |=0

(x − z)σ gσ (x),

|σ|=n+1

(2.4)

18

Chapter 2

where gσ ∈ C ∞ (U ). In fact, using Hadamard’s lemma for each function gi in the decomposition (2.3), the function f can be represented in the form f (x) = f (z) +

n

(xi − zi )gi (z) +

i=1

n

(xi − zi )(xj − zj )gij (x).

i,j=1

Repeating this procedure for the functions gij , etc., we shall obtain the decomposition f (x) = f (z) +

n

(x − z)τ ατ +

|τ |=1

(x − z)σ gσ (x),

|σ|=n+1

where ατ are constants and gσ ∈ C ∞ (U ). Applying to this equality all kinds of operators of the form ∂ |τ | /∂xτ (z), |τ | n, we see that ατ =

1 ∂ |τ | f (z). τ ! ∂xτ

2.10 Corollary. Let f (x) ∈ C ∞ (R) and f (0) = 0. Then the function f (x)/x belongs to C ∞ (R). In fact, by Hadamard’s lemma, any smooth function f (x) ∈ C ∞ (R) is representable in the form f (x) = f (0) + xg(x), where g(x) ∈ C ∞ (R). If, in addition, f (0) = 0, then f (x)/x = g(x). 2.11 Lemma. If f ∈ C ∞ (Rn ) and f (z) = f (y) = 0, where z and y are two diﬀerent points of the space Rn , then the function f can be represented as a sum of products gi hi , where gi (z) = 0, hi (y) = 0. By a linear coordinate change, the problem can be reduced to the case z = (0, . . . , 0, 0), y = (0, . . . , 0, 1). By Hadamard’s lemma 2.8, any function f satisfying f (z) = 0 is representable in the form f (x) =

n

xi gi (x).

i=1

Let us represent the function gi (x) in the form gi (x) = gi (x) − αi + αi , where αi = gi (y). Since in the product xi gi (x)−αi the ﬁrst factor vanishes at the point z, while the second one vanishes at the point y, the problem reduces to the case of a linear function n αi xi . f (x) = i=1

The condition f (y) = 0 means that αn = 0. Representing now xi , i < n, in the form xi = xn xi − xi (xn − 1), we conclude the proof of the lemma.

Special Smooth Functions on Rn

19

2.12 Exercises. 1. Show that any function f (x, y) ∈ C ∞ R2 vanishing on the coordinate cross K = {x = 0} ∪ {y = 0} is of the form f = xy g(x, y), g(x, y) ∈ C ∞ R2 . 2. Does a similar result hold if the cross is replaced by the union of the x-axis and the parabola y = x2 ?

3 Algebras and Points

3.1. This chapter is a mathematical exposition of the algebraic approach to manifolds, which was sketched in intuitive terms in Chapter 1. Here we give a detailed answer to the following fundamental question: Given an abstract R-algebra F, ﬁnd a set (smooth manifold) M whose R-algebra of (smooth) functions can be identiﬁed with F. Further, F will always be a commutative, associative algebra with unit over R, or brieﬂy, an R-algebra. All R-algebra homomorphisms α : F1 → F2 , i.e., maps of F1 into F2 preserving the operations α(f + g) = α(f ) + α(g),

α(f · g) = α(f ) · α(g),

α(λf ) = λα(f ),

are assumed unital (i.e., α sends the unit in F1 into the one in F2 ). We stress that the elements of F, also called “functions,” are not really functions at all; they are abstract objects of an unspeciﬁed nature. The point is to turn these objects into real functions on a manifold. In order to succeed in this undertaking, we shall successively impose certain conditions on F. The key terms will be geometrical (Section 3.7), complete (Section 3.27), and smooth (Section 4.1) R-algebras. We begin with simple illustrations of how an abstractly deﬁned R-algebra can acquire substance and become a genuine algebra of nice functions on a certain set. 3.2. Example. Suppose F is the R-algebra of all inﬁnite sequences of real numbers {ai } = (a0 , a1 , a2 , . . .) such that ai = 0 for all i, except perhaps a ﬁnite number. The sum operation and multiplication by elements of R is deﬁned term by term (λ{ai } = {λai }, etc.). The product {ci } of two © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 3

21

22

Chapter 3

sequences {ai } and {bi } is deﬁned by the formula ak bl . ci = k+l=i

Can this algebra F be realized as an algebra of nice functions on some set M? We hope the reader has guessed the answer. By putting {ai } → ai xi i0

(this sum is always ﬁnite), we obtain an R-algebra isomorphism F → R[x] of F onto the R-algebra of polynomials in x, R[x]. Thus any sequence {ai } ∈ F may be viewed as the function on M = R given by x → i0 ai xi . 3.3. Exercise. Suppose that the R-algebras F1 and F2 , as linear spaces, are isomorphic to the plane R2 = {(x, y)}. Let the multiplication in F1 and F2 be respectively given by (x1 , y1 ) · (x2 , y2 ) = (x1 x2 , y1 y2 ), (x1 , y1 ) · (x2 , y2 ) = (x1 x2 + y1 y2 , x1 y2 + x2 y1 ). Find the set (manifold) Mi for which the algebra Fi , i = 1, 2, is the algebra of smooth functions, explicitly indicating what function on Mi corresponds to the element (x, y) ∈ Fi . Are the algebras F1 and F2 isomorphic? 3.4. We now return to our given abstract R-algebra F. Recalling the philosophy of Section 1.16 (a point of a manifold or state of a physical system is determined by all the relevant measurements), we introduce the following notations and deﬁnitions. Denote by M = |F| the set of all R-algebra homomorphisms of F onto R: M x : F → R, f → x(f ). The elements of M will sometimes be called R-points for the algebra F (they will indeed be the points of our future manifold), and |F| the dual space of R-points. Further, set F = f˜: M → R | f˜(x) = x(f ), f ∈ F . (3.1) The set F has a natural R-algebra structure given by f˜ + g˜ (x) = f˜(x) + g˜(x) = x(f ) + x(g), f˜ · g˜ (x) = f˜(x) · g˜(x) = x(f ) · x(g), λf˜ (x) = λf˜(x) = λx(f ). There is a natural map τ : F → F,

f → f˜.

(3.2)

Algebras and Points

23

We would like this map to be an isomorphism: Then we could view F as a realization of F in the form of an R-algebra of functions on the dual space M = |F|. But is this the case? f → f˜, is a homomorphism. 3.5. First we note that τ : F → F, Indeed, by deﬁnition of F (and because any x ∈ M is a homomorphism), f + g (x) = x(f + g) = x(f ) + x(g) = f˜(x) + g˜(x) = f˜ + g˜ (x). The other two veriﬁcations are similar, and we leave them to the industrious reader. It is also obvious that τ is surjective. Thus it remains to show that τ is injective. Unfortunately, this is not so in the general case. 3.6. Example. Suppose F is the R-algebra isomorphic (as a linear space) to the plane R2 = {(x, y) | x, y ∈ R} with the product (x1 , y1 ) · (x2 , y2 ) = (x1 x2 , x1 y2 + x2 y1 ). We shall show that the dual space M = |F| consists of a single point. This implies that τ is not injective, since F is then isomorphic to R, while F is not (F ⊃ {(x, 0), x ∈ R} ∼ = R). The element (1, 0) is obviously the unit of the algebra F, and any element (x, y) has an inverse if x = 0, namely (x, y)−1 = x−1 , −yx−2 . Hence the only ideal of the algebra F, other than the ideals {0} and F, is the ideal I = {(0, y) | y ∈ R}. The quotient algebra F/I is naturally isomorphic to R, the quotient map q : F → F/I = R being the projection (x, y) → x. This map is the only surjective R-algebra homomorphism F → R, so that M = {q}. 3.7. In order to be able to assert that τ : F → F is injective (and hence an isomorphism), certain conditionsmust be imposed on F. Note that τ will be injective iﬀ the ideal I(F) = x∈M Ker x is trivial. Indeed, f ∈ Ker τ ⇐⇒ τ (f ) = f˜ = 0 ⇐⇒ f˜(x) = x(f ) = 0 ∀x ∈ M Ker x = I(F), ⇐⇒ f ∈ x∈M

and therefore Ker τ = 0 ⇐⇒ I(F) = 0.

This motivates (mathematically) the following deﬁnition. Definition. An R-algebra F is called geometrical if I(F) = Ker x = 0. x∈|F |

(In the previous example, I(F) = I = {(0, y) | y ∈ R} = 0.) It is worth noticing that an algebra with empty dual space is not geometrical.

24

Chapter 3

Exercises. 1. Prove that the polynomial algebra R[x1 , . . . , xn ] is geometrical. 2. Let V be a ﬁnite-dimensional vector space over R and let G : V → V be a linear operator. Consider the R-algebra FG generated by Gk , k = 0, 1, . . . , as a vector space. Characterize the operators G for which FG is geometrical. 3.8. In Sections 3.5, 3.7, we have in fact proved the following theorem. Theorem. Any geometrical R-algebra F is canonically isomorphic to the R-algebra F of functions deﬁned on the dual space M = |F| of R-points (M x : F → R) by the rule f (x) = x(f ). Having this isomorphism in mind, we shall identify our abstract algebra F (which will usually be assumed geometrical) with the R-algebra F of functions on the dual space M = |F| once and for all. The notation F will be abandoned; the elements f ∈ F will often be viewed as functions M → R. 3.9 Exercises. Check which of the following algebras are geometrical: 1. The formal series algebra R[[x1 , . . . , xn ]]. 2. The quotient algebra R[x1 , . . . , xn ]/f k R[x1 , . . . , xn ], f ∈ R[x1 , . . . , xn ]. 3. The algebra of germs of smooth functions at 0 ∈ Rn . 4. The algebra of all smooth bounded functions. 5. The algebra of all smooth periodic functions (of period 1) on R (see Section 3.18). 6. The subalgebra of the previous algebra consisting of all even functions. 7. The algebras F1 and F2 described in Exercise 3.3. 8. The algebra of all diﬀerential operators in Rn with constant coeﬃcients (multiplication in this algebra is the composition of operators). 3.10. The algebra FS of functions corresponding to measuring devices of a classical physical system S (see Sections 1.9, 1.15) is always geometrical. This property is the mathematical formulation of the classical physical postulate asserting that if all the readings of two diﬀerent devices for all the states of the system S are the same, then these two devices measure the same physical parameter (i.e., only one of the devices is needed). 3.11 Proposition. For an arbitrary R-algebra F, the quotient R-algebra Ker p, F/I(F), where I(F) = p∈|F |

Algebras and Points

25

is geometrical and |F| = |F/I(F)|. Deﬁne the map ϕ : |F/I(F)| → |F| by assigning to each homomorphism b : F/I(F) → R the homomorphism ϕ(b) = a = b ◦ pr, where pr is the quotient map pr : F → F/I(F). We claim that ϕ is bijective. Obviously, b1 = b2 implies a1 = a2 , so ϕ is injective. Now suppose a ∈ |F|. Then Ker a ⊃ I(F). Hence the element b([f ]) = a(f ), where [f ] is the coset of the element f modulo I(F), is well deﬁned and determines a homomorphism b : F/I(F) → R. Clearly, we have a = ϕ(b), i.e., the map ϕ is surjective, so that ϕ identiﬁes |F| with |F/I(F)|. Suppose further that b ∈ |F/I(F)| and a = ϕ(b) = b ◦ pr. In that case Ker b = Ker a/I(F). Hence

Ker a/I(F) Ker b = I(F/I(F)) = b∈|F /I(F )|

=

a∈F

Ker a /I(F) = I(F)/I(F) = {0}.

a∈F

3.12. Given a geometrical R-algebra F, we intend to introduce a topology in the dual set M = |F| of R-points. From the physical point of view, two states s1 , s2 of a classical system S (two R-points) are near each other if all the readings of the relevant measuring devices are close, i.e., for all measuring devices D we must have fD (s2 ) ∈ ]fD (s1 ) − ε, fD (s1 ) + ε[ . Mathematically, we express this by saying that the topology in M , called the dual topology, is given by the basis of open sets of the form f −1 (V ), where V ⊂ R is open and f ∈ F. (The reader should recall at this point that the expression f −1 is meaningful only because we have identiﬁed F with an algebra of functions f : M → R.) Another way of saying this is the following: The topology in the dual space M = |F| is the weakest for which all the functions in F are continuous. 3.13 Proposition. The topology introduced in Section 3.12 in the dual space M = |F| is that of a Hausdorﬀ space. Suppose x and y are distinct points of |F|, i.e., diﬀerent homomorphisms of F into R. This means there is an f ∈ F for which f (x) = f (y), say f (x) < f (y). Then the sets f (x) + f (y) f (x) + f (y) , +∞ f −1 −∞, , f −1 2 2 are nonintersecting neighborhoods of the points x and y. When speaking of the “space” M = |F|, it is this topological (Hausdorﬀ) structure that will always be understood.

26

Chapter 3

3.14. In this section we assume that F0 is any R-algebra of functions on a given set M0 . Then there is a natural map θ : M0 → |F0 | assigning to each point a ∈ M0 the homomorphism f → f (a). In other words, θ(a)(f ) = f (a), a ∈ M0 , and therefore if an element f ∈ F0 , viewed as a function on the dual space |F0 |, vanishes on θ(M0 ), then f is the zero element of F0 . In particular, the algebra F0 will be geometrical, and we have the following result. Proposition. If F0 is a subalgebra of the R-algebra of continuous functions

on the topological space M0 , then the map θ : M0 → |F0 |, a → f → f (a) , is continuous. Suppose U = f −1 (V ) is a basis open set in |F0 |. By deﬁnition U consists of all the homomorphisms F0 → R that send the (ﬁxed) function f ∈ F to some point of the open set V ⊂ R. Then the inverse image θ−1 (U ) consists of all points a ∈ M0 such that f (a) ∈ V and is therefore an open subset of M. It should be noted that in our general situation (when F0 is any geometrical R-algebra, M0 = |F0 |), M0 is a topological space and the elements of F0 are continuous functions (see Section 3.12), so that the proposition proved above applies. Exercise. Describe the dual space for each of the following algebras: 1. R[x, y]/xyR[x, y];

2. R[x, y, z]/ x2 + y 2 + z 2 − 1 R[x, y, z]. 3.15. Example. Suppose F = R[x1 , . . . , xn ] is the R-algebra of polynomials in n variables. Every homomorphism a : F → R is determined by the “vector” (λ1 , . . . , λn ), where λi = a(xi ), since ⎛ ⎞

kn a⎝ ck1 ...kn xk11 · · · xknn ⎠ = ck1 ...kn a(x1 ))k1 · · · (a(xn ) k1 ,...,kn

k1 ,...,kn

=

ck1 ...kn λk11 · · · λknn .

k1 ,...,kn

Moreover, the map ck1 ...kn xk11 · · · xknn −→ F k1 ,...,kn

ck1 ...kn λk11 · · · λknn ∈ R

k1 ,...,kn

is a homomorphism of the algebra F into R for all λ1 , . . . , λn ∈ R. Thus the dual space |F| in this case is naturally identiﬁed with Rn = {(λ1 , . . . , λn )}. The topology deﬁned in |F| (see Section 3.12) coincides with the usual topology of Rn . Indeed, the sets f −1 (V ), where f is a polynomial and V ⊂ R is open, are open in Rn = |F|, since polynomials are continuous functions. Moreover, a

Algebras and Points

27

ball of radius r with center (b1 , . . . , bn ) in Rn = |F| is of the form f −1 (R+ ), where R+ is the positive half of R, if we take f to be n f (x1 , . . . , xn ) = r2 − (bi − xi )2 . i=1

Since such balls constitute a basis for the usual topology in Rn , the two topologies coincide. 3.16. Example. Suppose F = C ∞ (U ) is the R-algebra of inﬁnitely differentiable real-valued functions on an open subset U of Rn . Consider the map θ : U → |F|,

x → (f → f (x)).

We claim that the map θ is a homeomorphism, so that the dual space |C ∞ (U )| is homeomorphic to U . Since injectivity is obvious (elements of F being functions on U ), we ﬁrst prove the surjectivity of θ. Suppose p ∈ |F|, i.e., p : F → R is an R-algebra homomorphism onto R. Choose a smooth function fc ∈ C ∞ (U ) all of whose level surfaces are compact (such a function exists by Proposition 2.7). Then, in particular, the set L = fc−1 (λ), where λ = p(f ), is compact. Assume that p ∈ |F| does not correspond to any point of U . Then for any point a ∈ U there exists a function fa ∈ F for which fa (a) = p(fa ). The sets Ua = {x ∈ U | fa (x) = p(fa )},

a ∈ L,

constitute an open covering of L. Since L is compact, we can choose a ﬁnite subcovering Ua1 , . . . , Uam . Consider the function g = (f − p(f ))2 +

m

(fai − p(fai ))2 .

i=1

This is a smooth nonvanishing function on U , so that 1/g ∈ F. Since p is a (unital!) R-algebra homomorphism, we must have p(1) = p(g · (1/g)) = p(g) · p(1/g) = 1. But by the deﬁnition of g, p(g) = (p(f ) − p(f ))2 +

(3.3)

(p(fai ) − p(fai ))2 = 0,

which contradicts (3.3), proving the surjectivity of θ. The fact that θ is a homeomorphism is an immediate consequence of Proposition 3.14. In particular, we have proved that |C ∞ (Rn ) | = Rn . 3.17 Exercises. Describe the dual space for each of the following algebras:

1. C ∞ R3 / x2 + y 2 + z 2 − 1 C ∞ R3 .

28

Chapter 3

2. C ∞ R3 / x2 + y 2 − z 2 C ∞ R3 . 3. Smooth even functions on the real line. 4. Smooth even functions of period 1 on the real line. 5. Smooth functions of rational period (not necessarily the same) on the real line. 6. Functions deﬁned on the real line as the ratio of two polynomials p(x)/q(x), where q(x) = 0 for all x ∈ R. 7. The same functions as before, but with the additional requirement deg p(x) deg q(x). 8. Functions deﬁned on the real line as the ratio of two polynomials p(x)/q(x), where q(x) is not identically zero (i.e., deﬁned as rational functions).

9. The subalgebra {f ∈ C ∞ R2 | f (x + 1, y) = f (x, y)} of C ∞ R2 .

10. The subalgebra {f ∈ C ∞ R2 | f (x + 1, −y) = f (x, y)} of C ∞ R2 .

11. The subalgebra {f ∈ C ∞ R2 | f (x, y + 1) = f (x, y) = f (x + 1, y)}

of C ∞ R2 . 12. The subalgebra {f ∈ C ∞ (R3 \ 0) | f (x, y, z) = f (λx, λy, λz),

∀ λ = 0}

of C ∞ (R3 \ 0).

2

∞ R | f (x + 1, −y) = f (x, y) = f (x, y + 1)} 13. The subalgebra {f ∈ C 2

∞ R . of C 14. The subalgebra {f ∈ C ∞ (R3 \ 0) | f (x, y, z) = f (λx, λy, λz), ∞

3

∀λ ∈ R+ }

of C (R \ 0). 3.18. Example. Suppose F consists of all periodic smooth functions of period 1 on the line R. Then, as usual, each point a ∈ R determines the homomorphism F → R, f → f (a). But diﬀerent points can give rise to the same homomorphism; this happens iﬀ the distance between the points is an integer. We claim that there are no homomorphisms other than the ones determined by the points a ∈ R. The proof is similar to the one in the previous section. Namely, if the homomorphism p : F → R is not determined by any point, then for any a ∈ R there exists a function fa ∈ F such that p(fa ) = fa (a). From the open covering of the closed interval [0, 1] by sets of the form Ua = {x ∈ R | fa (x) = p(fa )}, a ∈ R,

Algebras and Points

29

choose a ﬁnite subcovering Ua1 , . . . , Uan . The function m g= (fai − p(fai ))2 i=1

does not vanish anywhere on [0, 1] and, by periodicity, anywhere on R. Hence 1/g ∈ F, etc., just as in Section 3.16. Thus in our case |F| can be identiﬁed with the quotient space R/Z, where Z is the subgroup of integers in R. Of course, R/Z = S 1 is the circle. Thus we have shown rigorously that smooth periodic functions of period 1 on R are actually functions on the circle, which is in accord with our intuitive understanding of such functions. 3.19. Now suppose F1 and F2 are two geometrical R-algebras and ϕ : F1 → F2 is an R-algebra homomorphism. Then for the dual spaces of R-points |F1 | and |F2 | the dual map |ϕ| arises: |ϕ| : |F2 | → |F1 |,

x → x ◦ ϕ.

We claim that the map |ϕ| is continuous. Indeed, take a basis open set U = f −1 (V ) ⊂ |F1 |, where f ∈ F1 and V ⊂ R is open. Then U consists of all points x ∈ |F1 | such that f (x) ∈ V . The inverse image of U by |ϕ| consists of all points y ∈ |F2 | such that |ϕ|(y) ∈ U , i.e., f (|ϕ|(y)) ∈ V . But f (|ϕ|(y)) = f (y ◦ ϕ) = (y ◦ ϕ)(f ) = y(ϕ(f )) = ϕ(f )(y) (the reader should check each of these relations!). Therefore, the set |ϕ|−1 (U ) consists of all points y ∈ |F2 | for which ϕ(f )(y) ∈ V ; thus the set |ϕ|−1 (U ) is open. 3.20. If ϕ1 : F1 → F2 , ϕ2 : F2 → F3 are R-algebra homomorphisms of geometrical R-algebras F1 , F2 , F3 , then obviously |ϕ2 ◦ ϕ1 | = |ϕ1 | ◦ |ϕ2 | and | idFi | = id|Fi | . Further, if ϕ : F1 → F2 has an inverse homomorphism ϕ−1 , then −1 ϕ = |ϕ|−1 . In particular, if ϕ is an isomorphism, then |ϕ| is a homeomorphism. 3.21. Having started from an abstract geometrical R-algebra F, we have constructed the (Hausdorﬀ) topological space M = |F| (the dual space of R-points), for which F is a subalgebra of the algebra of all continuous functions. It might now seem that we need only postulate that F be locally isomorphic to C ∞ (Rn ) i.e., cover M with a family of sets Ei , M = Ei , such that the restriction of F to each Ei is isomorphic to C ∞ (Rn ) , and our program of deﬁning a manifold in terms of its R-algebra of functions will be carried out. Unfortunately, things are not as simple as they appear at ﬁrst glance: certain technical diﬃculties, related to the notion of restriction, must be overcome before we succeed in implementing our program.

30

Chapter 3

3.22. Example. Suppose F = C ∞ (R) and R+ ⊂ R is the set of positive real numbers. We would like to obtain the algebra of smooth functions on R+ as a “restriction” of the algebra F. But consider the function x → 1/x on R+ ; it is certainly a smooth function on R+ , but clearly is not the restriction of any function f ∈ F = C ∞ (R). How can such functions be obtained from F? 3.23. Definition. Suppose F is a geometrical R-algebra and A ⊂ |F| is any subset of its dual space |F|; the restriction F A of F to A is the set of all functions f : A → R such that for any point a ∈ A there exists a neighborhood U ⊂ A and an element f¯ ∈ F such that the (ordinary) restriction of f to U coincides with the restriction of f¯ (understood as a function on |F|) to U . Obviously, F A is an R-algebra. Now we can return to Example 3.22. We claim that the function x → 1/x belongs to C ∞ (R)R . +

Indeed, for any point a > 0 there exists (see Section 2.5) a function α ∈ C ∞ (R) that vanishes when x a/3 and equals 1 whenever x 2a/3. For f¯ take the function that vanishes when x 0 and equals α(x)/x when x > 0. Obviously, f¯ is smooth and coincides with the function x → 1/x in the neighborhood ]2a/3, 4a/3[ of the point a. In a similar way we can show that any smooth function on R+ belongs to C ∞ (R)R ; i.e., we have + C ∞ (R)R = C ∞ (R+ ). +

This statement has the following generalization. 3.24 Proposition. If F = C ∞ (U ),where U ⊂ Rn is not empty and open, while V is open in U = |F|, then F V = C ∞ (V ). The identiﬁcation U = |F| in the statement of the proposition was established in Section 3.16. Suppose f ∈ C ∞ (V ) and x ∈ V . By 2.6 there exists a neighborhood W of x such that the point W ⊂ V and a function g ∈ C ∞ (Rn ) such that g W = f W . If g¯ = g U , then we will also have g¯W = f W . Thus f ∈ C ∞ (U ) ; i.e., C ∞ (V ) ⊂ C ∞ (U ) . V

V

The inverse inclusion immediately follows from the deﬁnition of the algebra C ∞ (U )V . Exercise. For the subsets A ={1/n | n = 1, 2, 3, . . . } and B = A ∪ {0} in R describe the restrictions R[x]A and R[x]B . 3.25. In the general case, in which F is any geometrical R-algebra and A is a subset of the dual space |F|, we can assign to every function f ∈ F its restriction to A ⊂ |F|, which obviously belongs to F A . Thus we obtain

Algebras and Points

the restriction homomorphism ρA : F → F A ,

31

f → f A .

(Here as usual the element f ∈ F is viewed as a function on the dual space |F|.) Proposition. Suppose i : F1 → F2 is an isomorphism of two geometrical algebras, A2 ⊂ |F2 |, A1 = |i|(A2 ). Then the map

F1 A → F2 A , f → f |i|A 1

2

2

is an isomorphism. The proof is a straightforward veriﬁcation of deﬁnitions.

3.26. Now, in the most important particular case A = |F|, we can consider the restriction homomorphism ρ : F → F |F | . Since F is assumed geometrical (diﬀerent elements f ∈ F are identiﬁed with diﬀerent functions on |F|), ρ is injective. Surjectivity, surprisingly enough, is not obvious: By the deﬁnition given in Section 3.23, F |F | consists of all functions that are locally like those of F, but it is not clear why all such functions belong to F. Indeed, this is not always the case. 3.27. Example. Suppose F is the subalgebra of the algebra C ∞ (Rn ) consisting of functions each of which is less in absolute value than some polynomial. Then the dual space |F| is homeomorphic to Rn . The proof is similar to the one given in Section 3.16, except that the function f with compact level surfaces must be chosen so that it belongs to F but f (x) → ∞ as x → ∞; e.g., we can take f : x → x2 + 1. Then 1/g(x) → 0 as x → ∞, so that 1/g also belongs to F, and the proof proceeds as 3.16. in Section Thus F |F | = F Rn coincides with the algebra C ∞ (Rn ) of all smooth functions on Rn , since any function f ∈ C ∞ (Rn ) in a neighborhood of some point a ∈ Rn coincides with the function f θ, where θ is a smooth function that vanishes outside the ball of radius 2 and center a and equals 1 inside the concentric ball of radius 1 (see Section 2.5) and, obviously, f θ ∈ F. Hence ρ : F → F |F | = C ∞ (Rn ) cannot be surjective. 3.28. Definition. A geometrical R-algebra F is said to be complete if the restriction homomorphism ρ : F → F |F | is surjective (and is therefore an isomorphism), i.e., if any function |F| → R locally coinciding with elements of F is itself an element of F. It is clear that the algebras C ∞ (U ), where U ⊂ Rn is open, are complete (see Section 3.24). The algebra in the previous example (Section 3.27) is not complete. Exercise. Determine which of the following algebras are complete. 1. F = R[x].

32

Chapter 3

2. The algebra of all smooth bounded functions. 3. The algebra of all smooth periodic functions (of period 1) on R (see Section 3.18). 3.29. We now return to the general situation in which A is a subset of the dual space |F| of a geometrical R-algebra F. It is natural to ask the following question: Can the set F A be identiﬁed with A ⊂ F? Proposition. Suppose F is a geometrical R-algebra and A ⊂ |F|. Then the map μ : A → F A , (μ(a)) (f ) = f (a), is a homeomorphism onto a subset of the space F A . Since all the elements of F, understood as functions on the dual space |F|, are continuous, F A is a subalgebra of the algebra of continuous functions on A, and therefore μ is continuous by Proposition 3.14. Further, μ is injective: If a1 and a2 are distinct points of A, there is a function f0 ∈ F taking diﬀerent values at these points; but then f 0 A , which belongs to the algebra F A , has the same values as f0 at a1 and a2 , and hence these points determine diﬀerent homomorphisms F A → R. To prove that the inverse map μ−1 : μ(A) → A is continuous, consider a basis open set in A of the form A ∩ f −1 (V ), where f ∈ F and V ⊂ R is open. It is mapped onto the set μ(A) ∩ (f A )−1 (V ), i.e., onto an open subset of μ(A). This proposition immediately implies that A ⊂ B ⊂ |F| ⇒ F B A = F A . Should μ(A) coincide with |F A |, we would have a positive answer to the question put at the beginning of this section. Proposition 3.24 implies that this is true for algebras F = C ∞ (U ), where U ⊂ Rn is open, when A ⊂ |F| = U is also open. However, this is false in the general case. 3.30. Example. Suppose F = R[x]. Let A = R+ ⊂ R be the positive reals. Then the restriction homomorphism ρA : R[x] → R[x]R is an iso+ morphism, since any nth degree polynomial is determined by its values at (n + 1) points, by its values on R+ . On the other hand, hence the map μ : R+ → R[x]R is the inclusion of R+ into R = R[x]R = R[x], so + + that here A = R+ cannot be identiﬁed with F = R. A

3.31 Exercise. Describe an R-algebra whose dual space is the conﬁguration space of one of the given hinge mechanisms (see Section 1.14).

Algebras and Points

33

3.32. In order to avoid situations like the one in Example 3.30, we need a condition that would guarantee the bijectivity of the map μ : A → F A , a → (f → f (a)). Definition. A geometrical R-algebra F is said to be closed with respect to smooth composition, or C ∞ -closed , if for any ﬁnite collection of its elements f1 , . . . , fk ∈ F and any function g ∈ C ∞ Rk there exists an element f ∈ F such that f (a) = g(f1 (a), . . . , fk (a))

for all a ∈ |F|.

(3.4)

Note that the function f ∈ F appearing in this deﬁnition is uniquely determined (since F is geometrical). For the case in which FS is the algebra of functions determined by measuring devices of a physical system S, the algebra FS is always C ∞ -closed. This is because the composite function (3.4) may be constructed by means of a device synthesizing the function g(f1 , . . . , fk ); see Section 1.10. Exercise. Determine which of the algebras listed in Exercise 3.28 are closed. 3.33. We shall now show that the map μ : A → |F A | (see Section 3.29) is surjective (and therefore a homeomorphism) for C ∞ -closed algebras F in the case of any basis open set A: A = {a ∈ |F| | α < h(a) < β}, α, β ∈ R, h ∈ F. (We shall not require this fact in more general form in what follows.) By Corollary 2.3, there exists a function g ∈ C ∞ (R) such that g ≡ 0 on R \ ]α, β[ and g > 0 on ]α, β[. Since F is C ∞ -closed, there is a function f ∈ F such that f (a) = g(h(a)) for all points a ∈ |F|. Then f (a) > 0 whenever a ∈ A, so that f A is an invertible element of the algebra F A . Further, suppose b ∈ |F| is the image of some point b ∈ F A under the natural map F A → |F|. If b ∈ / A, then

0 = f (b ) = f A (b), which contradicts the fact that f A is invertible. Thus μ(A) = F A . 3.34. Having in mind the results of Section 3.33, we would like to modify a given geometrical R-algebra F so as to obtain a C ∞ -closed algebra F. The most direct way to do that is the following. Identifying F with the corresponding algebra of functions on |F|, consider the set F of functions on |F| that can be represented in the form

g(f1 , . . . , fl ), where l ∈ N, f1 , . . . , fl ∈ F, g ∈ C ∞ Rl .

34

Chapter 3

The set F has an obvious R-algebra structure, and F is a subalgebra of F. Denote the natural inclusion F ⊂ F by iF . Since the composition of smooth functions is smooth, the algebra F is C ∞ -closed. It is also geometrical, being the algebra of certain functions on a set (see Section 3.14). Thus we have constructed a natural inclusion map iF : F → F for any geometrical R-algebra F into a C ∞ -closed R-algebra F, which we (temporarily) call the C ∞ -closure of F. This algebra possesses the following remarkable property. 3.35 Proposition. Any homomorphism α : F → F of a geometrical R-algebra F into a C ∞ -closed R-algebra F can be uniquely extended to a homomorphism α : F → F of its C ∞ -closure F. Assume that the required extension α exists, i.e., that α = α ◦ iF , where iF : F → F is the natural inclusion. Then, by Section 3.20, we have |α| = |iF | ◦ |α|. Here |α| denotes the dual map (see 3.19). Further, for any point a ∈ |F |, α(g(f1 , . . . , fl ))(a) = g(f1 , . . . , fl )(|α|(a)) = g(iF (f1 ), . . . , iF (fl ))(|α|(a)) = g(f1 , . . . , fl )(|iF |(|α|(a))) = g(f1 , . . . , fl )(|α|(a)) = g(f1 (|α|(a)), . . . , fl (|α|(a))) = g(α(f1 ), . . . , α(fl ))(a). Since F is geometrical, this implies α(g(f1 , . . . , fl )) = g(α(f1 ), . . . , α(fl )). If α exists, this last formula proves its uniqueness. To prove existence, we can use this formula as the deﬁnition of α, if we establish that the righthand side is well deﬁned, i.e., if we show that g(f1 , . . . , fl ) = g (f1 , . . . , fl ) implies g(α(f1 ), . . . , α(fl )) = g (α (f1 ) , . . . , α (fl )) . Since F is geometrical, it suﬃces to prove this at an arbitrary point a ∈ F . But g(α(f1 ), . . . , α(fl ))(a ) = g (α(f1 ) (a ) , . . . , α(fl ) (a )) = g(f1 (a), . . . , fl (a)) = g(f1 , . . . , fl )(a), where a = |α|(a ). Similarly, g (α (f1 ) , . . . , α (fl )) (a ) = g (f1 , . . . , fl ) (a). Comparing the last two formulas, we see that α is well deﬁned, concluding our proof. 3.36. It is remarkable that Proposition 3.35 entirely characterizes the C ∞ -closure F of a geometrical R-algebra F. To explain this in adequate terms, we need the following deﬁnition.

Algebras and Points

35

Definition. A C ∞ -closed geometrical R-algebra F together with a homomorphism i : F → F is called the smooth envelope of the R-algebra F if for any homomorphism α : F → F of F into a geometrical C ∞ -closed Ralgebra F there exists a unique homomorphism α : F → F extending α (i.e., such that α = α ◦ i). In other words, under the above assumptions, the diagram α

F i

F α

F can always be uniquely completed (by the dotted arrow α) to a commutative one. It now follows from Proposition 3.35 that the C ∞ -closure (see 3.34) is a smooth envelope of F. 3.37 Proposition. The smooth envelope of any F is unique up

R-algebra to isomorphism. More precisely, if the pairs, ik , F k , k = 1, 2, are smooth envelopes of F, there exists a unique isomorphism j : F 1 → F 2 such that i2 = j ◦ i1 . In other words, the following diagram commutes: j

F1

F2

j −1 i1

i2

F

First we note that for any given smooth envelope i, F of F any homomorphism α : F → F satisfying α ◦ i = i is the identity, α = idF . (Indeed, by Deﬁnition 3.36, the “solution” of the “equation” α ◦ i = i is unique, but this equation has the obvious solution idF .) Further, according to the same deﬁnition for (i1 , F1 ), the homomorphism i2 : F → F 2 can be uniquely represented in the form i2 = j1 ◦ i1 , where j1 : F 1 → F 2 is a homomorphism. Similarly, i1 = j2 ◦i2 , where j2 : F 2 → F 1 is a homomorphism. Hence i2 = j1 ◦ i1 = j1 ◦ (j2 ◦ i2 ) = (j1 ◦ j2 ) ◦ i2 . By the remark at the beginning of the proof, this implies j1 ◦ j2 = idF 2 . Similarly, j2 ◦ j1 = idF 1 . Thus j1 and j2 are isomorphisms inverse to each other, and we can put j = j1 to establish the proposition. (The uniqueness of the isomorphism j follows from the deﬁnition of smooth envelopes.) A direct consequence of this proposition is that the temporary term “C ∞ -closure” (see 3.34) is characterized by its universal property expressed in Proposition 3.35 and therefore coincides with the term “smooth envelope.” It is the latter term that will be used from now on.

36

Chapter 3

3.38. In accordance with its deﬁnition (see Section 3.36), the smooth envelope F of a geometrical algebra F is an object that plays a universal role in its interactions (i.e., isomorphisms) with the “world” of C ∞ -closed geometrical algebras. We can say that the smooth envelope is the “ambassador plenipotentiary” of the algebra F in this “world,” and that F interacts with the latter exclusively via this ambassador. The reader may have noticed that the arguments used in Section 3.37 are very general in nature. The art of ﬁnding and using such arguments is one of the main facets of category theory, familiarly known as abstract nonsense. We feel that one has to get used to it before learning it mathematically, so we shall not develop the theory, but use many of its standard arguments and tricks (e.g., see Sections 6.4, 6.6, 6.16, 6.17). In the set-theoretic approach to mathematics, one studies the inner nature of mathematical objects, i.e., point sets supplied with certain structures. It is a biology of species. On the other hand, the categorical approach is a kind of sociology: one is no longer interested in the properties of individual objects, but in their relationships (called “morphisms” in the theory) with other objects of the same or similar type. One can also say that the categorical approach is similar to the experimental method in the natural sciences, when objects are not studied per se, but are analyzed in terms of their interaction with other objects. 3.39 Exercises. Find the smooth envelope of (1) the algebra R[x1 , . . . , xn ]; (2) the algebra of functions on the line R of the form f (x) =

n k=0

where ai (x) ∈ C ∞ (R).

ak (x)|x|k ,

x ∈ R,

4 Smooth Manifolds (Algebraic Definition)

4.1. A complete (Section 3.27) geometrical (Section 3.7) R-algebra F is called smooth if there exists a ﬁnite or countable open covering {Uk } of the dual space |F| such that all the algebras F U (Section 3.23) are isomorphic k to the algebra C ∞ (Rn ) of smooth functions in Euclidean space. The (ﬁxed positive) integer n is said to be the dimension of the algebra F. Smooth n-dimensional algebras are our main object of study; they can be viewed as R-algebras of smooth functions on n-dimensional smooth manifolds. From the viewpoint of formal mathematics, the R-algebra F entirely determines the corresponding manifold M as the dual space M = |F| of its R-points (Section 3.8) and is most convenient to work with: all of diﬀerential mathematics applies neatly to F, so that the space M is not formally required. Nevertheless, in order to be able to visualize M as a geometrical object, we must learn to work simultaneously with the smooth algebra F and the space M = |F| of its R-points. Learning this will be the main goal of the present chapter. Considering F and M = |F| simultaneously, we say that we are dealing with a smooth manifold. Although the second object in this pair is determined by the ﬁrst, we make a concession to our geometrical intuition (and to traditional terminology) and say that F is the algebra of smooth functions on the manifold M (= |F|). 4.2. A somewhat more general concept is that of a smooth algebra with boundary. In this case, for each element Uk of the covering {Uk } we require the algebra F U to be isomorphic either to C ∞ (Rn ) or to C ∞ (RnH ), where k

RnH = {(r1 , . . . , r2 ) ∈ Rn | r1 0},

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 4

37

38

Chapter 4

and C ∞ (RnH ) consists of the restrictions (in the usual sense) of all functions from C ∞ (Rn ) to the set RnH . Exercise. Prove that the algebras C ∞ (Rn ) and C ∞ (RnH ) are not isomorphic. The points of the space |F| that correspond to the boundary of the half-space RnH in the identiﬁcation Uk = RnH are called boundary points. Exercise. Prove that the set of boundary points ∂|F| has a natural structure of a smooth manifold (without boundary). As above, emphasizing the geometrical viewpoint on this concept, we shall say that F is the algebra of smooth functions on the smooth manifold M (= |F|) with boundary. 4.3 Lemma. If a geometrical R-algebra F is isomorphic to C ∞ (Rn ) or C ∞ (RnH ), then it is C ∞ -closed (see Section 3.32). The lemma immediately follows from the following stronger statement. If i : F1 → F2 is an isomorphism of geometrical R-algebras, and F1 is C ∞ -closed, then so is F2 . The veriﬁcation of this statement is quite similar to the uniqueness proof of α in Section 3.35, and the reader should have no diﬃculty in carrying it out. 4.4 Proposition. Smooth algebras are C ∞ -closed. (The same is true for smooth algebras with boundary.) Let F be a smooth R-algebra (possibly with boundary), let l ∈ N, g ∈ C ∞ Rl , f1 , . . . , fl ∈ F, and let {Uk } be the covering that appears in Deﬁnition 4.1 (or 4.2). Consider the function h : |F| → R,

h(a) = g(f1 (a), . . . , fl (a)). By Lemma 4.3, for any k there exists an hk ∈ F U such that k

∀a ∈ Uk ,

hk = g(f1 (a), . . . , fl (a)).

Thus in a neighborhood of each point the function coincides with a function from F. Since by 4.1 (or 4.2) F is complete (Section 3.28), it follows that h ∈ F. 4.5. Example. Suppose F is the algebra of smooth periodic functions on the line R of period 1: F = {f ∈ C ∞ (R) | f (r + 1) = f (r),

∀r ∈ R}. ∞

Being a subalgebra of the geometrical algebra C (R), the algebra F is itself geometrical (see Section 3.19). It is not diﬃcult to prove that F is

Smooth Manifolds (Algebraic Definition)

39

complete. (In Section 4.22 we shall present a general argument implying the completeness of all the algebras considered in Examples 4.5–4.8.) It was shown in 3.18 that the space |F| is the circle S 1 . Now consider the functions g1 , g2 ∈ F, g1 (r) = sin2 πr,

g2 (r) = cos2 πr,

and the open covering of the circle |F| by the sets Ui = {r ∈ |F| | gi (x) = 0},

i = 1, 2.

It is easy to establish bijections Ui ↔ ]0, 1[ that correspond to the isomorphisms F U ∼ (∼ = C ∞ (R)). = C ∞ ]0, 1[ i Thus the algebra F is a smooth algebra of dimension 1, and the manifold it determines is the circle S 1 = |F|. In a similar way one establishes that the R-algebra F = f ∈ C ∞ (R)2 | f (r1 + 1, r2 ) = f (r1 , r2 ) ∀(r1 , r2 ) ∈ R2 is a smooth algebra of dimension 2. In this case the space M = |F| is homeomorphic to the cylinder. 4.6 Exercise. Carefully review the previous example and ﬁnd the mistake in the following argument: since a 1-periodic smooth function takes the same ﬁnite value at the end points of the closed interval [0,1], while the algebra C ∞ ]0, 1[ contains unbounded functions as well as functions that have diﬀerent limits at the points 0 and 1, the algebra F ]0,1[ (where F is the algebra of 1-periodic smooth functions on R) cannot be isomorphic to C ∞ ]0, 1[ .

r1

r2

Figure 4.1. The M¨ obius band.

40

Chapter 4

4.7. Examples. I. F = {f ∈ C ∞ R2 | f (r1 , r2 ) = f (r1 + 1, −r2 ) for all (r1 , r2 ) ∈ R2 }. The space |F| is called the open M¨ obius band. II. F = {f ∈ C ∞ (Π) | f (r1 , r2 ) = f (r1 + 1, −r2 ) ∀(r1 , r2 ) ∈ Π}, where Π is the strip {(r1 , r2 ) ∈ Rr | −1 r2 1}. This is a smooth algebra with boundary, and the space |F| is known as the (closed) M¨ obius band (see Figure 4.1). III. F = f ∈ C ∞ R2 | f (r1 , r2 ) = f (r1 + 1, −r2 ) = f (r1 , r2 + 1) ∀(r1 , r2 ) ∈ R2 . Using the functions g1 (r1 , r2 ) = sin2 πr1 ,

h1 (r1 , r2 ) = sin2 πr2 ,

g2 (r1 , r2 ) = cos2 πr1 ,

h2 (r1 , r2 ) = cos2 πr2 ,

we can cover the space |F| by the four open sets Uik = {(r1 , r2 ) ∈ |F| | gi (r1 , r2 ) = 0, hk (r1 , r2 ) = 0}, i, k = 1, 2. For each of these sets one can immediately construct a homeomorphism on the open square corresponding to the isomorphism of the R-algebra F U on the algebra of smooth functions on the open square. Therefore, ik F U ∼ = C ∞ R2 , and F is a smooth R-algebra of dimension 2. The space ik |F| is known as the Klein bottle.

Figure 4.2. The Klein bottle.

It is useful to visualize how the squares Uik are “glued together” when they are embedded into |F|. The beginning of this process is pictured on the left-hand side of Figure 4.2. What happens if the process is continued in 3-space is shown on the right-hand side of the same ﬁgure (the little circular self-intersection does not really occur in the Klein bottle; it is due to the fact that the latter does not ﬁt into 3-space).

Smooth Manifolds (Algebraic Definition)

4.8 Exercise. Prove that the R-algebra F = f ∈ C ∞ R2 | f (r1 , r2 ) = f (r1 + 1, −r2 ) = f (−r1 , r2 + 1), (r1 , r2 ) ∈ R2

41

is smooth. Find as many geometrical descriptions of the topological space |F| as you can. Prove, for example, that |F| is homeomorphic to the space whose “points” are the straight lines of R3 passing through the origin (0, 0, 0); see Section 1.3. 4.9. Suppose F is the algebra of smooth functions on a manifold M = |F| with boundary. Recalling Deﬁnition 4.2, we say that a point p ∈ M is a boundary point of M if it corresponds to a boundary point of RnH in the identiﬁcation Uk = RnH . The set of all boundary points is denoted by ∂M and called the boundary of M . Exercise. 1. Prove that if the boundary ∂M of the n-dimensional manifold M = |F| is nonempty, then F ∂M is a smooth algebra of dimension (n − 1) and ∂M has no boundary points (see the second exercise in Section 4.2). 2. Check algebraically that the manifold ∂|F| in Example 4.7, II, can be identiﬁed with the circle (cf. Sections 4.5 and 6.9). 4.10. Remark. It is far from obvious that the dimension of a smooth algebra is well deﬁned, i.e., that it does not on the choice of the depend C ∞ (Rn ) (see 4.1). This covering {Uk } and of the isomorphisms F U ∼ = k almost immediately follows from the fact that the algebras C ∞ (Rn ) and C ∞ (Rm ) are not isomorphic if n = m. The reader who has industriously worked his way through the previous examples undoubtedly feels that this is true. A more experienced reader will probably have no trouble in proving this fact by using “Sard’s theorem on singular points of smooth maps” (advanced calculus). As for us, we shall prove this result in Chapter 9. Until then, our skeptical readers may consider dimension to be an invariant of the covering {Uk } rather than that of the algebra F itself. 4.11. Definition. Suppose F is the algebra of smooth functions on the manifold M and N ⊂ M = |F| is a subset. If the algebra FN = F N is a smooth R-algebra, then we say that N is a smooth submanifold of the smooth manifold M and that FN is the algebra of smooth functions on the submanifold N . If the restriction homomorphism i : F → FN is surjective, the smooth submanifold N ⊂ M = |F| is called closed. 4.12. Let N be a closed submanifold of M . Were we being consistent in 4.11 when we spoke of FN as “the algebra of smooth functions on the manifold N ”? The answer to that question is given by the following result.

42

Chapter 4

Proposition. Suppose F is the algebra of smooth functions on the manifold M and N ⊂ M = |F| is a closed smooth submanifold. Then (i) N is closed as a subset of the topological space M ; (ii) N = |FN |. (i) Let a ∈ M N be a limit point of N , and U ⊂ M a neighborhood of a such that F U ∼ = C ∞ (Rn ). This isomorphism may be chosen so that the elements of F U corresponding to the coordinate functions r1 , . . . , rn ∈ C ∞ (Rn ) vanish at thepoint a. Consider the function on F a corresponding to 1/ r12 + · · · + rn2 . This function may be extended from the punctured neighborhood U a of a to a smooth function g on the submanifold M a. It is obvious that the restriction g N belongs to the algebra F N = FN but it does not belong to the image of the restriction homomorphism i : F → F N . This contradiction proves that a ∈ N . (ii) Consider the R-point b : FN → R and take the composition c = b ◦ i : F → R. Assume that c ∈ / N . Generalizing Proposition 2.5, let us construct a function f ∈ F such that f N ≡ 0, f (c) = 0. But then i(f ) = 0 and f (c) = 0: a contradiction. Therefore, the map b → c = b ◦ i is a surjection of |FN | onto N . Together with Proposition 3.29 this gives the result. 4.13. Examples. In R2 consider the set of points S 1 given by the equation r12 + r22 − 1 = 0. Let us check that the R-algebra FS 1 = C ∞ R2 S 1 is isomorphic to the algebra of smooth periodic functions of period 1 on the line R (see Example 4.5). First, note that FS 1 = C ∞ R2 |S 1 = C ∞ (R2 \ {0})|S 1 and consider the map w : R → S 1 ⊂ R2 ,

r → (cos 2πr, sin 2πr).

Clearly, the corresponding R-algebra homomorphism |w| : FS 1 → C ∞ (R), |w|(f )(r) = f (w(r)) = f (cos 2πr, sin 2πr),

r ∈ R,

∞ is injective and its image is contained in the subalgebra Cper (R) of smooth 1-periodic functions on R. To prove that |w| is surjective, consider the ∞ (R) → C ∞ (R2 \ {0}) deﬁned by homomorphism ι : Cper arg z ι(f )(r1 , r2 ) = f , z = r1 + ir2 . 2π Obviously, |w|(ι(f )|S 1 ) = f .

Exercise. 1. Show that any odd 2π-periodic smooth function is of the form g(x) sin x, where g(x) is an even 2π-periodic smooth function.

Smooth Manifolds (Algebraic Definition)

43

2. Is it true that any even 2π-periodic smooth function can be written as f (cos x) for f ∈ C ∞ (R)? 4.14. Recall that the restriction of elements of an R-algebra F to a subset N ⊂ |F| was not deﬁned in algebraic terms, but in geometrical ones (see Section 3.23). However, if N ⊂ |F| is a closed smooth submanifold, then there is a purely algebraic way to ﬁnd the algebra FN = F N . Namely, suppose AN ⊂ F is the set of elements of F that vanish on N , i.e., AN = {f ∈ F | ∀a ∈ N, f (a) = 0}. This is obviously an ideal of the algebra F, so we can consider the quotient algebra F/AN . There exists an obvious identiﬁcation of F/AN with the algebra FN = F N (see the proof of Proposition 4.12) for which the quotient map ϕ : F → F A becomes the restriction homomorphism p : F → FN : N

F N

F/AN p

ϕ

F 4.15. Example. If S1 isthe circle from Example 4.13, then AS 1 is the principal ideal in C ∞ R2 generated by the function r12 + r22 − 1. Let f ∈ AS 1 . Let us prove that f (r1 , r2 ) = g(r1 , r2 ) · r12 + r22 − 1 for a suitable function g ∈ C ∞ R2 . Since the algebra C ∞ R2 is complete, if suﬃces to construct g in a neighborhood of S 1 , say, in R2 \{0}. To this end, introduce the following auxiliary functions: 1−t , u(t, r1 , r2 ) = t + 2 r1 + r22 h(t, r1 , r2 ) = f (r1 · u(t, r1 , r2 ), r2 · u(t, r1 , r2 )) . Then, taking into account the fact that ∂u 1 r12 + r22 − 1 =1− 2 = , ∂t r12 + r22 + r12 + r22 r1 + r22 h(0, r1 , r2 ) = 0 (since f ∈ AS 1 ), h(1, r1 , r2 ) = f (r1 , r2 ),

44

Chapter 4

we obtain

1 f (r1 , r2 ) = 0

∂h (t, r1 , r2 )dt ∂t

1 =

0

∂f ∂f (r1 u, r2 u) + r2 (r1 u, r2 u) dt r1 ∂r1 ∂r2 2 2 · r + r − 1 . 1 2 r12 + r22 + r12 + r22

The ﬁrst factor in the last expression is the desired function g(r1 , r2 ). 4.16 Exercise. Let A = C ∞ R2 . Show that the algebra A/ y 2 − x3 A is not smooth. A traditional purely algebraic approach to solving such problems is based on the following two facts: 1. If an algebra is smooth, then its localization at a maximum ideal is isomorphic to the algebra of germs of smooth functions on some Rk (at the origin), see Example III on page 156; 2. The formal completion of the above local algebra is isomorphic to the algebra of formal power series. The task becomes much simpler if one uses certain elementary tools of diﬀerential calculus over commutative algebras (see Exercise 9.34). 4.17. Lemmas. It is not diﬃcult to generalize the statements in Sections 2.3–2.7 from Rn to arbitrary manifolds. In particular, for any arbitrary algebra F of smooth functions on a manifold M , the following statements hold: (i) For any open set U ⊂ M there exists a function f ∈ F such that

f (x) > 0 for all x ∈ U, f (x) = 0 if x ∈ / U. (ii) For any two nonintersecting closed subsets A, B ⊂ M there exists a function f ∈ F such that ⎧ ⎪ for all x ∈ A, ⎨f (x) = 0 f (x) = 1 for all x ∈ B, ⎪ ⎩ 0 < f (x) < 1 otherwise. (The reader possibly established a weaker statement when working through Proposition 4.12.) (iii) There exists a function f ∈ F all of whose level surfaces are compact.

Smooth Manifolds (Algebraic Definition)

45

4.18. In proving Lemma 4.17, the following statement, called the “partition of unity lemma,” may be useful: If {Uα } is a locally finite open covering of the space M = |F|, then there exist functions fα ∈ F such that fα (x) = 0 if x ∈ M Uα and fα (x) ≡ 1. α

(A locally finite covering {Uα } is a covering such that for any x ∈ M there exists a neighborhood U ⊂ M of x that intersects only a ﬁnite number of sets Uα .) We suggest that the reader try to prove this statement ﬁrst in the particular case where the covering {Uα } is supplied with isomorphisms F U ∼ = C ∞ (Rn ) (or with diﬀeomorphisms Uα ∼ = Rn ). α

4.19. Suppose F is the algebra of smooth functions on the manifold M . Consider the action of a group on this smooth manifold, i.e., a family Γ of automorphisms γ : F → F such that (i) γ1 , γ2 ∈ Γ ⇒ γ1 ◦ γ2 ∈ Γ, (ii) γ ∈ Γ ⇒ γ −1 ∈ Γ. Suppose F Γ ⊂ F is the subalgebra of invariant functions for this action, i.e., F Γ = {f ∈ F | γ(f ) = f

for all γ ∈ Γ}.

Γ

4.20 Lemma. Suppose F is the subalgebra of invariant functions of a Γ (see Section 3.4) contains a function group action of Γ on M = |F|. If F all of whose level surfaces are compact, then the algebra F Γ is geometrical. The proof will be a repetition, word for word, of 3.16, if we note that the operations applied there preserve Γ-invariance. In particular, if the set |F Γ | is compact, then the algebra F Γ is always geometrical. 4.21. In order to learn to visualize the algebra F Γ of Γ-invariant functions on a manifold |F| = M , let us consider the orbit Oa = {|γ|(a) | γ ∈ Γ} for each point a ∈ M . Denote the set of all orbits by N . Elements of F Γ can be understood as functions on N . Indeed, if b = |γ|(a) ∈ Oa and f ∈ F is a Γ-invariant function, then f (b) = f (a). In other words, each “point” of the set N (i.e., each orbit) determines an R-point of the algebra F Γ , so that we have the natural map N → F Γ , Oa → (f → f (a)). This map will be bijective if the two following conditions hold: (i) Any homomorphism a : F Γ → R can be extended to a homomorphism a ˜ : F → R (surjectivity).

46

Chapter 4

(ii) If b ∈ / Oa , then there exists an f ∈ F Γ such that f (b) = f (a) (injectivity). These two conditions are satisﬁed, for example, if the group Γ is ﬁnite. We leave their veriﬁcation in the case of Examples 4.24 and 4.25 below as an exercise for the reader. 4.22 Proposition. The algebra F Γ of Γ-invariant functions on a smooth manifold M = |F| is complete if conditions (i), (ii) of 4.21 and the assumptions of Lemma 4.20 hold. Each real-valued function f : F Γ → R determines, by means of the pro Γ ˜ jection M → N = F , a → Oa , the function Γ f : M → R. If f coincides in a neighborhood of each point b ∈ Oa ∈ F with some function belonging to F Γ , then the Γ-invariant function f˜ coincides with some smooth function (from F) in a neighborhood of each point a ∈ |F|. Since F is complete, f˜ ∈ F, so that f ∈ F Γ . 4.23. It is clear that the set of orbits N of a group action of Γ on the manifold M = |F| is the quotient set of M by an equivalence relation (the one identifying points within each orbit). Definition. Assume that N coincides with F Γ and the algebra F Γ is smooth (or smooth with boundary); then we say that F Γ is the algebra of smooth functions on the quotient manifold of M by the group action of Γ. Following tradition, we shall often denote the quotient manifold by M/Γ, although this is sometimes a misleading notation (just as is the notation |F Γ |).

Z

2 1

-3

-2

-1

0

R2

Figure 4.3. Group action of Z on R2 producing the M¨ obius band.

Smooth Manifolds (Algebraic Definition)

47

4.24. Examples. I. In Examples 4.5 and 4.7, I, we actually dealt with quotient manifolds of the line R and of the plane R2 by the discrete cyclic group Z of isometries (see Figure 4.3). II. In Examples 4.7, II, and 4.8 we took the quotient manifolds with respect to the action of isometry groups with two generators. The reader may try to depict how the map |i| : R2 → |F| winds the plane R2 about the Klein bottle (see Section 4.7, III). III. Consider the action on F = C ∞ (Rn ) of the free Abelian group Γ with n generators γ1 , . . . , γn , where γi is the parallel translation by the unit vector along the ith coordinate, i.e., γi (f )(r1 , . . . , rn ) = f (r1 , . . . , ri−1 , ri + 1, ri+1 , . . . , rn ) for all f ∈ F, (r1 , . . . , rn ) ∈ Rn . It is easy to see that F Γ is the subalgebra of all functions in C ∞ (Rn ) that are 1-periodic with respect to each variable. Generalizing the arguments carried out in Sections 4.5 and 4.7, the reader will easily check that the quotient of Rn with respect to this action of Γ will be a smooth manifold. This manifold is known as the n-dimensional torus and is denoted by T n . (It is easy to see that T 1 = S 1 ; the most popular case, n = 2, was mentioned in the Introduction: It is the surface of the doughnut.) 4.25. Examples. I. Consider the R-algebra F = C ∞ (Rn+1 \ {0}). The multiplicative group R+ of positive real numbers acts on F by the automorphism hλ : hλ (f )(r1 , . . . , rn+1 ) = f (λr1 , . . . , λrn+1 ) for all f ∈ F, (r1 , . . . , rn+1 ) ∈ Rn+1 \ {0}. It turns out that the corresponding quotient set by this group action is a smooth manifold. Generalizing the argument of Section 4.13, show that the quotient manifold Rn+1 /R+ can be identiﬁed with the closed submanifold 2 = 1. This is the in Rn+1 whose points satisfy the equation r12 + · · · + rn+1 n n-dimensional sphere S . Prove that the ideal AS n = {f ∈ F | f (a) = 0 ∀a ∈ S n } is the principal ideal generated by the function 2 (r1 , . . . , rn+1 ) → r12 + · · · + rn+1 − 1.

II. In the previous example replace the group R+ by the multiplicative group R of all nonzero real numbers (with the action described by the same formula). Prove that the quotient of M = |F| by this action is a smooth manifold. In the case n = 2 check that it can be identiﬁed with the projective plane (cf. Section 4.8 (ii)). In the general case this manifold is known as the n-dimensional real projective space and is denoted by RP n . III. RP n can also be obtained from S n , n 1, by taking the quotient with respect to the group action of Z2 = R /R+ . Geometrically, this quotient space can be visualized as obtained by “gluing together” all pairs of diametrically opposed points on the sphere. Nevertheless, note that RP 1 ∼ = S1.

48

Chapter 4

4.26 Exercises. 1. Suppose Γ is the automorphism group of the algebra F = C ∞ R2 with one generator γ: γ(f )(r1 , r2 ) = f (−r1 , −r2 )

for all (r1 , r2 ) ∈ R2 , f ∈ F.

Show that the algebra F Γ of Γ-invariant functions is not smooth (nor smooth with boundary). 2. Suppose Γ is the rotation group of the plane about the origin: cos ϕ − sin ϕ |γ| = , sin ϕ cos ϕ γ(f )(r1 , r2 ) = f (r1 cos ϕ − r2 sin ϕ, r1 sin ϕ + r2 cos ϕ), for all (r1 , r2 ) ∈ R2 , f ∈ F = C ∞ R2 . Show that the space F Γ is the closed half-line, so that F Γ is an algebra of smooth functions on a manifold with boundary. Why does this algebra not coincide with the whole algebra of smooth functions on the half-line? 4.27. Remarks. (i) For a number of reasons, the deﬁnition of a group action on a manifold given in Section 4.19 is not a very fortunate one; it should be regarded as preliminary. We shall give a satisfactory deﬁnition only in Section 6.10. (ii) We also do not possess any meaningful criterion for the smoothness of the algebra F Γ of Γ-invariant functions simple enough to mention here. The reader, however, will proﬁt by proving the following: If for any a ∈ M there exists a neighborhood U ⊂ M , U a, such that for all nontrivial γ ∈ Γ, |γ|(U ) ∩ U = ∅, then the algebra F Γ is smooth. (iii) In the next six sections we are also anticipating a bit. Since these sections are relatively diﬃcult, they may be omitted at ﬁrst reading, which should then be continued from the beginning of Chapter 5. 4.28. If a physical system consists of independent parts, then it is natural to think of any state of the system as being a pair (a1 , a2 ), where a1 and a2 are the corresponding states of the ﬁrst and second part. If the states ai are understood as points of the manifold Mi (i = 1, 2), the algebras Fi of smooth functions on the Mi being correctly deﬁned and well known, it may be useful to deﬁne the manifold M of states of the entire system by using these algebras. Exercise. Let R-algebras F1 and F2 be geometrical. Show that their tensor product F1 ⊗R F2 is geometrical too. Definition. The smooth envelope F = F1 ⊗R F2 of the tensor product F1 ⊗R F2 of geometrical R-algebras (see Section 3.36) is said to be the algebra of smooth functions of the Cartesian product of the smooth manifolds M1 and M2 .

Smooth Manifolds (Algebraic Definition)

49

In the next section this terminology will be justiﬁed. 4.29 Proposition. If F is the algebra of smooth functions on the Cartesian product of the manifolds M1 and M2 , then |F| is indeed homeomorphic to the Cartesian product of the topological spaces M1 = |F1 | and M2 = |F2 |. (We suggest that the reader return to this proof after having read Chapter 6.) Our goal is to identify the space M1 × M2 with the set of R-points of the algebra F = F1 ⊗R F2 . To each pair (a1 , a2 ) ∈ M1 ×M2 let us assign the homomorphism a1 ⊗a2 : F1 ⊗ F2 → R R

(f1 ⊗ f2 → f1 (a1 ) · f2 (a2 )),

fi ∈ Fi (i = 1, 2).

Note that the algebra R is C ∞ -closed (Section 3.32). Hence by the deﬁnition of smooth envelope, a1 ⊗ a2 can be uniquely extended to the homomorphism a1 ⊗ a2 : F = F1 ⊗ F2 → R, R

and we have constructed the map π : M1 × M2 → |F|

(a1 , a2 ) → a1 ⊗ a2 .

The map π is injective: If a1 ⊗ a2 coincides with a1 ⊗ a2 , then, using the fact that these homomorphisms coincide on elements of the form f1 ⊗ 1 and 1 ⊗ f2 , we immediately conclude that fi (ai ) = fi (ai ) for all fi ∈ Fi ,

i = 1, 2.

Since the algebras Fi are geometrical, this implies ai = ai and hence the injectivity of π. Its surjectivity follows from elementary properties of tensor products. Thus π identiﬁes M1 × M2 and |F| as sets. It remains to prove that π identiﬁes the standard product topology in M1 × M2 with the R-algebra topology (Section 3.12) in |F|. Consider the basis of the topology in M1 × M2 consisting of the sets U1 × U2 with Ui = {a ∈ Mi | αi < fi (a) < βi },

fi ∈ Fi , αi , βi ∈ R, i = 1, 2.

Then the sets V1 = U1 × M2 = {(a1 , a2 ) | α1 < a1 ⊗ a2 (f1 ⊗ 1) < β1 } , V2 = M1 × U2 = {(a1 , a2 ) | α2 < a1 ⊗ a2 (1 ⊗ f2 ) < β2 } , are open in the topology induced from |F| by π in M1 × M2 . Therefore, V1 ∩ V2 = U1 × U2 is open as well. Conversely, it follows from the construction of smooth envelopes (Section 3.36) that in order to obtain a basis of the topology in |F|, we can take any subset of functions in F as long as the subalgebra generated by this subset has a smooth envelope coinciding with F. In the given case it suﬃces to take the subset of functions of the form f1 ⊗ f2 , fi ∈ Fi , i = 1, 2.

50

Chapter 4

Consider the basis open set V = {(a1 , a2 ) ∈ M1 × M2 = |F| | α < f1 (a1 )f2 (a2 ) < β}, α, β ∈ R, corresponding to such a function.

Figure 4.4. A basic open set in the Cartesian product.

The set of points on the plane R2 satisfying the inequality α < r1 r2 < β is open in the sense that, together with any of its points, it contains a rectangle {(r1 , r2 ) | α1 < r1 < β1 , α2 < r2 < β2 } (see Figure 4.4). Hence V is the union of sets of the form {(a1 , a2 ) ∈ M1 × M2 | α1 < f1 (a1 ) < β, α2 < f2 (a2 ) < β2 } and is therefore open in M1 × M2 .

4.30 Example–Lemma. The smooth envelope of the R-algebra C ∞ Rk ⊗R C ∞ Rl is isomorphic to the R-algebra C ∞ Rk+l . Consider the R-algebra homomorphism i : C ∞ Rk ⊗R C ∞ Rl → C ∞ Rk+l , i(f ⊗ g)(r1 , . . . , rk+l ) = f (r1 , . . . , rk ) · g(rk+1 , . . . , rk+l ). We shall show that i satisﬁes the deﬁnition of smooth envelope (Section 3.36). Indeed, suppose Φ : C ∞ Rk ⊗ C ∞ Rl → F R

Smooth Manifolds (Algebraic Definition)

51

is a homomorphism to a C ∞ -closed (Section 3.32) R-algebra F. A homomorphism Φ : C ∞ Rk+l → F is a prolongation of Φ (i.e., Φ = Φ ◦i) if and only if for all (r1 , . . . , rk ) ∈ Rk , (s1 , . . . , sl ) ∈ Rl , g ∈ C ∞ Rk+l , we have Φ (g) = g(Φ(r1 ⊗ 1), . . . , Φ(rk ⊗ 1), Φ(1 ⊗ s1 ), . . . , Φ(1 ⊗ sl )). (Here, on the right-hand side, we regard ri and sj as functions on Rk and Rl ; the right-hand side is then well deﬁned because F is C ∞-closed.) Since the last formula is well deﬁned for any g ∈ C ∞ Rk+l , the required homomorphism exists and is unique. By the uniqueness theorem in Section 3.37, the lemma follows. 4.31 Proposition. If F1 , F2 are smooth R-algebras, then so is the algebra F = F1 ⊗R F2 (see Section 4.28). Since the proof of this proposition repeats the one given in Section 4.29, we only indicate the main ideas, leaving the details to the industrious reader. Suppose ai ∈ Mi = |F| and let Ui ai be neighborhoods such that C ∞ (Rni ), i = 1, 2. We would like to establish the isomorphism Fi U ∼ = i ∼ F = C ∞ Rn1 +n2 . U1 ×U2

By 4.30, it suﬃces to show that F U ×U ∼ F1 U ⊗R F2 U . = 1 2 1 2 But as before, there is a homomorphism i : F1 ⊗R F2 → F U1

U2

U1 ×U2

,

which, as it turns out, satisﬁes the deﬁnition of a smooth envelope.

Exercises. 1. Let F1 be a smooth algebra with boundary (see Section 4.2) and let F2 be a smooth algebra. Mimicking the proof of the above proposition, show that the algebra F = F1 ⊗R F2 is smooth with boundary. 2. Does the previous assertion remain valid if F2 is also a smooth algebra with boundary? 4.32. Example. I. The cylinder in Example 4.5, II, is the Cartesian product of S 1 and R. II. The n-dimensional torus T n (see 4.24, III) is the Cartesian product of T n−1 and S 1 . In particular, T 2 = S 1 × S 1 . 4.33 Exercise. The reader suﬃciently versed in topology will proﬁt a great deal by proving that the M¨ obius band (Example 4.7, I) is not the Cartesian product of R and S 1 .

5 Charts and Atlases

5.1. In this chapter the intuitive idea of “introducing local coordinates” is elaborated into a formal mathematical deﬁnition of a diﬀerentiable manifold. The deﬁnition, of course, turns out to be equivalent to the algebraic one given in the previous chapter, as will be proved in Chapter 7. The coordinate approach is more traditional, and is certainly more appropriate for practical applications (when something must be computed). However, it is less suitable for developing the theory, since it requires tedious veriﬁcations of the fact that the notions and constructions introduced in the theory by means of coordinates are well deﬁned, i.e., independent of the speciﬁc choice of local coordinates. In the coordinate approach, a manifold structure on a set is deﬁned by a family of compatible charts constituting a smooth atlas, much in the same way as the geopolitical structure on the Earth’s surface is described by the charts of a geographical atlas. The words in italics above will be given mathematical deﬁnition in subsequent sections. 5.2. A chart (U, x) on the set M is a bijective map x : U → Rn of a subset U ⊂ M onto an open set x(U ) of Euclidean space Rn . The integer n > 0 is the dimension of the chart. Examples. chart (U, id) II. If T 2 (Figure 5.1),

I. If U is an open set in Rn , then the identity map deﬁnes a on the set Rn . is the conﬁguration space of the plane double pendulum then (U, s), where π π U = (ϕ, ψ) ∈ T 2 | − < ϕ, ψ < 4 4

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 5

53

54

Chapter 5

ϕ

ϕ

A

ψ ψ

Figure 5.1. Double pendulum.

and s denotes the map s

U (ϕ, ψ) −→ (sin ϕ, sin ψ) ∈ R2 , is a chart on T 2 . (Note that by assigning to each position of the pendulum the coordinates of its end point A ∈ R2 , we do not obtain a chart, since this assignment is not bijective.)

z

y

x

Figure 5.2. Saddle surface.

III. If S is the saddle surface z = 1 + x2 − y 2 , then the vertical projection pr

S (x, y, z) −→ (x, y) ∈ R2

Charts and Atlases

55

of a neighborhood U ⊂ S of the point (0, 0, 1) deﬁnes a chart (U, pr) on S (Figure 5.2). 5.3. Given a chart (U, x) and a point a ∈ U , note that x(a) is a point of Rn ; i.e., we have x(a) = (r1 , . . . , rn ) ∈ Rn ; the number ri is called the ith coordinate of a, and the corresponding function (sending each a ∈ U to its ith coordinate) is the ith coordinate function (in the chart (U, x)); it is denoted by xi : U → R. A chart is entirely determined by its coordinate functions; in the literature, the expression local coordinates is often used to mean “chart” in this sense. 5.4. Two charts (U, x), (V, y) on the same set M are called compatible if the change of coordinate map, i.e., y ◦ x−1 : x(U ∩ V ) → y(U ∩ V ), is a diﬀeomorphism of open subsets of the space Rn (see Figure 5.3) or if U ∩ V = ∅. The compatibility relation is reﬂexive, symmetric, and transitive because of appropriate properties of diﬀeomorphisms in Rn so that the family of all charts on a given set M splits into equivalence classes (sets of compatible charts).

y

x

V U Figure 5.3. Compatible charts.

Examples. I. All the charts (U, id), where U is an open subset of Rn and n is ﬁxed, are compatible. II. The chart (U, s) on the double pendulum T 2 described in 5.2, II, is not compatible with the chart (U, c) deﬁned by c

U (ϕ, ψ) −→ (sin ϕ, g(ψ)),

56

Chapter 5

where g(ψ) equals sin ψ for negative ψ and 1 − cos ψ for nonnegative ψ. This is because the change of coordinates map c ◦ s−1 fails to be smooth at the point (0, 0) ∈ R2 . 5.5. A family A of compatible charts xα : Uα → Rn on the set M (where n 0 is ﬁxed and α rangesover some index set J) is said to be an atlas on M if the Uα cover M , i.e., α∈J Uα = M . The integer n is the dimension of A. An atlas is maximal if it is not contained in any other atlas. Obviously, any atlas is contained in a unique maximal atlas, namely the one consisting of all charts compatible with any of the charts of the given atlas. Two atlases are said to be compatible if any chart of one of them is compatible with any chart of the other. The last condition is equivalent to the fact that the union of these atlases is also an atlas. Note that any two compatible atlases, together with their union, are contained in the same maximal atlas. This compatibility relation is reﬂexive, symmetric, and transitive (by the corresponding properties of diﬀeomorphisms in Rn ). Consequently, the family of all atlases on a given set M splits into equivalence classes, and any such a class contains a unique maximal atlas. Examples. I. Rn has an atlas consisting of a single chart: (Rn , id).

Figure 5.4. Stereographic projections.

II. The sphere S 2 has an atlas consisting of two charts (e.g., stereographic projections from the north and south poles; see Figure 5.4). III. The double pendulum (see Example 5.2, II) also has two-chart atlases (try to ﬁnd one).

Charts and Atlases

57

5.6. The reader is perhaps wondering why we are not deﬁning manifolds as sets supplied with an atlas. Well, we won’t be. Because if we did, this excessively general deﬁnition would put us under the obligation to bestow the noble title of manifold upon certain ungainly objects. Such as I. The discrete line. This is the set of points of R with the discrete atlas consisting of all charts of the form ({r}, v), where r ∈ R and v(r) = 0 ∈ R0 . II. The long line. This is the disjoint union R= Rα α

of copies Rα of R, indexed by an ordered uncountable set of indices α (we can take α to range over R itself). The set R has a natural order and a natural topology (induced from R and disjoint union). It also has a natural atlas A = {(Rα , idα ) : α ∈ R}, where idα : Rα → R1 is the identiﬁcation of each copy Rα with the original prototype R1 = R. III. The line with a double point. This is the line R to which a point θ is added, while the atlas consists of two charts (U, x) and (V, y), where U =R, V ={θ} ∪ R {0},

x = id, y(θ) = 0, y(r) = r, r ∈ R {0}.

In other words, the line with a double point can be obtained if two copies of the line R are identiﬁed at all points with the same coordinates except for zero. 5.7. The following deﬁnitions are needed to exclude pathological atlases (of the types described in Section 5.6) from our considerations. We say that an atlas A on M satisﬁes the countability condition if it consists of a ﬁnite or a countable number of charts or if all its charts are compatible with those of such an atlas. An atlas A on M satisﬁes the Hausdorﬀ condition if for any two points a, b ∈ M there exist nonintersecting charts (U, x), (V, y) containing these points (a ∈ U , b ∈ V , U ∩ V = ∅) and compatible with the charts of A. Clearly, the discrete and long lines (see Examples 5.6, I, and II) do not satisfy the countability condition, while the line with a double point, 5.6, III, fails to meet the Hausdorﬀ condition. 5.8. Coordinate definition of manifolds. A set supplied with a maximal atlas Amax = {(Uα , xα )}, xα : Uα → Rn , n 0, satisfying the countability and Hausdorﬀ conditions is called an n-dimensional diﬀerentiable (or smooth) manifold. It will be proved in Chapter 7 that this deﬁnition is equivalent to the algebraic one given in Section 4.1. In order to determine a speciﬁc manifold (M, Amax ), we shall often indicate some smaller atlas A ⊂ Amax , since Amax is uniquely determined by any of its subatlases (see 5.5). In that case we denote our manifold

58

Chapter 5

by (M, A) and say that A is a smooth atlas on M . Note that the adjective “smooth” implicitly includes the compatibility, Hausdorﬀ, and countability conditions. 5.9. We now show that any smooth atlas A = {(Uα , xα )} on the manifold M determines a topological structure on the set M , carried over from the Euclidean topology in the sets xα (Uα ) ⊂ Rn by the maps x−1 α . To be more precise, a base of open sets in the space M is constituted by all the sets x−1 α (Bβ ), where the Bβ ’s are all the open Euclidean balls contained in all the xα (Uα ). It follows immediately from the deﬁnitions that M then becomes a Hausdorﬀ topological space with countable base. When the set M is supplied with an atlas, this topological structure is always understood. For example, when we say that the manifold M is compact or connected, we mean that it is a compact (connected) topological space with respect to the topology described above. It is easy to see that any chart (U, x) is a homeomorphism of an open subset U ⊂ M (in this topology) onto an open subset x(U ) ⊂ Rn . 5.10. Examples of manifolds from geometry. I. The sphere S n = {x : |x| = 1} ⊂ Rn+1 has a two-chart atlas given by stereographic projections (similar to the two-dimensional case; see 5.5, II). The sphere S n is the simplest compact connected n-dimensional manifold. II. The hyperboloid x2 + y 2 − z 2 = 1 in R3 has a simple four-chart atlas (U± , pzy ), (V± , pzx ), where U± = {(x, y, z) | x = ± 1 + z 2 − y 2 , ±x > 0}, V± = {(x, y, z) | y = ± 1 + z 2 − x2 , ±y > 0}, and the maps pzy , pzx are the projections on the corresponding planes. The corresponding two-dimensional manifold is connected but not compact. III. The projective space RP n is the set of all straight lines passing through the origin O of Rn+1 . For each such line l consider the following chart (Ul , pl ). The set Ul consists of all lines forming an angle of less than (say) 30◦ with l (Figure 5.5). To deﬁne pl , choose a basis in the n-plane l⊥ O perpendicular to l and ﬁx one of the half-spaces Rn+1 \ l⊥ ; to every line l ∈ Ul let the map pl assign the coordinates in l⊥ of the projection on l⊥ of the unit vector pointing into the chosen half-space and determining l . The set of all such charts (Ul , pl ) constitutes a smooth atlas, endowing RP n with the structure of a compact connected n-dimensional manifold. IV. The Grassmann manifold Gn,m is the set of all m-dimensional planes in Rn passing through the origin O. To construct a chart, let us choose in Rn a Cartesian system (x1 , x2 , . . . , xn ) and take for U the set of all

Charts and Atlases

59

l

l

⊥

Figure 5.5. Construction of a chart on RP n .

m-dimensional planes given in these coordinates by the system of equations xm+i =

m

aij xj ,

i = 1, . . . , n − m.

j=1

Let also the map p : U → Rm(n−m) take such a plane to the set of coefﬁcients aij appearing in the system above. Choosing diﬀerent Cartesian systems, one can construct diﬀerent charts covering together the entire space Gn,m . However, to cover it, it suﬃces to use only one Cartesian system and change the order of coordinates in it. Compatibility of these charts follows from the smooth dependency of solutions of a linear system on the system’s coeﬃcients. Thus we obtain an m(n − m)-dimensional manifold that generalizes the previous example, namely Gn,1 = RP n−1 . One can consider planes not only in Rn but also in any ﬁnite-dimensional vector space V . The manifold obtained in this case is denoted by GV,m . Exercise. How many connected charts do you need to obtain an atlas for the Klein bottle (cf. 4.7, III)? Prove that two is enough. 5.11. Examples of manifolds from algebra. I. The general linear group GL(n) of all linear isomorphisms of Rn has a one-chart atlas of dimension n2 obtained by assigning to each g ∈ GL(n) the n2 entries of its matrix written column by column in the form of a single column vector with n2 components. The corresponding manifold is not compact and not connected. II. The special orthogonal group SO(n) of all positive orthogonal matrices possesses a smooth atlas of (n(n−1)/2)-dimensional charts. Its construction is left to the reader, who might proﬁt by referring to Example 5.10, III.

60

Chapter 5

5.12. Examples of manifolds from mechanics. I. The conﬁguration space of the double pendulum (see 5.2, II, and 5.4, II), as the reader must have guessed by now, is the two-dimensional torus T 2 = S 1 × S 1 . II. The conﬁguration space of a thin uniform disk whose center is ﬁxed by a hinge (allowing it be inclined at all angles and directions in threespace) possesses a natural two-dimensional atlas. The reader is urged to ﬁnd such an atlas and compare it with that of the projective plane RP 2 . He/she will also appreciate that if one side of the disk is painted, then the corresponding atlas will be that of S 2 . III. The conﬁguration space (recall Section 1.1) of a solid freely rotating in the space about a ﬁxed point is a three-dimensional compact connected manifold. The reader is asked to ﬁnd an atlas for it and compare it with RP 3 and SO(3). 5.13 Exercise. Consider the following manifolds (already discussed informally in Chapter 1): 1. The projective space RP 3 . 2. The sphere S 3 with antipodal points identiﬁed. 3. The disk D3 with antipodal points of its boundary ∂D3 = S 2 identiﬁed. 4. The special orthogonal group SO(3). 5. The conﬁguration space of a solid freely rotating about a ﬁxed point in R3 . Show that they are all diﬀeomorphic by 1. Constructing atlases and diﬀeomorphisms between the charts. 2. Constructing isomorphisms of the corresponding smooth R-algebras. 5.14. Previously (see Section 4.2), we introduced the notion of manifold with boundary algebraically. Now we give the corresponding coordinate deﬁnition. This deﬁnition is just the same as that of an ordinary manifold (see Section 5.8), except that the notion of chart must be modiﬁed (by substituting RnH for Rn ). Namely, a chart with boundary (U, x) on the set M is a bijective map x : U → RnH of a subset U ⊂ M onto an open subset x(U ) of the Euclidean half-space RnH = {(r1 , . . . , rn ) ∈ Rn | rn 0}. Note that a chart in Rn is a particular case of a chart in RnH , since x(U ) may not intersect the “boundary” {(r1 , . . . , rn ) ∈ Rn | rn = 0} of the half-space.

Charts and Atlases

61

All further deﬁnitions (those of dimension, compatibility, atlases, etc.) remain the same, except that RnH must be substituted for Rn in the appropriate places. Repeating these deﬁnitions in this modiﬁed form, we obtain the coordinate deﬁnition of a manifold with boundary. The industrious reader will gain by actually carrying out these repetitions in detail; this is a good way to check that he/she has in fact mastered the main Deﬁnition 5.8. 5.15. If A = (Uα , xα ), α ∈ I, consists of charts with boundary, the set

∂M = m ∈ M | ∃α ∈ I, m = x−1 α (r1 , . . . , rn ), rn = 0 of points mapped by coordinate maps on the boundary (n − 1)-dimensional plane {rn = 0} of the half-space RnH is said to be the boundary of M . When ∂M is empty, we recover the deﬁnition of ordinary manifold (the open half-space {rn > 0} being homeomorphic to Rn ). Sometimes in the literature the term “manifold” is deﬁned so as to include manifolds with boundary; in that case the expression closed manifold is used to mean manifold (with empty boundary). Proposition. The boundary ∂M of an n-dimensional manifold with boundary M has a natural (n − 1)-dimensional manifold structure. Hint of the proof. Take the intersection of the charts of M with the boundary (n − 1)-plane of the half-space Rn−1 = {(r1 , . . . , rn ) ∈ RnH | rn = 0} to get an atlas on ∂M .

5.16. Examples of manifolds with boundary. I. The n-dimensional disk Dn = {x | x ≤ 1} ⊂ Rn is a n-dimensional manifold with boundary ∂Dn = S n−1 . II. If M is an n-dimensional manifold (without boundary) deﬁned by its atlas A, a manifold with boundary can be obtained from M by “removing an open disk from it.” This means that we take any chart (U, x) ∈ A, choose an open n-dimensional disk V in x(U ) ⊂ Rn , and consider the set M \ x−1 (V ), which has an obvious manifold-with-boundary structure; M \ x−1 (V ) is sometimes called a punctured manifold. III. Figure 5.6 presents the beginning of a complete list of all twodimensional manifolds (with boundary) whose boundary is the circle S 1 . They are called the 2-disk, the punctured torus, the punctured orientable surface of genus 2, . . . , the punctured orientable surface of genus k, . . . (upper row), the M¨ obius band, the punctured Klein bottle,. . . , the punctured non-orientable surface of genus k, . . . . The term “orientable,” which we do not discuss in the general case here, in the two-dimensional case means “does not contain a M¨ obius band.”

62

Chapter 5

Figure 5.6. Two-dimensional manifolds with S 1 as the boundary.

5.17. In the algebraic study of smooth manifolds, the fundamental concept was the R-algebra of smooth functions F. This R-algebra can also be deﬁned for a manifold M given by an atlas A. Definitions. A function f : M → R on the manifold M with smooth atlas A is called smooth if for any chart (U, x) ∈ A the function f ◦x−1 : x(U ) → R deﬁned on the open set x(U ) ⊂ Rn is smooth (i.e., f ◦ x−1 ∈ C ∞ (x(U ))). The set of all smooth functions on M is denoted by C ∞ (M ). This set has an obvious R-algebra structure, and will temporarily be called the R-algebra of smooth functions on M with respect to the atlas A. It is easy to establish that C ∞ (M ) is the same for any atlas A compatible with A. Moreover, we shall see that C ∞ (M ) is the same R-algebra as the one in the algebraic approach (C ∞ (M ) = F), but this will be proved only in Chapter 7). Exercise. 1. Describe the smooth function algebra (in the sense of the above deﬁnition) of the conﬁguration space for the double pendulum (see 5.12). 2. Same question for the case in which the ϕ-rod is shorter then the ψ-rod, so that the latter is blocked in its rotation when the point A hits the ϕ-axle (see Figure 5.1). Identify the corresponding smooth manifold.

Charts and Atlases

63

5.18. In the equivalence proof carried out in Chapter 7, we shall need the following proposition. Proposition. If M is a manifold A its smooth atlas, and C ∞ (M ) the R-algebra of smooth functions on M (with respect to A), then there exists a function f ∈ C ∞ (M ) all of whose level surfaces (i.e. the sets f −1 (λ), λ ∈ R) are compact subsets of M . This proposition generalizes Proposition 2.7 and can be proved by using the latter and the partition of unity lemma, stated in the appropriate form for manifolds with atlases (compare with Section 4.18).

6 Smooth Maps

6.1. Suppose F1 is the algebra of smooth functions on a manifold M1 , and F2 is the one on another manifold M2 . The map f : M1 → M2 is called smooth if f = |ϕ|, where ϕ : F2 → F1 is an R-algebra homomorphism. Recall that Mi = |Fi |, i = 1, 2, are the dual spaces to Fi (see Section 3.8), i.e., consist of all R-algebra homomorphisms x : Fi → R, i = 1, 2. Recall also that |ϕ| : |F1 | → |F2 | is the dual map deﬁned in 3.19 as |ϕ| : x → x ◦ ϕ and that all homomorphisms (see Section 3.1) are unital: ϕ(1) = 1. Exercise. Prove that ϕ is injective whenever f is surjective. Construct a counterexample to the converse statement (if you do not succeed, try again after reading Example 6.5 below). 6.2. Example. Suppose F is the algebra of smooth functions on the manifold M , and Γ is a group acting on M . The map p = |i| : M → M/Γ Γ

dual to the inclusion i : F → F of the algebra of Γ-invariant functions (see 4.19) into F is of course smooth. To be speciﬁc, consider the group Γ = Z acting on R by identifying points r1 and r2 whenever r1 − r2 ∈ Z. Denote by S 1 the set of equivalence classes, and let F be, as in 4.5, the algebra of smooth 1-periodic functions on the line R. The natural projection p : R → S 1 is a smooth map, since it coincides with |i|, where i is the inclusion F ⊂ C ∞ (R). We say that the map p winds the line R around the circle S 1 . 6.3. Example. Let F be the algebra of smooth functions on the manifold M , and N ⊂ M a smooth submanifold (Deﬁnition 4.11) of M . In this case © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 6

65

66

Chapter 6

the inclusion N → M is a smooth map, since it coincides with |ρ|, where ρ : F → FN = F N is the restriction homomorphism. To be speciﬁc, suppose that S 1 and F are the same as in Example 6.2 and let FC be the algebra of all real-valued functions of a complex variable z = x + iy smoothly depending on the variables x and y. Consider the inclusion i : S 1 → C [r] −→ e2πir , where [r] ∈ S 1 is the equivalence class of the point r ∈ R. Further, deﬁne the homomorphism β : FC → F (β(f )(r) = f e2πir , f ∈ FC , r ∈ R). The inclusion i coincides with |β| and is, therefore, a smooth map of S 1 into R2 = C. Notice that we have already met this inclusion in another coordinate representation; see Example 4.13. 6.4. Examples. Suppose F is the algebra of smooth functions on the open M¨ obius band (4.7, I) and F(S 1 ) the algebra of 1-periodic smooth functions on the line (“the circle” 3.18).

b=2 |β|

b=1

|α|

Figure 6.1. Maps from S 1 to the M¨ obius band.

I. Consider the homomorphisms α, β : F → F(S 1 )

(α(f )(r) = f (r, 0), β(f )(r) = f (2r, b)),

where f ∈ F, r ∈ R, and b = 0 is any real number.

Smooth Maps

67

Exercise. Why does the formula γ(f )(r) = f (r, b) not deﬁne a homomorphism γ : F → F(S 1 )? The smooth maps |α| and |β| are shown in Figure 6.1. Notice that the image of the map |β| is “twice as long” as that of |α|. II. There is a remarkable smooth map of the M¨obius band on the circle, namely the map g = |ξ|, where ξ : F S 1 → F (ξ(f )(r1 , r2 ) = f (r1 )). Note that α◦ξ = idF (S 1 ) . The map g = |ξ| may be visualized as “collapsing” the M¨ obius band to its central circle (Figure 6.2).

Figure 6.2. “Collapsing” the M¨ obius band.

6.5. Example Choose an irrational number λ ∈ R and consider the map f : R1 → R2 , Then f = |ϕ|, where

r → (r, λr).

ϕ : C ∞ R2 → C ∞ R1 (ϕ(g)(r) = g(r, λr)) for all g ∈ C ∞ R2 , r ∈ R. Denote by ϕ the restriction of the homomorphism ϕ to the subalgebra of doubly periodic functions (see Example 4.24, III, with n = 2). The image of the smooth map |ϕ| : R1 → T 2 is everywhere dense in the torus T 2 and, therefore, the homomorphism ϕ is injective. This example is interesting, since it shows that an algebra of functions “in several variables” may be isomorphic to a subalgebra of C ∞ R1 . When the number λ is taken to be the rational, the image of |ϕ| is compact. A particular case is shown in Figure 6.3. Try to guess what value of λ was taken there.

68

Chapter 6

Figure 6.3. A map of R to the torus.

6.6. Now that smooth maps of manifolds have been introduced, manifolds no longer appear as unrelated, separate objects; they have been brought together into something uniﬁed, called a category. Other examples of categories are groups and their homomorphisms, topological spaces and continuous maps, R-algebras and R-algebra homomorphisms, and linear spaces and linear operators. As we pointed out in Section 3.38, we shall not give any formal deﬁnitions from abstract category theory, but will often “think categorically.” In particular, let us point out two fundamental properties of the category of smooth manifolds and maps: (i) if a = |α| : M1 → M2 and b = |β| : M2 → M3 are smooth maps corresponding to the R-algebra homomorphisms α : F2 → F1 and β : F3 → F2 , then b ◦ a : M1 → M3 is a smooth map (since it corresponds to the composition α ◦ β, in inverse order, of the homomorphisms α and β); (ii) the identity map id : M1 = M → M2 = M is smooth, since it corresponds to the identity homomorphism id : F2 = F → F1 = F. The other categories mentioned above possess similar properties.

Smooth Maps

69

Suppose we are given a collection of maps possessing properties (i) and (ii). Then a typical “categorical trick” is to “inverse all arrows”; i.e., whenever a map A → B belongs to our collection, assume that there is a map B → A in a new, dual collection of maps, in which composition of maps is written in inverse order. Then we obtain the dual category, also satisfying (i), (ii). We have, in fact, been using that construction in passing from homomorphisms of smooth R-algebras to smooth maps of manifolds. The contents of Chapters 4 and 6 may be summarized as follows: A smooth manifold is a smooth R-algebra, understood as an object of the dual category. 6.7. Suppose M1 and M2 are manifolds. The smooth map a : M1 → M2 is said to be a diﬀeomorphism if there exists a smooth map b : M2 → M1 such that b ◦ a = idM1 , a ◦ b = idM2 . The manifolds M1 and M2 are called diﬀeomorphic if there exists a diﬀeomorphism of one onto the other. Note that two manifolds are diﬀeomorphic if and only if their algebras of smooth functions are isomorphic. The relation of being diﬀeomorphic is an equivalence relation and will be denoted by ∼ =. In Chapter 4, when we spoke of two manifolds being “the same” or “identical,” we actually meant that they were diﬀeomorphic; indeed, from the point of view of the theory, diﬀeomorphic manifolds are the same manifold presented in diﬀerent guises. 6.8. Examples. I. The argument in Example 4.13 can be understood as a proof of the fact that the two methods for constructing the circle (as the quotient space of R1 and as a submanifold of R2 ) result in diﬀeomorphic manifolds. II. A linear operator A : Rn → Rn will be a diﬀeomorphism if and only if det A = 0, i.e., if A is bijective. III. Suppose F1 and F2 are the algebras of smooth functions on manifolds M1 and M2 . In general, the bijectivity of the smooth map |α| : M1 → M2 (where α : F2 → F1 is an R-algebra homomorphism) is not suﬃcient for this map to be a diﬀeomorphism. As an example, we can take the map ϕ : R → R,

r → r3 .

In order to establish that various speciﬁc bijective smooth maps are in fact diﬀeomorphisms, one often uses the implicit function theorem 6.22. We shall not dwell on this here. 6.9 Exercises. 1. Prove that the boundary of the closed M¨ obius band (4.7, II) is diﬀeomorphic to the circle S 1 . 2. Following Example 4.13 and I above, construct a diﬀeomorphism between the two models of the sphere S n described in 4.25, I. 6.10. We now return to the topics of Chapter 4 in order to give, as promised, more satisfactory deﬁnitions of group action, quotient manifolds,

70

Chapter 6

and Cartesian products. These deﬁnitions will be “categorical” in character: They will be based on smooth maps and diagrams. Suppose F is the algebra of smooth functions on the manifold M , and Γ is some group of automorphisms of this R-algebra. The smooth map a : M → N of M into the manifold N is called Γ-invariant if a ◦ |γ| = a for all γ ∈ Γ. Obviously, the quotient map q of M on the quotient manifold M/Γ is Γ-invariant. (Of course, this is true, provided that M/Γ is a smooth manifold, i.e., if the algebra F Γ of invariant functions is a smooth R-algebra; see Deﬁnition 4.1 and Section 4.23.) It turns out that the quotient map is the “universal” Γ-invariant smooth map. Proposition. Let F(N ) be the algebra of smooth functions on the smooth manifold N and a : M → N be any Γ-invariant smooth map with respect to an action of Γ on M such that M/Γ is a smooth manifold. Then there exists a unique map b : M/Γ → N for which the following diagram is commutative: a

M

N

q

b

M/Γ We must prove the existence and uniqueness of an R-algebra homomorphism β : F(N ) → F Γ for which the diagram α

F i

F(N ) β

FΓ where a = |α| is commutative. Clearly, there is no more than one such β. It exists iﬀ α = α(F(N )) consists of Γ-invariant elements. But for a Γ-invariant map a, this is always the case: For all γ ∈ Γ, f ∈ F(N ) we have γ(α(f )) = α(f ) ◦ |γ| = f ◦ (a ◦ |γ|) = f ◦ a = α(f ).

6.11. Remark. The universal property characterizing the quotient map M → M/Γ determines the quotient manifold M/Γ uniquely up to diﬀeomorphism. The proof of this statement can be copied over from the uniqueness proof of smooth envelopes (Proposition 3.37), and we suggest that the reader carry it out. The underlying general principle for proofs of this type, which we do not wish to formalize here, is that “any universal property determines an object uniquely.” 6.12. Now we return to Cartesian products (Deﬁnition 4.28). Let Fl be the algebra of smooth functions on the manifold Ml , l = 1, 2. The projection

Smooth Maps

71

maps pl : M 1 × M 2 → M l ,

(a1 , a2 ) → al ,

l = 1, 2,

are smooth, since pl = |πl |, where the R-algebra homomorphism π is the composition il

σ

Fl → F1 ⊗R F2 → F1 ⊗R F2 = F(M1 × M2 ); here σ is the smooth envelope homomorphism (3.36), and f ⊗ 1, l = 1, il (f ) = for all f ∈ Fl . 1 ⊗ f, l = 2, The pair of projection maps (p1 , p2 ) possesses the following universal property. Proposition. For any smooth manifold N and any pair of smooth maps fl : N → Ml , l = 1, 2, there exists a unique smooth map f : N → M1 × M2 completing the commutative diagram N f1

f2

M1

M2

f p2

p1

M1 × M2 Denote the algebra of smooth functions on N by F(N ). Assume that fl = |ϕl |, where ϕl : Fl → F(N ), l = 1, 2, are the dual R-algebra homomorphisms. Our proposition will be proved if we establish the existence and uniqueness of the following diagram: F(N ) ϕ2

ϕ1 ϕ

F1

ψ

F2

π2

i1 π1

i2

F(M1 × M2 )

σ

F1 ⊗R F2

By the universal property of tensor products, there exists a unique homomorphism ψ shown in the diagram. Since the R-algebra F(N ) is smooth, while the smooth envelope homomorphism has the universal property stated in Proposition 3.37, the homomorphism ϕ is also well deﬁned and unique.

72

Chapter 6

6.13. Remark. Proposition 6.12 may be used to construct smooth maps from a third manifold M3 to the product M1 × M2 of two given ones. For example, a smooth map of a manifold N to S 1 × S 1 is a pair of smooth maps from N to S 1 , i.e., a pair of smooth functions on N deﬁned modulo 1. 6.14. The remainder of this section is a discussion of the notions and constructions developed in 6.1–6.13 carried out in the “coordinate language” introduced in Chapter 5.

M U

y f

x

V

N Figure 6.4. Smooth map in the “coordinate language.”

Suppose (M, A) and (N, B) are manifolds with smooth atlases A and B (see Section 5.8). The map f : M → N is called smooth (or diﬀerentiable) at the point a ∈ M if for some (hence for all) pairs of charts (U, x) and (V, y) compatible with the atlases A and B and covering the points a and f (a), respectively, the map y ◦f ◦x−1 , deﬁned in the neighborhood x(f −1 (V )∩U ) of the point x(a) ∈ Rn , is an inﬁnitely diﬀerentiable map of domains in Euclidean space (see Figure 6.4). The map f : M → N is called smooth if it is smooth at each point. In the coordinate language, a smooth bijective map f : M → N is called a diﬀeomorphism if the map f −1 is also smooth. 6.15. When working with speciﬁc maps of manifolds, we ordinarily use a coordinate representation. Actually, this means that we use the abovementioned map y ◦ f ◦ x−1 , which, being a map of Euclidean spaces, is

Smooth Maps

represented by certain functions yi = fi (x1 , . . . , xm ),

73

i = 1, . . . , n,

where m = dim M , n = dim N . For a smooth map (at a point), all the functions fi , i = 1, . . . , n, will also be smooth (respectively, at a point). It is easy to see that if a map is described by smooth functions in a family of pairs of charts compatible with the corresponding atlases (and the charts cover the entire manifold M ), then this map will be described by smooth functions for any pair of charts compatible with the same atlases (of course, charts for which the composition y ◦ f ◦ x−1 is undeﬁned are not taken into account). The equivalence of these coordinate deﬁnitions and the corresponding algebraic ones will be established in Chapter 7. 6.16. Examples. I. The map a : T 2 → R2 that assigns to each position of the double pendulum (Example 5.2, III) its end point A (see Figure 5.1) is smooth. Choose some ﬁxed position of the double pendulum; then all suﬃciently close positions of the pendulum are characterized by two angles x, y, so that we have a chart Φ U (position) → (x, y) that is compatible with the standard atlas of the torus. The manifold R2 can be covered by a single chart id : R2 → R2 . We can say that in the chosen local coordinates the map a is described by the formulas r1 = R cos x + r cos y, r2 = R sin x + r sin y. The rigorous meaning of the words in italics is that the formulas actually describe the map (id)−1 ◦ a ◦ Φ−1 ; hence it follows from Deﬁnition 6.14 that a is smooth. II. Choose a ﬁxed unit vector v ∈ S n−1 in Rn and consider the map fv : SO(n) → S n−1 ,

A → A(v).

Verify that fv is a smooth map by using the atlases described in Examples 5.10, I, and 5.11, II. We can also consider the map ϕ : SO(n) × S n−1 → S n−1 , (A, v) → A(v), and attempt to prove its smoothness, working with an atlas on the product SO(n)×S n−1 . The reader will be wise not to take this attempt too seriously, but to prove smoothness of ϕ eventually by using the algebraic deﬁnitions. 6.17. More examples I. Let us consider four-dimensional Euclidean space R4 as the algebra of quaternions H, denote by V ∼ = R3 the subspace of purely imaginary quaternions r1 i + r2 j + r3 k, and introduce the map

74

Chapter 6

(q, v) → qvq −1 .

(H {0}) × V → V,

It is easy to check that for each nonzero quaternion q, the matrix of the linear operator v → qvq −1 in the coordinates r1 , r2 , r3 is orthogonal, so that this is a map H {0} → SO(3). Two quaternions q1 and q2 determine the same transformation of the space V if and only if q1 = λq2 for some λ ∈ R {0}, so that we have obtained a bijective smooth map RP 3 → SO(3). Recall that RP 3 is precisely the quotient manifold of the punctured space R4 {0} = H {0} by the group R of all homotheties with center at the origin (see Example 4.25, II) and that the projection H {0} → RP 3 has the universal property (see Proposition 6.10). It is not useless to try to establish the smoothness of this map by using the atlases 5.10, III, and 5.11, II. It is more diﬃcult to prove the fact that it is a diﬀeomorphism (as mentioned in Section 1.6). II. The composition S 3 → R4 {0} → SO(3) → S 2 , where the ﬁrst arrow is the inclusion and the others are deﬁned in Examples 6.16, II, and 6.17, I, is called the Hopf map. Notice that the composition of the ﬁrst two arrows may be represented in the form ∼ =

S 3 → RP 3 → SO(3), where the map S 3 → RP 3 was described in Example 4.25, III, as the quotient map S 3 → S 3 /Z2 ∼ = RP 3 . Exercise. Try to show that the inverse image of any point of S 2 under the Hopf map h : S 3 → S 2 is a closed submanifold of S 3 , diﬀeomorphic to S 1 . 6.18. Example (of a smooth map of a manifold with boundary). Suppose D3 is the 3-dimensional closed disk with center at the origin O ∈ R3 and of radius π. To each point a ∈ D3 associate the rotation of R3 about the line joining a to the origin by the angle α = a . Thus we obtain the map g : D3 → SO(3). Clearly, diametrically opposed points on the boundary ∂D3 = S 2 of 3 D determine the same rotation. Thus the manifold SO(3) ∼ = RP 3 of orthogonal transformations of R3 can be represented as the disk D3 whose diametrically opposed boundary points have been glued together. The map g : D3 → SO(3) is a smooth surjective map of a manifold with boundary onto a manifold (without boundary). Exercise. Show that the image of the boundary S 2 = ∂D3 under this map is a smooth closed submanifold in SO(3) diﬀeomorphic to the manifold RP 2 .

Smooth Maps

75

6.19 Exercises. 1. Write out the formulas for the orthogonal projection of the unit sphere S n ⊂ Rn+1 onto the hyperplane Rn ⊂ Rn+1 in the charts indicated in 5.10 (i) and verify that this projection is a smooth map. 2. Prove that SO(4) ∼ = S 3 × SO(3). Hint: SO(3) may be understood as the set of orthonormed pairs {u, v} in R3 , SO(4) as the set of orthonormed triples {u, v, w} in R4 ; let R4 = H and R3 = V ⊂ H as in Example 6.17, I; investigate the map S 3 × SO(3) → SO(4),

(u, {v, w}) → {u, uv, uw}.

6.20. The collection of all smooth maps (in the sense of the coordinate Deﬁnition 6.13) also possesses the two properties concerning composition and identity maps mentioned in Section 6.6 (which is not surprising, since the coordinate deﬁnition is equivalent to the algebraic one; see Section 6.1). The class of all smooth manifolds in the sense of Section 5.8 together with the family of all smooth (Section 6.14) maps constitutes the category of coordinate manifolds. In what follows, we shall need some classical results of multidimensional local calculus, i.e., the theorems on implicit and inverse functions. We formulate them here in a form convenient for the subsequent exposition, without any proofs. The latter may be found in any advanced calculus course. 6.21 Inverse Function Theorem. Let a smooth map f = (f1 , . . . , fn ) : Rn → Rn possess a nondegenerate Jacobi matrix in a neighborhood of the origin 0 ∈ Rn : ∂fi (0) = 0. det f (0) = det ∂xj Then there exist open sets U 0 and V f (0) such that the map def ϕ = f U : U → V possesses a smooth inverse ϕ−1 : V → U . The Jacobi matrix of the latter at any point y ∈ V can be computed by the formula −1 (ϕ−1 ) (y) = ϕ (ϕ−1 (y)) . 6.22 Implicit Function Theorem. Let f = (f1 , . . . , fn+m ) : Rn × Rm → Rm

76

Chapter 6

be a smooth map with f (a, b) = 0 ∈ Rm possessing a nondegenerate (m×m) matrix of partial derivatives with respect to the variables xn+1 , . . . , xn+m : ∂fi det (a) = 0. ∂xn+j 1i,jm Then there exist open sets U and V , a ∈ U ⊂ Rn , b ∈ V ⊂ Rm , and a smooth function g : U → V such that f (x, g(x)) = 0 for all x ∈ U . Finally, we shall need another classical theorem (the theorem on the linearization of a smooth map), which is an amalgam of the previous two results. 6.23 Theorem. Let f : Rn → Rm , m n, be a smooth map such that f (0) = 0 and the rank of the matrix ∂fi (0) M= ∂xj 1im, 1jn

is equal to m. Then there exist neighborhoods U, V ⊂ Rn of 0 and a diﬀeomorphism ϕ : U → V such that (f ◦ ϕ)(x1 , . . . , xn ) = (xn−m+1 , . . . , xn ).

6.24. Remark. The three theorems above are “local”: their assumptions are formulated in Euclidean spaces, while the conclusions are valid for neighborhoods (of Euclidean spaces). Nevertheless, they may be applied to the “global” case of manifolds by considering separate coordinate neighborhoods. By uniqueness of the inverse (and implicit) functions in the corresponding neighborhood, the functions constructed in these theorems coincide on the common part of the two neighborhoods. Therefore, we can “glue” them together over the entire manifold. Thus, for example, the conclusion of the implicit function theorem may be formulated as follows: If f : M → N is a smooth map of manifolds, m = dim M > n = dim N , and the Jacobian of f has rank n at any point, then f −1 (z) ⊂ M is a submanifold for any z ∈ N . 6.25 Exercises. A plane hinge mechanism (see Section 1.14) is called generic if the conﬁguration with all the hinges positioned along the same straight line is impossible (i.e., there is no linear relation with coeﬃcients ±1 among the lengths of the rods). 1. Prove that the conﬁguration space of a generic hinge mechanism is a smooth manifold. 2. Show that the conﬁguration space of a generic pentagon (Section 1.14) is diﬀeomorphic either to the sphere with no more than 4 handles, or to the disjoint union of two tori, or to the disjoint union of two spheres.

7 Equivalence of Coordinate and Algebraic Definitions

7.1. The aim of this chapter is to prove that the two deﬁnitions of smooth manifold (4.1 and 5.8) and of smooth maps (Sections 6.1 and 6.13) yield the same concepts, thus showing that the coordinate approach and the algebraic one are equivalent. The equivalence of the two deﬁnitions of smooth manifold will be stated in the form of two theorems (7.2 and 7.7). The equivalence of the deﬁnitions of smooth maps is Theorem 7.16. 7.2 Theorem. Suppose F = C ∞ (M ) is the algebra of smooth functions on a manifold M defined by its smooth atlas A. Then F is a smooth R-algebra (in the sense of Section 4.1), and the map θ : M → |F|, θ(p) (f ) = f (p), is a homeomorphism. The proof will be in four steps. First we shall establish that the map θ : M → |F| is bijective (Section 7.3), then that it is a homeomorphism (Section 7.4); then we shall show that C ∞ (M ) is geometrical and complete (Section 7.5) and ﬁnally prove that C ∞ (M ) is smooth (Section 7.6). 7.3. The map θ : M → |F| is bijective. Injectivity is obvious: If p, q ∈ M are distinct points, then there exists a function f ∈ C ∞ (M ) such that f (p) = f (q) (e.g., any function positive in a small neighborhood of p not containing q and identically zero outside of it; see Proposition 2.4); for this function the values of the homomorphisms © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 7

77

78

Chapter 7

p, q : f → R diﬀer, since q(f ) = f (q) = f (p) = p(f ). To prove surjectivity, suppose p : F → R is any homomorphism; let f ∈ C ∞ (M ) be any function with compact level surfaces (see Proposition 5.17) and λ = p(f ). Suppose that none of the points of the compact set L = f −1 (λ) correspond to the homomorphism p. Then there exists a family of functions {fx | x ∈ L} such that fx (x) = p(fx ). Consider the covering of L by the open sets Ux = {q ∈ M | fx (q) = p(fx )} and choose a ﬁnite subcovering Ux1 , . . . , Uxm . Consider the function g = (f − λ)2 +

m

(fxk − p(fxk ))2 .

k=1

This is a smooth function on M that vanishes nowhere and therefore possesses a smooth inverse 1/g. Now we easily obtain the standard contradiction, familiar from the examples: p(g) = (p(f ) − λ)2 +

m

(p(fxk ) − p(fxk ))2 = 0,

k=1

1 1 1 = p(1) = p g · = p(g)p = 0. g g Hence the homomorphism p is given by some point of L ⊂ M , so that θ is surjective. 7.4. The map θ : M → |F| is a homeomorphism. Let U be an open set in |F|. Then, by Deﬁnition 3.12, the set U is the union of sets of the form f −1 (V ), where V ⊂ R is open. Since we have f ∈ F = C ∞ (M ) is smooth (hence continuous), the sets f −1 (V ) are open in the topology of M , and so is U . Conversely, for any open set U in the topology of M there exists a function f ∈ F such that U = f −1 (R+ ). In fact, Lemma 4.17 (i) remains obviously valid if in its statement M is a manifold in the sense of Section 5.8. But R+ ⊂ R is open, so that U is open in the topology of |F|. 7.5. The algebra F = C ∞ (M ) is geometrical and complete. The fact that F is geometrical is obvious, since the elements f ∈ F are real functions on the set M , so that only the identically zero function vanishes at all points of M . The fact that F is complete is also obvious: Any function f : M → R that is smooth in a neighborhood of every point is smooth on the entire manifold, i.e., belongs to F = C ∞ (M ) (see Section 5.17). 7.6. C ∞ (M ) is a smooth R-algebra.

Equivalence of Coordinate and Algebraic Definitions

79

To prove this, we shall construct a countable atlas A2 = {(Uk , xk )} such that xk (Uk ) = Rn . We begin with an arbitrary countable atlas A0 = {(Vl , yl )} compatible with the given atlas A. Each open set yl (Vl ) ⊂ Rn maybe represented as the countable union of open balls in ∞ Rn ; i.e., yl (Vl ) = i=1 Gli , where Gli = {r ∈ Rn | r − ali < rli }.

The family A1 = yl−1 (Gli ), yl y−1 (G ) | l, i ∈ N is obviously a countli l able atlas on M , compatible with A0 and hence with A. Now note that for any chart (U, x) on M , where x(U ) is an open ball in Rn , we can construct the chart (U, η ◦ x), where η ◦ x(U ) = Rn , by taking η to be any diﬀeomorphism of the ball x(U ) onto Rn . For instance, if x(U ) is the ball of radius ρ and center a, we can take η to be the bijective map η : Rn → x(U ) given by the formula sρ . η(s) = a + 2 ρ + s2 The map η has an inverse, namely η −1 (r) =

ρ(r − a) ρ2

− r − a2

,

and therefore the chart (U, η ◦ x) is compatible with (U, x). Carrying out this construction for every chart in A1 , we obtain the required atlas A2 = {(Uk , xk )}. Consider any chart (Uk , xk ) ∈ A2 . The set Uk is obviously open in the topological space M . Clearly, the restriction of the algebra F = C ∞ (M ) to this set consists of all functions f : Uk → R such that f ◦ x−1 k is smooth on −1 the assignment f → f ◦ x is an isomorphism of the R-algebra Rn ; hence k

C ∞ (M ) U onto C ∞ (Rn ). This means that F = C ∞ (M ) is smooth. k This concludes the proof of Theorem 7.2. 7.7 Theorem. Suppose F is any smooth R-algebra. Then there exists a smooth atlas A on the dual space M = |F| such that the map F → C ∞ (M ),

f → (p → p(f )), ∞

of the algebra F onto the algebra C (M ) of smooth functions on M with respect to A (see Section 5.17) is an isomorphism. The proof will require four steps. In the ﬁrst one (Section 7.8) we construct a chart x : U → Rn for each of the open sets U of a countable covering of |F|, using a lemma proved in the second step (Section 7.9). In the third step (Section 7.10) we show that any two such charts are compatible. In the fourth and ﬁnal step (Section 7.11) we show that f ∈ F iﬀ f ∈ C ∞ (M ) (where f , an abstract element of F, is identiﬁed with the function f : |F| → R, p → p(f )).

80

Chapter 7

7.8. Construction of a chart x : U → Rn . By the deﬁnition of smooth R-algebras (Section 4.1), there is an open covering of |F| by open sets U for each of which there exist an isomorphism

i : F U → C ∞ (Rn ). Taking any such U , we shall construct the required chart x : U → Rn . The composition

μ |ρ| U → F U → |F|,

where μ is the inclusion (see Section 3.29) and ρ : F → F U the restriction map (Section 3.25), as can easily be checked, with the inclusion

coincides

U ⊂ |F|. Since μ is a homeomorphism onto F U (this is proved in Lemma

7.9 below) and |ρ| ◦ μ is the inclusion U ⊂ |F|, it follows that |ρ| must be a homeomorphism onto U ⊂ |F|. Now let h : F → C ∞ (Rn ) be the composition

i ρ F → F U → C ∞ (Rn ) , h = i ◦ ρ. Consider the dual map |h| = |ρ| ◦ |i|. Since |C ∞ (Rn ) | = Rn (by Example 3.16) and i is an isomorphism,

it follows from Section 3.20 that |i|

n is a homeomorphism |i| : R → F U . But |ρ| is also a homeomorphism

(onto U ⊂ |F|), and so the composition |h| = |ρ| ◦ |i| is a homeomorphism |h| : Rn → U , where U ⊂ |F| is open. Hence we obtain the required chart by putting x = |h|−1 : U → Rn .

7.9 Lemma. The inclusion map μ : U → F U is a homeomorphism onto

F . U Suppose

μ is not surjective; i.e., assume that there exists a point a ∈ F U \ μ(U ); set a = |i|−1 (a), where i : F U → C ∞ (Rn ) is the isomorphism appearing in Section 7.8. We consider two cases, depending on whether a belongs to the closure of μ(U ) or not. First case: a ∈ μ(U ). Consider the function f : x → 1/x − a deﬁned on Rn \ {a} and the function g = f ◦ |i|−1 ◦ μ deﬁned on U . We claim that g ∈ F U . Indeed, any point r ∈ Rn \{a} possesses a neighborhood on which f coincides with a smooth function deﬁned on all of Rn . Taking the inverse images of such neighborhoods under the map |i|−1 ◦ μ, we see that any point q ∈ U possesses a neighborhood on which g coincides with a function

that g locally from F U . By the deﬁnition of F U (Section 3.23), this means

coincides with functions belonging to F; hence g ∈ F U , as claimed. But now if we consider the function i(g), which is a smooth function (on the entire space Rn ) coinciding with f on the set |i|−1 (μ(U )) whose closure contains a, we obtain a contradiction (since f “becomes inﬁnite” at a). Second case: a ∈ / μ(U ). Consider two smooth functions on Rn : the identically zero one, and a function f0 that vanishes on the closed set |i|−1 (μ(U )) and equals 1 at a (f0 exists by Corollary 2.5). These are distinct functions,

Equivalence of Coordinate and Algebraic Definitions

81

so that their pullbacks by i on F U are diﬀerent elements of this algebra, which is impossible, since both pullbacks vanish on U . This proves the surjectivity of μ. The fact that it is a homeomorphism follows from Proposition 3.29. 7.10. The charts constructed in Section 7.8 are compatible. Suppose x : U → Rn and y : V → Rn are two such charts, while

i : F U → C ∞ (Rn ) and j : F V → C ∞ (Rn ) are the corresponding R-algebra isomorphisms. Let W = U ∩ V = ∅ (the case W = ∅ is trivial). By Proposition 3.25 we have the isomorphisms

i : F → C ∞ (x(W )), j : F → C ∞ (y(W )), W

W

W

W

which give us the isomorphism t : C ∞ (x(W )) → C ∞ (y(W )); this, by Proposition 3.16, shows that |t| : y(W ) → x(W ), the change of coordinate map, is a diﬀeomorphism. 7.11. Final step. Suppose f ∈ F. To prove that f is smooth in the sense of C ∞ (M ) (see Section 5.16), we must show that the function f ◦ x−1 is a smooth function on Rn for each of the charts x : U → Rn constructed in 7.8. But we have (see 7.8) f ◦ x−1 = f ◦ |h| = f ◦ |ρ| ◦ |i| = i(ρ(f )) ∈ C ∞ (Rn ) ,

since i is the isomorphism i : F U → C ∞ (Rn ). Thus f ◦ x−1 ∈ C ∞ (Rn ). Conversely, let f ∈ C ∞ (M ). This means that for any chart (U, x) the function f ◦ x−1 = i(f ◦ |ρ|) belongs to C ∞ (Rn ) and is the image (by i) of an element of F U (namely f ◦ |ρ|). Thus f locally coincides with elements

of F U and hence of F. Since F is complete, f ∈ F. 7.12. Thus we have established the equivalence of the two deﬁnitions of smooth manifold: the algebraic one (Deﬁnition 4.1) and the coordinate one (Deﬁnition 5.8). Theorems similar to 7.2 and 7.7 are valid for manifolds with boundary. The proofs are similar (with obvious modiﬁcation here and there). The reader who wishes to check that he or she has mastered the contents of Sections 7.1–7.11 will beneﬁt by carrying them out in detail. 7.13 Definition. A smooth set is a pair (W, C ∞ (W )), where W is a closed subset W ⊂ M of a smooth manifold M , and C ∞ (W ) is the algebra of smooth functions on W deﬁned as follows: def

C ∞ (W ) = f W | f ∈ C ∞ (M ) . Exercise. Prove that theorems similar to 7.2 and 7.7 are valid for smooth sets as well. 7.14 Exercise. Describe the algebras C ∞ (W ) for the following cases: 1. W = K is the coordinate cross on the plane: K = (x, y) ⊂ R2 | xy = 0 .

2. W ⊂ R2 is given by the equation y = |x|.

82

Chapter 7

3. W is a triangle in R2 : W = T1 ∪ T2 ∪ T3 , where T1 = {(x, y) | 0 y 1, x = 0}, T2 = {(x, y) | 0 x 1, y = 0}, T3 = {(x, y) | x + y = 1, x, y 0}. 4. W is the triangle described in Problem 3 together with its interior part: W = {(x, y)} | x + y 1, x, y 0}. 5. W is the cone x2 + y 2 = z 2 in R3 . 6. W = Wi , i = 1, 2, 3, is one of the three one-dimensional pairwise homeomorphic polyhedra depicted in Figure 7.1 (W1 lives in R2 , while W2 and W3 are in R3 ). In Chapter 9 the reader will ﬁnd Exercise 9.36, 5, in which it will be required to prove that the algebras C ∞ (Wi ), i = 1, 2, 3, are mutually non-isomorphic. 7. W ⊂ R2 is the closure of the graph of the function y = sin x1 .

W2 W1

W3 Figure 7.1. 1-skeleton of the tetrahedron.

Of course, there are various alternative descriptions of the algebras in question. For instance, perhaps the most direct and constructive way to represent smooth functions on the triangle (see Problem 3 above) is by means of triples (f1 , f2 , f3 ), fi ∈ C ∞ ([0, 1]), such that f1 (1) = f2 (0), f2 (1) = f3 (0), f3 (1) = f1 (0). Try to give a similar description of smooth functions on the cross and on the polyhedra mentioned in Problems 1 and 6, respectively. 7.15. Smooth sets sometimes appear implicitly in various mathematical problems. We illustrate this in the following exercises. Exercises. 1. Show that the conﬁguration space of a hinge mechanism can be viewed as a smooth set. 2. Determine which of the smooth sets corresponding to quadrilaterals and pentagons listed in Exercise 1 from Section 1.14 are not smooth manifolds and describe their smooth function algebras (cf. Exercise 6.25).

Equivalence of Coordinate and Algebraic Definitions

83

3. Prove that the smooth set corresponding to a nonrigid nongeneric pentagon (see Exercise 6.25), e.g., (2; 2, 1, 1, 2), is not a smooth manifold and describe its singular points (a pentagon is called rigid if its conﬁguration space consists of one point, e.g., (4; 1, 1, 1, 1)). Remark. The smooth set corresponding to the quadrilaterals (5; 3, 3, 1) is “diﬀeomorphic” to a pair of tangent circles at a point, say z. Smooth functions on this set can be viewed as pairs (f1 , f2 ), where f1 , f2 ∈ C ∞ (S 1 ), such that f1 (z) = f2 (z) and f1 (z) = f2 (z). Try to give similar descriptions for smooth functions on smooth sets appearing in Exercises 2 and 3. 7.16. The coincidence of the two deﬁnitions of a smooth map given in 6.1 and 6.14 is guaranteed by the following theorem. Theorem. Let M and N be manifolds with smooth atlases A and B and smooth function algebras FM and FN , respectively. A map ϕ : M → N is smooth with regard to A and B (Section 6.14) if and only if ϕ∗ (FN ) ⊂ FM , where ϕ∗ : FN → FM ,

f → f ◦ ϕ.

The proof is given below, with Sections 7.17 and 7.18 corresponding to the “only if” and the “if” parts of the theorem, respectively. 7.17. If ϕ : M → N is smooth, then ϕ∗ (FN ) ⊂ FM . Suppose f ∈ FN and a ∈ M . Choose a chart (V, y) in a neighborhood of ϕ(a) and a chart (U, x) in a neighborhood of a, the charts (V, y) and (U, x) being compatible with B and A, respectively, and satisfying ϕ(U ) ⊂ V . Then ϕ∗ (f ) ◦ x−1 = (f ◦ ϕ) ◦ x−1 = f ◦ y −1 ◦ y ◦ ϕ ◦ x−1 is smooth as the composition of two smooth maps of Euclidean domains. Thus locally the map ϕ∗ (f ) coincides with a map locally coinciding with an element of FM . Since FM is complete (see 7.5), it follows that ϕ∗ (f ) ∈ FM . 7.18. If ϕ∗ (FN ) ⊂ FM , then ϕ : M → N is smooth. Choose arbitrary charts (U, x) ∈ A and (V, y) ∈ B such that ϕ(U ) ⊂ V . We must prove, according to Deﬁnition 6.13, that the local coordinates of the point ϕ(a) are smooth functions of the local coordinates of a ∈ U . In other words, the functions y i ◦ ϕ = ϕ∗ y i must be smooth. For every function y i , let us choose a function fi ∈ FN such that y i = fi |V . By the assumption ϕ∗ (FN ) ⊂ FM , the functions ϕ∗ (fi ) = fi ◦ ϕ are smooth, and thus the functions y i ◦ ϕ|U = ϕ∗ (fi )|U are smooth as well. 7.19. We have proved that two categories, the category of manifolds (as smooth atlases) with smooth maps in the sense of 6.13 and the category of manifolds (as smooth R-algebras) with smooth maps in the sense of 6.1, are equivalent. From now on we shall not diﬀerentiate between the two categories. The notation (M, F) or (M, C ∞ (M )) will be used to denote

84

Chapter 7

a manifold; the identiﬁcations M = |F| and F = C ∞ (M ) will always be implied. The reader acquainted with the notion of complex manifold has probably noticed already that, in general, such a manifold does not coincide with the complex spectrum of its algebra of holomorphic functions. For example, as is well known from the elementary theory of functions of a complex variable, all holomorphic functions on the Riemann sphere (and on any compact complex manifold) reduce to constants. For this reason, it might seem that the “spectral approach” adopted in this book is less universal than the standard one, based on charts and atlases. Nevertheless, the observability principle forces us to understand complex manifolds as smooth ones, but equipped with an additional (complex) structure. In other words, complex manifolds within this approach are understood as solutions of certain diﬀerential equations, while complex charts appear as local solutions of these equations. This viewpoint, going back to Riemann, has many advantages, despite its apparent lack of simplicity. For example, it can be generalized to any commutative algebra by the methods of the “algebraic” diﬀerential calculus described below in Chapters 9 and 12.

8 Points, Spectra, and Ghosts

8.1. Spectra of commutative k-algebras. In the previous chapters, we looked at commutative R-algebras from the “physical” point of view, interpreting them as observation instruments. Namely, a commutative R-algebra A allows to observe (or see) the topological space |A|. Moreover, we noted that the very construction of this space is in no way related to the speciﬁcs of the ﬁeld R and is of a more general algebraic nature. Indeed, let A be a commutative unitary (= with unit) algebra over an arbitrary ﬁeld k. (In what follows, if this is required by the context, the unit of the algebra A is denoted by 1A and, if the converse is not stated explicitly, the term “commutative algebra” means commutative algebra with unit.) Let us put def

|A| = {h : A → k | h ∈ Homk (A, k),

h = 0}.

The set |A| deﬁned in this way is called the k-spectrum or the dual space of the algebra A. It is endowed with the natural topology, whose base of open sets consists of sets of the form def

Ua = {h ∈ |A| | h(a) = 0},

a ∈ A.

The obvious relation Uab = Ua ∩Ub shows that any ﬁnite intersection of sets of the base is also a set of the base. In algebraic geometry, such a topology is known as the Zariski topology, and we shall use this term in our more general context. Similarly, using geometrical terminology, we will refer to the homomorphism h ∈ |A| as a point of the spectrum of the algebra A. © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 8

85

86

Chapter 8

S1

S1

S2

S2 = S1

S1

S2

S1

S2

S 2 = S1

S2

Figure 8.1. Conjunction and disjunction.

8.2. Spectra of Boolean algebras. The fact that commutative algebras over ﬁelds other than R can actually serve as “observation instruments” can be seen in other branches of mathematics far removed from physics. The ﬁrst of such examples is, of course, algebraic geometry, in which spectra of a certain class of commutative algebras (see below) are studied. Another remarkable example, appearing in mathematical logic, is that of Boolean algebras, which we begin to discuss here and will continue studying further on. Recall that a Boolean algebra B is a commutative algebra over the twoelement ﬁeld F2 , all of whose elements are idempotent, i.e., satisfy the condition b2 = b. A very visual example of a Boolean algebra is the algebra of all subsets of a given set S. In that case, the logical operations of conjunction and disjunction correspond to the operations of intersection and union of subsets: S1 ∧ S 2 = S 1 ∩ S 2 ,

S1 ∨ S 2 = S 1 ∪ S 2 ,

Si ⊂ S.

As to the operations that endow the collection of subsets of S with a commutative algebra structure are intersection (multiplication) and symmetric diﬀerence (addition): def

S1 · S 2 = S 1 ∩ S 2 ,

def

S1 + S2 = (S1 ∪ S2 ) \ (S1 ∩ S2 ),

Si ⊂ S.

Exercise. Show that the set of all subsets of a set S is a commutative algebra over F2 with respect to the above deﬁned operations of multiplication and addition. The zero and the unit of this algebra are the empty set and the entire set S, respectively. In the following proposition, we list some main properties of Boolean algebras. Proposition. Let B be a Boolean algebra. Then (a) If the element b ∈ B is invertible, then b = 1B ;

Points, Spectra and Ghosts

S1

S1 S2 = S1

S2

S2

S1

S1 + S2 = (S1

87

S2

S2 ) (S2

S1 )

Figure 8.2. Product and sum.

(b) If I ⊂ B is a maximal ideal, then the algebra B/I is isomorphic to F2 ; (c) Every element b ∈ B diﬀerent from 1B is contained in a certain maximal ideal. In particular, |B| = ∅; (d) The intersection of all the nontrivial maximal ideals of B is trivial. (e) The open sets Ub , b ∈ B from the base of the Zariski topology of the spectrum |B| are also closed ; (f) The topological space |B| is Hausdorﬀ, see Section 5.7. (a)–(b) The proof of the ﬁrst two points is left to the reader as an exercise. (c) The principal ideal b · B is proper, because it does not contain the element 1B − b = 0. Indeed, assuming that 1B − b = bb , we come to a contradiction: 0 = b(1B − b) = b2 b = bb = 1B − b. But b · B, as any proper ideal, is contained in some maximal ideal I. Hence b ∈ b · B ⊂ I. (d) Assume I is the intersection of all the maximal ideals of the algebra B, 0 = b ∈ I, and I is the maximal ideal containing 1B − b. But I does not contain b, or else it would contain 1B and, therefore, would coincide with B. But since I ⊂ I, it follows that b ∈ / I, contradicting the assumption. (e) Obviously, Ub = {h ∈ |B| | h(b) = 0} = {h ∈ |B| | h(b) = 1} = {h ∈ |B| | h(b − 1B ) = 0} (8.1) This shows that Ub is the complement of the open set Ub−1B and is therefore closed. (f) Let h1 , h2 ∈ |B| and h1 = h2 . The latter means that there exists an element b ∈ B such that h1 (b) = h2 (b). This implies h1 (b) = 1, h2 (b) = 0 or h1 (b) = 0, h2 (b) = 1. In the ﬁrst of those two cases, this is equivalent to h1 ∈ Ua and h2 ∈ U1B −b , in the second case, to h2 ∈ Ua and h1 ∈ U1B −b . But, as we saw above, the open sets Ub and U1B −b don’t intersect.

88

Chapter 8

8.3. Homomorphisms and maps of k-spectra. As we have previously seen, smooth maps of smooth manifolds and homomorphisms of their algebras are in a natural bijective correspondence. Actually, this is a particular case of the following more general and fundamental algebraic construction. Let H : A → B be a homomorphism of commutative k-algebras. If h ∈ |B|, then the homomorphism h ◦ H : A → k is a point of the spectrum of the algebra A. Thus we obtain the following map of k-spectra |H| : |B| → |A|,

h → h ◦ H.

It possesses the following properties. Proposition.

(a) The map |H| is continuous.

(b) If the homomorphism H is surjective, then the map |H| is injective. H

H

2 1 (c) Let A1 −→ A2 −→ A3 be a homomorphism of commutative k-algebras. Then |H2 ◦ H1 | = |H1 | ◦ |H2 |.

(a) It suﬃces to verify that the sets |H|−1 (Ua ), a ∈ A, are open in |B|. But h ∈ |H|−1 (Ua ) ⇔ h ◦ H ∈ Ua ⇔ h(H(a)) = 0 ⇔ h ∈ UH(a) , i.e., |H|−1 (Ua ) = UH(a) . (b) If h1 , h2 ∈ |B| h1 = h2 , then there exists an element b ∈ B such that h1 (b) = h2 (b). Suppose further that b = H(a). In view of the subjectivity of H, such an element a ∈ A exists and so (|H|(h1 ))(a) = h1 (H(a)) = h1 (b) = h2 (b) = h2 (H(a)) = (|H|(h2 ))(a), i.e., |H|(h1 ) = |H|(h2 ). (c) This statement is obviously equivalent to the associativity of the composition of maps: h ◦ (H2 ◦ H1 ) = (h ◦ H2 ) ◦ H1 , h ∈ |A3 |. Exercises. 1. Show that the injectivity of the homomorphism H does not imply, in general, the surjectivity of the map |H|. 2. Is the assertion converse to item (b) of Proposition 8.3 true? 3. Show that the surjectivity of |H| does not imply, in general, the injectivity of the homomorphism H. 4. Let A = Zm [x], where m is prime. Show that any continuous map S of the set |A| into itself has the form S = |H|, where H : A → A is some algebra homomorphism. 8.4. Geometrization of algebras. It is remarkable that elements of an algebra A can be represented by functions on its spectrum. Namely, to each element a ∈ A, let us assign the k-valued function fa : |A| → k,

def

fa (h) = h(a),

h ∈ |A|.

Points, Spectra and Ghosts

89

and denote by Γ(A) the set of all such functions. If a, b ∈ A, λ ∈ k, then fa+b = fa + fb , fab = fa fb , fλa = λfa , f1A = 1Γ(A) , This is equivalent to the map Γ = ΓA : A → Γ(A),

a → fa ,

being a k-algebra homomorphism. This mapis obviously surjective, but not necessarily injective. If fa ≡ 0, then a ∈ h∈|A| Ker h and conversely. In other words, def Ker ΓA = Inv(A) = Ker h. h∈|A|

In what follows, it will be convenient for us to use the shorter notation μh = Ker h. Obviously, μz = μhz if A = C ∞ (M ) and z ∈ M . The function fa is the geometrical realization of the element a of the abstract algebra A. Accordingly, it is natural to call the map ΓA the geometrization of the algebra A. It follows from the above that the zero function corresponds to elements a ∈ Inv(A). In that sense all such elements are invisible or non-observable . Definition. A commutative k-algebra A is said be geometrical, if ΓA is an isomorphism or, equivalently, if Inv(A) = 0. Exercises. 1. Let m be prime, let A = Fm [x1 , . . . xn ] be the polynomial algebra over the m-element ﬁeld Fm . Show that |A| consists of mn points and describe Inv(A) and Γ(A). 2. Let A = k(x1 , . . . , xn ) be the algebra of power series over a ﬁeld k of zero characteristic. Show that |A| is a singleton and describe Inv(A) and Γ(A). The next proposition shows that the homomorphism ΓA “kills” the invisible elements of the algebra A without changing its spectrum. Proposition. For any commutative algebra A, the algebra Γ(A) is geometrical and |Γ(A)| = |A|. Since the homomorphism ΓA is surjective, the map |ΓA | is injective (Proposition 8.3, (b)). Further, h ∈ |A| implies Inv(A) ⊂ Ker h and so h can be represented as the following composition of homomorphisms Γ

A A −→ Γ(A) =

This shows that h = |ΓA |(h ).

A h −→ k. Inv(A)

8.5. “Boolean observability.” Now we can readily describe what can be “seen” by means of Boolean algebras. The main fact here is

90

Chapter 8

Proposition. Boolean algebras are geometrical. In particular, the geometrization homomorphism ΓB : B → Γ(B) is an isomorphism for any Boolean algebra B. If I is a maximal ideal of the algebra B, then the quotient algebra B/I is canonically isomorphic to the ﬁeld F2 (Proposition 8.2, (b)). Therefore, the canonical projection B → B/I may be identiﬁed with a certain element of the spectrum |B|. On the other hand, if h ∈ |B|, then Ker h is a maximal ideal, because the range of the homomorphism h is a ﬁeld. Thus the correspondence h → Ker h is bijective and so Inv(B) = h∈|B| Ker h coincides with the intersection of all the maximal ideals of the algebra B, and this intersection is trivial (Proposition 8.2, (d)). Thus any Boolean algebra B may be identiﬁed with its geometrization Γ(B). Moreover, since the elements of the algebra Γ(B) are functions fb : |B| → F2 , b ∈ B, each such function uniquely determines the domain where it takes the value 1F2 . For the function fb , such a domain will obviously be Ub . Hence the assignment U

B b → Ub ⊂ |B| is bijective. This assignment, as can be readily seen, does the following. Theorem. fb1 + fb2 → (Ub1 ∪ Ub2 ) \ (Ub1 ∩ Ub2 ) fb1 fb2 → Ub1 ∩ Ub2 (intersection). fb1 + fb2 + fb1 fb2 → Ub1 ∪ Ub2

(symmetric diﬀerence).

(union).

The ﬁrst two of these maps show that the canonical base {Ub }b∈|B| in the Zariski topology in |B|, supplied with the operations of symmetric diﬀerence and intersection, is an algebra isomorphic to B (see Exercise 8.2). But since the Boolean operations of conjunction (∧) and disjunction (∨) in the algebra of subsets of the given set are actually the operations of intersection and union, it follows that the last two of the three maps deﬁned above show that the Boolean operations in B itself are the following b 1 ∧ b 2 = b1 b 2 ,

b1 ∨ b2 = b1 + b2 + b1 b2 .

This also explains how Venn diagrams arise (see pages 86–87), as geometrical representations of Boolean operations. The geometrical description of Boolean algebras given above, except for the abstract characteristic of their spectra, is basically the well-known Theorem of Stone. The interpretation of Boolean algebras as a means of observation makes the proof of that theorem simpler and more transparent. What we actually “observe” in this approach will be discussed in the Appendix.

Points, Spectra and Ghosts

91

8.6. Invisible elements and homomorphisms. The behavior of “invisible” elements under homomorphisms of k-algebras is described by the following proposition. Proposition. Suppose H : A → B is a homomorphism of k-algebras and h ∈ |B|. Then (a) H Ker |H|(h) ⊂ Ker h; (b) H(Inv(A)) ⊂ Inv(B); (c) if the map |H| is surjective, then Ker H ⊂ Inv(A); (d) if the algebra A is geometrical and the map |H| is surjective, then the homomorphism H is injective. (a) Since Ker |H|(h) = Ker(h ◦ H), we have (8.2) Ker H ⊂ Ker |H|(h) = H −1 (Ker h). This implies the inclusion (a). (b) We have Inv(A) = Ker h ⊂ h ∈|A|

Ker h =

h ∈|H|

Ker |H|(h) .

h∈|B|

Therefore, in view of (8.2), Ker |H|(h) H (Inv(A)) ⊂H h∈|B|

=H

h∈|B|

H −1 (Ker h) = Ker h = Inv(B). h∈|B|

(c) From (8.2) we have Ker H ⊂ h∈|B| Ker |H|(h) . On the other hand, h∈|B| Ker |H|(h) = Inv(A), because |H| is surjective. (d) This follows immediately from (c), since in the given situation we have Inv(A) = 0. Note that for a geometrical algebra A, assertion (d) of the previous proposition is dual to assertion (b) of Proposition 8.1. For nongeometrical algebras A this is, generally speaking, not true. For example, the homomorphism H = ΓA induces a bijection of the spectra (Proposition 8.4), but is not injective if the ideal Inv(A) is nontrivial.

8.7. What is a point? Making more visible, more geometrical, any algebraic, analytic, or some other formal mathematical context is not only natural, but can be very useful. When trying to establish some fact, it is necessary to organize in a deﬁnite way the chain of logical arguments and calculations. For such a task, the geometrical picture, if it is available, may prove very helpful. In this connection, the story behind the proof of Fermat’s Last Theorem is rather instructive. Attempts to prove it by means

92

Chapter 8

of various purely algebraic tricks were unsuccessful for over 350 years. The solution was found on the basis of methods of contemporary algebraic geometry, by reformulating the problem in terms of rational points on algebraic curves. In many cases the necessary visibility can be achieved by linking the speciﬁc objects under consideration to some appropriate spaces. Since spaces consist of points, the very notion of point becomes crucial. Above, considering commutative algebras, we found the appropriate notion of point by analyzing the observation mechanism in classical physics. However, this should not be regarded as the universal solution. To show this, we shall introduce some other approaches below, so as to give the reader a more complete picture. In this connection, here are some remarks. (i) Although the “readings of measuring devices” (see Section 1.9) that we have used to motivate our constructions are usually real numbers, it is often necessary to consider measurements of a more general nature (complex numbers, matrices, residues modulo some positive integer, etc.). A striking example is the complex phase method used in the elementary theory of electricity. (ii) If the solution of some problem (in physics or mathematics) reduces to the solution of algebraic equations with coeﬃcients in a ring A (or in a ﬁeld A which is not algebraically closed), then as a rule, it is useful to seek the solution in an extension B ⊃ A of the ring, rather than in the ring A itself. The simplest example is the use of complex roots of a polynomial with real coeﬃcients. A less trivial example is given by the so-called Pauli operators, which arise in the quantum mechanics of electrons as matrix solutions σ1 , σ2 , σ3 of the system of equations σ12 = σ22 = σ32 = −1, σ2 σ3 − σ3 σ2 = σ1 , σ1 σ2 − σ2 σ1 = σ3 ,

σ1 σ3 − σ3 σ1 = −σ2 .

(iii) Mathematicians like to avoid exceptions and strive for the aesthetic uniﬁcation of any theory; this often leads them to invent a language in which the exceptions turn out to be part of the general rule. Thus parallel lines “intersect at inﬁnity,” imaginary points and points at inﬁnity appear in many situations, in particular in connection with R-algebras (see Example 1.18). This trick of giving intuitive geometrical meaning to diﬀerent algebraic situations is particularly fruitful in algebraic geometry, whose ideas will be used extensively here. 8.8. Motivating example. If f ∈ R[x] is an irreducible polynomial of degree 2, then there obviously exists no R-point of the algebra R[x] that would be a root of this polynomial; i.e., there is no homomorphism α : R[x] → R such that α(f ) = 0.

Points, Spectra and Ghosts

93

However, there exist exactly two homomorphisms into the ﬁeld of complex numbers, α, α : R[x] → C, such that α(f ) = α(f ) = 0. This will be proved below (see Theorem 8.11), but we suggest that the reader try to prove this now as an exercise. Homomorphisms such as α and α should be viewed as complex points for the algebra R[x]. 8.9. Suppose k is an arbitrary ﬁeld, K ⊃ k a ring without zero divisors, and F a commutative k-algebra with unit. In view of Example 8.8, it is natural to deﬁne a point of the algebra F “over the ring K” as an epimorphism of k-algebras a : F → K. It is no less natural to consider two points ai : F → Ki ⊃ k, i = 1, 2, identical (equivalent) if there exists a k-algebra isomorphism i : K1 → K2 such that a2 = i ◦ a1 . Now we can give the following deﬁnition. Definition. An equivalence class of k-algebra epimorphisms a: F → K ⊃ k is called a K-point of the k-algebra F. The ring K is then referred to as the domain of the point a. Further, we shall often speak of points of the algebra without specifying their domain. The reader should keep in mind that each point of the k-algebra F has its own domain, and diﬀerent points may have diﬀerent (or identical) domains. 8.10. From the viewpoint of our physical interpretation (in terms of measurements), the isomorphism i appearing in Deﬁnition 8.9 can be construed as an equivalent change of the “system of observations,” which, of course, should not inﬂuence the collection of points (states) determined by the algebra F. Deﬁnition 8.9 also possesses a purely algebraic motivation, which will appear in the next section. 8.11 Theorem. Points of a commutative k-algebra F (understood in the sense of Deﬁnition 8.9) correspond bijectively to prime ideals of the algebra F, the correspondence being given by the map (a : F → K) → Ker a ⊂ F. The fact that the ideal Ker a is prime and depends only on the equivalence class of the homomorphism a and that diﬀerent equivalence classes correspond to diﬀerent ideals is obvious. To prove surjectivity, take an arbitrary prime ideal p ⊂ F. Then the ring F/p has no zero divisors, and the quotient map q : F → F/p is a point of the algebra F for which Ker q = p. The set of prime ideals of a k-algebra F is called the prime spectrum of F and is denoted by Spec F. According to the theorem, Spec F may be viewed as constituting the set of all points of the k-algebra F. Obviously,

94

Chapter 8

when k = R, we have Spec F ⊃ |F|, so that elements of Spec F generalize the notion of R-point; i.e., R-points in the sense of Section 3.4 are “points over the ﬁeld R” in the sense of Section 8.9. 8.12. Examples (based on Theorem 8.11). I. A smooth manifold has no points with values z ∈ C with Im z = 0. Suppose a : C ∞(M ) → C is a C-point and a(f ) = i ∈ C. Therefore, we have a 1 + f 2 = 0, and hence 1 + f 2 ∈ Ker a. On the other hand, the element 1 + f 2 is obviously invertible in C ∞ (M ) and as such does not belong to any proper ideal of C ∞ (M ). II. The set of all points of the polynomial algebra R[x] can be identiﬁed with the complex half-plane {z | z 0} to which a point ω, corresponding to the ideal {0} ⊂ R[x], has been added. Since R[x] is a principal ideal domain, any prime ideal in it is of the form R[x]f , where f is an irreducible polynomial. But then either f ≡ 0 or f = a(x − b) or f = a(x − c)(x − c), where a, b ∈ R, while c and c are conjugate complex numbers. Note that we have proved the statement mentioned at the end of Section 8.8. Exercises. 1. Let k be a ﬁeld, and X a formal variable; describe Spec k[X]. def k ∞ 2. Show that μ∞ z = k μz , z ∈ M , is a prime ideal in C (M ). 8.13. We have learned to assign a set of points, Spec F, to every k-algebra F. It is only natural to try to supply this set with a topology. In a similar situation, working with |F|, we introduced the topology induced from the topology in R, but in the case of an arbitrary k-algebra F we no longer have a ﬁxed ﬁeld R with a trustworthy topology. A new idea is needed to ﬁnd a topology in Spec F. Suppose C is a closed set in Rn , then there exists a function f ∈ C ∞ (Rn ) such that f (r) = 0 ⇐⇒ r ∈ C. The same is true for a closed set B in a manifold M : There exists a function f ∈ F = C ∞ (M ) such that f (p) = 0 ⇐⇒ p ∈ B (Section 4.17 (i)). This circumstance will be the basis for introducing a topology in Spec F. First note that any element f ∈ F may be viewed as a function on Spec F ranging in F mod p (which may have diﬀerent algebraic structures). Namely, for any prime ideal p ∈ Spec F, we put f (p) = f mod p. For any subset E ⊂ F, denote by V (E) ⊂ Spec F the set of all prime ideals containing E. In other words, if p ∈ V (E), then for any function f ∈ E we have f (p) = 0. We can now deﬁne the Zariski topology in the prime spectrum Spec F of an arbitrary k-algebra F as the topology whose basis of closed sets is the collection {V (E) | E ⊂ F}.

Points, Spectra and Ghosts

95

8.14. The deﬁnition of the Zariski topology in Spec F can also be given in terms of the closure operation. For any M ⊂ Spec F, consider the ideal IM = p∈M p and deﬁne the closure M of M as M = {p ∈ Spec F | p ⊃ IM }. This deﬁnition has a clear algebraic meaning: If the element f ∈ F vanishes on M (i.e., p ∈ M ⇒ f (p) = 0 ⇐⇒ f ∈ p), then it also vanishes on M ; conversely, if any f that vanishes on M is also zero on some point p ∈ Spec F, then p ∈ M . (The fact that this construction gives the same topology as 8.13 follows directly from the two deﬁnitions.) The reader has perhaps wondered what price we shall have to pay for the extreme generality of this construction. It turns out that the Zariski topology in Spec F is non-Hausdorﬀ. In particular, Spec F contains nonclosed points, which will be considered in the next section. Note, however, that every point of |F| is closed. 8.15 Exercises. 1. Prove that a one-point set {p} ⊂ Spec F is closed in the Zariski topology iﬀ the ideal p is maximal. . Compare the Zariski 2. Describe the Zariski topology of R[X] ∞ topologies of R[X] = R and of C (R) = R. 3. Describe the Zariski topology of Spec R[X]. 8.16 Proposition. The closure of a point q ∈ Spec F coincides with the set V (q) = {p ∈ Spec F | q ⊂ p} (in other words, {q} is the set of all common zeros of all the functions from the ideal q). By Deﬁnition 8.13, the set V (q) is closed. On the other hand, E ⊂ F, and q ∈ V (E) ⇒ q ⊃ E ⇒ V (q) ⊂ V (E). 8.17. Corollaries and examples. I. If a k-algebra has no zero divisors, then {0} ∈ Spec F and {0} = Spec F. For this reason, the ideal {0} is said to be the common point of the set Spec F. II. Suppose f ∈ k[x1 , . . . , xn ] = F is an irreducible polynomial in n variables over the algebraically closed ﬁeld k. Consider the prime ideal p = k[x1 , . . . , xn ]·f . The closure of the point p in the prime spectrum Spec F of the polynomial algebra F contains all the maximal ideals corresponding to the points of the hypersurface Hf = {(r1 , . . . , rn ) ∈ kn | f (r1 , . . . , rn ) = 0}. The point p is therefore called the common point of this hypersurface. (To each point (r1 , . . . , rn ) ∈ kn there corresponds the maximal ideal of F generated by all monomials x1 − r1 , . . . , xn − rn .)

96

Chapter 8

III. By Proposition 8.16, the closure of the point p in Example III above contains all the prime ideals containing p. From the point of view of “maximal ideal topology,” each such ideal determines a surface of “lesser dimension” contained in the hypersurface Hf and is the common point of this lesser surface. 8.18 Theorem. The prime spectrum Spec F of any k-algebra F is compact. Let us restate the theorem in terms of closed sets: if {Mα }α∈A is a family of closed sets such that α∈A Mα = ∅, then we can choose a ﬁnite subfamily Mα1 , . . . , MαN with empty intersection. We shall prove the theorem in this (equivalent) form. Without loss of generality, we can assume that Mα = V (Eα ), where Eα ⊂ F is some subset. Denote by [Eα ] ⊂ F the ideal generated by Eα and note that Mα = V (Eα ) = V [Eα ] . ∅= α∈A

α∈A

α∈A

But this means that the ideal α∈A [Eα ] is not contained in any prime ideal and hence in any maximal ideal of the algebra F. In other words, [Eα ] = F 1. α∈A

Therefore, we can ﬁnd α1 , . . . , αN ∈ A and fi ∈ [Eαi ], i = 1, . . . , N , such

N that i=1 fi = 1. But in this case N N Mαi = V [Eαi ] = V (F) = ∅. i=1

i=1

8.19. Let us now investigate how prime spectra behave under homomorphisms of their k-algebras. Suppose F1 , F2 are k-algebras and α : F2 → F1 is a k-algebra homomorphism. We claim that if p ⊂ F1 is a prime ideal, then so is α−1 (p). If q1 ∈ α−1 (p), then for any q2 ∈ F2 , we have α(q1 q2 ) ∈ α(α−1 (p)q2 ) = pα(q2 ) ⊂ p. Thus q1 q2 ∈ α−1 (p), and α−1 (p) is an ideal. To show that it is prime, let q1 , q2 ∈ F and q1 q2 ∈ α−1 (p). Then α(q1 )α(q2 ) ∈ p, and since p is prime, at least one of the elements α(q1 ), α(q2 ), say the ﬁrst, belongs to p. Then we see that q1 ∈ α−1 (α(q1 )) ⊂ α−1 (p); i.e., the ideal α−1 (p) is prime.

Points, Spectra and Ghosts

97

Now, to every k-algebra homomorphism α : F2 → F1 , we can assign the map of prime spectra |α| : Spec F1 → Spec F2 , p → α−1 (p). 8.20 Proposition. For any k-algebra homomorphism α : F2 → F1 the corresponding prime spectra map |a| : Spec F1 → Spec F2 is continuous in the Zariski topology. The proof is a straightforward veriﬁcation of deﬁnitions; we leave it to the reader. 8.21. Now, copying a similar deﬁnition for smooth manifolds (see Section 6.1), we can give the following deﬁnition: Definition. A map β : Spec F1 → Spec F2 is said to be smooth if there exists a k-algebra homomorphism α : F2 → F1 such that β = |α|. Thus the reader has met with another example of a category: the category of k-algebra prime spectra, in which the morphisms are smooth maps of spectra. 8.22. Suppose the ideal p ∈ Spec F of the k-algebra F satisﬁes F/p = k. Then it is easy to show that p is a maximal ideal. We also suggest that the reader work out the following exercises. Exercises.

1. The ideal p is maximal iﬀ F/p is a ﬁeld.

2. If F is ﬁnitely generated and p is maximal, then F/p is a ﬁnite algebraic extension of k. 8.23. Definition. The maximal spectrum Spm F of the k-algebra F is the set of maximal ideals of F. The previous section, as well as Sections 8.15–8.16, where we studied the closure of one-point sets in Spec F, suggests that instead of Spec F we should have been studying Spm F. Indeed, we would thus have avoided such pathology as non-closed points, and the set of domains of points would be more manageable. However, this is not quite reasonable because of the fact that unlike prime ideals, maximal ideals may have inverse images that are no longer maximal (see Example 8.24). Therefore, generally there is no map of maximal spectra that could naturally be assigned to a homomorphism of the corresponding algebras. In other words, the correspondence A → Spm A does not deﬁne a functor from the category of K-algebras to the category of topological spaces. 8.24. Example. Suppose F2 = k[x1 , x2 ] is the algebra of polynomials in two variables, F1 = k(x1 ) is the ﬁeld of rational functions, and α : F2 → F1 is the composition of the quotient epimorphism F2 = k[x1 , x2 ] → k[x1 , x2 ]/(x2 ) = k[x1 ] and the inclusion k[x1 ] → k(x1 ) = F1 .

98

Chapter 8

The ideal {0} ⊂ k(x1 ) is a maximal one; however, α−1 ({0}) = F2 · x2 is a prime ideal, but not a maximal one. This establishes the statement in italics in the previous section. 8.25 Exercises. 1. Let M be a compact manifold. Prove that any maximal ideal in C ∞ (M ) is of the form μz , z ∈ M (see Exercise 2 from Section 8.12). 2. Show that for any noncompact manifold M there exist maximal ideals in C ∞ (M ) diﬀerent from the μz ’s. 3. Show that any such maximal ideal, regarded as a C ∞ (M )-module or an algebra, has an inﬁnite number of generators. 8.26. In the remainder of this section, we shall discuss certain aspects of prime spectra that distinguish them from smooth manifolds. As we saw in Example 8.12, II, a simple spectrum may contain, besides “visible” points, certain points whose geometrical interpretation is not obvious (e.g., the prime ideal ω in that example). We shall begin with a few examples showing that “points” of that type may already appear in the maximal spectrum Spm F of a k-algebra. 8.27. Ghosts. The reader who has studied Example 8.12, I, in detail has undoubtedly noticed that the maximal ideals of the algebra of smooth functions on a compact manifold correspond bijectively to ordinary R-points of this algebra (i.e., to ordinary points of the manifold). For noncompact manifolds this is no longer the case. Indeed, suppose Ic is the ideal of all functions with compact support on a noncompact manifold. None of the maximal ideals containing Ic is the kernel of any R-point (the proof is left to the reader as an exercise). Such maximal ideals correspond to “points” of noncompact manifolds we have called ghosts. In Sections 8.29–8.31, we shall see how such ghosts can actually materialize. 8.28. Before continuing, the following remark is called for. Suppose F is the R-algebra of smooth functions on the compact manifold M . As was pointed out above, M = Spm F, but the reader should not think that M = Spec F. To that end, the reader should work out the following exercise. Exercise. A function f ∈ F is called ﬂat at the point a ∈ M if it vanishes with all its derivatives at that point. The set of all functions that are ﬂat at the given point a ∈ M constitutes an ideal. Prove that this ideal is prime (i.e., is a point of Spec F that corresponds to no point of M ). 8.29. Example (compactiﬁcations of the line). Let F be the R-algebra of smooth functions on the line R, and let Ic , as above, be its ideal of functions with compact support. One can try to describe the maximal ideals that contain Ic , but most of these ideals—ghosts—have no reasonable constructive description.

Points, Spectra and Ghosts

99

Nevertheless, there are two very nice sets of functions that contain Ic , namely the sets μ+∞ and μ−∞ : μ±∞ = f ∈ F | lim f (r) = 0 . r→±∞

These sets, unfortunately, are not ideals (in poetic language we can say that “were they ideals, they would be maximal ones”). This unpleasant circumstance can be overcome as follows. Put dk f dk f and lim exist F1 = f ∈ F | ∀k 0 lim r→+∞ dr k r→−∞ dr k and deﬁne μ±∞ by putting μ±∞ = F1 ∩ μ±∞ . Then, (i) μ+∞ and μ−∞ are maximal ideals of the algebra F1 containing the ideal Ic . (ii) If M = [0, 1] ⊂ R1 , and the algebra FM consists of the restrictions of all the functions from F to M , then the manifold with boundary M is diﬀeomorphic to |F1 |. (iii) Under this identiﬁcation, the ideal Ic becomes the ideal of functions that vanish at the end points of the closed interval [0, 1], while the ideals μ−∞ and μ+∞ become the points 0 and 1, respectively. (iv) This implies that all the maximal ideals of the algebra F1 , except μ+∞ and μ−∞ , correspond to ordinary points of the line R1 ; as for the ideals μ±∞ , they are “ghosts,” which are adjoined to the line in order to make it compact. (v) Prove that the algebra F1 is not isomorphic to the algebra FM . Try to ﬁnd another subalgebra of F, diﬀerent from F1 , isomorphic to FM . 8.30. Another example. Among the numerous other methods of compactifying the line, we consider only one more. Let dk f dk f F2 = f ∈ F1 | ∀k 0 lim = lim ∈R r→+∞ dr k r→−∞ dr k and put μ∞ = μ+∞ ∩ F2

(= μ−∞ ∩ F2 ).

It is easy to prove that the algebra F2 is isomorphic to C ∞ (S 1 ), the algebra of smooth functions on the circle. This can be visualized by imagining that the “inﬁnite end points” of the line are glued together by means of the “ghost” corresponding to the ideal μ∞ (see Figure 8.3). 8.31. Further examples. I. Denote by F the algebra of complex-valued functions of a complex variable, deﬁned, holomorphic, and bounded in the domain |z| < 1. Among the maximal ideals of the algebra F, we know, of

100

Chapter 8

Figure 8.3. Gluing “infinite end points.”

course, the ideals corresponding to points a, |a| < 1, namely μa = {f ∈ F | f (a) = 0} . It is also possible to deﬁne μa when |a| = 1 by putting μa = {f ∈ F | lim f (z) = 0}. z→a

Prove that the maximal spectrum of the algebra F consists of all the ideals μa , |a| 1. II. Find the maximal spectrum of the algebra F2 of complex-valued functions of two complex variables, analytic and bounded in the open polydisk {(z1 , z2 ) | |z1 | < 1, |z2 | < 1}.

9 Diﬀerential Calculus as Part of Commutative Algebra

9.1. The formal approach to observation procedures in classical physics, described in the previous chapters, leads to rather important conclusions. Let us ﬁrst note that the diﬀerential calculus is the natural language of classical physics. On the other hand, all information about some classical physical system is encoded in the corresponding algebra of observables. From this it follows that the diﬀerential calculus needed to describe physical problems is a part of commutative algebra. The basic aim of this chapter is to explain how to obtain a purely algebraic deﬁnition of diﬀerential operators using general facts known from the classical calculus. It is very important that the constructions obtained below can be used for arbitrary, not necessarily smooth, algebras. But perhaps the key point is somewhat diﬀerent. We shall see that the diﬀerential calculus is a formal consequence of arithmetical operations. This unexpected and beautiful fact plays an important role not only in mathematics itself. It also allows us to reconsider some paradigms reﬂecting the relationship of mathematics to the natural sciences and, above all, with physics and mechanics. By including observability in our considerations, we ensure that mathematics may be regarded as a branch of the natural sciences. 9.2. Let us start with the simplest notion of calculus, namely that of the derivative, which is the formal mathematical counterpart of velocity. From elementary mechanics we know that velocity is a vector, i.e., a directed segment. This point of view is hardly satisfactory, since it is completely unclear what a directed segment is in the case of an abstract (curved ) © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 9

101

102

Chapter 9

manifold. For this reason, the founders of diﬀerential geometry deﬁned the tangent vector as a quantity described in a given coordinate system as an n-tuple of numbers; when passing from one coordinate system to another, this n-tuple is to be transformed in a prescribed manner. This approach is also unsatisfactory, because it describes vectors in local coordinates and does not explain what they are in essence. In many modern textbooks on diﬀerential geometry one can ﬁnd another deﬁnition of a tangent vector, which does not use local coordinates: A tangent vector is an equivalence class of smooth curves tangent to each other at a given point of the manifold under consideration. But try to ﬁnd the sum of such classes or multiply a class by a number (i.e., try to introduce the structure of a linear space), and you will immediately see that this deﬁnition is inconvenient to work with. The principal reason why these deﬁnitions of a tangent vector are unsatisfactory is their descriptive nature: They say nothing about the functional role of this notion in the diﬀerential calculus. This role can be understood if algebras of observables are used. Let A be an algebra of observables. Then, by deﬁnition, M = |A| is the “manifold” of states for the corresponding system, and a particular state is an element h ∈ |A|. Therefore, a time evolution of the system’s state is described by a family ht . Consequently, the velocity of evolution at a moment t0 is def dht Δh = : A → R, (9.1) dt t=t0 where, by deﬁnition, dht 1 = lim (ht+Δt − ht ), Δt→0 dt Δt whatever the meaning of the limit above may be. In other words, the motion that we conceive is in fact the change in time in the readings given by the measuring devices, while the velocity of motion is the velocity of these changes. The translation of the above to geometrical language is accomplished by the correspondence M z ↔ h = hz ∈ |A|, where A = C ∞ (M ) and hz (f ) = f (z) for any f ∈ C ∞ (M ). After this translation, the family {ht } becomes a curve z(t) on the manifold M . This curve is such that ht = hz(t) , while the derivative dht dt is a tangent vector to z(t) and consequently to the manifold M . Formula (9.1) now acquires the form dhz(t) Δz = : A → R, (9.2) dt t=t0

Diﬀerential Calculus as Part of Commutative Algebra

103

where z = z(t0 ), and obviously Δz ↔ Δh . Thus, we arrive at the following interpretation: velocity of changes of system states

←→

tangent vector to the manifold M

where the system is described by the algebra A. The words tangent vector were used intuitively above. Our aim now is to deﬁne it rigorously, solely in terms of the algebra of observables A. The above interpretation allows us to do this in a natural way. It remains only to understand the mathematical nature of the operator Δh (or Δz ), informally deﬁned by formula (9.1) (respectively by (9.2)): Not all maps from A to R (for example, h) may be appropriately called velocities of state change. 9.3. The algebra of observables is a combination of two structures: that of a vector space and that of a multiplicative structure. The interaction of the operator Δh (or Δz ) with the ﬁrst structure is obvious: It is R-linear, i.e., ⎛ ⎞ k k Δh ⎝ λj fj ⎠ = λj Δh (fj ), λj ∈ R, fj ∈ C ∞ (U ), (9.3) j=1

j=1

where U is a coordinate neighborhood. Exercise. Check this. To understand how Δh interacts with the multiplicative structure of the algebra C ∞ (M ), one needs to compute the action of this operator on the product of two observables. We have dht (f g) d(ht (f )ht (g)) = Δh (f g) = t=t0 dt dt t=t0 dht (f ) dht (g) = ht0 (g) + ht0 (f ) dt t=t0 dt t=t0 = Δh (f )h(g) + h(f )Δh (g); i.e., Δh satisﬁes the following Leibniz rule: Δh (f g) = Δh (f )h(g) + h(f )Δh (g).

(9.4)

In geometrical form, this rule can be written as Δz (f g) = Δz (f )g(z) + f (z)Δz (g).

(9.5)

Thus, the rules (9.3) and (9.4) completely govern the interrelations between the operator Δh and the basic structures in the algebra of observables. Therefore, we have a good reason for giving the following deﬁnition:

104

Chapter 9

9.4. Deﬁnition. A map ξ : C ∞ (M ) → R is said to be a tangent vector to a manifold M at a point z ∈ M if it satisﬁes the two following conditions: (1) R-linearity: ⎛ ⎞ k k ⎝ ⎠ ξ λj fj = λj ξ(fj ), j=1

λj ∈ R, fj ∈ C ∞ (M ).

j=1

(2) The local Leibniz rule (or the Leibniz rule at a point z): ξ(f g) = f (z)ξ(g) + g(z)ξ(f ),

f, g ∈ C ∞ (M ).

Obviously, if we now deﬁne the sum of two tangent vectors and the multiplication of a tangent vector by a real number using the rules (ξ + ξ )(f ) = ξ(f ) + ξ (f ), (λξ)(f ) = λξ(f ),

λ ∈ R,

then in both cases the result will be an R-linear operator satisfying the Leibniz rule; i.e., we shall obtain a tangent vector again. In other words, the set Tz M of all tangent vectors at a point z ∈ M possesses a natural structure of a vector space over R. This space is called the tangent space of the manifold M at z. Remark. The zero vector 0z ∈ Tz M is just the zero map from C ∞ (M ) to R and as such coincides with 0z for any other point z . But it is natural to distinguish between vectors 0z and 0z , z = z , since they are tangent to M at two diﬀerent points (are subject to diﬀerent Leibniz rules). For a formally satisfactory explanation of this distinction see Section 9.52. 9.5. Let us now describe the operators ξ ∈ Tz M in local coordinates. Let U be a domain in Rn and ﬁx a local coordinate system x1 , . . . , xn . Assume that z = (z1 , . . . , zn ), y = y(Δt) = (z1 + α1 Δt, . . . , zn + αn Δt), αi ∈ R, in this system. Then obviously z = y(0) and

n f (y(Δt)) − f (z) ∂f df (y(t)) Δz (f ) = lim = = αi . Δt→0 Δt dt ∂xi z t=0 i=1

This is nothing but the derivation of the function f in the direction α = (α1 , . . . , αn ). Hence, the operator Δz is described by an n-tuple of real numbers. This observation explains the hidden meaning of the classical descriptive deﬁnition of tangent vectors. The arguments of Section 9.3 and 9.5 use the operator Δz , which was not deﬁned in conceptually correct way, and so these arguments are not rigorous

Diﬀerential Calculus as Part of Commutative Algebra

105

either. Nevertheless, they make it possible to deﬁne tangent vectors in terms of the algebra of observables (see Deﬁnition 9.4). Using this deﬁnition as a starting point, we can now compare our approach with the usual one. 9.6 Tangent Vector Theorem. Let M be a smooth manifold, z ∈ M , and let x1 , . . . , xn be a local coordinate system in a neighborhood U z. Then, in this coordinate system, any tangent vector ξ ∈ Tz M can be represented in the form n ∂ ξ= αi , αi ∈ R. ∂xi z i=1 In other words, the notions of tangent vector and of diﬀerentiation in a given direction coincide. The proof of this theorem consists of several steps. The ﬁrst of them is proposed to the reader. 9.7 Lemma–Exercise. Let f = const ∈ R; then ξ(f ) = 0.

9.8 Lemma. Tangent vectors are local operators, i.e., if two functions f, g ∈ F coincide on an open set U z, then for any tangent vector ξ ∈ Tz M the equality ξ(f ) = ξ(g) holds. To prove this statement, it suﬃces to check that if the equality f |U = 0 holds in some neighborhood U z, then ξ(f ) = 0. Indeed, in this case, by Corollary 2.5, there exists a function h ∈ C ∞ (M ) such that h(z) = 0 and h|M \U = 1. Consequently, f = hf , and, by the Leibniz rule, ξ(f ) = ξ(hf ) = f (z)ξ(h) + h(z)ξ(f ) = 0. 9.9 Lemma. The spaces Tz U and Tz M are naturally isomorphic for any open neighborhood U z. The embedding i : U ⊂ M of the open set U into the manifold M induces ∞ the map dz i : Tz U → Tz M as follows. Let ξ ∈ Tz U and f ∈ C (M ); set dz i(ξ)(f ) = ξ(f U ). (Check that dz i(ξ) is indeed a tangent vector.) Obviously, the map dz i is R-linear. Let us construct the inverse map. To this end, note ﬁrst that for any function g ∈ C ∞ (U ) one can ﬁnd a function f ∈ C ∞ (M ) coinciding with g in some neighborhood in U of the point z. Indeed, consider a function h ∈ C ∞ (U ) vanishing outside some compact neighborhood V0 ⊂ U and equal to 1 in a neighborhood V1 ⊂ V0 of the point z (see Section 2.5). Then for f ∈ C ∞ (M ) we can take a function vanishing outside U and coinciding with gh in U . Now deﬁne an A-linear map πU : Tz M → Tz U by setting πU (η)(g) = η(f ). It follows from Lemma 9.8 that the value η(f ) does not depend on the choice of the function f ; i.e., the homomorphism πU is well deﬁned. It is now easy to see that dz i ◦ πU = id and πU ◦ dz i = id.

106

Chapter 9

9.10. From the lemma above it follows that we may conﬁne ourselves to the case M = U ⊂ Rn . Moreover, we may assume the domain U to be starshaped with respect to the point z (i.e., such that y ∈ U implies [z, y] ⊂ U , the domain U contains the entire closed interval [z, y]). By Corollary 2.9, any smooth function f in a star-shaped neighborhood of z can be represented in the form n n ∂f (xi − zi ) (z) + (xi − zi )(xj − zj )gij (x). f (x) = f (z) + ∂xi i=1 i,j=1 Applying the tangent vector ξ to the last equality and using the Leibniz rule, we immediately see that for any derivation ξ ∈ Tz M , n ∂f αi (z), ξ(f ) = ∂x i i=1 where αi = ξ(xi − zi ) = ξ(xi ), which concludes the proof of the Tangent Vector theorem. 9.11. From the Tangent Vector Theorem it follows that Tz M is an n-dimensional vector space over R. In fact, by this theorem, the tangent vectors ∂ ∂ , ..., ∂x1 z ∂xn z generate the space Tz M for any coordinate neighborhood U of the point z. Let n ∂ αj ξ= ∂x j j=1 be a linear combination of the vectors ∂/∂xj . Since obviously αj = ξ(xj ), the vector ξ does not vanish if at least one of the coeﬃcients αj is not zero. Hence, the vectors ∂/∂xj are linearly independent and form a basis of the tangent space Tz U . The isomorphism dz i : Tz U → Tz M constructed during the proof of the theorem now shows that dim Tz M = n. Below we shall identify the vectors ∂/∂xj forming a basis of Tz U with the vectors dz i (∂/∂xj ) forming a basis of Tz M . It follows from the above that the dimension of the tangent space Tz M is equal to the number of local coordinates in any chart containing z. In other words, it is equal to the dimension of the manifold M . 9.12. Let y1 , . . . , yn be another local coordinate system in a neighborhood of the point z. Then in the corresponding basis of the tangent space ∂/∂y1 , . . . , ∂/∂yn , one has z

z

ξ=

n k=1

βk

∂ . ∂yk z

Diﬀerential Calculus as Part of Commutative Algebra

107

Further, in view of the classical chain rule (see any advanced calculus textbook), n n n n n ∂yk ∂ ∂yk ∂ ∂ αi αi αi ξ= = = , ∂xi z i=1 ∂xi ∂yk z ∂xi ∂yk z i=1 i=1 k=1

k=1

i.e., βk =

n i=1

αi

∂yk , ∂xi

k = 1, . . . , n.

The matrix that transforms the basis ∂ , ..., ∂x1 z

∂ , ∂xn z

corresponding to the local coordinates x1 , . . . , xn , to the basis ∂ ∂ , ..., , ∂y1 z ∂yn z corresponding to the coordinates y1 , . . . , yn , is the Jacobi matrix ⎞ ⎛ ∂y1 ∂y1 ... ⎜ ∂x1 ∂xn ⎟ ⎜ . .. ⎟ .. ⎟ . Jz = ⎜ . . ⎟ . ⎜ . ⎝ ∂y ∂yn ⎠ n ... ∂x1 ∂xn z The subscript z indicates that the elements of the Jacobi matrix are computed at the point z. We see that the coordinate change rules obtained here for tangent vectors are in agreement with the approach accepted in the tensor calculus. 9.13. The diﬀerential of a smooth map. It is quite natural that any smooth map of manifolds generates a map of tangent vectors. (Try to check this yourself by looking at Figure 9.1 on the next page and continuing the informal arguments of Section 9.2.) A rigorous construction is as follows. Let ϕ : M → N be a smooth map and ξ ∈ Tz M . Then the map η = ξ ◦ ϕ∗ : C ∞ (N ) → R is a tangent vector to the manifold N at the point ϕ(z). In fact, its R-linearity is obvious. In addition, for any f, g ∈ C ∞ (N ), we have η(f g) = ξ(ϕ∗ (f g)) = ξ(ϕ∗ (f )ϕ∗ (g)) = ξ(ϕ∗ (f ))(ϕ∗ (g)(z)) + (ϕ∗ (f )(z))ξ(ϕ∗ (g)) = η(f )g(ϕ(z)) + f (ϕ(z))η(g). Deﬁnition. The map dz ϕ : Tz (M ) → Tz (N ),

ξ → ξ ◦ ϕ∗ ,

ξ ∈ Tz M,

108

Chapter 9

ϕ

ο Figure 9.1. Mapping tangent vectors.

is called the diﬀerential of the map ϕ at the point z ∈ M . Obviously, the diﬀerential dz ϕ is a linear map. 9.14 Exercise. Prove that if ψ : N → L is another smooth map, then dz (ψ ◦ ϕ) = dϕ(z) ψ ◦ dz ϕ.

(9.6)

Prove also that if N = L and ψ = idN , then dz ψ = idTz N . Formula (9.6), applied to ψ = ϕ−1 , shows that dϕ(z) ϕ−1 = (dz ϕ)−1 . In particular, dz ϕ is an isomorphism whenever ϕ is a diﬀeomorphism. We can now return to the discussion of Section 4.10 and prove the following statement: 9.15 Proposition. The algebras C ∞ (M ) and C ∞ (N ) are not isomorphic if dim M = dim N . In particular, the algebras C ∞ (Rn ) and C ∞ (Rm ) are not isomorphic if m = n.

Diﬀerential Calculus as Part of Commutative Algebra

109

Indeed, let Φ : C ∞ (N ) → C ∞ (M ) be an isomorphism. In that case, ϕ = |Φ| : M → N is a diﬀeomorphism. As was observed earlier, the diﬀerential dz ϕ : Tz M → Tϕ(z) N is then an isomorphism for all z ∈ M . Therefore, dim M = dim Tz M = dim Tϕ(z) N = dim N.

9.16. Let us now describe dz ϕ in coordinates. As in Section 6.15, choose local charts (x1 , . . . , xn ) and (y1 , . . . , ym ) in M and N containing the points z and ϕ(z), respectively. Let ϕ∗ (yi ) = ϕi (x1 , . . . , xn ) be the functions describing the map ϕ in coordinates. Then, for g ∈ C ∞ (N ), we have

∂ ∂ ∗ dz (ϕ) (g) = ◦ ϕ (g) ∂xi z ∂xi z ∂g(ϕ1 (x1 , . . . , xn ), . . . , ϕm (x1 , . . . , xn )) = ∂xi z m ∂g(ϕ1 (x1 , . . . , xn ), . . . )) ∂ϕj (x1 , . . . , xn ) = ∂yj ∂xi z z j=1

m ∂ϕj (x1 , . . . , xn ) ∂g = (z) ϕ∗ ∂xi ∂yj z j=1 m m ∂ϕj (x1 , . . . , xn ) ∂g ∂ϕj ∂g ϕ(z) = (z) (ϕ(z)) ∂xi ∂xi ∂yj z ∂yj j=1 j=1 ⎡ ⎤ m ∂ϕ ∂ j ⎦ (g). =⎣ (z) ∂x ∂y ϕ(z) i j j=1

=

In other words,

dz (ϕ)

∂ ∂xi z

=

m ∂ϕj j=1

∂xi

(z)

∂ ; ∂yj ϕ(z)

(9.7)

i.e., the matrix of the linear map dz ϕ in the bases ∂ ∂ ⊂ Tϕ(z) (N ), ⊂ Tz (M ), ∂xi z ∂yj ϕ(z) respectively, is the Jacobi matrix ∂ϕi /∂xj z of ϕ at the point z. The subscript z indicates here that all the derivatives in this matrix are taken at the point z. Thus, the coordinate representation for the diﬀerential dz ϕ

110

Chapter 9

is of the form

⎞ ⎛ ∂ϕ1 β1 ⎜ .. ⎟ ⎜ ∂x1 ⎜ . ⎟ ⎜ . ⎜ ⎟=⎜ . ⎜ . ⎟ ⎜ . ⎝ .. ⎠ ⎝ ∂ϕ m βm ∂x1 ⎛

... ..

.

...

⎞ ⎛ ⎞ ∂ϕ1 α1 ⎜ .. ⎟ ∂xn ⎟ ⎜ ⎟ .. ⎟ ⎟ ⎜ . ⎟ . ⎟ .⎜ . ⎟, . ∂ϕm ⎠ ⎝ . ⎠ αn ∂xn z

(9.8)

where (α1 , . . . , αn ) and (β1 , . . . , βm ) are the coordinates of the vector v ∈ Tz (M ) and of its image dz ϕ(v) in the bases ∂ ∂ , , ∂xi z ∂yj ϕ(z) respectively. 9.17. The tangent manifold. As we know from elementary mechanics, any particular state of a mechanical system S is determined by its position (conﬁguration) and instantaneous velocity. If M = MS is the conﬁguration space (see Sections 1.1, 5.12) of this system, then, as follows from the arguments of Section 9.2, the notion of a tangent vector to the manifold MS is identical to the concept of the system’s state. More precisely, if we consider a tangent vector ξ ∈ Tz M , then z is the position of the system, while ξ is its instantaneous velocity. Thus, the set of all system states (pay attention to Remark 9.4) is def Tz M. TM = z∈M

We will equip this set with a smooth manifold structure in a natural way below. The object obtained is called the tangent manifold of the manifold M . Since the evolution of the system is uniquely determined by its initial state, diﬀerential equations describing possible evolutions should be equations on the manifold T M . Besides mechanics, tangent manifolds naturally arise in various branches of mathematics, and ﬁrst of all in diﬀerential geometry. 9.18. To introduce a smooth manifold structure on T M , we shall need the following simple facts: I. Any smooth map Φ : M → N generates the map of sets T Φ : T M → T N, taking a tangent vector ξ ∈ Tz M to dz Φ(ξ) ∈ TΦ(z) N . By (9.6), T (Φ ◦ Ψ) = T Φ ◦ T Ψ. II. If W ⊂ R is an open domain of the arithmetical space Rn , then T W is naturally identiﬁed with the domain W × Rn ⊂ Rn × Rn = R2n . n

Diﬀerential Calculus as Part of Commutative Algebra

111

Namely, if z ∈ W, z = (z1 , . . . , zn ) and ξ ∈ Tz W , then ξ ⇐⇒ (z1 , . . . , zn , α1 , . . . , αn ) if ξ =

n

αi

i=1

∂ . ∂xi

III. A natural map of sets πT = πT M : T M → M,

Tz M ξ → z ∈ M,

is deﬁned, taking any tangent vector to its point of application. IV. If U ⊂ M is an open subset, then πT−1 (U ) = Tz M = Tz U = T U. z∈U

z∈U

9.19. Let us now note that any chart (U, x) of the manifold M , by the above, generates the map T x : T U → T W ⊂ R2n ,

where W = x(U ).

This map is obviously a bijection. Using 9.18, IV, we can identify T U with πT−1 (U ) and obtain a 2n-dimensional chart πT−1 (U ), T x in T M . In other words, coordinate functions {xi , qj } associated with T x are such that xi for (z, ξ) ∈ T U is the ith coordinate of the point z, while qj is the jth component of the coordinate presentation for the vector ξ in the basis ∂/∂xi |z . Charts of this kind are called special . If the charts (U, x) and (U , y) are compatible on M , the corresponding special charts πT−1 (U ), T x and πT−1 (U ), T y are also compatible. In fact, the analytical form for the coordinate change (T y) ◦ (T x)−1 : T (W ) → T (W ),

W = y (U ) ,

is (see (9.12)) yi = yi (x),

βj =

n k=1

αk

∂yj (x), ∂xk

and consequently the map under consideration has the Jacobi matrix

J ∗ , 0 J where J is the Jacobi matrix for the coordinate change y ◦ x−1 , while the asterisk denotes an (n×n)–matrix. Hence, (T y)◦(T x)−1 is a diﬀeomorphism of open sets in R2n . Let A = {(Uk , xk )} be an atlas on M . Then, by the above, we see that T A = {(πT−1 (Uk ), T xk )} is a special atlas on T M . If two atlases A1 and A2 are compatible on M , then the corresponding special atlas T A1 is compatible with T A2 as well. If A is a countable and Hausdorﬀ atlas, then T A enjoys the same properties. For all these reasons, the smooth manifold

112

Chapter 9

structure on the set T M determined by the atlas T A does not depend on the choice of the atlas A on M . The set T M equipped with the described smooth structure is called the tangent manifold of the manifold M . Let us also note that the map πT : T M → M described in Section 9.18, III, is smooth. It is called the tangent ﬁber bundle or simply tangent bundle of the manifold M . This correspondence deﬁnes a functor.1 We use here the words ﬁber bundle for the ﬁrst time. The exact deﬁnition will be given in Section 11.10; Chapters 11 and 12 are completely devoted to the study of this notion. In the case under consideration, these words mean that the tangent spaces Tz M (ﬁbers of the projection πT ) are identical (i.e., diﬀeomorphic to each other) and “ﬁber” the tangent manifold T M . Moreover, these ﬁbers are mutually isomorphic vector spaces. Fiber bundles of this type are called vector bundles and studied in detail in Chapter 12. Exercise. Prove that the map dΦ : T M → T N , corresponding to the smooth map Φ : M → N , is smooth and the following diagram TM

dΦ

πT

M

TN πT

Φ

N

is commutative. 9.20. Sticking to the observability principle, it would be much more attractive to associate with any (smooth) algebra A an algebra T A such that |T A| = T |A|. That is why we have to conﬁne ourselves to the example of the cotangent manifold considered in the next sections. The question of the algebraic deﬁnition of cotangent manifolds is delayed to Chapter 10. 9.21. Besides the description of a mechanical system in terms of position– velocity, there exists another, often more convenient, description in terms of position–momentum. The fundamental relation p = mv, which ties the velocity and the momentum of a mass point, shows that momenta are linear functionals on the space of velocities. In other words, the momentum of a system S in a position z ∈ M = MS is a linear functional on the tangent space Tz M ; i.e., it is an element of the dual 1 The reader will recall that a covariant functor Φ from a category M to a category N is given by assignments Φ : Ob M → Ob N and Φ∗ : Mor(P, Q) → Mor(Φ(P ), Φ(Q)), such that Φ∗ (ξ ◦ η) = Φ∗ (ξ) ◦ Φ∗ (η) and Φ∗ (idM ) = idN . Contravariant functors are deﬁned similarly (reverse arrows and raise indices).

Diﬀerential Calculus as Part of Commutative Algebra

113

space Tz∗ M = HomR (Tz M, R). def

The space Tz∗ M is called the cotangent space to M at the point z, and its elements are called tangent covectors to the manifold M at the point z. So, the momenta of a mechanical system S are tangent covectors to the conﬁguration manifold M = MS . These and many other considerations lead us to the notion of cotangent manifold . To give a formal deﬁnition of the latter, let us consider the set def Tz∗ M T ∗M = z∈M

together with the natural projection πT∗ = πT ∗ M : T ∗ M → M,

Tz∗ M θ → z ∈ M.

9.22. A natural smooth manifold structure on T ∗ M can be deﬁned using the scheme applied already to T M . It is extremely helpful here to understand tangent covectors as diﬀerentials of functions on M . Namely, let z ∈ M and let f be a function from C ∞ (M ). Let us deﬁne a function dz f on Tz M by setting def

dz f (ξ) = ξ(f ),

ξ ∈ Tz M.

Exercise. Prove the following statements: 1. If f = const, then dz f = 0. 2. dz (f + g) = dz f + dz g, 3. dz (f g) = f (z)dz g + g(z)dz f . By the deﬁnition of the linear space structure on Tz M (see Section 9.4), the function dz f is linear. Therefore, dz f is a tangent covector at the point z, called the diﬀerential of the function f at z. If (U, x) is a chart containing the point z, then ∂ ∂x i dz xi (z) = δij , = ∂xj z ∂xj where δij is the Kronecker symbol. This shows that (dz x1 , . . . , dz xn ) is the basis of the space Tz∗ M dual to the basis ∂ ∂ , . . . , ∂x1 z ∂xn z in Tz M . Therefore, any covector θ ∈ Tz∗ M is uniquely written in the form n pi dz xi , pi ∈ R. θ= i=1

114

Chapter 9

It is useful to note that pi = θ

∂ . ∂xi z

In particular, if θ = dz f , then n ∂ ∂f ∂f (z) and hence dz f = (z)dz xi . θ = ∂xi z ∂xi ∂xi i=1 This formula justiﬁes the adopted terminology and shows that any covector θ can be represented in the form θ = dz f . Thus, tangent covectors at a given point are exhausted by diﬀerentials of functions at this point. 9.23. Any smooth map Φ : M → N generates the linear map ∗ N → Tz∗ M dz Φ∗ : TΦ(z) dual to the linear map dz Φ : Tz M → TΦ(z) N. ∞

If ξ ∈ Tz M and g ∈ C (N ), then, by deﬁnition, dz Φ∗ (dΦ(z) g)(ξ) = dΦ(z) g(dz Φ(ξ)) = dz Φ(ξ) (g) = ξ Φ∗ (g) = dz Φ∗ (g) (ξ). This means that dz Φ∗ (dΦ(z) g) = dz Φ∗ (g). Note also that, in the notation of Section 9.16, the matrix of the map dz Φ∗ ∗ in the bases {dz xi } and {dΦ(z) yj } in Tz∗ M and TΦ(z) N , respectively, is the transposed Jacobi matrix Jz = ∂yi /∂xj (z). 9.24. The construction of special charts on T ∗ M is accomplished in a way similar to that used above for T M . Instead of properties I–IV from Section 9.18, the following facts should be used: I. Any diﬀeomorphism Φ : M → N generates the bijection Φ∗ : T ∗ M → T ∗ N,

Tz∗ M θ → (dz Φ∗ )−1 (θ).

II. If W ⊂ Rn is an open domain, then T ∗ W = W × Rn ⊂ Rn × Rn = R2n , and the identiﬁcation T ∗ W = W × Rn follows the rule T ∗ W θ ⇐⇒ (z1 , . . . , zn , p1 , . . . , pn ), n where z = (z1 , . . . , zn ), θ = i=1 pi dz xi and (x1 , . . . , xn ) are the standard coordinates in Rn . Now let (U, x) be a chart on M and W = x(U ) ⊂ Rn . By duality, the natural identiﬁcation of Tz U with Tz M allows the identiﬁcation of Tz∗ U

Diﬀerential Calculus as Part of Commutative Algebra

115

with Tz∗ M . In turn, this leads to the identiﬁcation of T ∗ U with πT−1∗ (U ). Now using II, we obtain a special chart (πT−1∗ (U ), T ∗ x) on T ∗ M . Here T ∗ x denotes the system of coordinate functions {xi , pj }, where xi for (z, θ) taken from T ∗ U is the ith coordinate of the point z, while pj is the jth component of the decomposition of the vector θ in the basis dxi . def −1 If A = {(Uk , xk )} is an atlas on M , then T ∗ A = πT ∗ (Uk ), T ∗ xk is an atlas (of dimension 2n) on T ∗ M . If two atlases A1 and A2 are compatible on M , then so are the atlases T ∗ A1 and T ∗ A2 . For this reason, the atlas T ∗ A determines a smooth manifold structure on T ∗ M independent of the choice of a particular atlas A. The 2n-dimensional manifold T ∗ M thus obtained is called the cotangent manifold of the manifold M . Note also that the map πT ∗ : T ∗ M → M is smooth. It is called the cotangent bundle of M . 9.25. Any function f ∈ C ∞ (M ) generates the smooth map sdf : M → T ∗ M,

sdf (z) = dz (f ).

This map is characterized by the fact that any point z ∈ M is taken to a point in the ﬁber of the cotangent bundle πT−1∗ (z) = Tz∗ M over z. Such maps are called sections. This notion will be discussed in more detail in subsequent chapters; see Sections 11.12 and 12.7. Exercise. Describe sdf in special local coordinates. ∗ 9.26. Any map Φ : M → N generates a family dz Φ∗ : TΦ(z) N → Tz∗ M of maps taking cotangent spaces of points in M to those in N . Unfortunately, it does not allow one, in general, to construct a map of cotangent manifolds ∗ M. T ∗ N → T ∗ M that reduces to dz Φ∗ when restricted to TΦ(z) If dim M = dim N = n and the map Φ : M → N is regular at all points of the manifold, i.e., all diﬀerentials are isomorphisms, then one can deﬁne a smooth map T ∗ M → T ∗ N covering Φ and reducing to (dz Φ∗ )−1 when restricted to Tz∗ M . The map Φ∗ considered in Section 9.24 is its particular case.

9.27. Note that the cotangent space Tz∗ M can be deﬁned in a purely algebraic way, as it is done in algebraic geometry. Let μz be the ideal consisting of all functions vanishing at the point z: μz = {f ∈ C ∞ (M ) | f (z) = 0}. def

Proposition. There exists a natural isomorphism between Tz∗ M and the quotient μz /μ2z . Consider the quotient algebra Jz1 M = C ∞ (M )/μ2z def

116

Chapter 9

and the map d¯z : Jz1 M → Tz∗ M,

d¯z ([f ]) = dz f,

where f ∈ C ∞ (M ) and [f ] = f mod μ2z . Since dz f = 0 for z ∈ μ2z (see the exercise from Section 9.22), this map is well deﬁned. Obviously, it is R-linear and surjective (see Section 9.22) because any covector θ can be presented as the diﬀerential of some function, θ = dz f . The decomposition of the algebra C ∞ (M ) into the direct sum of linear spaces C ∞ (M ) = R ⊕ μz ,

f = f (z) + (f − f (z)),

gives the direct sum decomposition Jz1 M = R ⊕ μz /μ2z .

(9.9)

By the exercise from Section 9.22, the map dz annihilates the ﬁrst summand. On the other hand, Hadamard’s lemma (Lemma 2.8) shows that f ∈ μz and dz f = 0 imply f ∈ μ2z . Therefore, the restriction d¯z to μz /μ2z is an isomorphism. Corollary. dim Jz1 M = n + 1. Indeed, the above proposition allows us to rewrite equality (9.9) in the form Jz1 M = R ⊕ Tz∗ M.

9.28. The quotient algebra Jz1 M was useful in the proof of Proposition 9.27; as we shall see later, this algebra is one of the most important constructions of the diﬀerential calculus. It is called the algebra of ﬁrst-order jets (or of 1-jets) at the point z ∈ M for the algebra of smooth functions C ∞ (M ). The union J 1M = Jz1 M z∈M

can be endowed with a natural smooth manifold structure in a similar way as was done above for T ∗ M . Exercise. Develop the corresponding constructions in detail. Describe the special coordinates in J 1 M . The manifold J 1 M is called the manifold of ﬁrst-order jets for the manifold M . Similarly to tangent and cotangent manifolds, J 1 M is ﬁbered over M by means of the natural map πJ 1 : J 1 M → M,

πJ 1 ([f ]1z ) = z,

Diﬀerential Calculus as Part of Commutative Algebra

117

where [f ]1z denotes the image of the function f under the quotient map C ∞ (M ) → Jz1 M . Similar to πT and πT ∗ (see Sections 9.19 and 9.24, respectively), the map πJ 1 is also a vector bundle over M . Its ﬁbers are of dimension (n + 1). For any function f ∈ C ∞ (M ), one can consider the smooth map sj1 f : M → J 1 M,

sj1 f (z) = [f ]1z ,

which is a section of the bundle πJ 1 . The map πJ 1 ,T ∗ : J 1 M → T ∗ M,

πJ 1 ,T ∗ ([f ]1z ) = dz f,

which relates the manifold of 1-jets in the natural way to the cotangent manifold, is a one-dimensional vector bundle over T ∗ M . A remarkable feature of the manifold J 1 M is that it allows us to construct an exhaustive theory of ﬁrst-order partial diﬀerential equations in one unknown. In this theory, diﬀerential equations are interpreted as submanifolds in J 1 M . The Tangent Vector Theorem allowed us to make the ﬁrst step in understanding the diﬀerential calculus as a part of commutative algebra. The next step is to deﬁne tangent vectors to the spectrum of an arbitrary commutative algebra. Let A be an arbitrary unital commutative K-algebra. Denote by |A| its K-spectrum, i.e., the set of all (unital) K-homomorphisms from A to K. 9.29. Deﬁnition. A map ξ : A → K is called a tangent vector , or a derivation at a point h ∈ |A|, if it (i) is K-linear, i.e., ⎞ ⎛ k k λj fj ⎠ = λj ξ(fj ), ξ⎝ j=1

λj ∈ K,

fj ∈ A;

j=1

(ii) satisﬁes the Leibniz rule at h, i.e., ξ(f g) = f (h)ξ(f ) + g(h)ξ(f ),

f, g ∈ A.

This deﬁnition, in the case K = R and A = C ∞ (M ), coincides with the deﬁnition of tangent vectors to the manifold M (= |A|) at a point z ∈ M . To understand this fact, it suﬃces to recall the identiﬁcation M = |C ∞ (M )| and to treat z as the homomorphism hz : f → f (z), f ∈ C ∞ (M ). The set of all tangent vectors at a given point is naturally endowed with a K-module structure (or that of a vector space over K when K is a ﬁeld): def

1. (ξ1 + ξ2 )(a) = ξ1 (a) + ξ2 (a), a ∈ A; def

2. (kξ)(a) = kξ(a), k ∈ K, a ∈ A.

118

Chapter 9

Let us denote this K-module by Th A. If K = R and A = C ∞ (M ), then, under the above identiﬁcation of points z ∈ M with K-homomorphisms hz , the space Tz M will coincide with Thz A. Remark. In algebraic geometry, one considers various spectra of algebras, maximal, primitive, etc. Treating the symbol h in the previous deﬁnition in an adequate way, the reader will easily deﬁne tangent vectors for points of all these spectra. 9.30. Cotangent spaces of commutative algebra spectra. Proposition 9.27, revealing the purely algebraic nature of cotangent bundles, shows how to deﬁne the cotangent space of the spectrum |A| for an arbitrary commutative K-algebra A at some point h ∈ |A|. Namely, set Th∗ A = μh /μ2h , def

(9.10)

where μh is the kernel of the K-algebra homomorphism h : A → K. By deﬁnition, Th∗ A is a K-module. Its role is illustrated by the following proposition: Proposition. For any K-algebra A, the natural surjection of K-modules νh : HomK (Th∗ A, K) → Th A

(9.11)

is deﬁned. If K is a ﬁeld, then νh is an isomorphism. Let us ﬁrst note that any K-linear map ϕ : Th∗ A → K determines the tangent vector ξϕ ∈ Th A,

ξϕ (a) = ϕ([a − h(a) · 1A ]),

where [b] = b mod (check it). The correspondence ϕ → ξϕ in an obvious way determines the K-module homomorphism νh : HomK (Th∗ A, K) → Th A. μ2h

The map νh is a surjection. In fact, let ξ ∈ Th A. Consider the K-linear map ϕξ ([a]) = ξ(a), a ∈ μh . ϕξ : Th∗ A → K, The Leibniz rule implies ξ(μ2h ) ⊂ μh , and thus the map ϕξ is well deﬁned. Obviously, νh (ϕξ ) = ξ. If K is a ﬁeld, then νh is also an injection. In fact, now let a, b ∈ μh and [a] = [b]. Since K is a ﬁeld, one can always ﬁnd a linear function ϕ deﬁned on the vector K-space Th∗ A and satisfying ϕ([a]) = ϕ([b]), i.e., ξϕ (a) = ξϕ (b). 9.31. To ﬁnd an algebraic counterpart for the concept of the diﬀerential of a smooth map, let us note that to any (unital) K-algebra homomorphism F : A1 → A2 there corresponds a map of K-spectra, namely |F | : |A2 | → |A1 |, |A2 | h → h ◦ F ∈ |A1 |.

Diﬀerential Calculus as Part of Commutative Algebra

119

If, in addition, ξ ∈ Th (A1 ), then the map def

dh |F |(ξ) = ξ ◦ F : A1 → K is a tangent vector to the space |A1 | at the point |F |(h) = h ◦ F . Thus we obtain the K-linear map dh |F | : Th (A1 ) → Th◦F (A2 ) (prove this fact). If F = ϕ∗ , where ϕ : M2 → M1 is a smooth map and Ai = C ∞ (Mi ), i = 1, 2, then the diﬀerentials dh ϕ and dh |F | coincide. 9.32 Exercises. 1. Prove that TidK K = 0, where idK : K → K, is the only point of the K-spectrum for K. 2. Let i : K → A, k → k · 1A , be the canonical embedding and let ξ ∈ Th A. Prove that ξ i = 0; i.e., any derivation at a given point takes constants to zero. 3. Let F : A1 → A2 be a K-algebra epimorphism. Prove that the map dh |F | is a monomorphism for any point h ∈ |A2 |. 4. Let C 0 (M ) be the algebra of all continuous functions on M . Prove that Tz (C 0 (M )) = 0 for any point z ∈ M . 9.33. The advantages of the algebraic approach to the diﬀerential calculus can already be shown at this point, though so far we have succeeded only in giving the deﬁnition of tangent vectors. For example, we can deﬁne tangent spaces to manifolds with singularities and furthermore—to arbitrary smooth sets (see Section 7.13)—and obtain the simplest invariants of singular points. Some examples will be given below. The following statement, whose proof can be literally carried over from Section 9.9, will be quite useful in analyzing these examples. Below we use the notation of Sections 3.23–3.25. Proposition. Suppose F is an arbitrary geometrical R-algebra and let U ⊂ |F| be an open subset. Then the restriction homomorphism def ρU : F → FU = F U induces an isomorphism dh (ρU ) : Th (FU ) → TρU ◦h (F),

h ∈ |FU |.

A similar construction is valid for arbitrary K-algebras. 9.34 Exercises. 1. Let W = (x, y) ∈ R2 | y 2 = x3 be the semicubical parabola. Show that Tz W is two-dimensional for z = (0, 0) and onedimensional otherwise. An obvious consequence of this fact is that the algebra C ∞ (W ) = C ∞ R2 / y 2 − x3 C ∞ R2 is not smooth (cf. 4.16).

120

Chapter 9

2. Give an example of a smooth set whose tangent spaces are all onedimensional except for single point in which it is 3-dimensional. 9.35. Example. Suppose K is the coordinate cross on the plane (see Sec tion 7.14, 1), K = {(x, y) ⊂ R2 | xy = 0}, and F = C ∞ (K) = C ∞ R2 K is the algebra of smooth functions on K. Let us describe Tz C ∞ (K) for all points z ∈ K. Elements of the algebra C ∞ (K) may be understood as pairs (f (x), g(y)) of smooth functions on the line satisfying f (0) = g(0). In other words, C ∞ (K) = {(f (x), g(y)) | f (0) = g(0)}. Note that for any nonsingular point on the cross, i.e., for a point of the form (x, 0), with x = 0, or (0, y), with y = 0, the tangent space is onedimensional. Let us consider, say, z = (x, 0), x = 0, and U = {(x , 0) | xx > 0}. Then U is open by Proposition 9.33, we have in K = |F|, and consequently, Tz F = Tz F U . It remains to note that F U = C ∞ (R1+ ) = C ∞ (R1 ). For a basis vector in the space T(x,0) , we can take the operator d d df (x), , (f (x), g(y)) = dx (x,0) dx (x,0) dx while for tangent spaces of the form T(0,y) , we can take the operator d d dg (y). , (f (x), g(y)) = dy (0,y) dy (0,y) dy Now consider the point (0, 0). Obviously, the operators d d and dx (0,0) dy (0,0) will be tangent vectors at this point. They are linearly independent, and hence the space T(0,0) is at least two-dimensional. Since the natural restriction map τ : C ∞ (R2 ) → C ∞ (K) is an epimorphism, then by Problem 3 of Section 9.32, the kernel of the map d(0,0) τ : T(0,0) (C ∞ (K)) → T(0,0) (C ∞ (R2 )) = R2 is trivial. Therefore, the tangent space T(0,0) (C ∞ (K)) is isomorphic to R2 . Thus, the property of the point (0, 0) ∈ K to be singular manifests itself, in particular, in the fact that the dimension of the tangent space at this point is greater than for “normal” ones. Let us stress that the standard coordinate approach does not allow one to deﬁne tangent vectors at the point (0, 0). But if one tried to understand a tangent vector as an equivalence class of curves, then there would be no linear space structure in the set of such tangent vectors to K at this point.

Diﬀerential Calculus as Part of Commutative Algebra

121

n 9.36 Exercise. Let W ⊂∞R nbe a smooth set. Recall that by deﬁnition, ∞ C (W ) = {f W | f ∈ C (R )}. Describe Tz W for all points z ∈ W in the following cases (see Exercise 7.14): ! 1. W ⊂ R2 is given by the equation y = |x|.

2. W is the triangle in R2 : W = W1 ∪ W2 ∪ W3 , where W1 = {(x, y) | 0 y 1, x = 0}, W2 = {(x, y) | 0 x 1, y = 0}, W3 = {(x, y) | x + y = 1, x, y 0}. 3. W is the triangle from the previous problem together with the interior domain: W = {(x, y) | x + y 1, x, y 0}. 4. W is the cone x2 + y 2 = z 2 in R3 . 5. W = Wi , i = 1, 2, 3, is one of the one-dimensional homeomorphic polyhedra shown in Figure 7.1. Explain why the algebras C ∞ (Wi ), i = 1, 2, 3, are pairwise nonisomorphic. 6. W ⊂ R2 is the closure of the graph of the function y = sin 1/x. 9.37 Exercise. Let dim Tz W = 0, where W is a smooth set. Prove that z is an isolated point of W . This is no longer true for arbitrary algebras. Find an example of an algebra F such that |F| R, but dim Tz F = 0 for all z ∈ |F|. Can you construct another algebra F with |F| R and dim Tz F = 2 for some z ∈ |F|? And the same for all z ∈ |F|? 9.38. More complicated objects of the diﬀerential calculus, which can be constructed using tangent vectors, are vector ﬁelds. Vivid geometrical images of vector ﬁelds are provided by numerous ﬁelds of forces in mechanics and physics, velocity ﬁelds of continuous media, etc. A “ﬁeld” of arrows on a meteorological map may be considered as the velocity ﬁeld of moving air masses. Let us try to formalize this notion in the same spirit as was done for tangent vectors. The ﬁrst step in this direction is obvious: a vector ﬁeld on a manifold M is a family of tangent vectors {Xz }z∈M , where Xz ∈ Tz M . In terms of the algebra of observables C ∞ (M ), this means that we are dealing with the family of operators Xz : C ∞ (M ) → R,

z ∈ M.

In particular, such a family of operators assigns to each function f from C ∞ (M ) the set of numbers {Xz (f ), z ∈ M }, which a physicist would call a scalar ﬁeld, while a mathematician would just call it a function on M . Denoting this function by X(f ), we obtain by deﬁnition def

X(f )(z) = Xz (f ),

z ∈ M.

122

Chapter 9

In this notation, it becomes clear that the words vector ﬁeld must be understood as a sort of operation on the algebra C ∞ (M ): X : C ∞ (M ) → ? , where the question mark means some set of functions on M . A natural way to formalize the idea of smoothness of a vector ﬁeld X is to set ? = C ∞ (M ): X(f ) ∈ C ∞ (M ) for any function f ∈ C ∞ (M ). Thus, a smooth vector ﬁeld X on M is an operator acting on C ∞ (M ): X : C ∞ (M ) → C ∞ (M ). By the R-linearity of the maps Xz , of which the operator X “consists,” this operator is also R-linear. Moreover, the Leibniz rule for a tangent vector Xz at a point z implies X(f g)(z) = Xz (f g) = Xz (f )g(z) + f (z)Xz (g) = X(f )(z) g(z) + f (z) X(g)(z) = [X(f )g + f X(g)](z); i.e., the operator X satisﬁes the Leibniz rule X(f g) = X(f )g + f X(g), f, g ∈ C ∞ (M ).

(9.12)

The above motivates the following deﬁnition: Deﬁnition. An R-linear operator X : C ∞ (M ) → C ∞ (M ) satisfying the Leibniz rule (9.12) is called a smooth vector ﬁeld on the manifold M . Everywhere below the word “smooth” is omitted, since we shall deal with smooth vector ﬁelds only. 9.39. The above deﬁnition of a vector ﬁeld was formulated in terms of the base algebra C ∞ (M ) and completely satisﬁes the principle of observability. Moreover, we can now a posteriori justify the use of the words vector ﬁeld in this deﬁnition: We can associate with any vector ﬁeld X the family of tangent vectors {Xz ∈ Tz M }z∈M . Namely, setting (9.13) Xz (f ) = X(f )(z), z ∈ M, we easily see that the maps Xz : C ∞ (M ) → R thus deﬁned are R-linear and satisfy the Leibniz rule at any point z, i.e., they are tangent vectors at z. Exercise. Prove this fact. 9.40 Proposition. (Locality of vector ﬁelds.) Let X be a vector ﬁeld on M . If functions f, g ∈ C ∞ (M ) coincide on an open set U ⊂ M , then the functions X(f ), X(g) also coincide on U . In fact, by Lemma 9.8, one has Xz (f ) = Xz (g) for all z ∈ U . Hence, X(f )(z) = X(g)(z) for all z ∈ U . The interpretation of a vector ﬁeld as a family of tangent vectors allows one to consider the section of the tangent bundle sX : M → T M, z → Xz ∈ Tz M ⊂ T M, related to this ﬁeld.

Diﬀerential Calculus as Part of Commutative Algebra

123

Exercise. Prove that sX is a smooth section, and vice versa, any (smooth) section of the tangent bundle is of the form sX . Therefore, vector ﬁelds on M can be understood as sections of the tangent bundle. 9.41. Equality (9.13) shows that the family of tangent vectors {Xz }z∈M generated by X determines this vector ﬁeld uniquely. Using this fact, we can easily understand how vector ﬁelds are described in terms of local coordinates. In fact, if (U, x) is a chart on M and z ∈ U , then Xz , as a tangent vector at the point z ∈ U , can be presented in the form n ∂ Xz = αi (z) (9.14) . ∂xi z i=1 The notation αi (z) underlines the fact that the coordinates of the vector Xz depend on a point z ∈ U ; i.e., they are functions on U . By (9.13) and (9.14), we have n n ∂ ∂f (f ) (z), αi (z) (z) = αi X(f )(z) = Xz (f ) = ∂xi ∂xi i=1 i=1 and consequently X(f ) =

n

αi

∂ (f ). ∂xi

αi

∂ . ∂xi

i=1

Therefore, X=

n i=1

Note that all functions αi belong to C ∞ (U ). In fact, let z ∈ U . Consider V of the a function x ˜i ∈ C ∞ (M ) coinciding with xi in a neighborhood ∞ = αi , since ) ∈ C (M ). Further, X(˜ x ) point z. By deﬁnition, X(˜ x i i V x ˜i V = xi V , and consequently αi ∈ C ∞ (V ). Since z ∈ U is an arbitrary point, we see that αi is a smooth function on U . 9.42. Transformation of vector ﬁelds. Let X be a vector ﬁeld on M , and ϕ : M → N a smooth map. The diﬀerential dz ϕ takes any vector Xz to the tangent vector Yϕ(z) = dz ϕ(Xz ) ∈ Tϕ(z) N . In general, the family of tangent vectors {Yϕ(z) }z∈M does not constitute a vector ﬁeld on N . In fact, if u ∈ N \ ϕ(M ), then the vector Yu is undeﬁned, while a point u ∈ ϕ(M ) may have several inverse images, and thus the vector Yu may be deﬁned ambiguously (see Figure 9.2). So, as a rule, there are no maps of vector ﬁelds corresponding to maps of manifolds. But diﬀeomorphisms are exceptions from the general rule, and if ϕ is a diﬀeomorphism, then its action on a vector ﬁeld X can be deﬁned

124

Chapter 9

?

!!

Figure 9.2. Trying to map a vector ﬁeld.

by the formula Y = (ϕ−1 )∗ ◦ X ◦ ϕ∗ . Exercise.

1. Prove that Y is really a vector ﬁeld.

2. Prove that Yϕ(z) = dz ϕ(Xz ). Below (see Section 9.47) it will be shown that the image of a vector ﬁeld may be deﬁned in a reasonable way, provided that the notion of vector ﬁeld can be adequately generalized. 9.43. The deﬁnition of a vector ﬁeld given above is a particular case of the general algebraic notion of derivation, which is as follows. Let A be a commutative K-algebra. Deﬁnition. A K-linear map Δ : A → A is called a derivation of the algebra A if it satisﬁes the Leibniz rule Δ(ab) = aΔ(b) + bΔ(a)

∀ a, b ∈ A.

Let us denote the set of all derivations of A by D(A). Let Δ, ∇ ∈ D(A) and a ∈ A. Then obviously, Δ + ∇ ∈ D(A) and aΔ ∈ D(A). These operations endow D(A) with a natural A-module structure. Any derivation of the K-algebra A can be understood as a vector ﬁeld on |A|. To see this, it suﬃces to carry over formula (9.13) to the algebraic setting. Let Δ ∈ D(A) and h ∈ |A|. Put Δh = h ◦ Δ : A → K.

Diﬀerential Calculus as Part of Commutative Algebra

125

Then, obviously, the operator Δh , being the composition of two K-linear operators, is also K-linear, and Δh (ab) = h(Δ(ab)) = h(Δ(a)b + aΔ(b)) = h(Δ(a))h(b) + h(a)h(Δ(b)) = Δh (a)h(b) + h(a)Δh (b). Thus Δh ∈ Th (A). In what follows, for brevity we shall write D(M ) instead of D(C ∞ (M )). Let us note that the deﬁnition of a tangent vector Δh written in the form Δh (f ) = h(Δ(f )),

f ∈ A,

is identical to (9.13) if K = R, A = C ∞ (M ), while h = hz ∈ M = |A|, is, as usual, understood as a homomorphism taking f to f (z). If the algebra A is geometrical, then the system {Δh }h∈|A| of vectors tangent to |A| determines the “vector ﬁeld” Δ uniquely. 9.44. Just as in the case of tangent vectors (see Sections 9.33–9.36), one can construct the theory of vector ﬁelds for geometrical objects of a much more general nature than smooth manifolds. For example, using Section 9.33, it is possible to obtain a theory of vector ﬁelds on arbitrary closed subsets of smooth manifolds, just as is done for the manifolds themselves. In the examples below we use the notation introduced for smooth manifolds. The locality of vector ﬁelds (Proposition 9.40) is valid in this more general situation and is proved in the same way. 9.45. Example. Let us describe vector ﬁelds on the cross K using the notation of Section 9.35, where we studied tangent vectors. By Ax and Ay we denote the algebras of smooth functions on the line with ﬁxed coordinate functions x and y, respectively. The natural embeddings ix : Ax → C ∞ (K), iy : Ay → C ∞ (K),

f (x) → (f (x), f (0)), g(y) → (g(0), g(y)),

are deﬁned together with the projections πx : C ∞ (K) → Ax , (f (x), g(y)) → f (x), πy : C ∞ (K) → Ay , (f (x), g(y)) → g(y). Obviously, πx ◦ ix = id and πy ◦ iy = id. Therefore, if Δ ∈ D(K), then Δx = πx ◦ Δ ◦ ix ∈ D(Ax ) and Δy = πy ◦ Δ ◦ iy ∈ D(Ay ). Let us show that (9.15) Δ(f (x), g(y)) = (Δx (f ), Δy (g)) and the ﬁelds Δx , Δy vanish at the point 0. Indeed, let ϕ = ix (f ) = (f (x), c) ∈ C ∞ (K), c = f (0). Consider the point z = (0, y) ∈ K, y = 0. Then a suﬃciently small neighborhood U of the point z is an interval and ϕ|U ≡ c. By the locality of

126

Chapter 9

tangent vectors, one has Δz (ϕ) = 0. In other words, Δ(ϕ)(z) = 0 for all points of the y-axis except for the point (0, 0). By continuity, Δ(ϕ) is zero identically on the whole axis. Therefore, ix (πx (Δ(ϕ))) = Δ(ϕ), or ix (Δx (f )) = Δ(ix (f )). The last equality means that ix ◦ Δx = Δ ◦ ix . In addition, Δx (f )(0) = 0, since Δ(ϕ)(0, 0) = 0. Consequently, the vector ﬁeld Δx ∈ D(Ax ) vanishes at the point (0, 0). In a similar way, iy ◦ Δy = Δ ◦ iy , and the vector ﬁeld Δy ∈ D(Ay ) also vanishes at (0, 0). Note now that (f (x), g(y)) = ix (f ) + iy (g) − (c, c), c = f (0) = g(0), and Δ (c, c) = 0, since functions of the form (c, c) are constants in the algebra C ∞ (K). From this we eventually obtain that Δ (f (x), g(y)) = Δ(ix (f )) + Δ(iy (g)) = ix (Δx (f )) + iy (Δy (g)) = Δx (f ), Δy (g) , where the last equality is a consequence of Δx (f )(0) = Δy (g)(0) = 0.

Obviously, the inverse statement is also valid: Any pair of vector ﬁelds Δx ∈ D(Ax ), Δy ∈ D(Ay ) determines a vector ﬁeld Δ on K by formula (9.15), provided that these ﬁelds vanish at the point (0, 0). Exercise. Describe the A-modules of vector ﬁelds for the following algebras: 1. For all algebras considered in Exercise 9.36. 2. For the R-algebra C m (R1 ), m 1, of m-times diﬀerentiable functions on the line. (Hint: Start from the case m = 0.) 3. For the algebra A = K[X]/X l+1 K[X] of truncated polynomials, where K = R or Z/mZ. 4. For a Boolean algebra, i.e., a commutative algebra over the ﬁeld F2 = Z/2Z whose elements satisfy the relation a2 = a. 9.46. Vector ﬁelds on submanifolds. In the above considerations, we formalized two geometrical images: a manifold at one point of which an arrow “grows” and a manifold on which arrows “grow” at all points. Clearly, an intermediate situation also exists: arrows may grow at points of some submanifold (or, more generally, of a closed subset). Examples of this kind are velocity ﬁelds on a moving thread or on an oscillating membrane. Arguments similar to those that have led us to the deﬁnition of vector ﬁelds on a manifold M lead to the desired formalization in this case also.

Diﬀerential Calculus as Part of Commutative Algebra

127

Figure 9.3. Vector ﬁeld along the submanifold.

Let N ⊂ M be a submanifold of a manifold M . Deﬁnition. An R-linear map X : C ∞ (M ) → C ∞ (N ) is said to be a tangent (to M ) vector ﬁeld along N if X(f g) = X(f )g N + f N X(g).

(9.16)

The sum of two vector ﬁelds along N is obviously a vector ﬁeld along N . One can also deﬁne multiplication of such ﬁelds by functions from C ∞ (M ): def (f X)(g) = f N X(g), f, g ∈ C ∞ (M ). The set D(M, N ) of all tangent ﬁelds along N on the manifold M is a C ∞ (M )-module with respect to these operations. If z ∈ N , then the following analog of formula (9.13), Xz (f ) = X(f )(z),

(9.17)

determines a tangent vector to the manifold M at a point z, corresponding to the vector ﬁeld X along the submanifold N . Note that (9.17) makes no sense when z ∈ / N , and the deﬁnition above really introduces a ﬁeld of vectors along the submanifold N . 9.47. Vector ﬁelds along maps. Relation (9.16) may be also rewritten in the form X(f g) = X(f )i∗ (g) + i∗ (f )X(g), where i : N → M denotes the inclusion map. After this, it becomes clear that it still makes sense, provided that i is an arbitrary map of N to M .

128

Chapter 9

Deﬁnition. Let ϕ : N → M be a smooth map of manifolds. An R-linear map X : C ∞ (M ) → C ∞ (N ) is said to be a tangent (to M ) vector ﬁeld along the map ϕ if X(f g) = X(f )ϕ∗ (g) + ϕ∗ (f )X(g)

∀f, g ∈ C ∞ (M ).

(9.18)

∞

The set Dϕ (M ) of all vector ﬁelds along a given map ϕ is a C (M )-module if the multiplication of a ﬁeld X ∈ Dϕ (M ) by f ∈ C ∞ (M ) is deﬁned by def

the rule (f X)(g) = ϕ∗ (f )X(g). The C ∞ (M )-module Dϕ (M ) also becomes a C ∞ (N )-module if the multiplication of its elements by elements of the algebra C ∞ (N ) is deﬁned by def

(f X)(g) = f X(g),

f ∈ C ∞ (N ), g ∈ C ∞ (M ), X ∈ Dϕ (M ).

Exercise. Check that formula (9.17), in the context under consideration, allows one to assign to any point z ∈ M a tangent vector to M at the point ϕ(z). Any vector ﬁeld X ∈ Dϕ (M ) may be understood as an inﬁnitesimal deformation of the map ϕ. In fact, since Xz ∈ Tϕ(z) M , this vector can be naturally understood as an inﬁnitesimal shift of the image of z under the map ϕ. Vector ﬁelds along maps are also often called relative vector ﬁelds. Example. Let ϕ : N → M be an arbitrary smooth map, X ∈ D(N ), and Y ∈ D(M ). Then X ◦ ϕ∗ and ϕ∗ ◦ Y are vector ﬁelds along the map ϕ. It is appropriate to interpret the relative ﬁeld X ◦ ϕ∗ as the image of the ﬁeld X under the map ϕ (cf. Section 9.42). 9.48. An important example of a relative vector ﬁeld is the universal vector ﬁeld on M . It is constructed in the following way. Consider the tangent bundle πT : T M → M (see Section 9.17). Let ξ be a tangent vector to M , also understood as a point of the manifold T M . The universal vector ﬁeld Z on M is deﬁned as the following vector ﬁeld along the map πT : Z(f )(ξ) = ξ(f ),

f ∈ C ∞ (M ).

9.49 Exercises. 1. Show that Z(f ) ∈ C ∞ (T M ). (Hint: Use special local coordinates on T M .) 2. Check that Z is indeed a vector ﬁeld along πT . 3. Let X ∈ D(M ). Prove that X = s∗X ◦ Z,

(9.19)

where sX : M → T M , z → Xz ∈ Tz M , is the section of the tangent bundle corresponding to the vector ﬁeld X. Formula (9.19) explains

Diﬀerential Calculus as Part of Commutative Algebra

129

why the ﬁeld Z is called universal: Any vector ﬁeld M can be obtained from Z by using the appropriate section. 9.50. The only diﬀerence between the two deﬁnitions of vector ﬁelds discussed above is the interpretation of the Leibniz rule, i.e., the rule for diﬀerentiation of products. Let us rewrite it without specifying the range of the map X: X(f g) = f X(g) + gX(f ).

(9.20)

This formula makes sense when the product of the objects X(g), X(f ) and the functions f, g is deﬁned, or in the other words, when the range of X is a module over its domain. For this reason, the following deﬁnition exhausts everything discussed above in relation to tangent vectors and vector ﬁelds. Deﬁnition. Let A be a commutative K-algebra and let P be an arbitrary A-module. A K-linear map Δ : A → P is called a derivation of the algebra A with values in P if it satisﬁes the Leibniz rule (9.20), i.e., Δ(f g) = f Δ(g) + gΔ(f )

∀ f, g ∈ A.

The set D(P ) of all derivations of the algebra A with values in P carries a natural A-module structure. 9.51. Let X ∈ D(P ), and h : P → Q be an A-module homomorphism. Then h ◦ X ∈ D(Q). (Check this.) Moreover, the map D(h) : D(P ) → D(Q),

D(P ) X → h ◦ X ∈ D(Q),

is obviously an A-module homomorphism, and D(idP ) = idD(P ) , D(h1 ◦ h2 ) = D(h1 ) ◦ D(h2 ). This means that the correspondence P → D(P ) is a functor in the category of A-modules and their homomorphisms. This functor is one of the basic ones of the diﬀerential calculus. Some others will be discussed below. A complete and systematic description of the algebra of these functors together with speciﬁc features of its realization for concrete commutative algebras is the object of diﬀerential calculus in its modern meaning. Therefore, it may be asserted that the construction of the diﬀerential calculus, started by Newton and Leibniz, is not ﬁnished yet; it must be accomplished in the future. 9.52. Let us show how to specify Deﬁnition 9.50 in order to obtain the deﬁnitions of tangent vectors and various vector ﬁelds considered above. We shall also describe the procedures that assign to a vector ﬁeld a tangent vector at a ﬁxed point. In all our considerations here, we assume that K = R.

130

Chapter 9

I. The tangent vector to the manifold M at a point z: A = C ∞ (M ),

P = A/μz = R,

where μz is an ideal of the point z. II. A vector ﬁeld on the manifold M : A = P = C ∞ (M ). The tangent vector Xz ∈ Tz M is assigned to a vector ﬁeld X ∈ D(M ) and a point z ∈ M in the following way: D(A) X → Xz = h ◦ X ∈ D(A/μz ), where h : A → A/μz is the natural projection. In other words, we have Xz = D(h)(X). III. A vector ﬁeld along a submanifold (without boundary) N ⊂ M : A = C ∞ (M ), P = A/μN , where μN = {f ∈ C ∞ (M ) | f N = 0}. Let us consider the natural isomorphism C ∞ (N ) = A/μN . If we are given X ∈ D(M, N ) = D(P ) and z ∈ N , then μz ⊃ μN , and the natural projection h : P = A/μN → A/μz is deﬁned. Also we have Xz = h ◦ X = D(h)(X). IV. A vector ﬁeld along a map ϕ : N → M : A = C ∞ (M ),

P = C ∞ (N ),

ϕ = |F |,

where F : A → P is an R-algebra homomorphism. Note also that the A-module structure in the algebra P = C ∞ (N ) is deﬁned by the rule (f, g) → F (f )g = ϕ∗ (f )g,

f ∈ C ∞ (M ), g ∈ C ∞ (N ).

Let X ∈ Dϕ (M ). If z ∈ N and h : P → Q = P/μz is the natural projection, then Xz = h ◦ X. By the way, at this point let us answer a question that naturally arises: What is a continuous vector ﬁeld on M ? It is an element of the C ∞ (M )-module D(C 0 (M )), where C 0 (M ) is the algebra of continuous functions on M equipped with a natural C ∞ (M )-module structure. We stress that C 0 (M ) is regarded as C ∞ (M )-module, we consider derivations from C ∞ (M ) to C 0 (M ) and not from C 0 (M ) to C 0 (M ). Vector ﬁelds of class C m are deﬁned in a similar way. 9.53. To conclude our discussion of geometrical and algebraic problems related to the notion of vector ﬁeld, let us note that the module D(M ) of vector ﬁelds (or more generally, derivations of the algebra A) carries another important algebraic structure. Namely, D(A) is a Lie algebra due

Diﬀerential Calculus as Part of Commutative Algebra

131

to the following R-linear skew-symmetric operation satisfying the Jacobi identity. def

Proposition. The commutator [X, Y ] = X ◦ Y − Y ◦ X of two derivations X, Y ∈ D(A) is again a derivation. Indeed, [X, Y ](f g) = ( X ◦ Y − Y ◦ X)(f g) = X(f Y (g) + gY (f )) − Y (f X(g) + gX(f )) = X(f )Y (g) + f X(Y (g)) + X(g)Y (f ) + gX(Y (f )) − Y (f )X(g) − f Y (X(g)) − Y (g)X(f ) − gY (X(f )) = f X(Y (g)) − f Y (X(g)) + gX(Y (f )) − gY (X(f )) = f [X, Y ](g) + g[X, Y ](f ). Since the commutator is obviously skew-symmetric, it remains only to check the Jacobi identity. 9.54 Proposition. Let X, Y, Z ∈ D(A). Then [X, [Y, Z]] = [[X, Y ], Z] + [Y, [X, Z]]. Indeed, [[X, Y ], Z] + [Y, [X, Z]] = [[X, Y ], Z] − [[X, Z], Y ] = [X ◦ Y − Y ◦ X, Z] − [X ◦ Z − Z ◦ X, Y ] =X ◦ Y ◦ Z − Y ◦ X ◦ Z − Z ◦ X ◦ Y + Z ◦ Y ◦ X −X ◦Z ◦Y +Z ◦X ◦Y +Y ◦X ◦Z −Y ◦Z ◦X −X ◦Y ◦Z −Z ◦Y ◦X −X ◦Z ◦Y +Y ◦Z ◦X = X ◦ (Y ◦ Z − Z ◦ Y ) − (Y ◦ Z − Z ◦ Y ) ◦ X = X ◦ [Y, Z] − [Y, Z] ◦ X = [X, [Y, Z]]. 9.55. The local coordinate description of vector ﬁelds given in Section 9.41 shows that they are (scalar) ﬁrst-order diﬀerential operators. On the other hand, a ﬁrst-order (scalar) diﬀerential operator Δ of general form on the manifold M can be locally written as Δ=

n i=1

αi

∂ + β, ∂xi

αi , β ∈ C ∞ (U ).

Let us note that its free term β has an invariant meaning: β = Δ(1). Therefore, we can assert that Δ is a ﬁrst-order diﬀerential operator if and only if Δ − Δ(1) is a derivation. This gives us a coordinate-free deﬁnition of (scalar) linear ﬁrst-order diﬀerential operators. But being insuﬃciently “clever,” it does not allow us to guess a similar deﬁnition for operators of

132

Chapter 9

higher orders. Let us trim this deﬁnition: Note that the Leibniz rule for the derivation Δ − Δ(1) is equivalent to the following equality: Δ(f g) − f Δ(g) = gΔ(f ) − f gΔ(1),

or

[Δ, f ](g) = g[Δ, f ](1).

Thus, we have obtained the following result: 9.56 Proposition. An R-linear map Δ : C ∞ (M ) → C ∞ (M ) is a ﬁrst-order diﬀerential operator if and only if [[Δ, f ], g] = 0

∀f, g ∈ C ∞ (M ).

(9.21)

Let us note that (9.21) is equivalent to the fact that the commutator [Δ, f ] is a C ∞ (M )-homomorphism for any f ∈ A. Looking at the last equality, the reader has probably understood already how to deﬁne a diﬀerential operator of any order over an arbitrary commutative algebra A. Before stating this deﬁnition, let us observe that the expression [[Δ, f ], g] is not manifestly symmetric with respect to f and g, while in fact f and g enter this expression symmetrically: [[Δ, f ], g] = Δ ◦ f g + f gΔ − gΔ ◦ f − f Δ ◦ g = [[Δ, g], f ]. Therefore, we shall change our notation and for any element f ∈ A introduce the map δf : HomK (A, A) → HomK (A, A),

def

δf (Δ) = [Δ, f ].

By the above, the operators δf and δg commute, and condition (9.21) acquires the form (δg ◦ δf )(Δ) = 0

∀f, g ∈ A.

We can now give the following fundamental deﬁnition: 9.57. Deﬁnition. Let A be a K-algebra; then a K-homomorphism Δ : A → A is called a linear diﬀerential operator of order l with values in A if for any f0 , . . . , fl ∈ A we have the identity (δf0 ◦ · · · ◦ δfl )(Δ) = 0,

(9.22)

where δf (Δ) = Δ ◦ f − f ◦ Δ. Let us denote the set of all diﬀerential operators of order l acting from A to A by Diﬀ l A. Like D(A), this set is stable with respect to summation and multiplication by elements of the algebra A. Therefore, it is naturally endowed with an A-module structure. Moreover, another A-module structure can be introduced in it, by deﬁning the action of an element f ∈ A on an operator Δ as the composition Δ ◦ f . This structure is called the right one, and the action of an element f ∈ A on an operator Δ will be

Diﬀerential Calculus as Part of Commutative Algebra

133

denoted2 by f > Δ instead of Δ ◦ f . The set Diﬀ l A endowed with the module structure with respect to the right multiplication will be denoted by < Diﬀ > l A, the one with the left multiplication by Diﬀ l A or simply Diﬀ l A. The two multiplicative structures in Diﬀ l A commute and thus determine a bimodule structure, denoted by Diﬀ ♦ l A. 9.58 Exercises. 1. Prove the last statement. Namely, check that the set Diﬀ l A is stable with respect to right multiplication and that the left and right multiplications commute in Diﬀ l A. 2. Check whether the set D(A) is stable with respect to right multiplication. To deduce some natural and useful properties of diﬀerential operators, we shall need the following notation. Let I(n) = (1, 2, . . . , n) be the ordered set of the ﬁrst n integers, and I = (i1 , . . . , il ), l n, J = (j1 , . . . , jr ), r n, ordered subsets of I(n) . Let us set by deﬁnition |I| = l, |J| = r, aI = (ai1 , . . . , ail ), aI = ai1 · . . . · ail , and δaI(n) = δai1 ◦ · · · ◦ δail . The " multiindex obtained by ordering the union I J will be denoted by I + J. Exercise. Let A be a K-algebra, and Δ, ∇ K-linear maps from A to A. Then δaI(n) (Δ ◦ ∇) = δaI (Δ) ◦ δaJ (∇), ai ∈ A, (9.23) I+J=I(n)

δaI(n) (Δ)(b) =

(−1)|I| aI Δ(aJ b),

ai , b ∈ A.

(9.24)

I+J=I(n)

For the case Δ ∈ Diﬀ m A, m < n, the left-hand side of the last equality vanishes by Section 9.57, and the equality can be rewritten in the following form: (−1)|I| aI Δ(aJ b). (9.25) Δ(aI(n) b) = − I+J=I(n) , |I|>0

These formulas allow one to readily prove the following two important statements: 9.59 Proposition. Let ∇ and Δ be linear diﬀerential operators of orders l and m, respectively. Then their composition Δ ◦ ∇ is a linear diﬀerential operator of order l + m. Indeed, let us set n = m+l+1 in formula (9.23). Then each monomial on the right-hand side of the equality thus obtained will vanish by the deﬁnition of diﬀerential operators: Either |I| m + 1 and therefore δaI (Δ) = 0, or |J| l + 1 and, respectively, δaJ (∇) = 0. 2

one.

Sometimes the notation a+ Δ is also used for right structure and aΔ, for the left

134

Chapter 9

9.60 Proposition. Let ν ⊂ A be an arbitrary ideal, a ∈ ν k , Δ ∈ Diﬀ n A, and n < k. Then Δ(a) ∈ ν k−n . To prove the proposition, it suﬃces to conﬁne ourselves to the case a = a1 · · · ak , ai ∈ ν. Let k = n + 1. Consider equality (9.25) with b = 1. Then every summand on the right-hand side will contain at least one element ai ∈ ν and consequently will belong to ν itself. The passage from k = n + r to k = n + r + 1 is accomplished as follows. Let us use formula (9.25) again. Each of the summands on the right-hand side is of the form (9.26) ai1 · · · aim Δ(aj1 · · · ajk−m ). Note that ai1 · · · aim ∈ ν m , aj1 · · · ajk−m ∈ ν k−m . If m k − n, then the monomial 9.26 obviously belongs to ν k−n . Otherwise, if k − m > n, we see that Δ(aj1 · · · ajk−m ) ∈ ν k−m−n by the induction hypothesis, and the monomial (9.26) as a whole belongs to ν k−n . 9.61 Corollary. If functions f and g coincide in some neighborhood U z, then for any diﬀerential operator Δ (in the sense of Deﬁnition 9.57) of the equality Δ(f )(z) = Δ(g)(z) is valid. In other words, diﬀerential operators are local. for any l, Indeed, let Δ be an operator of order l. Since f − g ∈ μl+1 z then by Proposition 9.60 one has Δ(f − g) ∈ μz . Now let us prove that for algebras of smooth functions, Deﬁnition 9.57 coincides with the usual deﬁnition of linear diﬀerential operator. This corollary allows us to obtain, for any diﬀerential operator Δ ∈ Diﬀ l C ∞ (M ), its well-deﬁned restriction ΔU : C ∞ (U ) → C ∞ (U ) to any open domain U ⊂ M by setting ΔU (f )(z) = Δ(g)(z), f ∈ C ∞ (U ), g ∈ C ∞ (M ), z ∈ U, where g is an arbitrary function coinciding with f in some neighborhood of the point z. This deﬁnition implies ΔU (f U ) = Δ(f )U for f ∈ C ∞ (M ). Obviously, any operator is uniquely determined by its restrictions on charts of an arbitrary atlas. Now we can prove the following important statement: 9.62 Theorem. Let Δ ∈ Diﬀ l (C ∞ (M )), and let x1 , . .. , xn be local coordinates in a neighborhood U ⊂ M . Then the operator ΔU can be presented in the form (for the notation, see Section 2.8) l ∂ |σ| ΔU = ασ σ , ∂x

ασ ∈ C ∞ (U ).

|σ|=0

∞

Let z ∈ U and f ∈ C (M ). Consider an arbitrary star-shaped neighborhood Uz ⊂ U of the point z and, using Section 2.9, present the function f

Diﬀerential Calculus as Part of Commutative Algebra

135

in this neighborhood in the form f=

l (x − z)σ ∂ |σ| f (z) + h(x), ∂xσ σ!

|σ|=0

def

and (x − z)σ = (x1 − a1 )σ1 · · · (xn − an )σn . Therefore, where h(x) ∈ μl+1 z l ∂ |σ| f Δ(f )(z) = Δ U (f U )(z) = (z)ασ (z), ∂xσ |σ|=0

where def ασ (x) = ΔU

(x − z)σ σ!

and ασ (z) = ΔU

(x − z)σ σ!

. z

It remains to note that the functions ασ (x) are smooth by construction. To understand how the algebraic Deﬁnition 9.57 of diﬀerential operators works for the case in which the algebra A is not the smooth function algebra on a smooth manifold, let us do the following exercise: 9.63 Exercises. 1. Describe the modules of diﬀerential operators for the algebra C ∞ (K). (See Examples 9.35 and 9.45.) 2. Do the same for the algebra of truncated polynomials A = K[X]/X n K[X],

where

K = R, K = Zm .

(See Exercise 3 from Section 9.45.) 3. In the classical situation A = C ∞ (R), any diﬀerential operator of order > 1 may be represented as the sum of compositions of ﬁrstorder operators. May one assert the same thing for the algebras from the previous exercises? 9.64. Jets of order l at a point. Let us formulate an important consedenotes the (l + 1)st power of quence of Proposition 9.60. Recall that μl+1 z the ideal μz consisting of all functions on M vanishing at the point z. Corollary. Let Δ ∈ Diﬀ l C ∞ (M ), f, g ∈ C ∞ (M ), and z ∈ M . Then Δ(f )(z) = Δ(g)(z) if f = g mod μl+1 z . Indeed, in this case f − g ∈ μl+1 z , and consequently, Δ(f − g) ∈ μz , i.e., Δ(f − g)(z) = 0. It will be useful to consider this fact after introducing the vector space of lth order jets, or l-jets, of (smooth) functions on M at some point z (cf. Section 9.27): Jzl M = C ∞ (M )/μl+1 z . def

136

Chapter 9

The image of the function f under the natural projection C ∞ (M ) → C ∞ (M )/μl+1 = Jzl M z is called its jet of order l (or l-jet) at the point z and is denoted by [f ]lz . In means that [f ]lz = [g]lz , while these terms, the condition f = g mod μl+1 z the previous corollary asserts that Δ(f )(z) = Δ(g)(z) if [f ]lz = [g]lz . In other words, the map hΔ,z : Jzl M → R,

[f ]lz → Δ(f )(z),

is well deﬁned. It is obviously R-linear. Its importance is explained by the fact that it completely determines the operator Δ at the point z. Exercise. The map hΔ,z is a linear function on the space Jzl M . Find a basis of the space Jzl M in which the components of this function are the numbers ασ (z) appearing in Theorem 9.62. 9.65. The manifold of jets. The family {hΔ,z }z∈M of linear functionals uniquely determines the operator Δ, since Δ(f )(z) = hΔ,z ([f ]lz ).

(9.27)

Therefore, it makes sense to construct a new object combining the separate maps hΔ,z into a single whole. To do this, one ﬁrst needs to join their domains, in the same way as was done in Section 9.28 for l = 1: Jzl M. J lM = z∈M l

The set J M is equipped with a smooth manifold structure by a procedure similar to that used for T M and T ∗ M . The details of this construction will be described in Section 11.11. This smooth manifold is called the manifold of jets of order l (or of l-jets) of the manifold M . The map πJ l = πJ l M : J l M → M,

J l M ⊃ Jzl M θ → z ∈ M,

l ﬁbers the manifold J l M over M . By this deﬁnition, πJ−1 l (z) = Jz M . ∞ Moreover, to any function f ∈ C (M ), we can assign the section

sjl (f ) : M → J l M,

z → [f ]lz ∈ Jzl M ⊂ J l M,

of this bundle. This section is called the l-jet of f . Any operator Δ ∈ Diﬀ l C ∞ (M ) determines the map hΔ : J l M → M × R,

Jzl M θ → (z, hΔ,z (θ)).

Let πR : M × R → R be the canonical projection. Then, by (9.27), πR (hΔ ([f ]lz )) = Δ(f )(z), and consequently, Δ(f ) = πR ◦ hΔ ◦ sjl (f ) ,

(9.28)

Diﬀerential Calculus as Part of Commutative Algebra

137

where Δ(f ) is understood as a smooth map from M to R. This relation shows that all the information on the operator Δ is encoded in the map of smooth manifolds hΔ . It will be shown in Section 14.15 that sjl : f → sjl (f ) is a diﬀerential operator of order l, whose range of values is the set of sections of the bundle πJ l . It is natural to call this operator the universal diﬀerential operator of order l, since all concrete operators are obtained by composing this operator with maps from J l M to M × R. The speciﬁcs of maps of the form hΔ acting from J l M to M × R can be described in the following way. Consider the projection π : M × R → M,

(z, λ) → z.

It is a trivial bundle over M with ﬁber R (see Section 12.2). Then, as is easily seen, maps of the form hΔ are morphisms of the vector bundle πJ l to the vector bundle π (see Section 12.4), and in particular, they take the l −1 (z) = R. This map of ﬁbers obviously ﬁber πJ−1 l (z) = Jz M to the ﬁber π coincides with hΔ,z . All these facts reveal the fundamental role of vector bundles in the diﬀerential calculus. For this and many other reasons (some of them will appear in our subsequent exposition), the theory of vector bundles is a necessary part of the diﬀerential calculus over smooth manifolds. This theory will be considered in detail in Chapter 12. Due to the universality of the operator sjl expressed by formula (9.28), the manifolds J l M and their natural generalizations constitute an important part of the foundations of the modern theory of partial diﬀerential equations. The universal property of this operator is also revealed by the fact that the module of sections of the bundle πJ l : J l M → M (see Section 12.7) is the representing object for the functor Diﬀ l of the diﬀerential calculus in the category of geometrical C ∞ (M )-modules (see Section 12.44). 9.66. It was shown above for the case in which A = C ∞ (M ) is the algebra of smooth functions on the manifold M that Deﬁnition 9.57 is equivalent to the usual deﬁnition of a linear diﬀerential operator acting on functions and taking its values in functions on M (i.e., to the deﬁnition of scalar diﬀerential operators). In fact, the more general case, that of matrix differential operators, can also be described in purely algebraic terms. Let us recall that such an operator Δ is usually deﬁned as a matrix composed of diﬀerential operators, ⎞ ⎛ Δ1,1 . . . Δ1,m ⎜ .. ⎟ , .. Δ = ⎝ ... . . ⎠ Δk,1

...

Δk,m

138

Chapter 9

where Δi,j are scalar diﬀerential operators of order l, while the action of this operator on a vector function f¯ = (f1 , . . . , fn ) is deﬁned in the following natural way: ⎞⎛ ⎞ ⎛ ⎛ ⎞ f1 Δ1,1 . . . Δ1,m Δ1,1 (f1 ) + . . . + Δ1,m (fm ) ⎜ .. ⎟ .. ⎟ ⎜ .. ⎟ = ⎜ .. .. ⎝ . ⎠. . . ⎠⎝ . ⎠ ⎝ . Δk,1

...

Δk,m

fm

Δk,1 (f1 ) + . . . + Δk,m (fm )

In Chapter 12, it will be shown that vector functions of the above type can naturally be considered as sections of an m-dimensional vector bundle over M and that the category of all vector bundles over M is equivalent to the category of projective modules over the algebra C ∞ (M ). This fact, together with the observations of Section 9.50, leads one to believe that diﬀerential operators of general nature should be maps connecting modules over some base algebra A. It is remarkable that to deﬁne a general diﬀerential operator it suﬃces simply to repeat the scalar Deﬁnition 9.57. The only thing that matters here is that for any K-linear map of A-modules Δ : P → Q and any a ∈ A one can deﬁne the commutator def

δa (Δ) = [Δ, a] : P → Q, where the element a ∈ A is understood as the operator of multiplication by a applied to elements of the corresponding A-module. In other words, δa (Δ)(p) = Δ(ap) − aΔ(p),

p ∈ P.

So, one can hope that the following purely algebraic deﬁnition reduces to the usual notion of a (matrix) diﬀerential operator in the “standard” situation. 9.67. Deﬁnition. Let A be an arbitrary commutative K-algebra, and let P and Q be A-modules. A K-homomorphism Δ : P → Q is called a linear diﬀerential operator of order l acting from P to Q if for any a0 , . . . , al ∈ A one has (δa0 ◦ · · · ◦ δal )(Δ) = 0.

(9.29) ∞

The fact that under an adequate specialization (K = R, A = C (M ) with projective A-modules P and Q) the deﬁnition given above coincides with the usual one will be proved in Chapter 12, after we have established relations between vector bundles and projective modules. Let us denote the set of all diﬀerential operators of order l acting from P to Q by Diﬀ l (P, Q). This set is stable with respect to summation and to the ordinary (left) multiplication by elements of the algebra A: def

(aΔ)(p) = a · Δ(p),

a ∈ A, p ∈ P.

Therefore, it possesses a natural left A-module structure. One can also introduce another A-module structure, deﬁning the action of an element a ∈ A on the operator Δ as the composition Δ ◦ a. This structure is

Diﬀerential Calculus as Part of Commutative Algebra

139

called right, and the action of a ∈ A to Δ is denoted by a> Δ instead of Δ ◦ a. The set Diﬀ l (P, Q), as a module with respect to the right multiplication, will be denoted by Diﬀ > l (P, Q). Two multiplicative structures in Diﬀ l (P, Q) commute and thus determine the structure of a bimodule, ♦ denoted by Diﬀ ♦ l (P, Q). For the sake of brevity, we use the notation Diﬀ l Q for Diﬀ ♦ l (A, Q). If h : P → Q is an A-module homomorphism, then the correspondence Δ → h ◦ Δ, Δ ∈ Diﬀ l P , determines a homomorphism of the A-module Diﬀ l P to the A-module Diﬀ l Q. Therefore, the correspondence P → Diﬀ l P is a functor on the category of A-modules. Let us denote this functor by Diﬀ l . We obtain another example of a functor of the diﬀerential calculus (see Section 9.51). Such functors are deﬁned for all commutative unital algebras. If we choose an A-module P , we obtain an example of what may be called a relative functor of diﬀerential calculus, Diﬀ l (P, ·) : Q → Diﬀ l (P, Q). Formulas (9.23)–(9.25) and Proposition 9.59 are proved in the general situation exactly in the same way as for scalar operators. As to Proposition 9.60, its analog in the general case is the following. Proposition. Let I ⊂ A be an ideal, let P, Q be A-modules, p ∈ I k P , Δ ∈ Diﬀ n (P, Q), and n < k. Then Δ(p) ∈ I k−n Q. The proof is the same as in the scalar case. Exercises. 1. Check that Diﬀ 0 (P, Q) = Diﬀ > 0 (P, Q) = HomK (P, Q). 2. Consider the maps i> and i< of A-modules that are the identities on the underlying sets: i> : Diﬀ l (P, Q) → Diﬀ > i> (f ) = f, l (P, Q), i< : Diﬀ > l (P, Q) → Diﬀ l (P, Q),

i< (f ) = f.

Prove that these maps are diﬀerential operators of order l. 9.68. Let us note that any diﬀerential operator of order l is an operator of order m as well, provided l m. Therefore, we have a natural bimodule ♦ embedding Diﬀ ♦ l (P, Q) ⊂ Diﬀ m (P, Q). Let us denote the direct limit of the sequence of embeddings ♦ ♦ Diﬀ ♦ 0 (P, Q) ⊂ · · · ⊂ Diﬀ l (P, Q) ⊂ Diﬀ l+1 (P, Q) ⊂ · · ·

by Diﬀ ♦ (P, Q). As we saw above, the composition of two diﬀerential operators, if it is deﬁned, is again a diﬀerential operator. Therefore, the bimodule Diﬀ ♦ (P, P ) becomes a left (noncommutative) A-algebra with respect to the operation a(Δ∇) = (aΔ)∇. ♦

Moreover, Diﬀ (P, Q) can be regarded as a right Diﬀ ♦ (P, P )- and left Diﬀ ♦ (Q, Q)-module.

10 Symbols and the Hamiltonian Formalism

10.1. We now have everything needed to answer the question stated in Section 9.20: What is the algebra whose spectrum is the cotangent manifold T ∗ M ? Let us start with necessary algebraic deﬁnitions. Let K be a ﬁeld and let A be a K-algebra. The embedding of A-modules Diﬀ k−1 A ⊂ Diﬀ k A (see Section 9.68) allows us to deﬁne the quotient module def

Sk (A) = Diﬀ k A/ Diﬀ k−1 A, which is called the module of symbols of order k (or the module of k-symbols). The coset of an operator Δ ∈ Diﬀ k A modulo Diﬀ k−1 A will be denoted by smblk Δ and called the symbol of Δ. Let us deﬁne the algebra of symbols S∗ (A) for the algebra A by setting S∗ (A) =

∞ Sn (A). n=0

The operation of multiplication in S∗ (A) is induced by the composition of diﬀerential operators. To be more precise, for two elements smbll Δ ∈ Sl (A),

smblk ∇ ∈ Sk (A)

let us set by deﬁnition def

smbll Δ · smblk ∇ = smblk+l (Δ ◦ ∇) ∈ Sl+k (A). This operation is well deﬁned, since the result does not depend on the choice of representatives in the cosets smbll Δ and smblk ∇. Indeed, if, say, © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 10

141

142

Chapter 10

smbll Δ = smbll Δ , then Δ − Δ ∈ Diﬀ l−1 A and consequently (Δ − Δ ) ◦ ∇ ∈ Diﬀ l+k−1 A. Proposition. S∗ (A) is a commutative algebra. We must check that Δ ◦ ∇ − ∇ ◦ Δ = [Δ, ∇] ∈ Diﬀ l+k−1 A if Δ ∈ Diﬀ l A and ∇ ∈ Diﬀ k A. Let us use induction on l + k. In the case l + k = 0, i.e., for l = k = 0, the statement is obvious, since scalar diﬀerential operators of order zero are the operators of multiplication by elements of the algebra A, and this algebra is commutative. The induction step from l + k < n to l + k = n is based on the following formula, which is a particular case of (9.23): δa (Δ ◦ ∇ − ∇ ◦ Δ) = δa (Δ) ◦ ∇ + Δ ◦ δa (∇) − δa (∇) ◦ Δ − ∇ ◦ δa (Δ) = [δa (Δ), ∇] + [Δ, δa (∇)]. The orders of the operators δa (Δ) and δa (∇) are l−1 and k−1, respectively. By the induction hypothesis, the last expression is an operator of order k +l −2. Hence, the order of the operator [Δ, ∇] does not exceed l +k −1. Let us note that S0 (A) = A is a subalgebra of the algebra S∗ (A), and the operations of left (right) multiplication of diﬀerential operators by elements of the algebra A reduce to the left (right) multiplication by elements of this subalgebra. By the commutativity of the algebra S∗ (A), these multiplication operations coincide. 10.2. Now let smbll Δ ∈ Sl (A) and smblk ∇ ∈ Sk (A). Then, by the last proposition, [Δ, ∇] ∈ Diﬀ l+k−1 A. One can assign to the pair (smbll Δ, smblk ∇) the element def

{smbll Δ, smblk ∇} = smblk+l−1 [Δ, ∇] ∈ Sk+l−1 (A), which is well deﬁned, i.e., does not depend on the choice of representatives in the cosets smbll Δ and smblk ∇ (this is proved exactly in the same way as we proved that the multiplication in S∗ (A) is well deﬁned). The operation { · , · } is K-linear and skew-symmetric. It satisﬁes the Jacobi identity, since the commutator of linear diﬀerential operators satisﬁes this identity. Thus, S∗ (A) is a Lie algebra with respect to this operation. If s1 , s2 ∈ S1 (A), then {s1 , s2 } ∈ S1 (A) as well. In other words, S1 (A) ⊂ S∗ (A) is a Lie subalgebra of the Lie algebra of symbols S∗ (A). Exercises.

1. Let s = smbl1 Δ. Prove that the correspondence s ↔ Δ − Δ(1) ∈ D(A)

is well deﬁned and establishes an isomorphism between the Lie algebras S1 (A) and D(A).

Symbols and the Hamiltonian Formalism

143

2. Fix an arbitrary element s ∈ S∗ (A). Show that the map {s, · } : S∗ (A) → S∗ (A),

s1 → {s, s1 },

is a derivation of the algebra S∗ (A). 3. Let a1 , . . . , an ∈ A and Δ ∈ Diﬀ n A. Show that the map smbln (Δ) : A ⊗ . . . ⊗ A → A given by smbln (Δ) (a1 , . . . , an ) → δa1 ,...,an (Δ) is well deﬁned, depends only on the symbol of the operator Δ, and is linear in each argument. Further, instead of smbln (Δ) (a1 , . . . , an ), we write [a1 , . . . , an ]smbln (Δ) . 10.3. Assume that the ring K is an algebra over the ﬁeld of rational numbers Q. Then any element a ∈ A determines a K-algebra homomorphism Ξa : S∗ (A) → A,

smblk (Δ) →

[δak (Δ)] (1). k!

Let us check this fact. Note ﬁrst that δak (∇) = 0 if ∇ ∈ Diﬀ k−1 A. Therefore, the map Ξa is well deﬁned. Its K-linearity is obvious. Further, Ξa S (A) : S0 (A) = A → A is the identity map, and hence Ξa (1S∗ (A) ) = 1A 0 (unitarity!). Finally, if Δ ∈ Diﬀ k A, ∇ ∈ Diﬀ l A, then from (9.23) it follows that k+l k δak+l (Δ ◦ ∇) = δa (Δ) ◦ δal (∇). (10.1) k l Since δak (Δ) ∈ Diﬀ 0 A = A and δa (∇) ∈ Diﬀ 0 A = A are operators of multiplication by the elements δak (Δ) (1) and δal (∇) (1) of the algebra A, the multiplicativity of the map Ξa is a direct consequence of (10.1).

Proposition. Let I ⊂ A be an ideal, and Π : A → A/I the natural projection. Then Π ◦ Ξa = 0 if a ∈ I 2 . From formula (9.24) it follows that k

k k i k−i . δa (Δ) (1) = aΔ a i i=0

Therefore, if a ∈ I 2 , then k δa (Δ) (1) = Δ ak

mod I 2 .

If, in addition, Δ ∈ Diﬀ k A, then, by Proposition 9.60, Δ ak ∈ I k , k 1, and consequently Π(Ξa (smblk Δ)) = 0. For k = 0, the assertion is obvious.

144

Chapter 10

10.4. We can now describe the K-spectrum |S∗ (A)| of the algebra S∗ (A). Let h ∈ |A|. Then the composition def

γh,a = h ◦ Ξa : S∗ (A) → K is a K-algebra homomorphism and thus is a point of the spectrum |S∗ (A)|. Corollary. Let μh = Ker h and a − h(a) · 1A = b − h(b) · 1A

mod μ2h .

Then γh,a = γh,b . Note ﬁrst that δλ·1A = 0 for λ ∈ K. Therefore, Ξa = Ξa−h(a)·1A , Ξb = Ξb−h(b)·1A , and we may conﬁne ourselves to the case h(a) = h(b) = 0, which is equivalent to a, b ∈ μh . Further, by our assumptions, we have a = a − b ∈ μ2h . Since k

k s k−s k k (Δ)), δb (Δ) = δa+a (Δ) = δ (δ s a a s=0 by setting Δs =

k! δ k−s (Δ) (k − s)! a

and assuming that Δ ∈ Diﬀ k A, we ﬁnd that 1 k δa+a (Δ) (1) Ξb (smblk Δ) = k! k

1 k 1 s = δa (Δ) (1) + [δa (Δs )] (1) k! s! s=1 =Ξa (smblk Δ) +

k

Ξa (smbls (Δs )).

s=1

Therefore, Proposition 10.3 implies that h ◦ Ξb = h ◦ Ξa . Let us recall that by deﬁnition (see Section 9.30) the cotangent space to def

the spectrum |A| at a point h ∈ |A| is the quotient module Th∗ (A) = μh /μ2h . Corollary 10.4 makes it possible to construct a map ih : Th∗ (A) → |S∗ (A)|, by setting ih ([a]) = γh,a , a ∈ μh . In other words, to any “cotangent vector” to the “manifold” |A| there corresponds a point of the “manifold” |S∗ (A)|. Let us study this correspondence in more detail. 10.5 Proposition. Let K be a ﬁeld and assume that any tangent vector ξ ∈ Th A can be continued to a “vector ﬁeld ” X ∈ D(A); i.e., for any

Symbols and the Hamiltonian Formalism

145

ξ ∈ Th A there exists a derivation X ∈ D(A) such that ξ = h ◦ X. Then the map ih is injective. By (9.11), to any K-linear map ϕ : Th∗ A → K there corresponds a tangent vector ξϕ = νh (ϕ) ∈ Th A. Let now a, b ∈ μh and [a] = [b], where [g] = g mod μ2h . Since K is a ﬁeld, by Section 9.30, νh is an isomorphism and consequently ξϕ (a) = ξϕ (b). Let us continue the tangent vector ξϕ to a vector ﬁeld X ∈ D(A). Then the above inequality can be interpreted as h(X(a)) = h(X(b)). Now identifying D(A) with S1 (A) (see Exercise 1 from Section 10.2), we see that X(a) = Ξa (X), X(b) = Ξb (X), and thus the last inequality can be rewritten in the form (h ◦ Ξa )(X) = (h ◦ Ξb )(X). Hence we have γh,a = γh,b , which is equivalent to the desired inequality ih ([a]) = ih ([b]). Obviously, the assumptions of the proposition proved above hold for the algebra A = C ∞ (M ). Therefore, setting iz = ihz for a point z ∈ M , we obtain the following: 10.6 Corollary. The map iz : Tz∗ M → |S∗ (C ∞ (M ))| is injective.

Combining the maps iz for all points z ∈ M , we obtain the embedding i : T ∗ M → |S∗ (C ∞ (M ))|, iT M = iz . (10.2) z

10.7. Let us discuss some other facts useful in our subsequent study of the K-spectrum of the algebra S∗ (A). Suppose that h ∈ |S∗ (A)|. Let us identify A with S0 (A). Then obviously def h = h ∈ |A|. A

hA = h. Check that the projection Exercise. Show that h ∈ ih implies πT ∗ : T ∗ M → M is the geometrical analog of the map h → h in the case h = hθ , θ ∈ T ∗ M , then h = hz , where A = C ∞ (M ). (In other words, if z = πT ∗ (θ).)

Note now that if a ∈ A and X ∈ D(A) = S1 (A), then h(aX) = h(a) h(X). In particular, h(aX) = 0 if a ∈ μh . Therefore, the map of K-modules h : D(A)/μh D(A) → K,

h(X mod μh D(A)) = h(X),

is well deﬁned. On the other hand, we have the natural map τh : D(A)/μh D(A) → Th A,

X mod μh D(A) → h ◦ X.

Lemma. If A = C ∞ (M ), then the map τh deﬁned above is an isomorphism of vector spaces over R. From the spectrum theorem, Theorem 7.7, it follows that h = hz for some point z ∈ M , and consequently h ◦ X = Xz (see Section 9.52, II).

146

Chapter 10

Since any tangent vector ξ ∈ Th A = Tz M can obviously be continued to a vector ﬁeld on M , τh is a surjective map. Now the injectivity of τh means that Xz = 0 implies X ∈ μh D(A) = μz D(M ). This implication is easily proved using the following fact: if X=

n

αi (x)

i=1

∂ ∂xi

in a local coordinate system, then αi (z) = 0, i.e., the coeﬃcients αi belong to μz . The following fact may be used to rigorously complete the proof of the above lemma. Exercise. Let X ∈ D(M ) be such that Xz = 0, z ∈ M . Prove that n

X= fi Xi + Y, fi ∈ C ∞ (M ), Y, Xi ∈ D(M ), i=1

where fi (respectively, Xi ) coincides with αi (respectively, ∂/∂xi ), i = 1, . . . , n, in a neighborhood of z, while Y vanishes in this neighborhood. We can now completely describe the R-spectrum of the algebra of symbols S∗ (C ∞ (M )). 10.8 Theorem. The map i : T ∗ M → |S∗ (C ∞ (M ))| is an isomorphism; i.e., T ∗ M is the R-spectrum of the algebra S∗ (C ∞ (M )). The injectivity of the map i was proved in Corollary 10.6. Let us prove its surjectivity. Suppose that, in the notation of Section 10.7, we have h ∈ |S∗ (A)|, h = hA , and let z ∈ M be a point such that A = C ∞ (M ), h = hz . Then Th A = Tz M , and by Lemma 10.7 the map h can be understood as an R-linear map from Tz M to R, i.e., as a covector dz f ∈ T ∗ M (see Section 9.22). By the deﬁnition of h, we see that γhz ,f (X) = (hz ◦ Ξf )(X) = (hz ◦ X)(f ) = Xz (f ) = dz f (Xz ) = h(Xz ) = h(X). Thus, hS (A) = γhz ,f S (A) . The following lemma, whose assumptions hold 1 1 for the algebra C ∞ (M ) because the partition of unity lemma (4.18) applies, shows that h lies in the image of the map i. 10.9 Lemma. Let a K-algebra A be such that for all natural numbers l any diﬀerential operator of order l is representable as the sum of monomials of the form X1 ◦ · · · ◦ Xs , where Xi ∈ D(A), s l. In this case, if h1 , h2 ∈ |S∗ (A)|, h1 = h2 and h1 = h2 , A

h2 . then h1 =

A

S1 (A)

S1 (A)

Symbols and the Hamiltonian Formalism

147

Passing to symbols of diﬀerential operators, we see that the algebra S∗ (A) is generated by its submodule S1 (A) = D(A); i.e., any symbol s = smblk (δ) can be represented as the sum of monomials s1 · · · sk , where si = smbl1 Xj , Xj ∈ D(A). Since hi is an algebra homomorphism, we have hi (s1 · · · sk ) = hi (s1 ) · · · hi (sk ) and the required equality h1 (s) = h2 (s) follows from the fact that by our assumptions, h1 (s) = h2 (s) for all symbols s ∈ D(A) = S1 (A). 10.10 Exercise. Describe the R-spectrum of the algebra of symbols S∗ (C ∞ (K)) on the cross. Use a reasonable modiﬁcation of the constructions that allowed us to describe the spectrum of the algebra S∗ (C ∞ (M )). By Proposition 10.3, this spectrum can be naturally interpreted as the cotangent space for the cross.

10.11. The algebra of symbols in coordinates. Theorem 10.8 allows one to understand elements of the algebra S∗ (C ∞ (M )) as functions on T ∗ M . Let us describe this interpretation in special coordinates (see Section 9.24). Let U be a local chart on M . Then, by Section 9.24, T ∗ U is a local chart on T ∗ M . The localization of diﬀerential operators deﬁned on M to the domain U naturally generates the corresponding localization of the algebra S∗ (C ∞ (M )). This localization clearly coincides with S∗ (C ∞ (U )). Therefore, we can restrict ourselves to the interpretation of its elements as functions on T ∗ U . We shall use the notation of Section 10.8. Let θ = dz f ∈ T ∗ U , f ∈ C ∞ (U ), and Δ ∈ Diﬀ k C ∞ (M ). Denote by s = sΔ the function on T ∗ M corresponding to the symbol smblk Δ. Then, by deﬁnition, s(θ) = γhz ,f (smblk (Δ)). Let us identify, as above, a vector ﬁeld X ∈ D(U ) with its symbol. It was noted in Section 10.8 that γhz ,f (X) = Xz (f ). Therefore, s(θ) = Xz (f ). In particular, if X = ∂/∂xi , then s(θ) = ∂f /∂xi (z). Recall that by the deﬁnition of special coordinates (x, p) in T ∗ U , the coordinate pi (θ) is the ith component of the covector θ in the basis {dz xi } in Tz∗ M . Since θ = dz f in our case, pi (θ) = ∂f /∂xi (z) and consequently, s∂/∂xi = pi . Note further that the deﬁnition smblk (Δ) · smbll (∇) = smblk+l (Δ ◦ ∇),

148

Chapter 10

for Δ ∈ Diﬀ k C ∞ (M ), ∇ ∈ Diﬀ l C ∞ (M ), implies s(Δ◦∇) = sΔ · s∇ . Therefore, if Δ = |σ|h aσ ∂ |σ| /∂xσ , then

sΔ = aσ pσ , where pσ = pi11 · · · pinn if σ = (i1 , . . . , in ). |σ|=k

Thus the algebra S∗ (C ∞ (U )) is isomorphic to the algebra of ﬁberwise polynomials in the variables p1 , . . . , pn with coeﬃcients in the algebra C ∞ (U ). From this we immediately obtain the following: 10.12 Proposition. The algebra S∗ (C ∞ (M )) is isomorphic to the subalgebra of the algebra C ∞ (T ∗ M ) consisting of functions whose restrictions to the ﬁbers Tz∗ M of the cotangent bundle are polynomials. The algebra C ∞ (T ∗ M ) is isomorphic to the smooth closure of the algebra S∗ (C ∞ (M )). Exercise. 1. Prove that the restriction of s∗df : C ∞ (T ∗ M ) → C ∞ (M ) to the subalgebra S∗ (C ∞ (M )) coincides with the map Ξf : C ∞ (T ∗ M ) → C ∞ (M ) (see Section 10.3). 2. Let A be a K-algebra, h ∈ |A|, and S+ (A) = i>0 Si (A). Then the map ¯ : S∗ (A) → K, h ¯ = h, h ¯ h = 0, A S (A) +

¯ ∈ |S∗ (A)|. Show that when is a K-algebra homomorphism, i.e., h ∞ ¯ coincides with sdf for f ≡ 0 A = C (M ), the map |A| → |S∗ (A)|, h → h, ∗ (the canonical embedding of M into T M ). 3. Find an analog of the map sdf for arbitrary K-algebras. 4. Describe the algebra of symbols S∗ (C ∞ (K)) and realize it as the algebra of functions on the spectrum |S∗ (C ∞ (K))|. 10.13. The Hamiltonian formalism in T ∗ M and |S∗ (A)|. Now consider the case A = C ∞ (M ); let us describe, in special coordinates, the bracket {·, ·} introduced in Section 10.2. By the skew-symmetry of this bracket and because of the relation {s, s1 s2 } = {s, s1 }s2 + s1 {s, s2 }

(10.3)

(see the exercise from Section 10.2), it suﬃces to compute this bracket for the coordinate functions. Using the notation of the previous section, we have by deﬁnition {sΔ , s∇ } = s[Δ,∇] . ∞

Since [f, g] = [g, f ] = 0 for f, g ∈ C (U ),

∂

∂ ∂ ∂f = 0, , ,f = , ∂xi ∂xj ∂xi ∂xi

(10.4)

Symbols and the Hamiltonian Formalism

149

it follows by (10.4) that {f (x), g(x)} = 0,

{pi , pj } = 0,

{pi , f (x)} =

∂f (x) . ∂xi

(10.5)

Further, applying (10.3) to F = f (x)pσ , G = g(x)pτ , we obtain {F, G} = {f (x)pσ , g(x)pτ } = {f (x), g(x)}pσ pτ + {f (x), pτ }g(x)pσ + {pσ , g(x)}f (x)pτ + {pσ , pτ }f (x)g(x) = {pσ , g(x)}f (x)pτ − {pτ , f (x)}g(x)pσ , where pσ = pi1 . . . pin , pτ = pj1 . . . pin . By the last equality in (10.5), {pσ , g(x)} =

n

∂pσ ∂g , ∂pi ∂xi i=1

{pτ , f (x)} =

n

∂pτ ∂f . ∂pi ∂xi i=1

As a result of these computations, we ﬁnally obtain the formula: n

∂F ∂G ∂G ∂F {F, G} = − , ∂pi ∂xi ∂pi ∂xi i=1

(10.6)

which is the standard Poisson bracket on T ∗ M . Moreover, this formula def shows that the derivation XF = {F, ·} of the algebra of symbols (which is, geometrically, a vector ﬁeld on its R-spectrum, i.e., on T ∗ M ) is of the form n

∂F ∂ ∂F ∂ XF = − . (10.7) ∂pi ∂xi ∂xi ∂pi i=1 Thus XF is the Hamiltonian vector ﬁeld on T ∗ M with Hamiltonian F . These facts justify calling the bracket {·, ·} on the algebra of symbols S∗ (A) of an arbitrary K-algebra A the Poisson bracket, while the derivations {s, ·}, s ∈ S∗ (A), are naturally called Hamiltonian vector ﬁelds on |S∗ (A)|. This is another evidence in favor of treating diﬀerential calculus as a part of commutative algebra (this treatment being a consequence of the observability principle). The reader can now enjoy constructing Hamiltonian mechanics on smooth sets or over arithmetic ﬁelds. Exercise. 1. Let F = F (x, p) ∈ C ∞ (T ∗ M ). Check that ﬁnding functions satisfying the condition s∗df (F ) = 0, f ∈ C ∞ (M ), is equivalent to solving the Hamilton–Jacobi equations. Find the analog of these equations for an arbitrary K-algebra A. 2. Find the analog of the Hamilton–Jacobi equations on the cross K. 10.14. Thus we see that the diﬀerential calculus is a consequence of the classical observability principle and is developed simply and naturally if one keeps this fact in mind. The commutativity of the algebra of observables is a formalization of the fundamental idea of classical physics: the independence of observations. A more sophisticated realization of this idea by

150

Chapter 10

means of commutative graded algebras (traditionally called superalgebras) does not involve overcoming any additional diﬃculties. All deﬁnitions and constructions of the diﬀerential calculus are carried over to this case and to any other case in which commutativity can be treated in a reasonable way. This is done in detail in Chapter 21. As is known, in quantum physics, one has to reject the principle of independence of observations. Nevertheless, it would not be correct to try to quantize the diﬀerential calculus in order to describe quantum phenomena by a simple change from commutative algebras to noncommutative ones. The reader will see this by trying systematically to carry over the constructions of this chapter to noncommutative algebras. The failure of such attempts becomes really catastrophic when one tries to repeat subtler and deeper constructions in the noncommutative situation. These and many other reasons show that it is hardly possible to obtain a mathematically adequate quantum principle of observability by using the language of noncommutative algebra (or noncommutative geometry). There are serious reasons to believe now that this aim can be reached in a natural way by using the language of the secondary diﬀerential calculus, which is a sort of synthesis of the usual (= primary) diﬀerential calculus with homological algebra. In any case, we have no doubt that this calculus is the natural language for the geometry of nonlinear partial diﬀerential equations.

11 Smooth Bundles

11.1. The inner structure of points. The concept of observability developed in this book assumes that a point is an elementary object that can be individualized with the help of a given set of instruments, i.e., a given algebra of observables A. We know, however, that points of the physical manifold where we live may have inner parameters, such as temperature, color, and humidity. To give a precise mathematical meaning to this phrase, we need the notion of ﬁbration or ﬁber space, which is the main protagonist of the present chapter. A priori, there are two possibilities: the concept of inner structure can be either relative or absolute. The inner structure is relative if it can be expressed within the classical framework by simply adding new instruments. The mathematical meaning of this construction is that the algebra of observables A is extended to a bigger algebra B by means of the inclusion i : A → B. The inner structure of a point z ∈ |A| is then described by its inverse image |i|−1 (z) ⊂ |B| under the map of R-spectra |i| : |B| → |A|. Example. The 3-dimensional world R3 can be made colored if to the set of instruments measuring a point’s coordinates, we add one more instrument measuring the color, i.e., the frequency of electromagnetic waves. In alge braic language, this means that we pass from the algebra A = C ∞ R3 to the algebra B = A ⊗R C, where C is the algebra of smooth functions on the “manifold of colors,” which we identify naturally with R1+ . The inclusion i : A → B is deﬁned by the rule A a → a ⊗ 1C ∈ A ⊗R C. © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 11

151

152

Chapter 11

Returning to the general case, let us note that |i|−1 (z) = |Bz |, where Bz = B/(μz · B). The inner structure of a point z ∈ |A| is thus observable by means of the algebra Bz . The assumption that all points z ∈ |A| have the same inner structure means that all algebras of additional observables Bz are the same, i.e., isomorphic to each other. If this condition is fulﬁlled in a certain regular manner (see Section 11.9), then the map |i| : |B| → |A| is referred to as a locally trivial smooth bundle. In the above example, this condition holds, and all algebras Bz are isomorphic to C. At ﬁrst glance, relative inner structures do not add anything new to the classical scheme of observability, because any such structure can be reduced to a standard one through an appropriate extension of the algebra A. This approach, however, is not convenient when the manifold M = |A| is considered as a display that shows the points with diﬀerent inner structures. For example, this is the case for real physical space. Moreover, the problem ceases to be a question of mere convenience if the inner structures of the points displayed on M are absolute in the sense that they cannot be described by the above classical approach. In particular, this is true of quantum phenomena that have survived many unsuccessful attempts of explanation based on so-called “hidden parameters.” Unless otherwise speciﬁed, all algebras in this chapter are assumed to be smooth. 11.2. Fibrations as extension of algebras. Before going on to bundles, we shall introduce the more general notion of a ﬁbration. In algebraic terms it can be deﬁned as follows. Definition. A smooth fibration is an injective homomorphism of smooth algebras i : A → B. The manifold |A| is called the base of the ﬁbration i, the manifold |B| is its total space, while the map |i| : |B| → |A| is referred to as the projection of the ﬁbration. 11.3. Examples. I. Product fibration. Let A and C be smooth algebras. Set B = A ⊗R C (we recall that the bar stands for the smooth envelope of an algebra; see Section 3.36) and deﬁne the inclusion i : A → B by the rule a → a ⊗ 1. A concrete example of this construction is the ﬁbration of the torus over the circle, deﬁned as the extension i : A → B, where B = {g ∈ C ∞ (R2 ) | g(x + 1, y) = g(x, y + 1) = g(x, y)} is the algebra of twice periodic functions in two variables and A is the subalgebra of B consisting of all functions that do not depend on y. Indeed, if A = {f ∈ C ∞ (R) | f (x + 1) = f (x)}, C = {f ∈ C ∞ (R) | f (y + 1) = f (y)}, then it is readily veriﬁed that A ⊗ C ∼ = B.

(11.1) (11.2)

Smooth Bundles

153

Figure 11.1. Product fibration.

II. The Klein bottle fibered over the circle (see Figure 11.2). This is the inclusion of the algebra of smooth periodic functions on the real line A = {f ∈ C ∞ (R) | f (x + 1) = f (x)} into the algebra B = {g ∈ C ∞ (R2 ) | g(x + 1, y) = −g(x, y + 1) = g(x, y)} by the rule f → g : g(x, y) = f (x).

Figure 11.2. Nontrivial fibration.

154

Chapter 11

III. The two-sheeted covering of the circle. This is the map of the algebra A = {f ∈ C ∞ (R) | f (x + 1) = f (x)} into itself deﬁned by the formula f → g : g(x) = f (2x). IV. One-sheeted fibration of the line over the circle. Consider B = C ∞ (R) and let A ⊂ B consist of all functions f ∈ B for which the function x → f (1/x) and all of its derivatives have ﬁnite limits as x → 0. One can see that A ∼ = C ∞ (S 1 ), and under a proper choice of this isomorphism, the inclusion A → B corresponds to the map 1 − t2 2t R → S 1 = (x, y) | x2 + y 2 = 1 : t → , . 1 + t 2 1 + t2 The image of this map is the entire circle with one point removed. V. The projection of the tangent space T M of a manifold M onto M . In Section 9.18, we described the map πT : T M → M , given by (z, ξ) → z. The corresponding homomorphism of smooth algebras C ∞ (M ) → C ∞ (T M ) is injective; therefore, πT can be regarded as the projection of the ﬁbration. The most important class of ﬁbrations consists of bundles (see Sections 11.9 and 11.10), deﬁned as locally trivial ﬁbrations. In the previous list, examples I–III, V possess this property, while example IV does not. To give a precise deﬁnition of local triviality, we need the notion of ﬁber and the procedure of localization. 11.4. Fiber of a fibration. Geometrically, the ﬁber of a given ﬁbration i : A → B over the point z ∈ |A| is the inverse image of z under the projection |i| : |B| → |A|. In examples I–II, the ﬁber over any point is a circle; in example III it is two points; in example IV, depending on the choice of the point z, the ﬁber is either empty or consists of one point. Finally, in example V, the ﬁber Tz M is isomorphic to the linear space Rn . Note that in examples I and II both the base space and the ﬁber of both ﬁbrations are the same, whereas the total spaces are diﬀerent. An algebraic deﬁnition of the fiber over a point a ∈ |A| can be given as follows: it is the quotient algebra of B over the ideal generated by the set i(μa ), where μa ⊂ A is the maximal ideal of the point a. 11.5. The category of fibrations. By deﬁnition, a morphism of a ﬁbration i1 : A → B1 into a ﬁbration i2 : A → B2 is an algebra homomorphism ϕ : B2 → B1 making the following diagram commutative A i2

i1

B1

ϕ

B2

Smooth Bundles

155

An equivalent deﬁnition in terms of spectra (see Sections 3.4 and 8.11) reads that the diagram |ϕ|

|B1 | |i1 |

|B2 | |i2 |

|A| commutes. This means that the map |ϕ| takes the ﬁbers of one ﬁbration into the ﬁbers of another: |ϕ|(|i1 |−1 (a)) ⊂ |i2 |−1 (a), or, equivalently, that ϕ(i2 (μa ) · B2 ) ⊂ i1 (μa ) · B1 for any point a ∈ |A|. The totality of all ﬁbrations over a smooth algebra A = C ∞ (M ) together with all morphisms between them constitutes the category of fibrations over M. If the homomorphism ϕ is an isomorphism, or, which is the same thing, the map |ϕ| is a diﬀeomorphism, then the ﬁbrations i1 and i2 are said to be equivalent. In the case where the homomorphism ϕ is surjective, which corresponds to a proper embedding of manifolds |B1 | → |B2 |, the ﬁbration i1 is called a subfibration of the ﬁbration |i2 |. The simplest example of a ﬁbration with base M and ﬁber F is provided by the direct product M × F with the natural projection on the ﬁrst factor, or, in algebraic terms, the natural inclusion of the algebra A = C ∞ (M ) into the smooth envelope of the tensor product A ⊗ C, where C = C ∞ (F ). A ﬁbration equivalent (in the category of ﬁbrations) to a ﬁbration of this kind is referred to as a trivial fibration. Roughly speaking, a bundle is a ﬁbration that is locally isomorphic to a trivial ﬁbration, see Deﬁnition 11.9. This phrase will become an exact deﬁnition after we have explained the meaning of the word “locally.” 11.6. Localization. The aim of this section is to describe an algebraic construction that allows one to deﬁne the restriction of an algebra to an open set (see Section 3.23) in purely algebraic terms. Let A be a commutative ring with unit and let S ⊂ A be a multiplicative set, i.e., a subset of A, containing 1, not containing 0, and closed with respect to multiplication. In the set of all pairs (a, s), where a ∈ A, s ∈ S, we introduce the equivalence relation def

(a1 , s1 ) ∼ (a2 , s2 ) ⇐⇒ ∃s ∈ S : s(a1 s2 − a2 s1 ) = 0. The equivalence class of a pair (a, s) is denoted by as (or a/s) and called a formal fraction; we denote the set of all such classes by S −1 A. The sum and product of formal fractions are deﬁned by the ordinary formulas a1 a2 a1 a2 a1 a2 a1 s2 + a2 s1 · = , + = . s1 s2 s1 s2 s1 s2 s1 s2

156

Chapter 11

The resulting ring S −1 A is referred to as the localization of the ring A over the multiplicative system S. We leave the straightforward checks to the reader. There is a canonical homomorphism ι : A → S −1 A deﬁned by ι(a) = a/1. In general, ι is neither injective nor surjective. Suppose now that P is a module over A. In the same way as above, in the set of pairs (p, s) with p ∈ P , s ∈ S, we can introduce the equivalence relation def

(p1 , s1 ) ∼ (p2 , s2 ) ⇐⇒ ∃s ∈ S : s(s2 p1 − s1 p2 ) = 0. A formal fraction ps (or p/s) is the equivalence class of the pair (p, s). The set of all such classes, denoted by S −1 P , is referred to as the localization of P over S. Exercises. 1. Introduce the addition of two elements of S −1 P and the multiplication of an element of S −1 P by an element of S −1 A as follows: p1 p2 s2 p1 + s1 p2 a1 p2 a1 p2 + = , · = . s1 s2 s1 s2 s1 s2 s1 s2 Verify that these operations are well deﬁned and turn the set S −1 P into an (S −1 A)-module. 2. Let ϕ : P → Q be a homomorphism of A-modules. Prove that the map S −1 (ϕ) : S −1 P → S −1 Q, given by the formula p def ϕ(p) , p ∈ P, s ∈ S, = S −1 (ϕ) s s is well deﬁned and represents a homomorphism of (S −1 A)-modules. Summarizing, we can say that for a given multiplicative set S ⊂ A we have deﬁned a functor S −1 from the category of A-modules to the category of (S −1 A)-modules. Examples. I. If A has no zero divisors and S = A \ {0}, then S −1 A is the quotient ﬁeld of A. II. Let A = Z and let S be the set of all nonnegative powers of 10. Then S −1 A consists of all rational numbers that have a ﬁnite decimal representation. III. If M is a smooth manifold, A = C ∞ (M ), x ∈ M , and S = A \ μx , where μz = {f ∈ C ∞ (M )|f (z) = 0}, then S −1 A is the ring of germs of smooth functions on M at the point x (readers who are familiar with the notion of germ may prove this fact as an exercise; others may take it as the deﬁnition and try to understand its geometrical meaning). 11.7 Proposition. Let U be an open subset of the manifold M , A = C ∞ (M ),

and

S = {f ∈ A | f (x) = 0 ∀x ∈ U }.

Smooth Bundles

157

Then S −1 A is isomorphic to C ∞ (U ), and the canonical homomorphism

ι : C ∞ (M ) → C ∞ (U ) coincides with the restriction ι(f ) = f U . To prove this fact, consider the map α : S −1 A → C ∞ (U ) given by f f (x) , f ∈ A, s ∈ S, x ∈ U, α (x) = s s(x) which converts a formal fraction into the ordinary quotient of two functions. This map is well deﬁned, because the functions s ∈ S do not vanish anywhere in U . Suppose that α(f1 /s1 ) = α(f2 /s2 ). Then the function f1 s2 − f2 s1 is identically zero on U . By Lemma II of Section 4.17, there is a function s ∈ S that is identically zero outside of U . The product of these two functions is identically zero on all of M , i.e., s(f1 s2 − f2 s1 ) = 0 as an element of the algebra A. By deﬁnition, the two formal fractions f1 /s1 and f2 /s2 are equal. This proves that α is injective. The fact that it is surjective follows from the next lemma. Exercise. Give an example of a geometrical algebra A that is not smooth and of an open subset U ⊂ |A| such that the algebras S −1 A and A|U are not isomorphic. 11.8 Lemma. Suppose that U is an open subset of the manifold M and f ∈ C ∞ (U ). Then there exists a function g ∈ C ∞ (M ), having no zeros on U , such that the product f g can be smoothly extended to the entire manifold M and therefore f = α(f g/g). A rigorous proof of this fact can be obtained by the techniques of partition of unity described in Chapter 2. We leave the details to the reader. We are now in a position to give a precise algebraic deﬁnition of a bundle. Recall that a homomorphism of K-algebras ϕ : A → B gives rise to the operation of the change of rings: any B-module R can be regarded as an def A-module with multiplication a · r = ϕ(a)r, a ∈ A, r ∈ R. In particular, the algebra B itself can be regarded as an A-module. Therefore, for a given multiplicative set S ⊂ A, the (S −1 A)-module S −1 B is deﬁned. 11.9. Definition. Let A, B, and F be smooth algebras. An injective homomorphism i : A → B is called a bundle |B| over |A| with fiber |F | if every point z ∈ |A| has an open neighborhood Uz ⊂ |A| over which the localization of i is equivalent to the product ﬁbration Sz−1 A → Sz−1 A ⊗ F , where Sz ⊂ A is the multiplicative system of Uz , i.e., the set of all elements of A whose values at the points of Uz are nonzero. More exactly, there must exist an algebra isomorphism p : Sz−1 B → Sz−1 A ⊗ F that makes the

158

Chapter 11

following triangle commutative Sz−1 (i)

Sz−1 A

Sz−1 B p

j

Sz−1 A

⊗ F,

where j is the map taking every element a to a ⊗ 1. (Recall once again that the bar over the notation of an algebra means that we take its smooth envelope.) A bundle π is trivial if and only if the previous condition (local triviality axiom) is fulﬁlled for Uz = M . The geometrical deﬁnition of a bundle, which follows, is simpler. The equivalence of the two deﬁnitions is quite obvious. 11.10. Definition. Let E and M be smooth manifolds. A smooth map π : E → M is said to be a fiber bundle, or bundle for short, if there exists a manifold F such that the following condition holds: any point x ∈ M has a neighborhood U ⊂ M for which there exists a trivializing diﬀeomorphism ϕ : π −1 (U ) → U × F that closes the commutative diagram ϕ

π −1 (U ) π

U ×F p

U, where p is the projection of the product on the ﬁrst factor. Under these conditions, M , F , and E are referred to as the base, the fiber, and the total space of the bundle π, respectively. The dimension of the bundle is by deﬁnition equal to the dimension of the ﬁber. The whole F thing is conventionally written as E → M . The total space of the bundle π is usually denoted by Eπ . The ﬁber bundle π over a point x ∈ M is the set πx = π −1 (x). The ﬁber over any point, equipped with the structure of the submanifold of E, is diﬀeomorphic to the ﬁxed “outer” ﬁber F . 11.11. Some more examples. In the examples that follow, we give only the geometrical construction of the bundle, leaving it to the reader to describe the corresponding extension of smooth algebras. R I. The open M¨ obius band is a line bundle over the circle: M → S 1 . 2 If M is viewed as the band [0, 1] × R ⊂ R with the points (0, y) and (1, −y) identiﬁed for any y ∈ R, and the circle S 1 is viewed as the segment [0, 1] with identiﬁed end points 0 and 1, then the projection π : M → S 1 is simply π(x, y) = x. A visual representation of this bundle is given in Figure 6.2, where the M¨ obius band embedded into R3 squeezes to its midline; the ﬁbers are the segments perpendicular to the midline.

Smooth Bundles

159

Let us check local triviality in this example. If a ∈ S 1 is an inner point of the segment [0, 1], then for the neighborhood

U we can take the inter val ]0, 1[. If a = 0, then we put U = x ∈ S 1 x = 1/2 and deﬁne the diﬀeomorphism ϕ : π −1 (U ) → U × R as follows: (x, y), ifx < 12 , ϕ(x, y) = (x, −y), if x > 12 . II. The bundle of the line over the circle (cf. Section 6.2). Representing the circle S 1 as the set of complex numbers of modulus 1, we deﬁne the map exp : R1 → S 1 by t → eit . Any neighborhood of x ∈ S 1 , but not the whole circle S 1 , can be taken as the neighborhood U in the condition of local triviality. The ﬁber in this example is the zero-dimensional manifold Z. Bundles with zero-dimensional ﬁbers are called coverings. III. Another example of a covering is provided by the map S n → RP n that assigns to a point x ∈ S n ⊂ Rn the line in Rn+1 passing through that point and the origin (see the deﬁnitions in Section 5.10). This bundle is trivial over the comFigure 11.3. exp : R1 → S 1 plement to any hyperplane RP n \ RP n−1 . Its ﬁber consists of two points. IV. The bundle of unit tangent vectors to the sphere π : T1 S 2 → 2 S . The space of this bundle

T1 S 2 = {(x, y) ∈ R3 × R3 |x| = 1, |y| = 1, x ⊥ y} is a submanifold of R6 . Making the orthogonal group act on a ﬁxed unit vector to the sphere, we obtain a diﬀeomorphism T1 S 2 ∼ = SO(3). The total space of the bundle under study thus provides another (ﬁfth!) realization of the manifold considered at the beginning of the book in Examples 1.1–1.4. The ﬁber is the circle S 1 . To prove local triviality, we shall show that this bundle is trivial over any open hemisphere S 2 . Indeed, if U is a hemisphere and S 1 is its boundary, then we can identify the points of U and S 1 with the vectors drawn from the center of the sphere and set ϕ(x, z) = x×z (cross product of vectors) for x ∈ U and z ∈ S 1 . The two vectors x ∈ U and z ∈ S 1 are never collinear; hence the map ϕ : U × S 1 → π −1 (U ) is a diﬀeomorphism. V. The composition of the maps S 3 → RP 3 and RP 3 → S 2 deﬁned in Examples III and IV above is a bundle of S 3 over S 2 with ﬁber S 1 , called

160

Chapter 11

the Hopf fibration, compare with Section 6.17, II. We leave it to the reader to check its local triviality. VI. Tautological bundle over a Grassmannian. Suppose Gn,k is the Grassmann manifold (see Example IV in Section 5.10) whose points are k-dimensional linear subspaces of the n-dimensional space Rn ; let En,k be the set of all pairs (x, L) such that x ∈ L ∈ Gn,k ; En,k is viewed as a submanifold in Rn × Gn,k . The correspondence (x, L) → L deﬁnes a ﬁbration Θ = Θn,k : En,k → Gn,k , called the tautological bundle. Exercise. Show that the tautological bundle En,1 → Gn,1 = RP n−1 for k = 1 can be described as the projection π : RP n \ {L0 } → RP n−1 , where L0 is the (n + 1)th coordinate axis in Rn+1 (= “vertical” line), and π assigns to each “slanted” line its projection to Rn ∼ = {xn+1 = 0}. In particular, if k = 1 and n = 2, we obtain the ﬁbration of the M¨ obius band over the circle from Example III. Let us prove that the tautological ﬁbration is a bundle, i.e., possesses the property of local triviality. We shall use the open covering of the manifold Gn,k by the family of open sets UI ,

where I = {i1 , . . . , ik }, 1 i1 < · · · < ik n.

By deﬁnition, the neighborhood UI consists of all k-planes in Rn that do not , where I = {1, . . . , n}\I degenerate under the projection on RkI along Rn−k I and the symbol Rm , J = {j , . . . , j }, stands for the m-plane in Rn spanned 1 m J by the basic vectors numbered j1 , . . . , jm . If x ∈ L and L ∈ UI , then to the pair (x, L) ∈ En,k we assign the pair (x, L), where x ∈ RkI ∼ = Rk is n−k k the projection of x along RI onto RI . This assignment is a trivializing diﬀeomorphism for the tautological ﬁbration over the set UI . This is why Θn,k is in fact a bundle. All bundles listed above are nontrivial. For example, the M¨ obius band is nonorientable and therefore not diﬀeomorphic to the cylinder S 1 × R. The nontriviality of the bundle of unit tangent vectors (Example IV) can be proved by using the basic facts about fundamental groups. In fact, the manifolds RP 3 and S 2 × S 1 are not diﬀeomorphic, because their fundamental groups are diﬀerent: π1 (RP 3 ) = Z2 , π1 (S 2 × S 1 ) = Z. The same argument shows the nontriviality of the Hopf ﬁbration (Example V). VII. The tangent bundle πT : T M → M (see Section 9.19). If (U, x) is a chart on M , then the corresponding trivializing diﬀeomorphism U × Rn → πT−1 (U )

Smooth Bundles

161

is the composition of natural identiﬁcations U × Rn ←→ T U

and T U ←→ πT−1 (U ),

described in Section 9.18 II and IV: n ∂

(z, q) → qi

∈ Tz M ⊂ T U, where q = (q1 , . . . , qn ). ∂xi z i=1 Depending on the manifold M , its tangent bundle can be either trivial or nontrivial. A manifold with a trivial tangent bundle is called parallelizable. For example, any Lie group is parallelizable. Exercise. Among the examples of manifolds considered earlier in this book, ﬁnd some that are parallelizable and some that are not. VIII. The cotangent bundle πT ∗ : T ∗ M → M (Section 9.24). Just as in the previous case, the required trivialization U × Rn → πT−1∗ (U ) can be obtained from the identiﬁcations U × Rn ←→ T ∗ U

and T ∗ U ←→ πT−1∗ (U ),

described in Section 9.24: n (z, p) → pi dz xi ∈ Tz∗ M ⊂ T ∗ U,

where p = (p1 , . . . , pn ).

i=1

IX. The bundle of l-jets of functions πJ l : J l M → M . In the chart l (U, x) on M , the trivializing map U × RN → πJ−1 l (U ) = J (U ), where N is the total number of diﬀerent derivatives of order ≤ l, 1 z, pl → [fpl ]lz ∈ Jzl M ⊂ J l U, where fpl = pσ (x − z)σ , σ! σ≤l

and p = (pσ ) is the vector with components pσ , |σ| ≤ l, arranged in the lexicographic order of the subscripts. Formula (2.4) shows that l-jets of functions fpl at the point z exhaust Jz M . The functions xi , i = 1, . . . , n, and pσ , |σ| ≤ l, constitute a local coordinate system in πJ−1 l (U ), and such special charts form an atlas of J l M . See Section 14.20. l

Exercise. Find the value of N deﬁned above. 11.12. Sections. A section of a bundle π : E → M is a smooth map s : M → E that assigns to every point x ∈ M an element of the ﬁber over this point: s(x) ∈ πx . In other words, the map s is such that π ◦ s = idM . The set of all sections of the bundle π is denoted by Γ(π). In algebraic language, a section of the bundle i : A → B is represented by an algebra homomorphism σ : B → A, left inverse to i, i.e., such that σ ◦ i = idA .

162

Chapter 11

Examples. I. The bundle π : R1 → S 1 , described in Example 11.13, I, has no sections. Indeed, let f : S 1 → R1 be a smooth map and π ◦ f = idS 1 . The last equality implies that the restriction π f (S 1 ) : f (S 1 ) → S 1 is a diﬀeomorphism of the set f (S 1 ) ⊂ R1 onto S 1 . However, the image of a continuous map f : S 1 → R1 is a certain segment [a, b] ⊂ R and thus cannot be homeomorphic to the circle. II. The bundle T1 S 2 → S 2 (Example 11.11, IV) has no sections. This fact is sometimes referred to as the hedgehog theorem. Indeed, the existence of a section f : S 2 → T1 S 2 would lead to the following construction of a diﬀeomorphism ϕ : S 2 × S 1 → T1 S 2 : put ϕ(x, α) equal to the vector, obtained from f (x) by a rotation through angle α in a ﬁxed direction (say, counterclockwise, if the sphere is viewed from the outside). III. The sections of the trivial bundle over M with ﬁber F are in one-toone correspondence with smooth maps from M to F . IV. The sections of the tangent bundle over M are naturally interpreted as vector ﬁelds on M (see Section 9.40). In a similar way, the sections of the cotangent bundle over M are naturally associated with ﬁrst-order diﬀerential forms, or 1-forms. This notion is introduced and discussed below in Sections 14.4–14.8. A similar situation takes place for jet bundles as well; see Sections 9.65, 14.14–14.15. Exercise. Describe the sections sX , sdf , and sjl (f ) of the tangent, cotangent, and l-jet bundles (see 9.40, 9.25, and 9.65, respectively) in terms of special local coordinates. V. This example concerns some remarkable sections of the tautological bundle over the Grassmannian Θn,k : En,k → Gn,k (see Example VI in Section 11.11). Let Mk,n be the space of (k × n) matrices of rank k. If J = (j1 , . . . , jk ), 1 ≤ j1 < · · · < jk ≤ n, and M ∈ Mk,n , then MJ denotes the (k×k) matrix formed by the columns of M with numbers j1 , . . . , jk . At least

one of the minors |MJ | in the matrix M is diﬀerent from zero; therefore, J |MJ |2 > 0. Fix a multiindex I = (i1 , . . . , ik ) and consider the following function on Mk,n : |MI | νI (M) = . 2 J |MJ | Obviously, νI ∈ C ∞ (Mk,n ), and for any g ∈ GL(k, R), we have νI (gM) = |g|−1 νI (M).

(11.3)

I be the adjoint to the matrix MI , i.e., the one formed by Further, let M the minors of order k − 1 of MI . Denote by Matk,n the space of all (k × n) matrices over R. The map mI : Mk,n → Matk,n ,

I M, mI (M) = νI (M)M

(11.4)

Smooth Bundles

163

is GL(k, R)-equivariant, i.e., satisﬁes mI (gM) = mI (M),

g ∈ GL(k, R).

(11.5)

It is evidently smooth. Furthermore, mI (M) ∈ Mk,n if |MI | = 0; otherwise, mI (M) = 0. Consider the natural projection μ : Mk,n → Gn,k , where μ(M) is the subspace of Rn spanned by the rows of the matrix M. The two conditions: μ(M ) = μ(M) and M = g(M) for some g ∈ GL(k, R) are equivalent. Therefore, μ(mI (M)) = μ(M) if |MI | = 0. On the other hand, the last condition means that μI (M) ∈ UI (see Example VI in Section 11.11). We are now in a position to deﬁne the section sI,i : Gn,k → En,k of the tautological bundle Θn,k by setting sI,i (L) = (ith line of the matrix mI (M), L), where M is any matrix such that mI (M) = L. By virtue of (11.5), this construction is well deﬁned. Formula (11.4) ensures that the sections sI,i are smooth. 11.13. Subbundles. A bundle η : Eη → M is said to be a subbundle of the bundle π : Eπ → M (notation: η ⊂ π) if (i) the total space Eη is a submanifold of Eπ ; (ii) the map η is the restriction of π on Eη ; (iii) for any point x ∈ M the ﬁber ηx is a submanifold of the ﬁber πx . Exercise. Give an algebraic deﬁnition of subbundles. Examples. I. The tautological bundle over the Grassmannian (Example VI from 11.11) is a subbundle of the trivial bundle Rn × Gn,k → Gn,k . II. The bundle of unit tangent vectors of the sphere (Example IV from 11.13), which is a subbundle of the tangent bundle of the sphere, has no proper subbundles. 11.14. Whitney sum. Given two bundles η and ζ over one and the same manifold M , one can construct a new bundle π whose ﬁber over an arbitrary point x ∈ M is the Cartesian product of the ﬁbers of η and ζ: πx = ηx × ζx . This bundle π is called the direct sum, or Whitney sum, or Whitney product, of the bundles η and ζ and denoted by η ⊕ ζ.

164

Chapter 11

To give this construction an exact meaning, we must explain how the individual ﬁbers are put together to make a smooth manifold. As always, there are two ways to do this. The algebraic deﬁnition of the Whitney sum reads as follows: If the two given bundles correspond to algebra extensions i : A → B and j : A → C, then their Whitney sum is represented by the homomorphism i ⊗ j : A → B ⊗A C that takes every element a to a(1 ⊗ 1) = i(a) ⊗ 1 + 1 ⊗ j(a). Note that the tensor product of the algebras B and C is taken over the algebra A, not over the ground ring. This reﬂects the fact that it is the ﬁbers that get multiplied in this construction, not the total spaces of the bundles. The geometric construction of the Whitney sum consists in the following. The total space of the bundle η ⊕ ζ is deﬁned as Eη⊕ζ = {(y, z) ∈ Eη × Eζ | η(y) = ζ(z)}, and the projection as the map that takes the pair (y, z) to the point η(y) = ζ(z). Exercise. Check that these deﬁnitions of the Whitney sum are equivalent and the ﬁber of the resulting bundle over an arbitrary point x is in a natural bijection with the manifold ηx × ζx . There is one more useful description of Whitney sum. The map η × ζ : Eη × Eζ → M × M,

(e1 , e2 ) → (η(e1 ), ζ(e2 )),

is a bundle with ﬁber ηu × ζv over the point (u, v), where u, v ∈ M . The diagonal MΔ = {(z, z) | z ∈ M } ⊂ M × M is a submanifold in M × M , identiﬁed with M via the map z → (z, z). The total space Eη⊕ζ and the projection η ⊕ ζ are identiﬁed with the manifold (η × ζ)−1 (MΔ ) and the map η × ζ (η×ζ)−1 (MΔ ) , respectively. The restrictions of p η : Eη × E ζ → Eη −1

to the submanifold (η × ζ)

and pζ : Eη × Eζ → Eζ

(MΔ ) give rise to smooth surjective maps

pη : Eη⊕ζ → Eη

and pζ : Eη⊕ζ → Eζ .

def

def

If s ∈ Γ(η ⊕ ζ), then sη = pη ◦ s ∈ Γ(η), sζ = pη ◦ s ∈ Γ(ζ), and s(z) = (sη (z), sζ (z)) ∈ ηz × ζz . This establishes a natural bijection between the sets of sections Γ(η ⊕ ζ) = Γ(η) × Γ(ζ).

(11.6)

11.15. Examples of direct sums. I. Denote by α : S11 ∪ S21 → S 1 , where Si1 are copies of S 1 , i = 1, 2, and by β : S 1 → S 1 the trivial and the nontrivial two-sheeted coverings of the circle, respectively. Let A be the algebra of

Smooth Bundles

165

smooth functions on the circle, i.e., the algebra of smooth periodic functions on the line. In algebraic terms, the map α is described by the injection i : A → A⊕A, f → (f, f ), while β corresponds to the map j : A → A taking f (x) to f (2x). Then 1. α ⊕ α is a trivial four-sheeted covering of the circle; 2. α ⊕ β ∼ = β ⊕ β is a four-sheeted covering of the circle, whose total space consists of two connected components, each of which represents a nontrivial two-sheeted covering. To understand this fact geometrically, it is suﬃcient to sketch the behavior of the four points of the ﬁber after one complete turn of the base circle. Exercise. Prove these facts algebraically, considering the tensor products of algebras A1 and A2 over A, where A1 = A2 = A = C ∞ (S 1 ) and A1 , A2 are equipped with the A-module structure induced by the inclusions i and j.

Figure 11.4. Direct sum of two M¨ obius bands.

II. The direct sum of two M¨ obius bands, considered as bundles over the circle, is trivial. A visual proof of this fact is shown in Figure 11.4. Let us represent the M¨obius bundle as a subbundle of the trivial bundle over the circle with ﬁber R2 . Then the lines perpendicular to its ﬁbers constitute another M¨ obius bundle. III. The bundle of 1-jets (see Section 9.28) is the direct sum of the trivial 1-dimensional bundle M × R → M and the cotangent bundle πT ∗ (see Section 9.24). 11.16. Pullback bundle. Given a bundle π : Eπ → M and a smooth map f : N → M , we can attach a copy of the ﬁber πf (y) to every point y ∈ N . The union of all these ﬁbers constitutes the total space of the bundle, f ∗ (π),

166

Chapter 11

induced from π by means of the map f or the pullback of π by f . There are two ways to turn this intuitive picture into a precise deﬁnition. Geometrically, the total space of the pullback bundle is deﬁned as def

Ef ∗ (π) = {(y, z) | y ∈ N, z ∈ Eπ , π(z) = f (y)}. The projection f ∗ (π) acts as follows: f ∗ (π)(y, z) = y. Let us check its local triviality. For a point b ∈ N , we set a = f (b) and choose a neighborhood U of the point a in M such that π is trivial over U . Let ψ : π −1 (U ) → U × F be the trivializing diﬀeomorphism and let ψ be its composition with the projection U × F → F . Set χ(y, z) = y, ψ(z) . Then χ : (f ∗ (π))−1 f −1 (U ) → f −1 (U ) × F is the required diﬀeomorphism. The restriction of the bundle π to a submanifold N ⊂ M is a particular case of pullback bundle. It is deﬁned as follows:

def

π N = π π−1 (N ) : π −1 (N ) → N. (11.7)

Exercise. Check that π N = i∗ (π), where i : N → M is the inclusion map. The algebraic deﬁnition of the pullback bundle can be stated as follows. Let i : A → B be a bundle, understood as an algebra extension, and let ϕ : A → A1 be the algebra homomorphism corresponding to the smooth map |ϕ| : |A1 | → |A|. Consider the algebras A1 and B as A-modules with multiplication deﬁned via i and ϕ. Then the pullback bundle |ϕ|∗ (i) is the natural homomorphism A1 → A1 ⊗A B. Remark. The commutative diagram A i

B

ϕ

A1 |ϕ|∗ (i)

A1 ⊗A B

shows that the notion of pullback bundle is a generalization of the Whitney sum. Exercises. 1. Prove the equivalence of the geometrical and the algebraic deﬁnitions of the pullback bundle. 2. Show that a vector ﬁeld along a map of manifolds ϕ : N → M (see Section 9.47) can be interpreted as a section of the induced bundle ϕ∗ (πT M ) in the same way as an ordinary vector ﬁeld is interpreted as a section of the tangent bundle (see Section 9.40).

Smooth Bundles

167

Figure 11.5. Bundles fn∗ (µ) for n = 1, 2, 3.

11.17. Examples. I. Let μ be the M¨ obius band bundle over the circle (11.11, III) and fn : S 1 → S 1 the n-sheeted covering of the circle (representing S 1 = {z ∈ C | |z| = 1}, one can set fn (z) = z n , n ∈ Z). Then μ, n odd, fn∗ (μ) = 1 IS , n even. (See Figure 11.5; here and below, IM denotes the trivial bundle over M with ﬁber R.) II. Triviality criterion in terms of pullback bundles. A bundle is trivial if and only if it is equivalent to the pullback bundle obtained from a bundle over one point. Exercise. Prove this fact. 11.18. Deﬁne the canonical morphism κ : Ef ∗ (π) → Eπ by κ(y, z) = z. The map κ is included into the commutative diagram Ef ∗ (π)

κ

∗

π

f (π)

N

Eπ

f

M

and therefore is an f -morphism from the bundle f ∗ (π) to the bundle π. More generally, given two bundles π over M and η over N and a smooth map f : N → M , then an f -morphism, or a morphism over f , from η

168

Chapter 11

into π is a smooth map ψ : Eη → Eπ that makes the following diagram commutative: Eη

ψ

Eπ

η

N

π

M.

f

The notion of f -morphism generalizes the notion of a morphism of bundles over M : the latter is nothing but a morphism over the map idM . Example. The map T Φ : T M → T N , arising from a map Φ : M → N (Section 9.18), is a Φ-morphism from πT M into πT N . The pair (f ∗ (π), η) has the following universal property: for any bundle η over M and any f -morphism ψ : η → π there exists a unique smooth map χ that makes commutative the diagram Eη χ η

ψ

Ef ∗ (π)

κ

Eπ

f ∗ (π)

N

f

π

M.

The proof is easy. For an arbitrary y ∈ Eη , both projections of χ(y) onto N and Eπ are uniquely deﬁned due to the commutativity of the diagram. This implies the uniqueness of χ. Existence follows from the explicit formula χ(y) = (η(y), ψ(y)). There is a natural map f: Γ(π) → Γ(f ∗ (π)) called the lifting of sections. By deﬁnition, for any s ∈ Γ(π) the value of the section f(s) at the point y ∈ N is equal to the value of s at f (y). The precise formula is f(s)(y) = (y, s(f (y))) ∈ Ef ∗ (π) . The section f(s) is called the lift of s along f . 11.19. Regular morphisms. Working with manifolds whose points have one and the same inner structure, it is natural to introduce the class of morphisms that preserve this structure. More speciﬁcally, an f -morphism ψ is said to be regular if for any point z ∈ N the map of ﬁbers ψz : ηz → πf (z) is a diﬀeomorphism. Proposition. Let ψ : η → π be a regular morphism of bundles over the map f : N → M . Then the canonical morphism χ : η → f ∗ (π) defined in the previous section is an equivalence of bundles over N .

Smooth Bundles

169

For any z ∈ N , the ﬁber map χz : ηz → f ∗ (π)z is identiﬁed with the map ψ : ηz → πf (z) by the construction of χ, and therefore is a diﬀeomorphism. As a consequence, the map χ : Eη → Ef ∗ (π) is a diﬀeomorphism, too. The proposition shows that the class of bundles related to a given bundle π by means of regular morphisms is exhausted by the bundles pulled back from π. This observation leads to the tempting idea of building, for a given type of ﬁbers, a universal bundle such that any bundle with this ﬁber could be the pullback of the universal bundle via a suitable smooth map. After an appropriate concretization, this idea can be implemented. Here is an example.

M

ϕ

R2n G2n,n ϕ (M)

Figure 11.6. The Gauss map.

11.20. Example (Gauss map). Let M be an n-dimensional manifold. By the Whitney immersion theorem, M can be immersed into R2n , i.e., there exists a map ϕ : M → R2n such that for any point z ∈ M the diﬀerential dz ϕ : Tz M → Tϕ(z) R2n is injective. Denote by ra , a ∈ R2n , the linear shift in R2n through the vector −a: R2n v → v − a ∈ R2n . Let G2n,n be the Grassmann manifold of n-dimensional linear subspaces in TO R2n , where O = (0, . . . , 0) ∈ R2n . The Gauss map g : M → G2n,n takes every point z ∈ M to the image of the corresponding tangent space Tz M under the map dz (rϕ(z) ◦ ϕ) : Tz M → TO R2n (see Figure 11.6). The map g is covered by the morphism of bundles γ : πT M → Θ2n,n , where Θ2n,n : E2n,n → G2n,n is the tautological bundle described in Example VI of Section 11.11. Indeed, if ξ ∈ Tz M , then γ(ξ) = dz (rϕ(z) ◦ ϕ)(ξ), g(z) ∈ E2n,n .

170

Chapter 11

Therefore, by Proposition 11.19, the tangent bundles of all n-dimensional manifolds can be obtained as pullbacks from the tautological bundle over the Grassmannian G2n,n . This fact plays an important role in the study of manifolds. In particular, it lies at the foundation of the theory of characteristic classes.

12 Vector Bundles and Projective Modules

12.1. We have seen in the previous chapter that for a given bundle, the ﬁber over a point of the base space describes the inner structure of this point. The ﬁber may have a certain mathematical structure. For example, the ﬁbers of the tangent bundle have the natural structure of a linear space, and this structure has an evident physical meaning. Indeed, if the manifold M is the conﬁguration space of a mechanical system (see Section 9.22), then for a ﬁxed point a ∈ M , the tangent vector is interpreted as the velocity vector of the system having conﬁguration a. Bundles of this kind, where ﬁbers are vector spaces, are called vector bundles (see Section 12.2 for an exact deﬁnition). They form an interesting and important class of bundles. In particular, besides tangent bundles, this class contains cotangent bundles and jet bundles, which are fundamental objects of study in the geometrical theory of diﬀerential equations. Vector bundles have a simpler algebraic description than bundles of general type. It turns out that under certain natural regularity conditions the extensions of algebras A → B that correspond to vector bundles are of the form A → S(P ), where S(P ) is the symmetric algebra of a certain A-module P , appropriately completed. The study of vector bundles over a manifold M is thus reduced to the study of a certain class of modules over the algebra A = C ∞ (M ). We begin with the geometrical deﬁnition of a vector bundle. After an investigation of the basic properties of vector bundles, we shall prove the fundamental theorem on the equivalence of the notions of a vector bundle over M and a ﬁnitely generated projective module over A, and then explain how symmetric algebras of modules appear in this context. © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 12

171

172

Chapter 12

12.2. Geometrical definition of a vector bundle. A ﬁber bundle π : E → M with ﬁber V is said to be a vector bundle if (1) V is a vector space (over R); (2) For any point x ∈ M , the ﬁber πx is a vector space; (3) The vector property of local triviality holds: for any point x ∈ M there exists a neighborhood U ⊂ M , x ∈ U , and a trivializing diﬀeomorphism ϕ : π −1 (U ) → U × V , linear on every ﬁber, i.e., such that all maps ϕy : πy → V , y ∈ U , are linear. The dimension or rank of a vector bundle is the dimension of its ﬁber. Zero-dimensional vector bundles are called zero bundles and denoted by OM . Trivial one-dimensional bundles are called unit bundles and denoted by IM . 12.3. Adapted coordinates in vector bundles. Suppose (U, x) is a chart on M satisfying Condition (3) of Deﬁnition 12.2, while ϕ : π −1 (U ) → U × V is the corresponding trivializing diﬀeomorphism, and ξ ∈ V . The map −1 sϕ (U ), ξ : U →π

U x → ϕ−1 (x, ξ) ∈ π −1 (U ),

is a section of the ﬁbration π U . If v1 , . . . , vm is a basis of the vector space V , then the sections ej = sϕ vj , 1 j m, have the property that at every point z ∈ U the vectors e1 (z) = ϕ−1 (z, v1 ), . . . , em (z) = ϕ−1 (z, vm ) −1 form a basis of the space πU (z) ∼ = V . Below (see Section 12.7), we shall show that the totality of all sections of a vector bundle π over M has a natural C ∞ (M )-module structure induced by the linear structure in the ﬁbers πz . In this sense we can say that in the chosen coordinate chart U the module of sections of the bundle π|U is free and ej = sϕ vj , 1 j m, is its free basis. Now suppose that z ∈ U and (x1 , . . . , xn ) are coordinate functions on U . A point y ∈ πz ⊂ π −1 (U ) is deﬁned by the set of n + m numbers (x1 , .. . , xn , u1 , . . . , um ), where (x1 , . . . , xn ) are coordinates of the point z and u1 , . . . , um are coordinates of the point y with respect to the basis e1 |z , . . . , em |z . The functions x1 , . . . , xn , u1 , . . . , um form a coordinate system on π −1 (U ), called the adapted coordinates of the bundle π. Accordingly, the chart π −1 (U ), x1 , . . . , xn , u1 , . . . , um on the manifold E is referred to as an adapted chart. Finally, an atlas made up of adapted charts is also called adapted. It is readily veriﬁed that if two charts (U, x) and (U , x ) on M are compatible, then the corresponding adapted charts on E are compatible, too. (The reader is invited to check this fact as an exercise.) In particular, this means that every atlas on the manifold M gives rise to an adapted atlas on the total space E.

Vector Bundles and Projective Modules

173

12.4. Morphisms of vector bundles. A morphism of vector bundles α : π → η over M is a bundle morphism α : Eπ → Eη , which is ﬁberwise linear (i.e., the map αz , the restriction of α to the ﬁber over z, is R-linear for any point z ∈ M of the base space). The set of all morphisms from π to η is denoted by Mor(π, η), and the category of vector bundles thus arising will be denoted by VBM . For the local study of morphisms, the following point of view is convenient: a morphism of trivial vector bundles over M is the same thing as an operator-valued function on the manifold M . An exact statement of this observation is contained in the obvious lemma that follows. 12.5 Lemma. Let π : M × V → M and η : M × W → M be trivial vector bundles. To every fiberwise linear map ϕ : M × V → M × W one can assign a family of linear operators ϕ : M → Hom(V, W ) by setting the value of the operator ϕ(x) on the vector v ∈ V equal to the W -component of the element ϕ(x, v) ∈ M × W . Then the following conditions are equivalent: (a) the map ϕ is smooth (i.e., ϕ is a bundle morphism); (b) the map ϕ is smooth (the space Hom(V, W ) is endowed with the structure of a smooth manifold, because it is a finite-dimensional real vector space). 12.6. Examples of vector bundles. I. The M¨ obius band ﬁbered over the circle (Example 11.11, III) can be viewed as a vector bundle if its ﬁbers are regarded as one-dimensional linear spaces. Representing the algebra of functions on the M¨ obius band as the subalgebra B ⊂ C ∞ (R2 ) distinguished by the condition f (x + 1, y) = f (x, −y), we can deﬁne this bundle by the inclusion of algebras A → B that takes a function f to the function g, g(x, y) = f (x); here A = {f ∈ C ∞ (R) | f (x + 1) = f (x)} is the algebra of functions on the circle. II. The tangent bundle πT : T M → M (see Section 9.19). The trivializing diﬀeomorphisms described in Section 11.11, VII, are ﬁberwise linear. Therefore, the tangent bundle is a vector bundle. III. The cotangent bundle πT ∗ : T ∗ M → M (see Section 9.24). As in the previous example, the trivializing diﬀeomorphisms of Section 11.11, VIII, are obviously linear. IV. The trivializations described in Section 11.11, IX, are also ﬁberwise linear. Therefore, the bundle of l-jets πJ l : J l M → M is a vector bundle. Note that in the last three examples, the special coordinate systems deﬁned in 9.19, 9.24, and 11.11, IX, respectively, are adapted. Exercise. Check whether the maps (see 9.28) [f ]m πl,m : J l M → J m M, [f ]lz → z , l

∗

τl : J M → T M, are vector bundles.

[f ]lz

→ dz (f ),

l m, l 1,

174

Chapter 12

12.7. Module of sections. A remarkable property of vector bundles is that their sets of sections possess a module structure over the algebra of smooth functions. The resulting interrelation between the bundles and modules is of fundamental importance. Note ﬁrst of all that the set of sections of any vector bundle is nonempty: It always contains the zero section s0 . By deﬁnition, the value of s0 at any point z ∈ M is the zero of the vector space πz . Using the linear structure in the ﬁbers πx , one can introduce two operations in the set of sections of a vector bundle: addition and multiplication by a function on the manifold, (s1 + s2 )(z) = s1 (z) + s2 (z),

(f s)(z) = f (z)s(z),

for any sections s, s1 , s2 , any smooth function f ∈ C ∞ (M ), and any point z ∈ M . The deﬁnition immediately implies that the sum of two sections and the product of a section and a smooth function are again (smooth) sections. These operations turn the totality of all smooth sections of a vector bundle π into a C ∞ (M )-module, called the module of sections and denoted by Γ(π). The next lemma clariﬁes the relationship between the global and the pointwise approaches to the sections of a vector bundle. As before, we denote by μz the maximal ideal of the algebra C ∞ (M ), deﬁned by μz = {f ∈ C ∞ (M ) f (z) = 0} and called the ideal of the point z. 12.8 Lemma. Let π be a vector bundle over a manifold M and z ∈ M . Then (a) for any point y ∈ πz there is a section s ∈ Γ(π) such that s(z) = y; (b) if s ∈ Γ(π) and s(z) = 0, then there exist functions fi ∈ μz and sections si ∈ Γ(π) such that s can be written as a finite sum s = fi si . (a) For a trivial bundle, the assertion evidently holds. Therefore, by local triviality, there is a neighborhood U of z and a section sU ∈ Γ(π U ) satisfy ing sU (z) = y. In order to obtain a global (i.e., deﬁned over all M ) section of the bundle π possessing the same property, it remains to multiply sU by a smooth function whose support is contained in U and that has value 1 at the point z (see Chapter 2). (b) First suppose that the bundle is trivial. In this case (see Section 12.3) there are sections e1 , . . . , em ∈ Γ(π) whose values at every point z ∈ M form a basis of the linear space πz . A given section s can be expanded over the basis: m fi ei . s= i=1

Vector Bundles and Projective Modules

175

The equality s(z) = 0 implies that fi (z) = 0 for all i = 1, . . . , m; therefore, fi ∈ μz , as required. In the case of an arbitrary bundle, there is a neighborhood U of the point z, sections ei ∈ Γ π U , and functions fi ∈ C ∞ (U ) satisfying the equality m sU = i=1 fi ei . Choose a function f ∈ C ∞ (M ) such that supp f ⊂ U and f (z) = 1. Extending the function f fi ∈ C ∞ (U ) (respectively, the section f ei ∈ Γ π U ) as the identical zero function outside of U , we obtain a smooth function (respectively, section) on the entire manifold M . Keeping the notation f fi and f ei for the extensions, we can write f 2s =

m

(f fi )(f ei ),

i=1

and therefore

s= 1−f

2

s+

m

(f fi )(f ei ).

i=1

It remains to note that the functions 1 − f 2 , f f1 , . . . , f fm belong to μz . The lemma just proved above has a compact reformulation in terms of exact sequences. Recall that a sequence of A-modules αi−1

α

· · · → Pi−1 → Pi →i Pi+1 → · · · is said to be exact at the term Pi if Ker αi = αi−1 . The sequence is called exact if it is exact at every term. 12.9 Corollary. For any vector bundle π, the sequence 0 → μz Γ(π) → Γ(π) → πz → 0, where the first arrow is the inclusion, while the second assigns to every section its value at point z ∈ M , is exact. Hence Γ(π)/μz Γ(π) ∼ = πz . To every element h ∈ |A| we can assign the ideal μh = Ker h ⊂ A. The above result justiﬁes the following deﬁnition: The fiber Ph of an A-module P over a point h ∈ |A| is the quotient module P/μh P . The value ph of an element p ∈ P at the point h is the image of p under the natural projection P → Ph . For the case in which A = C ∞ (M ) and h = hz for z ∈ M , we def

can write Pz = P/μz P in the same sense and speak of the value pz of an element p ∈ P at the point z. Exercise. Show that for the module of vector ﬁelds P = D(M ) over the algebra of smooth functions A = C ∞ (M ), we have D(M )z = Tz M , and the value of an element X ∈ D(M ) at the point z is just the vector of the ﬁeld X at this point (see 9.39). (In other words, the notation Xz in both cases has the same meaning.)

176

Chapter 12

12.10. For the analysis of modules of sections, the following fact is important: Proposition. Suppose that the sections s1 , . . . , sl ∈ Γ(π) have the property that for every point z ∈ M the vectors s1 (z), . . . , sl (z) span the fiber πz . Then these sections generate the module Γ(π). Let k be the dimension of the bundle π. For an ordered set of integers I = (i1 , . . . , ik ), 1 i1 < · · · < ik l, put UI = {z ∈ M | si1 (z), . . . , sik (z) ∈ πz are linearly independent}. Evidently, the set UI is open, and the sections si1 |UI , . . . , sik |UI generate the C ∞ (UI )-module Γ(π|UI ). Moreover, I UI = M . Indeed, for any point z ∈ M , we can choose a basis si1 (z), . . . , sik (z) among the vectors s1 (z), . . . , sl (z) that span the ﬁber πz . This means that z ∈ UI . For a section s ∈ Γ(π), we have s|UI =

k

λI,α ∈ C ∞ (UI ).

λI,α siα |UI ,

α=1 ∞

Now let μI ∈ C (M ) be a function which is strictly positive inside UI , vanishes outside of this set, and has the property that the functions μI λI,i inside UI , νI,i = 0 outside UI , are smooth on M . Then the function μ = I μI is everywhere positive on M and μI s = νI,i si . i

Therefore, s=

νI,i 1 si . μI s = μ μ I

I,i

12.11. Geometrization of modules. With every A-module P over a commutative K-algebra A we can associate a geometrical object

|P | = Ph (or |P | = Pz , if A = C ∞ (M )), h∈|A|

z∈M

together with a natural projection onto |A|: π

P |P | ⊃ Ph ph → h ∈ |A|.

The A-module Ph = P/μh P can be also viewed as a module over A/μh , hence, by virtue of the isomorphism A/μh = K, as a K-module. The projection πP looks very much like a bundle and, as we shall show below, is equivalent to a vector bundle if A = C ∞ (M ) and the module P

Vector Bundles and Projective Modules

177

is projective and ﬁnitely generated. We shall refer to such projections as pseudo-bundles. In the case P = D(M ), every element X ∈ D(M ) corresponds to a section sX : z → Xz ∈ Tz M = D(M )z of the tangent bundle πT . This construction is of general nature and can be used for arbitrary pseudobundles by assigning the map sp : |A| → |P |,

h → ph ,

to an element p ∈ P . This allows us to visualize the elements of an arbitrary module P as sections of the pseudo-bundles |P | much in the same way as the elements of an arbitrary algebra A were viewed as functions on its spectrum |A|. One of the main goals of the present chapter is to show that vector bundles are obtained from projective modules just as smooth manifolds are obtained from smooth algebras. Maps sp : |A| → |P | are referred to as sections of the pseudo-bundles πP . (There is no other way to distinguish a reasonable class among all maps s : |A| → |P | such that πP ◦ s = id|A| .) The set Γ(P ) of all sections of the pseudo-bundles πP forms an A-module with respect to the natural operations (sp1 + sp2 ) = sp1 +p2 , p1 , p2 ∈ P, (asp ) = sap , a ∈ A, p ∈ P. To every A-module P we thus assign the A-module Γ(P ) of sections of the pseudo-bundles πP . Our aim can now be stated more precisely: We want to show that for A = C ∞ (M ) and every projective ﬁnitely generated A-module P , the pseudo-bundles πP is a vector bundle and the two modules P and Γ(P ) are naturally isomorphic. Exercise. Prove that the assignment P → Γ(P ) is a functor in the category of A-modules. ∞ If P is a C (M )-module and its element p ∈ P belongs to the intersection z∈M μz P , then the value of p at every point z ∈ M is zero. Such elements can be called invisible, or unobservable. Indeed, by the principle of observability, the class p mod μz P should be viewed as a certain component of the inner structure of the point z ∈ M , and the fact that p belongs to all subspaces μz P means that this component is unobservable. A C ∞ (M )-module P is said to be geometrical, if z∈M μz P = 0, i.e., if all elements of P are observable.

Exercise. Prove that P is geometrical if and only if the two modules P and Γ(P ) are isomorphic. The algebraic paraphrase of the above discussion is as follows. The map Γ = ΓP : P → P μz P = Γ(P ) z∈M

178

Chapter 12

kills all unobservable elements of P . Therefore, the quotient module Γ(P ) deﬁned in this way can be called the geometrization of P . The assignment P → Γ(P ) deﬁnes a functor from the category Mod C ∞ (M ) of all C ∞ (M )modules to the category GMod C ∞ (M ) of geometrical C ∞ (M )-modules. In some situations, it is suﬃcient to use the smaller category GMod C ∞ (M ) instead of the bigger category Mod C ∞ (M ). Exercise. Show that the subcategory GMod C ∞ (M ) ⊂ Mod C ∞ (M ) is stable under the operations and Hom: if P and Q are both geometrical C ∞ (M )-modules, then the modules P ⊗ Q and HomC ∞ (M ) (P, Q) are also geometrical. The behavior of the ﬁber Ph when the point h ∈ |A| varies provides important information about the module P . For example, one can speak of the support of a module, supp P = {h ∈ |A| Ph = 0} ⊂ |A|, where the bar means closure in the Zariski topology. Exercises. 1. The tangent space Tz M at a point z ∈ M can be considered as a C ∞ (M )-module with multiplication deﬁned by the rule (f, ξ) → f (z)ξ, f ∈ C ∞ (M ), ξ ∈ Tz M . Show that the support of this module consists of one point, namely z. 2. Prove that the support of the C ∞ (M )-module D(M, N ) (module of vector ﬁelds along a submanifold N ⊂ M , see Section 9.46) coincides with N . The geometrization of A-modules helps to visualize and thus better understand various algebraic constructions. For example, the structure of an A-module homomorphism f : P → Q is displayed through the family of its values at diﬀerent points of the spectrum of |A|. By the value of F at the point h ∈ |A| we understand the map of quotient modules Fh : Ph → Qh , well deﬁned because F (μh P ) ⊂ μh Q. In the geometrical situation, when A = C ∞ (M ), we can use the notation Fz , Pz , Qz instead of Fhz , Phz , Qhz . 12.12. Topology in |P |. The set |P | can be turned into a topological space by using an appropriate generalization of the ideas used in Chapter 9 to prove that the cotangent manifold T ∗ M is the R-spectrum of the symbol algebra S∗ . In the situation under study, a natural candidate for the role of such an algebra is the symmetric algebra S(P ∗ ) of the module P ∗ = HomA (P, A). By deﬁnition, Sk (P ∗ ), S(P ∗ ) = k0

where Sk (P ∗ ) is the kth symmetric power of the module P ∗ . An element f ∈ P ∗ = S1 (P ∗ ) can be viewed as a function on |P | by setting def

f (ph ) = f (p)

mod μh ∈ A/μh = K,

h ∈ |A|, p ∈ P.

Vector Bundles and Projective Modules

179

Since f is a homomorphism, the value f (ph ) does not depend on the choices made. For a generic element f1 ⊗ . . . ⊗ fk ∈ (P ∗ )⊗k , where (P ∗ )⊗k denotes the kth tensor power of the A-module P ∗ , we put def

(f1 ⊗ · · · ⊗ fk )(ph ) = f1 (ph ) · · · fk (ph ) ∈ K. From this formula, elements of (P ∗ )⊗k can be understood as symmetric functions on |P |. For an element ω ∈ (P ∗ )⊗k of the form ω = f1 ⊗ · · · ⊗ fi ⊗ . . . ⊗ fj ⊗ · · · ⊗ fk − f1 ⊗ · · · ⊗ fj ⊗ · · · ⊗ fi ⊗ · · · ⊗ fk the corresponding function ω is identically zero. The quotient algebra of the complete tensor algebra (P ∗ )⊗k (P ∗ )⊗ = k0

over the ideal generated by such elements is the symmetric algebra S(P ∗ ). Below, we discuss this idea in detail for modules over the algebra of smooth functions C ∞ (M ) and show that for any vector bundle π, the two spaces |Γ(π)| and Eπ coincide. Denote by F(|P |) the K-algebra of functions on |P | that correspond to elements of the algebra S(P ∗ ). With the help of this algebra, we can turn the set |P | into a topological space with the Zariski topology, in which the basic closed sets are zero sets of functions belonging to F(|P |). Exercise. Show that the maps πP : |P | → |A| and sp : |A| → |P |, p ∈ P , are continuous in this topology. Using the Zariski topology in |P |, one can widen the class of sections of the pseudo-bundles πP . Namely, a continuous section of πP is a continuous map s : |A| → |P | such that πP ◦s = id|A| . The set of all continuous sections of πP will be denoted by Γ0 (P ). Exercise. Show that the structure of a K-linear space in each ﬁber Ph ⊂ |P | induces the structure of an A-module in Γ0 (P ). 12.13. The functor of sections. The assignment π → Γ(π) that associates the module of sections Γ(π) to a given vector bundle π can be made into a functor from the category of A-modules to itself as follows. Let α ∈ Mor(π, η). Put Γ(α)(s) = α ◦ s for any s ∈ Γ(π). Then Γ(α) : Γ(π) → Γ(η) is a C ∞ (M )-module homomorphism, and the assignment α → Γ(α) has the property of functionality: (i) Γ(idπ ) = idΓ(π) for any π, β

α

(ii) Γ(α ◦ β) = Γ(α) ◦ Γ(β) for any pair of morphisms ζ → π → η.

180

Chapter 12

Exercise. Let P = Γ(π), Q = Γ(η), α ∈ Mor(π, η), and F = Γ(α). Prove that the map Fz : Pz → Qz (see Section 12.11) is canonically identiﬁed with αz : πz → ηz under the identiﬁcations Pz = πz , Qz = ηz , described in Lemma 12.8. The study of the functor Γ that relates the geometry of vector bundles with the algebra of rings and modules is the main point of the present chapter. This functor allows one to express the geometrical properties of vector bundles and operations with them in algebraic language. Here is a simple example. Proposition. A vector bundle π is trivial if and only if the module Γ(π) is free. Indeed, choose a trivializing diﬀeomorphism ϕ : Eπ → M × V and a basis v1 , . . . , vn of the linear space V . Let ei (x) = ϕ−1 (x, vi ). Then the set {e1 , . . . , em } is a free basis of the module Γ(π). Conversely, suppose that the module Γ(π) is free with a basis e1 , . . . , em ; then we can deﬁne a diﬀeomorphism ϕ : Eπ → M × Rn by setting m λi ei (x) = (x; λ1 , . . . , λm ). ϕ i=1

Remark. For any free C ∞ (M )-module P of finite type, i.e., with a ﬁnite set of generators, there exists a vector bundle whose module of sections is isomorphic to P . Indeed, a free C ∞ (M )-module of rank m is isomorphic to Γ(π), where π is the product bundle M × Rm → M . 12.14. Projective modules. It is natural to suppose that section modules of vector bundles must possess certain speciﬁc properties originating from the fact that all ﬁbers of a given bundle are the same. These properties serve as a formalization of our doctrine that the inner structures of all points are identical (see Section 11.1). We shall see that an adequate description can be given by using the notion of projectivity. A module P over a commutative ring A is said to be projective if it has the following property: for any epimorphism of A-modules ϕ : Q → R and any homomorphism ψ : P → R there is a homomorphism χ : P → Q such that ϕ ◦ χ = ψ, i.e., the diagram Q χ

P

ϕ ψ

R

0 commutes. The homomorphism χ is called the lift of ψ along ϕ. Let us give several equivalent deﬁnitions of projectivity.

Vector Bundles and Projective Modules

181

12.15 Proposition. The following properties of an A-module P are equivalent: (a) P is projective; (b) any epimorphism ϕ : Q → P of an arbitrary A-module onto P splits, i.e., there exists a homomorphism χ : P → Q such that ϕ ◦ χ = idP ; (c) P is isomorphic to a direct summand of a free A-module; (d) the functor HomA (P, ·) : Q → HomA (P, Q) on the category of A-modules is exact, i.e., it preserves the class of exact sequences. (a) =⇒ (b). It suﬃces to set R = P and ψ = idP in the deﬁnition of projectivity. (b) =⇒ (c). Let us consider a free module Q with free basis {ep }p∈P equipotent to the set P and construct an epimorphism ϕ : Q → P by putting ϕ(ep ) = p for every p ∈ P . By virtue of (b), there is a homomorphism χ ∈ Hom(P, Q) such that χ ◦ ϕ = idP . Then P ∼ = χ and Q = χ ⊕ Ker ϕ. Indeed, any element a ∈ Q can be written as χ(ϕ(a)) + a − χ(ϕ(a)) ; here the ﬁrst summand belongs to χ and the second to Ker ϕ. On the other hand, if a ∈ χ ∩ Ker ϕ, then a = χ(p), p ∈ P , and 0 = ϕ(a) = ϕ(χ(p)) = p. Therefore, a = 0. (c) =⇒ (d). Note that for a free module R, the functor HomA (R, ·) is exact. This follows from the fact that a homomorphism of a free module R into another module is uniquely determined by its values on the basis elements, and these values can be arbitrary. Now suppose that R = P ⊕ Q and ϕk

S = {· · · → Sk → Sk+1 → · · · } is an exact sequence of A-modules. Then the sequence HomA (R, S), · · · → HomA (R, Sk ) → HomA (R, Sk+1 ) → · · · , is exact, too. This sequence decomposes into the direct sum of the sequence HomA (P, S) of the form · · · → HomA (P, Sk ) → HomA (P, Sk+1 ) → · · · , and the sequence HomA (Q, S) of the form · · · → HomA (Q, Sk ) → HomA (Q, Sk+1 ) → · · · . In other words, every term of the sequence HomA (R, S) is the direct sum of the corresponding terms of two sequences HomA (P, S) and HomA (Q, S), and every homomorphism of the sequence HomA (R, S) is the direct sum of the corresponding homomorphisms in HomA (P, S) and HomA (Q, S). It remains to apply the following simple observation: The direct sum of two sequences of modules is exact if and only if both summands are exact.

182

Chapter 12

Finally, to prove the implication (d) =⇒ (a) it suﬃces to apply the property (d) to the exact sequence ϕ

0 → Ker ϕ → Q → R → 0.

Exercise. Suppose that P ⊂ R, where P is projective and R is free. Is it true that there exists a submodule Q ⊂ R such that R = P ⊕ Q? 12.16. Examples of projective modules. I. Over a ﬁeld F, all F-modules are projective, because they are all free. II. Over the ring of integers Z, all projective modules are free (although in this case not all modules are free). III. The simplest example of a module that is projective, but not free, is provided by the group Z, considered as a module over the ring Z ⊕ Z with multiplication (a, b) · x = ax. IV. Modules of sections of vector bundles are projective. This will be proved later, in Theorem 12.32. Exercise. Describe the projective modules over the ring of residues Z/mZ and over the matrix ring. 12.17. Subbundles. We say that a vector bundle η : Eη → M is a subbundle of a vector bundle π : Eπ → M (denoted by η ⊂ π) if (i) the total space Eη is a submanifold of Eπ ; (ii) the projection η is the restriction of π to Eη ; (iii) for any point x ∈ M the ﬁber ηx is a linear subspace of the ﬁber πx . Examples. I. The zero subbundle: its total space coincides with the image of the zero section. II. The tangent bundle of the 2-sphere does not contain one-dimensional subbundles. To prove this fact, suppose that such a subbundle ξ exists. Then for a smooth oriented closed curve Γ ⊂ S 2 , we can deﬁne an integer invariant ν(Γ) equal to the number of half-turns made by the tangent vector to the curve with respect to the ﬁbers of ξ. The number ν(Γ) does not change under smooth deformations of the curve Γ, and it changes its sign when the orientation of the curve changes. If Γ+ is a small positively oriented circle and Γ− the same circle with the negative orientation, then it is evident that ν(Γ+ ) = 2 and ν(Γ− ) = −2. But on the sphere the curve Γ+ can be smoothly deformed into Γ− . This contradiction proves our assertion. 12.18. The local structure of subbundles. Suppose that at every point z ∈ M a linear subspace ηz of the ﬁber πz is given. This will deﬁne a subbundle of π if the following two properties hold: (i) the set z∈M ηz must be a submanifold of Eπ ; (ii) the local triviality property must hold for the family {ηz }.

Vector Bundles and Projective Modules

183

In the case of a trivial bundle π, these requirements can be stated in the form of the following simple lemma. 12.19 Lemma. Let π : M × Rn → M be the projection on the first factor, and let a k-dimensional linear subspace ηz ⊂ πz ∼ = Rn be given at every point z ∈ M . Denote by η : M → Gn,k the map η(z) = ηz , where the plane ηz is regarded as a point in the Grassmann manifold Gn,k . Then the following two conditions are equivalent: (a) The family {ηz }z∈M defines a subbundle η ⊂ π. (b) The map η is smooth. (a) =⇒ (b). Pick a point a ∈ M and a basis e1 , . . . , ek of the space ηa . By Lemma 12.8, there exist sections s1 , . . . , sk ∈ Γ(η) such that si (a) = ei . Since the sections si are smooth, there is a neighborhood U of a such that the vectors s1 (z), . . . , sk (z) are linearly independent for all z ∈ U . Therefore, these vectors form a basis of the space ηz . In the neighborhood U , the map η can be represented as the composition of two maps U z → s1 (z), . . . , sk (z) → L s1 (z), . . . , sk (z) ∈ Gn,k , where L s1 (z), . . . , sk (z) stands for the linear span of the vectors s1 (z), . . . , sk (z). Therefore, the map η is smooth. (b) =⇒ (a). Suppose that the plane ηz ∈ Gn,k smoothly depends on the point z. Using the standard coordinate system in Gn,k (see 11.11, VI), we can choose a basis s1 (z), . . . , sk (z) of ηz that smoothly depends on the point z, if z belongs to a certain neighborhood U of a point a ∈ M . Deﬁne a map ϕ : U × Rk → U × Rn by k λi si (z) . ϕ(z, λ1 , . . . λk ) = z, i=1

The linear independence of the vectors s1 (z), . . . , sk (z) implies that ϕ is of maximal rank. By the implicit function theorem (see Theorem 6.23 and Remark 6.24), the image ϕ is a submanifold of the space U × Rn . According to the construction of the map ϕ, this means that {ηz }z∈M deﬁnes a subbundle of π. The lemma describes the structure of subbundles of a trivial vector bundle and thus the local structure of subbundles of any vector bundle. We shall now apply the lemma to investigate the conditions under which the kernel and the image of a bundle morphism ϕ ∈ Mor(π, η) are subbundles in π andη, respectively. By the kernel (respectively, image) we understand the set z∈M Ker ϕz ⊂ Eπ , together with the restriction to this set of the projection π (respectively, the set z∈M ϕz ⊂ Eη , together with the restriction of η). 12.20 Proposition. For a vector bundle morphism ϕ : π → η over a manifold M , the following conditions are equivalent:

184

Chapter 12

(a) dim Ker ϕx does not depend on x. (b) dim ϕx does not depend on x. (c) Ker ϕ is a subbundle of π. (d) ϕ is a subbundle of η. The implications (c) =⇒ (a) and (d) =⇒ (b) are evident. The equivalence (a) ⇐⇒ (b) follows from the fact that the sum dim Ker ϕx + dim ϕx is equal to the dimension of the ﬁber of π and is thus constant. Let us prove that (b) implies (d). Since assertion (d) is local, it suﬃces to prove it in a neighborhood of an arbitrary point of the base space. Let U ⊂ M be a neighborhood of the given point such that the bundles π U and η U are trivial. Therefore, we can assume that we deal with a morphism ϕU of trivial bundles acting from πU : U × V → U to ηU : U × W → U . To this morphism there corresponds a smooth map ϕ U : U → Hom(V, W ) sending each point x to the operator ϕ(x) whose value on the vector v ∈ V is equal to the W -component of the element ϕ(x, v) ∈ M × W . By assumption, the rank of ϕ x does not depend on the point x; denote it by r. Suppose that at a given point a ∈ U the vectors v1 , . . . , vr have the property that their images under ϕa are linearly independent. Then, by continuity, there is a neighborhood of a where the vectors ϕx (v1 ), . . . , ϕx (vr ) form a basis of ϕx that smoothly depends on x. Now by Lemma 12.19, ϕ is a subbundle of η. The previous argument can also be applied to prove the implication (a) =⇒ (c). If a family of operators ϕx ∈ Hom(V, W ) smoothly depends on x and has constant rank r, then ϕx , regarded as a point of the Grassmannian GW,r , is a smooth function of x (by virtue of Lemma 12.19). Note that Ker ϕx = Ann ϕ∗x (we recall that Ann ϕ∗x denotes the set of mutual zeros of all linear functionals from ϕ∗x ). The smooth dependence of ϕ∗x on x follows from the fact that the components of this operator in appropriate bases are equal to the components of ϕx . Hence ϕ∗x is a smooth function of x. It remains to note that the map Ann : GV ∗ ,r → GV,dim V −r sending every subspace to its annihilator is smooth.

12.21. Direct sum of vector bundles. In the case of vector bundles, the construction of the direct sum (Section 11.14) must agree with the linear structure in the ﬁbers. We say that a vector bundle π is the direct sum of two subbundles η and ζ (notation: π = η ⊕ ζ) if its ﬁber over every point x ∈ M is the direct sum of two subspaces: πx = ηx ⊕ ζx . If two vector bundles η and ζ over one manifold M are given, we can construct a vector bundle π that decomposes into a direct sum of two subbundles isomorphic to η and ζ. Such a vector bundle π is deﬁned uniquely up to isomorphism; it is called the outer direct sum or Whitney sum of the

Vector Bundles and Projective Modules

185

bundles η and ζ and is also denoted by η ⊕ ζ. As in the case of arbitrary locally trivial bundles, the total space of η ⊕ ζ can be deﬁned as Eη⊕ζ = {(y, z) ∈ Eη × Eζ | η(y) = ζ(z)}, and the projection is the map sending a point (y, z) to η(y). The ﬁber of the Whitney sum over a point x ∈ M is in a natural bijection with the space ηx ⊕ ζx ; this bijection endows (η ⊕ ζ)x with the structure of a vector space. 12.22. Examples of direct sums. I. Let M be a submanifold of a Euclidean space E. Then the trivial bundle E × M → M is a direct sum of two subbundles: the tangent subbundle πT : T M → M and the normal subbundle ν : N M → M . The ﬁber νz of the normal subbundle over a point z ∈ M is by deﬁnition the orthogonal complement of the tangent space Tz M in Tz E, the latter being identiﬁed with E. It is interesting to note that, for example, for the sphere S 2 ⊂ R3 , the normal bundle ν is trivial, but the tangent bundle πT is not (since it does not have nonvanishing sections; see Example 11.11, III). We see that the direct sum of a trivial and a nontrivial bundle can be trivial, a fact seems quite unexpected at ﬁrst glance. II. The Whitney sum of the M¨ obius band with the trivial one-dimensional (unit) bundle is nontrivial. Indeed, the total space of this sum is the product [0, 1]×R2 ⊂ R3 with points (0, y, z) and (1, −y, z) identiﬁed for any y, z ∈ R. This manifold is nonorientable and therefore is not diﬀeomorphic to S 1 ×R2 . 12.23 Proposition. If π = η ⊕ ζ, then Γ(π) = Γ(η) ⊕ Γ(ζ) (direct sum of submodules). The assertion follows from (11.6) by virtue of the natural identiﬁcations Γ(η) × Γ(ϕ) and Γ(η) ⊕ Γ(ϕ). The following proposition is important because of its relationship with the property of projectivity. 12.24 Proposition. Every subbundle of a vector bundle has a direct complement. The proof is based on a standard technical trick: the introduction of a scalar product. A scalar product on a vector bundle π is by deﬁnition a scalar product in every ﬁber x ∈ M , smoothly depending on x. The smoothness requirement can be stated as follows: the scalar product of any two smooth sections is a smooth function. Remark. After studying the construction of the tensor product of vector bundles (Section 12.35), the reader will see that the scalar product on a vector bundle π is nothing but a smooth function on the manifold Eπ⊗π whose restriction to every ﬁber is linear and positive deﬁnite. Example. A scalar product on the tangent bundle is the same thing as a Riemannian metric on the given manifold. 12.25 Lemma. On any vector bundle there exists a scalar product.

186

Chapter 12

To prove the lemma, note that, for a trivial bundle, the problem has a trivial solution: it suﬃces to supply every ﬁber with one and the same scalar product. Now let {Ui }i∈I be a covering of the manifold M with open sets, each of which trivializes π, and let gi be a scalar product on the vector bundle π U . Choose a partition of unity {ei }i∈I , subjected to the covering i {Ui }i∈I (see 4.18), and for y1 , y2 ∈ πx set g(y1 , y2 ) = ei (x)gi (y1 , y2 ), where the summation ranges over all indices i for which x ∈ Ui . All the necessary properties of the function g can be veriﬁed in a straightforward way. Let us continue the proof of Proposition 12.24. Let π be a bundle over M and η ⊂ π. Choose a scalar product g on π. Then in the ﬁber over any point x ∈ M , we can consider the subspace ηx⊥ , the orthogonal complement to η in π with respect to the scalar product g. We only have to check that ηx⊥ smoothly depends on x. Since this is a local property, we can assume that the bundle π is trivial, π : M × V → M . Then we have smooth maps η : M → GV,k (see 12.19) and g : M → (V ⊗ V )∗ (scalar product). Denote by N ⊂ (V ⊗ V )∗ the set of all symmetric positive deﬁnite bilinear forms. The map GV,k × N → GV,n−k sending a pair (L, ϕ) to the orthogonal complement of L with respect to the scalar product ϕ is smooth. The map x → ηx⊥ is the composition M → M × M → GV,k × N → GV,n−k , where the ﬁrst arrow is the diagonal map x → (x, x), the second arrow is the direct product of the maps η and g, and the third arrow is the map introduced above. The resulting map is smooth, and by Lemma 12.19, the proposition is proved. 12.26. Pullback vector bundles. The geometrical construction of the pullback vector bundles does not diﬀer from the same construction in the general case (see 11.16). Example. Let π : E → M be an arbitrary bundle and suppose that λ = πT ∗ : T ∗ M → M is the cotangent bundle of its base space. Then the sections of the pullback bundle π ∗ (λ) are called horizontal 1-forms on the total space E (see Section 14.4). Pay attention to the fact that Γ(π ∗ (λ)) is embedded into the module of 1-forms on the manifold E. The following assertion is one of the key facts in the theory of vector bundles. 12.27 Theorem. Every vector bundle with connected base can be obtained as the pullback of the tautological bundle over an appropriate Grassmann manifold. Let η : Eη → M , dim M = n, dim Eη = n + k. Denote by oz : ηz → Tz (ηz ) ⊂ Tz (Eη )

Vector Bundles and Projective Modules

187

the canonical identiﬁcation of the vector space ηz with its tangent space at zero. By Whitney’s theorem, there exists an immersion φ : Eη → R2(n+k) of the manifold Eη into a Euclidean space. This means that all diﬀerentials dy φ : Ty (Eη ) → Tφ(y) R2(n+k) ,

y ∈ Eη ,

are injective.

R2(n+k)

Eη

M

φ

G2(n+k),k Figure 12.1. The Gauss map.

The Gauss map g : M → G2(n+k),k assigns to every point z ∈ M the k-dimensional subspace of the tangent space TO (R2(n+k) ) ∼ = R2(n+k) which is the image of the subspace Tz (ηz ) ⊂ Tz (Eη ) under the map dz (rφ(z) ◦ φ) : Tz (Eη ) → TO R2(n+k) (see Figure 12.1). Here z is understood as a point of the manifold Eη , the zero element in the ﬁber ηz , and ra : v → v − a (for a, v ∈ R2(n+k) ) is the translation of the space R2(n+k) by the vector −a. The map g is covered by the morphism of vector bundles γ : η → Θ2(n+k),k ,

γz = dz (rφ(z) ◦ φ) ◦ oz ,

where Θ2(n+k),k denote the tautological bundle (see p. 160). For every point z ∈ M the map γz is an isomorphism of the ﬁber ηz onto the ﬁber at g(z) ∈ G2(n+k),k . Now Theorem 11.19 shows that the bundles η and g ∗ (Θ2(n+k),k ) are isomorphic. 12.28 Corollary. The C ∞ (M )-module Γ(π), π : Eπ → M , has a system of generators consisting of no more than N elements, where N = N (n, k) is a natural number that depends only on the dimensions of the base and the fiber (n and k respectively) of the bundle π. Note that the sections sI,i of the tautological bundle Θm,l described in Example V of Section 11.12 satisfy the assumptions of Proposition

188

Chapter 12

12.10. Therefore, the sections f(sI,i ) of any pullback bundle f ∗ (Θm,l ) (see Section 11.16) also satisfy these assumptions and thus generate the module Γ(f ∗ (Θm,l )). Theorem 12.27 implies that any k-dimensional vector bundle over a manifold of dimension n is the pullback of the tautological bundle Θ2(n+k),k via the Gauss map. By Proposition 12.10, the sections g(sI,i ), where 1 i k, I = {i1 , . . . , ik } ⊂ {1, . . . , 2(n + k)}, generate the module Γ(π). The number of these sections is 2(n + k) k = N (n, k). k We have completed all preparations necessary to state and prove the two main theorems of this chapter (Theorems 12.29 and 12.32), which give an exhaustive description of the section modules of vector bundles. 12.29 Theorem. For any pair π, η of vector bundles over a manifold M , the section functor Γ determines a one-to-one correspondence Mor(π, η) ∼ = HomC ∞ (M ) (Γ(π), Γ(η)). We must prove that for any C ∞ (M )-homomorphism of modules F : Γ(π) → Γ(η) there exists a unique bundle morphism ϕ : π → η such that Γ(ϕ) = F . First we prove the uniqueness. Suppose that ϕ, ψ ∈ Mor(π, η) and Γ(ϕ) = Γ(ψ). This means that ϕ, ψ : Eπ → Eη and ϕ ◦ s = ψ ◦ s for any section s ∈ Γ(π), i.e., ϕ(s(x)) = ψ(s(x)) for all s ∈ Γ(π) and all x ∈ M. According to Lemma 12.8(a), every point of Eπ can be represented as s(x), which implies that ϕ = ψ. Now suppose that a homomorphism F : Γ(π) → Γ(η) is given, and we must deﬁne the corresponding map ϕ : Eπ → Eη , i.e., we must deﬁne its value ϕ(y) ∈ Eη at an arbitrary point y ∈ Eπ . The point y belongs to a certain ﬁber: y ∈ πx . By Lemma 12.8(a), we can choose a section s ∈ Γ(π) such that s(x) = y and put ϕ(y) = F (s)(x). We must check (a) that ϕ is well deﬁned, (b) that ϕ is a bundle morphism, and (c) the equality Γ(ϕ) = F . (a) Let s1 and s2 be two sections of π such that s1 (x) = s2 (x). Lemma 12.8(b) shows that s1 − s2 ∈ μx Γ(π). Therefore, F (s1 − s2 ) ∈ μx Γ(η) and F (s1 )(x) = F (s2 )(x). (b) The only thing worth verifying here is the smoothness of the map ϕ : Eπ → Eη . It is suﬃcient to prove that the map ϕU = ϕπ−1 (U ) is smooth

Vector Bundles and Projective Modules

189

for an arbitrary open set U ⊂ M , over which both π and η are trivial. Note that ϕU (t(x)) = FU (t)(x), (12.1) where t ∈ Γ(π U ) and FU : Γ(π U ) → Γ(η U ) is the localization of the homomorphism F on the subset U (that is, over the multiplicative system of functions that do not vanish at the points of U ; see Section 11.6). Choosing some bases of the free C ∞ (U )-modules Γ(π U ) and Γ(η U ), we can write the homomorphism FU as a matrix over the ring C ∞ (U ). By virtue of equation (12.1), the same matrix gives the coordinate representation of the morphism ϕU (Lemma 12.5). Since the elements of the matrix belong to C ∞ (U ), the map ϕU is smooth. (c) For any s ∈ Γπ , we have (Γ(ϕ)(s))(x) = (ϕ ◦ s)(x) = ϕ(s(x)) = F (s)(x), i.e., Γ(ϕ)(s) = F (s).

12.30 Lemma. If ϕ : ζ → π is a vector bundle morphism and ϕz is an ∼ πz for any point z ∈ M, then ϕ is a isomorphism of vector spaces ζz = bundle isomorphism. The only thing we must check is that the inverse map ϕ−1 : Eπ → Eζ is smooth. By the inverse function theorem (Section 6.21), it suﬃces to show that the diﬀerential dy ϕ at any point y ∈ Eζ is an isomorphism of the corresponding tangent spaces. Since the dimensions of the manifolds Eζ and Eπ are equal, the dimensions of the tangent spaces Ty Eζ and Tϕ(y) Eπ are equal too. Therefore, for the diﬀerential dy ϕ to be an isomorphism, it is suﬃcient that it be injective. To check the latter, note that we must have dϕ(y) π ◦ dy ϕ = dy ζ, because π ◦ ϕ = ζ. Now suppose that dy ϕ(v) = 0. Then dyζ(v) = 0, which means that v ∈ Ty (ζz ), where z = ζ(y). By assumption, ϕζz = ϕz is an isomorphism between the ﬁbers ζz and πz . This implies v = 0. 12.31 Corollary. If F ∈ HomC ∞ (M ) (Γ(ζ), Γ(π)) and for any point x ∈ M the induced map Fx : Γ(ζ)/μx Γ(ζ) → Γ(π)/μx Γ(π) is an isomorphism of vector spaces, then F is a module isomorphism. This fact is easily reduced to the lemma just proved. Indeed, we have F = Γ(ϕ) for an appropriate morphism ϕ ∈ Mor(ζ, π) and Fx = ϕx at every point x ∈ M . 12.32 Theorem. Suppose M is a connected smooth manifold and let P be a C ∞ (M )-module. Then P is isomorphic to the module of sections Γ(π) of a smooth vector bundle π over M if and only if P is finitely generated and projective.

190

Chapter 12

(i) Recall, ﬁrst of all, that for any vector bundle π over M , the module Γ(π) is finitely generated (Corollary 12.28). (ii) Let us prove that the module Γ(π) is projective. According to (i), there exists a ﬁnite system of sections s1 , . . . , sN that generate the module Γ(ξ). Let Q be a free C ∞ (M )-module of rank N with generators e1 , . . . , eN and let F : Q → Γ(ξ) be the homomorphism given by F (ei ) = si for i = 1, . . . , N . By construction, F is epimorphic. Note that Q = Γ(η) for a certain trivial vector bundle η; hence by Theorem 12.29 there is a bundle morphism ϕ ∈ Mor(π, η) such that F = Γ(ϕ). Lemma 12.8 implies that at every point z ∈ M the map ϕz is surjective, since F is an epimorphism. Therefore, we can apply Proposition 12.20 and infer that Ker ϕ is a subbundle in η. By Proposition 12.24, there is a direct decomposition η = Ker ϕ ⊕ ζ for an appropriate subbundle ζ of π. Proposition 12.23 implies that Γ(η) = Γ(Ker ϕ) ⊕ Γ(ζ); i.e., the module Γ(ζ) is a direct summand of a free module. By Proposition 12.15, Γ(ζ) is projective. Now let us check that the map ϕ restricted to the total space of the subbundle ζ gives an isomorphism between ζ and π. Indeed, ϕz : ζz → πz is a linear isomorphism for any point z ∈ M , and we can use Lemma 12.30. The isomorphism of bundles ζ ∼ = π implies the isomorphism of modules Γ(ζ) ∼ = Γ(π) and thus assertion (ii). (iii) We shall now prove that any projective module of finite type is isomorphic to the module of sections of a smooth vector bundle. Suppose that P is a projective C ∞ (M )-module with a ﬁnite number of generators. Then (see Proposition 12.15 and the remark in Section 12) we can write Γ(η) = P ⊕ Q, where η is the trivial bundle over M , P and Q are submodules of Γ(η), and P ∼ = P . Since we are considering P only up to isomorphism, in the sequel by abuse of notation we suppress the prime and write P instead of P . Let Pz = {p(z) | p ∈ P }. This is an R-linear subspace in ηz . The subspace Qz is deﬁned similarly. We claim that ηz = Pz ⊕ Qz . Indeed, let y ∈ ηz . Choose a section s ∈ Γ(η) such that s(z) = y and represent it as p + q, where p ∈ P , q ∈ Q. Then y = p(z) + q(z) ∈ Pz + Qz . On the other hand, suppose that y ∈ Pz ∩ Qz , i.e., y = p(z) = q(z), where p ∈ P , q ∈ Q. Then (p − q)(z) = 0, hence by Lemma 12.8(b) we have fi si = fi pi + fi qi p−q = i

i

i

for a certain choice of fi ∈ μz , pi ∈ P , qi ∈ Q. Since P ∩ Q = 0, the last equation implies p = i fi pi . Therefore, p(z) = 0, i.e., y = 0, and thus ηz = Pz ⊕ Qz . We must verify that the union of all subspaces Pz constitutes a subbundle in η and that P coincides with the module of its (smooth) sections. We shall

Vector Bundles and Projective Modules

191

show ﬁrst that dim Pz does not depend on z. Let dim Pz = r for some point z ∈ M and let p1 , . . . , pr ∈ P be a set of sections whose values at z span the subspace Pz . The continuity of sections implies the linear independence of the vectors p1 (y), . . . , pr (y) for all points y in a neighborhood U of the point z. Therefore, dim Py dim Pz . A similar argument for the submodule Q shows that dim Qy dim Qz in a neighborhood of z. Since the sum dim Py + dim Qy is constant, we see that dim Py is a locally constant function of y. Since M is connected, it is a global constant. Having in mind Lemma 12.5, we shall now prove that the subspace Pz , viewed as a point of GV,r , where V is the ﬁber of η, smoothly depends on z (here we can assume that η is a trivial bundle, because the problem under consideration is local). Indeed, let p1 (a), . . . , pr (a) be a basis of the space Pa at a certain point a ∈ M . Then the vectors p1 (z), . . . , pr (z) form a basis of the linear space Pz for all points z belonging to some neighborhood of a. We see that the family Pz locally has a basis that smoothly depends on z and thus represents a smooth family of points in the Grassmannian GV,r . Denote the bundle with ﬁbers Pz by π. By construction, P ⊂ Γ(π). Let us prove the reverse inclusion. If s ∈ Γ(π) ⊂ Γ(η), then there are elements p ∈ P and q ∈ Q such that s = p + q. Since Pz ∩ Qz = 0, the equation p(z) + q(z) = s(z) implies that q(z) = 0 for any z ∈ M . Hence q = 0 and s = p ∈ P. We have shown that P = Γ(π). This completes the proof of the theorem. Remark. Our proof shows that the module Γ(ξ) is projective also in the case of a disconnected base manifold. On the other hand, any projective module over C ∞ (M ) is evidently reduced to the direct sum of modules Γ(πα ), where πα is a vector bundle over a connected component Mα of the manifold M . The dimension of the bundle πα may vary between the connected components. 12.33. Equivalence of the two categories. Theorems 12.29 and 12.32, taken together, establish the equivalence of the category VBM of vector bundles over the manifold M and the category Modpf C ∞ (M ) of projective ﬁnite-type modules over the algebra C ∞ (M ). This result is in full parallel with the result of Section 7.19 about the equivalence between the category of smooth manifolds and the category of smooth R-algebras. It can be used in either direction, i.e., by applying algebra to geometry and vice versa. Here is a simple example: Proposition. For any vector bundle π there is a vector bundle η such that the direct sum π ⊕ η is a trivial bundle. This fact, which plays a fundamental role in K-theory is surprising from the geometrical viewpoint, can be reduced to the mere deﬁnition of projective modules by the application of Theorem 12.32.

192

Chapter 12

Below (in Section 12.38) we shall give an example of an algebraic statement (the tensor square of a one-dimensional projective module over C ∞ (M ) is isomorphic to C ∞ (M )), which becomes evident after a geometrical trick (introduction of a scalar product on the corresponding vector bundle). In the next sections, we discuss two operations on vector bundles, namely the construction of tensor products and pullback bundles, and the corresponding operations on projective modules. 12.34 Proposition. The tensor product of two projective modules over a commutative ring is a projective module. The well-known property (whose proof we leave to the reader as an algebraic exercise) of Hom−⊗-associativity reads HomA (P ⊗A Q, R) ∼ = HomA (P, HomA (Q, R)) for any three A-modules P , Q, R. In other words, this means that the functor HomA (P ⊗ Q, ·) is the composition of functors HomA (P, ·) and HomA (Q, ·). It remains to use the equivalence of (a) and (d) from Proposition 12.15. 12.35. Tensor product of vector bundles. Let η and ζ be vector bundles over a manifold M . The ﬁber of the new bundle π = η ⊗ ζ (the tensor product of η and ζ) over a point z ∈ M is, by deﬁnition, the linear space πz = ηz ⊗ ζz . The structure of smooth manifold on the total space Eπ = z∈M πz is introduced in the following way. Every point of the base manifold M has a neighborhood U over which both bundles η and ζ are trivial. The sets π −1 (U ) form a covering of the space Eπ . Let V be the “outer” ﬁber of the bundle η, let W be the “outer” ﬁber of the bundle ζ, while ϕ : η −1 (U ) → U × V and ψ : ζ −1 (U ) → U × W are the corresponding trivializing diﬀeomorphisms. The chart χ : π −1 (U ) → U × (V ⊗ W ) is constructed as follows. Let u ∈ π −1 (U ). Then u ∈ ηz ⊗ ζz for some point z ∈ M . The maps ϕ and ψ, restricted to the ﬁber over z, give isomorphisms ηz → V and ζz → W and hence an isomorphism ηz ⊗ ζz → V ⊗ W . Denote the image of an element u under this isomorphism by v and put χ(u) = (z, v). Exercise. Prove that these charts form a smooth atlas on Eπ . The bundle η ⊗ ζ is thus deﬁned. Note that it satisﬁes the axiom of local triviality by construction. For brevity, we denote by η ⊗k the kth tensor power of a bundle η. Example. If πT and πT ∗ are the tangent and cotangent bundles of a manifold, then πT⊗k ⊗ πT⊗l∗ is the bundle of tensors of type (k, l) (k times contravariant and l times covariant).

Vector Bundles and Projective Modules

193

12.36 Proposition. Let P and Q be projective modules over a commutative ring A. Then the module HomA (P, Q) is also projective. If both P and Q are finitely generated, then the module HomA (P, Q) is also of finite type. If R and S are (ﬁnitely generated) free modules, then HomA (R, S) is also (ﬁnitely generated and) free. Indeed, let {ri } and {sj } be the free generators of R and S, respectively. Then the A-homomorphisms hi,j ∈ HomA (R, S), deﬁned by the rule hi,j (ri ) = δi,j sj , where δi,j is the Kronecker delta, constitute a basis of HomA (R, S). Now suppose that R and S are free A-modules that contain the modules P and Q, respectively, as direct summands. Let αP : P → R

and

βP : R → P,

αP ◦ βP = idP ,

(and similarly with αQ and βQ for Q) be the injections and projections that realize the corresponding decompositions of R and S into direct sums. Then the A-homomorphism HomA (P, Q) h → αQ ◦ h ◦ βP ∈ HomA (R, S) is the injection of HomA (P, Q) into the free module HomA (R, S), whose image, together with the kernel of the projection HomA (R, S) H → βQ ◦ H ◦ αP ∈ HomA (P, Q), turns the module HomA (P, Q) into a direct summand in the free A-module HomA (R, S). There is a natural map (12.2) ι : P ∗ ⊗A Q → HomA (P, Q), where P ∗ = HomA (P, A), deﬁned by the formula ι(p∗ ⊗ q)(p) = p∗ (p)q. If the modules P and Q are projective and ﬁnitely generated, then ι is an isomorphism. This fact has an evident proof for ﬁnitely generated free modules and can be generalized to arbitrary projective modules of ﬁnite type by an argument similar to the proof of Proposition 12.36. 12.37. To every vector bundle ζ, we can associate the dual bundle ζ ∗ , dual to ζz . The whose ﬁber ζz∗ over a point z ∈ M is the vector space precise construction of a smooth atlas on the set Eζ ∗ = z∈M ζz∗ repeats the corresponding construction for the case of the cotangent bundle from the tangent bundle given in Section 9.24. Example. πT∗ = πT ∗ . For any vector bundle ζ, there is a natural pairing Γ(ζ) × Γ(ζ ∗ ) → C ∞ (M ),

def

(s, s∗ )(z) = (s(z), s∗ (z)),

where s ∈ Γ(ζ), s∗ ∈ Γ(ζ ∗ ), z ∈ M . Exercise. Using local triviality, verify that (s, s∗ )(z) is a smooth function for any smooth sections s and s∗ .

194

Chapter 12

For the bundle ζ = πT M , this pairing, by the equality (πT M )∗ = πT ∗ M , turns into the pairing Γ(πT M ) × Γ(πT ∗ M ) → C ∞ (M ). A metric on ζ allows us to identify ζz and ζz∗ for any point z ∈ M . Lemma 12.30 implies that vector bundles ζ and ζ ∗ are isomorphic. Given a pair of vector bundles η and ζ, one can deﬁne the bundle Hom(η, ζ) with ﬁber HomR (ηz , ζz ) over every point z ∈ M in the same way. The construction follows that of the tensor product of two vector bundles (see Section 12.35). Another possibility for deﬁning the bundle Hom(η, ζ) is to explicitly reduce it to the constructions of the tensor product and the dual bundle, using the natural isomorphism of vector spaces HomR (ηz , ζz ) = ηz∗ ⊗ ζz . 12.38. Example. The tensor square of the M¨ obius band, viewed as a one-dimensional vector bundle over the circle (Example 12.6, III), is IS 1 , the trivial one-dimensional bundle. This result can either be understood directly from the construction of the M¨ obius band and the deﬁnition of the tensor product (we recommend to the reader to carry it out) or be deduced from a more general fact: Proposition. The tensor square of any one-dimensional bundle over an arbitrary manifold is isomorphic to the unit bundle. To prove the latter, we use the above-mentioned isomorphisms ζ ⊗ζ ∼ = Hom(ζ, ζ). = ζ∗ ⊗ ζ ∼ If dim ζ = 1, then the bundle Hom(ζ, ζ) is trivial: the trivializing diﬀeomorphism ϕ : M × R → EHom(ζ,ζ) can be deﬁned by the formula ϕ(a, λ)y = λy for all a ∈ M , λ ∈ R, y ∈ π −1 (a). Note, ﬁnally, that tensor multiplication introduces a group structure into the set V 1 (M ) of equivalence classes of one-dimensional vector bundles over the manifold M . The order of any nonunital element of this group is 2. Thus, V 1 (S 1 ) = Z2 , the two elements being represented by the trivial bundle and the M¨ obius band. In the next theorem, we establish the relationship between the functors Γ, ⊗, and Hom. 12.39 Theorem. The functor Γ preserves tensor products and homomorphisms, i.e., Γ(π ⊗ η) ∼ = Γ(π) ⊗ Γ(η), Γ(Hom(π, η)) ∼ = HomC ∞ (M ) (Γ(π), Γ(η)), for any two vector bundles π and η over the manifold M (here and below, tensor products of C ∞ (M )-modules are computed over the ring C ∞ (M )).

Vector Bundles and Projective Modules

195

The natural isomorphisms Hom(π, η) ∼ = π ∗ ⊗ η,

HomC ∞ (M ) (Γ(π), Γ(η)) ∼ = Γ(π ∗ ) ⊗ Γ(η)

(see Section 12.36) together with the natural identiﬁcations (π ∗ )∗ = π, Γ((π)∗ )∗ = Γ(π) show that it suﬃces to prove the assertion only for tensor products. We shall construct a map from Γ(π) ⊗ Γ(η) to Γ(π ⊗ η) and verify that it is an isomorphism. Let s ∈ Γ(π), t ∈ Γ(η). Deﬁne the section s ⊗ t ∈ Γ(π ⊗ η) by the formula (s ⊗ t)(x) = s(x) ⊗ t(x). From the construction of the bundle π ⊗ η, we see that s ⊗ t is a smooth section of π ⊗ η. The assignment of s ⊗ t to the pair (s, t) is homomorphic with respect to both arguments and thus deﬁnes a C ∞ (M )-module homomorphism ι : Γ(π) ⊗ Γ(η) → Γ(π ⊗ η). Let us show that the value of this homomorphism at a point x ∈ M , ιx : Γ(π) ⊗ Γ(η)/μx Γ(π) ⊗ Γ(η) → πx ⊗ ηx , is an isomorphism of vector spaces. Note that ιx si ⊗ ti = si (x) ⊗ ti (x), where the square brackets denote the equivalence class of an element in the quotient space. (i) Surjectivity of ιx . An arbitrary element of the space πx ⊗ ηx has the form yi ⊗ zi with yi ∈ πx , zi ∈ ηx . By Lemma 12.8(a), there are sections si ∈ Γ(π), ti ∈ Γ(η) that take values yi and zi , respectively, at the point x. Then ιx [ si ⊗ ti ] = y i ⊗ zi . (ii) Injectivity of ιx . We must show that if si ∈ Γ(π), ti ∈ Γ(η), and si (x) ⊗ ti (x) = 0 in the space πx ⊗ ηx , then there exist sections pi ∈ Γ(π), qi ∈ Γ(η) and functions fi ∈ μx such that si ⊗ ti = fi pi ⊗ qi (equality in Γ(π) ⊗ Γ(η)). The following lemma clariﬁes the structure of zero elements in the tensor product of vector spaces. 12.40 Lemma. Let V and W be vector spaces over a certain field. Suppose that vi ∈ V , wi ∈ W are nonzero vectors and m vi ⊗ wi = 0 in V ⊗ W. i=1

Then there exist a natural number k, 1 k m, and elements aij , 1 i k, k < j m, of the ground field such that after an appropriate

196

Chapter 12

renumeration {1, . . . , m} → {1, . . . , m}, the same one for both {vi } and {wi }, the following relations hold : vj =

k i=1

wi = −

aij vi ,

m

j = k + 1, . . . , m;

aij wj ,

i = 1, . . . , k.

j=k+1

Indeed,if the elements v1 , . . . , vm are linearly independent, then the equality vi ⊗ wi = 0 implies that all vectors wi are zero. If not, choose from v1 , . . . , vm a maximal linearly independent subset. Let itk be v1 , . . . , vk . Expand vj for j = k+1, . . . , m in terms of this basis: vj = i=1 aij vi . Then m

vi ⊗ wi =

i=1

k i=1

=

k

j=k+1

vi ⊗ wi +

i=1

m

m k

vi ⊗ wi +

m

aij vi ⊗ wj

i=1

aij wj ,

j=k+1

whence wi = − j=k+1 aij wi . The lemma is proved. Applying the lemma in the current situation, we obtain sj =

k i=1

ti = −

aij si + sj ,

m

j = k + 1, . . . , m;

aij ti + ti ,

i = 1, . . . , k;

sj ∈ μx Γ(π); ti ∈ μx Γ(η).

j=k+1

Therefore, m

⎞ ⎛ k k m m si ⊗ ti = si ⊗ ⎝− aij tj + ti ⎠ + aij si + sj ⊗ tj

i=1

i=1

=

k i=1

j=k+1

si ⊗ ti +

m

j=k+1 i=1

sj ⊗ tj ∈ μx Γ(π) ⊗ Γ(η) ,

j=k+1

as desired. We see that the morphism ι is an isomorphism for every point x ∈ M . Using the corollary of Lemma 12.30, we want to conclude that ι is a module isomorphism. For this corollary to apply, both modules must be modules of sections of smooth vector bundles. In our case, only the module Γ(π)⊗Γ(η) is to be checked in this respect. But by Theorem 12.32 and Proposition 12.34, it is projective. Hence, by Theorem 12.32, this module is isomorphic

Vector Bundles and Projective Modules

197

to a certain module of sections, and thus the corollary can be used. This completes the proof of the theorem. 12.41. Change of rings. To translate the construction of the induced bundle into algebraic language, we must understand what relations between modules over diﬀerent rings arise when a homomorphism from one ring to another is given. Let ϕ : A → B be a ring homomorphism, and P a module over A. The homomorphism ϕ allows us to view B as an A-module with multiplication def of any element b ∈ B by a ∈ A deﬁned as ab = ϕ(a)b and hence deﬁne an A-module ϕ∗ (P ) = B ⊗A P . Setting b1 · (b2 ⊗ p) = b1 · b2 ⊗ p, we convert ϕ∗ (P ) into a B-module. The assignment P → ϕ∗ (P ) extends to a functor from Mod A to Mod B, called the change of rings functor denoted by Φ∗ . Proposition. The change of rings functor preserves projectivity. We shall show that the projectivity of an A-module P implies the projectivity of the B-module ϕ∗ (P ), using property (d) from Proposition 12.15. For an arbitrary B-module Q, there is an isomorphism of Abelian groups HomA (P ⊗A B, Q) ∼ = HomA (P, HomA (B, Q)).

(12.3)

More exactly, the elements γ ∈ HomA (P ⊗A B, Q)

and δ ∈ HomA (P, HomA (B, Q))

that correspond to each other under this isomorphism are related by the equations γ(p ⊗ b) = δ(p)(b),

p ∈ P, b ∈ B.

In particular, if γ ∈ HomB (P ⊗A B, Q), then for any p ∈ P , b1 , b2 ∈ B we have γ(p ⊗ b1 b2 ) = b1 γ(p ⊗ b2 ), and so δ(p)(b1 b2 ) = b1 · δ(p)(b2 ), i.e., δ(p) ∈ HomB (B, Q) ∼ = Q. The converse argument is also valid. Therefore, the isomorphism (12.3) induces an isomorphism HomB (P ⊗A B, Q) ∼ = HomA (P, Q). This isomorphism is natural with respect to Q, i.e., it extends to an isomorphism of functors on the category of B-modules with values in the category of Abelian groups, HomB (P ⊗A B, ·) ∼ = HomA (P, ·), which, by Proposition 12.15, implies that the B-module P ⊗A B is projective. 12.42. Algebraic formulation of pullback bundles. We now establish the algebraic meaning of the pullback procedure of vector bundles. Let

198

Chapter 12

ϕ : N → M be a smooth map of manifolds, Φ = ϕ∗ : C ∞ (M ) → C ∞ (N ), the corresponding homomorphism of function rings, and let Φ∗ be the change of rings functor. According to Proposition 12.41, the functor Φ∗ preserves projectivity; besides, it preserves the ﬁnite type property. Therefore, the functor Φ∗ can be restricted to the subcategory of ﬁnitely generated projective modules: Φ∗ : Modpf C ∞ (M ) → Modpf C ∞ (N ). 12.43 Theorem. For any smooth map of manifolds ϕ : N → M and any vector bundle π over M , there is an isomorphism of C ∞ (N )-modules Γ(ϕ∗ (π)) ∼ = Φ∗ (Γ(π)). This isomorphism can be chosen to be natural with respect to π, so that the functors Γ ◦ ϕ∗ and Φ∗ ◦ Γ are isomorphic, and the functor diagram VBM

ϕ∗

VBN

Γ

Γ

Modpf C ∞ (M )

Φ∗

Modpf C ∞ (N )

is commutative. Below we will refer to the lifting of sections ϕˆ deﬁned in Section 11.18. Let A = C ∞ (M ), B = C ∞ (N ). The map B × Γ(π) → Γ(ϕ∗ (π)), which sends the pair (f, s) to the section f · ϕ(s), ˆ is homomorphic over A with respect to either argument (here the B-module Γ(ϕ∗ (π)) is viewed as an A-module with multiplication introduced via the ring homomorphism Φ). Therefore, this map deﬁnes an A-homomorphism ν : B ⊗A Γ(π) → Γ(ϕ∗ (π)). Note that in fact, ν is a homomorphism not only over A, but also over B. Indeed, for f, g ∈ B and s ∈ Γ(π) we have ν(f g ⊗ s) = f g ϕ(s) ˆ = f ν(g ⊗ s). Let us prove that ν is an isomorphism. The module B ⊗A Γ(π) is ﬁnitely generated and projective; hence by Theorem 12.32 it is isomorphic to the module of sections of a bundle over N . Using Lemma 12.30, we can consider the value of the homomorphism ν at a point w ∈ N : νw : B ⊗A Γ(π)/μw ⊗A Γ(π) → (ϕ∗ (π))w ∼ = πϕ(w) . Using the identiﬁcation of (ϕ∗ (π))w with πϕ(w) , we obtain νw ([g ⊗ s]) = g(w)s(ϕ(w)). The map νw is epimorphic, because for any z ∈ πϕ(w) we can, by Lemma 12.8(a), ﬁnd a section s such that s(ϕ(w)) = z and hence νw ([1 ⊗ s]) = z.

Vector Bundles and Projective Modules

Now let us check that νw is monomorphic. Let of B ⊗A Γ(π) such that gi (w)si (ϕ(w)) = 0.

i gi

199

⊗ si be an element

i

Set gi (w) = βi ∈ R and put g i = gi − βi . The previous equation can be rewritten as i βi si ∈ μϕ(w) Γ(π), i.e., βi si = fj tj , where fj ∈ μϕ(w) , ti ∈ Γ(π). i

j

By the deﬁnition of the A-module structure in B, g ⊗ f t = ϕ∗ (f )g ⊗ t

for all g ∈ B, f ∈ A, t ∈ Γ(π).

Therefore, the following transformations are valid: gi ⊗ si = g i ⊗ si + βi ⊗ si = g i ⊗ si + 1 ⊗ βi si i

i

=

i

g i ⊗ si + 1 ⊗

i

=

i

g i ⊗ si +

i

i

fj tj

j

ϕ∗ (fj ) ⊗ tj ∈ μw ⊗ Γ(π)

j

(the last inclusion holds because ϕ∗ (fj ) ∈ ϕ∗ (μϕ(w) ) ⊂ μw ). We see that νw is an isomorphism at any point w ∈ N ; hence ν is an isomorphism of B-modules. Its naturality with respect to π is evident. The theorem is proved. Exercise. Show that the C ∞ (N )-module Dϕ (M ) consisting of vector ﬁelds along the map ϕ : N → M (see Section 9.47) is naturally isomorphic to the module Γ(ϕ∗ (πT )). What is the algebraic meaning of the lifting of sections, i.e., what is the map Γ(π) → B ⊗A Γ(π) it is described by? It is easy to see from the deﬁnitions that this is the map that takes each element s ∈ Γ(π) to 1 ⊗ s. 12.44. Pseudo-bundles and geometrical modules. Let M be a smooth manifold and F = C ∞ (M ). Theorem 12.32 gives a geometrical meaning to the notion of ﬁnitely generated projective F-module. We want to ﬁnd out, in the spirit of Sections 12.11–12.12, to what extent arbitrary modules over F possess a geometrical interpretation. In Section 12.11, to an arbitrary module P over an arbitrary commutative K-algebra A we assigned a pseudo-bundle πP . For the function algebra F(|P |) on the total space of the bundle, we took the symmetric algebra S(P ∗ ) of the module P . In the case of the algebra of smooth functions A = C ∞ (M ), it is natural to take the smooth envelope F(|P |) = S(P ∗ ) instead of just S(P ∗ ), as we did in Section 10.12 for the cotangent bundle and the algebra of symbols.

200

Chapter 12

12.45 Exercises.

1. Show that the maps

πP : |P | → |A|,

sp : |A| → |P |

(p ∈ P ),

deﬁned in Section 12.11, are continuous in the Zariski topology deﬁned by the function algebra S(P ∗ ). 2. Let P = D(C ∞ (K)) be the module of vector ﬁelds on the cross (see Exercises 7.14, 9.35, 9.45, 10.10). Describe |P |. From now on, by continuous sections of a pseudo-bundles πP , we shall understand sections, continuous in the Zariski topology, corresponding to the smooth envelope of the symmetric algebra F(|P |) = S(P ∗ ). The set of all such sections forms a module over the ring A = C ∞ (M ), which we denote by Γc (π). The assignment p → sp deﬁnes a C ∞ (M )-module homomorphism S : P → Γc (π). By Theorem 12.32, for a projective ﬁnitely generated module P , this homomorphism is a monomorphism, and its image coincides with the submodule of smooth sections in Γc (π). We pass to examples of geometrical and nongeometrical modules over algebras of smooth functions. 12.46. Examples. A. Geometrical nonprojective modules. I. A smooth map of manifolds ϕ : M → N gives rise to a homomorphism of the corresponding smooth function rings ϕ∗ : B → A and thereby turns A into a B-module. An easy check shows that this module is always geometrical. However, it is projective only in exceptional cases (for instance, if ϕ is a diﬀeomorphism). The simplest example of a geometrical nonprojective module of this kind is obtained if M is the manifold consisting of one point. Exercise. Describe all smooth maps ϕ for which the B-module A is projective. II. The ideal μa of any point a ∈ M , viewed as a C ∞ (M )-module, is obviously geometrical. It turns out that this module is projective if and only if dim M = 1. Indeed, the value of the module μa at a point b ∈ M is the quotient space μa /μa μb . Its dimension is dim M, if b = a, dim μa /μa μb = 1, if b = a. The ﬁrst equality follows from the fact that μa /μ2a is the cotangent space of the manifold M at the point a (see Section 9.27). The second one is a consequence of Lemma 2.11.

Vector Bundles and Projective Modules

201

The ﬁber dimension of the vector bundle corresponding to μa is thus constant in the case dim M = 1 and nonconstant in the case dim M > 1. There exist two nondiﬀeomorphic connected one-dimensional manifolds: the line R1 and the circle S 1 . What vector bundles correspond to μa in each case? The answer, at ﬁrst sight unexpected, is that for the line it is obius the trivial one-dimensional bundle IR1 , while for the circle it is the M¨ band bundle described in Example 12.6, I. Here is the proof. Let F = C ∞ (R1 ) and let a ∈ R1 be an arbitrary point. Hadamard’s Lemma 2.10 implies that the map F → μa that sends each function f to the product (x − a)f establishes the module isomorphism F → μa . Therefore, μa ∼ =F ∼ = Γ(IR1 ). This argument does not apply to F = C ∞ (S 1 ) and a ∈ S 1 , because in this case μa is not a principal ideal: there is no smooth function on the circle that vanishes only at one point and has a nonzero derivative at this point. The isomorphism between μa and the module of sections of the obius band, can nontrivial vector bundle π, whose total space Eπ is the M¨ be deﬁned as follows. We know that the tensor square of π is isomorphic to IR1 (see Example 12.38); hence there is a well-deﬁned multiplication Γ(π) × Γ(π) → F. Fix a section f0 ∈ Γ(π) with a single simple zero at the point a (i.e., f0 (a) = 0, f0 (a) = 0, f0 (b) = 0 for b = a). Then the map that sends every section f ∈ Γ(π) to the product f0 f ∈ F establishes the required isomorphism Γ(π) → μa . B. Nongeometrical modules. (see III. The C ∞ (M )-module of lth order jets Jzl M = C ∞ (M )/μl+1 z Section 9.64) is not geometrical if l 1. This is due to the facts that: (i) μz · Jzl M = Jzl M , if z = z; (ii) μz · Jzl M = μz /μl+1 z . Leaving the proof to the reader, we deduce that

μz · Jzl M = μz /μl+1 z . z ∈M

The last module is composed of all “invisible” elements of Jzl M (see 12.11) and is nontrivial for l 1. However, C ∞ (M )-modules Tz M = D(M )/μz D(M ) (see Lemma 10.7) and Tz∗ M = Λ1 (M )/μz Λ1 (M ) are geometrical. Indeed, if P is one of them, then μz · P = P for z = z and μz · P = 0 (prove that) and therefore z ∈M μz · P = 0. IV. Let A = C ∞ (R) and let I ⊂ A be the ideal that consists of all functions with compact support. The reader can prove, by way of exercise, that the quotient module P = A/I has the property P = x∈R μx P , so that the corresponding map S (see Section 12.45) is identically zero. The module P in this example consists entirely of invisible (unobservable) elements.

202

Chapter 12

12.47. Vector bundles as pseudo-bundles. We are now in a position to keep the promise given previously and explain the relationship between the algebraic treatment of a vector bundle as a module and the treatment of a pseudo-bundle as an embedding of smooth algebras. Proposition. The algebra of functions on the total space of a vector bundle π is isomorphic to the smooth envelope of the complete symmetric algebra of the module of sections Γ(π). Since the modules of sections of a given bundle π and its conjugate π ∗ are isomorphic (see the remark in Section 12.38), it suﬃces to construct an isomorphism of the algebra C ∞ (Eπ ) with the smooth envelope of the symmetric algebra of Γ(π ∗ ). Such an isomorphism can be built in a natural way. Indeed, every section s ∈ Γ(π ∗ ) = S 1 (Γ(π ∗ )) deﬁnes a function on Eπ that is linear on the ﬁbers. Elements of S 2 (Γ(π ∗ )) correspond to functions on Eπ that are quadratic on the ﬁbers; elements of S 3 (Γ(π ∗ )) correspond to functions on Eπ that are cubic on the ﬁbers; etc. Such functions are obviously smooth. The whole symmetric algebra S(Γ(π ∗ )) can be considered as the algebra of all functions on Eπ polynomial on every ﬁber of π. The construction of the smooth envelope (Section 3.36) extends the set of polynomial functions to the set of all smooth functions. 12.48. In this book, we deal with smooth manifolds, smooth functions, smooth vector ﬁelds, smooth sections of vector bundles, etc. What can we say about similar objects that are not inﬁnitely smooth, but for instance of class C m ? These notions of the standard calculus can be treated in the algebraic framework, using diﬀerent functional algebras and using the change of rings procedure. Exercises. 1. Let P be the module of smooth (of class C ∞ ) sections of a vector bundle πP . For an arbitrary m 0, give an algebraic deﬁnition of the module of sections of this bundle belonging to the class C m (e.g., to C 0 , i.e., continuous sections). 2. Give an algebraic deﬁnition of vector ﬁelds (diﬀerential operators) of class C m on a smooth manifold M .

13 Localization

13.1. Localization of diﬀerential operators. Locality is the most important property of diﬀerential operators. It is precisely this property that explains the fundamental role that the diﬀerential calculus plays in the natural sciences, especially in physics. Indeed, the experimental information on the basis of which various theories are constructed is, by its very nature, local is space, in time and in many other respects. Therefore, the mathematics that processes this information must be localizable. Ignoring this principle leads to substantial diﬃculties in the construction of new theories, as we can observe, for instance, in contemporary quantum ﬁeld theory. The importance of using a localizable mathematical apparatus in classical physics does not directly jump to the eye mainly because the standard, historically arisen coordinate description of diﬀerential operators Δ=

aσ (x)

∂ |σ| ∂xσ

is automatically localized, i.e., can be applied to any function independently of the domain of the variable in which the function is deﬁned. Therefore, in this “classical” context, the question of localization is not a problem. In the general framework of the diﬀerential calculus on commutative algebras, the localization property of diﬀerential operators means, in particular, that it is possible to localize them to any domain “observable” by the given commutative algebra, for example on open (in the Zariski topology) subsets of its spectrum. In this chapter, we ﬁrst present the necessary general facts having to do with the localization of algebras and modules © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 13

203

204

Chapter 13

(see Sections 11.6–11.8) and conclude the chapter with the derivation of a general formula on the localization of diﬀerential operators. 13.2. Universality property of localization. We shall say that, in an A-module P , division by an element s ∈ A is possible if the k-linear operator p → sp, p ∈ P , is invertible. In that case, s−1 p denotes the element of the module P such that s(s−1 p) = p. If division by the elements s1 and s2 is possible in P , then obviously it is possible by s1 s2 . Therefore, all the divisors s ∈ A of the module P constitute a multiplicative set Sdiv (P ) ⊂ A. Lemma. Let S ⊂ Sdiv (P ) be a multiplicative set. Then P carries the structure of an S −1 A-module. Let us deﬁne an S −1 A-module structure in P by the formula (a/s)p = s−1 (ap), a ∈ A, s ∈ S, p, def

about which, however, we must check that it is well deﬁned. Suppose that a/s = a1 /s1 , i.e., s (as1 − a1 s) = 0 for some s ∈ S. We must verify that s−1 (ap) = s−1 1 (a1 p). But this follows from the sequence of equalities −1 s−1 (ap) = s−1 (ap)] = (ss1 )[s−1 1 (a1 p) ⇐⇒ (ss1 )[s 1 (a1 p)] ⇐⇒ ⇐⇒ (s1 a − saa )p = 0 ⇐⇒ s (as1 − a1 s)p = 0 .

The fact that P with this structure is an S −1 A-module is obvious.

Proposition. Suppose P and Q are A-modules, and division by elements of the multiplicative set S in Q are possible. Then the correspondence HomA (P, Q) H → S −1 H ∈ HomS −1 A (S −1 P, Q), def

where (S −1 H)(p/s) = s−1 H(p) is an S −1 A-module isomorphism. First of all, we must prove that the map S −1 H is well deﬁned. This, however, can be proved exactly like the similar statement in the previous lemma. Further, the veriﬁcation of the fact that S −1 H is a homomorphism of S −1 A is obvious and we omit it. Finally, noting that H = S −1 H ◦ ιS,P , where ιS,P : P → S −1 P is the canonical homomorphism (p → (p, 1)), we see that the correspondence H → S −1 H is bijective. The homomorphism ιS takes elements s ∈ S to invertible elements of the algebra S −1 A and this property uniquely characterizes the pair (ιS , S −1 A). Namely, we have: Corollary. Let H : A → B be a homomorphism of commutative algebras with unit (over a ﬁeld k) such that all elements H(s), s ∈ S, are invertible. Then H can be uniquely presented in the form H = HS ◦ ιS , where S : S −1 A → B is a k-algebra homomorphism. From the homomorphism H, the algebra B acquires an A-module structure and, by the previous lemma, the structure of an S −1 A-module as well. Therefore, the homomorphism S −1 H : S −1 A → B of S −1 A-modules is well

Localization

205

deﬁned. It is easy to check that it is also an algebra homomorphism, so it only remains to put HS = S −1 H. Exercise. Let S ⊂ A be a multiplicative subset. A homomorphism of commutative k-algebras ι : A → C will be called S-invertible, if all the elements ι(s), s ∈ S, are invertible. An S-invertible homomorphism ι is called universal, if for any S-invertible homomorphism H : A → B there exists a unique k-algebra homomorphism H1 : C → B for which we have H = h1 ◦ ι. Show that there exists an isomorphism ϕ : S −1 A → C such that ι = ϕ ◦ ιS . In other words, any homomorphism (of the type speciﬁed above) is equivalent to the homomorphism ιS . 13.3. Geometrical portrait of localization. In our previous exposition, we were able to appreciate the usefulness of the fact that tangent vectors and the other diﬀerential operators that we considered can be localized. There we understood localization as the passage from the algebra C ∞ (M ) to the algebra C ∞ (U ), as they are naively described in Sections 3.23 and 3.24. The algebraic formalization of that “experimental material”, which is well known in algebraic geometry, was presented in Section 11.6. It will now be appropriate to illustrate this procedure geometrically. Let S ⊂ A be a multiplicative subset. With it, we associate the subset def Us ⊂ |A|. US = {h ∈ |A| | h(s) = 0, ∀s ∈ S} = s∈S

Proposition. The map |ι| : |S

−1

A| → |A| is an embedding and (|ι|) = US .

If S is ﬁnitely generated, then |S −1 A| is an open (in the Zariski topology) subset of |A|. ¯ ∈ |S −1 A|, h = |ι|(h) ¯ and s ∈ S, then If h ¯ ¯ ¯ ¯ 1k = h(1S −1 A ) = h(s/1 · 1/s) = h(s/1) h(1/s) . Therefore, ¯ ¯ ¯ h(s/1) = h(s) = 0, ∀s ∈ S, h(1/s) = h(s)−1 , h(a/s) = h(a)h(s)−1 . ¯ i.e., |ι| The last one of these equalities shows that h uniquely determines h, is an embedding. Besides, we see that (|ι|) ⊂ {h ∈ Speck A | h(s) = 0, ∀s ∈ S} . ¯ ∈ |S −1 A| by Conversely, if h(s) = 0, ∀s ∈ S, we deﬁne a certain element h −1 ¯ setting h(a/s) = h(a)h(s) . The reader will easily verify that this element ¯ is well deﬁned, and the existence of such an h proves the converse inclusion. Finally, noting that Uab = Ua Ub , we see that (|ι|) =

s∈S

Us =

m i=1

Usi

206

Chapter 13

if the elements s1 , . . . , sm generate S. The equality s(a1 s2 −a2 s1 ) = 0 that describes the equivalence relation in the deﬁnition of formal fractions a/s also has a clear geometrical interpretation. As is easily seen, it is equivalent to the equality fa1 /fs1 = fa2 /fs2 for functions deﬁned on US (see Section 8.4). It is intuitively clear that US does not deﬁne S, nor S −1 A. One of the reasons for this is illustrated by the following example. Example. Let M be the closed interval [−1, 1], while U = {−1 < x < 1} is the corresponding open one and A = C ∞ (M ). Then, as we have seen, U = US if S = {f ∈ C ∞ (M ) | f (x) = 0, ∀x ∈ U }. On the other hand, U = US1 , where S1 is generated (multiplicatively) by the function ϕ = (x2 − 1)|M . Here the algebras S −1 A and S1−1 A are not isomorphic. Indeed, suppose that g/ϕk ∈ S1−1 A, g ∈ A. Then ϕk /1 · g/ϕk = g|U is a bounded function on U , since g ∈ C ∞ (M ). However, there does not exist any function ψ ∈ C ∞ (M ) such that ψ l /1 · μ becomes a bounded function on U for an any value of the degree l independent of the element μ ∈ S −1 A. For example, this will be the case for μ = 1/s, 0 < s ∈ S, if 1/s tends to inﬁnity as x → 1 faster than any degree of |ψ k |. Exercise. 1. Give an example of a multiplicative set S ⊂ C ∞ (M ) for which US = ∅. 2. Is there such a multiplicative set in the polynomial algebra k[x1 , . . . , xn ]? 13.4. US and SV . Subsets of the spectrum of an algebra A are a natural source of useful multiplicative systems. Namely, to an arbitrary nonempty subset V ⊂ |A|, we can associate the multiplicative system def

SV = {s ∈ A | h(s) = 0, ∀h ∈ V } . In other words, SV is generated by elements a ∈ A such that the function fa on |A| (see Section 8.4) does not vanish at points of the subset V . If V is empty, then SV is deﬁned as the set of those elements of the algebra A which are not divisors of zero. In all these cases, SV is nonempty because it always contains 1A . Conversely, to any multiplicative system S in A, we can associate the subset def

US = {h ∈ |A| | h(s) = 0, ∀s ∈ S} .

Localization

207

We have already dealt with these two constructions previously in the case when the subset V is open. Further, we shall see that they are useful not only in this case. Iterating these constructions, we obtain a chain of the form V =⇒ SV =⇒ USV =⇒ . . . ,

S =⇒ US =⇒ SUS =⇒ . . .

(13.1)

Obviously, V ⊂ USV and S ⊂ SUS . The examples presented below show that, generally speaking, USV and SUS are wider than V and S, respectively. Here the chains (13.1) already stabilize at the third step. Proposition. Let V = USV and S = SUS . Then SV = SV and US = US . Note that SV1 ⊂ SV2 if V1 ⊃ V2 . But since V ⊃ V , it follows that SV ⊂ SV . On the other hand, if Sˆ = SV , then USˆ = USV = V and Sˆ = SUSˆ = SV . Therefore, SV = Sˆ ⊂ Sˆ = SV and so SV = SV . The equality US = US is proved exactly in the same way. 13.5. Examples. In the examples in this series, we have assumed that A = k[x1 , . . . , xn ] and so |A| = kn , while in examples I-V, it is also assumed that V = kn \ {0}. I. According to its deﬁnition, SV consists, in this case, of polynomials whose restriction to V is everywhere nonzero. Let f be such a polynomial and L ⊂ V be a straight line. Then f |L is a polynomial on that line. It the ﬁeld k is algebraically closed, e.g., k = C, then this polynomial has roots, provided its degree is positive. Therefore, f |L is a nonzero constant. But since this is true for all lines not passing through (0, . . . , 0), it follows that f is also a nonzero constant. Thus, in the case under consideration, SV consists of nonzero constants and so USV equals kn , i.e., is wider, than V . II. SV becomes much richer if the ﬁeld k is not algebraically closed. For example, if k = R, then SV contains, besides nonzero constants, positive deﬁnite quadratic forms in the variables x1 , . . . , xn . Since each such form vanishes at the point (0, . . . , 0), it follows that USV does not contain that point and so coincides with V . Thus, in this case, USV = V . III. Let f (t) ∈ k[t] be a polynomial of degree l without roots in the ﬁeld k, and let ϕ(x) = ϕ(x1 . . . , xn−1 ) ∈ k[x1 . . . , xn−1 ] be a homogeneous polynomial of degree m that vanishes only at the point (0, . . . , 0) ∈ kn−1 . Then the homogeneous polynomial g(x1 , . . . , xn ) = xlm n f (ϕ(x1 /xn , . . . , xn−1 /xn )) vanishes only at the point {0} ∈ kn . Indeed, this is obvious if xn = 0, while g(x1 , . . . , xn−1 , 0) = ϕ(x1 , . . . , xn−1 )l . Besides, the above construction yields, by induction, polynomials f with the required properties, beginning with the polynomial x1 ∈ k[x1 ]. Therefore, for the same reason as in the previous example, we have USV = V . Thus, a nontrivial widening of the set V at the second step occurs only when the ﬁeld k is algebraically closed.

208

Chapter 13

IV. Since A is an algebra without zero divisors, S = A \ 0A is a multiplicative system, while S −1 A is its fraction ﬁeld. Therefore, |S −1 A| = ∅ and so US = ∅ and SUS = S. V. Suppose the ﬁeld k is not algebraically closed and the polynomial g(x1 , . . . , xn ) vanishes only at the point {0} ∈ kn (see Example III above). If S = {g k }k≥0 , then US = kn \ {0} and so SUS consists of all the polynomials that do not vanish at the points of kn \ {0}. Thus, SUS is wider than S. VI. If V = kn , then SV consists of polynomials that do not vanish anywhere. If the ﬁeld k is algebraically closed , then SV consists of nonzero constants only. For ﬁelds that are not algebraically closed, this is no longer so, since there exist polynomials of positive degree without zeros in kn . For instance, such is the polynomial f = g(ψ(x)), where f is the polynomial from Example III, while the polynomial ψ ∈ k[x1 , . . . , xn ] is arbitrary. 13.6. Germs. As we have already seen (Proposition 11.7), the localization of the algebra A with respect to a multiplicative set SU is the analog of the passage from the algebra C ∞ (M ) to the algebra C ∞ (U ) under the condition that the subset U ⊂ |A| is open. Now if, on the other hand, the set U ⊂ M is closed, then the result of such a localization is the algebra of germs of smooth functions along the subset U . This will be shown below. Therefore, it is appropriate to regard the algebra of germs of an arbitrary algebra A as the analog of the algebra SU−1 A for some closed subset U ⊂ |A|. It should be noted that in the case of smooth manifolds, this construction is of the most interest when U is a smooth submanifold. So, let N ⊂ M be a closed subset. (Here we have replaced U by N because the letter U is systematically used to denote open sets.) Let us call two functions f and g from C ∞ (M ) equivalent, if they coincide in some neighborhood U of the set N and write f ∼N g. By a germ of a function f ∈ C ∞ (M ) along N we mean its equivalence class w.r.t. ∼N . Arithmetical operations with functions can obviously be carried over to germs. Thus, we have deﬁned the algebra of germs of functions along a closed subset N ⊂ M ; ∞ (N ). we denote this algebra by Cgerm −1 ∞ ∞ (N ) and SN C (M ) are canoniProposition. (a) The algebras Cgerm cally isomorphic.

(b) Let a subset V ⊂ |A| be closed and let ι : A → SV−1 A be the canonical isomorphism. Then |ι| = V . (a) Let U be an open set containing N . The equality f |U = g|U is −1 ∞ equivalent to ι(f ) = ι(g), where ι : C ∞ (M ) → SN C (M ) is the canonical homomorphism. Indeed, the last relation means that ϕ · (f − g) = 0 for some function ϕ ∈ SN . Since, by deﬁnition, ϕ does not vanish anywhere on N , we have N ⊂ Uϕ = {z ∈ M | ϕ(z) = 0}

and f |Uϕ = g|Uϕ .

Localization

209

The last equality is obvious since ϕ(z) = 0 and ϕ(z)(f (z) − g(z)) = 0 imply f (z) = g(z). Finally, we must establish that the algebra S −1 C ∞ (M ) consists only of the germs of smooth functions along N . This is equivalent to ι being an epimorphism. Let us prove this. Let f /ϕ ∈ S −1 C ∞ (M ). Then to the pair (f, ϕ) representing this formal fraction, we can associate the function ψ ∈ C ∞ (Uϕ ), ψ(z) = f (z)/ϕ(z). If the function f ∈ C ∞ (M ) coincides with ψ on some open set U ⊃ N , then f /ϕ = f /1A = ι(f ), i.e., we have g(f − ϕf ) ≡ 0 for some function g ∈ SN . But since the function f − ϕf is identically zero on U , we can choose, in the role of g, any function which is strictly positive on U and is zero outside of U . The existence of functions f and g with the required properties was established previously. (b) The fact that V ⊂ |ι| immediately follows from the proposition in Section 13.3, so that it only remains to show that (|A| \ V ) ∩ |ι| = ∅. But the set (|A| \ V ) is open and so it is the union of some open sets from the base. Let us show that if Ua is one of them, then Ua ∩ ( |ι|) = ∅. def

Note that V ⊂ Va = |A| \ Ua . If h0 ∈ Ua , then λ = h0 (a) = 0. Therefore, a0 = a − λ1A , implies h(a0 ) = −λ = 0 provided h ∈ Va , so that we have a0 ∈ SVa ⊂ SV . On the other hand, h0 (a0 ) = 0 and the proposition from / |A|. Section 13.3 implies h0 ∈ 13.7. Example IV from Section 11.6 shows that quite diﬀerent multiplicative systems can lead to the same localization of a given algebra A. So, it is of interest to ﬁnd out when this can occur. In its turn, this question is a particular case of a more general and less natural question: How are two localizations of the same algebra related? In what follows, we shall begin to study this question, as well as related ones. We will need to consider formal fractions corresponding to diﬀerent multiplicative systems. The fact that such a fraction belongs to a given localization of the algebra A under consideration will depend on whether or not the denominator of the fraction belongs to that system. The multiplicative system itself will be denoted by the capital letter S ¯ while its elements will be denoted with additional indications, e.g., Sα or S, by a lower case s with the same indicator and, possibly, an additional index, e.g., sα,i or s¯i . Finally, ιS : A → S −1 A will denote the canonical homomorphism . 13.8. Localizations of a localization. Let V ⊂ U ⊂ M be open subsets of a manifold M . Then the algebra C ∞ (V ) is a localization of both the algebra C ∞ (M ) and the algebra C ∞ (U ) (Proposition 11.7). Now since C ∞ (U ) is a localization of the algebra C ∞ (M ), it follows that C ∞ (V ) is a “localization of a localization”. This coincidence of iterated localizations is not only true for multiplicative sets of the form SU on smooth manifolds, and in what follows we will show that it takes place for arbitrary multiplicative sets.

210

Chapter 13

Let S and S+ be multiplicative subsets in A and S −1 A, respectively. Further, let us consider the set S¯ of elements a ∈ A such that the pair (a, s) for some s is a formal fraction belonging to S+ . Since the pair (s, s) ¯ The set S¯ is multiplicative. represents the unit 1S −1 A ∈ S+ , we have S ⊂ S. Indeed, if ai /si ∈ S+ , i = 1, . . . , m, then (a1 . . . am )/(s1 . . . sm ) = 0, which is equivalent to sa1 . . . am = 0, for all s ∈ S. This shows that S¯ is closed under multiplication and does not contain the zero element. −1 (S −1 A) is naturally Proposition. “The localization of localization” S+ −1 ¯ isomorphic to the algebra S A.

In what follows, we use the notation a+ /s+ for formal fractions that ¯ Let the pair (a/s, a+ /s+ ) belong to S+ . Thereby we indicate that a+ ∈ S. −1 (S −1 A). Then the map η represent the formal fraction f from S+ S −1 (S −1 A) f → as+ /a+ s ∈ S¯ +

is the required isomorphism. To establish this, we must ﬁrst show that the map η is well deﬁned. Let (a /s , a+ /s+ ) be another pair representing the fraction f . This means that aa+ a a+ a ˆ+ · − ˆ+ (aa+ s s+ − a a+ ss+ ) = 0, = 0 ⇐⇒ s0 a sˆ+ ss+ s s+ for some s0 ∈ S. We must now show that the formal fractions as+ /a+ s and a s+ /a+ s are equal, i.e., s¯(aa+ s s+ − a a+ ss+ ) = 0 ¯ But for that it suﬃces to put s¯ = s0 a ¯ for some s¯ ∈ S. ˆ+ ∈ S. −1 ¯ An arbitrary element of the algebra S A can be expressed in the form a/a+ s = as+ /a+ ss+ . It is the image under the map η of the formal fraction represented by the pair (a/ss+ , a+ /s+ ). Hence, η is surjective. Finally, if ¯ In other η(f ) = 0, then (a+ s+ ) · (as+ ) = 0 for some element a+ s+ ∈ S. words, sˆ(aa+ = 0) for sˆ = s+ s+ ∈ S. This implies (a+ /s+ ) · (a/s) = aa+ /ss+ = 0 and so f = 0. Hence, the map η is injective. For further applications, it is convenient to denote the multiplicative set S¯ by S −1 (S+ ). Corollary.

(a) Let a multiplicative set S1 in A be such that the set def

ιS (S1 ) = {ιS (s1 ) | s1 ∈ S1 } is multiplicative in S −1 A. Then the set S1 is contained in S −1 (S+ ). (b) The set ιS (S −1 (S+ )) is multiplicative in S −1 A. (c) The localization of the algebra S −1 A via the multiplicative set ιS (S −1 (S+ )) coincides with the localization of the algebra A via the −1 (S −1 A). set S −1 (S+ ) and hence coincides with S+

Localization

211

The ﬁrst two statements and the ﬁrst part of the third one are immediate consequences of the construction of the set S −1 (S+ ), while the remaining part is the assertion of the proposition from Section 13.8. 13.9. Compatibility of multiplicative sets. Assertion (c) of the previous corollary shows that any localization of the algebra S −1 A coincides with its localization via a multiplicative set of the form ιS (S ), where S is some multiplicative set in the algebra A. In this connection, it makes sense to characterize those multiplicative sets in A which are taken to multiplicative sets of the algebra S −1 A by the homomorphism ιS . Let us call two multiplcative sets S1 and S2 of the algebra A compatible, if s1 s2 = 0 for all (s1 , s2 ) ∈ S1 × S2 . If the algebra A is geometrical, then the inequality s1 s2 = 0 is equivalent to the existence of an element h ∈ |A| such that 0 = h(s1 s2 ) = h(s1 )h(s2 ). In its turn, this is equivalent to US1 ∩ US2 = ∅. Since, in the algebra of subsets of a given set, the intersection operation plays the role of multiplication, one can use the geometrical interpretation of the compatibility relation given above in order to naturally introduce the product of compatible multiplicative sets by setting def

S1 S2 = {s1 s2 | (s1 , s2 ) ∈ S1 × S2 } . Note that S1 S2 = S2 S1 and S1 , S2 ⊂ S1 S2 . Besides, if the multiplcative sets S1 S2 and S3 are compatible, then s1 s2 s3 = 0 for all (s1 , s2 , s3 ) ∈ S1 × S2 × S3 . This obviously implies the compatibility of the sets Si and Sj , as well as that of the sets Si and Sj Sk , where i, j, k are diﬀerent indices taking values in {1, 2, 3}. In particular, we have associativity, (S1 S2 )S3 = S1 (S2 S3 ), for the product of compatible multiplicative sets. Proposition. (a) The set ιS1 (S2 ) is multiplicative in the algebra S1−1 A if and only if the multiplcative sets S1 and S2 are compatible in the algebra A. (b) If two multiplcative sets S1 and S2 are compatible, then −1 −1 −1 −1 ιS1 (S2 ) (S1 A) = (S1 S2 )−1 A = ιS2 (S1 ) (S2 A) . (a) The multiplicativity of the set ιS1 (S2 ) means that 0 = ιS1 (s2 ) = s2 /1 for all s2 ∈ S2 . In its turn, the inequality s2 /1 = 0 means that s1 s2 = 0 for all s1 ∈ S1 . (b) In view of the commutativity of the “product” S1 S2 , it suﬃces to prove only the ﬁrst of these two equations. To do that, we only need to note that the set S −1 (S+ ) for S = S1 and S+ = ιS1 (S2 ) coincides with S1 S2 and use the proposition from Section 13.8. Exercise. 1. Show that the multiplicative sets S1 and S2 are compatible whenever US1 ∩ US2 = ∅.

212

Chapter 13

2. Give an example of compatible multiplicative sets S1 and S2 for which US1 ∩ US2 = ∅. 3. Show that the compatibility relation is not transitive in general. 13.10. Noncompatible multiplicative sets. The multiplicative sets S1 and S2 are called noncompatible, if there exist elements si ∈ Si , i = 1, 2, such that s1 s2 = 0. The condition US1 ∩US2 = ∅ is geometrically necessary for the noncompatibility of S1 and S2 (see Exercises 1 and 2 from the previous section). For noncompatible multiplicative sets, there is an analog of the operation of union for nonintersecting sets. Namely, to each pair s1 ∈ S1 , s2 ∈ S2 such that s1 s2 = 0, let us associate the element s1 + s2 . The set consisting of all products of elements of that form and of the unit element 1A will be denoted by S1 S2 . Let us show that it is multiplicative, i.e., that an arbitrary product of the form w = (s1,1 + s2,1 ) · . . . · (s1,m + s2,m ), where si,j ∈ Si and s1,j s2,j = 0, is nonzero. Let s = s1,1 · . . . · s1,m . Then sw = (s21,1 + s1,1 s2,1 ) · . . . · (s21,m + s1,m s2,m ) = s21,1 · . . . · s21,m ∈ S1 . Therefore, sw = 0 and so w = 0. Examples. I. Let U1 , U2 be open subsets of a manifold M such that U1 ∩ U2 = ∅, and let Si = SUi , i = 1, 2. Then S1 S2 consists of all functions f1 + f2 , fi ∈ C ∞ (M ), such that f1 does not vanish anywhere on U1 , is identically zero on U2 , and similarly for f2 . So, US1 S2 = U1 ∪ U2 . II. Let K be the coordinate cross on the plane (see Exercise 1 in xk }k≥0 and Section 2.11) and let x ¯ = x|K , y¯ = y|K . Let also S1 = {¯ y k }k≥0 .Then US1 and US2 are the coordinate axes with the point S2 = {¯ (0, 0) deleted, while S1 S2 consists of the unit 1K and all possible functions of the form x ¯k + y¯l , k ≥ 1, l ≥ 1. Here US1 S2 = K \ {(0, 0)}. 13.11. Natural homomorphisms of localizations. The algebra homomorphism H : S1−1 A → S2−1 A is called natural if ιS2 = H ◦ ιS1 . Proposition. (a) A necessary and suﬃcient condition for the existence of the natural homomorphism H : S1−1 A → S2−1 A is the invertibility of all elements ιS2 (s1 ), s1 ∈ S1 . (b) The natural homomorphism is unique (if it exists). (c) The element ιS2 (s1 ) is invertible if and only if it divides some element s2 ∈ S2 , i.e., if there exists an element a ∈ A such that s1 a = s2 . (d) If the natural homomorphism H exists, then the multiplcative sets S1 and S2 are compatible. def

(a)–(c) Further it will be convenient to use the notation a/1i = ιSi (a) which stresses that a/1i is a formal fraction in Si−1 A. The elements ιS1 (s1 ) are obviously invertible, and therefore, their images under the

Localization

213

homomorphism H, i.e., the elements ιS2 (s1 ), are also invertible. Therefore, H(a/s1 ) =H((a/11 ) · (1/s1 )) = H(ιS1 (a))H(ιS1 (s1 )−1 ) =ιS2 (a)[ιS2 (s1 )]−1 , which proves the uniqueness of the homomorphism H. Further, the invertibility of the element ιS2 (s1 ) = s1 /12 means that for some a ∈ A and s2 , s2 ∈ S2 , we have (s1 /12 ) · (a /s2 ) = 1/12 ⇐⇒ s2 (s1 a − s2 ) = 0 =⇒ s1 a = s2 ,

where a = a s2 , s2 = s2 s2 . Conversely, the relation s1 a = s2 implies (s1 /12 ) · (a/s2 ) = 1/12 , i.e., [ιS2 (s1 )]−1 = a/s2 . This proves assertion (c). Finally, assuming that the elements ιS2 (s1 ) are invertible, we deﬁne the homomorphism H by setting def

H(b/s1 ) = ιS2 (b) · (a/s2 ),

where

s1 a = s2 .

We leave to the reader the routine veriﬁcation of the fact that h is well deﬁned. (d) The fact that s1 s2 = 0 for all s1 ∈ S1 , s2 ∈ S2 follows from the relation s1 s2 a = ss2 s2 = 0, where s1 a = s2 . Example. If S1 ⊂ S2 , then, in view of the fact that any element ιS2 (s1 ) is (obviously) invertible, the natural homomorphism H : S1−1 A → S2−1 A −1 (S −1 A), taking exists. In particular, the homomorphism ιS+ : S −1 A → S+ the “localization” to the “localization of localization”, is identiﬁed with a similar homomorphism for S1 = S and S2 = S¯ (see Section 13.8). In particular, if U1 ⊃ U2 are open sets of a manifold M , then the restriction homomorphism C ∞ (U1 ) → C ∞ (U1 ), f → f |U2 , is natural. If U1 ⊃ U2 , then there does not exist any natural homomorphism from C ∞ (U1 ) to C ∞ (U2 ). The fact that the natural homomorphism H : S1−1 A → S2−1 A is a more delicate version of the restriction of functions follows from the relation VS1 ⊃ VS2 . Indeed, in this case any element s1 ∈ S1 divides some element s2 ∈ S2 , so that h(s1 ) = 0, if h(s2 ) = 0. Moreover, as can be easily seen, fa/s1 |VS2 = fH(a/s1 ) . 13.12 Exercise. Prove that the composition of natural homomorphisms is also a natural homomorphism. 13.13. Equivalence of multiplicative sets. Two multiplicative sets S1 and S2 are said to be equivalent if the localizations S1−1 A and S2−1 A of the algebra A coincide, i.e., are naturally isomorphic. Corollary. (a) Two multiplicative sets S1 and S2 are equivalent if and only if any element s1 ∈ S1 divides some element s2 ∈ S2 and conversely.

214

Chapter 13

(b) The homomorphism ιS : A → S −1 A is a isomorphism if and only S consists of invertible elements. (c) Two localizations S1−1 A and S2−1 A are naturally isomorphic, if

Sij = ιSi (Sj ), i = j, is a multiplicative set in Si−1 A such −1 (Si−1 A) is an that the canonical homomorphism ιSij : Si−1 A → Sij isomorphism. def

(d) If two multiplicative sets S1 and S2 are equivalent, then the sets Si and S1 S2 , i = 1, 2, are also equivalent. Assertions (a)–(c) directly follow from the proposition in the previous section. To prove (c), one must additionally have in mind assertion (b). (d) It suﬃces to prove that any element s1 s2 ∈ S1 S2 divides, say, some element from S1 , since the converse is obvious. But s2 divides some element s1 ∈ S1 , hence s2 a = s1 and so s1 s2 a = s1 s1 ∈ S1 . Examples. I. Let U1 and U2 be diﬀeomorphic open domains in a manifold M and let Si = SU1 , i = 1, 2. Then the localizations Si−1 C ∞ (M ) = C ∞ (Ui ) of the algebra C ∞ (M ) are naturally isomorphic only when U1 = U2 , although as abstract algebras they are always isomorphic. xk }k≥0 , S2 = {¯ y k }k≥0 be given in the II. Let A = C ∞ (K) and S1 = {¯ notation of Example II from Section 13.10. Then S1 and S2 not equivalent, although the algebras S1−1 A and S2−1 A are isomorphic. III. Let s ∈ A be a nonnilpotent element and let l, k be nonnegative integers.Then the multiplcative sets {slm }m≥0 and {skm }m≥0 are equivalent if lk > 0, and nonequivalent, if k + l = 0, while the element s is noninvertible. 13.14. Maximal multiplicative sets. Note that any pair of equivalent multiplicative sets is contained in some equivalent multiplicative set (assertion (d) of the proposition from the previous section). In connection with this, one could suppose that there exists a maximal multiplicative set, containing all those equivalent to it. This is indeed so, moreover, the construction of maximal sets is very simple. Proposition. Let S be some multiplicative set in an algebra A and let Sˆ be the sets of divisors of all elements of S. Then the set Sˆ is multiplicative and contains any multiplicative set equivalent to S. ˆ Then ab = s, a b = s for some b, b ∈ A, s, s ∈ S. Since Let a, a ∈ S. (aa )(bb ) = ss ∈ S, it follows that Sˆ is closed w.r.t. multiplication and ˆ Thus, does not contain zero. Obviously, S ⊂ Sˆ and, in particular, 1A ∈ S. the set Sˆ is multiplicative. If a multiplicative set S1 is equivalent to S and s1 ∈ S1 , then s1 b1 = s ˆ for some b1 ∈ A, s ∈ S. Hence, s1 ∈ S. It is important to note that there is a bijection between localizations of the algebra A and maximal multiplicative sets. Examples. The following multiplicative sets are maximal.

Localization

215

I. The set SV for any V ⊂ |A|. II. The set of polynomials of the form λϕk , ϕ ∈ k[x1 , . . . , xn ] under the condition that 0 = λ ∈ k and the polynomial ϕ is irreducible. III. The set A \ I, where I is an ideal. 13.15. Geometricity and localization. The geometricity property of algebras or modules is not preserved in general under localization, and the reasons for that should be understood. First of all, let us note that the canonical homomorphism ι = ιP : P → S −1 P generates a linear map of k-vector spaces ιP,h : Ph → (S −1 P )h¯ , ph → (ι(p))h¯ , Lemma.

¯ ∈ |S −1 A|, h = |ι|(h) ¯ . h

¯ = S −1 (Ker h). 1. Ker h

2. ιP,h is an isomorphism. ¯ The relation h(a/s) = h(s)−1 h(a) immediately implies the ﬁrst assertion. To prove the second one, we ﬁrst note that the map ιP,h is surjective because (p/s)h¯ = (1/s)h¯ (p/1)h¯ = h(s)−1 ι(p)h¯ . Further, if ιP,h (ph ) = 0, then ¯ · S −1 P ⇔ ι(p) = ι(p) ∈ Ker h

(ai /si ) · (pi /si ),

i

si , si

pi

where ai ∈ Ker h, ∈ S and ∈ P . Bringing this sum to a common denominator, we obtain

(ai /1) · (pi /s) = ai pi /s , where s ∈ S . Thus ι(p) = p /s , where p = ai pi ∈ Ker h · P , or, equivalently, we can write s˜p = s˜s p for some s˜ ∈ S. Therefore, h(˜ ss )ph = (˜ ss p)h = (˜ sp )h = h(˜ s)ph = 0 But since we have h(˜ ss ) = 0, it follows from Proposition 13.3 that ph = 0, i.e., the map ιP,h is injective. ¯ ∈ |S −1 A| associated Corollary. The family of vector spaces (S −1 P )h¯ , h −1 −1 with the S A-module S P can be canonically identiﬁed with the family {Ph }, h ∈ |ι|. Exercise. Prove that the S −1 A-module S −1 P is projective, provided the A-module P is projective. 13.16. Let U ⊂ M be an open subset of a manifold M , let A = C ∞ (M ) and P = Γ(π), where π is a vector bundle over M . Proposition. The SU−1 A-module SU−1 P and the C ∞ (U )-module Γ(π|U ) are naturally isomorphic.

216

Chapter 13

Recall that C ∞ (U ) = SU−1 A, where the equality is understood as the natural identiﬁcation (Proposition 11.7). Here the ﬁber π −1 (z) is identiﬁed def

with Pz = Phz , while the element p ∈ P , with the section σp : M z → pz = phz ∈ π −1 (z) def

and this section is smooth. Besides, there are natural identiﬁcations (SU−1 P )z ↔ Pz ↔ π −1 (z), if z ∈ U (obtained in Section 13.15). This shows that (1) the set of maps U z → (SU−1 P )z is identical to the family of ﬁbers of the bundle π|U ; (2) the section σp |U coincides with the section U z → ιP (p)z . But since any element of the module SU−1 P is of the form p/f, f ∈ SU , it follows that the corresponding section of the vector bundle π|U is 1/f ·σp |U . This establishes an isomorphism between the SU−1 A-module SU−1 P and the C ∞ (U )-module Γ(π|U ). 13.17 Proposition. Let P be a geometrical A-module and let S ⊂ A be a multiplicative subset.Then (a) p/s ∈ Inv(S −1 P ) ⇔ ι(p) ∈ Inv(S −1 P ), here Inv is deﬁned in Section 8.4; (b) ι(p) ∈ Inv(S −1 P ) ⇔ (|ι|) ∩ Up = ∅. Here ι(p) = 0 only when we additionally have Up ∩ Us = ∅, ∀s ∈ S; (c) if the given algebra A is geometrical, 0 = a ∈ A and S = SUa , then the algebra S −1 A is also geometrical. (a) Since (p/s)h¯ = h(s)−1 ι(p)h¯ = h(s)−1 ιP,h (ph ) and h(s)−1 = 0, the assertion follows from the fact that ιP,h is an isomorphism (see the lemma in Section 13.15) for all h ∈ (|ι|). (b) First note that ¯ ∈ |S −1 A|} , {ι(p) ∈ Inv(S −1 P )} ⇔ {0 = ι(p)h¯ = ιP,h (ph ), ∀h and since ιP,h is an isomorphism, the last one of these two assertions is equivalent to ph = 0, for any h ∈ (|ι|). But this means precisely that we have (|ι|) ∩ Up = ∅. Further, ι(p) = 0 means that sp = 0 for some s ∈ S. Therefore, ι(p) = 0 if and only if sp = 0 for all s ∈ S. But the module P is geometrical, so this is equivalent to (sp)h = h(s)ph = 0 for some h = hs ∈ |A|. Therefore, hs ∈ Up ∩ Us , i.e., Up ∩ Us = ∅. (c) Since the algebra A is geometrical, we have Ua = ∅ provided a = 0. Also note that a ∈ SUa and Us ⊃ Ua for all s ∈ SUa . Hence, ∩s∈SUa Us = Ua and Proposition 13.3 implies (|ι|) = Ua . If ι(a ) ∈ Inv(AUa ), then item (b) implies Ua ∩ Ua = ∅. On the other hand, since a ∈ SUa , we again see,

Localization

217

by the same proposition, that the intersection Ua ∩ Ua must be nonempty if a = 0. Therefore, a = 0. Corollary. Suppose the ﬁeld k is not algebraically closed and A is some k-algebra. Suppose also that a1 , . . . , am ∈ A, ai = 0, for all i, and U = Ua1 ∪ · · · ∪ Uam . Then the algebra SU−1 A is geometrical. In this case U = Ua for a = ϕ(a1 , . . . , am ), where ϕ(x1 , . . . , xm ) is a polynomial that vanishes only at the point (0, . . . , 0) ∈ km (see Example III in Section 13.5). Example. Let M be a manifold and let ∅ = V ⊂ M be a closed subset. The set of functions S = {f ∈ C ∞ (M ) | Uf ⊃ V }, obviously, is multiplicative and |ιS | = ∩f ∈S Uf = V . Then ∅ = ιS (g) ∈ Inv(S −1 C ∞ (M )) if g|V = 0, but g|U = 0 for any open set U ⊃ V —this follows from assertions (a) and (b) of the previous proposition. In particular, this shows that the algebra S −1 C ∞ (M ) is not geometrical and does not contain any nilpotent elements. Exercise. 1. Let A = k[x1 , . . . , xn ], where k is a ﬁeld of zero characteristic. Show that the algebra S −1 A is geometrical if the set S is ﬁnitely generated. Is this assertion true if k has ﬁnite characteristics? 2. In the polynomial algebra from the previous exercise, ﬁnd a multiplicative system S such that the algebra S −1 A is not geometrical. 3. Let the algebras A and S −1 A, as well as the A-module P , be geometrical. Will the S −1 A-module S −1 P necessarily be geometrical? 13.18. Localization of diﬀerential operators. Let Δ ∈ Diﬀ k (P, Q). Suppose that there exists a localization ΔS : S −1 P → S −1 Q of the operator Δ corresponding to the localization S −1 A of the algebra A, i. e., an operator such that Δ

P

Q S −1

S −1

S −1 P

S

−1

Δ

S −1 Q

commutes. Let us put s˜ = ι(s), s ∈ S. Applying formula (9.24) to ΔS for a1 = s˜−1 , a2 = · · · = ak+1 = s˜, we obtain

k i 0 = [(δs˜) δs˜−1 ](ΔS ) = (−1) s˜ ◦ ΔS ◦ s˜k−i−1 + i 0≤i≤k k + (−1)j s˜j−2 ◦ ΔS ◦ s˜k−j+1 , j−1 k

i

1≤j≤k+1

or, equivalently,

218

Chapter 13

(−1)i−1

0≤i≤k+1

k + 1 i−1 ◦ ΔS ◦ s˜k−i = 0 . s˜ i

The last equality may be rewritten in the form k + 1 j−1 (−1)k−j ◦ ΔS ◦ s˜k−j . s˜ s˜k ◦ ΔS ◦ s˜−1 = j 0≤j≤k

Dividing it by s˜k and applying the obtained result to p ∈ P , we obtain k p k + 1 Δ(si p) Lk (s, Δ)(p) = ΔS (−1)i = , (13.2) i+1 i + 1 s s sk+1 i=0 where k

Lk (s, Δ) =

(−1)i

i=0

k + 1 k−i ◦ Δ ◦ si . s i+1

(13.3)

This heuristic consideration brings us to the following deﬁnition. 13.19. Deﬁnition. The localization of an operator Δ ∈ Diﬀ k (P, Q) Δ : P → Q over the multiplicative set S ⊂ A is deﬁned as the operator given by formula (13.2). However, we must prove that this notion is well deﬁned, i.e., does not depend on the choice of the pair (p, s) representing the formal fraction p/s. To do that, we shall need certain properties of the operator Lk (s, Δ). 13.20 Lemma. The following relations hold for the operators Lk (s, Δ), Δ ∈ Homk (P, Q): (1) Lk (s, a ◦ Δ) = a ◦ Lk (s, Δ), a ∈ A; (2) Lk (s, Δ ◦ a) = Lk (s, Δ) ◦ a, a ∈ A; (3) Lk (s, Δ) = sk Δ − Lk−1 (s, δs (Δ)); (4) Lk (s, Δ) ◦ s = sk+1 Δ; (5) Lk (ss1 , Δ) ◦ s1 = sk+1 Lk (s, Δ); 1 (6) Lk (s, Δ) − s ◦ Lk−1 (s, Δ) = (−1)k δsk (Δ). The ﬁrst two relations are obvious, the third can be proved by a simple direct calculation that we leave to the reader. In its turn, the fourth relation follows from the third by induction over k. Indeed, assuming that (4) holds for k − 1, for any k-linear map, and, in particular, for δs (Δ), we have Lk (s, Δ) ◦ s =sk Δ ◦ s − Lk−1 (s, δs (Δ)) ◦ s =sk ◦ Δ ◦ s − sk δs (Δ) = sk+1 Δ.

Localization

219

Relation (5) is also proved by induction over k. The base of induction is obvious, while the induction step from k − 1 to k is carried out in the following way. First using relation (3), then the induction hypothesis, and ﬁnally relations (1)–(2), we obtain Lk (ss1 , Δ) ◦ s1 = sk sk1 ◦ Δ ◦ s1 − Lk−1 (ss1 , δss1 (Δ)) ◦ s1 =sk sk1 ◦ Δ ◦ s1 − sk1 ◦ Lk−1 (s, δss1 (Δ)) = sk sk1 ◦ Δ ◦ s1 − sk1 ◦ Lk−1 (s, s1 ◦ δs (Δ) + δs1 (Δ) ◦ s) ◦ Lk−1 (s, δs (Δ)) − sk1 Lk−1 (s, δs1 (Δ) ◦ s. = sk sk1 ◦Δ ◦ s1 − sk+1 1 Further, taking into account Lk−1 (s, δs1 (Δ)) ◦ s = sk δs1 (Δ) and using relation (4), we can rewrite the last of these equalities as (3)

sk+1 (sk Δ − Lk−1 (s, δs (Δ))) = sk+1 Lk (s, Δ). 1 1 Finally, relation (6) follows from the fact that k k−i Lk (s, Δ) − s ◦ Lk−1 (s, Δ) = (−1) ◦ Δ ◦ si s i i=0 k

i

if we take into account the relation k k k−i (−1)i ◦ Δ ◦ si = (−1)k δsk (Δ) s i i=0 (see (9.24)). 13.21 Proposition. Let Δ ∈ Diﬀ k (P, Q). Then (a) Deﬁnition 13.19 well deﬁnes ΔS , which is a diﬀerential operator of order ≤ k over the algebra S −1 A; (b) ιQ ◦ Δ = ΔS ◦ ιP ; (c) if ∇ ∈ Diﬀ l (R, Q), then (Δ ◦ ∇)S = ΔS ◦ ∇S ; (d) if A = C ∞ (M ),P = Γ(π1 ), Q = Γ(π2 ) and the set U ⊂ M is open, then ΔSU coincides with the restriction ΔU : Γ(π1 |U ) → Γ(π2 |U ) of the operator Δ to U in the usual sense of the word. (a). Relation (5) of Lemma 13.20 implies that the pairs (Lk (ss1 , Δ)(s1 p), (ss1 )k+1 ) and (Lk (s, Δ)(p), sk+1 ) are equivalent. This proves that the right-hand side of relation (13.2) does not change when the pair (p, s) representing the formal fraction p/s in the left-hand side is replaced by the proportional pair (s1 p, s1 s). On the other hand, the pairs (p1 , s1 ) and (p2 , s2 ) representing the same formal fraction are proportional to the pair (ss1 p2 , ss1 s2 ) = (ss2 p1 , ss1 s2 ),

220

Chapter 13

where the element s ∈ S satisﬁes s(s2 p1 − ss1 p2 ) = 0. This proves that the right-hand side of relation (13.2) does not depend on the choice of the pair representing the formal fraction p/s. To complete the proof of the fact that 13.19 well deﬁnes ΔS , we must also prove that this deﬁnition does not depend on k, i.e., that Lk (s, Δ)(p) Ll (s, Δ)(p) = sk+1 sl+1 if the operator operator Δ is of order l < k. This, obviously, needs to be done only in the case l = k − 1, but in that case, it directly follows from item (6) of Lemma 13.20. Now let us establish that ΔS is a diﬀerential operator of order ≤ k over the algebra S −1 A. In the subsequent formulas and computations needed for this, it will be convenient to use the more compact notation a ˜ instead of ι(a) for a ∈ A. For the proof, we shall use induction on the order k of the localized operator. The expression (13.2) when k = 0 has the following form: ΔS (p/s) = Δ(p)/s. Since in this case Δ is a homomorphism of A-modules, it follows that a p

ap Δ(ap) aΔ(p) ΔS · =ΔS = = s s ss s s s s p

a a Δ(p) = · ΔS . = · s s s s This shows that ΔS is a homomorphism of S −1 A-modules, i.e., a diﬀerential operators of zero order. To carry out the induction step, we shall use the relation δa/s (X) = δa·1/s (X) = δa˜ (X) ◦ s−1 + a ˜ ◦ δs−1 (X),

(13.4)

where a ∈ A, s ∈ S and X is a k-linear map of S −1 A-modules. Using this formula, we obtain 0 = δ1S−1 A (X) = δs/s (X) = δs˜(X) ◦ s−1 + s˜ ◦ δs−1 (X), and therefore, δs−1 (X) = −s−1 ◦ δs˜(X) ◦ s−1 .

(13.5)

The formulas (13.4) and (13.5) reduce the computation of δa/s to that of δ˜b , b = a, s, which is obvious enough. Namely, using relations (1) and (2) of Lemma 13.20 and the fact that the fraction Lk (s, ∇)/sk+1 does not depend on the order of the operator ∇ provided it is not greater than k, we obtain p

p L (s, δ (Δ))(p) k b = , b ∈ A. = δ (Δ) δ˜b (ΔS ) b S s sk+1 s This implies δ˜b (ΔS ) = δb (Δ)S . Together with formulas (13.4) and (13.5) for = ΔS , this yields the required formula δa/s (ΔS ) = δa (Δ)S ◦ s−1 − a ˜s−1 ◦ δs (Δ)S ◦ s−1 .

(13.6)

Localization

221

Indeed, the diﬀerential operators δa (Δ) and δs (Δ) are of order ≤ k − 1. Therefore, by the induction hypothesis, their localization also has the same order and so, in the right-hand side of formula (13.6), an operator of order ≤ k − 1 over the algebra S −1 A appears. Thus, the operator δa/s (ΔS ) is also of order ≤ k − 1 for all p/s ∈ S −1 A, which implies that the operator ΔS is of order ≤ k. (b) This is obvious. (c) Since p

Ll (s, ∇)(p) Lk (sl+1 , Δ)(Ll (s, ∇)(p)) ) = ΔS ΔS (∇S , = s sl+1 s(k+1)(l+1) it suﬃces to prove skl Lk+s (s, Δ ◦ ∇) = Lk (sl+1 , Δ) ◦ (Ll (s, ∇) Now division by ι(s) is possible in S −1 A, and so it suﬃces to prove that skl Lk+s (s, Δ ◦ ∇) ◦ s = Lk (sl+1 , Δ) ◦ (Ll (s, ∇) ◦ s. (13.7) Item (4) in Lemma 13.20 allows us to bring the right-hand side of this hypothetical equality to the form Lk (sl+1 , Δ) ◦ (Ll (s, ∇) ◦ s) = Lk (sl+1 , Δ) ◦ sl+1 ◦ ∇ = s(k+1)(l+1) Δ ◦ ∇. Similarly, we have Lk+s (s, Δ ◦ ∇) ◦ s = sk+l+1 Δ ◦ ∇, which proves (13.7). (d). Note that the relation Lk (1, Δ) = Δ is actually relation (4) from Lemma 13.20 for s = 1. Hence, p L (1, Δ)(p) Δ(p) k = = . (13.8) ΔS 1 1 1 In the case under consideration, we have S = SU and we can identify Γ(π|U ) and S −1 R if R = Γ(π) (see 13.16). Here p/1 = p|U , Δ(p)/1 = Δ(p)|U . Therefore, 13.8 may be rewritten in the form ΔSU (p|U ) = Δ(p)|U . But since Δ(p)|U = Δ|U (p|U ), we have ΔSU (p|U ) = Δ|U (p|U ). From this we can conclude that ΔSU = Δ|U . Indeed, ΔSU is a diﬀerential operator over the algebra S −1 C ∞ (M ) = C ∞ (U ) and so it is a diﬀerential operator in the usual sense of the word (see Section 14.22). Therefore, this operator and hence the operator Δ|U are local (Proposition 14.22). But since in some neighborhood of an arbitrary point z ∈ U any element of the module Γ(π1 |U ) may be represented in the form p|U , it follows that the operators coincide in this neighborhood. But the point z was arbitrary, so they coincide everywhere. Remark. If the element s ∈ A is invertible, then formula (13.2) can be rewritten in the form

222

Chapter 13

−1

Δ s

p =

k i=0

i

(−1)

k + 1 −(i+1) Δ(si p) = s−(k+1) Lk (s, Δ)(p), (13.9) s i+1

describes the result of applying the operator Δ to the “nonformal” fraction p/s = s−1 p. The reader can once again appreciate the advantage of the algebraic approach by trying to ﬁnd this formula on the basis of the classical deﬁnition of diﬀerential operators. 13.22. Localization is well deﬁned. Deﬁnition 13.19 describes the localization of diﬀerential operators using the description of the localized algebra itself in terms of multiplicative sets S. Therefore, we must show that the result does not depend on that description. Proposition. Let the multiplcative sets S1 and S2 be equivalent and let η : S1−1 A → S1−1 A be the natural isomorphism. Then η ◦ ΔS1 = ΔS2 ◦ η. Since the composition of natural isomorphisms is also natural, see Exercise 13.12, it suﬃces to prove this only in the case when S2 is a maximal multiplicative set equivalent to S1 . In this case, S1 ⊂ S2 and any element of S2−1 A may be represented by a formal fraction whose denominator belongs to S1 . Therefore, the expressions for ΔS1 (p/s1 ) and ΔS2 (p/s1 ) in the deﬁning formula 13.2 are exactly identical except that in the ﬁrst one the formal fraction is understood as an element of S1−1 A, while in the second one, as an element of S2−1 A. But in that case, the required assertion becomes obvious,

14 Diﬀerential 1-forms and Jets

14.1. Duality: vectors and covectors. In Chapter 9, considering vector ﬁelds of various types, we came to a generalization that covers them all, namely to the notion of P -valued vector ﬁeld or derivation of the main algebra A with values in an arbitrary A-module P (see Section 9.50). As we have seen, the set of all such vector ﬁelds constitutes an A-module, denoted by D(P ). In this way, we obtain the functor D (see 9.51). On the other hand, geometry, mechanics, and physics naturally require the introduction of notions dual to tangent vectors and to vector ﬁelds. In elementary diﬀerential geometry, this is the notion of cotangent vector, i.e., the diﬀerential of a function at the given point (Sections 9.21 and 9.22). We have previously noted the duality between velocity and momentum in mechanics. We also noted previously that tangent vectors and covectors at a point z of a manifold M are dual to each other in the sense of elementary linear algebra. Recall that tangent covectors are linear functionals on the tangent space Tz M , while tangent vectors are linear functionals on the cotangent space Tz∗ M . This is equivalent to saying that a pairing Tz M ⊕ Tz∗ M → R is deﬁned, i.e., a bilinear form ξ, η ∈ R, where ξ ∈ Tz M, η ∈ Tz∗ M . Similarly, for the quantity dual to the notion of the vector ﬁeld, it is natural to take the notion of covector ﬁeld, which is a family of tangent covectors ω = {ωz }z∈M , ωz ∈ Tz∗ M smoothly depending on z. One of the possible deﬁnitions of smoothness here consists in that, for any vector ﬁeld © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 14

223

224

Chapter 14

X ∈ D(M ), the function M z → Xz , ωz ∈ R is smooth. If this function is denoted by ω(X), then, by deﬁnition, we have ω(X)(z) = Xz , ωz . It is easy to verify that ω(f X) = f ω(X) for f ∈ C ∞ (M ). This implies that the covector ﬁeld ω can also be deﬁned as a C ∞ (M )–linear map ω : D(M ) → C ∞ (M ),

X → ω(X),

∞

i.e., as an element of the C (M )–module D∗ (M ) = HomC ∞ (M ) (D(M ), C ∞ (M )) def

Moreover, we have the isomorphism D(M ) = HomC ∞ (M ) (D∗ (M )), C ∞ (M )),

(14.1)

establishing the duality of vector and covector ﬁelds in the sense of linear algebra. However, the duality understood in this sense no longer holds for vector ﬁelds of other types, as will be shown below, where we shall be using the simpliﬁed notation P ∗ = HomA (P, A) for any P . 14.2 Lemma. Let N ⊂ M be a nowhere dense closed subset of a manifold M . If P is a geometrical C ∞ (M )-module and supp P ⊂ N (see 12.11), then P ∗ = 0. Let a function f ∈ C ∞ (M ) satisfy f |N =0. Then f p = 0, for all p ∈ P . Indeed, (f p)z = f (z)pz ∈ Pz (see Section 12.11) and so (f p)z = 0, provided / N . Thus, (f p)z = 0, for all z ∈ M . z ∈ N . On the other hand, pz = 0 if z ∈ So, since the module P is geometrical, this implies f p = 0. Now let γ ∈ P ∗ . If f |N = 0, then f γ(p) = γ(f p) = 0. Besides, since the set M \ N is open, it follows that the function f may be chosen so that f (z) = 1 if z ∈ M \ N . Therefore, γ(p)(z) = 0, i.e., γ(p)|M \N = 0. Now since M \ N is everywhere dense, this implies that γ(p) = 0, for all p ∈ P , i.e., we have γ = 0. Corollary. Let N ⊂ M be a nowhere dense closed subset of a manifold ∗ M . Then D(M, N )∗ = 0 and so (D(M, N )∗ ) = 0 (see 9.46). It suﬃces to note that supp D(M, N ) = N . Exercise. 14.2.

1. State and prove the general algebraic version of Lemma

2. Suppose the smooth map ϕ : M → N is not a diﬀeomorphism. Can it happen that (D(ϕ)∗ )∗ = D(ϕ) (see 9.47), if we regard D(ϕ) as a I. C ∞ (M )-module? II. C ∞ (N )-module?

Diﬀerential 1-forms and Jets

225

14.3. What instead of duality? By historical inertia, the duality (14.1) has remained in the physics literature, where vectors and covectors are often called contravariant and covariant vectors, respectively. On the contrary, mathematicians prefer to call covector ﬁelds diﬀerential forms of ﬁrst order or simply 1-forms. In view of the importance of the duality that we are now discussing, the following natural question arises: Does this notion have a replacement that would work for all kinds of vector ﬁelds? Thanks to the fact that we are living at a time when mathematicians are familiar with the notions of functor and representing object of a functor, this question does have a positive answer, and we now begin to prepare to explain this. 14.4. Diﬀerential 1-forms. Thus, we begin with the commonly accepted notion of diﬀerential 1-form. Fix a manifold M . The module of sections Γ(πT ∗ ) of the bundle πT ∗ = πT ∗ M is called the module of diﬀerential 1-forms of the manifold M and denoted by Λ1 (M ). The elements of this module, i.e., smooth sections of the bundle πT ∗ , are referred to as diﬀerential 1-forms on the manifold M . According to Section 9.25, any function f ∈ C ∞ (M ) gives rise to a section of πT ∗ deﬁned by sdf : M → T ∗ M,

sdf (z) = dz (f ).

Sections of this kind are called diﬀerentials of smooth functions. The diﬀerential of a function f , viewed as a purely algebraic object, will be denoted by df . The same thing, viewed geometrically, as a map from M to T ∗ M , will be denoted by sdf (the graph of the function of df ). Exercise 9.22 implies that the map d : C ∞ (M ) → Λ1 (M ),

f → df,

is a derivation called the universal derivation of the algebra C ∞ (M ) with values in the C ∞ (M )-module Λ1 (M ). The origin of this term will be clariﬁed later. Now let (U, x) be a chart on M and let πT−1∗ (U ), T ∗ x be the corresponding chart on T ∗ M . In Section 9.24, we denoted by T ∗ x the system of coordinate functions {xi , pj }, where the value of xi at (z, θ) ∈ T ∗ U is the ith coordinate of the point z, while pj is the jth component in the expansion of the covector θ in the basis dxi . Within this chart, every smooth section s has the coordinate representation xi = xi ,

i = 1, . . . , n,

pj = pj (x),

j = 1, . . . , n,

where pj (x) ∈ C ∞ (U ). For simplcity, the section sdxi ∈ Γ(T ∗ U ) = Λ1 (U ) will be denoted by dxi . It follows that the sections dxi , i = 1, . . . , n, form a basis of the free C ∞ (U )-module Λ1 (U ). In particular, the restriction of the diﬀerential form ω ∈ Λ1 (M ) to U belongs to Λ1 (U ) and can, therefore,

226

Chapter 14

be written as ω=

pi (x)dxi .

i

For the diﬀerential of a function df , we can write ∂f dxi df = ∂xi (see the exercise in Section 9.25). Vector ﬁelds on M can be viewed as sections of the tangent bundle (see Section 9.40), and the pairing Γ(πT M ) × Γ(πT ∗ M ) → C ∞ (M ) becomes D(M ) × Λ1 (M ) → C ∞ (M ).

(14.2)

Returning to local coordinates, we recall that for any point z ∈ U ⊂ M the basis dz x1 , . . . , dz xn of the linear space Tz∗ M is by deﬁnition dual to the basis ∂ ∂ ,..., ∂x1 z ∂xn z of the space Tz M . If X is a vector ﬁeld and ω is a 1-form represented in special local coordinates by ∂ αi (x) , ω= pi (x)dxi , X= ∂xi i i then the result of the pairing (X, ω) restricted to U is the function αi pi (x) ∈ C ∞ (U ). X, ωU = i

In particular, for ω = df we obtain ∂f αi = X(f )U . X, df U = ∂xi i 14.5. The universal 1-form. Recall that any 1-form λ ∈ Λ1 (M ), just as the diﬀerential of a function df ∈ Λ1 (M ), can be understood as a section of the cotangent bundle π : T ∗ M → M : sλ : M → T ∗ M, M x → (x, λ|x ) ∈ T ∗ M. It determines the dual maps s∗λ : C ∞ (T ∗ M ) → C ∞ (M ) and s∗λ : Λ1 (T ∗ M ) → Λ1 (M ).

Diﬀerential 1-forms and Jets

227

Deﬁnition. A form ρ ∈ Λ1 (T ∗ M ) is said to be a universal 1-form, if for every 1-form λ ∈ Λ1 (M ) we have s∗λ (ρ) = λ. Proposition. Such a form ρ exists and is unique. An arbitrary point of the manifold T ∗ M can be written as (x, df |x ), where x ∈ M , and f ∈ C ∞ (M ) is some function. Let us put: ρ|(x,df |x ) = d(π ∗ (f ))|(x,df |x ) , def

Let λ ∈ Λ1 (M ) and let f ∈ C ∞ (M ) be a function such that λ|x = df |x . Then s∗λ (ρ)|(x,df |x ) = s∗λ (d(π ∗ (f ))|(x,df |x ) ) = d(s∗λ (π ∗ (f ))|x ) = df |x = λ|x . The uniqueness of ρ is a consequence of the fact that if, for some form ρ ∈ Λ1 (T ∗ M ) and any form λ ∈ Λ1 (M ), we have s∗λ (ρ ) = 0, then ρ = 0. Exercise. Show that in the system of special coordinates (see Section n 9.24), ρ = i=1 pi dxi . 14.6. Universal derivation. The deﬁnition of the module of diﬀerential 1-forms Λ1 (M ) given above was descriptive. Here we intend to give a conceptual deﬁnition of the same notion. Let M be a certain category of modules over a k-algebra A, and Λ ∈ M. A derivation δ : A → Λ is called universal in the category M if for any module P from M the correspondence HomA (Λ, P ) h → h ◦ δ ∈ D(P ) is an isomorphism of the A-modules HomA (Λ, P ) and D(P ). Proposition. The universal derivation, if it exists, is unique up to isomorphism; i.e., if (δ , Λ ) is another universal derivation, then there exists an isomorphism of A-modules γ : Λ → Λ such that δ = γ ◦ δ. The universality of the derivation δ implies that there exists a homomorphism γ : Λ → Λ such that δ = γ ◦ δ. Similarly, δ = γ ◦ δ for an appropriate γ ∈ HomA (Λ , Λ). Therefore, δ = γ ◦ γ ◦ δ and δ ⊂ Λ0 , where Λ0 = {ω ∈ Λ | γ (γ(ω)) = ω} ⊂ Λ. Let α : Λ → Λ/Λ0 be the natural projection. Then 0 = α ◦ δ ∈ D(Λ/Λ0 ), hence α = 0 by the universality of δ. Since α is surjective, it follows that Λ = Λ0 and γ ◦ γ = idΛ . Symmetrically, γ ◦ γ = idΛ ; hence γ and γ are mutually inverse. 14.7 Theorem. The pair (d, Λ1 (C ∞ (M ))) is the universal derivation in the category of geometrical C ∞ (M )-modules.

228

Chapter 14

Let us prove that the natural homomorphism ηP : HomC ∞ (M ) Λ1 (M ), P → D(P ),

h → h ◦ d,

is an isomorphism if the module P is geometrical. To this end, we shall construct the inverse homomorphism νP : D(P ) → HomC ∞ (M ) Λ1 (M ), P , X → hX . 1 We use the fact that any 1-form ω ∈ Λ (M ) can be written as ω = i fi dgi (this will be independently proved below; see Corollary 14.17). Put def fi X(gi ) hX (ω) = i

and let us check that hXis well-deﬁned. Let z ∈ M and f = i ci gi , where ci = fi (z) ∈ R. Then we can write ωz = i ci dz gi = dz f . Moreover, hX (ω)(z) = fi (z)Xz (gi ) = Xz ci gi = Xz (f ). i

i X

Here Xz denotes the composition C ∞ (M ) → P → Pz (see the end of Section 12.11), therefore, the value of hX (ω) at an arbitrary point z ∈ M is well-deﬁned, i.e., it does not depend on the choice of the representation ω = f dg . Since P is geometrical, this implies that hX (ω) is well-deﬁned. i i i In the case ω = dg, we have by deﬁnition hX (dg) = X(g), which means that X = hX ◦ d ⇔ ηP ◦ νP = idD(P ) . If X = h ◦ d, then hX (ω) = fi h(dgi ) = h fi dgi = h(ω), i

and so hX = h ⇔ νP ◦ ηP = idD(P ) . The theorem yields the pairing Λ1 (M ) × D(P ) → P,

i

(ω, X) → hX (ω),

whose result can be written as ω(X) = hX (ω). Exercise. Show that for P = C ∞ (M ) this pairing coincides with the one deﬁned in Section 14.4. 14.8. Conceptual deﬁnition of diﬀerential forms. The theorem proved in the previous section suggests a conceptual approach to the theory of diﬀerential forms over an arbitrary algebra A. Namely, diﬀerential 1-forms are deﬁned as elements of the A-module Λ, the target of the universal derivation which we now denote dM : A → Λ. We stress that this module depends on the choice of the category of A-modules M (see Section 14.6). This module is called the representing object for the functor D in this category and is denoted by ΛM .

Diﬀerential 1-forms and Jets

229

Exercise. Indicate a category of C ∞ (M )-modules in which the functor D is not representable, i.e., does not determine a universal derivation. 14.9 Theorem. In the category of all A-modules over an arbitrary commutative K-algebra A, the functor D is representable. We denote the corresponding universal derivation in that category by dalg : A → Λalg (A). The proof that follows is a typical category theory argument. Consider for all a ∈ A. Let Λ generated by the symbols da 0 be the free A-module Λ the submodule spanning all relations of the form d(ka) − k da,

+ b) − da − db, d(a

− bda, d(ab) − adb

k ∈ K, a, b ∈ A.

Let us put Λ 0 Λalg (A) = Λ/

and

mod Λ 0, dalg a = da

a ∈ A.

− bda ∈Λ 0 , it follows that dalg (ab) − adalg b − bdalg a = 0, Since d(ab) − adb is free, any and so dalg is a derivation. Further, since the A-module Λ → q ∈ Q can be uniquely extended to an map of its generators ψ : da → Q. In particular, this allows to assign to any A-homomorphism ψ : Λ ˜X : Λ → P determined by the map X ∈ D(P ) the A-homomorphism h ˜ X annihi It is obvious that h da → X(a) of the generators of the module Λ. ˜ lates the generators of the A-module Λ0 . Therefore, hX can be represented as the composition Λ pr hX −→ Λ = Λalg (A) −→ P, Λ0 where pr is the natural projection, while hX is an A-homomorphism. Here = hX (pr(da)) = hX (dalg a), i.e., X = hX ◦ dalg . ˜ X (da) X(a) = h 14.10. Is this any good? Now the following natural question arises: Does the previous algebraic construction of 1-forms coincide with what is well known and approved in diﬀerential geometry? In other words, does Λalg (A) coincide with Λ1 (M ) if A = C ∞ (M )? More precisely, since the diﬀerential d : C ∞ (M ) → Λ1 (M ) is a derivation, the universality of the pair (Λalg (A), dalg ) implies that there exists a unique homomorphism, namely hd : Λalg (A) → Λ1 (M ) such that d = hd ◦ dalg . In these terms, our question can be stated as follows: Is hd an isomorphism? The answer turns out to be negative, as the next statement implies. Proposition. Let M = R. Then dalg ex −ex dalg x = 0, i. e., dalg ex −ex dalg x is an example of an invisible element. First let us prove that x and ex are algebraically independent, i.e., the function fn (x)enx + · · · + f1 (x)ex + f0 (x),

230

Chapter 14

where the fi (x) are polynomials over R, is nonzero if at least one of these polynomials may be nonzero. But this easily follows from the fact that enx increases more rapidly than any polynomial or than ekx , k < n, for x → ∞. Further, let us consider the localization of the main algebra A = C ∞ (R) w.r.t. the multiplicative set consisting of the polynomials P (x, ex ). Then the algebra A1 deﬁned in this way contains the ﬁeld F of rational functions in the variables x and y = ex , and we consider the derivation δ = ∂/∂y of this ﬁeld. If, moreover, I is a maximal ideal of the algebra A1 , then the ﬁeld F¯ = A1 /I contains the isomorphic image of the ﬁeld F , which we identify with F itself. It is not too hard to prove that any derivation on a subﬁeld can be extended to a derivation of the whole ﬁeld for a ﬁeld of characteristic zero. In particular, we can extend the derivation δ to the derivation δ¯ of the ﬁeld F¯ . Then ¯ x) = δ(x) = 0, δ(¯

¯ y ) = δ(y) = 1, δ(¯

where z¯ = z mod I ∈ A1 /I = F¯ for z ∈ A1 . Finally, consider the composition δ

A → A1 → F¯ = A1 /I → F¯ , which we denote by ∂. Obviously, it is a derivation of the algebra A with values in the A-module F¯ and, besides, we have ∂(x) = 0 and ∂(ex ) = 1. But ∂ = h ◦ dalg , in view of the universality of the derivation dalg , where h : Λalg (A) → F¯ , is an A-homomorphism. Therefore, the assumption that dalg (ex ) = ex dalg (x) implies the equality ∂(ex ) = ex ∂(x), which is a contradiction, since ∂(ex ) = 1, whereas ∂(x) = 0. Exercise. Let the functions f, g ∈ C ∞ (M ) be algebraically independent and g(x) = φ(f (x)), where φ ∈ C ∞ (R). Prove that dalg (g) = φ (f )dalg (f ). 14.11. Geometrical diﬀerential forms. The fact that Λalg (A) and Λ1 (M ) do not coincide for the algebra C ∞ (M ) might lead us to suspect that the “conceptual” approach to diﬀerential forms is inadequate. However, considerations of observability here will help us out once again. Namely, let Λgeom (A) = Γ(Λalg (A)) and dgeom = ΓΛalg (A) ◦ dalg (see Section 12.11). Theorem. The pair (dgeom , Λgeom (A)) is the universal derivation of the algebra A in the category of geometrical A-modules. In particular, if A = C ∞ (M ), then Λ1 (M ) ∼ = Λgeom (A). Let the A-module P be geometrical and X ∈ D(P ). Then X = halg X ◦ dalg in view of the universality of dalg , where halg : Λ (A) → P is the correalg X sponding A-homomorphism. In its turn, since the module P is geometrical, it follows that halg X can be represented as the composition ΓΛalg (A)

hgeom

X P, Λalg (A) −→ Λgeom (A) −→

Diﬀerential 1-forms and Jets

231

where hgeom is an A-homomorphism (see Proposition in Section 14.6 and X ◦ dgeom . the discussion in Section 12.11), which implies that X = hgeom X Further, as can be easily seen, the uniqueness of the homomorphism follows from the fact that the A-module Λgeom (A) is generated by hgeom X elements of the form dgeom a, a ∈ A. But this is indeed so, because the elements dalg a, a ∈ A, generate Λalg (A), while ΓΛalg (A) is surjective. Corollary. Let f (x) ∈ C ∞ (R). Then dalg (f (x)) − f (x)dalg x ∈ Inv(Λ1alg (A)). Indeed, Γ(dalg (f (x))−f (x)dalg x) = dgeom (f (x))−f (x)dgeom x and dgeom can be identiﬁed with the ordinary diﬀerential d in view of the uniqueness of the universal derivation. 14.12. Universal derivation in other categories. Universal derivations may be constructed in other categories of A-modules. To see this, we will consider some examples of such categories. Let N be a complete subcategory of the category Mod A of all A-modules, i.e., a category such that MorN (P, Q), P, Q ∈ N ⊂ Mod A coincides the set of morphisms MorMod A (P, Q) in Mod A. The simplest such category is the following one. Let A = C ∞ (M ) and z ∈ M . Consider the category GModz C ∞ (M ) whose objects are the geometrical C ∞ (M )-modules whose support is the single point z. In other words, they are modules P such that μz P = 0, or, equivalently, P = Pz . Then dz : C ∞ (M ) → T ∗ M,

f → dz f,

is the universal derivation (or diﬀerential ) in this category. Indeed, if X ∈ D(P ), then X(1) = 0 and X(μ2z ) ⊂ μz P = 0. Therefore, the value X(f ) is uniquely determined by the coset of the function (f − f (z)) ∈ μz by mod μ2z . But, as we have seen (see Section 9.27), this coset can be identiﬁed with dz f . Exercise. Let h ∈ |A| and GModh A be the category of A-modules whose support is the point h. Prove that if k is a ﬁeld, then dh : A → Th∗ A, dh a = (a − h(a)1A ) mod μ2h , is the universal derivation in the category GModh A. Let W be a closed subset in the Zariski topology of |A| and let ModW A and GModW A be the categories whose objects are all, or, respectively, all the geometrical A-modules with support in W . Both of them may be considered as generalizations of GModh A. Let IW = {a ∈ A | h(a) = 0, ∀h ∈ W }. Obviously, IW is an ideal of the algebras A and C ∞ (N ) = C ∞ (M )/IN if N is a submanifold of the manifold M . Lemma. Let k be a ﬁeld, let A be a k-algebra, let P and Q be A-modules and let the subset W ⊂ |A| be closed. Then

232

Chapter 14

(i) if W = supp(P ), then IW · P ⊂ Inv(P ), in particular, IW · P = 0 if the module P is geometrical ; (ii) supp Q/(IW · Q) ⊂ W ; (iii) if the module P is geometrical and W = supp P , then any homomorphism ϕ : Q → P can be uniquely represented as at the composition of homomorphisms pr

ϕ

Q −→ Q/(IW · Q + Inv(Q)) −→ P, where pr is the natural projection. (i) Let a ∈ IW , p ∈ P and h ∈ |A|. Let us show that (ap)h = h(a)ph = 0. Indeed, ph = 0 if h ∈ |A| \ W , and h(a) = 0 if h ∈ W . (ii) Let h ∈ |A| \ W and q ∈ Q. Since |A| \ W is open, it follows that h ∈ Ua ⊂ |A| \ W for some a ∈ A. The latter means that h(a) = 0 and a ∈ IW . Therefore, (a q)h = qh , where a = 1/h(a) · a ∈ IW , and so Q = IW · Q + μh · Q. It remains to note that Q/(IW · Q) h = Q/(IW · Q + μh · Q). (iii) It suﬃces to prove that (IW · Q + Inv(Q)) ⊂ Ker ϕ. Since we have ϕ(IW · Q) ⊂ IW · P , it follows that IW · Q ⊂ Ker ϕ in view of (i). Besides, ϕ(Inv(Q)) ⊂ Inv(P ) = 0, since the module P is geometrical. Proposition. The composition dgeom pr A −→ Λ1geom (A) −→ Λ1geom (A)/ IW · Λ1geom (A) , denoted by dW geom , where pr is the factorization map, is the universal derivation in the category GModW A. Let the A-module P be geometrical, supp P ⊂ W and X ∈ D(P ). Then, in view of the universality of the diﬀerential dgeom , we have X = hX ◦dgeom , where hX : Λ1geom (A) → P is a homomorphism. In its turn, assertion (iii) of the previous lemma shows that hX can be represented as the composition hW pr X P Λ1geom (A) −→ Λ1geom (A)/ IW · Λ1geom (A) , −→ W W for the natural choice of hW X , and so X = hX ◦ dgeom . Moreover, the uniqueness of this representation follows from the fact that the quotient module Λ1geom (A)/ IW · Λ1geom (A) is an object of the category GModW A, as statement (ii) of the previous lemma asserts. Similar facts hold for other functors of the diﬀerential calculus. For the functor Diﬀ l , they are discussed below.

14.13. Behavior of diﬀerential forms under smooth maps of manifolds. The universality property is the cause of the fact that 1-forms, just like functions, are naturally transformed by smooth maps. Let F : M → N

Diﬀerential 1-forms and Jets

233

be a smooth map of manifolds. Note that the composition F∗

C ∞ (N ) −→ C ∞ (M ) −→ Λ1 (M ) d

is a derivation of the algebra C ∞ (N ) with values in the C ∞ (N )-module Λ1 (M ). We recall that the C ∞ (N )-module structure in Λ1 (M ) is deﬁned by (f, ω) → F ∗ (f )ω,

f ∈ C ∞ (N ), ω ∈ Λ1 (M ).

Exercise. Prove that Λ1 (M ) is a geometrical C ∞ (N )-module. By Theorem 14.7, there exists a C ∞ (N )-homomorphism hd◦F ∗ : Λ1 (N ) → Λ1 (M ) such that hd◦F ∗ ◦ d = d ◦ F ∗ . For the sake of brevity, we shall write F ∗ instead of hd◦F ∗ . We thus have an R-linear map F ∗ : Λ1 (N ) → Λ1 (M ), with the following properties: (i) F ∗ ◦ d = d ◦ F ∗ ; (ii) F ∗ (f ω) = F ∗ (f )F ∗ (ω), if f ∈ C ∞ (N ), ω ∈ Λ1 (N ). In view of (i) and (ii), we see that F ∗ (ω) = F ∗ (fi )dF ∗ (gi ),

if ω =

i

fi dgi .

i

Exercise. Show that G

F

1. (F ◦ G)∗ = G∗ ◦ F ∗ , where L → M → N ; 2. (F ∗ )−1 = (F −1 )∗ ; 3. F ∗ (ω)z (ξ) = ωF (z) (dz F (ξ)), or, equivalently, F ∗ (ω)z = (dz F )∗ (ωF (z) ), where z ∈ M, ξ ∈ Tz∗ (M ) and ω ∈ Λ1 (N ). 14.14. The jet algebras J l (M ). We return to the case A = C ∞ (M ) and for every natural l deﬁne the C ∞ (M )-module J l (M ) as the module of sections of the vector bundle πJ l : J l M → M (see Example IX in Section 11.11). The elements of this module are referred to as l-jets on the manifold M . According to Section 9.65, every function f ∈ C ∞ (M ) gives rise to the section sjl (f ) : M → J l M,

z → [f ]lz .

Sections of this kind are called l-jets of smooth functions. Note that the multiplication in the algebra C ∞ (M ) induces an algebra structure in each ﬁber of the bundle πJ l . In fact, let z ∈ M, f, g ∈ C ∞ (M ),

234

Chapter 14

l+1 and h ∈ μl+1 z . Then f g = f (g + h) mod μz . Therefore, the formula def

[f ]lz · [g]lz = [f g]lz gives a well-deﬁned product in the ﬁber Jzl M . This multiplication induces a C ∞ (M )-algebra structure in the module J l (M ) = Γ(πJ l ). Using this multiplication, we can give a more transparent coordinate expression of jets. Let δxi = jl (xi ) − xi jl (1) and δ σ = δxi1 · · · δxik if σ = (i1 , . . . , ik ), 0 < |σ| l. We also put δx∅ = jl (1), and then the sections δ σ , |σ| l, form a basis of the vector bundle πJ l over U . This follows from the fact that the l-jets of the polynomials (x − z)σ , |σ| l, form a basis of Jzl M . We see that a jet of order l on the manifold M in the special coordinate system corresponding to a local chart (U, x) can be written as |σ|l ασ δxσ , where ασ ∈ A. 14.15. Jet algebras J l (M ) as representing objects. With respect to diﬀerential operators in the algebra C ∞ (M ), the jet algebras J l (M ) play a role similar to the role of the module of diﬀerential forms Λ1 (M ) with respect to derivations. This fact can be proved by an argument very close to the one that we used in Section 14.4 (also see Theorem 14.19 below). If we forget the geometrical meaning of the module J l (M ) as the module of sections of the bundle J l M and view its elements from a purely algebraic standpoint, we can still denote the jet of a function f ∈ C ∞ (M ) by jl (f ); otherwise, we denote it by sjl (f ) . 14.16 Theorem. There is a ﬁnite set of functions f1 , . . . , fm ∈ C ∞ (M ) whose l-jets jl (f1 ), . . . , jl (fm ) generate the C ∞ (M )-module J l (M ). If M = Rk , then as such we can take the set of all monomials xσ with |σ| l, where x = (x1 , . . . , xk ) are the usual coordinates in Rk . In the general case, choose an appropriate k ∈ N and consider an immersion F : M → Rk (Whitney’s immersion theorem). Then the family F ∗ (xσ ), |σ| l, will have the required property. Indeed, by Corollary 12.28, it suﬃces to show that for every point z ∈ M the l-jets [F ∗ (xσ )]lz generate the vector space Jzl M . Since F is an immersion, the diﬀerential dz F is injective; hence (dz F )∗ is surjective. Therefore, among the diﬀerentials dz F ∗ (xi ), i = 1, . . . , k, there are n = dim M linearly independent ones, say dz F ∗ (x1 ), . . . , dz F ∗ (xn ). The functions F ∗ (x1 ), . . . , F ∗ (xn ) form a local system of coordinates near the point z. As Corollary 2.9 to the generalized Hadamard lemma shows, the monomials of degree l in these variables generate the space Jzl M . 14.17 Corollary. There is a ﬁnite set of functions f1 , . . . , fm from C ∞ (M ) whose diﬀerentials df1 , . . . , dfm generate the C ∞ (M )-module Λ1 (M ).

Diﬀerential 1-forms and Jets

235

Let f1 , . . . , fm be the functions whose jets generate J 1 (M ). Then their diﬀerentials generate Λ1 (M ). Indeed, the canonical direct decomposition Jz1 M = R ⊕ Tz∗ M (see the proof of Corollary 9.27) shows that the bundle πJ 1 is the direct sum of the two bundles IM : M × R → M and πT ∗ . Passing to the modules of sections and using Proposition 12.23, we infer that J 1 (M ) = C ∞ (M ) ⊕ Λ1 (M ). Therefore, the images of the elements that generate J 1 (M ) under the projection J 1 (M ) → Λ1 (M ) generate Λ1 (M ). It remains to notice that the image of j1 (f ) is df . 14.18 Proposition. The R-homomorphism jl : C ∞ (M ) → J l (M ), given by f → jl (f ) is a diﬀerential operator of order l, i.e., it satisﬁes Deﬁnition 9.57. We must prove that δg0 ◦ · · · ◦ δgl (jl ) = 0 for any functions g0 , . . . , gl . Let θ ∈ J l (M ) and Δ = θ · jl : C ∞ (M ) → J l (M ) → M,

Δ(f ) = (θ ◦ jl )(f ).

Then δg (Δ)(f ) = θ · jl (gf ) − gθ · jl (f ) = θ · jl (g) · jl (f ) − θ · gjl (f ) = jl (g) − gjl (1) · θ · jl (f ) = δl (g) · θ · jl (f ) = δl (g) · Δ(f ), where δl (g) = jl (g) − gjl (1). Therefore, δg0 ◦ · · · ◦ δgl (jl ) (f ) = δl (g0 ) · · · δl (gl ) · jl (f ). The required fact follows from the equality δl (g0 ) · · · δl (gl ) = 0. To prove the latter, note that the image of the element δl (g) under the natural projection J l (M ) → J 0 (M ) = C ∞ (M ),

jl (f ) → j0 (f ),

is equal to δ0 (g) = 0. This means that the value of the section δl (g) at any point z ∈ M belongs to μz Jzl M = μz /μl+1 z . Hence l l+1 l+1 δl (g0 ) · · · δl (gl ) ∈ μl+1 = 0. z Jz M = μz /μz

The element δl (g0 ) · · · δl (gl ) is thus the zero section of the vector bundle πJ l , i.e., the zero element of the module J l (M ). The operator jl : C ∞ (M ) → J l (M ) is referred to as the universal differential operator of order l on the manifold M . The origin of the word “universal” will become clear in a little while. If P is a geometrical C ∞ (M )-module, then there is a natural pairing Diﬀ l P × J l (M ) → P. Indeed, suppose that Δ ∈ Diﬀ l P , Θ ∈ J l (M ), z ∈ M , and let f ∈ C ∞ (M ) be a smooth function such that Θ(z) = [f ]lz . Put (Δ, Θ)(z) = Δ(f )(z) ∈ Pz .

236

Chapter 14

By virtue of Corollary 9.64, the value (Δ, Θ)(z) does not depend on the choice of f . Exercise. As in the proof of Theorem 14.7, show that the totality of all values Δ(f )(z) ∈ Pz uniquely determines the element (Δ, Θ) ∈ P . To an arbitrary operator Δ ∈ Diﬀ l P , we can assign the homomorphism of C ∞ (M )-modules hΔ : J l (M ) → P by putting hΔ (Θ) = (Δ, Θ). It follows from the deﬁnition of the pairing that we have (Δ, jl (f )) = Δ(f ). Therefore, jl ◦ hΔ = Δ. On the other hand, if h : J l (M ) → P is an arbitrary C ∞ (M )homomorphism, then the composition Δh = h ◦ jl : C ∞ (M ) → P, according to 9.67, 9.59, is a diﬀerential operator of order l that satisﬁes hΔh = h. We thus arrive at the following important result: 14.19 Theorem. For any geometrical C ∞ (M )-module P , the assignment HomC ∞ (M ) (J l (M ), P ) h → h ◦ jl ∈ Diﬀ l P deﬁnes a natural isomorphism of C ∞ (M )-modules HomC ∞ (M ) (J l (M ), P ) ∼ = Diﬀ l P. In other words, the functor Diﬀ l deﬁned on the category of geometrical C ∞ (M )-modules is representable, with representing object J l (M ). This theorem explains why the diﬀerential operator jl : C ∞ (M ) → J l (M ) is called universal. As in the case of diﬀerential forms, the assumption that P is geometrical is essential: without it, Theorem 14.19 is not valid. The signiﬁcance of Theorem 14.19 is also due to the fact that it shows how to correctly introduce the notion of the jet in the diﬀerential calculus over any commutative K-algebra A. To do this, one must, ﬁrst of all, choose the corresponding category of A-modules, say M (see Section 14.6) and then l (A) as the range of values of the universal deﬁne the module of l-jets JM M l (A). The universality of the operator jlM diﬀerential operator jl : A → JM means that for any module P in the category M, the correspondence l HomA (JM (A), P ) h → h ◦ jl ∈ Diﬀ l P l determines a natural A-module isomorphism Hom(JM (A), P ) ∼ = Diﬀ l P . l Exercise. Describe the module of l-jets JM (A) and the module of 1-forms Λ1M (A), where A is the algebra of smooth functions on the cross K, while M is the category of geometrical modules over this algebra.

14.20. Jet bundles. Suppose that P is a projective C ∞ (M )-module, say, the module of sections of a vector bundle πP : E → M , and μz ∈ C ∞ (M ) is the maximal ideal corresponding to the point z ∈ M . Note that μl+1 z P is

Diﬀerential 1-forms and Jets

237

def

a submodule of P , and let Jzl P = P/μl+1 z P be the quotient module. The image of the element p ∈ P under the natural projection will be denoted by [p]lz ∈ Jzl P . The vector space Jzl P is a module over the algebra Jzl M with respect to the multiplication def

[f ]lz [p]lz = [f p]lz ,

f ∈ C ∞ (M ), p ∈ P.

Exercise. Prove that this multiplication is well deﬁned. def

Put J l P = z∈M Jzl P . Our nearest aim is to equip the set J l P with the structure of a smooth manifold in such a way that the natural projection πJ l P : J l P → M,

Jzl P → z ∈ M,

will deﬁne a vector bundle structure over M on the smooth manifold J l P . This vector bundle will be called the bundle of jets of order l (or l-jets) of the bundle πP . On the total space E of the bundle πP , there is an adapted atlas (see Section 12.3). Its charts are of the form (π −1 (U ), x, u), where (U, x) is a chart of the corresponding atlas on M and u = (u1 , . . . , um ), m = dim πP , are the ﬁber coordinates. Then, according to Proposition 12.13, the local ization PU = Γ(πP U ) is a free C ∞ (U )-module. Let e1 , . . . , em be its basis. The restriction of an element q ∈ P to U can be written as m q U = f i ei ,

where f i ∈ C ∞ (U ).

i=1

In other words, in the adapted coordinates, any section of the vector bundle is represented by a vector function (f 1 , . . . , f m ) in the variables (x1 , . . . , xn ). This implies that [q]lz is uniquely determined by the m-uple ([f 1 ]lz , . . . , [f m ]lz ), and therefore, the collection of numbers ∂ |σ| f j x1 , . . . , xn , u1 , . . . , um , . . . , piσ , . . . , |σ| l, pjσ = , ∂xσ z uniquely determines the point [q]lz ∈ πJ−1 uj , . . . , piσ , . . . are l P (U ), where the special local coordinates of the l-jet of the function f j in J l M (see Section 11.11, IX). The functions xi , uj , pjσ ,

1 i n, 1 j m, 0 < |σ| l,

l form a coordinate system in the domain πJ−1 l P (U ) ⊂ J P . Charts of this l type will be referred to as special charts on J P .

Exercise. Show that the special charts on J l P that correspond to compatible charts on M are compatible as well. In other words, these charts form an atlas, thus deﬁning the structure of a smooth manifold in J l P .

238

Chapter 14

The projection πJ l P : J l P → M,

Jzl P → z ∈ M,

is evidently smooth. The special charts described above are direct products of the form U × RmN , where N is the number of all derivatives of order l (see Example IX in Section 11.11). Therefore, πJ l P is a locally trivial mN bundle. Moreover, the trivializing diﬀeomorphisms πJ−1 l P (U ) → U × R are linear on each ﬁber, so that πJ l P is a vector bundle over M . 14.21. Let A = C ∞ (M ), z ∈ M , and let P and Q be C ∞ (M )-modules corresponding to the vector bundles πP and πQ over M , respectively. We have the following generalization of Corollary 9.61. 14.22 Proposition. Suppose that p1 , p2 ∈ P , Δ ∈ Diﬀ l (P, Q), and p1 − p2 ∈ μl+1 z P . Then Δ(p1 )(z) = Δ(p2 )(z). In particular, if the elements p1 , p2 ∈ P coincide in a neighborhood U z, then Δ(p1 )(z) = Δ(p2 )(z) for any diﬀerential operator Δ ∈ Diﬀ(P, Q). In other words, the diﬀerential operators that act on sections of vector bundles are local. Indeed, if p1 − p2 ∈ μl+1 z P , then Δ(p1 − p2 ) ∈ μz Q by Proposition 9. Now, if the two sections p1 and p2 ∈ P coincide in a neighborhood of U z, then p1 − p2 ∈ μl+1 z P for any l. This proposition allows us to well deﬁne the localization ΔU : P U → QU , ΔU (¯ p)(z) = Δ(p)(z), p¯ ∈ P U , p ∈ P, z ∈ U, for any diﬀerential operator Δ ∈ Diﬀ(P, Q) and any open set U ⊂ M , where p is an arbitrary element of the module P coinciding with p¯ in a certain neighborhood of the point z. According to this deﬁnition, ΔU (pU )(z) = Δ(p)U (z) if p ∈ P . The operator Δ is uniquely determined by its restrictions to the charts of an arbitrary atlas of the manifold M . Now ﬁx a system of local coordinates x1 , . . . , xn in a neighborhood U ⊂ M so that both vector bundles πP U and πQ U are trivial. Let e1 , . . . , em and ε1 , . . . , εk be bases of the modules P U and QU , respectively. Then the restriction of the elements p ∈ P and q ∈ Q to U can be represented as m f i ei , pU = i=1

k qU = g r εr ,

where f i , g r ∈ C ∞ (U ).

r=1

Fixing the bases e1 , . . . , em and ε1 , . . . , εr , we can deﬁne the following C ∞ (U )-linear maps: αi : C ∞ (U ) → P U , f → f ei , 1 i m, βj : QU → C ∞ (U ),

k r=1

g r εr → g j ,

1 j r.

Diﬀerential 1-forms and Jets

239

def

The composition Δi,j = βj ◦ Δ ◦ αi : C ∞ (U ) → C ∞ (U ) is, according to Sections 9.67 and 9.59, a diﬀerential operator of order l. For scalar diﬀerential operators, we have already proved that the algebraic deﬁnition 9.57 coincides with the conventional one. Now we action of the see that the operator ΔU on pU , i.e., on a vector function f 1 , . . . , f m , is given by ⎛ ⎞⎛ ⎞ ⎛ ⎞ Δ1,1 . . . Δ1,m f1 Δ1,1 (f1 ) + . . . + Δ1,m (fm ) ⎜ .. ⎟ .. ⎟ ⎜ .. ⎟ = ⎜ .. .. ⎝ . ⎠. . . ⎠⎝ . ⎠ ⎝ . Δk,1

...

Δk,m

Δk,1 (f1 ) + . . . + Δk,m (fm )

fm

It follows that the standard notion of matrix diﬀerential operator is a particular case of the general algebraic deﬁnition 9.67, since for the scalar diﬀerential operators, like Δi,j , this fact has already been established (see Section 9.62). Matrix diﬀerential operators are the coordinate description of diﬀerential operators (in the sense of Deﬁnition 9.67) that send the sections of one vector bundle to the sections of the other over the algebra A = C ∞ (M ). 14.23. Jet modules. The module of smooth sections of the vector bundle πJ l P is called the module of l-jets of the bundle πP and is denoted by J l (P ). The elements of this module are called geometrical l-jets of the module P or simply l-jets. It is worth noting that the C ∞ (M )-module J l (P ) is also a J l (M )-module with respect to the following multiplication, def

(θ · Θ)(z) = θ(z)Θ(z),

θ ∈ J l (M ), Θ ∈ J l (P ),

where the multiplication Jzl M ×Jzl P → Jzl P that appears on the right-hand side was deﬁned in Section 14.20. As in the scalar case, any element of the module P gives rise to a section jl (p) of the bundle πJ l P deﬁned by jl (p)(z) = [p]lz . Suppose that in the adapted coordinates p is represented by the vector function f 1 , . . . , f m . Then in the corresponding special coordinates on J l P , the section jl (p) takes the form of the vector function ∂ |σ| f i 1 m , . . . , |σ| l. f ,...,f ,..., ∂xσ The coordinate expression of jl (p) shows, ﬁrst, that this section is smooth, i.e., that jl (p) ∈ J l (P ), and, second, that the R-linear map jl : P → J l (P ),

p → jl (p),

is a diﬀerential operator of order l. Exercise. Give a coordinate-free proof of these facts. The operator jl is referred to as the universal diﬀerential operator of order l in the bundle πP .

240

Chapter 14

14.24 Proposition. Let P be a projective C ∞ (M )-module. There exists a ﬁnite set of elements p1 , . . . , pm ∈ P such that the corresponding l-jets jl (p1 ), . . . , jl (pm ) generate the C ∞ (M )-module J l (P ). Let p¯1 , . . . , p¯k ∈ P be a ﬁnite system of generators of the module P (see Corollary 12.28). Let f1 , . . . , fs be a ﬁnite set of functions whose l-jets generate J l (M ) (Proposition 14.16). Then the l-jets jl (fi p¯j ),

1 i s, 1 j k,

generate J l (P ). To establish this fact, it is suﬃcient, by Proposition 12.10, to show that the l-jets [fi p¯j ]lz generate the ﬁber Jzl P for any z ∈ M . We know that any element of this ﬁber has the form [p]lz for a certain p ∈ P . Let p= gj p¯j , gj ∈ C ∞ (M ), and [gj ]lz = λji [fj ]lz , λji ∈ R. j

Then [p]lz =

j

i

[gj p¯j ]lz =

[gj ]lz [¯ pj ]lz =

j

λji [fi ]lz [¯ pj ]lz =

i,j

λji [fi p¯j ]lz .

i,j

Suppose that Q is a geometrical C ∞ (M )-module. Following the approach used in Section 14.18, we deﬁne the pairing (·, ·) : Diﬀ l (P, Q) × J l (P ) → Q. For a point z ∈ M and a jet Θ ∈ J l (M ), we can choose an element p ∈ P such that Θ(z) = [p]lz . For an arbitrary diﬀerential operator Δ ∈ Diﬀ l (P, Q), let us put (Δ, Θ)(z) = Δ(p)(z) ∈ Qz . By virtue of Proposition 14.22, the value (Δ, Θ)(z) does not depend on the choice of the l representative in the class [p]z . If Θ = i hi jl (pi ) (see Proposition 14.24), then for p we can take the element i λi pi , where λi = hi (z) ∈ R. Therefore, Δ(p)(z) = λi Δ(pi )(z) = hi Δ(pi ) (z). i

i

Since the module Q is geometrical, it follows that hi Δ(pi ) ∈ Q. (Δ, Θ) = i

This proves the existence of the pairing. Proceeding as at the end of Section 14.18, we assign to any operator Δ ∈ Diﬀ l (P, Q) the homomorphism of C ∞ (M )-modules hΔ : J l (P ) → Q,

hΔ (Θ) = (Δ, Θ).

It follows from the deﬁnition of the pairing that (Δ, jl (p)) = Δ(p). Therefore, hΔ ◦ jl = Δ. On the other hand, if h : J l (P ) → Q is an arbitrary

Diﬀerential 1-forms and Jets

241

C ∞ (M )-homomorphism, then the composition Δh = h ◦ jl : P → Q is, according to Sections 9.59 and 9.67, a diﬀerential operator of order l. Also, evidently, hΔh = h. We have thus established the following important fact. 14.25 Theorem. Let a projective C ∞ (M )-module P be given. For any geometrical C ∞ (M )-module Q, the correspondence HomC ∞ (M ) (J l (P ), Q) h → h ◦ jl ∈ Diﬀ l (P, Q) deﬁnes a natural isomorphism of C ∞ (M )-modules HomC ∞ (M ) (J l (P ), Q) ∼ = Diﬀ l (P, Q). In other words, the functor Q → Diﬀ l (P, Q) is representable in the category of geometrical C ∞ (M )-modules, and the C ∞ (M )-module J l (P ) is its representing object. The last theorem makes it possible to change our point of view and deﬁne the module J l (P ) (together with the operator jl : P → J l (P )) as the representing object of the functor Q → Diﬀ l (P, Q) in the category of geometrical C ∞ (M )-modules. This approach is conceptual and thereby immediately extends to arbitrary algebras and categories of modules. Of course, the question of existence must be answered separately in each particular case. Exercise. Prove that the A-module Diﬀ l (P, Q) is geometric, provided that the module Q is geometrical.

15 Functors of the Diﬀerential Calculus and their Representations

15.1. In this chapter, we shall describe the main functors of the diﬀerential calculus in commutative algebras, the corresponding objects and the natural transformations that represent them, as well as the main rules that they satisfy. In the previous chapter, we used, for the same goal, standard constructions coming from diﬀerential geometry, whereas now we change our approach to the categorial algebraic one. The latter turns out to be much more ﬂexible and universal and works perfectly in situations where the classical geometrical approach leads to diﬃculties, sometimes unsurmountable. Moreover, thanks to this approach, many generalizations are carried out automatically. One of the most important applications of this algebraic approach to the diﬀerential calculus is the contemporary theory of nonlinear diﬀerential equations, in which the analogs of the standard diﬀerential geometrical structures are studied on inﬁnite-dimensional manifolds of special type. These are the inﬁnite extensions of diﬀerential equations and, in particular, the spaces of jets of inﬁnite order, their diﬀerential coverings, and other inverse limits of projective systems of smooth manifolds. Manifolds of this class do not possess a norm, nor a suﬃciently strong topology or any other structure traditionally used in the construction of the diﬀerential of a function and other elements of the diﬀerential calculus on ﬁnite-dimensional manifolds. Classical tradition itself lies in the belief that a necessary prerequisite to the construction of the basic elements of the diﬀerential calculus in any context is the notion of limit. Another important application, where the approach to the diﬀerential calculus developed in this book appears to have no valid alternative, initially arose in theoretical physics. Its diﬀerent parts are speciﬁed by the preﬁx “super”, for instance, super-manifolds, super-symmetries, and so on. It eventually turned out that this “super-mathematics”, which initially © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 15

243

244

Chapter 15

seemed enigmatic and mysterious, actually enters in a natural way in the much wider context of the diﬀerential calculus over graded commutative algebras as a very particular case. This important generalization is the subject matter of the last chapter of this book. As we have already seen, such basic structures of the diﬀerential calculus as vector ﬁelds, diﬀerential 1-forms, jets, and even the entire Hamiltonian formalism, may be constructed in any commutative algebra. Moreover, in that context, geometrical portraits of these algebras arise on their spectra. Actually, all this occurs for all the structures of the diﬀerential calculus, in particular—all the diﬀerential geometrical structures. Thus, commutative algebras become objects of diﬀerential geometry. This possibility is inherent in the structure of the calculus of functors of the diﬀerential calculus and the present chapter is an introduction to that calculus. Informally, it might be called the “logic” of the diﬀerential calculus. In what follows, we replace the long expression “functor of the diﬀerential calculus” by the term diﬀunctor. This is a new term and is introduced here for the ﬁrst time. Since it is impossible to give a general deﬁnition of this notion within the framework of the present book, in what follows we will limit ourselves to the description of several main examples of diﬀunctors and show how to work with them. In a certain sense, these diﬀunctors generate the algebra of diﬀunctors. Further, talking of diﬀunctors in general, we simply stress that some fact or some construction holds for all diﬀunctors, and not only those that are explicitly described in this chapter. Below we will be considering modules over a certain commutative k-algebra A, where, for simplicity, we assume that k is a ﬁeld. In the cases where this must be speciﬁed, we denote the operator of multiplication of elements of a module P by an element a ∈ A by aP . The fact that O is an object of the category K is expressed, by abuse of notation, as O ∈ Ob K. Remark. Let P , Q be A-modules and ϕ : P → Q be a k-linear map; then ϕ ◦ aP and aQ ◦ ϕ, a ∈ A, will also be maps of P to Q: aQ ◦ ϕ (p) = aϕ(p), p ∈ P. ϕ ◦ aP (p) = ϕ(ap), In what follows, we will often omit the lower indices indicating the modules on which the operators of multiplication by elements of the algebra A work, especially in the cases where these modules have cumbersome notations such as Diﬀ l (Diﬀ k (C ∞ (M ))). It seems to us that the expression a ◦ ϕ, ϕ ∈ Diﬀ l (Diﬀ k (C ∞ (M ))), looks simpler and—most important—is easier to understand than aDiﬀ k (C ∞ (M )) ◦ ϕ, ϕ ∈ Diﬀ l (Diﬀ k (C ∞ (M ))). We hope that this will not confuse the reader, since what speciﬁc modules are understood is always clear from the context. 15.2. Representing objects and homomorphisms. In this section, we recall the notion of representing the object and the general categorial

Functors of the diﬀerential calculus and their representations

245

scheme in accordance with which to a natural transformation of functors one assigns a morphism of their representing objects. Here we assume, since this is the only case needed in what follows, that everything takes place in the category Mod A and, in particular, that representing objects are A-modules, while morphisms are homomorphisms of A-modules. First recall that a representing object for a covariant functor Φ from some category of A-modules M to itself is an A-module OΦ such that for any P ∈ Ob M there is an isomorphism iΦ = iΦ,P : Φ(P ) → HomA (OΦ , P ). Here we assume that for any homomorphism h : P → Q, we have the commutative diagram Φ(P )

Φ(h)

(15.1)

Φ(Q)

iΦ,P

iΦ,Q

HomA (OΦ , P )

HomA (OΦ ,h)

HomA (OΦ , Q) ,

where HomA (OΦ , h)(ϕ) = h ◦ ϕ, ϕ ∈ HomA (OΦ , P ). In other words, the above means that the system of isomorphisms iΦ allows to identify the functor Φ with the functor HomA (OΦ , ·) : P → HomA (OΦ , P ). Let Φi be a certain functor and OΦi be its representing A-module, i = 1, 2. Assume also that F is a natural transformation of the functor Φ1 to the functor Φ2 . This means that for any arbitrary homomorphism of A-modules h : Q → R we have the two following commutative diagrams Φ1 (Q)

Φ1 (h)

Φ1 (R)

F (Q)

Φ2 (Q)

F (R) Φ2 (h)

Φ2 (R)

Φ1 (OΦ1 )

Φ1 (h)

Φ1 (P )

F (OΦ1 )

(15.2)

F (P )

Φ2 (OΦ1 )

Φ2 (h)

Φ2 (P ) .

From now on we simplify the notation OΦ1 to O1 and similarly for Φ2 . The diagram on the right is a particular case of the one on the left for Q = O1 , R = P . From the latter, using the isomorphisms iΦ1 ,O1 and iΦ1 ,P , we obtain the diagram HomA (O1 , O1 )

HomA (O1 ,h)

iF (O1 )

HomA (O2 , O1 )

HomA (O1 , P )

(15.3)

iF (P ) HomA (O2 ,h)

HomA (O2 , P ) ,

where we have used the notation iF (Q) = iΦ2 ,Q ◦ F (Q) ◦ i−1 Φ1 ,Q . Note further that HomA (O1 , h)(idO1 ) = h. Hence, the commutativity of diagram (15.3) for the identity homomorphism idO1 ∈ HomA (O1 , O1 )

246

Chapter 15

means that def

iF (P )(h) = h ◦ HF , where HF = iF (O1 )(idO1 ) ∈ HomA (O2 , O1 ). Thus, we obtain Proposition. If two functors Φi , i = 1, 2, are given each of which is identiﬁed with the functor HomA (Oi , ·), then any natural transformation F : Φ1 → Φ2 is described by the homomorphism of the representing objects HF : O2 → O1 : Φ1 (P ) ∼ = HomA (O1 , P ) h → h ◦ HF ∈ HomA (O2 , P ) ∼ = Φ2 (P ). The homomorphism HF is called representing for the transformation F . Exercise. Show that any homomorphism H : O2 → O1 determines a natural transformation of the functor Φ1 to the functor Φ2 . 15.3. Exact sequence of functors. In what follows, we will meet sequences of transformations of diﬀunctors Fi−2

Fi−1

F

Fi+1

i Φi+1 −→ . . . . . . −→ Φi−1 −→ Φi −→

(15.4)

Such a sequence is called exact (respectively, is said to be a complex ), if for any A-module P the following sequence of A-homomorphisms is exact (resp., is a complex): Fi−2 (P )

Fi−1 (P )

Fi+1 (P )

Fi (P )

. . . −→ Φi−1 (P ) −→ Φi (P ) −→ Φi+1 (P ) −→ . . . ;

(15.5)

recall that this means that for all indices i we have the condition Im Fi−1 (P ) = Ker Fi (P ) (resp., Im Fi−1 (P ) ⊂ Ker Fi (P )). Let Oi be an A-module representing the functor Φi , and let Hi−2

Hi−1

Hi+1

H

i Oi+1 ←− . . . , . . . ←− Oi−1 ←− Oi ←−

def

Hi = HFi ,

(15.6)

be the corresponding (15.4) sequence of representing homomorphisms. Proposition. The sequence of functors (15.4) is exact (resp., is a complex) if and only if such is the sequence of representing modules and homomorphisms (15.6). Identify the functors Φi and HomA (Oi , ·). In that case the sequence (15.5) acquires the form Fi−2,P

Fi−1,P

Fi,P

. . . −→ HomA (Oi−1 , P ) −→ HomA (Oi , P ) −→ Fi,P

Fi+1,P

−→ HomA (Oi+1 , P ) −→ . . . ,

(15.7)

where Fi,P (ϕ) = ϕ ◦ Hi , ϕ ∈ HomA (Oi , P ). First let us show that the exactness of (15.4) implies the exactness of (15.6). Let P = Oi−1 and ϕ = idOi−1 . Then, Fi−1,P (ϕ) = Hi−1 . But Fi,P ◦ Fi−1,P = 0, so that Hi−1 ◦ Hi = Fi,P (Hi−1 ) = 0, therefore,

Functors of the diﬀerential calculus and their representations

247

Im Hi ⊂ Ker Hi−1 . Further, let P = coker Hi and ϕ : Oi → P be the natural projection. Then Fi,P (ϕ) = ϕ ◦ Hi = 0, i.e., ϕ ∈ Ker Fi,P , and, in view of the exactness of (15.7), we have ϕ = Fi−1,P (ψ) = ψ ◦ Hi−1 for some ψ ∈ HomA (Oi−1 , P ). But Im Hi = Ker ϕ = Ker(ψ ◦ Hi−1 ) ⊃ Ker Hi−1 , i.e., Ker Hi−1 ⊂ Im Hi . Together with the previous inclusion, this shows that Ker Hi−1 = Im Hi and so, the sequence (15.6) is exact at the term Oi . How let us show that the exactness of (15.6) implies the exactness of (15.4). For that it suﬃces to verify the exactness of the sequence (15.7) for any module P . If ϕ ∈ Ker Fi,P , then ϕ ◦ Hi = 0, which is equivalent to Im Hi ⊂ Ker ϕ. But Im Hi = Ker Hi−1 in view of the exactness of the sequence (15.6), and so we obtain Ker Hi−1 ⊂ Ker ϕ. Finally, the last inclusion allows to represent ϕ in the form ϕ = ψ ◦ Hi−1 for some element ψ ∈ HomA (Hi−1 , P ). But since ψ ◦ Hi−1 = Fi−1,P (ψ), we see that Ker Fi,P ⊂ Im Fi−1,P . Conversely, if ϕ = Fi−1,P (ψ) = ψ◦Hi−1 , then Fi,P (ϕ) = ψ◦Hi−1 ◦Hi = 0, so that Ker Fi,P ⊃ Im Fi−1,P . Therefore, Ker Fi,P = Im Fi−1,P . Finally, note that the proof of the fact that the sequences (15.4) and (15.6) are actually complexes is part of the proof of their simultaneous exactness given above. If a sequence of functors F

F

0 1 Φ1 −→ Φ2 0 → Φ0 −→

(15.8)

is exact, then it is natural to call the functor Φ0 the kernel of the transformation of functors F1 . Similarly, one deﬁnes the image of a transformation of functors. One of the useful and simple applications of Proposition 15.3 is the proof of the existence of the representing object for the kernel functor. Corollary. Let the functor Φ0 be the kernel of the natural transformations of functors F : Φ1 → Φ2 and let Oi be the representing object for the functor def Φi , i = 1, 2. Then O0 = coker HF is the representing object for the functor Φ0 . If H0 : O1 → O0 = coker HF is the natural projection, then the sequence H

H

0 F O1 ←− O2 0 ← O0 ←−

(15.9)

is exact and Proposition 15.3 shows that the functor HomA (O0 , ·) is the kernel of the natural transformation of functors HomA (O1 , ·) → HomA (O2 , ·), generated by the homomorphism HF , which can be identiﬁed with the kernel of the transformation F .

248

Chapter 15

15.4. Composition of diﬀunctors and their representations. The description of representing objects for the composition of diﬀunctors is based on the following general algebraic fact, which will be constantly used in this chapter. For commutative algebras, it can be stated as follows: Let A and B be commutative algebras over a common ﬁeld k and let P ∈ Mod A, R ∈ Mod B. Let also Q be an (A, B)-bimodule, i.e., a k-vector space possessing the structure of a left A-module and of a right B-module, and let these structures be compatible, (aq)b = a(qb), a ∈ A, b ∈ B, q ∈ Q. Then P ⊗A Q and HomB (Q, R) are obviously also (A, B)-bimodules. In this situation, we have an isomorphism of (A, B)-bimodules ι : HomB (P ⊗A Q, R) ∼ = HomA (P, HomB (Q, R)),

(15.10)

speciﬁed by the relation (ι(ϕ))(p) (q) = ϕ(p⊗A q), p ∈ P, q ∈ Q, ϕ ∈ HomB (P ⊗A Q, R). (15.11) Below the isomorphism ι will be used in the situation when A = B, and we will speak of A-bimodules rather than (A, A)-bimodules. Forgetting the right A-module structure of Q, we obtain its left A-module, denoted Q< , similarly, its right A-module, denoted Q> . Depending on the situation, the indices “>” and “ , R) denotes the k-vector space of homomorphisms of the A-module Q> to the A-module R, whose A-module structure is generated by the left multiplication in the bimodule Q. The symbols Hom< A (R, Q> ), < Q> ⊗ R, etc., are understood similarly. 15.5. Diﬀerentially closed categories. An example of how one may prove the existence of representing objects for diﬀunctors over any commutative algebra is the construction of the representing object for the diﬀunctor D, i.e., for the module of diﬀerential 1–forms presented in Section 14.4. Actually, this is a very general construction. However, it is insuﬃciently constructive to be used for concrete calculations, which is nicely illustrated by the example of algebraic diﬀerential 1-forms for the algebra C ∞ (M ) (see Section 14.10). The reason for this is that the category of all C ∞ (M )-modules is too wide, and we should limit ourselves only to “observable” modules, i.e., geometrical modules. The necessity of restricting the category of modules in one way or another also arises in many other situations, so that we must face the problem of the actual existence of objects that represent diﬀunctors in the given category of modules over the given algebra. The following deﬁnition allows us to adequately formalize the setting of that problem. 15.6. Deﬁnition. A category of A-modules M is called diﬀerentially closed if

Functors of the diﬀerential calculus and their representations

249

(a) the result of applying any diﬀunctor to an object of the category M is again an object of this category; (b) the natural transformation of diﬀunctors applied to objects of the category M yield morphisms in this category; (c) objects representing diﬀunctors in the category M are objects of this same category. Now the question that concerns us can be stated as follows: Is the category of A-modules under consideration diﬀerentially closed or not? A simple but conceptually important fact is that the notion of diﬀerential closedness is in natural agreement with the classical mechanism of observability. Formally this means the following. Proposition. The category of geometrical modules over a geometrical algebra is diﬀerentially closed. The fact that the module O is the representing object for some covariant diﬀunctor Φ means that we have a natural isomorphism Φ(P ) ∼ = HomA (O, P ). If the module P is geometrical, then the canonical isomorphism HomA (O, P ) → HomA (Γ(O), P ) (see the proposition in Section 8.6) shows that Γ(O) represents the diﬀunctor Φ in the category GMod A. Thus, everything is reduced to the proof of the existence of representing objects for the diﬀunctors in the category of all A-modules. But such a proof presupposes the description of the whole algebra of diﬀunctors, and this, as we already pointed out at the beginning of this chapter, is beyond the scope of this book. For the diﬀunctors that will be discussed further in this chapter and their various combinations, the question of existence will be solved as we go along. Hence, from the formal point of view, the above proposition can be regarded as proved only for those diﬀunctors. Note that the general proof of existence is of simple algebraic character, in the spirit of how it was carried out for diﬀerential 1-forms (see Section 14.6). 15.7. Suﬃcient condition for diﬀerential closedness. For our aims, a condition suﬃcient for diﬀerential closedness is given in the next proposition. Proposition. A category of A-module M is diﬀerentially closed if (i) the algebra A belongs to M; (ii) the module representing the diﬀunctor Diﬀ < k in the category M is an object of the same category;

250

Chapter 15

(iii) M is closed with respect to the operations ⊗A and HomA ; (iv) submodules of any module belonging to the category M also belong to M. We will carry out the proof only for the diﬀunctors studied later in this chapter for the same reason as for the proof of the proposition from Section 15.6. The construction of representing objects for diﬀunctors is a step-by-step procedure, at each step of which the representing object of the subsequent, more complicated diﬀunctor, is constructed on the basis of the previous ones. This requires the application of the operations ⊗A and HomA and the consideration of certain submodules of modules that arise in the process. The fact that this can be done for the diﬀunctors studied below can be seen directly from their description. The starting point of the process is the construction of the module of jets as representing objects for the diﬀunctors Diﬀ < k , k ≥ 0. Here we must pay attention to the fact that Diﬀ < 0 is a functor equivalent to the identity, whose representing object is the algebra A (see Section 15.8). Thus, the algebra A itself must be an object of the category M. 15.8. The diﬀunctors Diﬀ < k and D(k) . Let P be an A-module and let k be some nonnegative integer. Then the correspondence Q → Diﬀ < k (P, Q), (P, Q) is deﬁned in Section 9.67, is a covariant functor, denoted where Diﬀ < k Diﬀ < k,P , from the category Mod A to itself. Here to the homomorphism of A-modules F : Q → R corresponds the homomorphism < < < Diﬀ < k,P (Q) = Diﬀ k (P, Q) Δ → F ◦ Δ ∈ Diﬀ k (P, R) = Diﬀ k,P (R).

This is one of the main diﬀunctors. Recall (see Section 9.57) that the diﬀunctor Diﬀ < k,A is denoted simply by < < Diﬀ < and we write Diﬀ (P ) instead of Diﬀ (A, P ). Its important chark k k acteristic is that it splits naturally. Namely, to each element p ∈ P we, associate the homomorphism of A-modules pA : A → P, pA (a) = ap. The correspondence p → pA establishes an isomorphism of A-modules P → HomA (A, P ) = Diﬀ < 0 (P )

(15.12)

or, in other words, the equivalence of the functor Diﬀ < 0 = HomA (A, ·) to the identity. Besides, the homomorphism1 < < D< k,P : Diﬀ k (P ) → Diﬀ 0 (P ),

Δ → Δ(1A ),

(15.13)

that determines the splitting < Diﬀ k (P ) = Ker(D< k,P ) ⊕ Diﬀ 0 (P )

1

The cyrillic letter D looks like Δ and reads as D.

(15.14)

Functors of the diﬀerential calculus and their representations

251

is deﬁned. The correspondence P → Ker(D< k,P ) is a new diﬀunctor, which will be denoted by D(k) . Thus, D(k) (P ) = {Δ ∈ Diﬀ < k (P ) | Δ(1A ) = 0}.

(15.15)

In this connection, let us note that D(1) = D. In this new notation, and having in mind the isomorphism (15.12), we can write the splitting (15.14) as Diﬀ < k (P ) = D(k) (P ) ⊕ P.

(15.16)

The above may also be summarized as an exact split sequence of diﬀunctors D

k (Q) → D(k) ,

Δ → Δ − Δ(1A ),

is a diﬀerential operator of order ≤ k − 1. An important property of the diﬀunctors Diﬀ < k and D(k) , as well as of all the others, is their compatibility with the mechanism of observability. Proposition. If the module Q is geometrical, then the module Diﬀ < k (P, Q) is also geometrical, i.e., the functor Diﬀ < k,P takes objects of GMod A to objects of GMod A. The proof is a consequence of the obvious inclusion < Ker h · Diﬀ < k (P, Q) ⊂ Diﬀ k (P, Ker h · Q),

h ∈ |A|,

which immediately implies that < Inv(Diﬀ < k (P, Q)) ⊂ Diﬀ k (P, Inv ·Q) = 0.

If l ≤ k, then, since we obtain a natural which may be called transformation of the diﬀunctors an embedding. The left and right multiplication of diﬀerential operators by some element a ∈ A induce the following natural transformations: ⊂ Diﬀ < k (P, Q), < Diﬀ < l,P → Diﬀ k,P ,

Diﬀ < l (P, Q)

< la , ra : Diﬀ < k,P → Diﬀ k,P :

la (Δ) = aQ ◦ Δ, ra (Δ) = Δ ◦ aP , Δ ∈ Diﬀ < l (P, Q). We will call them left and right product of the diﬀunctor Diﬀ < k,P by the element a ∈ A, respectively.

252

Chapter 15

15.9. The diﬀunctor Diﬀ > k . Now we consider the diﬀunctor > Diﬀ > k,Q : P → Diﬀ k (P, Q),

which is the right analog of the diﬀunctor Diﬀ < k,P . In this situation, to the homomorphism F : P → R corresponds the homomorphism > > > Diﬀ > k,Q (R) = Diﬀ k (R, Q) Δ → Δ ◦ F ∈ Diﬀ k (P, Q) = Diﬀ k,P (P ),

which means that Diﬀ > k,P is a contravariant functor. Further, note that the operators of change of structure < > id< k : Diﬀ k (P, Q) → Diﬀ k (P, Q), > < id> k : Diﬀ k (P, Q) → Diﬀ k (P, Q),

which are identity maps of the set Diﬀ k (P, Q), are diﬀerential operators of order ≤ k. Indeed, δa (id< k ) = (aQ − aP )|Diﬀ k (P,Q) or, equivalently, < δa (id< k )(Δ) = δa (Δ) ∈ Diﬀ k (P, Q),

Δ ∈ Diﬀ < k (P, Q),

and therefore [(δa1 ◦ δa0 ◦ · · · ◦ δak )(id< k )](Δ) = (δa1 ◦ δa0 ◦ · · · ◦ δak )(Δ) = 0. Similarly, one can show that id> k is an operator of order ≤ k. < < Further, consider the homomorphism D< k = Dk,P : Diﬀ k P → P (see (15.14)). Then the operator def

< > < D> k = Dk ◦ idk : Diﬀ k P → P ,

which coincides with D< k as a map of sets, being the composition of operators of order 0 and k, is an operator of order k. The importance of this operator is in its co-universality, a property that we will now clarify. Let Δ ∈ Diﬀ k (P, Q). The map hΔ : P → Diﬀ > k Q,

p → Δ ◦ p,

where

def

(Δ ◦ p)(a) = Δ(ap),

is obviously a homomorphism of A-modules, which makes the following diagram commutative P

hΔ

Diﬀ > k Q Δ

(15.19)

D> k

Q. > Δ Indeed, (D> k ◦ h )(p) = Dk (Δ ◦ p) = (Δ ◦ p)(1A ) = Δ(p).

Proposition. The map > Diﬀ > k (P, Q) → HomA (P, Diﬀ k Q),

is an isomorphism of A-modules.

Δ → hΔ ,

(15.20)

Functors of the diﬀerential calculus and their representations

253

The map > HomA (P, Diﬀ > k Q) → Diﬀ k (P, Q),

h → D> k ◦ h,

is obviously inverse to the map (15.20). (Check that it is a homomorphism of A-modules.) The map (15.20) may be regarded as a map of left A-modules: > < Diﬀ < k (P, Q) → HomA (P, Diﬀ k Q).

(15.21)

Obviously, it is also an isomorphism. The isomprphism (15.20) means precisely that the module Diﬀ > k Q co> represents the diﬀunctor Diﬀ > , while the operator D is co-universal. k,Q k Here the preﬁx “co” stresses the duality with respect to the space of jets (see Section 15.9), which represents the diﬀunctor Diﬀ < k,P . 15.10. Prolongation of co-representing homomorphisms. The composition operation of diﬀerential operators acquires its functorial expression in iterations of the diﬀunctor Diﬀ > k . In their turn, the relations between them are described by prolongations of co-representing homomorphisms. The necessity of the operation of prolongation stems from the following commutative diagram: D> l

Diﬀ > l P

(15.22)

P

hΔ (l)

Δ

hΔ

> Diﬀ > l (Diﬀ k Q)

D> l

Diﬀ > k Q

D> k

Q,

Δ def (hΔ ◦D> l ) . The lower left triangle where Δ ∈ Diﬀ > k (P, Q) and h(l) = h of this diagram is the diagram (15.19) for the operator hΔ ◦ D> l . The Δ is called the l-th prolongation homomorphism h . homomorphism hΔ (l) def

Similarly, the operator Δ(l) = Δ ◦ D> l is called the l-th prolongation of the operator Δ. Since Δ(l) is an operator of order ≤ k + l, we have the next commutative diagram, whose upper triangle is the diagram (15.19) for the operator Δ(l) : Diﬀ > l P D> l

P

Δ(l)

h

Diﬀ > k+l Q

Δ(l) Δ

D> k+l

Q.

(15.23)

254

Chapter 15

The diagram > Diﬀ > l (Diﬀ k Q)

> Cl,k

Diﬀ > k+l Q

(15.24)

D> k+l

D> l D> k

Diﬀ > k Q

Q

is a particular case of the previous one for the operator Δ = D> k . Namely, > > Cl,k = Cl,k (Q) = h∇ , where

> > > ∇ = D> k ◦ Dl : Diﬀ l (Diﬀ k Q) → Q.

This diagram can be understood as the deﬁnition of the homomorphism > Cl,k , which is called the gluing homomorphism and is the required functorial interpretation of the composition operation of diﬀerential operators. Finally, combining the diagrams (15.22)–(15.24), it is not diﬃcult to construct the following commutative diagram. Diﬀ > l P Δ(l)

h

Diﬀ > k+l Q

D> l Δ(l) D> k+l

(15.25)

P Δ

Q

According to the deﬁnition, hΔ(l) (∇) = Δ ◦ ∇. It is obvious from the = hΔ(m) . description that hΔ(l) |Diﬀ > mP 15.11 Exercise. According to the deﬁnition, the module Diﬀ > P (see Section 9.68) is ﬁltered by the submodules Diﬀ > l P . Let Δ ∈ Diﬀ k (P, Q). Recall that a homomorphism of a ﬁltered module is deﬁned as a homomorphism that shifts the ﬁltration by a certain constant. Such, in particular, is the A-homomorphism > > hΔ = hΔ(l) . hΔ ∗ : Diﬀ P → Diﬀ Q, ∗ |Diﬀ > l P Describe the structure of the ﬁltered Diﬀ(A)-module in the A-module Diﬀ R and show that hΔ ∗ is a homomorphism of Diﬀ(A)-modules. We now pass to modules of jets and related constructions, which are dual to those discussed in the last three sections. 15.12. Modules of jets. The object representing the diﬀunctor Diﬀ < k,P in a category of A-modules M is called the module of jets of k-th order or k-jets in the category M. This object, denoted J k (P ; M), is the range of the diﬀerential operator jkM : P → J k (P ; M). universal in M. This means that for any module Q ∈ Ob M, each operator Δ ∈ Diﬀ k (P, Q) can be uniquely represented in the form Δ = hΔ ◦jkM ,

Functors of the diﬀerential calculus and their representations

255

where k h Δ = hM Δ : J (P ; M) → Q

is a homomorphism of A-modules. Recall that representing objects are unique up to equivalence if, of course, they exist. Let us ﬁrst construct the module of k-jets in the category of all A-modules Mod A. To this end, we consider the A-bimodule A ⊗k P whose left (right) multiplication is deﬁned as (a, a ⊗k p) → aa ⊗k p (a ⊗k p, a) → a ⊗k ap, respectively. To an element a ∈ A assign the (A, A)-homomorphism δ a : A ⊗k P → A ⊗k P,

δ a (a ⊗k p) = a ⊗k ap − aa ⊗k p.

Further, let μk+1 be a submodule of the bimodule A ⊗k P generated by all elements of the form (δ a0 ◦ δ a1 ◦ · · · ◦ δ ak )(a ⊗k p). Put def

J k (P ) = jk : P → J Ck,l (Δ) (a) = [Δ(a)](1) = jlM (1) ⊗ jkM (a) .

Therefore, according to Proposition 15.2, the homomorphism representing the gluing transformation Ck,l will be the A-homomorphism Jl (M) ⊗A J Ck,l (Q) (Δ) = 0 if and only if Im Δ ⊂ D(l) (Q). The set of these operators will be denoted as follows: def > Df (l,k) (Q) = {Δ ∈ Diﬀ > k (Diﬀ l (Q)) | Im Δ ⊂ D(l) (Q)}. Thus, > Ker Ck,l (Q) = Df (l,k) (Q).

(17.5)

The space Df (l,k) (Q) possesses a natural A-bimodule structure, in which the left and right multiplication are deﬁned by (left)

Δ → aD(l) (Q) ◦ Δ,

(right) Δ → Δ ◦ aA .

The left and right A-modules of the A-bimodule Df (l,k) (Q) are denoted by > Df < (l,k) (Q) and Df (l,k) (Q), respectively. Exercise. Show that the identical transformation of change of module structure > Df < (l,k) (Q) → Df (l,k) (Q)

and

< Df > (l,k) (Q) → Df (l,k) (Q)

296

Chapter 17

is a diﬀerential operator of order ≤ k (see Exercise in 9.67). Thus, we have constructed the new diﬀunctors < Df < (l,k) : Q → Df (l,k) (Q)

and

> Df > (l,k) : Q → Df (l,k) (Q).

(Check this!) In what follows, it will be convenient to describe these diﬀunctors by the following formulas: < > Df < (l,k) = Diﬀ k (D(l) ⊂ Diﬀ l ) (17.6) > > Df > (l,k) = Diﬀ k (D(l) ⊂ Diﬀ l ). Accordingly, their values on the module Q are expressed by the formulas < > Df < (l,k) (Q) = Diﬀ k (D(l) (Q) ⊂ Diﬀ l (Q)) (17.7) > > Df > (l,k) (Q) = Diﬀ k (D(l) (Q) ⊂ Diﬀ l (Q)). Such descriptions are useful in that they explicitly indicate various connections between diﬀerent diﬀunctors. For example, we see directly that > < < Diﬀ < k (D(l) (Q) ⊂ Diﬀ l (Q)) ⊂ Diﬀ k (D(l) (Q)), i.e., that Df (l,k) (Q) is a submodule of the module Diﬀ < k (D(l) (Q)) and we observe a natural embedding < → Diﬀ of diﬀunctors Df < k ◦D(l) . Similarly, the same description shows (l,k) < > that Df (l,k) (Q) is not a submodule of the A-moduleDiﬀ < k (Diﬀ l (Q)), while the natural embedding < > Df < (l,k) (Q) → Diﬀ k (Diﬀ l (Q)) is a diﬀerential operator. (Find the order of this operator!) The above can be presented in the form of exact sequences of A-homomorphisms • > 0 → Df < (l,k) (Q) → Diﬀ k (Diﬀ l (Q))

0→

Df > (l,k) (Q)

→

> Diﬀ > k (Diﬀ l (Q))

< C(k,l) (Q)

−→

> C(k,l) (Q)

−→

Diﬀ < k+l (Q),

(17.8)

Diﬀ > k+l (Q),

Df > (l,k) (Q)

> where = Ker C(k,l) (Q). Since the sequences (17.8) are exact for any A-moduleQ, we obtain exact sequences of diﬀunctors < C(k,l)

• > < 0 → Df < (l,k) → Diﬀ k (Diﬀ l ) −→ Diﬀ k+l , > C(k,l)

(17.9)

> > > 0 → Df > (l,k) → Diﬀ k (Diﬀ l ) −→ Diﬀ k+l .

17.5. The diﬀunctors D(l,k) and related chain complexes. The diﬀunctor D(l,k) can be deﬁned by the formula def

Q → D(l,k) (Q) = D(k) (D(l) (Q) ⊂ Diﬀ > l (Q)),

(17.10)

which means that here we are considering diﬀerential operators Δ : A → Diﬀ > l (Q) of order ≤ k such that Δ(1A ) = 0 and Im Δ ⊂ D(l) (Q).

Spencer Complexes and Diﬀerential Forms

297

The module D(l,k) (Q), unlike Df (l,k) (Q), is supplied with only one A-module structure; in it, the multiplication by the element a ∈ A of the operator Δ : A → Diﬀ > l (Q) is deﬁned by the rule (a, Δ) → aD(l) (Q) ◦ Δ. Here aD(l) (Q) denotes the operator of multiplication by the element a in D(l) (Q). Since D(l) (Q) is a submodule of the module Diﬀ < l (Q), it follows of multithat aD(l) (Q) is the restriction to D(l) (Q) of the operator aDiﬀ < l (Q) plication by a ∈ A of elements of the module Diﬀ < (Q). Therefore, D (l,k) (Q) l < is a submodule of the moduleDf (l,k) (Q) and thereby we have deﬁned the embedding diﬀunctor D(l,k) → Df < (l,k) . Further, let us denote by D(k) (Diﬀ > l (Q)) the A-module that coincides (Q)) as a vector space over k and in which multiplication with D(k) (Diﬀ > l by an element a ∈ A is deﬁned by the rule (a, Δ) → aDiﬀ < ◦ Δ. l (Q) Then, D(k) (Diﬀ > l (Q)) can be regarded as a submodule of the module ◦ Diﬀ •k (Diﬀ > l (Q)). Denote by C(k,l) (Q) the restriction of the gluing homo< morphism C(k,l) (Q) to D(k) (Diﬀ > l (Q)) and note that, using the same considerations as in the proof of (17.5), we can derive the equality ◦ Ker C(k,l) (Q) = D(k,l) (Q).

(17.11)

< < ◦ (Q) maps D(k) (Diﬀ > Note that C(k,l) l (Q)) to D(k+l) (Q) ⊂ Diﬀ k+l (Q). For this reason, the composition

D(k) (Diﬀ > l (Q))

◦ C(k,l) (Q)

−→

D< k+l

Diﬀ > k+l (Q) −→ Q

is trivial (is the map to zero). Now, taking all the above into account, we obtain a sequence of homomorphisms 0 → D(l,k) (Q) → D(k) (Diﬀ > l (Q))

◦ C(k,l) (Q)

−→

D< k+l

Diﬀ < k+l (Q) −→ Q → 0, (17.12)

in which the composition of any two successive homomorphisms is trivial. In other words, the sequence (17.12) is a chain complex. Moreover, the nontrivial homology of this complex can occur only at the term Diﬀ < k+l (Q). This is an obvious consequence of the fact that the homomorphism D< k+l ◦ is surjective, while D(l,k) (Q) is the kernel of the homomorphism C(k,l) (Q). Note that the complex (17.12) is the result of applying the sequence ◦ C(k,l)

D< k+l

< 0 → D(l,k) → D(k) ◦ Diﬀ > l −→ Diﬀ k+l −→ idMod A → 0

(17.13)

of natural transformations of diﬀunctors to the module Q, where idMod A is the identity functor on the category Mod A. Denote this sequence by Sp(k,l) . Then the complex (17.12) should be denoted by Sp(l,k) (Q). This notation indicates that we are dealing with the functor Q → Sp(l,k) (Q) from the category Mod A to category of chain complexes of A-homomorphisms.

298

Chapter 17

This is an example of a type of diﬀunctor that diﬀers from those studied previously. One other example of a diﬀunctor of this type is the sequence Spen(l,k) : < C(k,l)

< > < 0 → Df < (l,k) → Diﬀ (k) (Diﬀ l ) −→ Diﬀ k+l → 0.

(17.14)

From this point of view, the chain complex (17.8) is the value of this diﬀunctor on the module Q. For this reason, it should denoted by Spen(l,k) (Q). Such “complex”-valued diﬀunctors, as well as other diﬀunctors, may be related by natural transformations. An example of such a transformation is the transformation Sp(l,k) → Spen(l,k) described by the following diagram: D(l,k)

D(k) (Diﬀ > l )

◦ Ck,l

Diﬀ < k+l

D< 2

(17.15)

idMod A

=

Df < (l,k)

Diﬀ •k (Diﬀ > l )

< Ck,l

Diﬀ < k+l

0,

in which the two vertical transformation on the left are the embeddings of diﬀunctors described above. Exercise. 1. Show that the homology of the complex (17.12) is trivial if < and only if the homomorphism C(k,l) (Q) is surjective. ∞ 2. Let A = C (K) (see Sections 9.45 and 9.45), Q = A and k = l = 1. Compute the homology of the complex (17.12) and ascertain that it is nontrivial. 17.6. Modules, representing the diﬀunctors D(l,k) and Df < (l,k) . Now we can describe the modules representing the diﬀunctors D(l,k) and Df < (l,k) in a diﬀerentially closed category M. To do this, it is ﬁrst necessary to describe the representing module of the diﬀunctor D(k) (Diﬀ > l ). It can be calculated by using the isomorphism (15.10) and Proposition 15.20 in the following way: (k)

> < D(k) (Diﬀ > l (P )) = HomA (ΛM , Diﬀ l (P )) (k)

> l = Hom< A (ΛM , HomA (J< (M), P )) (k)

(k)

l l = Hom< A (< J> (M) ⊗A ΛM , P ) = HomA (J< (ΛM ; M), P ), (k)

where the reduced jets ΛM are deﬁned in Section 15.21. Thus, the module (k) J l ) in the category M. Further, it follows directly from the proposition in Section 15.2 and formula (15.41), that the homomorphism (k,l)

(k)

C◦,M : J A a d a − a j1 (1A ) ⊗A d a = j1M (a dM a) − a j1M (dM a).

(17.24)

Having in mind (17.23), as well as the fact that the module Im S2,0 is generated by elements of the form j1M (dM b), b ∈ A, we see that d∗M (dM a ⊗A dM a) + d∗M (dM a ⊗A dM a ) = j1M (dM aa ) − a j1M (dM a) − aj1M (dM a ) ∈ Im S2,0 .

(17.25)

Therefore, (S1,1 ◦ d∗M )(dM a ⊗A dM a + dM a ⊗A dM a ) = 0. On the other hand, the representing homomorphism of the composition of diﬀunctors d

w1,1 : D2 → D< (Diﬀ < 1 ) −→ D(D)

Spencer Complexes and Diﬀerential Forms

303

is (w1,1 )∗M = S1,1 ◦ d∗M . Since w1,1 is an embedding of diﬀunctors, the homomorphism (w1,1 )∗M : Λ1M ⊗A Λ1M → Λ2M is surjective and elements of the form dM a ⊗A dM a+dM a⊗A dM a generate its kernel. In other words, this kernel consists of the symmetric part of the tensor product Λ1M ⊗A Λ1M , so that Λ2M = Λ1M ⊗A Λ1M /Λ1M A Λ1M = Λ1M ∧A Λ1M . But the last equality is the standard deﬁnition of the exterior square of the module Λ1M . Now, since the exterior square coincides with (w1,1 )∗M : Λ1M ⊗A Λ1M → Λ2M = Λ1M ∧A Λ1M ,

(17.26)

we come to the conceptual deﬁnition of the exterior product of 1-forms. Deﬁnition. The exterior product of ﬁrst degree diﬀerential forms is deﬁned as the homomorphism (w1,1 )∗M , representing the transformation of diﬀunctors w1,1 : D2 → D(D). In other words, ω1 ∧ ω2 = (w1,1 )∗M (ω1 ⊗A ω2 ), def

ω1 , ω2 ∈ Λ1M .

(17.27)

Remark. At this point, the reader may want to ask what all these complications are needed when the standard descriptive deﬁnition is so simple? The answer is that the conceptual deﬁnition automatically carries over to various nonstandard situations, whereas it is diﬃcult, if not impossible, to guess the correct generalization of the descriptive deﬁnition. Situations of this type arise, for example, in the diﬀerential calculus over graded commutative algebras, which will be studied in the concluding chapter of this book. 17.12. The exterior diﬀerential. We shall deﬁne the exterior diﬀerential acting on 1-forms by putting (see 17.18) M M 1 2 dM 1 = d(1,1) = S1,1 ◦ j1 : ΛM → ΛM . def

Having in mind (17.23), note that M j1M (a dM a) = j1M (a ) ⊗> A d a M = d∗M (dM a ⊗A dM a) + a j1M (1A ) ⊗> A d a)

= d∗M (dM a ⊗A dM a) + a j1M (dM a) = d∗M (dM a ⊗A dM a)

mod (Im S2,0 ).

Therefore, since Im S2,0 = Ker S1,1 , it follows that M M M ∗ M M dM 1 (a d a) = S2,0 (j1 (a d a)) = S2,0 (dM (d a ⊗A d a))

= (w1,1 )∗M (dM a ⊗A dM a) = dM a ∧A dM a.

304

Chapter 17

Thus, we have established the analog of the well-known property of the exterior diﬀerential: M M M dM 1 (a d a) = d a ∧A d a.

In particular, this implies that Exercise.

dM 1

◦d

M

(17.28)

= 0.

1. Show that the derivation ð2 : a → dM a ∧ dM , where (dM a ∧ dM )(b) = dM a ∧ dM b, def

is the element of the module D2 (Λ2M ) = HomA (Λ2M , Λ2M ), which corresponds to the identity homomorphism idΛ2M . Accordingly, the biderivation (a, b) → dM a ∧ dM b is universal in the sense that for any Δ ∈ D2 (P ), P ∈ Ob M, there exists a unique homomorphism of A-modules h : Λ2M → P such that Δ(a) (b) = h(dM a ∧ dM b) . 2. Describe the module of diﬀerential 2-form on the “cross”, i.e., when A = C ∞ (K) and M = GMod C ∞ (K). 17.13. The Spencer second-order jet complex. The results of the two previous sections can be reduced to the following commutative diagram. 0

Λ2M

S1,1

= j0M

Λ2M

J Df < (σ,k) = Diﬀ k (Dσ ⊂ Df σ ), def

D(σ,k) = D(k) (Dσ ⊂ Df > σ ).

def

> > Df > (σ,k) = Diﬀ k (Dσ ⊂ Df σ )

(17.31)

The meaning of these formulas is exactly the same as that of (17.10), (17.6). It is easy to verify that the pair of diﬀunctors deﬁned in this way satisﬁes the conditions formulated above, which were needed to continue the inductive procedure. (Check this!)

306

Chapter 17

The construction (17.31) may be generalized by deﬁning the diﬀunctors denoted by the symbol Φτ (Dσ ⊂ Df > σ ), where Φτ is one of the diﬀunctors > or Df , while τ = (τ , . . . , τ Dτ , Df < 1 l ). The meaning of this symbol becomes τ τ clear by induction on the length l of the sequence τ , assuming that σ is ﬁxed. For l = 1, this symbol coincides with Φ(σ,k) provided τ = (k). The inductive step is described by following formula. def

> > > Φ(k,τ ) (Dσ ⊂ Df > σ ) = Φ(k) (Dτ (Dσ ⊂ Df σ ) ⊂ Df τ (Dσ ⊂ Df σ )) .

The existence of the diﬀunctors Φτ (Dσ ⊂ Df > σ ), in particular, allows to deﬁne the embedding of diﬀunctors > < D(σ,k) = Dσ (D(k) ⊂ Diﬀ > (17.32) k ) → Dσ (Diﬀ k ) , which, according to the scheme 15.28, leads to the important natural diﬀerential operator that we shall discuss in the next section. 17.16. The modules representing Dσ and Df σ . The representing objects for the diﬀunctors Dσ and Df σ in any diﬀerentially closed category M exist and are denoted by ΛσM and LσM , respectively. This is proved in essentially the same way (see above) as for the diﬀunctors D(l,k) and Df (l,k) . However, the proof requires some new technical details that lie outside the scope of this book. Therefore, the existence of these objects will be proved later in a simpler (in this respect) particular case, when the sequence σ consists only of ones. In the framework of this approach, we shall obtain classical diﬀerential forms on the manifold M together with the exterior diﬀerential and all the rest, choosing, for the main category M, the category GMod C ∞ (M ). On the other hand, the deﬁnition of exterior diﬀerential itself and its higher analogs is not complicated at all, provided we assume that the existence of the representing modules ΛσM has been established. Indeed, the embedding of diﬀunctors (17.32) yields the following generalization of the diagram (17.18): (σ,k)

ΛM

πσ,k

dM σ,k

J k ).

(17.36)

Recall that Df k+1 (P ) is an A-bimodule and its left and right A-module > structures are denoted by Df < k+1 (P ) and Df k+1 (P ), respectively. More pre> cisely, if the operator Δ : A → Df k (P ) belongs to Df k+1 (P ), then the result ◦ Δ or, of its multiplication from the left by a ∈ A is the operator aDf < k (P )

308

Chapter 17

which is the same, the operator aDk (P ) ◦ Δ. The result of its multiplication from the right by a ∈ A is the operator Δ ◦ aA . Let Q be an A-bimodule, while Q< and Q> are the corresponding left and right A-modules. The operators I < : Q< → Q> and I > : Q> → Q< , which are, as maps of sets, identity maps, are called change of structure operators. Recall also that if p ∈ P , then the symbol pA denotes the homomorphism A → P , deﬁned by the rule pA (a) = ap. Lemma. For any k ≥ 0: (1) The embedding Dk+1 (P ) → Df < k+1 (P ) is a homomorphism of A-modules. (2) The change of structure operators in the A-bimodule Df k+1 (P ) are diﬀerential operators of ﬁrst order. (3) The projection Π>

Df > k+1 (P ) Δ −→ Δ − Δ(1) ∈ Dk+1 (P ) is a homomorphism of A-modules. (4) If Δ ∈ Df k+1 (P ), then the decomposition Δ = Δ(1) + (Δ − Δ(1)) establishes an isomorphism of A-modules Dk (P ) ⊕ Dk+1 (P ) ∼ = Df < k+1 (P ). These statements are easily proved by induction. For k = 0, they are obvious, since D1 = D and Df 1 = Diﬀ 1 . Further, assuming that they hold for k ≥ 0, let us show that they are valid for k + 1 as well. The statement (1) obviously follows from the deﬁnition of left multiplication. > Let I < : Df < k+1 (P ) → Df k+1 (P ) be the operator changing the left structure in the A-module to its right structure. Now if we assume that Δ : A → Df > k (P ) is a k-linear map, then a straightforward calculation shows that δa1 ,...,al (I < ) (Δ) = (−1l )δa1 ,...,al (Δ). In particular, if Δ ∈ Df k+1 (P ), then Δ is a diﬀerential operator of ﬁrst order or less and, therefore, δa1 ,a2 (I < ) (Δ) = δa1 ,a2 (Δ) = 0. Since Δ is arbitrary, we have δa1 ,a2 (I < ) = 0, i.e., I < is a ﬁrst order differential operator. This same argument can also be carried over, word for word, to case of the operator I > . Finally, since a> (Δ) = Δ ◦ a (recall that the operator of right > multiplication by a in Df > k+1 (P ) is denoted by a ), it follows that Π> (a> Δ) = Δ ◦ a − Δ(a).

Spencer Complexes and Diﬀerential Forms

309

Now if Δ ∈ Df k+1 (P ), then we also have Δ ∈ D(Dk (P )) and, therefore, Δ ◦ a − Δ(a) = aΔ. This proves statement (3), while statement (4) is obvious.

Exercise. Let (Δ0 , Δ) ∈ Dk (P ) ⊕ Dk+1 (P ). Let us put def a ∗ (Δ0 , Δ) = aΔ0 + Δ(a), aΔ , a ∈ A. Show that the operation of multiplication (deﬁned above) by elements of the algebra A supplies Dk (P )⊕Dk+1 (P ) with a second A-module structure and the module obtained in this way is isomorphic to the module Df > k+1 (P ). 17.18. Polyderivations. As we have already seen in Section 17.7, elements of the module D2 (P ) can be naturally interpreted as P -valued skew-symmetric biderivations, i.e., as bivectors in diﬀerential-geometrical terminology. Below, we will show that the elements of the module Dk (P ), k > 2, may also be interpreted similarly. Arguing by induction, which we can begin from the previously considered case k = 2, we assume that to any element ∇ ∈ Dl (P ), l ≤ k we can assign the skew-symmetric l-form (a1 , . . . , al ) → a1 , . . . , al ∇ ∈ P, a1 , . . . , al ∈ A, which is k-linear derivation in each argument. To do this, let Δ ∈ Dk+1 (P ). Since Δ : A → Df > k P is a derivation, we have Δ(ab) = aDf > · Δ(b) + bDf > · Δ(a) = Δ(b) ◦ a + Δ(a) ◦ b. k (P ) k (P )

(17.37)

On the other hand, Δ(ab) ∈ Dk (P ), since ImΔ ∈ Dk (P ). Therefore, (1) = 0. In view Δ(ab) : A → Df > k−1 (P ) is a derivation and so Δ(ab) of (17.37), this is equivalent to the equality Δ(b) (a) + Δ(a) (b) = 0. This condition can be conveniently rewritten in the form (17.38) [a, b]Δ + [b, a]Δ = 0, def where [a, b]Δ = Δ(a) (b). Note that (a, b) → [a, b]Δ ,

a, b ∈ A,

is a k-bilinear skew-symmetric form with values in Dk−1 (P ) and, moreover, it is a biderivation. Finally, let us put def

a1 , a2 , a3 , . . . , ak+1 Δ = [. . . [[Δ(a1 )](a2 )] . . . ](ak+1 ).

(17.39)

The P -valued form (17.39) is obviously k-linear and is a k-linear derivation in each argument. On the other hand, we have

a1 , a2 , a3 , . . . , ak+1 Δ = a2 , . . . , ak+1 Δ(a1 )

a3 , . . . , ak+1 (Δ(a1 ))(a2 ) = a3 , . . . , ak+1 [a1 ,a2 ]Δ .

(17.40)

310

Chapter 17

By the induction hypothesis, the second expression in this chain of equalities is skew-symmetric in the whole family of arguments, except the ﬁrst. Besides, in view of (17.37), the last expression is skew-symmetric in the ﬁrst two arguments. From these two facts, it obviously follows that the forms

·, . . . , ·Δ are skew-symmetric. The skew-symmetric k-linear k-forms that are at the same time derivations in each argument, are usually called P -valued k -derivations of the algebra A or polyderivations, if it is not necessary to specify the multiplicity k. All such k-derivations obviously constitute an A-module that will be denoted by Dk (P ). Thus, the formula (17.39) may be regarded as the deﬁnition of the homomorphism (17.41) iD,D : Dk (P ) → Dk (P ), Δ → ·, . . . , ·Δ . Moreover, to each P -valued k-derivation we can assign an element of the module Dk (P ) simply by reversing the previous arguments (see the LemmaExercise from Section 17.7). This shows that iD,D is an isomorphism. 17.19. A bit of philosophy. Possibly, at this point the reader may want to ask why do we need these complicated manipulations with the diﬀunctors Dk , while the equivalent notion of k-derivation is quite elementary. The general answer to this question is that this notion, as well as any other descriptive notion, does not explicitly contain any information on connections with other natural constructions of the diﬀerential calculus. In other words, working only with descriptions, it is diﬃcult, if not impossible, to discover the structure of relationships of the corresponding notions, i.e., to understand what may be called the algebra of their calculus. For example, it is hardly possible to guess the existence of higher analogs of diﬀerential forms and of the de Rham complexes starting from the standard deﬁnitions. What is more, the descriptive deﬁnitions often turn out to be “incorrect” in the sense that they cannot be generalized to various singular, inﬁnite-dimensional, and other nonstandard situations. Such is, for example, is the well-known deﬁnition of tensors. Concerning the concrete diﬀunctors Dk , it will be shown below that they lead directly to the construction of Spencer complexes and, as a consequence, to diﬀerential forms and de Rham complexes over arbitrary commutative algebras. 17.20. Polyvectors. A polyderivation over the algebra C ∞ (M ) with values in geometrical modules has a geometrical interpretation as a polyvector, or polyvector ﬁeld on a manifold M . Recall that according to the standard deﬁnition, by a k-vector on M we mean a C ∞ (M )-linear skew-symmetric k-form whose arguments are diﬀerential forms of ﬁrst order on M . Here we assume that the values of such a form are sections of a vector bundle π or, as a rule, simply functions. This interpretation is merely the expression of the fact that to each k-derivation of the type mentioned above we can assign a k-vector and

Spencer Complexes and Diﬀerential Forms

311

vice versa. The corresponding construction for k = 2 was already carried out in Section 17.9. Now we shall study this situation in a wider context. Let α(ω1 , . . . , ωk ) ∈ P,

ω1 , . . . , ωk ∈ Λ1M (A),

(17.42)

be an A-linear skew-symmetric polylinear form with values in P ∈ Ob M. Then χα (a1 , . . . , ak ) = α(dM a1 , . . . , dM ak ), def

a1 , . . . , ak ∈ A,

is a P -valued k-derivation of the algebra A. The family of all k-forms (17.42) obviously constitutes an A-module that will be denoted as Pk,M (P ). Then the map χ : Pk,M (P ) → Dk (P ),

α → χα ,

is a homomorphism of A-modules. Since the module Λ1M (A) is generated by the diﬀerentials dM a, a ∈ A, the form χα is trivial if and only if the form α is trivial. Therefore, the homomorphism χ is injective, but can it be inverted? To answer this question, consider a k-derivation β ∈ Dk (P ). If the form α = χ−1 (β) exists, then we necessarily have α(dM a1 , . . . , dM ak ) = β(a1 , . . . , ak ). Here, in order to deﬁne the value of α(ω1 , . . . , ωk ), we must represent each 1-form ωi as ωi =

mi

bij dM aij ,

aij , bij ∈ A,

(17.43)

j=1

and put def

α(ω1 , . . . , ωk ) =

b1j1 . . . bkjk β(a1j1 , . . . , akjk ).

(17.44)

j1 ,...,jk

The fact that this is well deﬁned must be veriﬁed, i.e., we must show that the right-hand side of the deﬁnition (17.44) does not depend on how the form ωi is expressed in (17.43). This is equivalent to saying that, for each i ∈ {1, . . . , k}, this right-hand side does not depend on the representation of the form ωi under the condition that for all other forms ωj such representations are ﬁxed. In view of the skew-symmetry of summands of (17.44), we can assume that i = 1. Thus, regarding that the representations (17.43) for the forms ωj , j > 1, as ﬁxed, let us consider the derivation X b2j2 . . . bkjk β(a, a2j2 , . . . , akjk ). a → j2 ,...,jk

312

Chapter 17

In view of universality of the module Λ1M (A), there exists a unique homomorphism hX : Λ1M (A) → P such that X = hX ◦ dM . Here hX (ω) = bj X(aj ), j

M where ω = j bj d aj . This shows that hX (ω) does not depend on the representation of the form ω as linear combination of diﬀerentials. To conclude the proof, it remains to note that hX (ω1 ) exactly coincides with the right-hand side of the equality (17.44). Thus, we have established natural isomorphisms

Dk (P ) ∼ = Dk (P ) ∼ = Pk (P ) .

(17.45)

This gives us three diﬀerent views of the same notion. Which of them is more convenient depends on the context. For example, the natural symplectic structure on the cotangent space T ∗ M of a manifold M may also be understood as the Poisson bracket on it, as well as a Poisson bivector. As we showed in Chapter 10, its analog for an arbitrary commutative algebra can be naturally obtained in the form of a biderivation, i.e., of the bracket {·, ·} on the algebra of symbols S∗ (A), deﬁned in Section 10.2. On the other hand, it is more convenient to work with the corresponding Poisson bivector by using of Schouten brackets [13]. 17.21. The algebra of polyderivations. Similarly to its symmetric analogs (see Section 16.7), A-valued polyderivations, as well as polyvectors, can be multiplied. The product of a k-vector α by an l-vector β is the (k + l)-vector, denoted by α ∧ β, and deﬁned by the following formula: def α(ωi1 , . . . , ωik ) · β(ωj1 , . . . , ωjl ) , (17.46) [α ∧ β](ω1 , . . . , ωk+l ) = |I|=k

where the sequences I = {i1 , . . . , ik } and J = {j1 , . . . , jl } are given in increasing order and I ∪ J = {1, 2, . . . , k + l}. Thus the direct sum Pk (A) P∗ (A) = k≥0

acquires an algebra structure. Moreover, formula (17.46) is also meaningful when β ∈ Dl (P ). In this case, α ∧ β ∈ Pk+l (P ). Exercise. Show that 1. β ∧ α = (−1)kl α ∧ β, if α ∈ Pk (A), β ∈ Pl (A); 2. P∗ (A) is an associative algebra, while P∗ (P ) is a P∗ (A)-module with respect to the multiplication (17.46).

Spencer Complexes and Diﬀerential Forms

313

The natural isomorphisms (17.45) allow us to carry over the (graded) algebra structure of P∗ (A) to Dk (A) D∗ (A) = (respectively, to D∗ (A) =

(respectively, D∗ (P ) = D∗ (A)-module).

k≥0

Dk (A)), as well as to regard D∗ (P ) = Dk (P ) k≥0

k≥0

k≥0

Dk (P )) as a D∗ (A)-module (respectively, a

17.22. The diﬀunctors Dk,l . The interpretation of elements of the module Dk (P ) as k-derivations allows us to deﬁne the diﬀunctors def

Dk,l = Dk (Dl ⊂ Df > l ). By deﬁnition, elements of the module Dk,l (P ) are k-derivations β of the algebra A with values in the module Df > l (P ) such that we have β(a1 , . . . , ak ) ∈ Dl (P ). Proposition. The diﬀunctors Dk,l and Dk+l are equivalent. We must indicate a natural isomorphism between the modules Dk,l (P ) and Dk+l (P ). To do this, we will show that any element β ∈ Dk,l is uniquely deﬁned by a (k + l)-derivation and vice versa. By deﬁnition, β(a1 , . . . , ak ) ∈ Dl (P ). Consider the form def

∇β (a1 , . . . ak+l ) = ak+1 , . . . , ak+l β(a1 ,...,ak )

(17.47)

(see deﬁnition (17.39) for Δ = β(a1 , . . . , ak )). This form is skew-symmetric in the ﬁrst k and last l arguments and is a derivation in each of them. Besides, β(a1 , . . . , ak−1 , ak ak+1 ) = β(a1 , . . . , ak−1 , ak ) ◦ ak+1 + β(a1 , . . . , ak−1 , ak+1 ) ◦ ak , (17.48) since, in particular, β is a derivation in the last argument. On the other hand, β(a1 , . . . , ak−1 , ak ak+1 ) ∈ Dl (P ) = D< (Dl−1 (P ) ⊂ Df > l−1 (P )) Therefore, we have β(a1 , . . . , ak−1 , ak ak+1 )(1A ) = 0 and, applying the operators from (17.48) to 1A , we obtain β(a1 , . . . , ak−1 , ak )(ak+1 ) + β(a1 , . . . , ak−1 , ak+1 )(ak ) = 0.

(17.49)

314

Chapter 17

Finally, from (17.40) and (17.49), it follows that

ak+1 , ak+2 , . . . , ak+l β(a1 ,...,ak−1 ,ak ) + ak , ak+2 , . . . , ak+l β(a1 ,...,ak−1 ,ak+1 ) = ak+2 , . . . , ak+l β(a1 ,...,ak−1 ,ak )(ak+1 ) + ak+2 , . . . , ak+l β(a1 ,...,ak−1 ,ak+1 )(ak ) = 0 or, equivalently, ∇β (a1 , . . . , ak−1 , ak , ak+1 , ak+2 , . . . , ak+l ) + ∇β (a1 , . . . , ak−1 , ak+1 , ak , ak+2 , . . . , ak+l ) = 0. (17.50) Thus, the form ∇β is skew-symmetric in the k-th and (k + 1)-st arguments. This, together with the properties of the skew-symmetric form ∇β , proves its skew-symmetry in all the arguments. Thus, we have constructed a map Dk,l (P ) → Dk+l (P ),

β → ∇β ,

which is obviously a homomorphism of A-modules. Conversely, the equality (17.47) may be regarded as the deﬁnition of the form β, assuming that the form ∇β is given. Then, inverting the order of the steps of the proof of the formula, starting from (17.48) and ending at (17.50), it is easy to see that the form β deﬁned in this way is indeed an element of the moduleDk,l (P ). 17.23 Exercise. In Section 17.15, we inductively deﬁned the diﬀunctors > Φτ (Dσ ⊂ Df > σ ). Show that the diﬀunctor Dτ (Dσ ⊂ Df σ ) deﬁned in this way is equivalent to Dk,l if the sequences τ and σ consist of k and l ones, respectively. 17.24. Existence of diﬀerential forms. Now everything is ready to prove the existence of representing objects ΛkM for the diﬀunctors Dk in a diﬀerentially closed category of A-modules M, i.e., to prove the existence of modules of diﬀerential forms. To do this, it is natural to use induction on k, having in mind that for k = 1, 2 these modules are already constructed (see Sections 17.10 and 17.6). Therefore, it remains to show how to conunder the assumption that the modules Λm struct the module Λk+1 M are M already constructed for m ≤ k, k ≥ 2. Do do this, by the corollary from Section 15.3, it suﬃces to represent the diﬀunctor Dk+1 as the kernel of some transformation of diﬀunctors for which the existence of representing modules has been established. The ﬁrst thing that we shall need is the following transformation of diﬀunctors, which we will denote by tk,l : > > > Dk (Diﬀ > l ) = Dk−1 (D(Diﬀ l ) ⊂ Diﬀ 1 (Diﬀ l )) tk,l

> −→ Dk−1 (Diﬀ > 1 (Diﬀ l )).

(17.51)

Spencer Complexes and Diﬀerential Forms

315

Here the expression Dm (Φ(Diﬀ > n (P ))) stands for the family of m-derivation (P )) supplied with the A-module structure induced with values in Φ(Diﬀ > n by left multiplication in the A-bimodule Diﬀ n (P ). k Further, let us consider the Spencer transformation Sk+l deﬁned by the following commutative diagram: k Sk+l

Dk (Diﬀ > l ) tk,l

Dk−1 (Diﬀ > l+1 )

(17.52)

< > Dk−1 (C1,l ) > > Dk−1 (Diﬀ 1 (Diﬀ l ))

We stress that the A-module structure in the lower term of the diagram is induced by the left multiplication in Diﬀ l . Proposition. The following equality holds: k Ker Sk+l = Dk−1 (D(l,1) ⊂ Df > (l,1) ), k in particular, Ker Sk+1 = Dk+1 . > > First note that Ker C1,l = Diﬀ > 1 (D(l) ⊂ Diﬀ l ). Therefore, > > > Ker Dk−1 (C1,l ) = Dk−1 (Diﬀ > 1 (D(l) ⊂ Diﬀ l )) = Dk−1 (Df (l,1) )

(see (17.10)). But since tk,l is an embedding, while Df > (l,1) is a submodule > > of the module Diﬀ 1 (Diﬀ l ), it follows from (17.51) that k > Ker Sk+l =(Im tk,l ) ∩ (Ker Dk−1 (C1,l )) > > > =Dk−1 (D(Diﬀ l ) ∩ Df (l,1) ) ⊂ Df > (l,1) = Dk−1 (D(l,1) ⊂ Df (l,1) ) .

This proves the ﬁrst assertion, while the second one follows from the relation > D(1,1) = D2 , Df > (1,1) = Df 2

and Dk−1 (D2 ⊂ Df > 2 (P )) = Dk+1

(see the Proposition from Section 17.22). The second assertion of the proposition just proved above may be described by the exact sequence of diﬀunctors k Sk+1

> 0 → Dk+1 → Dk (Diﬀ > 1 ) −→ Dk−1 (Diﬀ 2 ) .

(17.53)

The homomorphism k+l l k Sk,M : J l+1 (Λk−1 M ; M) → J (ΛM ; M),

(17.54)

which represents the Spencer transformation (17.52) is called the Spencer operator (in the category M). In these terms, the corollary from Section 15.3 applied to the exact sequence (17.53), proves the required fact. k+1 Corollary. The A-module Λk+1 M = coker Sk,M is the representing object of the diﬀunctor Dk+1 in the category M.

316

Chapter 17

Accordingly, the exact sequence of homomorphisms k+1 Sk,M

k−1 1 k 2 0 ← Λk+1 M ← J (ΛM ; M) ←− J (ΛM ; M)

(17.55)

is the representation of the exact sequence of diﬀunctors (17.53). 17.25. The representing module for the diﬀunctor Dk . A technically simpler, but conceptually less satisfactory way of deﬁning diﬀerential forms is to use the equivalence of the diﬀunctors Dk and Dk . Indeed, as will be shown below, the representing module for Dk may be constructed as the range of the universal k-derivation, simply by generalizing the construction of the diﬀerential 1-form from Section 14.4 to the case k > 1. In view of the equivalence iD,D of these diﬀunctors (see (17.41)), the representing module for Dk is isomorphic to the representing module for Dk and so can serve as the deﬁnition of the module of diﬀerential k-forms. We begin by recalling that, according to its deﬁnition, the tensor proddef uct A⊗k m = A ⊗k . . . ⊗k A (m times) is the universal object for k-linear m-forms on the algebra A understood as a k-vector space. This means that any such form ϕ with values in a k-vector space V can be represented in a unique way as follows ϕ(a1 , . . . , αm ) = hϕ (a1 ⊗k . . . ⊗k am ) , where hϕ : A⊗k m → V is a k-linear map. In what follows, we regard A⊗k m as a k-vector space and assign to it the A-module A ⊗k A⊗k m , in which multiplication by an element a ∈ A is deﬁned by the rule (a, a ⊗k a1 ⊗k . . . ⊗k am ) → aa ⊗k a1 ⊗k . . . ⊗k am . Let P ∈ Ob M and Fm (P ) be the vector space of k-linear m-forms on the module P . If ϕ ∈ Fm (P ), then aϕ ∈ Fm (P ), where def

(aϕ)(a1 , . . . , αm ) = aϕ(a1 , . . . , αm ) . This operation of multiplication by elements of the algebra A transforms Fm (P ) into an A-module. Here, as can be easily seen, the m-form Φm : (a1 , . . . , am ) → 1A ⊗k a1 ⊗k . . . ⊗k am ∈ A ⊗k A⊗k m is universal in the sense that for any form ϕ ∈ Fm (P ) there exists a unique A-homomorphism hϕ : A ⊗k A⊗k m → P such that ϕ(a1 , . . . , am ) = hϕ (1A ⊗k a1 ⊗k . . . ⊗k am ) . (17.56) Further, in the A-module A ⊗k A⊗k m , we consider the submodule R generated by all possible elements of the two following types: 1R . 1A ⊗k a1 ⊗k . . . ⊗k am + (−1)σ 1A ⊗k aσ(1) ⊗k . . . ⊗k aσ(m) , where σ is a permutation, while (−1)σ is its sign. 2R . 1A ⊗k a1 ⊗k . . . ⊗k ai ai ⊗k . . . ⊗k am −ai ⊗k a1 ⊗k . . . ⊗k ai ⊗k . . . ⊗k am −ai ⊗k a1 ⊗k . . . ⊗k ai ⊗k . . . ⊗k am , 1 ≤ i ≤ m.

Spencer Complexes and Diﬀerential Forms

317

def

Proposition. The quotient module1 Lm (A) = A ⊗k A⊗k m /R is the representing module for the diﬀunctor Dk in the category ModA. First, for convenience, let us introduce the following notation: def

Lm (A) a∂a1 |∂a2 | . . . |∂am = a ⊗k a1 ⊗k a2 ⊗k . . . ⊗k am

mod R,

assuming that ∂a1 | . . . |∂am = 1A ·∂a1 | . . . |∂am . From the fact that elements of the form 1R and 2R belong to the submodule R, it follows that (a1 , . . . , am ) → ∂a1 | . . . |∂am

(17.57)

is an m-derivation of the algebra A with values in the module Lm (A). Moreover, any m-derivation (a1 , . . . , am ) → ϕ(a1 , . . . , am ) ∈ P with values in an A-module P is, in particular, a k-linear m-form on the algebra A. Further, since the form ϕ is skew-symmetric and is a derivation in each argument, it follows that the homomorphism hϕ (see (17.56)) annihilates elements of the form 1R and 2R and, therefore, annihilates the whole submodule R. Hence, hϕ can be represented as the composition pr

hϕ

A ⊗k A⊗k m −→ Lm (A) −→ P , where pr is the quotient map, while hϕ is an A-homomorphism such that hϕ (a · ∂a1 | . . . |∂am ) = aϕ(a1 , . . . , am ).

(17.58)

As to the case of an arbitrary diﬀerentially closed category of A-modules M, the existence of an A-module representing in M the diﬀunctor Dk is obtained by an appropriate factorization of the module Lm (A) (see the Proposition above). This module will be denoted by Lm M (A). In this situation, the homomorphism m m iM D : LM (A) → ΛM (A) ,

which represents the equivalence of diﬀunctors iD,D : Dk → Dk in the category M, is called the canonical isomorphism of modules Lm M (A) and Λm M (A). 17.26. Universal polyderivation. The formula (17.58) shows that the polyderivation (17.57) is universal. Let us consider the polyderivation (a1 , . . . , am ) → ∂ M a1 | . . . |∂ M am , where ∂ M a1 | . . . |∂ M am = prM (∂a1 | . . . |∂am ), def

1

The Cyrillic letter L, which resembles the Greek Λ, reads as the Latin L.

(17.59)

318

Chapter 17

while prM : Lm (A) → Lm M (A) is the natural projection expressing the universality of the polyderivation (17.57); then this polyderivation will be universal in the category M. The range of the universal polyderivation ∗ (17.59) may be carried over to Λm M (A) by means of the isomorphism iD,D . This can be done directly as follows. Having in mind the isomorphism m m Dm (Λm M (A)) = HomA (ΛM (A), ΛM (A)) ,

let us consider the element Δm,M ∈ Dm (Λm M (A)), which under this isomorphism corresponds to the identity map of the module Λm M (A). The polyderivation corresponding to it in the notation of Section 17.18 is given by (a1 , . . . , am ) → a1 , . . . , am Δm,M . Its universality follows from the fact that, under the isomorphism i∗D,D , it corresponds to the identity homomorphism of the representing module LkM (A), which in its turn corresponds to the universal polyderivation (17.59) (see (17.58)). In other words, we have the equality M M

a1 , . . . , am Δm,M = iM D (∂ a1 | . . . |∂ am ) .

(17.60)

17.27. The exterior product. Note that by (15.10), the A-module LkM (A) ⊗A LlM (A) represents the diﬀunctor Dk ◦ Dl in any diﬀerentially closed category of A-modules M. The equivalence iD,D assigns to the natural embedding of diﬀunctors Dk+l = Dk,l = Dk (Dl ⊂ Df > l ) −→ Dk ◦ Dl the embedding νk,l : Dk+l → Dk ◦ Dl . Namely, if ϕ ∈ Dk+l , then ψ = νk,l (ϕ) is deﬁned by the rule ψ(a1 , . . . , ak ) : (ak+1 , . . . ak+l ) → ϕ(a1 , . . . , ak , ak+1 , . . . ak+l ) .

(17.61)

According to Section 15.2, the transformation νk,l is represented in the category M by the corresponding homomorphism of representing modules λk,l : LkM (A) ⊗A LlM (A) → Lk+l M (A). More precisely, λk,l : (∂a1 | . . . |∂ak )⊗A (∂ak+1 | . . . |∂ak+l ) → ∂a1 | . . . |∂ak |∂ak+1 | . . . |∂ak+l . (17.62) Indeed, in view of the the isomorphism k+l k Dk (Dl (Lk+l M (A)) = HomA (LM (A), Dl (LM (A)))

the transformation λk,l assigns to the element ∂a1 | . . . |∂ak ∈ LkM (A) the l-derivation LlM (A) ∂ak+1 | . . . |∂ak+l → ∂a1 | . . . |∂ak |∂ak+1 | . . . |∂ak+l ∈ Lk+l M (A).

Spencer Complexes and Diﬀerential Forms

319

Then, if ϕ ∈ Dk+l (P ) and ϕ(a1 , . . . , ak , ak+1 , . . . ak+l ) = F (∂a1 | . . . |∂ak |∂ak+1 | . . . |∂ak+l ), where F ∈ HomA (Lk+l M (A), P ) = Dk+l (P ), then the homomorphism F ◦ λk,l , due to the existence of the isomorphism Dk (Dl (P )) = HomA (LkM (A), Dl (P )) understood as an element of the module HomA (LkM (A), Dl (P )), obviously coincides with νk,l (ϕ) (see 17.61 above). Finally, using the canonical isomorphism n n n = k, l , iM D : LM (A) → ΛM (A), let us carry over the homomorphism λk,l to diﬀerential forms: λk,l : ΛkM (A) ⊗A ΛlM (A) → Λk+l M (A). 17.28. Deﬁnition. The exterior product of diﬀerential forms ω ∈ ΛkM (A) and ρ ∈ ΛlM (A) is deﬁned as the element def

ω ∧ ρ = λk,l (ω ⊗A ρ) ∈ Λk+l M (A) . Let us introduce the useful intermediate notation M M k daM 1 | . . . |dam = iD (∂a1 | . . . |∂am ) ∈ ΛM (A).

Then, from the deﬁnition of the representing homomorphism from Section (15.2), it follows that M M M (daM 1 | . . . |dak ) ∧ (dak+1 | . . . |dak+l ) = M M M daM 1 | . . . |dak |dak+1 | . . . |dak+l .

(17.63)

The right-hand side of this formula does not depend on the choice of the position of the symbol ∧ between the symbols dai in its left-hand side. This obviously implies the associativity of the exterior product, since the module M M Λm M (A) is generated by elements of the form d a1 | . . . |d am . Note that the symbol dM a1 | . . . |dM am inherits the properties of the symbol ∂a1 | . . . |∂am . In particular, it is skew-symmetric w.r.t. the arguments of dM ai . Therefore, (17.63) implies that (dM a1 | . . . |dM ak ) ∧ (dM ak+1 | . . . |dM ak+l ) = (−1)kl (dM ak+1 | . . . |dM ak+l ) ∧ (dM a1 | . . . |dM ak ) . (17.64) For the same reason as above, this implies the skew-symmetry of the exterior product: λl,k = (−1)kl λk,l . 17.29. The algebra of diﬀerential forms. The exterior product operation transforms the direct sum def Λ∗M = Λ∗M (A) = ΛkM (A) k≥0

320

Chapter 17

into a graded associative algebra. From the usual point of view, multiplication in this algebra is skew-commutative. However, in the “graded sense” this algebra is commutative. The exact meaning of this notion will be given in Chapter 21. As to now, we shall only indicate the remarkable fact that the diﬀerential calculus may also be developed over graded commutative algebras, for example, over the algebra of diﬀerential forms itself. In particular, we can construct the algebra of diﬀerential forms over the algebra of diﬀerential forms Λ∗M (A) and, continuing this process, arrive at a new mathematical object, namely the algebra of iterated diﬀerential forms. Notation. Further, for the two most important categories, we shall use the following simpliﬁed notation: def

Λi (A) = ΛiModA ,

def

ΛiΓ (A) = ΛiGModA

and similarly for i = ∗. 17.30. The isomorphism λM m . Recall that the exterior degree of an A-moduleP , denoted by P ∧m , is the quotient module of its tensor degree P ⊗m by the submodule Rm generated by all possible elements of the form pσ(1) ⊗A . . . ⊗A pσ(m) − (−1)σ p1 ⊗A . . . ⊗A pm , pi ∈ P , where σ is a permutation of the indices 1, . . . , m. Obviously, P ∧m is the universal object for skew-symmetric A-polylinear forms on P . The image of the element p1 ⊗A . . . ⊗A pm under the factorization homomorphism P ⊗m → P ∧m is denoted by p1 ∧ . . . ∧ pm and is called the exterior product of the elements p1 , . . . , pm . Proposition. The map

1 ∧m λM → Λm m : ΛM (A) M (A) ,

M M M M given by λM m (d a1 ∧ . . . ∧ d am ) = d a1 | . . . |d am , is an isomorphism.

Let us prove this statement by induction, ﬁrst noting that for m = 1 it is tautological. Now let us assume that it is valid for some m > 1 and let us consider the (m + 1)-form on Λ1M (A) M 1 (ω0 , ω1 , . . . , ωm ) → ∧M 1,m ω0 ⊗A (λm (ω1 ∧ . . . ∧ ωm )) , ωi ∈ ΛM (A) , with values Λm+1 M (A). Using the inductive hypothesis and the deﬁnition of exterior product ∧M 1,m , we see that (dM a0 , dM a1 , . . . , dM am ) M M M M M M 3 → λM 1,m d a0 ⊗A (d a1 | . . . |d am ) = d a0 |d a1 | . . . |d am . Moreover, this form is skew-symmetric and A-linear in each argument. Therefore, in view of the previously mentioned universality of the exterior ∧(m+1) degree Λ1M (A) , there exists a homomorphism 1 ∧(m+1) λM → Λm+1 m+1 : ΛM (A) M (A),

Spencer Complexes and Diﬀerential Forms

321

taking dM a0 ∧ dM a1 ∧ . . . ∧ dM am to dM a0 |dM a1 | . . . |dM am . On the other hand, (a0 , a1 , . . . , am ) → dM a0 ∧ dM a1 ∧ . . . ∧ dM am is a (m + 1)-derivation of the algebra A with values in the A-module 1 ∧(m+1) ΛM (A) . Therefore, having in mind the universality of the (m + 1)-derivation (a0 , a1 , . . . , am ) → dM a0 |dM a1 | . . . |dM am , we see that there is a homomorphism 1 ∧(m+1) m+1 νM : Λm+1 , M (A) → ΛM (A) taking dM a0 |dM a1 | . . . |dM am to dM a0 ∧ dM a1 ∧ . . . ∧ dM am . This shows m+1 are mutually inverse isomorphisms. that λM m+1 and νM It will be useful for the reader to compare the above proof with the one given in Section 17.10 for m = 2 and based on a direct calculation. 1 ∧m m , Since the isomorphism λM m identiﬁes the modules ΛM (A), ΛM (A) M M M (A), it follows that the temporary symbol d a |d a | . . . |d a and Λm 0 1 m M can be replaced by the standard one, i. e., dM a0 ∧ dM a1 ∧ . . . ∧ dM am . 17.31. The Kahler diﬀerentials. It should be especially noted that in algebraic geometry, the module of diﬀerential k-forms over a commutative ∧k , while Ω1 , as the algebra, usually denoted by Ωk , is deﬁned as Ω1 module of Kahler diﬀerentials. From our point of view, Ω1 = Λ1ModA . This approach looks quite natural from the historical perspective, however, it has the drawback that it misses certain important natural connections of the calculus of diﬀerential forms with other structures of the diﬀerential calculus and, above all, with the Spencer complexes, which will be discussed below. And this approach cannot be carried over directly to tensors of other types and leads to deﬁnite diﬃculties in passing to graded algebras. Under it, there remain “shadows” of the dependence of diﬀerential forms on the choice of the category of modules.

18 The (Co)Chain Complexes Coming from the Spencer Sequence

In this chapter, we continue our study of the Spencer sequence and the related diﬀunctors. We discover that the Spencer Diﬀ-sequence is a chain complex and that the Spencer jet complex is a cochain complex. Further, along the same lines, we obtain the de Rham cochain complex and continue our study of diﬀerential forms, moving in the direction of their conceptual deﬁnition. 18.1. The diﬀunctors of Spencer complexes. The exact sequence of diﬀunctors (17.53) is the initial segment of the Spencer Diﬀ-sequence of order k + 1, which is formed by the Spencer transformations r , 0 ≤ r ≤ k + 1, Sk+1

(see (17.52)). This whole sequence is of order m, it is denoted by Spenm , and has the following form: S m−1

S m−2

m m > < < 0 → Dm → Dm−1 (Diﬀ > 1 ) −→ Dm−2 (Diﬀ 2 ) −→ . . .

S l+1

Sl

S l−1

m m m > Dl (Diﬀ > . . . −→ m−l ) −→ Dl−1 (Diﬀ m−l+1 ) −→ . . .

S2

(18.1)

S1

m m < D1< (Diﬀ > . . . −→ m−1 ) −→ Diﬀ m → 0.

Theorem. The Spencer Diﬀ-sequence is a chain complex. For m = 0, we assume that D0 = id (the identity functor), so that the sequences Spen0 and Spen1 are of the form 0 → id → 0 and

0 → D = D1 → Diﬀ 1 → 0,

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 18

323

324

Chapter 18

respectively. Since there is nothing to prove in these two cases, we assume that m > 1. l+1 l ⊂ Ker Sm . To this end, it suﬃces to show We must show that Im Sm < ◦ l l−1 ◦ that Im Dl (C1,m−l−1 ) ⊂ Ker Sm , since Im Sm ⊂ Im Dl< (C1,m−l−1 ) (see (17.52)). To do this, we shall use the fact that Dl+1 = Dl−1 (D2 ⊂ Df > 2 ), so that > > > Dl+1 (Diﬀ > m−l−1 ) = Dl−1 (D2 (Diﬀ m−l−1 ) ⊂ Df 2 (Diﬀ m−l−1 )) .

(18.2)

Here > > > D2 (Diﬀ > m−l−1 )) = D(D(Diﬀ m−l−1 )) ⊂ Diﬀ 1 (Diﬀ m−l−1 ))), > > > > > Df > 2 (Diﬀ m−l−1 )) = Diﬀ 1 (D(Diﬀ m−l−1 )) ⊂ Diﬀ 1 (Diﬀ m−l−1 )))

(18.3)

On the other hand, by deﬁnition (see (17.52)) l+1 > Sm = Dl (C1,m−l−1 ) ◦ tl+1,m−l−1 .

Since tl+1,m−l−1 is an embedding, we can assume that the transformation > l+1 is induced by the action of the gluing transformation C1,m−l−1 on Sm the right-hand side of the equality (18.2). Taking into account the fact > transforms the diﬀunctor D(Diﬀ > that C1,m−l−1 m−l−1 ) into D(m−l) , and > > (Diﬀ ) into Diﬀ , we see that, by (18.3), it induces the Diﬀ > 1 m−l−1 (m−l) transformations > D2 (Diﬀ > m−l−1 )) → D(D(m−l) ⊂ Diﬀ m−l ) = D(m−1,1) , > > > > Df > 2 (Diﬀ m−l−1 )) → Diﬀ 1 (D(m−l) ⊂ Diﬀ m−l ) = Df (m−l,1) .

Finally, in view of (18.2), these transformations induce the transformation of functors > l Dl+1 (Diﬀ > m−l−1 ) → Dl−1 (D(m−1,1) ⊂ Df (m−l,1) ) = Ker Sm

(see the Proposition from Section 17.24). The projection Dm,P : Diﬀ < m P → P can be regarded as the value on the < module P of the natural transformation of diﬀunctors Dm : Diﬀ < m → Diﬀ 0 (see (15.13). Exercise. Show that the composition of the transformations of diﬀunctors 1 D m ◦ Sm vanishes. This fact allows us to obtain the extended Spencer Diﬀ-sequence by D

l+r )

l Sk+l+r

Dk (Diﬀ > l )

Dk−1 (Diﬀ > l+r+1 )

l Sk+l

(18.4)

Dk−1 (Diﬀ > l+1 )

In view of (18.4), the natural embeddings of diﬀunctors Dl (Diﬀ m−l ) → Dl (Diﬀ m−l+r ), 0 ≤ l ≤ m, deﬁne an embedding of the Spencer sequence Spenm into Spenm+r , denoted by spm,m+r , which is a homomorphism of chain complexes. The direct limit of the chain of embeddings sp0,1

Spen0 −→ · · ·

spm−1,m

−→

Spenm

spm,m+1

−→

Spenm+1

spm+1,m+2

−→

... ,

(18.5)

denoted by Spen∗ , is called the inﬁnite Spencer Diﬀ-sequence. It is a chaincomplex-valued diﬀunctor, ﬁltered by the chain-complex-valued diﬀunctors Spenm . 18.3. Spencer Diﬀ-complexes. Applying the sequence of diﬀunctors (18.1) to an A-module P , we obtain a ﬁnite or inﬁnite Spencer Diﬀcomplex, denoted by Spenm (P ) or Spen∗ (P ), respectively. The complex Spen∗ (P ) is ﬁltered by the subcomplexes Spenm (P ). The extended version of the complex Spenm (P ) ends with the fragment D< m,P

Diﬀ < m P −→ P → 0,

0 ≤ m ≤ ∞.

l l It should be noted that the diﬀerentials Sm = Sm (P ) of the chain complexes Spenm (P ), 0 ≤ m ≤ ∞, are A-homomorphisms. Therefore, the homology groups of these chain complexes are A-modules, and so one can speak of their supports in |A|. If P is a projective module over the algebra C ∞ (M ), then the zero-dimensional homology group of the complex Spenm (P ) is isomorphic to P , while the other groups are trivial. If P is not projective, then the homology groups of nonzero dimension, generally speaking, are nontrivial and their support consists of those points of the space |A| where projectivity fails. In particular, for manifolds with singularities, this support coincides with the set of singular points. Moreover, by means of Spencer Diﬀ-complexes, we can construct important homology invariants of diﬀerential operators. Namely, let Δ ∈ Diﬀ k (P, Q). The homomorphism > > hΔ ∗ : Diﬀ P → Diﬀ Q

(see Section 15.10) obviously induces a homomorphism of sequences spΔ : Spen∗ P → Spen∗ Q,

326

Chapter 18

which is a homomorphism of A-modules that shifts the ﬁltration by k, i.e., spΔ (Spenm (P )) ⊂ Spenm+k (Q), m ≥ 0.

(18.6)

Exercise. Show that spΔ is a morphism (chain map) of complexes. Further, denote the homology group of the complex Spen∗ (R) at the term Di (Diﬀ > R) by Hi (Spen∗ (R)). Being a chain map, spΔ induces a homomorphism of the homology group (18.7) Hi (spΔ ) : Hi (Spen∗ (P )) → Hi (Spen∗ (Q)) . Since spΔ is a homomorphism of A-modules, then so are the maps Hi (spΔ ). This is a homology invariant of the operator Δ. Moreover, one may consider the complexes Ker(spΔ ), Im(spΔ ), and coker(spΔ ). Their homology is also an invariant of the operator Δ. Homology groups of this kind play an important role in the contemporary geometrical theory of diﬀerential equations (see [3]). 18.4. Spencer δ-complexes. A standard construction of the homological algebra relates a certain spectral sequence to the ﬁltration of the complex Spen∗ (P ) by the subcomplexes Spenm (P ). Its detailed study lies outside of the scope of this book, and here we will only describe its initial term. By deﬁnition, it consists of the quotient complexes def

Smlm (Q) = Spenm (Q)/ Spenm−1 (Q), which are called Spencer dδ-complexes or δ-Diﬀ-complexes. The homology groups of these complexes are called the δ-homology groups of the module Q of order m. They are A-modules and constitute the terms of the above mentioned spectral sequence that follow the initial term. In view of (18.6), the homomorphism spΔ induces a homomorphism of quotient complexes sdm Δ : Smlm (P ) → Smlm+k (Q) and therefore, a homomorphism of δ-homology Hi (sdm Δ ) : Hi (Smlm (P )) → Hi (Smlm+k (Q)) . Obviously, the maps Hi (sdm Δ ) are homomorphisms of A-modules. As in Section 18.3, the groups Hi (Smlm (Q)) are invariants of the module Q, while the A-homomorphisms Hi (sdm Δ ) and the homology of the chain complexes m m Ker(sdm Δ ), Im(sdΔ ), and coker(sdΔ ) are invariants of the operator Δ. Exercise. Let A = C ∞ (M ), Q = Γ(π), and m > 0. Show that the homology groups Hi (Smlm (Q)), i > 0, are trivial and deduce from this that H0 (Spenm (Q)) = Q and Hi (Spenm (Q)) = 0, i > 0. 18.5. Spencer jet cochain complexes. Since the representing objects for the diﬀunctors Diﬀ < k and Dl in a diﬀerentially closed category exist, it follows that, according to Section 17.16, the representing objects for Dl (Diﬀ > k)

The (Co)Chain Complexes Coming from the Spencer Sequence

327

also exist. Therefore, the complex representing the extended diﬀunctor Spenm has the following form: m S1,M

iM

m S2,M

m J 1 (Diﬀ l )) in the k−1 l 1 category M is represented by the A-module J< (J< (ΛM ; M); M). Indeed, if P ∈ Ob M, then, in view of the isomorphism (15.10), and the results of

328

Chapter 18

Sections 15.20 and 17.3, we see that > > > k−1 Dk−1 (Diﬀ > 1 (Diﬀ l (P ))) = HomA (ΛM , Diﬀ 1 (Diﬀ l (P ))) l 1 = HomA (Λk−1 M , HomA (< J> (M) ⊗A J< (M), P ))

= HomA (< J>l (M) ⊗A < J>1 (M) ⊗A Λk−1 M ,P) It remains to notice that l < J> (M)

k−1 l 1 ⊗A < J>1 (M) ⊗A Λk−1 M = J< (J< (ΛM ; M); M).

> Further, in view of the fact that the transformation Dk−1 (C1,l ) is induced > by the transformation C1,l , the homomorphism k−1 l 1 J resenting C1,l . Therefore, from the results of Section 17.3, we see that M M M > M > jl+1 (ω) = jl+1 (1) ⊗> A ω → jl (1) ⊗A j1 (1) ⊗A ω M M M = jlM (1) ⊗> A j1 (ω) = jl (j1 (ω)),

(18.10)

where ω ∈ Λk−1 M (see 15.41). In its turn, the transformation tk,0 is induced by the embedding Dk → Dk−1 (Diﬀ > 1 ), represented by the homomorphism k k Sk,M : J 1 P ) = HomA (J< (ΛM ; M), P )

assigns the homomorphism M M j1M (a) ⊗> A (d a1 ∧ . . . ∧ d ak ) → ϕ(a, a1 , . . . , ak )

332

Chapter 18

or, which is the same, j1M (adM a1 ∧ . . . ∧ dM ak ) → ϕ(a, a1 , . . . , ak ).

(18.19)

Λk+1 M

In particular, let us now suppose that P = while ϕ = ϕid is the (k + 1)-derivation that corresponds to idΛk+1 under the canonical M

k+1 k+1 isomorphism Dk+1 (Λk+1 M ) = HomA (ΛM , ΛM ). Obviously,

ϕid (a, a1 , . . . , ak ) = dM a ∧ dM a1 ∧ . . . ∧ dM ak (see Section 17.30). It follows from the proposition in Section 15.2 that > k+1 k ∈ HomA (J 1 (ΛkM ; M), Λk+1 the homomorphism Sk,M M ) = Dk (Diﬀ 1 (ΛM )) is the image of the (k + 1)-derivation ϕid under the composition of homomorphisms tk+1,0

k+1 k+1 HomA (Λk+1 M , ΛM ) = Dk+1 (ΛM ) −→ tk+1,0

k+1 k+1 1 k −→ Dk (Diﬀ > 1 (ΛM )) = HomA (J (ΛM ; M), ΛM ),

(18.20)

where the equality sign stands for the canonical isomorphisms. Therefore, from (18.19), we see that k Sk,M : j1M (adM a1 ∧ . . . ∧ dM ak ) → dM a ∧ dM a1 ∧ . . . ∧ dM ak . k M But since dM k = Sk,M ◦ j1 , it follows that M M M M M dM k (ad a1 ∧ . . . ∧ d ak ) = d a ∧ d a1 ∧ . . . ∧ d ak .

(18.21)

dM k ,

This is the required description of the diﬀerential which is quite suﬃcient, since the diﬀerential forms adM a1 ∧. . .∧dM ak are additive generators of the module ΛkM . Further, as a rule, we omit the index k and write dM instead of dM k . Similarly, just as in the case of smooth manifolds, we call the form ω ∈ ΛkM closed if dM ω = 0, and exact if ω = dM ρ. Corollary. (a) Any exact form ω ∈ ΛkM is the sum of forms dM a1 ∧ . . . ∧ dM ak .

(b) dM (adM ω) = dM a ∧ dM ω. Assertion (a) is a direct consequence of formulas (18.21) and the fact that the forms a1 ∧ dM a2 ∧ . . . ∧ dM ak are additive generators of Λk−1 M . According to this same assertion, the form dM ω is the sum of the forms dM a0 ∧ dM a1 ∧ . . . ∧ dM ak , and assertion (b) again directly follows from formulas (18.21). Exercise. Is the form dalg ex −ex dalg ∈ Λ1alg considered in Section 14.10 closed? 18.10 Proposition. Let ω ∈ ΛkM and ρ ∈ ΛlM . Then dM (ω ∧ ρ) = dM ω ∧ ρ + (−1)k ω ∧ dM ρ.

(18.22)

The (Co)Chain Complexes Coming from the Spencer Sequence

333

Since the forms adM a1 ∧ . . . ∧ dM am = adM (a1 dM a2 ∧ . . . ∧ dM am ) M are additive generators of Λm M , we can assume that ω = ad ω1 and M M M ρ = a d ρ1 . In this case, we have ω ∧ ρ = aa d ω1 ∧ d ρ1 . Here the form dM ω1 ∧ dM ρ1 is exact (Corollary (a) from the previous section). Having in mind this corollary and formula (18.21), we see that

dM (ω ∧ ρ) = dM (aa ) ∧ dM ω1 ∧ dM ρ1 = a dM a ∧ dM ω1 ∧ dM ρ1 + adM a ∧ dM ω1 ∧ dM ρ1 = (dM a ∧ dM ω1 ) ∧ (a dM ρ1 ) + (−1)k (adM ω1 ) ∧ (dM a ∧ dM ρ1 ) = dM ω ∧ ρ + (−1)k ω ∧ dM ρ.

Formula (18.22) is an example of the graded Leibnitz rule, while the exterior diﬀerential dM is a graded derivation of degree 1 of the algebra of diﬀerential forms. 18.11. Substituting vector ﬁelds into diﬀerential forms. As we mentioned before, in standard diﬀerential geometry, a diﬀerential form of degree k is understood as a skew-symmetric C ∞ (M )-polylinear function ω(X1 , . . . , Xk ) of vector arguments Xi ∈ D(M ). In these terms, the important operation iX : Λk (M ) → Λk−1 (M ) of substituting vector ﬁelds into diﬀerential forms has a simple description: def

iX (ω)(X1 , . . . , Xk−1 ) = ω(X, X1 , . . . , Xk−1 ) . Here we assume that iX |Λ0 = 0. (Recall that Λ0 = C ∞ (M ).) Stated in that way, however, this operation is not convenient for generalizations to polyvectors and commutative algebras of general form. As we saw in Section 17.14, the conceptually correct approach consists of regarding iX as a homomorphism that represents a certain transformation of diﬀunctors iX : Dk−1 → Dk . We will describe this approach now, in the general situation of polyderivations. Let ∇ ∈ Dk (A) (see Section 17.18) and Δ ∈ Dl (P ). Using the notation of Section 9.58, and having in mind that n = k + l, let us put def i∇ (−1)I,J ∇(aI )Δ(aJ ), (18.23) l (Δ)(a1 , . . . , ak+l ) = I+J=I n , |I|=k

where (−1)I,J stands for the sign of the permutation of indices {1, . . . , k + l} → {i1 , . . . , ik , j1 , . . . , jl }; here I = {i1 , . . . , ik }, J = {j1 , . . . , jl }. It is obvious that i∇ l (Δ) ∈ Dk+l (P ), while i∇ l : Dl → Dk+l is a transformation of diﬀunctors. It is also easy k+l k to see that the homomorphism iM,l ∇ : ΛM → ΛM , which represents this transformation in some diﬀerentially closed category M ⊂ ModA, is given

334

Chapter 18

by the formula M M iM,l ∇ (d a1 ∧ . . . ∧ d ak+l ) =

(−1)I,J ∇(aI )dM (aJ ),

(18.24)

I+J=I k+l , |J|=l def

where dM (aJ ) = dM aj1 ∧ . . . ∧ dM ajl if J = (j1 , . . . , jl ). By deﬁnition, we can write iM,l ∇ = 0 if l < 0. Finally, bringing together the homomorphisms M,l {i∇ }, we obtain a “global” substitution operator or inner product M,l def iM i∇ : Λ∗M → Λ∗M . (18.25) ∇ = l≥0

Below, we will use the simpliﬁed notation i∇ = iModA and iΓ∇ = iGModA . ∇ ∇ In particular, if ∇ = X ∈ D(A), we use the standard notation iX . 18.12. Exercise. 1. Show that M m M k iM X (ω ∧ ρ) = iX (ω) ∧ ρ + (−1) ω ∧ iX (ρ), ω ∈ ΛM .

(18.26)

2. Let X1 , . . . , Xk ∈ D(A). Show that M M M iM Xk ◦ iXk−1 ◦ · · · ◦ iX1 = iX1 ∧...∧Xk .

(18.27)

In particular, iX ◦ iY = −iY ◦ iX . 18.13. Naturality of diﬀerential forms. Let ϕ : A → B be a homomorphism of commutative algebras, while M and K are ϕ-connected categories (see Section 15.24) of A- and B-modules, respectively. Then according to the approach of Section 15.25, there must exist a homomorphism of A-modules ϕ∗ = ϕ∗M,K : ΛkM → ΛkK , representing the diﬀunctor Dk in the categories M and K, respectively. As pointed out in that section, to this ϕ . This is easy to do if end it is necessary to describe the transformation TD k we interpret Dk as the polyderivation diﬀunctor, i.e., as Dk . Indeed, from ϕ simply assigns to any k-derivation of the algebra B this point of view, TD k (b1 , . . . , bk ) → ψ(b1 , . . . , bk ) ∈ Q, where Q ∈ Ob K, the k-derivation of the algebra A (a1 , . . . , ak ) → ψ(ϕ(a1 ), . . . , ϕ(ak )) ∈ Qϕ . Using the isomorphisms λkM and λkK (see Section 17.30), the required map ∧k ∧k → Λ1K which is the may be alternatively deﬁned as the map Λ1M k-th exterior product of the previously deﬁned (see the diagram (15.50)) map ϕ∗M,K : Λ1M → Λ1K . Under this approach, it becomes obvious that ϕ∗ = ϕ∗M,K : Λ∗M → Λ∗K is a homomorphism of exterior algebras, i.e., ϕ∗ (ω ∧ ρ) = ϕ∗ (ω) ∧ ϕ∗ (ρ), ω, ρ ∈ Λ∗M .

(18.28)

The (Co)Chain Complexes Coming from the Spencer Sequence

335

18.14 Exercise. Show that the two above deﬁnitions of the map ϕ∗ are equivalent. According to either one of these deﬁnitions, we have ϕ∗ (dM a1 ∧ . . . ∧ dM ak ) = dK (ϕ(a1 )) ∧ . . . ∧ dK (ϕ(ak )) . Therefore, in view of (18.21), we can write: M M ∗ M M M ϕ∗ (dM k (ad a1 ∧ . . . ∧ d ak )) = ϕ (d a ∧ d a1 ∧ . . . ∧ d ak )

= dK (ϕ(a)) ∧ dK (ϕ(a1 )) ∧ . . . ∧ dK (ϕ(ak )) K K = dK k (ϕ(a)d (ϕ(a1 )) ∧ . . . ∧ d (ϕ(ak ))) ∗ M M = dK k (ϕ (ad a1 ∧ . . . ∧ d ak )) , K ∗ i.e., ϕ∗ (dM k (ω)) = dk (ϕ (ω)) where ω is a diﬀerential form given by

ω = adM a1 ∧ . . . ∧ dM ak .

But such forms additively generate ΛkM , and so it follows that K ∗ ϕ∗ ◦ d M k = dk ◦ ϕ ,

(18.29)

i.e., we have proved the naturality of the exterior diﬀerential. 18.15. Example. It is remarkable that the operation of substituting vector ﬁelds into diﬀerential forms can be naturally carried over to relative vector ﬁelds. Namely, let X : A → B be such a ﬁeld with respect to the algebra homomorphism ϕ : A → B, i.e., a k–homomorphism satisfying the relative Leibnitz rule X(ab) = ϕ(a)X(b) + ϕ(b)X(a), and let M and K be ϕ-connected categories of A- and B-modules, respectively, (see Section 15.24). In a non-conceptual way, this can be done simply by putting iX (adK a1 ∧ . . . ∧ dK ak ) = ϕ∗ (a)

k

(−1)i−1 X(ai )dK ϕ∗ (a1 ) ∧ . . .

(18.30)

i=1 K ∗

. . . ∧ d ϕ (ai−1 ) ∧ dK ϕ∗ (ai+1 ) ∧ . . . ∧ dK ϕ∗ (ak ). (see (18.25)). However, this deﬁnition requires checking that iX is well deﬁned. To avoid this, let us modify the approach of Section 18.11 so that it will include the relative case. We can show that iX (ω ∧ ρ) = iX (ω) ∧ ϕ∗ (ρ) + (−1)i ϕ∗ (ω) ∧ iX (ρ) . ∞

(18.31)

∞

If M , N are smooth manifolds, A = C (N ), B = C (M ), and F : M → N is a smooth map, then in the standard notation formula (18.31) acquires the form: iX (ω ∧ ρ) = iX (ω) ∧ F ∗ (ρ) + (−1)i F ∗ (ω) ∧ iX (ρ),

(18.32)

336

Chapter 18

18.16. Description of the algebra of diﬀerential forms. The exact description of the algebra Λ∗M (A) depends on how the algebra A itself is presented. Below we indicate the main way of doing this in terms of the generators of this algebra. Proposition. Let the algebra A be generated by the elements {aα }α∈ℵ . Then (a) the algebra Λ∗M (A), regarded as an A-module, is generated by elements of the form dM aα1 ∧ . . . ∧ dM aαk , k ∈ N, αi ∈ ℵ; (b) if the algebra homomorphism ϕ : A → B is surjective and the categories M and K are ϕ-connected, the algebra homomorphism ϕ∗M,K : Λ∗M (A) → Λ∗K (B) is also surjective; ϕ (c) the kernel of the homomorphism ϕ∗M,K contains the ideal IM,K of the algebra Λ∗M (A) generated by Ker ϕ ∪ Ker(ϕ∗M,K |Λ1M ). Here the homogeneous components Λkϕ of the graded quotient algebra ϕ Λ∗ϕ = Λ∗M (A)/IM,K def

acquire the structure of a B-module, while the B-module Λkϕ is isomorphic to ΛkK (B) if it belongs to the category K. The latter always holds if M = Mod A, K = Mod B. (a) Since ΛkM (A) = (Λ1M )∧k (A), it suﬃces to show that the A-module Λ1M (A) is generated by the diﬀerentials dM αα , α ∈ ℵ. But this obviously follows from the fact that the algebra A, regarded as a k-module (i.e. k-vector space), is generated by elements of the form aα1 . . . aαm , αi ∈ ℵ, whose diﬀerentials are linear combinations of the diﬀerentials dM αα with coeﬃcients from A. (b) Note that the algebra B is generated by the elements of ϕ(aα ) since ϕ is surjective. Therefore, in view of item (a), (18.29), and (18.28), the algebra ΛK is generated (as a B-module) by elements of the form dK (ϕ(aα1 )) ∧ . . . ∧ dK (ϕ(aαk )) = ϕ∗ (dM aα1 ) ∧ . . . ∧ ϕ∗ (dM aαk ) = ϕ∗ (dM aα1 ∧ . . . ∧ dM aαk ) .

(18.33)

(c) The ﬁrst part of this statement is obvious. Note further that the ideal ϕ IM,K is homogeneous, i.e., ϕ,k ϕ ϕ,k def ϕ = IM,K , IM,K = IM,K ∩ ΛkM (A), IM,K k∈N ϕ and let us consider the quotient algebra Λ∗ϕ = Λ∗M (A)/IM,K . Therefore, def ϕ,k Λkϕ , where Λkϕ = ΛkM (A)/IM,K . Λ∗ϕ = k∈N

The (Co)Chain Complexes Coming from the Spencer Sequence

337

ϕ,k Since Ker ϕ · ΛkM (A) ⊂ IM,K , it follows that Ker ϕ · Λkϕ = 0, and so Λkϕ may be regarded as an (A/ Ker ϕ)-module, or, equivalently, as a B-module. ϕ,1 = Ker(ϕ∗M,K |Λ1M ), and we obtain an isomorphism But, by deﬁnition, IM,K Λ1ϕ = Λ1K (B) of k-vector spaces which, for the reasons indicated, may be regarded as an isomorphism of B-modules. Thus, we can identify the image [dM aα ] of the element dM aα under the natural projection Λ1M → Λ1ϕ with dK bα , where bα = ϕ(aα ). Hence,

(bα1 , . . . , bαk ) → [dM aα1 ] ∧ . . . ∧ [dM aαk ] ∈ Λkϕ ,

bαi = ϕ(aαi ),

is a k-derivation of the algebra B that we will denote by ψ(b1 , . . . , bk ). Besides, the homomorphism ϕ∗M,K induces the projection [ϕ] : Λkϕ → ΛkK ,

[dM a1 ] ∧ . . . ∧ [dM ak ] → dK (ϕ(a1 )) ∧ . . . ∧ dK (ϕ(ak )) .

Now if Λkϕ ∈ Ob K, then the k-derivations ψ can be obtained from the universal k-derivation in the category K by means of a B-homomorphism [ψ] : ΛkK → Λkϕ such that dK b1 ∧ . . . ∧ dK bk → [dM a1 ] ∧ . . . ∧ [dM ak ],

if bi = ϕ(ai ) .

This immediately implies [ϕ] and [ψ] are homomorphisms inverse to each other. 18.17. Examples. In the examples presented below, we show how the proposition from the previous section allows to concretely describe the differential forms of various algebras. Here we use the simpliﬁed notation Λi (A) for ΛiModA and ΛiΓ (A) for ΛiGModA and omit the indices M and K in ϕ ϕ,k IM,K and in IM,K . I. Let A = k[x1 , . . . , xn ] and M = Mod A. Then the A-module Λ∗M (A) is free, and the forms that read dM xi1 ∧ . . . ∧ dM xik , 1 ≤ i1 < · · · < ik ≤ n, constitute its free basis. The A-modules Λ∗M (A) are geometrical if k is a ﬁeld of zero characteristic, and nongeometrical otherwise. II. Let A = F2 [x1 , . . . , xn ]. Recall that the geometrization Γ(A) of the algebra A is the Boolean algebra Bn of subsets of a set of n elements. It is freely generated by the elements x ¯i = xi mod Inv(A), i = 1, . . . , n, ¯i . Therefore, the algebra of geometrical satisfying the condition x ¯2i = x diﬀerential forms Λ∗Γ (A), which, as an A-module, is the geometrization of the algebra Λ∗ (A) described above turns out to be none other than the algebra of skew-symmetric polynomials in the variables d¯ xi , i = 1, . . . , n, with coeﬃcients in Bn . III. Let A = k[x] and B = k[x]/(xm ). The algebra B is generated by elements of the form m−1 λi θi , where θ = x mod xm , θm = 0. i=0

338

Chapter 18

Then the proposition from the previous section shows that Λi (B) = 0 if i > 1, while the A-module I ϕ,1 is generated by the elements mxm−1 dx and xm dx. If we assume, for the sake of simplicity, that m1k = 0, it will follow that the B-module Λ1 (B) is generated by the diﬀerential dθ and obeys the relation θm−1 dθ = 0. In other words, the forms θi dθ, 0 ≤ i ≤ m − 2, constitute a basis of Λ1 (B) regarded as a vector space over k. Since the ideal Inv(B) is generated by the element θ, the geometrization of the B-module Λ1 (B), i.e, the B-module Λ1Γ (B) is a one-dimensional space (over k) k·dΓ θ. This example is interesting in that it shows that diﬀerential forms of higher degrees may be nontrivial while the spectrum of the algebra itself is “zero-dimensional”. IV. A = k[x1 , x2 ] and B = k[x1 , x2 ]/(x21 +x22 −1) (the algebra of functions on the “k-circle”). If the characteristic of the ﬁeld k is not 2, then the A-module I ϕ,1 is generated by the 1-forms ωi = f (x1 , x2 )dxi , i = 1, 2, ω = x1 dx1 + x2 dx2 = 1/2 · df (x1 , x2 ) , where f = f (x1 , x2 ) = x21 + x22 − 1. These forms together with the form ρ = x1 dx2 − x2 dx1 generate the A-module Λ1 (A), which can be seen from the fact that dx1 = x1 ω − x2 ρ − f dx1 and similarly for dx2 . This implies that Λ1 (B) = B · ϕ∗ (ρ) is a free one-dimensional module. In its turn, the A-module I ϕ,2 is generated by the forms f dx1 ∧ dx2 = dx1 ∧ ω2 , x1 dxi ∧ dx2 = −dx2 ∧ ω, x2 dx1 ∧ dx2 = dx1 ∧ ω. But since dx1 ∧ dx2 = x1 (x1 dx1 ∧ dx2 ) + x2 (x2 dx1 ∧ dx2 ) − f dx1 ∧ dx2 ∈ I ϕ,2 , we have Λ2 (B) = 0. If the characteristic of the ﬁeld k is 2, then df = 0, and therefore, the A-module I ϕ,1 is generated by the 1-forms f dxi , i = 1, 2,, while I ϕ,2 is generated by the 2-form f dx1 ∧ dx2 . Thus, in this case, the B-modules Λ1 (B) and Λ2 (B) are free, while {dϕ(x1 ), dϕ(x2 )} and {dϕ(x1 ) ∧ dϕ(x2 )} constitute their free bases, respectively. Finally, the modules Λi (B) are trivial for obvious reasons if i > 2, independently of the characteristic of the ﬁeld k, and since all the other modules Λi (B), 0 ≤ i ≤ 2, are free, they are geometrical if and only if the algebra B is geometrical. V. A = k[x, y] and B = k[x, y]/(xy) (“k-cross”). The A-module I ϕ,1 is generated by the 1-forms xydx, xydy, xdy + ydx = d(xy). From this it is easy to derive that Λ1 (B), regarded as a k-vector space, is the direct sum Λ1 (B) =(k[¯ x] · d¯ x) ⊕ (k[¯ y ] · d¯ y ) ⊕ (k · ρ), where

x ¯ = ϕ(x), y¯ = ϕ(y), ρ = x ¯d¯ y = −¯ y d¯ x,

The (Co)Chain Complexes Coming from the Spencer Sequence

339

while the B-module structure is ruled by the relations x ¯y¯ = 0, x ¯ρ = y¯ρ = 0. This implies that def

def

Λx = (k[¯ x] · d¯ x) ⊕ (k · ρ) and Λy = (k[¯ y ] · d¯ y ) ⊕ (k · ρ) are submodules of Λ1 (B) so that Λ1 (B) = Λx + Λy and Λx ∩ Λy = (k · ρ). It is obvious that |B| = {(λ, 0)}λ∈k } ∪ {(0, μ)}μ∈k ⊂ k2 (the union of the “x-axis” and the “y-axis” k2 ). A direct calculation shows that the k-vector space Λ1 (B)h , h ∈ |B|, is one-dimensional if h = (0, 0) and that the B-module Λ1 (B) is geometrical if k is a ﬁeld of zero characteristic. Here the k-vector space Λ1 (B)(0,0) is two-dimensional, while the images of the forms d¯ x and d¯ y constitute its basis. This implies, in particular, that the B-module Λ1 (B) is not projective and so the algebra B is not smooth. On the other hand, ρ∈x ¯Λ1 (B) ∩ y¯Λ1 (B) ∈ Inv Λ1Γ (B) , so that the form ρ is not geometrical. In its turn, this implies that the form dρ = d¯ x ∧ d¯ y is also nongeometrical, and therefore ΛiΓ (B) = 0, i > 1. VI. The considerations developed above can be applied word for word to describe the geometrical diﬀerential forms on the cross K (see Section 9.45). In particular, ΛiΓ (K) = 0, i > 1, while Λ1Γ (K) consists of forms expressed as f (¯ x)d¯ x and g(¯(y))d¯ y . Here y¯ · f (¯ x)d¯ x=0=x ¯ · g(¯(y))d¯ y. Exercise. Let P = C ∞ (K)h . Show that the A-module Diﬀ > 1 P is not geometrical and deduce from this fact that Λ2Γ (K) = 0. 18.18. Description of cosymbols. Recall that the group of cosymbols k CsmlM k (P ) ⊂ J (P ; M) of an A-module P has the additive generators δa1 ,...,ak (jkM )(p),

a1 , . . . , ak ∈ A,

p ∈ P.

Here it is important to note that the operator δa1 ,...,ak (jkM ) : P → J k (P ; M) is of order zero and so is an A-module homomorphism, both with respect to the right and the left A-module structure of the bimodule J k (P ; M). In other words, we have the relation δa1 ,...,ak (jkM )(ap) = aδa1 ,...,ak (jkM )(p),

a ∈ A,

p ∈ P.

(18.34)

Despite its simplicity, it is precisely this relation which allows one to give an exact description of cosymbols for a wide class of A-modules (see the example in the next section). Furthermore, the following fact simpliﬁes this description. Exercise. Let the algebra A be generated by the elements a0i , while the A-module P is generated by the elements p0j . Show that the elements of

340

Chapter 18

the form δa0i

1

,...,a0i

k

(jkM )(p0j ) generate the submodule CsmlM m (P ) of the A-

module J k (P ; M). (Compare with item (a) of the proposition in Section 18.16.) 18.19. Exercise-Example: cosymbols for the cross. Describe the module CsmΓm (K) of geometrical symbols of the algebra C ∞ (K).

19 Diﬀerential Forms: Classical and Algebraic Approach

The last three chapters of this book diﬀer, both in the style of exposition and the character of their contents, from the previous ones. The topics considered in each of them could be developed into separate books, perhaps into several volumes, or form the basis of further research. The present chapter deals with diﬀerential forms and their applications. We see how diﬀerential forms can be conceptually redeﬁned and then naturally generalized to become an inherent and very important part of the diﬀerential calculus over commutative algebras. The applications include the deﬁnition and study of symplectic, Poisson, and contact manifolds, Monge–Amp`ere equations, the Lie derivative, inﬁnitesimal symmetries, derivations of vector bundles, Killing ﬁelds, Jacobi brackets, etc. 19.1. Relationship of classical diﬀerential forms with dRM (A). First of all, let us state and prove the following Proposition. If A = C ∞ (M ) and M = GMod A, then the cochain complex dRM (A) is identical to the standard de Rham cochain complex (of the given manifold M ) deﬁned in Section 18.8. Recall that the C ∞ (M )-module of diﬀerential forms of degree k on the manifold M is denoted by Λk (M ). As we noted previously (see Section 14.11), we have Λ1 (M ) = Λ1M (the natural identiﬁcation). On the other ∧k ∧k hand, we know that Λk (M ) = Λ1 (M ) . Besides, we have ΛkM = Λ1M (by the proposition in Section 17.30). This allows us to identify ΛkM with Λk (M ). © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 19

341

342

Chapter 19

Further, we see that in the notation of Section 14.11, dgeom = dM 0 , so that ∞ : C (M ) → Λ1M the theorem from the same section allows us to identify dM 0 ∞ 1 with the standard diﬀerential d = dgeom : C (M ) → Λ (M ). Moreover, the formula (18.21) shows that the action of the diﬀerential dM k can be . Since a similar formula is expressed in terms of the diﬀerential dM = dM 0 valid for the standard exterior diﬀerential dk : Λk (M ) → Λk+1 (M ) on M , this allows us to identify dM k with dk . Let us stress that the rules of behavior of diﬀerential forms in diﬀerential geometry are simply the manifestation of the general naturality conditions that we studied in Section 18.13. In that context, we considered homomorphisms ϕ = F ∗ , where F : M → N is a smooth map and the categories of geometrical modules played the role of the categories M and K. Here the image of the diﬀerential form ω ∈ Λk (N ) under the homomorphism F ∗ is denoted by F ∗ (ω) and called its inverse image under the map F. Besides, by the restriction of the form ρ ∈ Λk (M ) to a submanifold L ⊂ M we mean the form i∗ (ρ) ∈ Λk (L), where i : L → M is the inclusion map. Traditionally, it is denoted by ρ|L . Exercise. Describe the algebra of diﬀerential forms on the cross K as this was done in example V from Section 18.17. 19.2. Standard description of diﬀerential forms on manifolds. Differential forms on a manifold M are standardly deﬁned in diﬀerential geometry as skew-symmetric C ∞ (M )-polylinear functions ω(X1 , . . . , Xk ) ∈ C ∞ (M ) of vector ﬁelds Xi ∈ D(M ) (see Section 16.13). As we have already noted, def

ω(X1 , . . . , Xk ) = (iXk ◦ · · · ◦ iX1 )(ω), . In particular, in view of (18.24), this implies that (df1 ∧ . . . ∧ dfk )(X1 , . . . , Xk ) = det Xi (fj ) . (19.1) And, in general, all the operations with diﬀerential forms are deﬁned and described in a similar way. For instance, the deﬁnition of the exterior product looks as follows: (ω ∧ ρ)(X1 , . . . , Xk+l ) ¯ def = (−1)(I,I) ω(Xi1 , . . . , Xik )ρ(Xj1 , . . . , Xjl ) , (19.2) I

¯ where I = {i1 , . . . , ik } ⊂ {1, . . . , k + l}, I¯ = {1, . . . , k + l} \ I and (−1)(I,I) is the sign of the corresponding permutation. Similarly, def (−1)i−1 Xi ω(X1 , . . . , Xˆi , . . . , Xk+1 ) dω(X1 , . . . , Xk+1 ) =

+

i i+j

(−1)

ω([Xi , Xj ], X1 , . . . , Xˆi , . . . , Xˆj , . . . , Xk+1 ).

i 2. 2. Show that Ω∧n is a volume form if Ω is a symplectic form on a 2n-dimensional manifold. 3. Show that Cartesian product of symplectic manifolds is also a symplectic manifold. 4. Show that the Poisson bracket introduced in Section 10.13 is nondegenerate. 5. Let A = k[x1 , . . . , x2n ] and M = ModA. Describe all the symplectic triples of the form (A, Ω, ModA). One obvious solution is provided by the form Ωn = dx1 ∧ dx2 + · · · + dx2n−1 ∧ dx2n . 6. Under the conditions of the previous item, let k = Fp . Describe all the symplectic triples of the form (A, Ω, GModA). 19.7. Contact manifolds: origins. Contact geometry, or the geometry of contact manifolds, became a complete theory dealing with (nonlinear) ﬁrst-order diﬀerential equations for functions in one variable in the work of Sophus Lie. From the contemporary point of view, such an equation E has the form F = 0, where F ∈ C ∞ (J 1 P ), while P is a one-dimensional projective C ∞ (M )-module. The solution of the equation E in the usual understanding is an element p ∈ P such that j1 (p)∗ (F ) = 0 (see Section

346

Chapter 19

14.20). From the geometrical point of view, E is a hypersurface in J 1 P whose solutions are the elements p ∈ P such that Im j1 (p) ⊂ E. 19.8 Exercise. In the coordinates on J 1 P speciﬁed in Section 14.20, describe the condition j1 (p)∗ (F ) = 0 and verify that it is actually a ﬁrst-order diﬀerential equation in the classical sense. With the manifold J 1 P , one can naturally associate the contact module CΛ = {ρ ∈ Λ1 (J 1 P ) | j1 (p)∗ (ρ) = 0, ∀p ∈ P } def

and the contact distribution def

C = {X ∈ D(J 1 P ) | iX (ρ) = 0, ∀ρ ∈ CΛ} . Since C is a submodule of the module D(J 1 P ), the vector bundle associated with C is a subbundle of the tangent bundle T (J 1 P ). In particular, its ﬁber Cz at the point z ∈ J 1 P is a subspace of the tangent space Tz J 1 P , while the family z → Cz is a distribution on the manifold J 1 P in the sense in which that term is used in diﬀerential geometry. 19.9 Exercises. 1. Show that in the coordinates on J 1 P speciﬁed in Section 14.20, the contact distribution C is locally generated by the vector ﬁelds ∂ ∂ ∂ + pi , , i = 1, . . . , n . ∂xi ∂u ∂pi 2. In the same coordinates, let ω = du −

n

pi dxi .

i=1

Show that the (local) section s of the bundle J 1 P will have the form s = j1 (p) if and only if s∗ (ω) = 0. 3. Show that ω ∧ (dω)∧n is a volume form on J 1 P , i.e., ω ∧ (dω)∧n z = 0, ∀z ∈ J 1 P. A consequence of these exercises is the fact that the contact module CΛ is projective and one-dimensional. In its turn, this implies that the module C is also projective and its dimension is 2n. In other words, the tangent subspace Cz of its associated geometrical distribution is of dimension 2n. The characteristic property of the contact distribution C will be stated following a useful digression. Example. Note that J 1 P = J 1 M if P = C ∞ (M ). On J 1 M , there exists a global analog of the diﬀerential form ω from the previous exercises, because J 1 M = T ∗ M ⊕ R; namely, it is the diﬀerential form ωM = πR∗ (du) − πT ∗ M (ρ), where ρ is the universal 1-form on T ∗ M , u is the coordinate on R, while πT ∗ M and πR are the projections of J 1 M on T ∗ M and R, respectively.

Diﬀerential Forms: Classical and Algebraic Approach

347

19.10. Digression: curvature of distributions. A distribution on the manifold M is a submodule D of the module D(M ). The quotient module def

κ = κD = D(M )/D is said to be normal for the distribution D, while the κ-valued form def

RD (X, Y ) = [X, Y ] mod D,

X, Y ∈ D,

is called its curvature. This form is obviously skew-symmetric and, as is easily seen, is C ∞ (M )-bilinear. With the curvature form, we naturally associate the curvature map ˆ D : D → HomC ∞ (M ) (D, κD ), R ˆ D (X) : Y → RD (X, Y ), R

X, Y ∈ D.

(19.5)

The distribution D is called regular if the modules D and κD are projective. As a rule, in diﬀerential geometry, only regular distributions are considered. Geometrically, the distribution D is characterized by the family of vector spaces Dz , z ∈ M , constituting |D| (see Section 12.11). Since D ⊂ D(M ), we can deﬁne the natural map γz = γz,D : Dz → D(M )z = Tz M . The geometrical portrait of the distribution D is obtained from the family of subspaces Im γz,D ⊂ Tz M . It is precisely in this way that it is convenient to visualize any distribution (see Figures 19.1 and 19.2). If γz is an embedding, then z is a regular point of the distribution D. If not, it is said to be a singular point. If D is regular, then all the points of the manifold M are regular and the geometrical image of the distribution D is a subbundle of the tangent bundle. The dimension (or rank ) of a regular distribution is deﬁned as the dimension of that subbundle. In the general case, the manifold M is divided into domains of regular points complemented by the nowhere dense set of singular points.

Figure 19.1. One-dimensional distribution with two singular points.

The approach to the theory of distributions brieﬂy sketched in this section is presented here in a form that can be automatically carried over to

348

Chapter 19

Figure 19.2. Two-dimensional distribution.

arbitrary commutative algebras. The deﬁnitions and constructions of this section were precisely motivated by this consideration. Besides, this allows to develop the theory of distributions with singularities even in the classical context of smooth manifolds, which would be quite arduous within the framework of the old purely geometrical approach. For example, we obtain an interesting class of distributions with singularities by stipulating that the module D be projective, at the same time omitting the requirement that the normal module γD be projective. Remark. Just as in 19.7, we can construct the distribution Ck on the manifold J k P ; this distribution is the higher analog of a contact structure (see the next section) and plays a fundamental role in the geometrical theory of diﬀerential equations. Exercises. 1. Verify that the deﬁnitions and constructions of this section can be generalized to arbitrary commutative algebras simply by replacing D(M ) by D(A). 2. Give examples of distributions with singularities for which D is projective. 3. Show that the distribution D is regular if and only if all the maps γz,D are injective. 19.11. More exercises. Let ω ∈ Λ1 (M ) and αω : D(M ) → C ∞ (M ), X → iX ω . def

With the form ω, let us associate the distribution Dω = Ker αω . Then, def obviously, κω = κDω ∼ = Im αω ⊂ C ∞ (M ). Therefore, we can assume that def the curvature form Rω = RDω is C ∞ (M )-valued. Show that 1. the distribution Dω is regular if and only if the form ω is nonsingular, i.e., ωz = 0, ∀z ∈ M ;

Diﬀerential Forms: Classical and Algebraic Approach

349

2. αm (Rω (X, Y )) = dω(X, Y ), X, Y ∈ Dω ; in other words, the curvature form Rω is the restriction of the form dω to the distribution Dω ; 3. the construction of the distribution Dω can be carried over automatically to arbitrary commutative algebras. 19.12. Contact manifolds: abstract algebraic approach. The form ω ∈ Λ1 (N ) on a manifold N of dimension 2n + 1 is said to be contact if ω ∧ (dω)∧n is a volume form on N . It is easy to see that all the covectors ωz , z ∈ M, are nonzero. If ω = f ω, then ω ∧ (dω )∧n = f ω ∧ (df ∧ ω + f dω)∧n = f n+1 ω ∧ (dω)∧n . Therefore, ω ∧ (dω )∧n is also a volume form on N provided that the function f does not vanish anywhere. In that case, the contact forms are considered to be equivalent. A one-dimensional projective submodule CΛ ⊂ Λ1 (N ) is referred to as a contact structure (or as a so-called Cartan structure) if it is locally generated by contact forms. In these terms, a contact manifold is a manifold with a contact structure on it, i.e., the pair (N, CΛ). If N possesses a global contact form ω, then we can write CΛ = C ∞ (N ) · ω. Example. The well-known Darboux lemma asserts that for any contact form there exist local canonical coordinates, in which the form can be expressed as dx2n+1 −

n

xn+i dxi .

i=1

In other words, all contact forms of the same dimension are locally equivalent. Exercise. Indicate some contact form on the 3-sphere. The deﬁnition of contact manifold may be convenient from the computational point of view, however, we will use the conceptual deﬁnition, which is as follows. Deﬁnition. A regular distribution D is said to be contact if the curvature ˆ D is locally an isomorphism. In this case (M, D) is said to be a map R contact manifold. In order to verify that this deﬁnition is equivalent to the generally ˆ D is an isomorphism, then the dimension of the accepted one, note that if R projective module HomC ∞ (M ) (D, κD ) (see (19.5)) equals dim D and, therefore, κD is a one-dimensional projective module. Now since the modules D and HomC ∞ (M ) (D, C ∞ (M )) are isomorphic, the module κD is isomorphic to C ∞ (M )). Therefore, locally there exists a 1-form ω ∈ Λ1 (M ) such that D = Dω (see Section 19.11) and so dω(X, Y ) = −ω([X, Y ]) for X, Y ∈ D. But since the conditions ω(Z) = 0 and Z ∈ D are equivalent, this implies

350

Chapter 19

that dω|D = f κD , where f ∈ C ∞ (M ) is nonzero everywhere. This shows ˆ D is a local isomorphism, and by arguing that ω is a contact form. But R as above, we conclude that D is locally a contact form. 19.13. Example: the secret of the Monge–Amp` ere equation. Recall that the general two-dimensional scalar Monge–Amp`ere equation reads N (uxx uyy − u2xy ) + Auxx + Buxy + Cuyy + D = 0,

(19.6)

where N, A, B, C, D are certain functions of the variables x, y, u, ux , uy . Concrete instances of this class of equations were studied by many authors, beginning with Monge’s 1784 paper. The intimate connection of this equation with contact geometry was discovered by Sophus Lie, who, in particular, established that this class is invariant under contact transformations, i.e., under diﬀeomorphisms of contact manifolds preserving the contact structure. Below, we indicate the previously unnoticed main cause of this, thereby once again demonstrating the importance of “conceptually correct” deﬁnitions. The curvature form of the distribution D allows us to distinguish the class of self-adjoint endomorphisms of the module D. Namely, an endomorphism A : D → D is called self-adjoint if RD (AX, Y ) = RD (X, AY ), ∀X, Y ∈ D. Self-adjoint endomorphisms constitute a Jordan algebra 1 , denoted Snd(D), in which multiplication is deﬁned by the formula S◦T +T ◦S , 2 and idD is the unit. It is also obvious that Snd(D) is a submodule of the C ∞ (J 1 (P ))-module End(D). Now let us consider the contact distribution C on J 1 P from Section 19.7. We shall say that an element p ∈ P is invariant with respect to the endomorphism T ∈ Snd(C) if the vector ﬁeld T (X), X ∈ C, is tangent to the image of the section j1 (p) whenever the vector ﬁeld X is tangent to it. The search for elements of the module P invariant w.r.t. T yields a system of ﬁrst-order diﬀerential equations, because T is a homomorphism, while j1 is an operator of ﬁrst order. In particular, let us put dim M = 2 and pass to the classical notation, taking (x = x1 , y = x2 , u, p = p1 , q = p2 ) to be the standard local chart on J 1 P (see Section 14.20). In these coordinates, def

S∗T =

1 A Jordan algebra is a nonassociative algebra over a field whose multiplication satisfies the following axioms: 1. xy = yx (commutative law), 2. (xy)(xx) = x(y(xx)) (Jordan identity).

Diﬀerential Forms: Classical and Algebraic Approach

351

the basis of the locally free module C is given by the vector ﬁelds ∂ ∂ +p , ∂x ∂u

∂ ∂ +q , ∂y ∂u

∂ , ∂p

∂ , ∂q

and the endomorphism T with the matrix ⎛ ⎞ B −2A 0 −2N ⎜ 2C −B 2N 0 ⎟ ⎜ ⎟ ⎝ 0 2D B 2C ⎠ −2D 0 −2A −B in this basis is self-adjoint. After that, the reader will easily verify that equation (19.6) exactly expresses the invariance condition of the function u = u(x, y) w.r.t. the operator T as deﬁned above. It is useful to note that T 2 = Δ·I, where I is the unit matrix and Δ = B 2 −4AC +4N D. Note that equation (19.6) is elliptic (respectively, parabolic or hyperbolic) if Δ < 0 (respectively, Δ = 0 or Δ > 0). The submodule A of the algebra Snd(C) generated by T and idC is, moreover, a Jordan subalgebra of this algebra. Obviously, the invariance condition for the element p w.r.t. the endomorphisms T ∈ A is equivalent to its invariance w.r.t. T . The condition of A-invariance is conceptually preferable to the condition of T -invariance, since it does not depend on the choice of local coordinates on J 1 P . The initial endomorphism T can be locally normalized so that the endomorphism T0 = f T will satisfy the condition T02 = idC , where = ±1, 0. This, of course, is possible in those domains of the manifold J 1 P where Δ does not change sign. Finally, all the above can be carried over to any ﬁve-dimensional contact manifold (N, D) by setting the problem of ﬁnding so-called Legendre manifolds invariant w.r.t. the given two-dimensional (as a projective module) C ∞ (N )-module of the Jordan subalgebra A ⊂ Snd(D). (In this situation, Legendre manifolds are two-dimensional submanifolds L ⊂ N such that RD |L = 0.) 19.14. Remarks. 1. The invariance condition described above implies that the solutions of the Monge–Amp`ere equation are so-called ζ-complex curves. By deﬁnition, these are curves over the algebras of ζ-complex “numbers” x + yζ, x, y ∈ R, for which ζ 2 = ±1 or 0 depending on the type of equation under consideration; thus, for an elliptic Monge–Amp`ere equation, ζ 2 = −1, and so its solutions are complex curves in the ordinary sense. 2. The interpretation of the Monge–Amp`ere equations given above indicates new approaches to their solution, their classiﬁcation, and so on. 3. The fact that the Monge–Amp`ere equation gives a description in coordinates of the problem of ﬁnding invariant elements p ∈ P in the sense indicated above has a deep meaning in the theory of geometrical singularities of multivalued solutions of nonlinear diﬀerential equations. Here we can only mention that these equations coincide with equations describing singularities of their solutions of certain types, and the singularities themselves are classiﬁed by the algebras of ζ-complex numbers.

352

Chapter 19

19.15. The Lie derivative: general idea. Any vector ﬁeld X on a manifold can be interpreted as the velocity of a certain ﬂow. One can also imagine that the ﬂow captures and carries a certain geometrical quantity G, deforming it along the way. On the intuitive level, the Lie derivative of this quantity along the vector ﬁeld X, denoted by LX (G), is the velocity of its deformation under the action of the ﬂow. We can intuitively think of the ﬂow as a family of diﬀeomorphisms Ft : M → M, F0 = idM depending on “time.” However, in order to “observe” it, it is necessary to pass to the family of transformations of the observables Ft∗ : C ∞ (M ) → C ∞ (M ). Once this is done, it is natural to deﬁne the velocity of the ﬁeld {Ft } as the derivative ∗ dFt∗ def dFt (f ) X= |t=0 , X(f ) = |t=0 , f ∈ C ∞ (M ) . dt dt The next manipulation shows that X is indeed a vector ﬁeld on M : [dFt∗ (f )Ft∗ (g)] dFt∗ (f g) = X(f g) = dt dt t=0 t=0 dFt∗ (f ) dFt∗ (g) = ·g+f · = X(f )g + f X(g) . dt dt t=0 t=0 As an example, let us compute the Lie derivative of LX (Y ) of a ﬁeld Y ∈ D(M ). The image Yt = Ft (Y ) of the ﬁeld Y under the diﬀeomorphism Ft is deﬁned by the relation Ft∗ ◦ Yt = Y ◦ Ft∗ . Let us take the derivatives of both sides of this equation, having in mind that, by deﬁnition, we have LX (Y ) = dYt /dt|t=0 . Since d(Ft∗ ◦ Yt )t=0 = d(Ft∗ )/dtt=0 ◦ Y + idC ∞ (M ) ◦dYt /dtt=0 = X ◦ Y + LX (Y ), it follows that X ◦ Y + LX (Y ) =

Y ◦ Ft∗ = Y ◦ X. dt t=0

Thus, we obtain LX (Y ) = [Y, X] .

(19.7)

This formula clariﬁes the conceptual meaning of the commutator of vector ﬁelds. In particular, it implies the remarkable relation LY (X) = −LX (Y ), which is not at all obvious a priori from the deﬁnition of the Lie derivative. Besides, it allows to deﬁne the Lie derivative of vector ﬁelds for arbitrary commutative algebras, when the heuristic considerations described above lose their meaning. Sometimes, in order to avoid misunderstandings, we use the notation LΨ X , where the index Ψ indicates the nature of the quantity to which the Lie derivative is applied. For example, LD X denotes the Lie derivative of a denotes the Lie derivative of diﬀerential forms. vector ﬁeld, while LΛ X Exercises.

D D 1. Using 19.7, show that [LD X , LY ] = L[X,Y ] .

Diﬀerential Forms: Classical and Algebraic Approach

353

2. From the general idea underlying the notion of Lie derivative, carry Ψ Ψ out a heuristic argument showing that [LΨ X , LY ] = L[X,Y ] . 19.16. The Lie derivative of diﬀerential forms. If ω ∈ Λk (M ), then the family Ft∗ (ω) is a deformation of the diﬀerential form ω under the action of the ﬂow Ft . Therefore, according to the informal deﬁnition, dFt∗ (ω) (19.8) LX (ω) = , ω ∈ Λk (M ) . dt t=0 However, this deﬁnition, although it clariﬁes the meaning of the Lie derivative, does not indicate any method of its practical computation and further manipulations. Hence, it is necessary to ﬁnd the analog of formula (19.7) for diﬀerential forms. To do this, we will use the result of the following simple exercise. Exercise. From the deﬁnition 19.8, show that 1. LX (f ) = X(f ), f ∈ C ∞ (M ) ; 2. LX (ω ∧ ρ) = LX (ω) ∧ ρ + ω ∧ LX (ρ), ω, ρ ∈ Λ∗ (M ) ; 3. d ◦ LX = LX ◦ d. These formulas imply that LX (f dg1 ∧ . . . ∧ dgk ) =LX (f )dg1 ∧ . . . ∧ dgk + dg1 ∧ . . . ∧ dX(gi ) ∧ . . . ∧ dgk . i

Therefore, the Lie derivative LX is entirely determined by the three properties listed in the above exercise. Further, let us note that LX (f ) = X(f ) = (iX ◦ d)(f ), LX (dg) = dLX (g) = dX(g) = (d ◦ iX )(dg) . From this, one might suppose that LX = d ◦ iX + iX ◦ d. This is indeed the case; to prove that formula, it suﬃces to verify that the operator deﬁned in that way satisﬁes relations 1–3 in the above exercise. The ﬁrst of them is obvious, the third immediately follows from d2 = 0. The second one, as the calculation below shows, can be established by successive applications of formulas (18.22) and (18.26) under the assumption that ω belongs to

354

Chapter 19

Λk (M ): (d ◦ iX + iX ◦ d)(ω ∧ ρ) =d(iX (ω) ∧ ρ + (−1)k ω ∧ iX (ρ)) + iX (dω ∧ ρ + (−1)k ω ∧ dρ)) =d(iX (ω)) ∧ ρ + (−1)k−1 iX (ω) ∧ dρ + (−1)k dω ∧ iX (ρ) + ω ∧ d(iX (ρ)) + iX (dω) ∧ ρ + (−1)k+1 dω ∧ iX (ρ) + (−1)k iX (ω) ∧ dρ + ω ∧ iX (dρ) =(d ◦ iX + iX ◦ d)(ω) ∧ ρ + ω ∧ (d ◦ iX + iX ◦ d)(ρ) . (19.9) Thus the previous heuristic considerations lead us to the following deﬁnition. Deﬁnition. Let X ∈ D(A). The operator LX = dM ◦ iX + iX ◦ dM is called the Lie derivative of diﬀerential forms in the category M. As before, a direct consequence of this deﬁnition is the formula LX (ada1 ∧ . . . ∧ dak ) =X(a)da1 ∧ . . . ∧ dak + ada1 ∧ . . . ∧ dX(ai ) ∧ . . . ∧ dak .

(19.10)

i

19.17. Relative Lie derivative. The previous deﬁnition can be immediately generalized to relative vector ﬁelds. Namely, if X : A → B is such a ﬁeld with respect to the algebra homomorphism ϕ : A → B, while M and K are ϕ-connected categories, then we can put LX = dM ◦ iX + iX ◦ dK : Λ∗K (A) → Λ∗M (B) . def

(19.11)

Exercise. Verify the validity of the following formulas 1. dM ◦ LX = LX ◦ dK ; 2. LX (a) = X(a), a ∈ A; 3. LX (ω ∧ ρ) = LX (ω) ∧ ϕ∗ (ρ) + ϕ∗ (ω) ∧ LX (ρ), ω, ρ ∈ Λ∗ (A); 4. LaX = aLX + dM a ∧ iX ,

def

(dM a ∧ iX )(ω) = dM a ∧ iX (ω).

Recall that a map of smooth sets (see Section 7.13). F : M → N is called smooth if F ∗ (C ∞ (N )) ⊂ C ∞ (M ). Proposition. Let Ft : M → N, − < t < , be a family of smooth maps of dF ∗ smooth sets and let X = dtt |t=0 be the corresponding relative vector ﬁeld. Then dFt∗ (ω) LX (ω) = , ω ∈ Λ∗Γ (C ∞ (N )). dt t=0 The proof directly follows from the fact that the operators LX and dFt∗ ∗ ∞ dt t=0 acting on ΛΓ (C (N )) coincide on forms of degree ≤ 1 and the

Diﬀerential Forms: Classical and Algebraic Approach

355

ﬁrst three formulas of the previous exercise are satisﬁed for both of them. 19.18 Proposition. Let X, Y ∈ D(A), a ∈ A and ω ∈ ΛkM . Then the following relations hold : (a) [LX , LY ] = L[X,Y ] , (b) [LX , iY ] = i[X,Y ] . Using formula (19.10) these two relations can be proved by a direct calculation. A more economical proof consists in the obvious veriﬁcation of these relations for forms of degree ≤ 1. Then, in order to generalize this to higher degree forms, it suﬃces to check that the operators [LX , LY ] and [LX , iY ] act on the exterior product of diﬀerential from according to the same rules as the operators L[X,Y ] and i[X,Y ] , respectively. But this last fact follows from the result of the exercise in Section 19.17. 19.19. Let us deﬁne the Lie derivative of a diﬀerential form by the following formula: LX (ω)(X1 , . . . , Xk ) = ω(X1 , . . . , [Xi , X], . . . , Xk ) . (19.12) X(ω(X1 , . . . , Xk )) + i

It is easily checked by induction (using the relation (see (19.18)) that iX1 ◦ LX = LX ◦ iX1 − [LX , iX1 ] = LX ◦ iX1 − i[X,X1 ] . Namely, if ω ∈ ΛkM , then this relation implies that LX (ω)(X1 , . . . , Xk ) = (iX1 ◦ LX )(ω) (X2 , . . . , Xk ) = LX (iX1 (ω))(X2 , . . . , Xk ) + ω([X1 , X], X2 , . . . , Xk ) . (19.13) For k = 1, the last relation immediately yields the required result. Assuming further that the formula that we are proving holds for forms of degree k ≥ 1, we obtain LX (iX1 (ω))(X2 , . . . , Xk ) = X(iX1 (ω))(X2 , . . . , Xk ) iX1 (ω))(X2 , . . . , [Xi , X], . . . , Xk ) = X(ω(X1 , X2 , . . . , Xk )) + 2≤i≤k

+

(ω(X1 , X2 , . . . , [Xi , X], . . . , Xk ) .

2≤i≤k

Comparing this last relation to formula (19.13), we conclude the proof. 19.20. Inﬁnitesimal symmetries: general idea. The Lie derivative is part of the natural language of inﬁnitesimal symmetries of mathematical structures and objects of various nature. On the purely intuitive level, a symmetry of some object O is simply a transformation (of a speciﬁc type)

356

Chapter 19

of the object into itself, usually having a speciﬁc term associated to it. For example, a symmetry of some algebraic object is usually called an automorphism, while the symmetry of manifold is its diﬀeomorphism. The symmetries of a given object O obviously constitute a group, which it is natural to denote by Sym O. Now let Ft : O → O, F0 = idO , be a family of symmetries depending on “time.” Then it would be appropriate to call dFt /dt|t=0 an inﬁnitesimal symmetry of the object O. Of course, this very rough idea must be clariﬁed. For example, the notion of derivative for a family of diﬀeomorphisms Ft : M → M makes no sense because the diﬀerence Ft+Δt − Ft is meaningless. But it will acquire a meaning if we recall that M = |C ∞ (M )| and Ft = |Ht |, where Ht is an automorphism of the algebra C ∞ (M ). Indeed, in this case the classical formula 1 dHt def (Ht+Δt − Ht ) = lim Δt→0 Δt dt makes sense, ﬁrst of all because the set of automorphisms of the algebra C ∞ (M ) constitute a vector space over R. Assuming that the following limit 1 def dHt Θ = (HΔt − idC ∞ (M ) ) = lim dt t=0 Δt→0 Δt exists, it is not hard to see that Θ : C ∞ (M ) → C ∞ (M ) is an R-linear map and Θ(f g) = Θ(f )g + f Θ(g). Thus, the heuristic considerations developed above lead us to the idea of deﬁning the inﬁnitesimal symmetry of a manifold M or the algebra C ∞ (M ) as the vector ﬁeld Θ ∈ D(M ). In this form, the idea can be carried over immediately to the general case of commutative algebras, whereas this cannot be done from the heuristic considerations that led to it. Thus it is natural to call any “vector ﬁeld” X ∈ D(A) an inﬁnitesimal symmetry of the algebra A. Comparing the ideas of inﬁnitesimal symmetry (IS) with the idea of Lie derivative (see Section 19.15), we come to the conclusion that a ﬁeld X ∈ D(M ) should be referred to as an IS of the quantity G provided that LX (G) = 0. For example, the ﬁeld X is the IS of the ﬁeld Y ∈ D(M ) if [X, Y ] = 0, or, respectively, the IS of the form ω ∈ Λk (M ) if LX (ω) = 0. In particular, if Ω is a symplectic form, then it is an IS of the symplectic manifold (M, Ω). Or if X is the velocity ﬁeld of a moving liquid, while ω = dx ∧ dy ∧ dz is the Euclidean volume form, then X is an IS of ω if and only if the liquid is incompressible. Exercises. Show that (1) a Hamiltonian ﬁeld Xf is the IS of a symplectic manifold (M, Ω); (2) divX = 0 if and only if X is the IS of the Euclidean volume form. These examples and similar ones show that inﬁnitesimal symmetries appear in diﬀerent guises in various circumstances in diﬀerential geometry, mechanics, and physics. For example, a Killing ﬁeld in Riemannian geometry (see Example V in Section 19.21) is an IS of Riemann manifolds.

Diﬀerential Forms: Classical and Algebraic Approach

357

Besides, such examples show that the notion of inﬁnitesimal symmetry, as well as the notion of Lie derivative, automatically carry over to objects of the diﬀerential calculus in commutative algebras. 19.21. Specifying the deﬁnitions of inﬁnitesimal symmetry. The exact deﬁnition of the notion of inﬁnitesimal symmetry in various concrete contexts is not always obvious and may require additional computations. From this point of view, we must ﬁrst determine what the ordinary notion of inﬁnitesimal symmetry is in the given context, which, as a rule, is not difﬁcult. Then we must diﬀerentiate the family of such symmetries depending on “time” and declare that the obtained result is an inﬁnitesimal symmetry. The following examples illustrate this procedure. Examples. I. Poisson manifolds. From the intuitive geometrical point of view, a symmetry of a Poisson manifold (M, {·, ·}) is a diﬀeomeorphism F preserving the bracket {·, ·}). But, as we have seen, it is more appropriate to speak of the symmetry of the Poisson algebra (C ∞ (M ), {·, ·}), i.e., of isomorphisms H : C ∞ (M ) → C ∞ (M ) that preserve this bracket. This clearly means that H({f, g}) = {H(f ), H(g)}. Further, following our procedure, let us consider the family of symmetries Ht , H0 = id |C ∞ (M ) , and formally diﬀerentiate this deﬁnition of symmetry w.r.t. t:

dHt (f ) dHt ({f, g}) Ht (g) = , Ht (g) + H(f ), . dt dt dt The last equation for X = d/dt|t=o acquires the form X({f, g}) = {X(f ), g} + {f, X(g)} and should be regarded as a symmetry of the given Poisson manifold. Moreover, this “computational deﬁnition” can be immediately carried over to arbitrary Poisson algebras. Namely, a derivation X ∈ D(A) is said to be an inﬁnitesimal symmetry of the Poisson algebra (A, {·, ·}), if X({a1 , a2 }) = {X(a1 ), a2 } + {a1 , X(a2 )}, ∀a1 , a2 ∈ A .

(19.14)

Note that Hamiltonian ﬁelds Xa , a ∈ A, are inﬁnitesimal symmetries of the given Poisson algebra, since in the present case the condition (19.14) is identical to the Jacobi identity for the elements a, a1 , a2 . II. Distributions. From the geometrical point of view, a symmetry of a distribution (M, D) is a diﬀeomorphism F : M → M which preserves the module D. This means that the image F (Y ) of any ﬁeld Y ∈ D also belongs to D. Further, diﬀerentiating, the inclusion Ft (Y ) ∈ D, where Ft , F0 = id |M , is the family of symmetries of the distribution D, we obtain the inclusion dFt (Y )/dt|t=0 ∈ D. But dFt (Y )/dt = LX (Y ) = [Y, X], where X = dFt∗ /dt|t=0 (see Section 19.15). Thus, a vector ﬁeld X ∈ D(M ) such that [Y, X] ∈ D should be called an inﬁnitesimal symmetry of the distribution D provided Y ∈ D or, symbolically, if [X, D] ⊂ D. An inﬁnitesimal symmetry of a contact distribution is called a contact ﬁeld.

358

Chapter 19

As before, these deﬁnitions can be automatically carried over to arbitrary commutative algebras. (Do that!) III. Tensors. A symmetry of a covariant tensor T of rank k on a manifold M , understood as a C ∞ (M )-polylinear function T (X1 , . . . , Xm ) on D(M ) (see Section 16.13) can be naturally considered as being a diﬀeomorphism F : M → M such that F ∗ (T (X1 , . . . , Xm )) = T (F (X1 ), . . . , F (Xm )) . Diﬀerentiating this equality w.r.t. t for the family of symmetries Ft and using, as in the previous example, the fact that dFt∗ dFt (Y )/dt = LX (Y ) = [Y, X], where X = , dt t=0 we obtain the following deﬁnition of an inﬁnitesimal symmetry X for the tensor T : m X(T (X1 , . . . , Xm ) = T (X1 , . . . , [Xi , X], . . . , Xm ) . (19.15) i=1

Moreover, as is easily seen, similar considerations lead to the formula LX (T )(X1 , . . . , Xm ) = X(T (X1 , . . . , Xm ))− m − T (X1 , . . . , [Xi , X], . . . , Xm ).

(19.16)

i=1

The deﬁnition (19.15) automatically carries over to arbitrary smooth algebras, while the formula (19.15) in that case may be taken for the deﬁnition of the Lie derivative of the tensor T . Recall that a nondegenerate symmetric covariant tensor g(X1 , X2 ) of rank 2 is said to be a pseudo-Riemannian metric on the manifold M . In the literature, the inﬁnitesimal symmetries of this tensor are called Killing ﬁelds. VI. Modules. It is natural to understand a symmetry of a vector bundle π over a manifold M as a diﬀeomorphism F of the manifold which is covered by a morphism of the bundle π to itself, the latter being an isomorphism on ¯ H) consisting of each ﬁber. The algebraic paraphrase of this is a pair (H, ∞ ¯ of an an automorphism H of the algebra C (M ) and an automorphism H ∞ R-vector space P , where P is a C (M )-module, the two automorphisms ¯ p) = H(f )H(p), ¯ being related by the formula H(f f ∈ C ∞ (M ), p ∈ P . We will obtain an inﬁnitesimal version of this by diﬀerentiating at t = 0 ¯ 0 = idP , H0 = ¯ t , Ht ) under the assumption H the family of symmetries (H ¯ ¯ t (p). As a idC ∞ (M ) together with the connecting relation Ht (f p) = Ht (f )H result, we obtain the formula ¯ p) = X(f )p + f X(p), ¯ X(f

(19.17)

Diﬀerential Forms: Classical and Algebraic Approach def

359

def

¯ = dH ¯ t /dt|t=0 , X = dHt /dt|t=0 . Here X, as we have already where X ¯ ∈ Diﬀ 1 (P, P ), which follows directly from the seen, is the vector ﬁeld X relation (19.17). The previous argument shows that in the general algebraic context, a pair ¯ X), X ¯ ∈ Diﬀ 1 (P, P ), X ∈ D(A) (X, obeying the relation ¯ ¯ X(ap) = X(a)p + aX(p),

a ∈ A, p ∈ P.

(19.18)

should be called an inﬁnitesimal symmetry of the A-module P . In the geometrical context, such a pair is known as a Der-operator or a derivation of the vector bundle π. Such operators arise in diﬀerent situations of diﬀerential geometry. For example, such are covariant derivatives in the theory of connections. 19.22 Exercises. 1. Give an example of a Poisson manifold/algebra whose inﬁnitesimal symmetry is not exhausted by Hamiltonian ﬁelds. 2. Let (N, ω) be a contact manifold (see Section 19.12). Show that a ﬁeld X ∈ D(N ) is contact if and only if LX (ω) = f ω for some function f ∈ C ∞ (N ). 3. Describe Killing ﬁelds in the Euclidean metric g=

n i=1

dxi dxi (=

n

dx2i ).

i=1

19.23. The Lie algebra of inﬁnitesimal symmetries. The set sym O of inﬁnitesimal symmetries of some object O has a natural Lie algebra structure. This can be seen by applying the general procedure for ﬁnding inﬁnitesimal symmetries from Section 19.21. Finite symmetries of some object O obviously form a group under composition. If Ht and Gt are families of symmetries of an object O depending on “time” and deﬁned over some algebra C ∞ (M ), then Ht ◦ Gt is also such a family. Therefore, d(Ht ◦ Gt )/dt|t=0 is also an IS of such an object if H0 = G0 = idC ∞ (M ) . Now if X = dHt /dt|t=0 , Y = dGt /dt|t=0 , then d(Ht ◦ Gt )/dt|t=0 = (dHt /dt ◦ Gt + Ht ◦ dGt /dt)|t=0 = X + Y, i.e., X + Y ∈ sym O. This shows how an additive structure arises in sym O, thereby transforming sym O into an Abelian group. Similarly, the multiplication of inﬁnitesimal symmetries by numbers is introduced by the relation dHλt /dt|t=0 = λX, λ ∈ R. Further, the relation −1 d(Gs ◦ Ht ◦ G−1 s )/dt|t=0 = Gs ◦ X ◦ Gs

360

Chapter 19

shows that Gs ◦ X ◦ G−1 s is a family of inﬁnitesimal symmetries depending on s. Finally, the relation d(Gs ◦ X ◦ G−1 s )/ds|s=0 = [Y, X] implies that [Y, X] is also an IS. This is how the Lie bracket is introduced in sym O. The fact that the operations deﬁned above satisfy the axioms deﬁning the notion of Lie algebra is also proved by diﬀerentiating the appropriate expressions. We leave this veriﬁcation to the reader. In cases where the IS of the quantity G is expressed by the formula LX (G) = 0, the fact that the inﬁnitesimal symmetries constitute a Lie algebra is a consequence of the formula [LX , LY ] = L[X,Y ] that we established above for vector ﬁelds and diﬀerential forms (see Sections 19.15 and 19.18). 19.24. Example: description of contact ﬁelds. The deﬁnition of contact manifold that we have been using has the advantage, among other things, of yielding a simple description of its inﬁnitesimal symmetries, i.e., contact vector ﬁelds. The Lie algebra of these ﬁelds for a contact manifolds (M, C) will be denoted by Cont C. Obviously, Cont C is an R-vector space. Here it should be noted that the ﬁeld f Z, f ∈ C ∞ (M ), is not a contact ﬁeld in general even if the ﬁeld Z is contact. The R-linear map c : Cont C → κ = Cont /C Cont C Z

→

(Z mod C) ∈ κ

plays the key role in the description of contact ﬁelds. First of all, it is injective. Indeed, Ker c = (Cont C)∩ C. Hence, if Z ∈ Ker c, then [Z, C] ⊂ C, because the ﬁeld Z is contact. Now since Z belongs to C, the inclusion ˆ C (Z) = 0 (cm. (19.5)). But since [Z, C] ⊂ C is equivalent to the equality R ˆ RC is a local isomorphism, it follows that Z = 0. If the C ∞ (M )-module κ is free, then the map c will also be surjective. To verify this, let us assign, to each ﬁeld Z ∈ D(M ), the homomorphism of C ∞ (M )-modules φZ : C → κ, φZ (X) = [X, Z](mod C) . Recall also that the curvature map ˆ C : C → HomC ∞ (M ) (C, κ) R is an isomorphism if the module κ is free. Therefore, we can ﬁnd a ﬁeld ˆ C (Y ) = φZ . But R ˆ C (Y ) = −φY , so that we will have Y ∈ C such that R φZ − φY = φZ−Y = 0. Now the last equality is equivalent to [X, Z − Y ] ∈ C, ∀X ∈ C, and so we can conclude that Z − Y ∈ Cont C. Thus, the map c is an isomorphism of vector spaces if the module κ = κC is free. But, since in the general case the module κ is projective, it follows that c is a local isomorphism.

Diﬀerential Forms: Classical and Algebraic Approach

361

It is also important to note that c allows to carry over the C ∞ (M )-module structure from κ to Cont C. Hence Cont C is a submodule of the module κ. Finding conditions for the coincidence of Cont C and κ is a topological problem, but this is hardly a topic to be discussed here. If Z ∈ Cont C and c(Z) = ν, then ν is said to be the generating section of the contact ﬁeld Z, and is denoted by Xν . Since the module κ is onedimensional, the local coordinates of its elements can be identiﬁed with functions; we then use the notation Xf , f ∈ C ∞ (M ). In its turn, the algebra Lie structure in Cont C is transferred by c to κ, because c is a local isomorphism, while the map κ ν → Xν ∈ D(M ) is a ﬁrst-order diﬀerential operator (see the exercise below). The bracket on κ deﬁned in this way is denoted by {·, ·}C , so that by deﬁnition, we have [Xμ , Xν ] = X{μ,ν}C ,

μ, ν ∈ κ .

This bracket is a ﬁrst-order diﬀerential operator in each argument. In the literature such a bracket is often called the Jacobi bracket. 19.25 Exercises.

1. Show that the ﬁeld

Z = [f, [g, χ]](ν) = Xf gν − f Xgν − gXf ν + f gXν , where f, g ∈ C ∞ (M ), μ, ν ∈ κ is a contact ﬁeld that belongs to C and deduce from this that the map κ ν → Xν ∈ D(M ) is a ﬁrst-order diﬀerential operator. 2. Express the ﬁeld Xν in the canonical coordinates of Section 14.20, assuming that ∂/∂u is a basis vector ﬁeld of κ. 3. Express the bracket {·, ·}C in the same coordinates. def

4. Let ν0 ∈ κ be the (local) basis of κ and deﬁne (locally) Xf = Xf ν0 . Consider the subalgebra C ∞ (M )0 = {f ∈ C ∞ (M ) | Xν0 (f ) = 0} ⊂ C ∞ (M ) . def

Show that {C ∞ (M )0 , C ∞ (M )0 }C ⊂ C ∞ (M )0 and prove that the restriction of the bracket {·, ·}C to the subalgebra C ∞ (M )0 transforms it into a Poisson algebra. 19.26. Conclusion. At ﬁrst glance, this chapter might seem to be merely a technical compilation of algebraic generalizations of many classical notions from diﬀerential geometry, mechanics, physics, and diﬀerential equations. But it is actually much more—it opens the way to many concrete applications of the theory developed in this book to these and other branches of the physico-mathematical sciences, in particular in practically important situations when the classical theory of smooth manifolds is

362

Chapter 19

not applicable because smoothness requirements no longer hold and singularities appear. It turns out that the diﬀerential calculus on commutative algebra, as developed here, is the tool that works in these situations just as well as the classical theory works in the smooth case.

20 Cohomology

20.1. In the previous chapters, starting from conceptually deﬁned diﬀerential forms, we were able to deﬁne the de Rham and Spencer complexes (Chapter 18) and construct the main functors of the diﬀerential calculus (Chapters 17 and 19). The next natural step in this direction is to study the cohomology of these complexes, which carries a wealth of useful information. In this chapter, we present a series of examples and constructions that show how these cohomology theories may be eﬀectively applied, and we demonstrate the techniques used in working with them. 20.2. De Rham cohomology of commutative algebras. Let M be a diﬀerentially closed category of A-modules. Denote by def

i (M) = HdR

Ker dM i Im dM i−1

the i-th de Rham cohomology group of the algebra A in the category M (see Section 18.8). In the most important cases, we simplify this notation to def

i i HdR (A) = HdR (Mod A),

def

i HΓi dR (A) = HdR (GMod A)

Proposition 18.10 allows to carry over, in the standard way, the exterior multiplication operation of diﬀerential forms (see Section 17.18) to the cohomology groups. def

[ω] ∧ [ρ] = [ω ∧ ρ], ω ∈ ΛkM , ρ ∈ ΛlM , © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 20

363

364

Chapter 20

where [ν] ∈ H m (M) denotes the cohomology class of the closed form ν. This multiplication is obviously skew-commutative, i.e., [ρ]∧[ω] = (−1)kl [ω]∧[ρ]. Therefore, it transforms the direct sum ∗ i (M) = HdR (M) HdR i≥0

into a skew-commutative algebra, or, equivalently, into a Z-graded commutative algebra (Z-graded commutative algebras will be discussed in the next chapter, see 21.6). Since dM 0 (1A ) = 0, the cohomology class [1A ] is well deﬁned and is the unit of this algebra. Let ϕ : A → B be a homomorphism of commutative algebras, let M and K be ϕ-connected diﬀerentially closed categories of A- and B-modules, respectively. The naturality condition of the exterior diﬀerential (see Section 18.13) is tantamount to the fact that the family of maps ϕ∗ : ΛiM → ΛiK , i ≥ 0, determines a homomorphism of the corresponding de Rham complexes that we denote by ϕdR = ϕM,K dR : dRM (A) → dRK (B) . Accordingly, the cohomology group homomorphism induced by ϕdR is denoted by i i ϕidR : HdR (M) → HdR (K),

∗ ∗ ϕ∗dR : HdR (M) → HdR (K) .

Since ϕ∗ : Λ∗M → Λ∗K is a homomorphism of algebras of diﬀerential forms, it follows that ∗ ∗ ϕ∗dR : HdR (M) → HdR (K)

(20.1)

is also a homomorphism of algebras with unit. In particular, if A = B and ϕ = idA , then any subcategory K of the category M is tautologically ϕ-connected with it and conversely. Therefore, in this case, it is natural to refer to the homomorphism (20.1) as the comparison homomorphism of the de Rham cohomology of the algebra A in the categories M and K. For example, we have the most obvious comparison homomorphism ∗ (A) → HΓ∗ dR (A). HdR We conclude this section with the following statement, which has now become obvious: Proposition. The de Rham cohomology groups of the algebra C ∞ (M ) in the category of geometrical modules coincide with the ordinary de Rham cohomology groups of the manifold M , i.e.,, HΓ∗ dR (C ∞ (M )) = H ∗ (M ) . This automatically follows from the isomorphism (constructed earlier) between the algebras Λ∗GMod C ∞ (M ) and Λ∗ (M ) that identiﬁes dGMod C ∞ (M ) and d = dM .

Cohomology

365

20.3. Exercises. Compute the de Rham cohomology of the algebra A in the categories ModA and GModA if 1. A = C(x) and A = R(x) (the ﬁeld of rational functions on C and R, respectively); 2. A = F2 (x) (the ﬁeld of rational functions on the ﬁeld F2 ); 3. A = S −1 (C∞ (R2 ), where S is the multiplicative set generated by the coordinate functions x and y; 4. A = C∞ (S 1 ) (use the fact that A is isomorphic to the algebra of periodic functions on R1 ); 5. A = F2 [x, y] (use the fact that A is a module over the algebra F2 [x2 , y 2 ]); 6. A = F2 [x, y]/(x2 + y 2 − 1); 7. A = C ∞ (K) (only in the category GModA), where K is coordinate cross in R2 . 20.4. Residue theory. It is useful to have a procedure that would allow to determine whether or not a closed form determines a nonzero de Rham cohomology class. Below, we explain how it is possible, for certain algebras, to construct such a procedure on the basis of an algebraic analog of Leray’s residue theory. Let A be an algebra without zero divisors, let M be a diﬀerentially closed category over A, and let Λ∗M be the algebra of diﬀerential forms in that category. Leray condition. Suppose s ∈ A is an irreversible nonnilpotent element and S = {sn }n0 . Recall (see 15.30) that Λ∗S −1 M = S −1 Λ∗M . To simplify the notation, the image of Λ∗M under the localization map S −1 : Λ∗M → Λ∗S −1 M will be denoted by Λ∗ . Set Λ∗s = Λ∗ /(sΛ∗ + ds ∧ Λ∗ ). def

Strictly speaking, the right-hand side of the previous equality should have been written as Λ∗ /(S −1 (s)Λ∗ + dS −1 M S −1 (s) ∧ Λ∗ ), but, in order to simplify the notation and make the formulas more readable, we will write simply s instead of S −1 (s), dropping the subscript of the exterior derivation operator. The image of a diﬀerential form ω ∈ Λ∗s under the factorization map Λ∗ → Λ∗s is called the restriction of ω to the ideal S −1 (sA) ⊂ S −1 A and is denoted by ω|s . The choice of the term “restriction” in this situation will become clear in the following example. Example. Suppose that s ∈ k[x1 , . . . , xn ] is an irreducible polynomial, def A ⊂ A¯ = k(x1 , . . . , xn ) is the algebra of all fractional-rational functions not containing s in the denominator, and M = Mod A. In this case, we ∗ ∗ ¯ Λ∗ = Λ∗ have S −1 A = A, Mod A , Λs = ΛMod A/νs , where νs is the principal

366

Chapter 20

ideal generated by s. (Check this.) From the geometrical point of view, the elements of Λ∗s may be regarded as the restrictions of the forms from Λ∗ to the submanifold in kn determined by the equation s = 0. Below we assume that for the form ds ∈ Λ∗ , the Leray condition holds, i.e., the following complex ∧ds

∧ds

∧ds

∧ds

∧ds

0 → A −→ Λ1 −→ . . . −→ Λi−1 −→ Λi −→ . . . is acyclic. For simplicity, we set p = ∞ if char k = 0. Exercise. Under the conditions of the previous example, verify whether the Leray condition holds for the polynomial s = x2 + y 2 − 1 ∈ k[x, y]. 20.5 Proposition. Let ω ∈ Λ∗ and ϕ = ω/sn , n < p. Suppose that there exists a form α ∈ Λ∗ such that dϕ = α/sn ; then there exist forms ψ, θ ∈ Λ∗ such that ϕ = ds/sn ∧ ψ + θ/sn−1 . Since dϕ = α/sn , it follows that nds ∧ ω = sdω − sα

(20.2)

Therefore, we have ds ∧ dω = ds ∧ α. By the Leray condition, there exists a form θ ∈ Λ∗ such that dω = ds ∧ nθ + α. Substituting the last relation in (20.2), we obtain nds ∧ ω = nds ∧ sω. Again using the Leray condition, we ﬁnd a form ψ ∈ Λ∗ such that ω = sθ + dsψ. Corollary. If the assumptions of Proposition 20.5 hold and n > 1, then there exist forms β, γ ∈ Λ∗ such that ϕ = d(β/sn−1 ) + γ/sn−1 . 20.6 Proposition. Let ω ∈ Λ∗ and let the form ϕ = ω/s be closed, then there exist forms ψ, θ ∈ Λ∗ such that ϕ = ds/s ∧ ψ + θ. The restrictions ψ|s , θ|s of the forms ψ, θ on S are uniquely determined, and ψ|s is closed. The forms ψ|s are called the residue forms of ϕ on S and are denoted by res[ϕ]s . 1. The the existence of the forms ϕ, θ follows from Proposition 20.5. 2. Uniqueness of ψ|s . It suﬃces to prove that ψ|s = 0 whenever ϕ = 0. If ds/s ∧ ψ + θ = 0, then ds ∧ θ = 0 and by the Leray condition there exists a form θ ∈ Λ∗p such that θ = ds ∧ θ . Hence, ds ∧ ψ + sθ = ds ∧ (ψ + sθ ). Using the lemma once more, we obtain ψ + sθ = ds ∧ θ , where θ ∈ Λl−2 p so that ψ|s = 0. The form ds/s∧θ = ds/s∧ϕ is closed, therefore by the previous argument θ|s = res[ds/s ∧ θ]s is uniquely deﬁned. 3. The form ψ|s is closed. Since dϕ = ds/s ∧ dψ + dθ = 0, it follows that dψ|s is the zero residue form on S and, by item 2 of the proof, is equal to zero.

Cohomology

367

For the statement and proof of this proposition, we followed F. Pham. If in the conditions of the previous proposition we have ϕ ∈ Λls , then, as is easily seen, res[ϕ]s = 0. Indeed, in that case, we may put θ = ϕ and ψ = 0. Then res[ϕ]s = ψ|s = 0. Obviously, if ϕ ∈ Λ∗ , then res[ϕ]s = 0. 20.7 Proposition. Let ϕ = ω/s, ω ∈ Λ∗ , ω be an exact form, and char k = 0, then res[ϕ]s is also exact. By deﬁnition, ω/s = d(α /sn ), α ∈ Λ∗ . If n > 1, then, by the corollary to Proposition 20.5, there exist forms β, γ ∈ Λ∗ such that β γ α + n−1 . = d sn sn−1 s Therefore, ω/s = d(γ/sn−1 ). Hence, by induction on n, we can ﬁnd a form α ∈ Λ∗ such that ω/s = d(α/s). By Proposition 20.5, there exist forms ψ, θ ∈ Λ∗ such that α/s = ds/s ∧ ψ + θ. Thus α ω ds =d = − ∧ dψ + dθ, s s s and res[ω/s]s = −dψ|s . 20.8 Proposition. Let ϕ = ω/sn , ω ∈ Λ∗ , 0 n < p, be a closed form. Then there exists a form α ∈ Λ∗ such that ϕ is cohomologous to α/s. If n = 0, then α = sω. The proof of this proposition for the case n > 1 is similar to the proof of Proposition 20.7. 20.9. Residues in the case of characteristic zero. Let char k = 0. If ϕ ∈ Λ∗S −1 M is a closed form, then, by Proposition 20.8, there exists a form α ∈ Λ∗ such that ϕ is cohomologous to α/s. The cohomology class of the form res[α/s]s is uniquely determined by Proposition 20.6. It is called the residue class of ϕ, and is denoted by Res[ϕ]s . Consider a cohomology class Θ ∈ H l (S −1 M). Let ϕ, ψ be arbitrary elements of Θ. Then, by Proposition 20.7, we have Res[ϕ]s = Res[ψ]s . Thus the k-homomorphism Ress : H l (S −1 M) → H l−1 (Λ∗s ),

Ress (Θ) = Res[ϕ]s , ϕ ∈ Θ,

is well deﬁned. 20.10 Theorem. The sequence 0 −→ H l (Λ∗ ) −→ H l (S −1 M) −→s H l−1 (Λ∗s ) i

Res

(20.3)

is exact. Obviously, Im i ⊂ Ker Ress . Let ϕ be an arbitrary representative of a cohomology class Θ ∈ H l (S −1 M). By deﬁnition, the kernel of the factorization homomorphism Λ∗ → Λ∗s is generated by s and ds. This implies

368

Chapter 20

that there exist forms α, β, γ, δ ∈ Λ∗ such that ds ∧ (dα + sβ + ds ∧ γ) + δ. s This means that the form ϕ is cohomologous to ds ∧ β + δ. Therefore Θ ∈ Im i. Remark. Let R ⊂ A be a multiplicative subset, and s ∈ / R. Suppose R−1 : Λ∗S −1 M → R−1 Λ∗S −1 M = Λ∗R−1 S −1 M is a monomorphism. Then the Leray condition holds for the element dR−1 S−1 M s ∈ R−1 Λ∗ , and the residue homomorphism ResR−1 s : H l (R−1 S −1 A; R−1 S −1 M) → H l−1 (Λ∗R−1 s ) ϕ=

can be constructed. The diagram H l (S −1 A; S −1 M)

H l (R−1 S −1 A; R−1 S −1 M)

Ress

ResR−1 s

H l−1 (Λ∗s )

H l−1 (Λ∗R−1 s )

is commutative. This means that it is not too diﬃcult to generalize most of the constructions described above to algebraic geometry, speciﬁcally to schemes and sheaves. Example: discrete valuation rings. A ring A with unit and without zero divisors is said to be a discrete valuation ring if there exists an element s ∈ A such that any ideal of the ring A is generated by some degree of the element s. In other words, all the ideals of the ring A are principal ideals generated by powers of one and the same element s, which is deﬁned up to multiplication by an invertible element. By deﬁnition, in A there exists a unique maximal ideal ν = {sa, a ∈ A}. An example of a discrete valuation ring is the ring of formal power series k[[x]] in one variable. Each element a ∈ A may be written in the form αsn , n 0, where α is an invertible element. As above, let us consider the multiplicative set S = {sn }n0 and note that the fraction ﬁeld A¯ of the ring A is nothing but S −1 A. For the category M, let us take the category Mod A of all A-modules. In that case, we have Λ∗ = Λ∗ (A) and ∗ ¯ ¯ To simplify the notation, we put Λ∗ −1 S −1 Mod A = Mod A. S Mod A = Λ (A). 20.11 Exercises. 1. Prove that Λ∗s does not depend on the choice of the generating element s and so we have the right to use the notation Λ∗ν . 2. Prove that Λ∗ν is isomorphic to Λ∗ (A/ν). 20.12. Deﬁnition. Let Ω = {ωj } be a ﬁnite subset of Λ∗ . A ﬁnite subset G = {gi } is called a ν-system of Ω if the following conditions hold: (1) the elements dgi are linearly independent; (2) if dG is a subalgebra of L generated by dgi , then Ω ⊂ dG.

Cohomology

369

It is not diﬃcult to prove the following two statements. 20.13 Proposition. The forms dgi are linearly independent over A¯ iﬀ the elements gi are algebraically independent over the ﬁeld of constants k. The simple proof of this proposition is left to the reader.

20.14 Proposition. For any ﬁnite subset of Ω = {ωj } ⊂ Λ∗ , there exists a ν-system. 20.15 Proposition. ds/s ∈ / Λ∗ . Assume the converse. Then there exist elements fi , gi ∈ A such that k ds = fi dgi (20.4) s i=1 By Propositions 20.13, 20.14, the elements gi are algebraically independent over k. On the other hand, it follows from the last condition that the elements {s, g1 , . . . , gk } are algebraically dependent over k. Let F (x, y1 , . . . , yk ) ∈ k[x, y1 , . . . , yk ] be an irreducible polynomial over k such that F (s, g1 , . . . , gk ) = 0. It is easily shown that (20.4) is true iﬀ the polynomial F (x, y1 , . . . , yk ) can be presented in the form F (x, y1 , . . . , yk ) = xF (x, y1 , . . . , yk ) + C, where F (x, y1 , . . . , yk ) ∈ k[x, y1 , . . . , yk ], 0 = C ∈ k. Then sF (s, g1 , . . . , dk ) + C = 0, and the element s is invertible in the algebra A: F (s, g1 , . . . , dk ) 1 =− . s C This contradiction proves the proposition. 20.16 Proposition. For any ﬁnite subset of Ω = {ωj } ⊂ Λ∗ , there exists a ν-system containing s. Let {gi } be any ν-system for the set Ω. If the set {s, gi } is algebraically independent over k, then it is just the required ν-system. Otherwise, there exists a nontrivial constraint k αi dgi = 0, α0 = 0, αi ∈ A. α0 ds + i=1

Each of the nonzero coeﬃcients αi may be presented in the form αi = βi sri , βi ∈ A. Let r = mini>0 {ri } and l be an integer such that rl = r. The assumption r0 < rl contradicts Proposition 20.15, and thus the set G\{gl } is the required ν-system. 20.17 Proposition. The Leray condition holds for the form ds.

370

Chapter 20

Let ω ∈ Λ∗ . By the previous proposition, for any ω there exists a ν-system {s, gi }. Obviously, if ds ∧ ω = 0, then h(i1 ,...,ik ) ds ∧ dg i1 ∧ · · · ∧ dg ik . ω= (i1 ,...,ik )

¯ we can assign its residue res[ϕ]s . Let Now to any closed form ϕ ∈ Λ∗ (A) s be another generating element of the ideal ν. For ds , the Leray condition also obviously holds.

20.18 Proposition. res[ϕ]s = res[ϕ]s . Indeed, s = αs , where α is an invertible element in A. Hence, d(αs ) ds dα ds ∧ψ+θ = ∧ ψ + θ. ∧ ψ + θ = ∧ψ+ s αs s α Thus, res[ϕ]s does not depend on the choice of the element s generating the ideal ν and we can speak of the residue form res[ϕ]ν of the closed form ϕ on the ideal ν and of the residue class Res[θ]ν of the cohomology class ¯ θ ∈ H ∗ (A). Bringing together all the above concerning discrete valuation rings, we come to the conclusion that for such rings we have Theorem 20.10, and the exact sequence (20.3) takes the form Res

i

¯ −→ν H l−1 (A/ν). 0 −→ H l (A) −→ H l (A)

(20.5)

The example that follows is interesting from two points of view. First, in it we consider a non-Noetherian algebra. Second, this example shows that the validity of the Leray condition depends on the choice of the “coeﬃcient” category M. Let A be a non-Noetherian local algebra, i. e., an algebra with unique maximal ideal ν. Assume that ν is principal, let s be a generator of ν, let As = S −1 A, and let A¯ be the ﬁeld of fractions of the algebra A. Denote by M, Ms the complete subcategories of Mod A, Mod As , respectively, such that objects of M and Ms have no torsion. It is easy to prove that the categories M and Ms are diﬀerentially closed. The natural embeddings A ⊂ As ⊂ A¯ induce the homomorphisms j

Λ∗M → Λ∗Ms → Λ∗M(A) ¯ . i

Since Λ∗M and Λ∗Ms are torsion-free, the homomorphisms i and j have no kernels. 20.19 Proposition. Under the above described conditions, the Leray condition holds. Consider any diﬀerential form ω= h(i1 ,...,ik ) dg i1 ∧ · · · ∧ dg ik . (i1 ,...,ik )

Cohomology

371

Put def ¯ ¯ def R = k(s, h(i1 ,...,ik ) , . . . , g 1 , . . . ) and R = R ∩ A.

Therefore, R is a discrete valuation ring, whence the Leray condition holds for Λ∗Mod R . As we know, the homomorphism α : Λ∗Mod R → Λ∗Mod R¯ has no ¯ ⊂ A¯ is separable, therefore the homomorphism kernel. The extension R α : Λ∗Mod R¯ → Λ∗Mod A¯ also has no kernel. Let ik i1 ω = h(i1 ,...,ik ) dgMod R ∧ · · · ∧ dgMod R . (i1 ,...,ik )

If ds∧ω = 0, then dMod R s∧ω = 0. By the Leray condition for the complex Λ∗Mod R , there exists a form λ ∈ Λ∗Mod R such that ω = dMod R s ∧ λ . Set λ = β ◦ α(λ ). Then ω = β ◦ α(ω ) = ds ∧ λ. 20.20. Residues in the case of positive characteristic. Suppose that char k = p > 0. For n ≡ 0 mod p, Propositions 20.5, 20.8 do not hold. For this reason, the construction of residue forms and residue classes described above does not work. Instead of it, it is possible to construct a spectral sequence whose terms play a similar role. The interested reader can ﬁnd the details of this construction in [26]. Concluding this excursion in residue theory, note that multiple (iterated) residues may be constructed in a similar way. 20.21. Relative cohomology. The naturally deﬁned restriction map Λ∗ (M ) → Λ∗ (N ) is surjective if the submanifold N ⊂ M is closed (see Sections 4.11, 4.12). Its kernel is obviously a subcomplex of the de Rham complex of manifold M and is denoted by . . . → Λ(i−1) (M, N ) → Λi (M, N ) → Λi+1 (M, N ) → . . . It is the complex of relative diﬀerential forms of the manifold M modulo N . Its cohomology groups are denoted by H i (M, N ). H ∗ (M, N ) = i

Elements of Λ∗ (M, N ) are called relative diﬀerential forms. In the general algebraic context, such a situation arises when there is a surjective homomorphism of commutative algebras ϕ : A → B, and the categories M and K are ϕ-connected. The kernel of the homomorphism ϕdR (see Section (20.1)) will be denoted by dRi (M, K), (20.6) dR(M, K) = i

and is called the complex of ϕ-relative diﬀerential forms (in the category M). Sometimes, we shall also use the notation dR(ϕ), which is more convenient in certain contexts. In particular, we should note the case of the

372

Chapter 20

comparison of categories, especially when K is a subcategory of M. A typical example is M = Mod A, K = GModA. Accordingly, (ϕ, M, K)-relative cohomology groups of the cochain complex 20.6 are denoted by ∗ i (M, K) = HdR (M, K) . (20.7) HdR i

Thus we have obtained a short exact sequence of complexes ϕM,K

ϕdR

dR dR(M) −→ dR(K) → 0 0 → dR(M, K) −→

(20.8)

It is known from basic cohomology theory that to any short exact sequence of complexes one can canonically associate a long exact sequence of cohomology groups which, in the case of the sequence 20.8, will be of the form ∂ i−1

i−1 i (K) −→ HdR (M, K) · · · → HdR

M,K H(ϕdR )

−→

H(ϕdR )

i HdR (M) −→ ∂i

H(ϕdR )

i+1 i −→ HdR (K) −→ HdR (K) → . . .

(20.9)

Here H(ϕM,K dR ) and H(ϕdR ) are homomorphisms of cohomology groups induced by the homomorphisms of the complexes ϕM,K and ϕdR , respecdR tively, while ∂ i is the connecting homomorphism. Recall that the connecting i (K), θ ∈ dRi (K). Then, homomorphism is deﬁned as follows. Let [θ] ∈ HdR M,K since 20.8 is exact, we have θ = ϕdR (θ ), θ ∈ dRi (M). But dK (θ) = 0, and so ϕdR (dM (θ )) = (dM (ϕdR (θ )) = 0, in view of the exactness of (20.8), we obtain dM (θ ) ∈ Im(ϕM,K dR ). Finally, we put def

i+1 ∂ I ([θ]) = [] ∈ HdR (M, K),

where

dM (θ ) = ϕM,K dR () .

The reader will easily check that the cohomology class [] does not depend on the choice of the diﬀerential forms used in the above construction. The long exact sequence (20.9) describes the relationship between the cohomology of a complex and the cohomology of its kernel and image under certain maps. For this reason, sequences of this type are one of the main technical instruments used for the practical computation of cohomology, and not only cohomology. 20.22. Example: the algebraic Newton–Leibnitz formula. Here we present a simply example of the application of the sequence (20.9) to cohomological integration theory. The necessity of such a theory becomes apparent in the framework of standard diﬀerential geometry, where integration in the sense of measure theory is not possible for a number of reasons. From this point of view, in this section we shall discuss the simplest particular case of the general Stokes formula, namely the Newton–Leibnitz formula.

Cohomology

373

Let k be a ﬁeld of zero characteristic. Then |k[x]| = k (as sets). Also let λ, μ ∈ k, λ = μ, and let I = {λ} ∪ {μ} be a two-point set in |k[x]|. The algebra of k-valued functions on I can be naturally identiﬁed with the algebra of ordered pairs (α, β) ∈ k2 , λ → α, μ → β with coordinate-wise addition and multiplication. Denote this algebra by k[λ,μ] , thereby stressing that we regard I as an oriented “interval” [λ, μ] in |k[x]|. Now let us describe the exact sequence 20.9 in the category of geometrical modules for the surjection ϕ : k[x] → k[λ,μ] , using the simpliﬁed notation H i (k[x]), H i (k[λ,μ] ) and H i (k[x]), k[λ,μ] ) for its terms. It is obvious that all the nontrivial terms are grouped together in the fragment HdR (ϕ)

∂

0 → H 0 (k[x]) −→ H 0 (k[λ,μ] ) −→ H 1 (k[x], k[λ,μ] ) → 0. Here H 0 (k[x]) consists of constants and is therefore canonically isomorphic to k, while H 0 (k[λ,μ] ) is canonically isomorphic to k2 . Besides, [HdR (ϕ)](ν) = (ν, ν). Therefore, H 1 (k[x], k[λ,μ] ) ∼ = coker HdR (ϕ). It is remarkable that among these isomorphisms there exist exactly two canonical ones. Namely, ∂(α, β) = β − α and ∂(α, β) = α − β. On the other hand, according to the deﬁnition, we have ∂(α, β) = [df (x)], where the polynomial f (x) ∈ k[x] satisﬁes f (λ) = α and f (μ) = β. Therefore, it is natural to regard the cohomology class of the 1-form df (x) as a relative form, either as the number f (μ) − f (λ), or as f (λ) − f (μ), depending on the orientation of the interval I. For this reason, it is appropriate to denote the relative μ cohomology class [df (x)] by the symbol λ df (x). This can be written in the form of the Newton–Leibnitz formula (over the ﬁeld k) μ f (x) dx = f (μ) − f (λ) λ

which under the cohomological approach is transformed from a theorem to the deﬁnition of the integral. Exercise. Investigate the sequence 20.9 for the algebra k[x] in the case when in |k[x]|, instead of two, there is an arbitrary ﬁnite number of points. 20.23. Homotopy formula. One of the most eﬀective methods for calculating cohomology is based on the use homotopy operators. Historically, this method appeared in the old algebraic topology quite naturally. In terms of de Rham cohomology on manifolds or sets, it looks as follows. Smooth maps of smooth sets F, G : M → N are homotopic if there exists a smooth family Ft : M → N, 0 ≤ t ≤ 1 such that F = F0 and G = F1 . Proposition. (Homotopy theorem.) If two maps F and G are homotopic, then the corresponding cohomology group homomorphisms H i (F ∗ ), H i (G∗ ) : H i (N ) → H i (M ) coincide. Below the symbol Ft∗ is understood as the homomorphism of algebras (corresponding to the map Ft ) of the geometrical diﬀerential forms

374

Chapter 20

Λ∗ (N ) → Λ∗ (M ). Recall that

dFt∗ dt t=0 is a vector ﬁeld w.r.t. the map F = F0 . Similarly, dFt∗ XFs = dt t=s is an Fs -relative vector ﬁeld. The desired homotopy operators are direct consequences of the generalized Newton–Leibnitz formula 1 dFt∗ ∗ ∗ ∗ ∗ dt, (20.10) F − G = F1 − F 0 = dt 0 XFt =

which in its turn follows from the usual formula for functions, because (Ft∗ )z : Λ(N )∗z → Λ∗ (M ), z ∈ N , is a map of ﬁnite-dimensional vector spaces, while Ft may be regarded as a family of such maps continuously depending on the parameter z. Now, using the proposition from Section 19.17 and the fact that the diﬀerentials dM and dN do not depend on t, we see that 1 1 dFt∗ dt = (iXt ◦ dN + dM ◦ iXt ) )dt dt 0 0 1 1 = iXt dt ◦ dN + dM ◦ iXt dt . (20.11) 0

0

As the result, we come to the homotopy formula F ∗ − G∗ = H ◦ dN + dM ◦ H,

where

def

H =

0

1

iXt dt;

(20.12)

applying it to the closed form ω ∈ Λi (N ) we see that F ∗ (ω) − G∗ (ω) = dM (H(ω)). This implies that the image of the cohomology class of a closed form does not change when we pass to a homotopic map. 20.24. A few words about algebraic topology. Contemporary algebraic topology arose from concrete problems in the diﬀerential calculus. Thus, Poincar´e, developing the theory of integration on manifolds, proposed to use their triangulation, thereby reducing the problem to a known Euclidean case. This in its turn led to the use of simplicial complexes, simplicial (co)homology, and so on. Poincar´e was the ﬁrst to use modern terminology and apply homotopy considerations to compute what was basically de Rham cohomology of Euclidean spaces. Below, we present a few examples of computations of this kind. Also see Section 20.30, where it is indicated how one can deﬁne cohomology with arbitrary coeﬃcients on the basis of de Rham cohomology. More generally, the subject matter of this chapter may be regarded as a ﬁrst step in the return of algebraic

Cohomology

375

topology to its forgotten sources in the more general context of the diﬀerential calculus over commutative algebras. 20.25. Homotopy operators. The importance of homotopy techniques for the computation of (co)homology deserves to be distinguished by a special deﬁnition. Deﬁnition. Let Kα = {Kαi , dα }, α = 1, 2, be cochain complexes. The cochain maps ψα : K1 → K2 , α = 1, 2, are said to be homotopic if there exists a map of cochain groups H : K1 → K2 , H(K1i ) ⊂ K2i−1 such that ψ1 − ψ 2 = d1 ◦ H + H ◦ d 2 . Here the map H is called a homotopy operator or simply a homotopy. From this algebraic point of view, formula (20.12) relates geometric homotopies (which are easy to see only for R-algebras) to algebraic homotopies. Exercise. 1. Generalize the notion of geometrical homotopy to maps of pairs (manifold, submanifold) and prove the analog of the Homotopy Theorem (see 20.23) for relative de Rham cohomology. 2. Let us call a commutative diagram of homomorphisms A

ϕ

(20.13)

α

α

B

A

ψ

B

a morphism of α to α . Generalize the notion of algebraic homotopy to morphisms of surjections of commutative algebras over the same ﬁeld and prove the Homotopy Theorem for relative de Rham cohomology. 20.26. Examples. I. Poincar´e Lemma. Poincar´e himself stated this lemma as asserting that any closed diﬀerential form of positive dimension in n-dimensional Euclidean space Rn is exact. To prove this, he considered the family of homotheties Ft = {x → tx, x = (x1 , . . . , xn ), 0 ≤ t ≤ 1}. Further the argument goes as follows. First notice that F0 = ζ ◦ ψ, where ψ : Rn → {0} is the map to the singleton {0}, while ζ : {0} → Rn is obviously an embedding. But since F1 = idRn , the left-hand side of the homotopy formula reads idΛ∗ (Rn ) −ψ ∗ ◦ ζ ∗ . Now the homomorphism ψ ∗ : Λi (Rn ) → Λi ({0}), i > 0, is obviously trivial, and so the homotopy formula tells us that ω = dH(ω), ω ∈ Λi (Ri ), if dω = 0. II. By introducing the notion of deformation retract, the above proof of the Poincar´e lemma carries over almost word for word to a much more general situation. In topology, a subspace W is said to be a deformation

376

Chapter 20

retract of U ⊃ W if there exists a family of continuous maps Ft , : U → W, 0 ≤ t ≤ 1, such that F0 |W = idW , Im(F0 ) = W, F1 = idU . For example, {0} is a deformation retract of Rn . The left-hand side of the homotopy formula for cochains of the complex K of “ordinary” cohomology of U has the form idK −F0∗ since F1 = idU . This means that F0∗ induces the identity map of the cohomology H ∗ (K) of the complex K. Further, let F¯0 : U → W be the restriction of the map F0 to U . Then F0 = ι ◦ F¯0 , where ι : W → U is the inclusion map and so F0∗ = F¯0∗ ◦ ι∗ . This implies that the cochain map ι∗ induces the following injective map in cohomology: H ∗ (U ) → H ∗ (W ), while F¯0∗ (in the opposite direction) is surjective. On the other hand, ι∗ ◦ F¯0∗ = id∗W , since F¯0 ◦ι = idW , which clearly implies the surjectivity of the cohomology homomorphism H ∗ (W ) → H ∗ (U ) induced by the cochain map ι∗ and the injectivity of the cohomology map H ∗ (U ) → H ∗ (W ) induced by F¯0∗ . In this way, we obtain an isomorphism between the cohomology groups of the given space U and those of its deformation retract W . The above argument carries over easily to the general algebraic context. For example, for de Rham cohomology, we can consider a surjective homomorphism ϕ : A → B, ϕ|B = id B, of the algebra A on its subalgebra B, together with ϕ-connected categories M and K of A- and B-modules, respectively, and a given (algebraic) homotopy H : ΛiM (A) → Λi−1 K (B) connecting idA and ϕ. It is useful to note that the requirements imposed on f and B may be replaced by a single one only, namely ϕ2 = ϕ if we put B = Im ϕ. From the same considerations as above, the homotopy formula corresponding to these data establishes an isomorphism between the coho∗ ∗ (A) and HK (B). This motivates us to refer to the mology algebras HM projection ϕ as a deformation retract in the general algebraic situation. Here, however, we lose the easily understandable geometrical mechanism for constructing the homotopy H, which must be searched for, depending on the nature of the algebra A. Finally, let us note that the above is applicable to any natural complex of the diﬀerential calculus, including the Spencer complex. III. Let us present several typical examples of deformation retracts. First of all, they are given by ﬁber bundles whose ﬁbers are contractible. The latter means that such a space (ﬁber) has the singleton as its retract. The image of any section of such a ﬁber bundle is a deformation retract of its total space. For instance, it is the zero section of any vector bundle, the central circle of the M¨obius strip (see Figure 6.2 on page 67) or of the solid torus (Figure 20.1). The axis of an inﬁnite solid cone (Figure 20.1) is its deformation retract, and the crossing point is that of the cross. Note also that the perpendicular projection of the inﬁnite solid cone onto its axis gives a simple example of a Serre ﬁbration (i.e., a ﬁbration satisfying the

Cohomology

377

Figure 20.1. Three deformation retracts.

so-called homotopy lifting property) whose ﬁbers are homotopy equivalent, but not homeomorphic, so that this ﬁbration is not a ﬁber bundle. IV. The classical Newton–Leibnitz formula. Each of the endpoints of the interval I = [a, b] is its deformation retract, hence we have H 0 (I) = R and H i (I) = 0, i > 0, so that the argumentation in 20.22 carries over, words for word, to the algebra C ∞ (I) and to the relative cohomology of the pair (I, ∂I). Thus we have shown that the classical Newton–Leibnitz formula may be regarded as the deﬁnition of the integral b df (x), f (x) ∈ C ∞ (I). a

20.27. The suspension isomorphism. It is well known that, in algebraic topology, there are some very useful constructions carried out on any topological space, such as the suspension and the loop space. The ﬁrst of these is interesting in that it allows to compare the cohomology groups of spaces of diﬀerent dimensions (we already saw this for retractions). Recall that the suspension ΣX over the space X is obtained from the product X × I, I = [0, 1], when its top (X × {1}) and bottom (X × {0}) are contracted to singletons u1 and u0 , so that BX = (X × {1}) (X × {0}) is contracted to the set {u1 , u0 } (see Figure 20.2). In that case, we have the following natural isomorphism H i (X × I, BX; R) ∼ = H i (ΣX, {u1 , u0 }; R) between the “ordinary” relative cohomology groups with coeﬃcients in any Abelian group R.

378

Chapter 20

Figure 20.2. Suspension.

Below, for a smooth set X, we compute the relative de Rham cohomology groups H i (X × I, BX). To this end, let us consider the following fragment of the exact sequence for the pair (X × I, BX) (see 20.9): ı

∂

j

H i (X × I) → H i (BX) −→ H i+1 (X × I, BX) → H i+1 (X × I).

(20.14)

But X ≡ X × {}, ∈ I, is obviously a deformation retract of X × I, so that we have the canonical isomorphism H j (X × I) ∼ = H j (X). Therefore, I ∼ having in mind this isomorphism, we have H (BX) = H j (X)×H j (X) and ı(ϑ) = (ϑ, ϑ). Moreover, the homomorphism j is trivial, since the restriction of a cocycle representing ϑ ∈ H j (X × I, BX) on X × {}, = 0, 1, is trivial by deﬁnition, and, on the other hand, H j (X × I) ∼ = X × {}, = 0, 1. Now H i+1 (X × I, BX) ∼ = coker ∂, while coker ∂ ∼ = H i (BX)/ Im ı ∼ = [H i (X) × H i (X)]/Δ ∼ = H i (X), where Δ ∼ = Im ı ∼ = H i (X) is the diagonal in H i (X) × H i (X). In other i+1 words, H (X × I, BX) can be naturally identiﬁed with the antidiagonal in H i (X)×H i (X), which by deﬁnition consists of pairs (ϑ, −ϑ), ϑ ∈ H i (X). Thus, the suspension isomorphism ⇑X : H i (X) → H i+1 (X × I, BX), in view of the isomorphisms speciﬁed above, has the form ϑ → (ϑ, −ϑ), which is not very informative. The reader can remedy this situation by doing the following exercise. 20.28 Exercises. 1. Noting that the exterior product of a relative cohomology class by an absolute one is a relative class of the same pair or a surjection, prove that ⇑X (ϑ) = pr∗1 (dx) ∧ pr∗2 (ϑ), pr1 and pr2

Cohomology

379

Figure 20.3. Homotopy from the square to the disk.

are the projections of X × I onto I and X, respectively, while dx is understood as the relative cohomology class of the pair (I, ∂I). 2. Consider the cube ∇k = I1 ×, . . . , ×Ik , where Ii is a copy of the closed interval I with the standard coordinate xi . Verify that H i (∇k , ∂ ∇k ) = 0 if i = k, and H k (∇k , ∂ ∇k ) = R · d¯ x1 ∧ · · · ∧ d¯ xk ,

x ¯i = pr∗i (xi ),

where pri : ∇k → Ii is the natural projection.

k 3. Let Dk = { i=1 x2i ≤ 1} ⊂ Rk be the k-dimensional disk. Show that H i (Dk , ∂Dk ) ∼ = H i (∇k , ∂∇k ), using the representation of the pairs shown in Figure 20.3 and the visually obvious geometrical homotopies that the picture suggests. 4. Compute the cohomology groups of the k-dimensional sphere S k , regarding it as the boundary of the disk Dk+1 .

380

Chapter 20

5. Iterating the suspension construction k times, one obtains the isomorphism ⇑kU : H i (U ) → H i+k (U × ∇k ), U × ∂ ∇k ), (20.15) ϑ → pr∗∇ (¯ x1 ∧ · · · ∧ d¯ xk ) ∧ pr∗U ϑ, where pr∗∇ and pr∗U are the natural projections of U × ∇k ) on ∇k and U , respectively. Verify this. Now if we replace, using item 3 of the Exercise, ∇k by Dk in (20.15), then we obtain the so-called Thom isomorphism for the Cartesian product by the disk. In the general case, this isomorphism takes place for ﬁbrations with ﬁber the disk, and plays an important role in the theory of characteristic classes. Exercise. In the example from Section 20.22, if we replace the algebra k[x] by k[x1 , . . . , xn ], then the result does not change. Verify this directly, or by constructing the homotopy needed to prove this fact. 20.29. Homotopy complex: general idea. There is a remarkable analogy between homotopy operators and cochain maps. Namely, in the notation of Section 20.23, the relation d1 ◦ ψ − ψ ◦ d2 = 0 deﬁnes a cochain map. Similarly, it is natural to interpret the relation d1 ◦ H + H ◦ d2 = 0, where H : K1 → K2 , H(K1i ) ⊂ K2i−1 , as a cochain map of degree −1. This map obviously induces the following map on the cohomology level: H i (Hi ) : H i (K1 ) → H i−1 (K2 ). At ﬁrst glance it would seem that there is no sense in comparing cohomology groups of diﬀerent dimensions. However, in connection with the suspension isomorphism, when the cohomology groups of the suspension over a space are mapped to another space, such a situation arises. Similarly, the map of a space to the suspensions over another space induces a homomorphism of the cohomology groups of the second space to those of the ﬁrst one, raising the dimension by 1. Moreover, having in mind the isomorphism of iterated suspension or the Thom isomorphism, we see that the comparison of cohomology groups of diﬀerent dimensions make sense, and so it is natural to include in our considerations the “higher” analogs of homotopy operators, which change the degree of cochains by a ﬁxed integer m. If we assume that H : K1 → K2 , H(K1i ) ⊂ K2i+m , and put def

ð = d1 ◦ H − (−1)m H ◦ d2 , then, obviously, ð2 = 0 and so we have obtained a homotopical complex. Here the fact that ψ is a cochain means that ðψ = 0, i.e., that ψ is a cocycle of degree zero of that complex. Similarly, the fact that cochain maps ψ1 are ψ2 homotopic means that ψ1 − ψ2 = ðH, i.e., ψ1 and ψ2 lie in the same cohomology class of the homotopic complex. The general scheme described above, however, needs to be developed, because it leaves unanswered the question of the nature of the maps

Cohomology

381

H : K1 → K2 under consideration. For instance, they can simply be linear maps of vector spaces or diﬀerential operators of some kind. In the next section, in order to illustrate this aspect, we shall consider the de Rham complex of smooth manifolds or smooth sets. 20.30. De Rham cohomology with arbitrary coeﬃcients. A suﬃciently complete understanding of the meaning and the role of de Rham cohomology is a thing of the future, when there will be enough “experimental material” obtained from computations and applications. But at this point, we already have and can suspect a lot of unexpected and interesting developments. For example, de Rham cohomology can be nontrivial for algebras A with a ﬁnite spectrum |A|. Or it seems quite natural to study the structure of those algebras related by homomorphisms that induce isomorphisms in de Rham cohomology or, possibly, in diﬀerential (co)homology of other types, say, Spencer cohomology. In this connection, we must stress that the language of the diﬀerential calculus in commutative algebras allows to carry over to this new context the entire standard apparatus of algebraic topology. For example, using it, one can show that the cohomology of a simplicial complex S with real coefﬁcients is isomorphic to the de Rham cohomology of the algebra of smooth functions on S in the category of geometrical modules over that algebra. By deﬁnition, that algebra consists of functions on S whose restriction to any closed simplex is smooth in the ordinary sense. Moreover, in the corresponding de Rham complex, we will be able to distinguish the integer subcomplex , whose cohomology will be identical to the ordinary cohomology of S with integer coeﬃcients. In its turn, this will allow us to deﬁne the cohomology of S with arbitrary coeﬃcients. In other words, the historical process that began with the combinatorial topology of simplicial complexes and much later led to the well-known de Rham theorem can be transformed into the cornerstone of the de Rham cohomology of commutative algebras even on the basis of this familiar material. 20.31 Proposition. In the geometrical category M = GMod C ∞ (M ), the Spencer δ-cohomology of a smooth manifold M , i.e., the cohomology groups of the cochain complexes Csmm M (see Section 18.7) and the cohomology groups of the Spencer jet complexes Spenm M are trivial if m ≥ 1. Besides, ∞ H 0 (Csm0M ) = H 0 (Spenm M ) = C (M ), m ≥ 0.

Since the diﬀerentials of the Spencer jet complexes are homomorphisms of C ∞ (M )-modules, it suﬃces to prove the required statement locally. If x = {(x1 , . . . , xn )} are local coordinates in some domain U ⊂ M , then the C ∞ (U )-module Λm−k (C ∞ (U )) CsmlkΓ (C ∞ (U )) Γ C ∞ (U )

382

Chapter 20

is generated by elements of the form f (x)(dxi1 · · ·dxik )⊗C ∞ (U ) (dxj1 ∧. . .∧dxjm−k ), f (x) ∈ C ∞ (U ). (20.16) This expression is rather ambiguous, since the symbols dxi in the product dxi1 · · · dxik denote 1-cosymbols, whereas they appear in the product dxj1 ∧ . . . ∧ dxjm−k as 1-forms. To stress this diﬀerence, we denote these xj = dxj for dxi by diﬀerent symbols, e.g., x¯i = dxi for 1-cosymbols and d¯ 1-forms. In this notation, elements of (20.16) at some point z ∈ U can be written in the form c¯ xi1 . . . x ¯ik · d¯ xj1 ∧ . . . ∧ d¯ xjm−k c ∈ R, i.e., as diﬀerential forms over the polynomial algebra R[¯ x1 , . . . , x ¯n ]. Here ∞ (C (M )) at the the forms corresponding to elements of the complex Csmm Γ point z constitute a subcomplex, which is invariant w.r.t. the “Poincar´e homotopy” (see the ﬁrst example in Section 20.26). Thus the required asser∞ tion concerning the cohomology groups of the complexes Csmm Γ (C (M )) is a consequence of the Poincar´e lemma. ∞ As to the cohomology groups of the complexes Spenm Γ (C (M )), they can be easily computed from the long exact cohomology sequence corresponding to the short exact sequence of complexes m m−1 ∞ ∞ 0 → Csmm (C ∞ (M )) → 0. Γ (C (M )) → SpenΓ (C (M )) → SpenΓ

Indeed, this long exact sequence consists of fragments of the form m ∞ i ∞ 0 → H i (Csmm Γ (C (M ))) → H (SpenΓ (C (M ))) → ∞ (C ∞ (M ))) → H i+1 (Csmm → H i (Spenm−1 Γ (C (M ))) → 0, Γ ∞ i+1 ∞ in which the groups H i (Csmm (Csmm Γ (C (M ))) and H Γ (C (M ))) are trivial if, as we noted before, m ≥ 1. This obviously implies that the m−1 ∞ i (C ∞ (M ))) are isomorphic. groups H i (Spenm Γ (C (M ))) and H (SpenΓ Therefore, 0 ∞ i ∞ H i (Spenm Γ (C (M ))) = H (SpenΓ (C (M ))),

while the complex Spen0Γ (C ∞ (M )) has the form 0 → C ∞ (M ) → 0 and hence has the cohomology groups indicated above. 20.32 Exercises. 1. Let P be a geometrical projective module over the geometrical algebra A. Let also P l (resp., P ∧k ) be the l-th symmetric (resp., the k-th exterior) degree of P . Then the elements (pi1 · · · pil ) ⊗A (pj1 ∧ . . . ∧ pjk ), generate the A-module P l ⊗A P ∧k . Show that def P l P ∧k SLm (P) = l+k=m

A

pis , pjr ∈ P.

Cohomology

383

is a cochain complex whose diﬀerentials are the following A-homomorphisms m : P (m−k+1 P ∧k−1 → P (m−k) P ∧k , δm−k,P A

(pi1 · · · pim−k+1 )

A

(pj1 ∧ . . . ∧ pjk−1 )

A

→

m−k+1

s=1

A

(pi1 · · ·p is · · · pim−k+1 )

(pis ∧ pj1 ∧ . . . ∧ pjk ).

Here we can assume, by deﬁnition, that P 0 = P ∧0 = A.

(20.17)

2. Prove that H i (SLm (P)) = 0 if i ≥ 1, and that H 0 (SLm (P)) = A by indicating the analog of the “Poincar´e homotopy” in this case (see the proof of Proposition 20.31). 3. Verify that for the smooth category M over the algebra A, we have m 0 H i (Spenm M (A)) = 0 if m ≥ 1, and H (SpenM (A)) = A. 20.33. Problem: Spencer cohomology of the cross K. In this book, we have repeatedly studied the properties of K. For the cross, the computation of Spencer cohomology reduces to that of the corresponding Spencer δ-cohomology via the short exact sequences (18.11) or the appropriate spectral sequence. We propose the following problem: calculate the geometrical Spencer δ-cohomology of the cross directly, and show that it is nontrivial because K has a singularity, namely its singular point. The solution of this problem, in our opinion, is worthy of publication in a serious journal. 20.34. Adjoint operators. In this section and the next ones, we assume that we are dealing with a diﬀerentially closed category M. (In our exposition, we follow [15].) To simplify the notation, the symbol M in the notation for diﬀunctors and their representing objects will be omitted. Let P be an A-module, and let us consider the sequence of maps d◦

d◦

d◦

(20.18) 0 → Diﬀ > (P, A) −→ Diﬀ > (P, Λ1 ) −→ Diﬀ > (P, Λ2 ) −→ . . . ,

i where, as in Section 9.68, we have Diﬀ > (P, Λi ) = n≥0 Diﬀ > n (P, Λ ) and def

d◦ (Δ) = d ◦ Δ, Δ ∈ Diﬀ > (P, Λi ). According to the deﬁnition of the right A-module structure in Diﬀ > (P, Λi ), the map d◦ is an A-homomorphism and, in view of the equality d ◦ d = 0, the sequence 20.18 is a cochain complex. Denote the cohomology module of this complex at the term Diﬀ > (P, Λn ) by Pn , n 0.

384

Chapter 20

With each diﬀerential operator Δ ∈ Diﬀ > (P, Q), we can associate the following morphism of complexes: ...

d◦

Diﬀ > (Q, Λk−1 )

d◦

Diﬀ > (Q, Λk )

Δ

...

◦

d

d◦

...

(20.19)

Δ

Diﬀ > (P, Λk−1 )

◦

d

Diﬀ > (P, Λk )

d◦

...,

where def Δ(∇) = ∇ ◦ Δ ∈ Diﬀ > (P, Λk ), ∇ ∈ Diﬀ > (Q, Λk ).

(20.20)

From the commutativity of the diagram (20.19) it follows that we have a map of the cohomology of the complex Diﬀ > (Q, Λ∗ ) to the cohomology of the complex Diﬀ > (P, Λ∗ ). n → Pn generated by the operator Deﬁnition. The operator Δ∗n : Q > Δ ∈ Diﬀ (P, Q) is called the (n-th) adjoint operator of the operator Δ. Denote by [X] the cohomology class of the operator X ∈ Diﬀ > (P, Λn ), ◦ d (X) = 0. Then, by deﬁnition, Δ∗n ([X]) = [X ◦ Δ]. n , Pn ). 20.35 Proposition. (1) If Δ ∈ Diﬀ k (P, Q), then Δ∗n ∈ Diﬀ k (Q (2) For any operators Δ ∈ Diﬀ > (P, Q) and ∇ ∈ Diﬀ > (Q, R), the following equality holds (∇ ◦ Δ)∗n = Δ∗n ◦ ∇∗n .

(20.21)

(1) Denoting, as above, the operator of right multiplication by a ∈ A by the symbol a> , we obtain δa (Δ∗n )([X]) = Δ∗n (a> [X]) − a> Δ∗n ([X]) = [X ◦ a ◦ Δ] − [X ◦ Δ ◦ a] = (a ◦ Δ)∗n ([X]) − (Δ ◦ a)∗n ([X]) = −δa (Δ)∗n ([X]).

In other words, δa (Δ∗n ) = −δa (Δ)∗n . Thus, δa0 ,...,ak (Δ∗n ) = δa0 ,...,ak (Δ)∗n = 0. (2) (∇ ◦ Δ)∗n ([X]) = [X ◦ ∇ ◦ Δ] = Δ∗n ([X ◦ ∇]) = Δ∗n (∇∗n ([X])).

Examples. 1. Let a denote an element of the algebra A, and at the same time, the operator of left multiplication by this element in the A-module P . Then a∗n ([∇]) = [∇ ◦ a] = a> [∇], i.e., a∗n = a. 2. To any element p ∈ P , we can assign the operator denoted by the same letter and given by p : A → P, a → pa, a ∈ A. Then p∗n ([∇]) = [∇ ◦ p], n ). Thus there exists a natural pairing p∗n ∈ HomA (Pn , A n , p, p = p∗n ( · , · : P ⊗ Pn → A p), p ∈ Pn .

Cohomology

385

20.36. Integral forms. Let Δ

Δk+1

k . . . −→ P k−1 −→ P k −→ P k+1 −→ . . .

k be a complex of diﬀerential operators. Replacing the modules P k by P n ∗ ∗ ∗ and taking into account the equalities Δk ◦ Δk+1 = (Δk+1 ◦ Δk ) = 0, we obtain the complex ∗

∗

Δk+1 k−1 Δk k+1 ←− Pnk ←− P ←− . . . , . . . ←− P n n

which is called (n-th) dual to the given one. The complex dual to the de Rham complex is called the complex of integral forms and is denoted as follows: δ

δ

δ

δ

δ

0 ←− Σn0 ←− Σn1 ←− . . . ←− Σnk ←− Σnk+1 ←− . . . , i and δ = d∗ . The module Σn = A n is called the (n-th) where Σni = Λ n 0 module of volume forms or the (n-th) Berezinian and is denoted by Bn or simply by B if the value of n is ﬁxed. The diﬀerential operator D : Diﬀ > (Λk ) → Λk , D(∇) = ∇(1) (see Section 15.8) induces a cohomology homomorphism : Bn → H n (Λ∗ ), since to any volume form ω ∈ Bn there corresponds a de Rham cohomology class, namely ω. This is yet another version of the operation of integration. Obviously, [∇] = ∇(1) mod dΛn−1 , ∇ ∈ Diﬀ(Λn ). 20.37 Proposition. δω = 0, where ω ∈ Σ1 . Putting ω = [∇], we obtain δω = [∇ ◦ d]. Therefore, δω = [(∇ ◦ d)(1)] = 0. n , the 20.38 Proposition. For any Δ ∈ Diﬀ(P, Q), p ∈ P and q ∈ Q following equality (“integration by parts formula”) holds

Δ(p), q =

q ). p, Δ∗ (

Let q = [∇], where ∇ : Q → Λn . Then, Δ(p), q = [∇ ◦ Δ(p)] = [∇ ◦ Δ ◦ p] q ). = p, [∇ ◦ Δ] = p, Δ∗ (

(20.22)

Note that in order to ﬁnd the volume form modules Bn , it is necessary to compute the cohomology of the complex (20.18) for P = A, i.e., of the

386

Chapter 20

following complex: d◦

d◦

d◦

d◦

0 → Diﬀ > A −→ Diﬀ > Λ1 −→ . . . −→ Diﬀ > Λn −→ . . . ,

(20.23)

Exercise. Let A = C ∞ (Rn ). Prove that in the category of geometrical A-modules, we have 0, if i = n; Bi = Λn , if i = n. The result of this exercise justiﬁes the term “volume form module” for n and explains why we have called the map Bn → H n (Λ∗ ) Bn = Σn0 = A “integration.” The origins of the term “Berezinian,” will be explained in the next chapter. The free module Bn = Λn is one-dimensional and its generator has the form dx1 ∧ dx2 ∧ . . . ∧ dxn , where x1 , . . . , xn are the coordinates in Rn . When we pass from the coordinates xi to the coordinates vj , the new generator is obtained from the old one via multiplication by the determinant det J( xv ) of the Jacobi matrix J( xv ). Terra incognita. As far as we know, no one has tried to apply the constructions of the algebraic theory of linear diﬀerential operators to p-adic functions (and to p-adic analysis). Apparently, in this direction, one can count on interesting results, which should also have applications to “p-adic mathematical physics.” Besides a simple description of p-adic diﬀerential operators, there is a particular interest in the study of diﬀerential and integral forms, of the abovementioned Spencer complexes, and of the cohomological approach to integration. In particular, this might lead to understand the conceptual nature of the Feynman integral. We hope that our book will attract the attention of researchers to these questions.

21 Diﬀerential Operators over Graded Algebras

Throughout the whole book, commutative algebras and modules over them were in the center of our considerations. In the seventies of the previous century, thanks to the pioneering work of Felix Berezin, it became clear that algebras with “anticommuting variables” are not only interesting per se, but also have important applications to physics and can readily be studied. Such algebras were called graded commutative (or supercommutative). We have already met with some examples of these algebras—ﬁrst of all, the algebras of diﬀerential forms and the de Rham cohomology of commutative algebras. Here we shall expand the list of examples and explain what the theory of diﬀerential operators for such algebras is like. The proof of many statements in this case will be a simple paraphrase of similar statements for commutative algebras; we shall omit these proofs, leaving them to the reader as exercises. The aim of this chapter is not only to give a systematic description of the diﬀerential calculus in the case of commutative graded algebras, but to show its usefulness. 21.1. Deﬁnition. Let G be a commutative (Abelian) semigroup with unit (which may sometimes be a group). An associative algebra with unit A over the ﬁeld k is called G-graded, if it can be represented in the form A = ⊕g∈G Ag , where the Ag satisfy Ag · Aq ⊂ Ag+q for all g, q ∈ G. If a ∈ Ag , then g is called the grading of a and is denoted by ω(a), while a is called a homogeneous element of grading g.

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4 21

387

388

Chapter 21

In what follows, when dealing with elements of graded algebras or with the other graded objects deﬁned below, we shall assume that they are homogeneous unless the converse is explicitly indicated. 21.2. Example. Second-order matrices. Consider the simplest noncommutative algebra—the algebra Mat(A, 2) of second-order matrices over a noncommutative algebra A of zero characteristic. Any such matrix can be uniquely presented as the sum of two matrices of certain special forms: x y a b c d = + . z w −b a d −c Denote the set of matrices of the ﬁrst type by M 0 , and of the second type, by M 1 . Let X1 , X2 ∈ M 0 , Y1 , Y2 ∈ M 1 . It is easy to check that X1 + X2 ∈ M 0 , Y1 + Y2 ∈ M 1 , X1 · X2 ∈ M 0 , Y1 · Y2 ∈ M 0 , X1 · Y1 ∈ M 1 . This means that Mat(A, 2) = M 0 ⊕ M 1 is a Z2 -graded algebra. Note that if A is the ﬁeld of real numbers, then M 0 is isomorphic to the ﬁeld of complex numbers. We can go further: divide pieces and take Z2 ⊕ Z2 for put a M 0,0 = 0 0 M 0,1 = a

each of the components M 0 and M 1 into two the grading semigroup. More precisely, let us 0 0 a , M 1,1 = , a −a 0 a a 0 , M 1,0 = . 0 0 −a

Exercise. Check that Mat(A, 2) = ⊕g∈Z2 ⊕Z2 M g is a (Z2 ⊕ Z2 )-graded algebra . 21.3. Example. Quaternions. The algebra of quaternions H may be deﬁned as the R-algebra generated by the four elements 1, i, j, k, whose products are given by the following table: · 1 i j k

1 1 i j k

i i −1 −k j

j j k −1 −i

k k −j i −1

(21.1)

As in the case of the algebra Mat(A, 2), there are two ways of representing of the set of quaternions as a graded algebra.

Diﬀerential Operators over Graded Algebras

389

In the ﬁrst one, we take Z2 for the grading group G and set H0 = R ⊕ R · i,

H1 = R · j ⊕ R · k.

Then H = H0 ⊕ H1 . This decomposition of the quaternion algebra in a direct sum is compatible with the grading and allows to regard it as a Z2 -graded algebra. (Check this!) In the second version, for the grading semigroup G, we take the direct sum Z2 ⊕ Z2 and deﬁne the direct summands as follows: H(0,0) = R · 1, H(1,0) = R · j,

H(1,1) = R · i, H(0,1) = R · k.

g Then H = g∈Z2 ⊕Z2 H and the multiplication of homogeneous quaternions is compatible with the grading, so that we may regard H as Z2 ⊕ Z2 -graded. (Check this also!) g 21.4. Deﬁnitions. 1. Let A = g∈G A be a G-graded k-algebra. The g q g+q for all left A-module P = g∈G P g is called G-graded, if Ag · P ⊂ P g, q ∈ G. Similarly, the right A-module P = g∈G P is called G-graded if P q · Ag ⊂ P g+q for all g, q ∈ G. If p ∈ P g , then g is called the grading of p and is denoted by ω(p), while p is said to be a homogeneous element of grading g. In what follows, unless indicated otherwise, elements of the graded algebras under consideration will be assumed homogeneous. g 2. Let P = g∈G P be a G-graded left (right) A-module and let the g g family Q ⊂ P , g ∈ G be such that Q = g∈G Qg is again a left (right) G-graded A-module. Modules of this form will be called left (right) graded submodules of the G-graded A-module P. def g g Exercise. Verify that the quotient module P/Q = g∈G P /Q is indeed a left (right) G-graded A-module. 3. If P and Q are G-graded A-modules, then the k-homomorphism φ : P → Q is said to be G-graded of degree g if φ(P g1 ) ⊂ Qg1 +g for any g1 ∈ G. In what follows, speaking of morphisms of graded modules, we shall consider only graded morphisms. Denote the k-module of all k-homomorphisms from P to Q by Homgk (P, Q). We shall denote the operator of multiplication by the element a ∈ A by the same letter a. If g+ω(a) (P, Q). φ ∈ Homgk (P, Q), then, obviously, φ◦a and a◦φ belong to Homk This means that the module Homk (P, Q) = Homgk (P, Q) (21.2) g∈G

possesses natural left as well as right A-module structures. These structures are compatible: a ◦ (φ ◦ b) = (a ◦ φ) ◦ b, so that Homk (P, Q) is supplied with a bimodule structure over A.

390

Chapter 21

4. The tensor product over k of a left G-graded A-module P and right G-graded A-module Q is a G ⊕ G-graded k-module: P g ⊗k Qr . P ⊗k Q = g,r∈G

The multiplication operation of elements of the tensor product P ⊗k Q by def elements of the algebra A from the left a(p ⊗k q) = ap ⊗k q and from the def

right (p ⊗k q)a = p ⊗k qa are compatible, so that P ⊗k Q is a bimodule over A. It is convenient to think of it as being G-graded: def P ⊗k Q = (P ⊗k Q)g , (P ⊗k Q)g = P r ⊗k Qt . r+t=g

g∈G

21.5. The parity form. Now suppose that the Abelian grading semigroup G is supplied with a parity form, i.e., a Z-bilinear symmetric map ·, · : G×G → Z2 . Here we assume that char k = 2. An important and often arising particular case of a parity form is the form that can be constructed from the parity homomorphism ρ : G → Z2 , g1 , g2 ρ = ρ(g1 )·ρ(g2 ). Parity forms constructed via the parity homomorphism will be called decomposable. If the grading semigroup G possesses a parity form ·, · and U, V are arbitrary G-graded objects, then to any homogeneous elements u ∈ U, v ∈ V, we can assign the element u, v ∈ Z2 by setting def

u, v = ω(u), ω(v) . If the parity form was constructed from the corresponding homomorphism, then it is sometimes convenient to use the notation |u| instead of ρ(ω(u)). In that case u, v = |u| · |v|. 21.6. Deﬁnition. Let a parity form ·, · be given. A G-graded algebra A is said to be graded commutative (we stress—not just commutative, but graded commutative), if ab = (−1)a,b ba for any homogeneous elements a, b ∈ A. A homogeneous element a ∈ A is called even, if a, b = 0 for all homogeneous b ∈ A, and odd, if a, a = 1. The fact that an element a is odd means that a2 = 0. If the parity form ·, · is decomposable, then the evenness of an element a ∈ A is equivalent to the condition |a| = 0, and its oddness, to |a| = 1, while the equality ab = −ba holds iﬀ both a and b are odd. In the general case, there may exist elements that are neither odd nor even (examples will appear below). If P is supplied with both a left and a right module structure over a graded commutative algebra A, then these structures are called compatible if ap = (−1)a,p pa for any homogeneous elements a ∈ A, p ∈ P . In that case P is said to be a two-sided module over A. Of course, to any right def A-module, we can assign a left one by setting ap = (−1)a,p pa for any

Diﬀerential Operators over Graded Algebras

391

a ∈ A, p ∈ P, and conversely. In what follows, unless the contrary is explicitly stated, all the modules are assumed two-sided. 21.7. Deﬁnition. Let a G-graded algebra A be commutative, and P, Q be G-graded modules over A. A k-homomorphism φ : P → Q with grading g = ω(φ) is said to be an A-homomorphism (or a homomorphism over A), if for any a ∈ A, p ∈ P, we have φ(ap) = (−1)a,φ aφ(p).

(21.3)

The set of all A-homomorphisms with ﬁxed grading g ∈ G from P to Q possesses a natural A0 -module structure (A0 being the 0-component of the graded algebra A) that we denote by HomgA (P, Q). Recall that the operator of multiplication by an element a ∈ A is denoted by the same letter a. If φ ∈ HomgA (P, Q), then, obviously, φ ◦ a and a ◦ φ will belong to g+ω(a) (P, Q). This means that the k-module HomA HomA (P, Q) = HomgA (P, Q) (21.4) g∈G

possesses natural left and right A-module structures. From condition (21.3), it follows that these two structures coincide and the module HomA (P, Q) of homomorphisms from P to Q is a two-sided A-module. For any G-graded A-module P, we obviously have HomA (A, P) = P.

(21.5)

21.8. Tensor product of graded modules. Let P, Q be G-graded A-modules. In the direct product P × Q, let us consider the submodule U generated by elements of the form (ap1 + bp2 , q1 ) − (ap1 , q1 ) − (bp2 , q1 ), (p1 , aq1 + bq2 ) − (−1)a,p1 (ap1 , q1 ) − (−1)b,p1 (bp1 , q2 ), a, b ∈ A, p1 , p2 ∈ P, q1 , q2 ∈ Q. The quotient module (P × Q)/U is called the tensor product of P and Q; it is denoted by P ⊗A Q, while the image of (p, q) under the corresponding factorization homomorphism is denoted by p ⊗ q. def Deﬁne the grading of an element p⊗q by putting ω(p⊗q) = ω(p)+ω(q). Then P ⊗A Q = (P ⊗A Q)g , g∈G

where (P ⊗A Q)g = {p ⊗ q | ω(p) + ω(q) = g}. The formula a(p ⊗ q) = ap ⊗ q supplies P ⊗A Q with the structure of a left A-module. Since, obviously, Ar · (P ⊗A Q)g ⊂ (P ⊗A Q)g+r ,

392

Chapter 21

it follows that P ⊗A Q is a G-graded A-module. For any G-graded A-module P, we have the following obvious isomorphism: A ⊗A P = P.

(21.6)

Exercises. 1. Modify the above construction of tensor product in the case when P is a left, while Q is a right G-graded A-module. 2. A map G of graded A-modules α : P × Q → T is called bilinear in the grading g if for any a1 , a2 , b1 , b2 ∈ A, p1 , p2 ∈ P, q1 , q2 ∈ Q, we have α(a1 p1 + a2 p2 , b1 q1 + b2 q2 ) = (−1)b1 ,p1 +a1 ,α+b1 ,α a1 b1 α(p1 , q1 ) + (−1)b2 ,p1 +a1 ,α+b2 ,α a1 b2 α(p1 , q2 ) + (−1)b1 ,p2 +a2 ,α+b1 ,α a2 b1 α(p2 , q1 ) + (−1)b2 ,p2 +a2 ,α+b2 ,α a2 b1 α(p2 , q2 ), where, as before, in the parity form, we write α instead of g. Prove that for bilinear maps P × Q → T , where T is an arbitrary G-graded A-module, the tensor product P ⊗A Q is a universal object, i.e., the map α may be presented as the composition of the factorization map P × Q → (P × Q)/U = P ⊗A Q and the homomorphism hα : P ⊗A Q → T . 3. Consider the map η : HomA (P ⊗k Q, U) → Homk (P, HomA (Q, U))

(21.7)

deﬁned by the following formula: η(h)(p)(q) = h(p ⊗k q),

h ∈ HomA (P ⊗k Q, U).

Prove that η is an isomorphism. 21.9. Before passing to examples, note that if the role of the graded semigroup is played by Z or Z+ , then there exist only two parity homomorphisms, namely, the trivial one ρ0 , ρ0 (n) = 0 for any n, and the homomorphism ρ2 reduction modulo 2: ρ2 (n) = n mod 2 for any n. Now let G and R be grading semigroups with parity forms ·, · G and ·, · R , respectively, then on G ⊕ R there exist at least three parity forms: (g1 , r1 ), (g2 , r2 ) 1 = g1 , g2 G , (g1 , r1 ), (g2 , r2 ) 2 = r1 , r2 R ,

g1 , g2 ∈ G, r1 , r2 ∈ R,

(g1 , r1 ), (g2 , r2 ) + = g1 , g2 G + r1 , r2 R . Of course, there may be other ones, and this is a good thing, as we shall see in what follows.

Diﬀerential Operators over Graded Algebras

393

21.10. Auxiliary constructions. Let us introduce some constructions, which we shall need to study the algebras Mat(A, 2n ), where A is a commutative algebra. For an arbitrary matrix X ∈ Mat(A, m), we deﬁne the matrices X i,j ∈ Mat(A, 2m), i, j ∈ Z2 , by putting X 0 0 X , X 1,1 = , X 0,0 = 0 X −X 0 0 X X 0 , X 1,0 = . X 0,1 = X 0 0 −X Let X, Y ∈ Mat(A, m) be arbitrary matrices. For the matrices X k,l , Y i,j ∈ Mat(A, 2m), let us write out the multiplication table: · 0,0

X X 1,1 X 1,0 X 0,1

Y 0,0 (XY )0,0 (XY )1,1 (XY )1,0 (XY )0,1

Y 1,1 (XY )1,1 −(XY )0,0 (XY )0,1 −(XY )1,0

Y 1,0 (XY )1,0 −(XY )0,1 0,0 (XY ) −(XY )1,1

Y 0,1 (XY )0,1 (XY )1,0 (XY )1,1 (XY )0,0

(21.8)

In other words, X i,j Y k,l = (−1)jk (XY )i+k,j+l .

(21.9)

21.11. Second-order matrices. Above (see Section 21.2), the algebra Mat(k, 2) of second-order matrices was presented as a Z2 -graded algebra: Mat(k, 2) = M 0 ⊕ M 2 , where a b c d 0 1 , M = . M = −b a d −c It is easy to check that the matrices from M 0 do not commute and do not anticommute with each other, nor with matrices from M1 , while the matrices from M1 do not anticommute with each other. Thus, with respect to this grading, Mat(k, 2) is not a Z2 -graded commutative algebra. Now let us regard Mat(A, 2) as a (Z2 ⊕ Z2 )-graded algebra and choose the following generators in it: 1 0 0 1 E = E 0,0 = , E 1,1 = , 0 1 −1 0 0 1 1 0 , E 1,0 = , E 0,1 = 1 0 0 −1 The above matrices obviously form a basis of Mat(A, 2) that we denote by E2 . Note that these matrices may be obtained from the construction of the previous section by putting X = E = 1 ∈ A = Mat(A, 2). Their multiplication table can be easily obtained from (21.8) by replacing the matrices X and Y by the matrix E = 1 ∈ A = Mat(A, 2)

394

Chapter 21

· 0,0

E E 1,1 E 1,0 E 0,1

E 0,0 E 0,0 E 1,1 E 1,0 E 0,1

E 1,1 E 1,1 −E 0,0 E 0,1 −E 1,0

E 1,0 E 1,0 −E 0,1 E 0,0 −E 1,1

E 0,1 E 0,1 E 1,0 E 1,1 E 0,0

(21.10)

From this we can see that all the squares of the generators are equal to ±E 0,0 , while E 0,0 commutes with the other generators, which, in turn, anticommute with each other. Comparing this table with the similar table (21.1) for quaternions, we see that they look very much alike. The main diﬀerence is that there are three units on the main diagonal of the table (21.10). The formula 21.9 in the case under consideration acquires the form: E i,j E k,l = (−1)jk (EE)i+k,j+l = (−1)jk E i+k,j+l .

(21.11)

Therefore, E i,j E k,l = (−1)jk+il E k,l E i,j .

(21.12)

On Z2 ⊕ Z2 , let us deﬁne the bilinear form ·, · 0,1 by setting def

(i, j), (k, l) 0,1 = jk + il,

(21.13)

E i,j E k,l = (−1)(i,j),(k,l)0,1 E k,l E i,j .

(21.14)

ﬁnally obtaining

Thus, the algebra Mat(k, 2) =

Mg

g∈Z2 ⊕Z2

is a graded commutative algebra with parity form ·, · 0,1 . We use the index 0, 1 in the notation of the parity form because the latter is given by the matrix E 0,1 . Indeed, let (i, j), (k, l) ∈ Z2 ⊕ Z2 , then 0 1 k k = (j i) = jk + il. (i, j), (k, l) 0,1 = (i, j) 1 0 l l Note in conclusion of this section that here elements of the form kE 0,0 , k ∈ A are even, while the other elements are neither even, nor odd. 21.12. Quaternions. The situation with the quaternion algebra is just the same as with the matrix algebra Mat(A, 2). Namely, the algebra Hg H= g∈Z2 ⊕Z2

is a graded commutative algebra, see [11], with parity form ·, · 0,1 . The veriﬁcation of this fact is left to the reader as an exercise.

Diﬀerential Operators over Graded Algebras

395

In the case under consideration, elements of H0,0 = R are even, elements of H1,1 = R · i, H1,0 = R · j, H0,1 = R · k are neither even, nor odd. 21.13. Fourth-order matrices. Let E i,j be an arbitrary basis matrix E i,j ∈ E2 of the algebra Mat(A, 2). Put i,j E 0 E i,j 0 i,j,1,1 = , E i,j,0,0 = , E −E i,j 0 0 E i,j i,j 0 E i,j E 0 i,j,1,0 = , E E i,j,0,1 = . E i,j 0 0 −E i,j As can easily be seen, matrices of the form E i,j,k,l constitute a basis 4 -graded. of the algebra Mat(k, 4), which we can now regard as Z 2 s=1 Denote this basis by E4 . It follows from formula 21.9 that E i1 ,j1 ,i2 ,j2 E k1 ,l1 ,k2 ,l2 = (−1)j2 k2 (E i1 ,j1 E k1 ,l1 )i2 +k2 ,j2 +l2

i2 +k2 ,j2 +l2 = (−1)j2 k2 (−1)j1 k1 E i1 +k1 ,j1 +l1 = (−1)j2 k2 +j1 k1 E i1 +k1 ,j1 +l1 ,i2 +k2 ,j2 +l2 . Similarly, E k1 ,l1 ,k2 ,l2 E i1 ,j1 ,i2 ,j2 = (−1)i2 l2 +i1 l1 E i1 +k1 ,j1 +l1 ,i2 +k2 ,j2 +l2 . From this we ﬁnally obtain E i1 ,j1 ,i2 ,j2 E k1 ,l1 ,k2 ,l2 = (−1)j1 k1 +i1 l1 +j2 k2 +i2 l2 E k1 ,l1 ,k2 ,l2 E i1 ,j1 ,i2 ,j2 . (21.15) 4 Now let us introduce the bilinear form ·, · 0,1 on s=1 Z2 by setting def

(i1 , j1 , i2 , j2 ), (k1 , l1 , k2 , l2 ) 0,1 = j1 k1 + i1 l1 + j2 k2 + i2 l2 . It is easily checked that the matrix of ⎛ 0 ⎜1 0,1,0,0 ⎜ E =⎝ 0 0

(21.16)

this bilinear form is ⎞ 1 0 0 0 0 0⎟ ⎟ 0 0 1⎠ 0 1 0

Now we can say that for any two homogeneous elements U, V of the

4 s=1 Z2 -graded algebra Mat(k, 4), we have the equality U V = (−1)U,V V U. (21.17)

4 and, therefore, the s=1 Z2 -graded algebra Mat(A, 4) is a graded commutative algebra with parity form ·, · 0,1 . Exercise. Show that, for any n, the matrix algebra Mat(A, 2n ) is a graded commutative algebra. (In [12] it is shown that a matrix algebra Mat(A, m) is graded commutative only if m = 2n .)

396

Chapter 21

21.14. Dioles. Let P be a module over a commutative algebra A. The module P may be regarded as an A-algebra by setting pq = 0 for any p, q ∈ P . Now let us deﬁne the Z-graded algebra A = A(P ) as ⎧ 0, j < 0, ⎪ ⎪ ⎪ ⎨A, j = 0, Aj = (21.18) ⎪ P, j = 1, ⎪ ⎪ ⎩ 0, j > 1. We take the trivial form for the parity form. Such Z-graded commutative forms will be called dioles. In what follows, we shall sometimes write A instead of A0 and P instead of A1 . Note that any Z-graded commutative algebra for which only A0 and A1 are nontrivial is a diole. The thoughtful reader has already noticed that, from the geometrical point of view, a diole appears when we consider the pair (manifold M , ﬁbration π over M ), then take the algebra of smooth functions C ∞ (M ) for the algebra A, and the module of sections Γ(π) for the module P . 21.15. More examples. Trivially graded algebra. Any commutative algebra A may be regarded as graded w.r.t. any Abelian semigroup G if we assume that all its components, except one, are equal to zero. Complex numbers. The algebra of complex numbers C = R ⊕ Ri is Z2 -graded commutative with trivial parity homomorphism ρ0 . Algebras related to diﬀunctors. Let A be a commutative algebra and M be a diﬀerentially closed category of A-modules. Then the algebra of symbols ∞ Si (A), S∗ (A) = i=0

(cm. ii 10.1) and the algebra of cosymbols ∞ CsmlM (A) = CsmliM (A) i=0

(see Section 17.29) are commutative Z-graded algebras with trivial parity homomorphism ρ0 . The algebra of diﬀerential forms ∞ Λ∗M = Λ∗M (A) = ΛkM (A), i=0

(see Section 17.29) and the de Rham cohomology algebra ∞ ∗ i HdR (M) = HdR (M) i=0

Diﬀerential Operators over Graded Algebras

397

(see Section 17.29) are commutative Z-graded algebras with parity homomorphism ρ2 (see Section 21.9). 21.16. Grassmann algebra. Let an n-dimensional vector space over V be k. The exterior algebra Λ(V ) = V ∧k has a natural Z-graded commutative algebra structure. It is called the Grassmann algebra and denoted from now on by F (for “Fermionic”). For the category of modules over F, we choose the category of ﬁnitely generated ones, which in this case amounts to saying that they are ﬁnite-dimensional over k. Fix a basis B = {ξ1 , . . . , ξn } of V ; then the algebra F is generated by the elements ξ1 , . . . , ξn of degree one and is subject only to the relations ξi · ξj = −ξj · ξi . This means that F is free as a Z-graded algebra. If σ ⊂ {1, . . . , n} is an ordered subset of length r, σ = {σ1 , . . . , σr }, we write ξσ = ξσ1 · · · ξσr and set ξ∅ = 1. Then every element of ω ∈ F has a unique representation ασ ξσ , ασ ∈ k ω= σ

g 21.17. Graded Lie algebras. Let V = g∈G V be a G-graded vector spaceover the ﬁeld k, let ·, · be a parity form on G, and let [ · , · ] : V k V → V be a bilinear operation on V with grading g ∈ G, i.e., an operation such that [V g1 , V g2 ] ⊂ V g1 +g2 +g

∀g1 , g2 ∈ G.

Then V is called a G-graded Lie algebra if the bracket [·, ·] is anti-symmetric [X, Y ] = −(−1)X,Y +g,g [Y, X] ∀X, Y ∈ V (21.19) and satisﬁes the Jacobi identity: [X, [Y, Z]] = [[X, Y ], Z] + (−1)X,Y +g,Y [Y, [X, Z]];

(21.20)

here, as before, g, Y means g, ω(Y ) . (This is the Leibnitz formula for the operator [X, · ].) For an even bracket [·, ·], the formulas (21.19) and (21.20) appear to be more usual. The last one can be modiﬁed by simple manipulations to the following “symmetric” form: (−1)X,Z [X, [Y, Z]] + (−1)Y,X [Y, [Z, X]] + (−1)Z,Y [Z, [X, Y ]] = 0.

(21.21)

It follows from the deﬁnitions that, in particular, for a G-graded commutative algebra A, the component A0 will be an ordinary Lie algebra over k. Each A-homomorphism φ : P → Q of degree r ∈ G may be understood as a family of A0 -homomorphisms φg : P g → Qg+r , g ∈ G such that φω(p)+ω(a) (ap) = (−1)a,r aφω(p) (p). 21.18. Poisson dioles. Just as we transformed an arbitrary A-module P into the Z-graded algebra A(P ) above (see Section 21.14), we can trans-

398

Chapter 21

form any commutative algebra into a Z-graded Lie algebra. To do this, we consider the diole A = A(D(A)) and deﬁne the Lie bracket [·, ·] on A as the extension of the commutator of vector ﬁelds to D(A) by setting [X, f ] = [f, X] = −X(f ),

[f, g] = 0 ∀f, g ∈ A, ∀X ∈ D(A).

The bracket on A deﬁned in this way has the grading −1, it is antisymmetric in the sense of (21.19) and satisﬁes the Jacobi identity (21.20), i.e., for any homogeneous elements x, y, z of the algebra A, we have: [x, [y, z]] = [[x, y], z] + (−1)(|x|+1)·|y| [y, [x, z]].

(21.22)

Indeed, if x, y, z are vector ﬁelds, then the sign in front of the last summand disappears and the above equality becomes the Jacobi identity for vector ﬁelds. If at least two elements of the three belong to A, then each summand of this equality vanishes. And, ﬁnally, when exactly one element belongs to A, while the two others are vector ﬁelds, the equality is a simple consequence of deﬁnitions. Indeed, let x = f ∈ A and y = Y, z = Z be vector ﬁelds. Let us compute the right-hand side of (21.22): [[f, Y ], Z] + (−1)(|f |+1)·|Y | [Y, [f, Z]] = [−Y (f ), Z] − [Y, −Z(f )] = Z(Y (f )) − Y (Z(f )) = −[Y, Z](f ) = [f, [Y, Z]]. Note further that the bracket is a biderivation (actually, it is graded derivation, see Section 21.23 below) with respect to A as well: [X, f g] = −X(f g) = −f X(g) − gX(f ) = f [X, g] + g[X, f ], [X, f Y ] = X(f )Y + f [X, Y ] = −[X, f ]Y + f [X, Y ]. A comment concerning the sign: it would seem that there should not be a minus sign in front of [X, f ]Y . Actually, the minus sign is correct. The Lie bracket is odd, so is the vector ﬁeld Y , hence the sign must change when we take the vector ﬁeld out of the brackets. The same construction was used for Lie algebroids, which have become a popular object of study in recent years, especially among mathematical physicists. Let a Lie algebroid be given, i.e., a vector bundle π : E → M over a manifold M together with a Lie bracket over the module Γ(π) of sections of the bundle and a homomorphism (the so-called anchor) ρ : Γ(π) → D(M ), given by S → Sρ such that for all sections S, T ∈ Γ(π) and f ∈ C ∞ M , we have [S, f T ] = Sρ (f )T + f [S, T ]. Just as above, deﬁne a Z-graded algebra A by putting A0 = C ∞ (M ), A1 = Γ(π), Ai = 0 for i = 0, 1 and assuming that ST = 0 for all S, T ∈ A1 . Deﬁne a Lie bracket [·, ·] on A as the extension of the given Lie bracket on Γ(π) by setting [S, f ] = [f, S] = −Sρ (f ),

[f, g] = 0,

f, g ∈ A, X ∈ D(A).

Diﬀerential Operators over Graded Algebras

399

Such a bracket on A has grading −1, is antisymmetric, and, for the same reasons as above, satisﬁes the Jacobi identity. Deﬁnition. A diole, supplied with a Lie algebra structure, with grading −1, such that for any elements S, T ∈ A1 and f ∈ A0 , we have [S, f T ] = −[S, f ]T + f [S, T ]. is called a Poisson diole. A Poisson diole will be called geometrical if A0 = C ∞ (M ) for some smooth manifold M and A1 is a projective C ∞ (M )-module. Obviously, any geometrical Poisson diole can be obtained from some Lie algebroid by means of the above construction, and vice versa. The deﬁnition of Lie algebroids seems simple enough, but this is not the case of the deﬁnitions of their morphisms. We will not deﬁne them here, but the reader will understand how complicated the deﬁnition must be, having in mind the completely diﬀerent nature of the objects that must enter into it and the various ways in which these objects behave under maps of the underlying manifold. Even the fact that the composition of morphisms of Lie algebroids is again a Lie algebroid is a suﬃciently cumbersome theorem. The situation with Poisson dioles is much simpler—their morphisms are simply homomorphisms of graded algebras that preserve the Lie bracket and the fact that the composition of morphisms is a morphism is obvious. In this situation, it is natural to restrict ourselves to the case of a ﬁxed base manifold M , which we need not do for Poisson dioles. 21.19. Main notations. Unfortunately, at this point we will have to introdef duce some rather cumbersome notation. We put I(n) = (1, . . . , n), by the letters I, J we will always denote ordered multiindices I = (i1 , . . . , ik ), 1 i1 < i2 < . . . < ik n, J = (j1 , . . . , jl ), 1 j1 < j2 < . . . < jl n. Then we put |I| = k, |J| = l. If I and J satisfy ir = js for all r k, s l, def

obtaining a nonordered multiindex (I, J) = (i1 , . . . , ik , j1 , . . . , jl ). The sum I + J will be deﬁned as the multiindex obtained by ordering the multiindex (I, J). Let A be a G-graded commutative algebra. Then we say that an l-linear map L : A × · · · × A → A is graded of degree g, if for any g1 , . . . , gl , we have L(Ag1 × · · · × Agl ) ⊆ Ag1 +···+gl +g , g1 , . . . , gl , g ∈ G. The degree g of a graded map L will be denoted by ω(L). Together with the degree, to any map L we shall assign its total degree deﬁned by the def formula ω(L) = (l − 1, ω(L)). The vector space of all polylinear graded maps is graded with respect to the semigroup Z × G. Let a1 , . . . , an ∈ A be homogeneous elements and let I = (i1 , . . . , in ) be an ordered multiindex; then the ordered set ai1 , . . . , ain will be denoted by

400

Chapter 21

a(I). Let σi = (1, . . . , i−1, i+1, i, . . . , n) be a transposition. Deﬁne the parity of a permutation a(σi ) = (a1 , . . . , ai−1 , ai+1 , ai , . . . , an ) as ai , ai+1 + 1. This parity will be denoted by |a(σi )|. An l-linear graded map L from A to A is called skew-symmetric if for any transposition σi L(a(I(n) )) = (−1)|a(σi )| L(a(σi )) = −(−1)ai ,ai+1 L(a(σi )),

(21.23)

and symmetric if L(a(I(n) )) = (−1)ai ,ai+1 L(a(σi )).

(21.24)

Denote by Altnk A the set of skew-symmetric n-linear maps from A to A and by Symnk A the set of symmetric maps from A to A. Put ∞ n-linear ∞ ∗ n ∗ n Alt A, Sym A = Sym A. Obviously, the vector Altk A = k k k n=1 n=1 spaces Alt∗k A and Sym∗k A are graded w.r.t. the group Z × G. For any element L ∈ Altnk A and an arbitrary permutation of n elements σ, we obviously have, L(a(I(n) )) = (−1)r L(a(σ)), where r is some element of Z2 . It is easy to see that it depends only on the graded permutation a(σ) and does not depend on L. We will call it the parity of the graded permutation a(σ) and denote it by |a(σ)|. In practice, the parity of any graded permutation can be computed by arbitrarily presenting it in the form of a composition of transpositions a(σi ). In particular, let I, |I| = n and J, |J| = l be such that the multiindex (I, J) is deﬁned; then, as can easily be seen, we have |a(I, J)| = |a(J, I)| + nl + ω(a(I)), ω(a(J)) (21.25) = |a(J, I)| + nl + a(I), a(J) , where ω(a(I)) = i∈I ω(ai ), ω(a(J)) = j∈J ω(aj ). Similarly, for any element L ∈ Symnk A¯ and an arbitrary permutation of n elements σ, we have L(a(I(n) )) = (−1)r L(a(σ)), where r is some element of Z2 that depends only on the graded permutation a(σ) and does not depend on L. We will call L(a(I(n) )) the symmetric parity (or s-parity) of the graded permutation a(σ) and denote it by |a(σ)|s . In practice, the symmetric parity of any graded permutation may be calculated by arbitrarily presenting the permutation as the composition of transpositions a(σi ). In particular, let I, |I| = n and J, |J| = l be such that the multiindex (I, J) is deﬁned; then |a(I, J)|s = |a(J, I)|s + a(I), a(J) .

(21.26)

21.20. Diﬀerential operators in graded commutative algebras. Now that we have collected enough examples of graded objects, we can pass to the main topic of this chapter—the diﬀerential calculus over graded objects.

Diﬀerential Operators over Graded Algebras

401

Let A be a G-graded commutative algebra over k, P and Q be G-graded A-modules. To each element a ∈ A, we assign the operators la , ra δa : Homk (P, Q) → Homk (P, Q), deﬁned as follows: la (ϕ)(p) = a ϕ(p), ϕ,a

ra (ϕ)(p) = (−1)

ϕ ∈ Homk (P, Q),

p ∈ P,

ϕ(ap),

δa (ϕ) = ra (ϕ) − la (ϕ). (Below, we will often use the old notation a< and a> instead of la and ra , respectively.) Having modiﬁed the deﬁnition of these δa , cf. Section 9.66, to ﬁt the case of graded algebras, we can easily carry over to this case the whole theory of k-linear diﬀerential operators. We shall now brieﬂy repeat the outline of this theory, leaving the proofs as exercises for the reader. Thus, the operators la and ra of left and right multiplication of elements of the k-module Homk (P, Q) by elements of the algebra A obviously commute. This allows us to deﬁne, in Homk (P, Q), the structure of a bimodule over the algebra A. As before, we shall write ϕ ◦ a instead of ra (ϕ) and a ◦ ϕ instead of la (ϕ), ϕ ∈ Homk (P, Q). Let a0 , . . . , as ∈ A; then we put δa0 ,...,as = δa0 ◦ . . . ◦ δas . Deﬁnition. An element Δ ∈ Homk (P, Q) will be called a k-linear differential operator (DO) of order s over A if, for any tuple of elements a0 , a1 , . . . , as ∈ A, δa0 ,...,as (Δ) = 0. The set of all DO’s of order s from P to Q is stable with respect to the left as well as to the right multiplication by elements of the algebra A and is therefore supplied with two natural G-graded A-module structures. The G-graded A-bimodule deﬁned by these two commuting structures will be denoted by Diﬀ ♦ s (P, Q). Obviously, > Diﬀ < 0 (P, Q) = Diﬀ 0 (P, Q) = HomA (P, Q)

and for any s 0, the module Diﬀ ♦ s (PQ) is a submodule of the A-bimodule Homk (P, Q). For s < 0, we put Diﬀ ♦ s (P, Q) = 0. We also

♦ write Diﬀ ♦ s (A, Q) = Diﬀ s Q. Whenever no misunderstanding can arise, we will simply write Diﬀ s (P, Q) instead of Diﬀ < s (P, Q). def

Exercises.

1. Verify that for the operator δa : Homk (P, Q) → Homk (P, Q),

402

Chapter 21

where a ∈ A, P, Q are A-modules, the following version of the Leibnitz formula holds: δab = a< δb + δa ◦ b> 2. Prove that for all ai ∈ A, p ∈ P and Δ ∈ Homk (P, Q), (−1)|I|+a(I),a(J) aI Δ(aJ p). δaI(n) (Δ)(p) =

(21.27)

(21.28)

I+J=I(n)

The last formula shows that for Δ ∈ Diﬀ n (P, Q), we have the graded analog of formula 9.25 (n) Δ(aI p) = − (−1)|I|+a(I),a(J) aI Δ(aJ p). (21.29) I+J=I(n) , |I|>0

21.21. Remark. This formula shows that if the algebra A is ﬁnitely generated, say by a1 , . . . , an , and the module P is ﬁnitely generated by p1 , . . . , pm , then any diﬀerential operator Δ : P → Q of order ≤ k is uniquely determined by its value on a ﬁnite number of elements of the form aI pi , where aI is the product of at most k generators of the algebra. Moreover if the algebra and the module are free over their generators, these values can be speciﬁed arbitrarily. This gives a method for computing diﬀerential operators in practice. Exercise. Consider the following maps i> and i< , which, as maps of sets, are identities > i> : Diﬀ < l (P, Q) → Diﬀ l (P, Q),

i> (f ) = f,

< i< : Diﬀ > l (P, Q) → Diﬀ l (P, Q),

i< (f ) = f.

Prove that they are diﬀerential operators of order l. Let Δ : P → Q be a DO of order n and of grading g, and let its r-th component be Δr : P r → Qr+g . Obviously, Δr is a DO of order n in the category of A0 -modules . If Δ ∈ Diﬀ ♦ n (P, Q), then, according to the deﬁnition, for any n-tuple of elements a1 , . . . , an ∈ A, we have δa1 ,...,an (Δ) ∈ HomA (P, Q). If P = Q = A, then HomA (A, A) = A and δa1 ,...,an (Δ) can be regarded as an element of the algebra A. In other words, the assignment (a1 , . . . , an ) → δa1 ,...,an (Δ) is an n-linear map from A to A. Since, as can be immediately checked, δa,b = (−1)ab δb,a , it follows that this map is symmetric. Let us put def

[a1 , . . . , an ]Δ = δa1 ,...,an (Δ)

(21.30)

Diﬀerential Operators over Graded Algebras

403

The natural embedding of A-modules Diﬀ k−1 A ⊂ Diﬀ k A allows us to deﬁne the G-graded quotient module def

Sk (A) = Diﬀ k A/ Diﬀ k−1 A, which is called the module of symbols of order k (or k-symbols). The coset of the operator Δ ∈ Diﬀ k A modulo Diﬀ k−1 A will be denoted by smblk Δ. We deﬁne the algebra of symbols of the algebra A by putting S∗ (A) =

∞

Sn (A).

n=0

The multiplication operation in S∗ (A) is induced by the composition of diﬀerential operators. More precisely, for elements smbll Δ ∈ Sl (A) and smblk ∇ ∈ Sk (A), we write by deﬁnition def

smbll Δ · smblk ∇ = smblk+l (Δ ◦ ∇) ∈ Sl+k (A). This operation is well deﬁned, since it does not depend on the choice of representatives from the classes smbll Δ and smblk ∇. Indeed, since, say, smbll Δ = smbll Δ , it follows that Δ − Δ ∈ Diﬀ l−1 A and, therefore, (Δ − Δ ) ◦ ∇ ∈ Diﬀ l+k−1 A. The modules Sn (A) are graded by elements of the semigroup G, and so the algebra S∗ (A) is Z ⊕ G graded. We put ω(smbll Δ) = ω(Δ). 21.22 Proposition. The algebra S∗ (A) is graded commutative: smbll Δ · smblk ∇ = (−1)smbll Δ,smbll ∇ smblk ∇ · smbll Δ.

If Δ ∈ Diﬀ n−1 A, then [a1 , . . . , an ]Δ = 0 for any a1 , . . . , an ∈ A. This means that the map [ ]Δ is determined not by the operator Δ itself, but by its symbol, and, therefore, to each symbol ς ∈ Sn we can assign the symmetric polylinear map [·, . . . , ·]ς : A × A × . . . × A → A. 21.23. Derivations. As above, we assume that P is a two-sided G-graded A-module. Deﬁnition. The map Δ : A → P is called a derivation with values in the A-module P if it satisﬁes the “graded” Leibnitz rule Δ(ab) = Δ(a)b + (−1)a,Δ aΔ(b) (21.31) The set of all derivations from A to P is denoted by D(P). This set possesses a natural left A-module structure. If Δ ∈ Diﬀ 1 P, then the map ¯ : A → P, Δ(a) ¯ Δ = Δ(a) − (−1)a,Δ aΔ(1) is a derivation. As can be easily seen, ¯ Δ(a) = (δa (Δ))(1).

(21.32)

404

Chapter 21

Suppose that Δ : A → Q is a derivation of grading g and Δr : Ar → Qr+g is its r-th component. Obviously, in the category of A0 -modules, the map Δ0 : A0 → Qg is a derivation, while Δr : Ar → Qr+g , r = 0, are DO’s of ﬁrst order. If a ∈ Ar , b ∈ Ag , then (21.33) Δr+g (ab) = Δr (a)b + (−1)a,Δ aΔg (b) 21.24. Diﬀerential calculus over Grassmann algebras. In this section, we shall see in detail what the calculus over Grassmann algebras F (see Section 21.16) looks like. We start by describing the modules of diﬀerential operators, the ﬁrst observation being that any k-linear map between F-modules is actually a diﬀerential operator of order ≤ 2n + 1. Proposition. Let P, Q be graded modules over the Grassmann algebra F, then Diﬀ(P, Q) = Diﬀ 2n+1 (P, Q) = Homk (P, Q) We need only to check that Homk (P, Q) ⊂ Diﬀ 2n+1 (P, Q), i.e., that δv1 ,...,v2n+1 ϕ = 0 for all v1 , . . . , v2n+1 ∈ F. The relation 21.27 shows that to this end it suﬃces to consider the case when the vi are generators of V . In view of 21.28, we have (−1)|I|+v(I),v(J) v I Δ(v J u), u ∈ F. δv1 ,...,v2n+1 ϕ(u) = δvI 2n+1 ϕ(u) = I+J=I(2n+1)

Now we see that each term of the right-hand side of the last equation contains a product of at least n + 1 elements vi , which is always zero. 21.25. Remark. Since we are considering ﬁnitely generated modules, it follows that HomZk (P, Q) = Homk (P, Q) and in particular, since HomZk 2 (P, Q) = Homk (P, Q), we obtain the equality Diﬀ ZF = Diﬀ ZF2 , i.e., here the notions of Z- or Z2 -graded diﬀerential operator coincide. In the case of diﬀerential operators from F to any A-module V , we can def say more. Let V ◦ = Homk (V, k). Proposition. There is a canonical identiﬁcation of k-vector spaces

l l ◦ Diﬀ k Q = Homk Λ V, Q = Λ V ⊗k Q l≤k

l≤k

which isgiven by restricting the diﬀerential operator Δ : F → Q of order ≤ k to l≤k Λl V ⊂ F. Note that Λl V = F/ Λl V, l≤k

l>k

Diﬀerential Operators over Graded Algebras

405

l and therefore l≤k Λ V has a natural F-module structure (the multiplication is just truncated ). With this identiﬁcation, the above isomorphism becomes an isomorphism of bimodules. This is a direct consequence of Remark 21.21. In particular, we obtain:

21.26 Corollary. Diﬀ P = Diﬀ n P = Homk (F, P) = F ◦

k

P.

We warn the reader that the natural inclusion Diﬀ k Q ⊂ Diﬀ k+1 Q does not coincide with the evident inclusion

l

l Homk Λ V, Q → Homk Λ V, Q . l≤k

l≤k+1

Now we are ready to describe derivations, diﬀerential forms, as well as the de Rham complex on any Grassman algebra F. Proposition. If P is a graded F-module, then there is a natural isomorphism of modules Dk (P) = Homk (S k V, P) = S k V ◦ ⊗k P obtained by restricting the arguments of the multi-derivation to elements of V ⊂ F. Corollary. The representative object of the functors Dk is Λk = S k V ⊗k F, so that Λ(F) = SV ⊗k F. Note that there are diﬀerential forms of arbitrary high order, which indicates that an attempt to construct integration theory on the basis of diﬀerential forms fails in the graded situation. Recall that in general if A is G-graded commutative, then Λ(A) is G ⊕ Z-graded commutative, with the parity form ·, · G⊕Z = ·, · G + ·, · ρZ . Moreover, the exterior diﬀerential d is of degree (0, 1) and the algebra Λ(A) is generated by the elements a ∈ A = Λ0 and da ∈ Λ1 . Using this and the previous corollary, we obtain: Proposition. The module Λ(F) is freely generated as a Z ⊕ Z-algebra by the elements ξi of degree (1, 0), where the ξi are the generators of V , and by the elements dξi of degree (1, 1). Explicitly, this means that we have the relations: ξi ξj = −ξj ξi ,

dξi dξj = dξj dξi ,

Consider the polynomial algebra def

B = SV =

i≥0

ξi dξj = −dξj ξi .

S i V,

406

Chapter 21

where S i V is i-th symmetric power of V , which physicists call “bosonic”. Now note that if we start with this algebra, we also obtain the algebra Λ(B) = SV ⊗k ΛV . So in the case of k-vector spaces, there is a canonical identiﬁcation φ : Λ(F) → Λ(B). If we denote our basis of V in the bosonic picture by x1 , . . . , xn , then this isomorphism is given by xi ↔ dξi ,

dxi ↔ ξi .

But this is not an isomorphism of graded algebras, since in Λ(B) we have xi dxj = dxj · xi , while in Λ(F) a sign appears when we exchange these two elements. Nevertheless, it is useful to interpret the operations from one picture in terms of operators in the other one, and we shall use this to compute the de Rham cohomology of Λ(F). In this ∂connection, recall that the Liouville . ﬁeld on a vector space V is l = i xi ∂x i Proposition. Let dF be the diﬀerential over Λ(F), then dF : Λ(B) → Λ(B) is an inclusion of the Liouville vector ﬁeld in Λ(B), i.e., dF (β) is an inclusion of l into β, β ∈ Λ(B). We shall denote the exterior diﬀerential in Λ(B) by d. Using the isomorphism φ, we see that dF (xi ) = dF (dF ξi ) = 0 = il (xi ) dF (dxi ) = dF ξi = xi = il (dxi ) It remains to show that dF is a derivation of degree −1 on the algebra Λ(B). We leave this veriﬁcation to the reader as an exercise. 21.27 Corollary. The de Rham cohomology of F is k, i = 0, i H (F) = 0, i = 0. 21.28. The Lie algebra of derivations. Deﬁnition. The commutator of two derivations Δ, ∇ ∈ D(A) is the operator [Δ, ∇] = Δ ◦ ∇ − (−1)Δ,∇ ∇ ◦ Δ. Exercises. Prove that: 1. [Δ, ∇] ∈ D(A) and ω([Δ, ∇]) = ω(Δ) + ω(∇), ω denotes the value of the grading. 2. With respect to the commutator, D(A) is a G-graded Lie algebra. 21.29. The Batalin–Vilkovisky bracket. Already from the algebraic deﬁnition of a diﬀerential operator, it is possible to extract notions and constructions which are extremely useful for physical applications. I. Batalin

Diﬀerential Operators over Graded Algebras

407

and G. Vilkovisky discovered that to any operator Δ of second order and odd grading such that Δ◦Δ = 0 it is possible to assign a bracket A×A → A satisfying a kind of Jacobi identity. In the framework of the algebraic approach, we will extend the class of operators for which such a bracket can be applied, and also construct its n-ary version (see [27]). Deﬁnition. A diﬀerential operator of odd grading Δ ∈ Diﬀ n A subject to the condition Δ ◦ Δ ∈ Diﬀ 2n−2 A.

(21.34)

will be called a Batalin–Vilkovisky operator. Further, smbln Δ is called the Batalin–Vilkovisky symbol. The bracket [ ·, . . . , · ]ς : A ⊗ . . . ⊗ A → A, where ς = smbln Δ (see Exercise 3 from Section 10.2), will be called the Batalin–Vilkovisky bracket. 21.30 Theorem. The Batalin–Vilkovisky bracket satisﬁes the following n-ary version of the Jacobi identity: (−1)|a(I,J)|s +|a(J)| [[a(I)]ς , a(J)]ς = 0. (21.35) I+J=I (2n−1) |I|=n

For the proof of the theorem we need the next lemma, which can be established by induction, using the fact that the operator δa acts on the composition of DO’s as a derivation: δa (Δ ◦ ∇) = δa (Δ) ◦ ∇ + (−1)a,Δ Δ ◦ δa (∇). (21.36) 21.31 Lemma. δa(I (l) ) (Δ ◦ ∇) =

(−1)|a(I,J)|s +(a(J),Δ δa(I) (Δ) ◦ δa(J) (∇)

(21.37)

I+J=I (l) 0|I|l

Now assume that ∇ = Δ ∈ Diﬀ n A and l = 2n−1; then in the decomposition (21.37) only the terms with |I| = n, |J| = n−1, and |I| = n−1, |J| = n will be nonzero, and so in this case the equality (21.37) can be given the following form: δ (2n−1) ) (Δ ◦ Δ) a(I = (−1)|a(I,J)|s +a(J),Δ+Δ,Δ (−1)Δ,Δ δa(I) (Δ) ◦ δa(J) (Δ) I+J=I (2n−1) |I|=n

+ (−1)δa(J) (Δ),δa(I) (Δ) δa(J) (Δ) ◦ δa(I) (Δ) (−1)|a(I,J)|s +a(J),Δ+Δ,Δ (−1)Δ,Δ [a(I)]Δ ◦ δa(J) (Δ) =

I+J=I (2n−1) |I|=n

+ (−1)δa(J) (Δ),[a(I)]Δ δa(J) (Δ) ◦ [a(I)]Δ

(21.38)

408

Chapter 21

If Δ is an operator of odd grading, then the last expression in the big brackets is nothing other than

δ[a(I)]Δ δa(J) (Δ) = δ[a(I)]Δ ,a(J) (Δ) = [[a(I)]Δ , a(J)]Δ , and equality (21.38) acquires the form δa(I (2n−1) ) (Δ ◦ Δ) = − (−1)|a(I,J)|s +|a(J)| [[a(I)]Δ , a(J)]Δ I+J=I (2n−1) |I|=n

The remark asserting that for the Batalin–Vilkovisky bracket of the Batalin–Vilkovisky operator the left-hand side of the equality vanishes concludes the proof of the theorem. The last equality also shows that the condition Δ ◦ Δ ∈ Diﬀ 2n−2 A is a necessary and suﬃcient condition for the validity of the Jacobi identity 21.35. 21.32. Proposition. If Δ is a Batalin–Vilkovisky operator of order n and ∇ is an even DO of order k such that ∇ ◦ ∇ ∈ Diﬀ 2k−2 A, then the composition Δ ◦ ∇ is a Batalin–Vilkovisky operator of order n + k. Since smbl(Δ ◦ ∇) = smbl(Δ ◦ ∇ − 12 [Δ, ∇]), it suﬀﬁces to prove that Δ ◦ ∇ − 12 [Δ, ∇] is a Batalin–Vilkovisky operator. 1 1 (Δ ◦ ∇ − [Δ, ∇]) ◦ (Δ ◦ ∇ − [Δ, ∇]) 2 2 1 1 1 = Δ ◦ ∇ ◦ Δ ◦ ∇ − [Δ, ∇] ◦ Δ ◦ ∇ − Δ ◦ ∇ ◦ [Δ, ∇] + [Δ, ∇] ◦ [Δ, ∇] 2 2 4 1 1 1 =Δ◦∇◦Δ◦∇− Δ◦∇◦Δ◦∇+ ∇◦Δ◦Δ◦∇− Δ◦∇◦Δ◦∇ 2 2 2 1 1 + Δ ◦ ∇ ◦ ∇ ◦ Δ + [Δ, ∇] ◦ [Δ, ∇] 2 4 1 1 1 = ∇ ◦ Δ ◦ Δ ◦ ∇ + Δ ◦ ∇ ◦ ∇ ◦ Δ + [Δ, ∇] ◦ [Δ, ∇] 2 2 4 Each of the summands in the last expression is a DO of order n + k − 2. 21.33. The Batalin–Vilkovisky bracket. Case n = 2. In that case |(a1 , a2 , a3 )|s = 0, |(a1 , a3 , a2 )|s = |a2 | · |a3 |, |(a2 , a3 , a1 )|s = |a1 | · |a2 | + |a1 | · |a3 | and Jacobi identity 21.35 acquires the form (−1)|a3 | [[a1 , a2 ]ς , a3 ]ς + (−1)a2 ,a3 +|a2 | [[a1 , a3 ]ς , a2 ]ς +(−1)a1 ,a2 +a1 ,a3 +|a1 | [[a2 , a3 ]ς , a1 ]ς = 0

(21.39)

Diﬀerential Operators over Graded Algebras

409

To each element a ∈ A, we can assign the Hamiltonian vector ﬁeld (=Hamiltonian derivation) Xa : A → A by putting Xa = (−1)|a|+1 [ · , a]ς ,

Xa (b) = (−1)|a|+1 [b, a]ς

(21.40)

21.34. Exercise. Verify that Xa is a derivation. Proposition. [Xa2 , Xa1 ] = −(−1)|a2 | X[a1 ,a2 ]ς

(21.41)

In each summand of (21.39), let us place a3 in the ﬁrst position and cancel out the expression (−1)a3 ,a1 +a3 ,a2 , obtaining [a3 , [a1 , a2 ]ς ]ς + (−1)|a2 | [[a3 , a1 ]ς , a2 ]ς + (−1)a2 ,a1 +|a1 | [[a3 , a2 ]ς , a1 ]ς = 0

(21.42)

According to deﬁnition (21.40), we have [a3 , a2 ]ς =(−1)|a2 |+1 Xa2 (a3 ) [[a3 , a2 ]ς , a1 ]ς =(−1)|a1 |+|a2 | Xa1 (Xa2 (a3 )) [[a3 , a1 ]ς , a2 ]ς =(−1)|a2 |+|a1 | Xa2 (Xa1 (a3 )) [a3 , [a1 , a2 ]ς ]ς =(−1)|[a1 ,a2 ]ς |+1 X[a1 ,a2 ]ς (a3 ) =(−1)|a2 |+|a1 | X[a1 ,a2 ]ς (a3 ) Substituting this expression into (21.42) and multiplying by (−1)|a1 | , we conclude that (−1)|a2 | X[a1 ,a2 ]ς (a3 ) + Xa2 (Xa1 (a3 )) − (−1)a2 ,a1 +|a2 |+|a1 |+1 Xa1 (Xa2 (a3 )) = (−1)|a2 | X[a1 ,a2 ]ς (a3 ) + [Xa2 , Xa1 ](a3 ) = 0, as claimed.

21.35. Derivation of dioles. Let A = A(P ) be a diole. Let us try to understand the meaning of the A-module D(A). We have Ai = 0, and if i = 0, 1, then any derivation Δ of grading r is a pair (Δ0 , Δ1 ), where Δ0 ∈ D(Ar ) and Δ1 ∈ Diﬀ A (P, Ar+1 ), and Δ0 (ab) = Δ0 (a)b + aΔ0 (b), ∀a, b ∈ A,

(21.43)

Δ (ap) = Δ (a)p + aΔ (p), ∀a ∈ A, p ∈ P.

(21.44)

1

0

1

Obviously, D(A)r is nontrivial only if r equals −1, 0 and 1. Consider these three cases. (1) D(A)1 . In that case, Δ1 ∈ Diﬀ A (P, A2 ) = 0 and condition (21.44) degenerates. Therefore, D(A)1 = D(P ). (2) D(A)0 . In that case, Δ0 ∈ D(A0 ) = D(A) and Δ1 ∈ Diﬀ A (P, A1 ) = Diﬀ A (P, P )

410

Chapter 21

and conditions (21.43), (21.44) are nondegenerate. The A-module consisting of pairs (Δ0 , Δ1 ), where Δ1 : P → P is a DO of ﬁrst order, while the map Δ0 : A → A is a derivation satisfying condition (21.44), is customarily denoted by Der P . Thus, D(A)0 = Der P . Note that if Δ0 = 0, then Δ1 is an A-homomorphism from P to P , and so HomA (P, P ) ⊂ Der P . (3) D(A)−1 . We have Δ0 ∈ D(A−1 ) = 0 and Δ1 ∈ Diﬀ A (P, A0 ) = Diﬀ A (P, A). The conditions (21.43), (21.44) degenerate into the following single condition Δ1 (ap) = aΔ1 (p), ∀a ∈ A, p ∈ P, so that D(A)−1 = HomA (P, A) = P ∗ . Thus, we have proved the following. 21.36 Proposition.

⎧ ⎪ 0, ⎪ ⎪ ⎪ ⎪ ⎪ ⎨D(P ), j D(A) = Der P, ⎪ ⎪ ⎪ HomA (P, A) = P ∗ , ⎪ ⎪ ⎪ ⎩0,

Exercises.

j j j j j

> 1, = 1, = 0, = −1, < −1.

(21.45)

1. Describe the A-module structure in D(A).

2. Describe the A-modules D(Q) to in the general case. 21.37. The module Der P from the geometrical viewpoint. To any diole A = (A, P ) we can assign its value at the point h ∈ |A| (where |A|, as before, denotes the k-spectrum of the algebra A) def

Figure 21.1. The ﬁber πz of the bundle π over z.

by writing Ah = (Ah = k, Ph ). Then the derivations with values in Ah are pairs of the form (Xh0 , Xh1 ), where Xh0 is the tangent vector at a point h ∈ |A|, while Xh1 is a map from the module P to the vector space Ph . In the case when A is the algebra of smooth functions on a smooth manifold M and P is the module of sections of a vector bundle π over M , we can identify M and |A|. Here Pz , z ∈ M , can be interpreted as the ﬁber πz of π over the point z. Further, Xz0 is the tangent vector at the point z, and Xz1 is “the set of tangent vectors to points of the ﬁber πz that are projected on Xz0 .”

Diﬀerential Operators over Graded Algebras

411

21.38. Representing objects. As in the nongraded case, the functors D and Diﬀ n (P, ·), where P is a graded module over A, are representable. Indeed, we have the following theorem. Theorem. (1) There exists a unique (up to isomorphism) A-module Λ(A) and a derivation of zero grading δ : A → Λ(A) such that for any A-module Q the following isomorphism exists: Hom(Λ(A), Q) = D(Q),

Hom(Λ(A) h → h ◦ δ ∈ D(Q).

(2) For any A-module P, there exists a unique (up to an isomorphism) A-module J n (P) and a DO of zero grading jn : P → J n (P) such that for any A-module Q the following isomorphism exists: Hom(J n (P), Q) = Diﬀ n (P, Q), Hom(J n (P) h → h ◦ jn ∈ Diﬀ n (P, Q). The proof of this theorem is left to the reader as an exercise. The module Λ(A) is called the module of diﬀerential forms of ﬁrst degree of the algebra A, the module J n (P), the module of jets of order n of the module P, while δ and jn are the universal derivations and universal DO of order n, respectively. Elements of the module Λ(A) are called diﬀerential forms of ﬁrst degree, while elements J n (P) are jets of order n of the A-module P. If a, b ∈ A, p ∈ P are homogeneous elements, then aδb ∈ Λ(A) and ajn (p) ∈ J n (P) will also be homogeneous elements, and for the grading ω we will have ω(aδb) = ω(a) + ω(b), ω(ajn (p)) = ω(a) + ω(p). 21.39. Representative objects and the category of modules over dioles. Let A = (A, P ) be a diole and Q = (Q0 , Q1 ) be an A -module. This means that (1) Q0 and Q1 are A-modules; (2) there exist an A-homomorphism Q : P A Q0 → Q1 ; the element Q (p A q), q ∈ Q0 is said to be the product of p by q and is denoted by p · q. In other words, in order to transform the pair of A-modules (Q0 , Q1 ) into an A-module, it suﬃces to deﬁne the A-homomorphism : P ⊗A Q0 → Q1 . For simplicity of notation, let us put Λ = Λ(A) and J 1 = J 1 (A), consider the pair (Λ, J 1 ), and deﬁne the map Λ : P ⊗A Λ → J 1 by setting def

Λ (p ⊗A da) = p · da = j1 (ap) − aj1 (p). Proposition. 1. bp · da = b(p · da). 2. p · d(ab) = p · adb + p · bda.

412

Chapter 21

By deﬁnition, bp · da = j1 (abp) − aj1 (bp) = aj1 (bp) + bj1 (ap) − abj1 (p) − aj1 (bp) = bj1 (ap) − abj1 (p) = b(p · da). Further, j1 (abp) = aj1 (bp) + bj1 (ap) − abj1 (p). Therefore, p · d(ab) = j1 (abp) − abj1 (p) = aj1 (bp) + bj1 (ap) − 2abj1 (p) = a(bj1 (p) − j1 (bp)) + b(aj1 (p) − j1 (ap)) = p · adb + p · bda. This proposition means that the above-constructed map is an A-homomorphism and so transforms Ω = (Λ(A), J 1 (P )) into an A-module. By deﬁnition, j1 (ap) = p · da + aj1 (p), i.e., d = (d, j1 ) ∈ D(Ω). Now let Δ = (Δ0 , Δ1 ) ∈ D(Q)l . By the universality properties of the operators d and j1 , there exist A-homomorphisms h0 : D(A) → Ql and h1 : J 1 (P ) → Ql+1 such that Δ0 = h0 ◦ d and Δ1 = h1 ◦ j1 . In order that the pair h = (h0 , h1 ) be an A-homomorphism from Ω to Q, it is necessary and suﬃcient that h1 (da · p) = h0 (da) · p. Let us verify this: p · h0 (da) = p · Δ0 (a) = Δ1 (ap) − aΔ1 (p)

= h1 (j1 (ap)) − h1 (aj1 (p)) = h1 j1 (ap)) − aj1 (p) = h1 (p · da). Note that if Ql = 0, then Δ1 ∈ HomA (P, Ql+1 ) ⊂ Diﬀ 1 (P, Ql+1 ) and h1 = Δ1 ◦ π1,0 , where π1,0 : J 1 (P ) → J 0 (P ) = P is the natural projection. Actually, we have proved that Ω is the representing object of the functor D, while d = (d, j1 ) is the corresponding derivation. In view of the uniqueness of the representing object, we obtain the following. Theorem. Λ(A) = (Λ(A), J 1 (P )), δ = d = (d, j1 ). Exercises. 1. Denote by Λ1 (P ) the submodule of the module J 1 (P ) generated by all elements of the form j1 (ap) − aj1 (p). Prove the equality Λ1 (P ) = Λ1 ⊗ P . 2. Prove the exactness of the short sequence 0 → Λ1 (P ) → J 1 (P ) → P → 0. (21.46) g To any A-module Q = g∈Z Q , we can assign the family of A-modules def Jn (Q) = J n (Qg ), n ∈ Z+ , g∈Z

Diﬀerential Operators over Graded Algebras

413

def

where p · jn (q) = jn (p · q), p ∈ P, q ∈ Qg . Theorem–Exercise. Prove that 1. The A-modules Jn (Q) are well deﬁned. 2. Jn (Q) = J n (Q). 21.40. Adjoint operators and integral forms. All the constructions of sections 20.34–20.38 can be easily carried over to the case of graded (see (20.20)), an commutative algebras. In the deﬁnition of the map Δ appropriate plus or minus sign appears: def = (−1)Δ,∇ ∇ ◦ Δ ∈ Diﬀ > (P, Λk ), ∇ ∈ Diﬀ + (Q, Λk ), (21.47) Δ(∇) and formulas (20.21)–(20.22) take the following form:

(∇ ◦ Δ)∗n = (−1)Δ,∇ Δ∗n ◦ ∇∗n , Δ,p Δ(p), q = (−1) q ) . p, Δ∗ (

(21.48) (21.49)

Exercise. Construct the cochain complex of integral forms for the Grassman algebra F. 21.41. The Berezinian for supermanifolds. In conclusion of this chapter, let us show how, for supermanifolds, the Berezinian (deﬁned earlier for nongraded commutative algebras in 20.36) naturally arises under the conceptually correct approach, and not as deus ex machina or a rabbit out of the magician’s hat in the classical approach. We will not give a formal deﬁnition of the notion of supermanifold (this would lead us out of the framework of this, book limiting ourselves to some “hand-waving.” As the simplest example of such a manifold, consider the superspace—the pair (Rn+k , An|k ), where An|k = C ∞ (Rn ) [ξ1 , . . . , ξk ] and the elements ξi are odd: ξi ξj = −ξj ξi . In other words, An|k consists of polynomials of odd variables ξ1 , . . . , ξk with coeﬃcients from the algebra of smooth functions on Rn . This superspace will be denoted by Rn|k , and n|k will be its superdimension. In the general case, the given supermanifold (M, A) of superdimension n|k can be represented as a smooth manifold of dimension n + k with a ring A of functions on it such that for any coordinate domain U the pair (U, A|U ) is isomoprphic to the superspace Rn|k . (We omit the details.) A typical example, in fact the one from which the study of supermanifolds began, is the pair T ∗ (M ), Λ∗ (M ) , where M is a smooth manifold, while Λ∗ (M ) is the de Rham algebra of diﬀerential forms. We have the following. Theorem. Let (M, A) be a supermanifold of superdimension n|k. Then (1) Bi = Ai = 0, if i = n. (2) Bn = An is a module of rank 1.

414

Chapter 21

(3) For any coordinate domain U, the module is Bn |U free. To each coordinate system {x, ξ} = {x1 , . . . , xn , ξ1 , . . . , ξk } on U a generator ex,ξ of the module Bn |U can be naturally associated so that under the change of coordinates {x, ξ} → {v, ψ} the following equality holds: x, ξ ex,ξ = Ber J ev,ψ ; v, ψ

x,ξ is the Jacobi matrix of the coordinate change, while Ber is the here J v,ψ Berezin determinant, which in the case of even matrices can be calculated by the following rule: BC Ber = det(B − CE −1 D)(det E)−1 , DE where B and E are square matrices of orders n and k, respectively. We will not prove this theorem. We have stated it here in order to demonstrate how the notion of Berezin determinant arises and to show what role it plays in “supermathematics.” Note that under the traditional approach, the appearance of the Berezin determinant is veiled in mystery: ﬁrst the Berezin determinant of an even B C matrix is deﬁned as the expression det(B − CE −1 D)(det E)−1 , D E then Ber M is deﬁned as a one-dimensional ﬁbration for which the local section over the domain U is of the form f (x) D(x), where f ∈ C ∞ (U) and D is the basis local section, which, under the change of coordinates D is multiplied by the Berezin determinant of the corresponding Jacobi matrix.

Afterword

If we continue on the path traced out by this book and analyze to what extent contemporary mathematics corresponds to the observability principle, we see that many things in our science are simply conceptually unfounded. This unavoidably leads to serious diﬃculties, which are usually ignored from force of habit even when they contradict our experience. If, for example, measure theory is the correct theory of integration, then why is it that all attempts to construct the continual integral on its basis have failed, although the existence of such integrals is experimentally veriﬁed? As the result of this, physicists are forced to use “unobservable” mathematics in their theories; as a result, considerable diﬃculties arise, say, in quantum ﬁeld theory, which some experts even regard as an inherent aspect of the theory. It is generally believed that the mathematical basis of quantum mechanics is the theory of self-adjoint operators in Hilbert space. But then why does Dirac writes that “physically signiﬁcant interactions in quantum ﬁeld theory are so strong that they throw any Schr¨ odinger state vector out of Hilbert space in the shortest possible time interval”? Having noted this, one must either avoid writing the Schr¨ odinger equation in the context of quantum ﬁelds theory or refuse to consider Hilbert spaces as the foundation of quantum mechanics. Dirac reluctantly chose the ﬁrst alternative, and this refusal was forced, since the mathematics of that time allowed him to talk about solutions of diﬀerential equations only in a very limited language (see the quotation at the beginning of the Introduction). On the other hand, since the Hilbert space formalism contains no procedure for distinguishing one vector from another, the observability © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4

415

416

Afterword

principle is not followed here. Thus the second alternative seems moreappropriate, but it requires specifying many other points, e.g., ﬁnding out how one can observe solutions of partial diﬀerential equations; this question, however, is outside the sphere of interests of the PDE experts: To them, even setting the question seems strange, to say the least. Thus the systematic mathematical formalization of the observability principle requires rethinking many branches of mathematics that seemed established once and for all. The main diﬃcult step that must be taken in this direction is to ﬁnd solutions in the framework of the diﬀerential calculus, avoiding the appeal of functional analysis, measure theory, and other purely set-theoretical constructions. In particular, we must refuse measure theory as integration theory in favor of the purely cohomological approach. One page suﬃces to write out the main rules of measure theory. The number of pages needed to explain de Rham cohomology is much larger. The conceptual distance between the two approaches shows what serious diﬃculties must be overcome on this road. At present it is clear that this road leads to the secondary diﬀerential calculus (already mentioned in the Introduction) and its main applications, e.g., cohomological physics. The reader may obtain an idea of what has already been done, and what remains to be done, in this direction by consulting the references appearing below.

Appendix A. M. Vinogradov Observability Principle, Set Theory and the “Foundations of Mathematics”

The following general remarks are meant to place the questions discussed in this book from the perspective of observable mathematics. Propositional and Boolean algebras. While the physicist describes nature by means of measuring devices with R-valued scales, the layman does so by means of statements. Using the elementary operations of conjunction, disjunction, and negation, new statements may be constructed from given ones. A system of statements (propositions) closed with respect to these operations is said to be a propositional algebra. Thus, the means of observation of an individual not possessing any measuring devices is formalized by the notion of propositional algebra. Let us explain this in more detail. Let us note, ﬁrst of all, that the individual observing the world without measuring devices was considered above only as an example of the main, initial mechanism of information processing, which in the sequel we shall call primitive. Thus, we identify propositional algebras with primitive means of observation. Further, let us recall that any propositional algebra A may be transformed into an unital commutative algebra over the ﬁeld Z2 of residues modulo 2 by introducing the operations of multiplication and addition as follows: def

pq = p ∧ q, def

p + q = (p ∧ q¯) ∨ (¯ p ∧ q), where ∨ and ∧ are the propositional connectives conjunction and disjunction, respectively, while the bar over a letter denotes negation. All elements © Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4

417

418

A. M. Vinogradov

of the algebra thus obtained are idempotent, i.e., a2 = a. Let us call any unital commutative Z2 -algebra Boolean if all its elements are idempotent. Conversely, any Boolean algebra may be regarded as a propositional algebra with respect to the operations def

p ∧ q = pq, def

p ∨ q = p + q + pq, def

p¯ = 1 + p. This shows that there is no essential diﬀerence between propositional and Boolean algebras, and the use of one or the other only speciﬁes what operations are involved in the given context. Thus we can restate the previous remarks about means of observation as follows: Boolean algebras are primitive means of observation. Boolean spectra. The advantage of the previous formulation is that it immediately allows us to discern the remarkable analogy with the observation mechanism in classical physics as interpreted in this book. Namely, in this mechanism one must merely replace the R-valued measurement scales by Z2 -valued ones (i.e., those that say either “yes” or “no”) and add the idempotence condition. This analogy shows that what we can observe by means of a Boolean algebra A is its Z2 -spectrum, i.e., the set of all its homomorphisms as an unital Z2 -algebra to the unital Z2 -algebra Z2 . Let us denote this spectrum by SpecZ2 and endow it with the natural topology, namely the Zariski one. Then we can say, more precisely, that Boolean algebras allow us to observe topological spaces of the form SpecZ2 , which we shall call, for this reason, Boolean spaces. In connection with the above, one may naturally ask whether the spectra of Boolean algebras possess any structure besides the topological one, say, a smooth structure, as was the case for spectra of R-algebras. The reader who managed to do Exercise 4 from Section 9.45 already knows that the diﬀerential calculus over Boolean algebras is trivial in the sense that any diﬀerential operator on such an algebra is of order zero, i.e., is a homomorphism of modules over this algebra. This means, in particular, that the phenomenon of motion cannot be adequately described and studied in mathematical terms by using only logical notions or, to put it simply, by using everyday language (recall the classical logical paradoxes on this topic). The Stone theorem stated below, which plays a central role in the theory of Boolean algebras, shows that the spectra of Boolean algebras possess only one independent structure: the topological one. In the statement of the theorem it is assumed that the ﬁeld Z2 is supplied with the discrete topology. Stone’s theorem. Any Boolean space is an absolutely disconnected compact Hausdorﬀ space and, conversely, any Boolean algebra coincides with

Observability Principle, Set Theory and the “Foundations of Mathematics”

419

the algebra of open-and-closed sets of its spectrum with respect to the set-theoretic operations of symmetric diﬀerence and intersection. Recall that the absolute disconnectedness of a topological space means that the open-and-closed sets form a base of its topology. The appearance of these simultaneously open-and-closed sets in Boolean spaces is explained by the fact that any propositional algebra possesses a natural duality. Namely, the negation operation maps it onto itself and interchanges conjunction and disjunction. Note also that Stone’s theorem is an identical twin of the Spectrum theorem (see Sections 7.2 and 7.7). Their proofs are based on the same idea, and diﬀer only in technical details reﬂecting the speciﬁcs of the diﬀerent classes of algebras under consideration. The reader may try to prove this theorem as an exercise, having in mind that the elements of the given Boolean algebra can be naturally identiﬁed with the open-and-closed subsets of its spectrum, while the operations of conjunction, disjunction, and negation then become the set-theoretical operations of intersection, union, and complement, respectively. It is easy to see that the spectrum of a ﬁnite Boolean algebra is a ﬁnite set supplied with the discrete topology. Thus any ﬁnite Boolean algebra turns out to be isomorphic to the algebra of all subsets of a certain set. “Eyes” and “ears.” After all these preliminaries, the role of “eyes” and “ears” in the process of observation may be described as follows. First of all, the “crude” data absorbed by our senses are written down by the brain and sent to the corresponding part of our memory. One may think that in the process of writing down, the crude data are split up into elementary blocks, “macros,” and so on, which are marked by appropriate expressions of everyday language. These marks are needed for further processing of the stored data. The system of statements constituting some description generates an ideal of the controlling Boolean algebra, thus distinguishing the corresponding closed subset in its spectrum. Supposing that to each point of the spectrum an elementary block is assigned, and this block is marked by the associated maximal ideal, we come to the conclusion that to each closed subset of the spectrum one can associate a certain image, just as a criminalist creates an identikit from individual details described by witnesses. Thus, if we forget about the “material” content of the elementary blocks (they may be “photographs” of an atomic fragment of a visual or an audio image, etc.) that corresponds (according to the above scheme) to points of the spectrum of the controlling Boolean algebra, we may assume that everything that can be observed on the primitive level is tautologically expressed by the points of this spectrum. Boolean algebras corresponding to the primitive level. It is clear that any rigorous mathematical notion of observability must come from some notion of observer, understood as a kind of mechanism for gathering and processing information. In other words, the notion of observability must be formalized approximately in the same way as Turing machines formalize

420

A. M. Vinogradov

the notion of algorithm. So as not to turn out to be an a priori formalized metaphysical scheme, such a formalization must take into account “experimental data.” The latter may be found in the construction and evolution of computer hardware and in the underlying theoretical ideas. Therefore it is useful to regard the individual mathematician, or better still, the mathematical community, in the spirit of the “noosphere” of Vernadskii, as a kind of computer. Then, having in mind that the operational system of any modern computer is a program written in the language of binary codes, we can say that there is no alternative to Boolean algebras as the mechanism describing information on the primitive level. For practical reasons, as well as for considerations of theoretical simplicity, it would be inconvenient to limit the size of this algebra by some concrete number, say the number of elementary particles in the universe. Hence it is natural to choose the free algebra in a countable number of generators. The notion of level of observability is apparently important for the mathematical analysis of the notion of observability itself, and we shall return to it below. How set theory appeared in the foundations of mathematics. As we saw above, any propositional algebra is canonically isomorphic to the algebra of all subsets of the spectrum of the associated Boolean algebra. If this spectrum is ﬁnite, then its topology is discrete. So we can forget about the topology without losing anything. Moreover, any concrete individual, especially if he/she is not familiar with Boolean algebras, feels sure that what she/he is observing are just subsets or, more precisely, the identikits which she/he deﬁned. Therefore, such an immediate “material” feeling leads us to the idea that the initial building blocks of precise abstract thinking are “points” (“elements”) grouped together in “families,” i.e., sets. Having accepted or rather having experienced this feeling of primitivity of the notion of set under the pressure of our immediate feelings, we are forced to place set theory at the foundation of exact knowledge, i.e., of mathematics. On the primitive level of ﬁnite sets, this choice, in view of what was explained above, does not contradict the observability principle, since any ﬁnite set can be naturally and uniquely interpreted as the spectrum of some Boolean algebra. However, if we go beyond the class of ﬁnite sets, the situation changes radically: The notion of observable set, i.e., of Boolean space, ceases to coincide with the general notion of a set without any additional structure. Therefore, our respect for the observability principle leads us to abandon the notion of a set as the formal-logical foundation of mathematics and leave the paradise so favored by Hilbert. One of the advantages of such a step, among others, is that it allows us to avoid many of the paradoxes inherent to set theory. For example, the analog of the “set of all sets” in observable mathematics is the “Boolean space of all Boolean spaces.” But this last construction is clearly meaningless, because it deﬁnes no topology in the “Boolean space of all Boolean spaces.” Or the “observable” version

Observability Principle, Set Theory and the “Foundations of Mathematics”

421

of the “set (not) containing itself as an element,” i.e., the “Boolean space (not) containing itself as an element” is so striking that no comment is needed. In this connection we should additionally note that in order to observe Boolean spaces (on the primitive level!) as individual objects, a separate Boolean space that distinguishes them is required. Observable mathematical structures (Boole groups). Now it is the time to ask what observable mathematical structures are. If we are talking about groups observable in the “Boolean” sense, then we mean topological groups whose set of elements constitutes a Boolean space. Such a group should be called Boolean. In other words, a Boole group is a group structure on the spectrum of some Boolean algebra. If we replace in this deﬁnition the notion of Boolean observability by that of classical observability, we come to the notion of Lie group, i.e., of a group structure on the spectrum of the classical algebra of observables. Observing observables: diﬀerent levels of observability. Just as the operating system in a computer manipulates programs of the next level, one can imagine a Boolean algebra of the primitive level (see above) with the points of its spectrum marking other Boolean algebras. In other words, this is a Boolean algebra observing other Boolean algebras. Iterating this procedure, we come to “observed objects,” which, if one forgets the multistep observation scheme, can naively be understood as sets of cardinality higher than ﬁnite or countable. For instance, starting from the primitive level, we can introduce into observable mathematics things that in “nonobservable” mathematics are related to sets of continual cardinality. In this direction, one may hope that there is a constructive formalization of the observability of smooth R-algebras, which, in turn, formalize the observation procedure in classical physics. Down with set theory? The numerous failed attempts to construct mathematics on the formal-logical foundations of set theory, together with the considerations related to observability developed above, lead us to refuse this idea altogether. We can note that it also contradicts the physiological basis of human thought, which ideally consists in the harmonious interaction of the left and right hemispheres of the brain. It is known that the left hemisphere is responsible for rational reasoning, computations, logical analysis, and pragmatic decision-making. Dually, the right hemisphere answers for “irrational” thought, i.e., intuition, premonitions, emotions, imagination, and geometry. If the problem under consideration is too hard for direct logical analysis, we ask our intuition what to do. We also know that in order to obtain a satisfactory result, the intuitive solution must be controlled by logical analysis and, possibly, corrected on its basis. Thus, in the process of decision-making, in the search for the solution of a problem, etc., the switching of control from one hemisphere to the other takes place, and such iterations can be numerous.

422

A. M. Vinogradov

All this, of course, is entirely relevant to the solution of mathematical problems. The left hemisphere, i.e., the algebro-analytical part of our brain, is incapable of ﬁnding the solution to a problem whose complexity is higher than, say, the possibilities of human memory. Indeed, from any assumption one can deduce numerous logically correct consequences. Therefore, in the purely logical approach, the number of chains of inference grows at least exponentially with their length, while those that lead to a correct solution constitute a vanishingly small part of that number. Thus if the correct consequence is chosen haphazardly at each step, and the left hemisphere knows no better, then the propagation of this “logical wave” in all directions will overﬁll our memory before it reaches the desired haven. The only way out of this situation is to direct this wave along an appropriate path, i.e., to choose at each step the consequences that can lead in a more or less straight line to the solution. But what do we mean by a “straight line”? This means that an overall picture of the problem must be sketched, a picture on which possible ways of solution could be drawn. The construction of such an overall picture, in other words, of the geometrical image of the problem, takes place in the right hemisphere, which was created by nature precisely for such constructions. The basic building blocks for them, at least when we are dealing with mathematics, are sets. These are sets in the naive sense, since they live in the right hemisphere. Hence any attempt to formalize them, moving them from the right hemisphere to the left one, is just an outrage against nature. So let us leave set theory in the right hemisphere in its naive form, thanks to which it is has been so useful. Inﬁnitesimal observability. Above we considered Boolean algebras as analog of smooth algebras. But we can interchange our priorities and do things the other way around. From this point of view, the operations or, better, the functors of the diﬀerential calculus, will appear as the analog of logical operations, and the calculus itself as a mechanism for manipulating inﬁnitesimal descriptions. In this way we would like to stress the inﬁnitesimal aspect related to observability. Some of the “primary” functors were described in this book. Their complete list should be understood as the logic algebra of the diﬀerential calculus. The work related to the complete formalization of this idea is still to be completed. In conclusion let us note, expressing ourselves informally, that in our imaginary computer, working with stored knowledge, the program called “diﬀerential calculus” is not part of its operating system, and so is located at a higher level than the primitive one (see above). This means that the geometrical images built on its basis cannot be interpreted in a material way. They should retain their naive status in the sense explained above. The constructive diﬀerential calculus, developed in the framework of “constructive mathematical logic,” illustrates what can happen if this warning is ignored.

References

[1] V.N. Chetverikov, A.B. Bocharov, S.V. Duzhin, N.G. Khor’kova, I.S. Krasil’shchik, A.V. Samokhin, Y.N. Torkhov, A.M. Verbovetsky, A.M. Vinogradov, in Symmetries and Conservation Laws for Diﬀerential Equations of Mathematical Physics, vol. 182, Translations of Mathematical Monographs, ed. by J. Krasil’shchik, A. Vinogradov (American Mathematical Society, Providence, 1999) [2] H. Goldschmidt, D. Spencer, On the nonlinear cohomology of Lie equations IV. J. Diﬀer. Geom. 13(4), 455–526 (1978) [3] H. Goldschmidt, Sur la structure des equations de Lie, I, II, III. J. Diﬀer. Geom. 6(3), 357–373 (1972); 7(1-2), 67–95 (1972); 11(2) 167– 223 (1976) [4] M. Henneaux, I.S. Krasil’shchik, A.M. Vinogradov (eds.), Secondary Calculus and Cohomological Physics, Contemporary Mathematics, vol. 219 (American Mathematical Society, Providence, 1998) [5] I. Krasil’shchik, Algebraic theories of brackets and related (co)homologies. Acta Appl. Math. 109, 137–150 (2010). https://doi. org/10.1007/s10440-009-9445-1, (Preprint. arXiv:0812.4676) [6] I.S. Krasil’shchik, V.V. Lychagin, A.M. Vinogradov, Geometry of Jet Spaces and Nonlinear Diﬀerential Equations, vol. 1, Advanced Studies in Contemporary Mathematics (Gordon and Breach, London, 1986)

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4

423

424

References

[7] I.S. Krasil’shchik, A.M. Verbovetsky, Homological Methods in Equations of Mathematical Physics (Open Education, Opava, 1998) [8] I.S. Krasil’shchik, A.M. Vinogradov (eds.), Algebraic aspects of differential calculus. A dedicated issue of Acta. Appl. Math. 49(3) (1997) [9] I.S. Krasil’shchik, A.M. Vinogradov, Nonlocal trends in the geometry of diﬀerential equations: symmetries, conservation laws, and B¨ acklund transformations. Acta Appl. Math. 15, 161–209 (1989) [10] I.S. Krasil’shchik, A.M. Vinogradov, What is the Hamiltonian formalism? Uspekhi Mat. Nauk 30(1) (1975) (in Russian, English translation in: London Math. Soc. Lecture Notes Ser., 60 (1981)) [11] V.V. Lychagin, Colour calculus and colour quantizations. Acta Appl. Math. 41, 193–226 (1995) [12] S. Morier-Genoud, V. Ovsienko, Simple graded commutative algebras. J. Alg. 323, 1649–1664 (2010) [13] J.A. Schouten, Uber Diﬀerentialkomitanten zweier kontravarianter Grossen. Proc. Ned. Akad. Wet. Amst. 43, 449–452 (1940) [14] A. Verbovetsky, Diﬀerential Operators Over Quantum Spaces. Acta Appl. Math. 49, 339–361 (1997) [15] A. Verbovetsky, Lagrangian formalism over graded algebras. J. Geom. Phys. 18, 195–214 (1996) [16] A.M. Vinogradov, A common generalization of the Schouten and Nijenhuis brackets, cohomology, and superdiﬀerential operators (Russian). Mat. Zametki 47(6), 138–140 (1990). English translation in Math. Notes, 47 (1990) [17] A.M. Vinogradov, Cohomological Analysis of Partial Diﬀerential Equations and Secondary Calculus, vol. 204, Translations of Mathematical Monographs (American Mathematical Society, Providence, 2001) [18] A.M. Vinogradov, From symmetries of partial diﬀerential equations towards secondary (“quantized”) calculus. J. Geom. Phys. 14, 146–194 (1994) [19] A.M. Vinogradov, Geometry of nonlinear diﬀerential equations. J. Sov. Math. 17, 1624–1649 (1981) [20] A.M. Vinogradov, Local symmetries and conservation laws. Acta Appl. Math. 2(1), 21–78 (1984)

References

425

[21] A.M. Vinogradov, Logic of diﬀerential calculus and the zoo of geometric structures. Geom. Jets Fields 110, 257–285 (Banach Center Publications, 2016), arXiv:1511.06861v1 (2015) [math.DG] [22] A.M. Vinogradov, A. Blanco, Green formula and transformation. Acta Appl. Math. 83, 149–166 (2004)

legendre

[23] A.M. Vinogradov, A. De Paris, Fat Manifolds and Linear Connections (World Scientiﬁc, 20081), xii+297 pp [24] A.M. Vinogradov, G. Vezzosi, On higher order analogues of de Rham cohomology. Diﬀ. Geom. Appl. 19, 29–59 (2003) [25] A.M. Vinogradov, L. Vitagliano, Iterated Diﬀerential Forms I–V, Dokl. Akad. Nauk, 407(1), 16–18 (2006); 407(2), 151–153 (2006); 413(1), 7–10 (2007); 414(1), 447–450 (2007); 416(2), 161–165 (2007). English translation: Doklady Mathematics, 73(2), 169–171 (2006); 73(2), 182– 184 (2006); 75(2), 177–180 (2007); 75(3), 403–406 (2007); 76(2), 673– 677 (2007) [26] M.M. Vinogradov, Residues without duality. Acta Appl. Math. 49(3), 281–291 (1997) [27] M.M. Vinogradov, n-ary Batalin–Vilkovisky brackets [seminar talk], Geometry of Diﬀerential Equations Seminar (Independent University of Moscow, Moscow, Russia, 2010). https://gdeq.org/Seminar talk, 3 March 2010

Index

Symbols C ∞ -closed geometrical R-algebra, 33 K-point of a k-algebra, 93 P -symbols, 277 Der-operator, 359 R-point, 22 A adapted coordinates, 172, 237 adjoint operator, 384 algebra Jordan, 350 of jets, 233 of cosymbols, 276 of de Rham cohomology of commutative algebra, 364 of diﬀerential forms, 320 of ﬁrst order jets at the point, 116 of smooth functions, 37 of the Cartesian product, 48 of symbols, 141 commutativity, 142 coordinates, 147 Lie algebra structure, 142

atlas, 56 compatibility, 56 countability condition, 57 dimension, 56 Hausdorﬀ condition, 57 maximal, 56 B base (of a ﬁbration), 152 Batalin-Vilkovisky bracket, 407 operator, 407 Berezinian, 385 biderivation universal, 304 billiards on a disk, 4 Boole groups, 421 Boolean algebra, 86, 417 Boolean space, 418 boundary of manifold, 41 point, 41 bracket Poisson, 344 bundle algebraic deﬁnition, 157 geometrical deﬁnition, 158 base, 158

© Springer Nature Switzerland AG 2020 J. Nestruev, Smooth Manifolds and Observables, Graduate Texts in Mathematics 220, https://doi.org/10.1007/978-3-030-45650-4

427

428

Index

f -morphism, 167 ﬁber of, 158 lifting of sections, 168 locally trivial smooth, 152 of l-jets, 136, 161 section, 162 pullback, 165, 197 regular f -morphism, 168 restriction, 166 section, 161 total space, 158 triviality criterion, 167 vector, 172 tensor product, 192 C category Modpf C ∞ (M ), 191 VBM of vector bundles, 173 diﬀerentially closed, 248 of coordinate manifolds, 75 of ﬁbrations, 154 of geometrical modules, 249 of smooth manifolds, 68 as smooth R-algebras, 68 change of rings, 157, 197 change of structure operator, 308 chart, 53 compatibility, 55 special, 111 coﬁltration bimodule, 274 commutative graded algebra, 390 complete geometrical R-algebra, 31 complex of integral forms, 385 conﬁguration space, 1 contact distribution, 346, 349 form, 349 canonical coordinates of, 349 manifold, 349 structure, 349 contact ﬁeld, 357 generating section of, 361 cosymbols, 274 cotangent bundle, 115, 161, 173 section, 162 cotangent manifold, 113, 115

special atlas, 114 cotangent space algebraic deﬁnition, 115 at a point, 113 of a manifold, 113 of a spectrum, 118 countability condition, 57 covector ﬁeld, 223 covering, 159 cross, 19, 81 algebra of symbols, 147 Hamiltonian mechanics, 149 jet algebra, 236 module of diﬀerential operators, 135 of vector ﬁelds, 200 smooth functions algebra, 81 tangent space, 120 vector ﬁeld, 125 curvature, 347 of distribution, 347 D de Rham cohomology, 331 de Rham complex, 304, 307, 330 inﬁnite analog, 307 de Rham complex of the algebra, 331 δ-cohomology, 328 δ-homology, 326 derivation at a point of the spectrum, 117 graded, 333 of an algebra, 124 of the algebra with values in a module, 129 derivation of vector bundles, 359 diﬀeomorphism, 69 diﬀerential exterior of 1-forms, 303 of a function, 225 at a point, 113 of a smooth map, 107 in coordinates, 109 diﬀerential 1-forms, 225, 228 morphisms, 232 diﬀerential forms of second degree higher analog, 299

Index diﬀerential operator in algebras, 132 in modules, 138 diﬀerentially closed category, 248 diﬃety, 265 diﬀunctor, 244 Dif< (k,l) , 295 Dif> (k,l) , 295 Diﬀ< k , 250 Diﬀ> k , 252 Difσ , 305 Dk , 307 D(k) , 250 Dk,l , 313 D(l,k) , 296 Dσ , 305 exact sequence, 296 diﬀunctors transformation natural, 294 dimension of a chart, 53 of a smooth R-algebra, 37 of an atlas, 56 diole, 396 direct sum of vector bundles, 184 discrete valuation ring, 368 distribution, 346 contact, 346, 349 curvature, 347 curvature map, 347 dimension, 347 regular, 347 point of, 347 singular point of, 347 domain of a K-point, 93 double pendulum, 53, 55, 60 dual complex, 385 dual space, 85 dual space |F|, 22 topology, 25 E exact sequence, 175 of diﬀunctors, 296 exterior diﬀerential higher analog, 306 naturality of, 335

429

of 1-forms, 303 exterior product, 302 of 1-forms, conceptual deﬁnition, 303 of diﬀerential k-forms, 319 F f -morphism of bundles, 167 regular, 168 ﬁber of a module, 175 over a point, 154 ﬁber bundle, 158 ﬁbration, 152 ﬂow, 352 form contact, 349 canonical coordinates of, 349 symplectic, 344 functionality, 179 functor D, 129 representing object, 228 S −1 , 156 Diﬀl , 137, 139 representing object, 137, 236 representing object in category of geometrical C ∞ (M )-modules, 241 Γ, 177, 179 HomA (P, ·), 181 exact, 181 gluing, 254 of change of rings, 197 functors transformation natural, 244 G Γ-invariant function, 45 Γ-invariant maps, 70 Gauss map, 169, 187 generating section of contact ﬁelds, 361 geometrical k-algebra, 89 C ∞ (M )-module, 177 and pseudo-bundles, 199

430

Index

R-algebra, 23 C ∞ -closed, 33 complete, 31 restriction, 30 smooth, 37 smooth envelope, 35 geometrization of modules, 176 of the algebras, 89 germs of smooth functions along the subset U , 208 at the point x, 156 ghosts, 98 gluing homomorphism, 254 gluing transformation, 294 graded algebra, 387 graded derivation, 333 graded module, 389 Grassmann algebra, 397 Grassmann manifold, 58 tautological bundle over, 160, 186 as a subbundle of the trivial bundle, 163 section, 162 group GL(n), 59 SO(3), 3, 74, 75 SO(4), 75 SO(n), 59 action, 45 H Hadamard’s lemma, 17 Hamiltonian formalism in T ∗ M , 148 vector ﬁelds, 149 Hausdorﬀ condition, 57 hedgehog theorem, 162 Hopf ﬁbration, 159 map, 74 I implicit function theorem, 75 inﬁnitesimal symmetry, 356 of algebras, 356 of distributions, 357 of manifolds, 356

of tensors, 358 inverse function theorem, 75 invisible element, 89, 177 J dδ-complex, 328 Jz1 M , 116 Jacobi bracket, 361 Jacobi matrix, 107 jet algebra Jz1 M , 116 Jzl M , 135 J l (M ), 233, 234 bundle J 1 M , 116 J l P , 236 J l M , 136, 161 geometrical, 256 manifold J 1 M , 116 J l M , 136 module Jk P ), 255 J (P ; M), 264 Jk (P ; M), 259 modules J l (P ), 239 naturalness, 265 of a function, 136 of inﬁnite order, 263 vector space Jzl P , 237 jets reduced, 262 jets of algebra, 259 Jordan algebra, 350 K k-derivation, 310 Kahler diﬀerential, 321 Killing ﬁelds, 358 Klein bottle, 40 ﬁbered over the circle, 153 L Legendre transformation, 290

Index Leibnitz rule graded, 333 Leibniz rule, 104, 117, 122, 124, 129 Lie derivative, 352 of diﬀerential forms, 353 lift, 168 lifting of sections, 168 line discrete, 57 long, 57 with a double point, 57 local coordinates, 55 localization, 155 of diﬀerential operators, 203, 218 of representing modules, 270 of smooth algebras, 155 locally trivial smooth bundle, 152 M manifold contact, 349 nonorientable, 343 orientable, 343 orientation of, 343 Poisson, 344 symplectic, 343 manifold of jets of ﬁrst-order, 116 of order l, 136 maximal multiplicative set, 214 spectrum, 97 M¨ obius band, 40, 66, 158, 167, 185 direct sum, 165 module of jets, 254 of sections, 174 of volume forms, 385 module of derivations, 124 Lie algebra structure, 130 with values in a module, 129 Monge–Amp`ere equation, 350 multiplicative set, 155 equivalence, 213 maximal, 214 noncompatibility, 212

431

N natural diﬀerential operator, 269 natural transformation of diﬀunctors, 294 of functors, 244 representing object, 246 non-observable element, 89 O observable and observability principle, 101, 149, 415, 417 operator adjoint, 384 orientation of manifolds, 343 P parallelizable manifold, 161 parity form, 390 phase space of billiard, 4 Poisson bracket, 344 manifold, 344 Poisson bracket, 149 Poisson diole, 399 polyderivation, 309, 310 universal, 317 polyvector, 310 ﬁeld, 310 prime spectrum, 93, 96 product ﬁbration, 152 projection (of a ﬁbration), 152 projective module, 180 projective space RP 3 , 3, 74 RP n , 47, 58 propositional algebras, 417 pseudo-bundles, 177 and modules, 199 pseudo-Riemannian metric, 358 pullback bundle, 165 algebraic deﬁnition, 166, 197 pullback vector bundles, 186 Q quotient manifold, 46 R reduced jets, 262

432

Index

regular f -morphism of bundles, 168 representing object for covariant functor, 245 for functor D, 228 Diﬀl , 137, 236 for functor in category of geometrical C ∞ (M )-modules Diﬀl , 241 for natural transformation of functors, 246 residue class, 367, 370 residue form, 366, 370 restriction homomorphism, 30 restriction of a bundle, 166 restriction of an R-algebra, 30 ring of germs of smooth functions, 156 rotating solid, 1, 60 S section of a bundle, 161 of a cotangent bundle, 115 of a pseudo-bundles, 177 self-adjoint endomorphism, 350 smooth algebra, 37, 282 dimension, 37, 41, 108 with boundary, 37 smooth category, 282 smooth envelope, 35 smooth function algebraic deﬁnition, 37 coordinate deﬁnition, 62 smooth manifold algebraic deﬁnition, 37 coordinate deﬁnition, 57 with boundary, 60 smooth map algebraic deﬁnition, 65 coordinate deﬁnition, 72 diﬀerential, 107, 109 of prime spectra, 97 smooth R-algebra, 37 smooth set, 81 tangent vectors, 121 special orthogonal group SO(3), 3

spectrum maximal, 97 prime, 93, 96 Spencer d − δ-complex, 326 Spencer δ-complex, 326 Spencer δ − Diﬀ-complex, 326 Spencer Diﬀ-complex, 324 of second order, 301 Spencer Diﬀ-sequence, 323 extended, 324 inﬁnite, 325 of 1-st order, 251 Spencer diﬀerentials, 327 Spencer jet cochain complex of order m, 327 Spencer jet cohomology, 327 Spencer jet complex of second order, 304 Spencer operator, 315 Spencer transformation, 315 Stone’s theorem, 418 subbundle, 163 of a vector bundle, 182 subﬁbration, 155 submanifold, 41 substitution of the derivation into the 2-form, 305 support of a module, 178 symbol of a diﬀerential operator, 141 symplectic algebra, 344 form, 344 manifold, 343 triple, 344 T tangent bundle, 112, 160, 173 section, 162 tangent covector, 113 tangent manifold, 110 special atlas, 111 special chart, 111 tangent space, 104 basis, 106 coordinate change, 107 tangent vector at a point of a manifold, 104 at a point of the spectrum, 117

Index theorem, 105 tangent vector ﬁeld along a map, 128 tangent vector ﬁeld along a submanifold, 127 tautological bundle, 160, 162, 186 Taylor expansion, 17 thermodynamics of an ideal gas, 8 topology in |P |, 178 in the dual space M = |F|, 25 Zariski, 94 total space (of a ﬁbration), 152 triviality criterion for bundles, 167 U universal 1-form, 227 biderivation, 304 polyderivation, 317 universal derivation algebraic case, 228 descriptive deﬁnition, 225 geometrical case, 227 universal diﬀerential operator, 137, 235 universal vector ﬁeld, 128 of an algebra, 290 V vector bundle, 172

adapted coordinates, 172 direct sum, 184 module of sections, 174 morphisms, 173 tensor product, 192 IM , 172 Whitney sum, 184 OM , 172 vector ﬁeld, 122 along maps, 127 as a smooth section, 123 on submanifolds, 126 transformation, 123 universal, 128 volume form, 343 W Whitney sum, 163 of vector bundles, 184 Z Zariski topology, 85, 94 zero section, 174

433