You are on page 1of 278

roUJNJ])A'II'll(Q)N~ (Q)1f ID>llIFIFlEBN'II'~ILlE mrndl ILlilE (GJm(Q)UJIF~




GE~METRiA mf:::]~;i~::\L
F. I. LL


Seott.Foresman and CGmpuy G~lIIinois


This textbook fills a long-standing gap. The beginning graduate student finds it hard to learn the basic material on differentiable manifolds. All the books he is referred to give a cursory treatment and quickly move on to more specialized topics. For this reason, Professor Warner's book is especially welcome. Here is a clear, detailed, and careful development of the fundamental facts on manifold theory and Lie groups. Numerous problems extend the theory and help the student master the subject. An added bonus is the sheaf-theoretic proof of the de Rham theorem and an elementary proof of the Hodge theorem. As far as I know, the latter proof is the only one in the literature easily accessible to the novice in analysis. 1. M. Singer

This book provides tho necessary foundation for students interested in any of the diverse areas of mathematics which require tho notion of a diKereatiable manifold. It is designed as a beginDiIl8graduate-level textbook and presumos a good undergraduate training in algebra and analysis plus some knowledge of point set topolo&)', covering spates, and the fundamental group. It is also intended for use as a reference book since it illCludes a number of items which are difficult to ferret out of the literature, in particular. thecompleteand self-contained proofs of the fundamental theorems of Hodge and de Rham. The core material is contained in Chapten 1, 2, and 4. This includes differentiable manifolds, tangent vecton, submanifolds, implicit function theorems, vector fields, distributions and the FrobeDius theorem, diKereatial forms, integration, Stokos' theorem, and de Rham cohomolo&),. Chapter 3 treats the foundations of Lie group theory, including the relationship between Lie groups and their Lie algebras, the exponential map, tho adjoint representation, and the closed subgroup theorem. Many examples are given, and many properties of tho classical groups are derived. The chapter concludes with a discussion of homogeneous manifolds. The standard reference for Lie group theory for over two decades has been Chevalley's Theory of Lie Groups, to which I am greatly indebted. For the de Rham theorem, which is the main goal of Chapter 5, axiomatic sheaf cohomolo&), theory is developed. In addition to a proof of the strong form of the de Rham theorem-the de Rham homomorphism given by integration is a ring isomorphism from the de Rham cohomolo&), ring to the differentiable singular cohomolo&), ring-it is proved that there are canonical isomorphisms of all the classical cohomolo&), theories on manifolds. The pertinent parts of all these theories are developed in the text. The approach which I have followed for axiomatic sheaf cohomology is due to H. Cartan, who gave an exposition in his S/minaire 1950/1951. For the Hodge theorem, a complete treatment of the local theory of elliptic operators is presented in Chapter 6, using Fourier series as the basic tool. Only a slight acquaintaDCC with Hilbert spaces is presumed. I wish to thank Jerry Kazdan, who spent a large portion of the summer of 1969 educating me to the whys and wherefores of inequalities and who provided considerable assistance with the preparation of this chapter. I also benefited from notes on lectures by J. J. Kohn and Stephen Andrea, from sevoraJ papen of Louis Nirenberg, and from Partial Differential


Equations by Bers, John, and Schechter, which the reader might wish to consult for further references to the literature. At the end of each chapter is a set of exercises. These are an integral part of the text. Often where a claim in a chapter has been left to the reader, there is a reminder in the Exercises that the reader should provide a proof of the claim. Some exercises are routine and test general understanding of the chapter. Many present significant extensions of the text. In some cases the exercises contain major theorems. Two notable examples are properties of the eigenfunctions of the Laplacian and the Peter- Weyl theorem, which are developed in the Exercises for Chapter 6. Hints are provided for many of the difficult exercises. There are a few notable omissions in the text. I have not treated complex manifolds, although the sheaf theory developed in Chapter S will provide the reader with one of the basic tools for the study of complex manifolds. Neither have I treated infinite dimensional manifolds, for which I refer the reader to Lang's Introduction to Differentiable Manifolds, nor Sard's theorem and imbedding theorems, which the reader can find in Sternberg's Lectures on Differential Geometry. Several possible courses can be based on this text. Typical one-semester courses would cover the core material of Chapters I, 2, and 4, and then either Chapter 3 or S or 6, depending on the interests of the class. The entire text can be covered in a one-year course. Students who wish to continue with further study in differential geometry should consult such advanced texts as Differential Geometry and Symmetric Spaces by Helgason, Geometry of Manifolds by Bishop and Crittenden, and Foundations of Differential Geometry (2 vols.) by Kobayashi and Nomizu. I am happy to express my gratitude to Professor 1. M. Singer, from whom I learned much of the material in this book and whose courses have always generated a great excitement and enthusiasm for the subject. Many people generously devoted considerable time and effort to reading early versions of the manuscript and making many corrections and helpful suggestions. I particularly wish to thank Manfredo do Carmo, Jerry Kazdan, Stuart Newberger, Marc Rieffel, John Thorpe, Nolan Wallach, Hung-Hsi Wu, and the students in my classes at the University of California at Berkeley and at the University of Pennsylvania. My special thanks to Jeanne Robinson, Marian Griffiths, and Mary Ann Hipple for their excellent job of typing, and to Nat Weintraub of Scott, Foresman and Company for his cooperation and excellent guidance and assistance in the final preparation of the manuscript. Frank Warner

2 5 8 11 22

Preliminaries Di1rerentiable Manifolds The Second Axiom of Countability Tangent Vecton and Differentials Submanifolds. Diffeomorphisms, and the Inverse Function Theorem Implicit Function Theorems Vector Fields Distributions and the Frobeaius Theorem Exercises



TENSORS AND nIFFERENnAL 54 62 69 13 77

Tensor and Exterior Algebras Tensor Fields and Differential Forms The Lie Derivative Dift'ereatialldeals



89 92

101 102
110 112 117




Lie Groups and Their Lie Algebras Homomorphisms Lie Subgroups Coverings Simply Connected Lie Groups Exponential Map Continuous Homomorphisms Clesed Subgroups The Adjoint Representation Automorphisms and Derivations of Bilinear Operations Homogeneous Manifolds Exercises

and Forms

~ 138 140 153 157



Orientation Integration on Manifolds de Rham Cohomology Exercises

~ 163 173 176 186 189 191



105 107 114 116

Sheaves and Presheaves Cochain Complexes Axiomatic Sheaf Cohomology The Classical Cohomology Theories Alexander-Spanier Cohomology de Rham Cohomology Singular Cohomology tech Cohomology The de Rham Theorem Multiplicative Structure Supports Exercises

120 122 127 240 143 150 251 160 162 265



The Laplace-Beltrami Operator The Hodge Theorem Some Calculus Elliptic Operators Reduction to the Periodic Case Ellipticity of the Laplace-Beltrami Operator Exercises BIBLIOGRAPHY INDEX OF NOTATION INDEX

After establishing some notational conventions which will be used throughout the book, we will begin with the notion of a differentiable manifold. These are spaces which are locally like Euclidean space and which have 'enough structure 10 that the basic concepts of calculus can be carried over. In this lint chapter we shall primarily be concerned with the analogs and implications for manifolds of the fundamental theorems of differential calculus. Later, in Chapter 4, we shall consider the theory of integration on manifolds. From the notion of directional derivative in Euc:lidean space we will obtain the notion of a tangent vector to a differentiable manifold. We wiD study mappings between manifolds and the effect that mappings have on tangent vectors. We wiD investigate the implications for mappings of manifolds of the classical inverse and implicit function theorems. We will see that the fUndamental existence and uniqueness theorems for ordinary differential equations translate into existence and uniqueness statements for integral curves of vector fields. The chapter closes with the Frobenius theorem, which pertains to the existence and uniqueness of integral manifolds of involutive distributions on manifolds.

1.1 Some Bale Notadon and TermfnolOlY Throughout this text we will describe sets either by listings of their elements, for example {aI, ... , a,,}, or by expressions of the form {x: P}, which denote the set of all x satisfying property P. The expression a e A means that a is an element of the set A.. If a set A is a subset of a set B (that is, a e B whenever a e A), we write A. c:: B. If A c:: Band B c:: A, then A equals B, denoted A-B. The negations of e, c:: and - are denoted by" ¢, and "respectively. A set A is a proper subset of B if A c:: B but A " B.

We will denote the empty .t by 1lJ. We will often denote a collection {U.: ex e A} of sets U. indexed by the set A simply by {U.} if explicit mention of the index set is not necessal)'. The "';on of the sets in the collection {U.: ex e A} will be denoted U U. or simply U U.. Similarly, their iIIt.r.ction will be denoted


n .... or simply n U•• .... U.

U. - {a: a belongs to lOme U.}. {a: a belongs to every U.}.


n U. -

The expression f: A - B means that / is a WUlppillg the set A into the of set B. When describing a mapping by describing its etrect on individual elements, we use the special arrow .....; thus "the mapping m .....(m) of A / into B" means that/is a mapping of the set A into the set B taking the element m of A into the element/em) of B. If U c: A, then/I U denotes the restriction all to U, and /(U) - {b e B: /(41) - b for some a e U}. If C c: B, then /-l(C) _ {a e A: /(41) e q. A mapping / is one-to-one (also denoted 1 : 1), or injective, if whenever a and b are distinct elements of A, then/(a) v&/(b). A mapping/is onto, or surjective, if/(A) - B. If/: A - B and g: C - D, then the composition go/ is the map g

0/: /-l(B n C) -


defined by go/(41) - g(f(a» for every a e/n C). For notational convenience, we shall not exclude the case in which /-l(B n C) - 1lJ. That is, given any two mappings / and g, we shall consider their composition go/ as being defined, with the understanding that the domain of go/ may well be the empty set. The cartesian product A x B 0/ two sets A and B is the set of aU pairs (a, b) of points a e A and be B. If /: A - C and g: B - D, then the cartesian product f x g 0/ the maps / and g is the map (a, b) .....(f(a), g(b» of Ax B into C x D. We shall denote the identity map on any set by "id." A diagram of maps such as



is called commutative if g 0/_ h. We shall always use the term function to mean a mapping into the real numbers.

Manlfoltb Let d ~ 1 be an integer. and let JR' - {a: a - (~ ••••• a,) where the a,

are real numben}.

Then JR' is the ti-dlmnuional Euclitktul 6JHlc.. In the cue d - 1. we denote the real lln« JRI simply by JR. The Drigilt (0•...• 0) in Euclidean space of any dimension will be denoted O. The notations [a,b] and (a,b) denote as usual the intervals of the real line a ~ t ~ b and a < t < b respectively. The function r,: JR' - JR defined by r,(a) - a" where a - (411, ••• , a~ e JR', is called the ith (canonical) coordinal. functiolt on JR'. The canonical coordinate function rl on JR will be denoted simply by r. Thus ria) - a for each a e JR. If/: X _JR', then we let

where I. is called the ith COmpoMltlfimctiolt 0/ f. If /: JR - JR and 1 e JR. then we denote the tkrirxztiv.

.!.I U) _ ~
dr , lim/(t

all at

t by


I-lim/(t ....

+ It) -/(t).

If f: JR" - JR., if 1 ~ i ~ It, and if the partial tkrivative all with rupect

ar, ,

1..1 _!l.1 _ .... ar"


r, at


, ••• ,


t by h

e JR", then we denote

, I,,) -/(t) .


... • "-I' I,

+ h, "tl' ...

If P e JR', then B.(r) win denote the open ball of radius r about p. The open ball of radius r about the origin will be denoted simply by B(r). C(r) will denote the open cuN with sides of length 2r about the origin in IR'. That is, C(r) - {(aI, ...• a,) e IR': 141,1< r for all I}. We shall use C to denote the compl.x mmrber ji.ld and C" to denote compl." n-spac«; CIt - {(ZI' ..• , z,,): z, e C for I ~i ~ It}. Unless we indicate otherwise, we shall always use the term neighborhood in the sense of opelt neighborhood. If A is a subset of a topological space, its dosure will be denoted by A. If qI is a function on a topological space X, the support of qI is the subset of X defined by supp II' We use the Kronecker inde"
qI-l(1R -




If ex .. (exl , ••• , -..) is a d-tuple of non-negative integers, then we set [ex] and

I ex"
arl·' ...

If ex -


(0, ... , 0), then we let

ar,,·· .



DIFFER.ENTIABLE 1.2 Definitions

f is differentiable


Let U c IR" be open, and letf: U _IR. We say that of class C" on U (or simply that f is C"), for k a nonnegative integer, if the partial derivatives exist and are continuous on U

for [at] ~ k. In particular,fis CO iff is continuous. Iff: U - IR", thenf is differentiable of class C" if each of the component functions h == r, of is C". We say that f is COO if it is C" for all k ~ O. 1.3 Definitions A locally Euclidean space M of dimension d is a Hausdorff topological space M for which each point has a neighborhood homeomorphic to an open subset of Euclidean space IR". If qI is a homeomorphism of a connected open set U c M onto an open subset of IRtI, qI is called a coordinate map, the functions X~ = r, 0 qI are called the coordinate functions, and the pair (U,qI) (sometimes denoted by (U, Xl , ••• , x.J) is called a coordinate system. A coordinate system (U, qI) is called a cubic coordinate system if qI(U) is an open cube about the origin in IR". If me U and qI(m) == 0, then the coordinate system is said to be centered at m. 1.4 Definitions A differentiable structure' of class C" (1 ~ k ~ eo] on a locally Euclidean space M is a collection of coordinate systems {(U., qI.): at e A} satisfying the following three properties: (a) (b)


U U. ==
• &,A


qI.o qI,-l

is C" for all at, peA.

qI.O qI-l

(c) The collection' is maximal with respect to (b); (U,qI) is a coordinate system such that qI 0 qI.-l and C" for all at e A, then (U,qI) e'.

that is, if are



If '. - {(U•• ,J: ex E A} is any coUection of coordinate syatema satisfying properties (a) and (b). then there is a unique differentiable stJ:UCture' containing ' •. Namely. let


- {(U.,): ,.


and , ••

,-1 ire

C· for all ,.


contains ' •• clearly satisfies (a). and it is euily checked that' satisfies (b), Now' is maximal by construction. and so , is a differentiable structure containing ' •. Cearly' is the unique such ttruc:ture.

We mention two other fundamental types of differentiable ItrUc:tures on locally Euclidean spaces. types that we shall not treat in this text. Damely. the structure of class C· and the complex analytic Itructure. For a diJIerentiable structure 0/ class C·. one requires that the compolitiona in (b) are locally given by convergent power series. For a compkx QMlytic structure on a ld-dimensional locally Euclidean space. one requires that the coordinate systems have range in complex d-space C' and overlap holomOrphically. A d-d_IUlonai diJIM'e"tiabk 1JUIIIifoklof cltw C·(limi1arly C· or complex analytic) is a pair (M,#) consiltins of a d-dimensional. MCODdcountable. locally Euclidean space M together with • differentiable Itructure , of class C·. We shall usually denote the differentiable manifold (M,#) simply by M. with the understandins that when we apeak of the ··differentiable manifold M" we are considerins the locally Euclidean space M with some given differentiable structure'. Our attention will be restricted solely to the case of class C"". so by diJIere"tiable we will always mean diJIerllltiable 0/ class C"". We also use the terminology smooth to indicate differentiability of class C"". We often refer to differentiable manifolds simply as lIUUIi/olds. with differentiabtlity of class C"" always implicitly assumed. A manifold can be viewed as a triple consisting of an underlyins point set. a second countable locally Euclidean topolOgy for this set. and a differentiable structure. If X is a set. by a manifokl Itructure 011 X we shall mean a choice of both a second countable locally Euclidean topolOgy for X and a differentiable structure. . Even though we shall restrict our attention to the C"" case. many of our theorems do. however. have C· veniona for k < 00. which are asentia11y no more complicated than the ones we shall obtain. They limply require that one keep track of degrees of differentiability. for differentiating a C· f\mction may only yield a function of class Ca-l if I ~k < 00. Unless we indicate otherwise. we shall always use M and N to denote differentiable manifolds. and M' win indicate that M is a manifold of dimension d.

1.5 Examplea
(a) The standard differentiable structure on Euclidean space JR' is obtained by taking' to be the maximal collection (with respect to 1.4(b» containing (JR'.i). where i: JR' - JR' is the identity map.

Differentiable (b) Let V be a finite dimensional


real vector space. Then V bas a natural manifold structure. Indeed, if {e,} is a basis of V, then the elements of the dual basis {r,} are the coordinate functions of a global coordinate system on V. Such a global coordinate system uniquely determines a ditrerentiabh structure' on V. This differentiable structure is independent of the choice of basis, since ditrerent bases give C" overlapping coordinate systems. In fact, the change of coordinates is given simply by a constant non-singular matrix.


Complex. II-Space C· is a real 2n-dimensional vector space, and so, by Example (b), has a natural structure as a 2n-dimensional real manifold. If {e,} is the canonical complex. basis in which e, is the a-tuple consisting of zeros except for a 1 in the ith spot, then {el
, ••• ,

e., .[:::iel, ••• , ../-ie.}


is a real basis for C", and its dual basis is the canonical coordinate system on C". (d) The d-sphere is the set

S' - {ae 1R0I+1:I al-



Let II - (0, ... ,0,1) and s - (0, ... ,0, -1). Then the standard ditrerentiable structure on S' is obtained by taking' to be the maximal collection containing (S' - n, p,.) and (S' - s, pJ, where P.. and P. are stereographic projections from nand s respectively.
(e) An open subset U of a ditrerentiable manifold (M,:'- M) is itself a ditrerentiable manifold with ditrerentiable structure



, u - {(U. n U, fII.1 U. n U): (U. ,fIIJ e'M}'

Unless specified otherwise, open subsets of ditrerentiable always be given this natural ditrerentiable structure. manifolds will


The gellerallinear group G/(II,IR) is the set of all n X II non-singular real matrices. If we identify in the obvious way the points of IR "I with II x II real matrices, then the determinant becomes a continuous function on IR"I. G/(n,lR) receives a manifold structure as the open subset of IR "I where the determinant function does not vanish.


Product manifolds.

Let (Ml ,'1) and (M, ,'.) be ditrerentiable manifolds of dimensions d1 and d. respectively. Then x M. becomes a differentiable manifold of dimension d1 + d., with ditrerentiable structure' the maximal collection containing




fII. x '1',:



1R'1 X IR'I): (U. ,fII.) e'I' (V, ,'1',) e'.}.



1.6 De8nldo... Let U c: M be open. We say that I: U - IR is a C"" fimction on U (denof#J Ie C""(U» if 1 fP-1 is C"" for each coordinate map fP on M. A continpous map ,,: ~ 7 Nis said to be differentiable olca C""(denof#J" e C""(M,N)orsimply"e CC) ifg is a C'" function on y-l(domain of g) for all C"" fuDctions g defined on open sets in N. Equivalently. the continuous map" is C'" if and only if fP T-1 is C'" for each coordinate map T on M and fP on N. Clearly the compositioD of two differentiable maps is again differentiable. Observe that a mapping ,,: M - N is CtCif and only if for each m e M there exists an open neighborhood U of m auch that U is C"".





The second axiom of countability has many consequences for manifolds. Among them, manifolds are normal, metrizable, and paracompact. Paracompactness implies the existence of partitions of unity, an extremely useful tool for piecing together global functions and structures out of local ones, and conversely for representing global structures as locally finite sums of local ones. After giving the necessary definitions, we sllall give a simple direct proof of paracompactness for manifolds, and sllall then derive the existence of partitions of unity. It is evident that manifolds are regular topologieal spaces and their normality follows easily from this and the paracompactness.. We sballieave the proof that manifolds are normal as an exercise. For the fact that manifolds are metrizable, see [13]. A collection {U.} of subsets of M is a cover of a set U. is open. A subcollection of the U. which still covers is called a subcover. A refinement {V,} of the cover {U.} is a cover such that for each P there is an at such tbat V, c: U.. A collectioD {A.} of subsets of M is locally jinlte if whenever m e M there exists a neighborhood W of m such that W n A. " f2J for only finitely many Ct. A topological space is paracompact if every open cover has an open locally finite refinement. 1.7 Deftnido...
W c: M if W c:

U U.. It is an open cover if each



1.8 DeftnidOD A partition 01 ""ity on M is a collection {qI,: I e I} of C"" functions on M such that (a) (b) The collection of supports {supp qI,: i e I} is locally finite. ~ qI,(p) - 1 for all p e M, and qI,(p) ~ 0 for all p e M and I e I.


A partition of unity {qI,: i e I} is subordinate to the cover {U.: at e A} if for each I there exists an at such that supp qI, c: U.. We say that it is subordinate to the cover {U,: I e I} with the same index set as the partition 01 ""Ity if supp qI, c: U, for each t et.

Second Axiom

0/ Countabllity

1.9 "",rna Let X be a topological 'pac. which Is locally compact (.ach point has at least OM compact Mighborhood), Hausdorff, and second countable (manifolds, for ."ample). Then X is paracompact. In fact, .ach open cover has a countable, locally finite r.fiMment consisting of open s.ts with compact closures.
PROOF We prove first that there exists a sequence {G,: i-I, of open sets such that G, is compact,

2, ... }


G, c: Gl+l,

Let {U,: i-I, 2, ... } be a countable basis of the topology of X consisting of open sets with compact closures. Such a basis can be obtained by starting with any countable basis and selecting the subcollection consisting of basic sets with compact closures. The fact that X is Hausdorff and locally compact implies that this subcollection is itself a basis. Now, let G1 - U1• Suppose that
G. - U1U ••• U USt'


Then letji+l be the smallest positive integer greater thanjl; such that

and define
Gi+l =



This defines inductively a sequence {GI;}satisfying (1).


, I I I I I





Let {U.: a e A} be an arbitrary open cover. The let G, - GI-l is compact and contained in the open let Gft-I - G,.... For each i ~ 3 chOOllCa finite subcover of the open COVet {U. n (Gft-l - G,...>: a e A} of G, - G'_l' and cboose a fiDite subcover of the open cover {U. n G.: a e A} of the compact set G•• This collection of open sets is easily seen to be a countable, locally finite refinement of the open cover {UJ, and consists of open seta with compact closures.

1.10 Lemma There exist I a ftOIt-Mgfltlve C"" Jimction , on IR' which «JUIIu 1 on the closed cube C(l) and uro 011 the complement of the open cube

We need only let,


be the product


° rJ ...


° r~,

where h is a non-negative C"" function on the real line which is 1 on [-1,1) and zero outside of (-2,2). To construct such an h, we start with the function (2)
f(t) -

f' -I'.


t>O t~O

which is non-negative, C"", and positive for t

> O.

Then the function


h(t) - get


is non-negative, C"", and takes the value 1 for t ~ 1 and the value zero for t ~ O. We obtain the desired function h by setting

+ 2)g(2 -


1.11 Theorem (ExJstence of Parddoas of Unity) Let M be a differentiable manifold and {U.: a e A} an open cover of M. Then there existl a countable partition of unity {,,: i-I, 2, 3, ... } subordinate to the cover {UJ with supp " compactfor each I. If one doel not require compact supportl, then there il a partition of unity {,J subordinate to the cover {U.} (tlult is, supp ,. c: UJ with at mOlt countably many of the ,. not identically zero.

Let the sequence {G,} cover M as in 1.9(1), and set G. - e. For p e M, let I. be the largest integer such that p .eM - G, . Choose an a. such that p e U. , and let (V,T) be a COOrdinate system centered at p such that V c: U... n (G' .. I - G,) and such that T(V) contains + the closed cube C(2). Define




"" -



on V elsewhere


Tangent V.ctors and Differentials


where fII is the fun~on 1.10(1). Then,," is a CO" function on M which has the value I 011 some open neighborhood Willofp, and has compact support lying in V c: U•• n (G'.+I _;. G. ). For each i ~ 1, choose a finite set of points p in M whose corresponding Willneighborhoods cover G. - G'_l' Order the corresponding "" functions in a sequence J - 1, 2, 3. . . •. The supports of the form a locally finite family of subsets of M. Thus the function




,,-I~ 1-1
fII. --.


is a well-defined CO" function on M, and moreover ,,(P) p e M. For each i-I, 2, 3, ..• define


for each


Then the functions {fII,: i-I, 2, 3, •.. } form a partition of unity subordinate to the cover {U.} .with supp fII. compact for each t. If we let fII. be identically zero if no fII. has support in U., and otherwise let fII. be the sum of the fII. with support in U. , then {fII.} is a partition of unity subordinate to the cover {U.} with at most countably many of the fII. not identically zero. To see that the support of fII. lies in U., observe that if 091 is a loCallyfinite family of closed sets, then U A - U A. Observe, however. that the support of fII. is not necessarily compact.
Ae." Ae."

". ,,-


CoroUary Let G be open in M, and let A be closed in M, with A'c: G. Then there exists a Coo function fII: M - IR such that (a) 0

s fIICp) s I for

all p e M.

(b) fIICp) - I ifp e A. (c) supp fII c: G.


There is a partition of unity {fII,lI'} subordinate to the cover {G, M - A} of M with supp fII c: G and supp 11' c: M - A. Then fII is the desired function.


1.12 A vector v with components VI , •••• v, at a point p in Euclidean space IR' can be thought of as an operator on differentiable functions. Specifically, iff is differentiable on a neighborhood of p. then V assigns to f the real number v(f) which is the directional derivative off in the direction v at p. That is, (1)




1biJ operation of the vector important properties. (2)


on diff'erentiabie functions satisfies two


! I


+ 19) -

v(/) + Av(g)j,

- f(p)v{t)

+ g(p)l1(/).

whenever f and g are differentiable near p. and 1 is a real number. The fint property says that II acts linearly on functions. and the 8eCond says that II is a tUrivalion. This motivates our definition of tangent vectors on manifolds. They will be directional clerivativel. that is. linear derivations on functions. The operation of taking derivatives depends only on local properties of functions. properties in arbitrarily small neighborhoods of the point at which the derivative is being taken. In order to express most conveniently this dependence of the derivative on the local nature of functions, we introduce the notion of germs of functions. 1.13 De8DitioD. Let m E M. Functions f and g defined on open sets containing m are said to have the same prm at m if they agree on some neighborhood of m. Thil introduces an equivalence relation on the Ceo functions defined on neighborhoods of m, two functions being equivalent if and only if they have the same germ. The eguivalence classes are called P""'". and we denote the set of lermI at m by F", • Iff is a C"" function on a neighborhood of m, then f will denote its aerm- The operations of addition. scalar multiplication, and multiplication of functions induce on 1", the structure of an algebra over IR. A &erm f bas a well-defined value f(m) at m, namely. the value at m of any representative of the germ. Let F", c: 1", be the set of germs which vanish at m. Then F", is an ideal in 1", • and we let F"," denote its ktb power. F,.· is the ideal of 1", consisting of all finite linear combinations of k-fold products of elements of F",. These form a descending sequence of ideals 1", ::;,F", ::;,F",· ::;, F,.· ::;,.... 1.14 De1lnidOD tion of the algebra (a) (b)

1",. That is. for all r. 1 e 1 and 1 e IR. + la) - v(f) + 1v(l).

A tangent vector

II at

the point m eM is a linear deriva-

v(f· I) - f(m)v(l)

+ I(m)v(f).

M", denotes the set of tangent vectors to Mat m and is called the tangent space to Mat m. Observe that if we define (II + w)(f) and (1I1)(f)by


+ w)(f) -

1I(f) + w(f) (1I1)(f)- A(v(f)

whenever II, we M", and A e IR, then II + wand 1I1again are tangent vectors at m. So in this way M", becomes a real vector space. The fundamental property of the vector space M'" • which we shall- establish in 1.17. is that its dimension equals the dimension of M. This definition of tangent vector

____ -_L


Tangent Vectors and Differentials


is not suitable in the Ck case for 1 ~ k < co. (We will discuss the CJ: case further in 1.21.) We give this definition of tangent vector for several reasons. One reason is that it is intrinsic; that is, it does not depend on coordinate systems. Another reason is that it generalizes naturally to higher order tangent vectors, as we shall see in 1.26. 1.15 If e is the germ of a function with the constant value c on a neighborhood of m, and if v is a tangent vector at m, then v(c) .. 0, for
v(c) .. cv(I),


vel) .. v(l .1) = Iv(l)

+ IV(I) ..


1.16 Lemma M",isnaturallyisomorphicwith(F",/F",I)*. denotes dual vector space.)


(The symbol

If v e M"" then v is a linear function on F", vanishing on F",I because of the derivation property. Conversely, if Ie (F",/F",I)*, we define a tangent vectorv(atm by setting vtCf)- I({f - f(m)}) forf e F",. (Here f(m) denotes the germ of the function with the constant value f(m), and { } is used to denote cosets in F.,./F ",I.) Linearity of v( on F'" is clear. It is a derivation since v(f·g) .. I({(f - f(m»(ggem»~+ f(m)(a - gem»~ + (f - f(m»g(m)}) - I({f .... - f(m.(m)}) g

- I({(f - f(m»(g - gem»~})+ f(m)t({a - gem)}) + g(m)t({f - f(m)}) - f(m)v(g) + g(m)vtCf).

Thus we obtain mappings of M", into (F",/F",I)*, and vice versa. It is easily checked that these are inverses of each other and thus are iso morphisms. 1.17 Theorem dim (F.,./F",I) - dim M. IRd,

The proof is based on the following calculus lemma [31]. Lemma (1)
If g is of class Cit (k ~ 2) on a convex open set U about pin then for each q e U, g(q) -

g(p) +

1-1or, "

i :g I

(r,(q) - rl(p»

+ I (r,(q)

- r,(p»h(q)

- rJ(p»f1(1

'n particular, if g e C.." then the second summation in (I) determines element of F"I since the integral as a function of q is of class COD •


- t) ~

or, urJ









---... --~'~I
~ ~-



"'., on a neighborhood of m. where" e CO. Thus



f au;r,

1.17 Let (U,f) be a coordinate system about m with eofunctions Xl, ••• , x. (4 - dim M). Let f e F. . Apply f-1, and compose with f to obtain

1 I

I (~- ~m»+ I '.1


(x, - x,(m) )(XI - XAm»h







(x. - x.(m»modF.1.

Hence ({XI - x.(m)}: 1- 1, .•. ,tI} spans F.JF.I. Consequendy dim F"JF.I ~ 4. We claim that these elements are linearly independent. For suppose that


But this implies that

aa rl
for j - 1•...• Corollary

(ID,(r, - r,(f{m»»



4, which implies that the

must an be zero.

dim M. - dim M.

1.18 In practice we win treat tangent vectOR as operating on functions rather than on their germs. If / is a differentiable function defined on a neighborhood of m, and v eM. , we define

v(f) - v(f).

Thus v(f) - vCg) whenever / and g agree on a neighborhood of m, and clearly (2) v(j + Ag) - v(f) + Av(g) (A e IR), v(j· g) -/(m)v(g) + g(m)v(f).

where / + Ag and /. g are defined on the intersection of the domains of definition of/and g.

1.19 De8nfdon
Xl' .••

Let (U.f) be a coordinate system with coordinate ,X, , and let me U. For each 1e (I, .•. ,d), we define

i .

Tangent Vectors and Differentials


a tangent vector (8/8x()I. eM. by setting (1)

8 ( 8x(


(f) -

_8(fo ,-1)


for each function / which is c- on a neighborhood of m. We interpret (1) as the directional derivative of/at m in the x( coordinate direction. We also use the notation

N..I - (.!.I )(f). 8~. 8~.

Remarks on 1.-l9 Clearly «%xJI ...)(f) depends only on the germ of/at m, and (a) and (b) of 1.14 are satisfied; so (O/OXi) is a tangent vector at m. Moreover, {(%x()I ...: i-I, ... ,d} is a basis of M Indeed, it is the basis of M... dual to the basis {{XI - xl(m)}: i -= I, ,d} of F"JF.I since



If v eM., then

o~ • (x, -


x,(m» -


Simply check that both sides give the same results when applied to the functions (XI - xi(m». (c) Suppose that (U,!p) and (V,¥,) are coordinate systems about m, with coordinate functions Xl , ..• , Xd and Yl , ••• ,Y d respectively. Then it follows from remark (b) that

-} ,


s. I...- (-18Yli I... .!.I i 8x( ox(.· 8Yi

Observe that (%x() depends on !p and not only on X(. In particular, if Xl were equal to Y1 , it would not necessarily follow that 0/OX1 equals

.. .. •

8/0Y1 .


If we apply Definition 1.19 to the canonical coordinate system rl , ... ,r. on IR·, then the tangent vectors which we obtain are none other than the ordinary partial derivative operators (alar,).

1.21 Our proof of the finite dimensionality of F"JF.I certainly fails in the CJ:case for k < co since the remainder term in the lemma of 1.17 will not be a sum of products of CJ: functions, and the lemma doesn't even make sense in the Cl case. In fact, it turns out (see [21» that F"JF".I is always infinite dimensional in the CJ:case for I ~ k < co. There are various ways to define tangent vectors in the CJ: case in order that dim M.,. - dim M (all of which work in the ClIO case, too). One way is to define a tangent vector v at m as a mapping which assigns to each function (defined and differentiable of class CJ:on a neighborhood of m) a real number v{f) such that if (U,!p) is a coordinate system on a neighborhood of m. then there exists a list of real numbers (aI, ... , a.) (depending on !p) such that
v(J) - ~a,


a(/o ,-1) f



Then the space M. of tangent vectors again turns out to be finite dimensional, with a basis {(alax')I.}. 1.22 The DUl'erentlal Let,,: M - N be ClIO, and let III diffemltial 01 tp at m is the linear map
(I) dtp: M. -Nyl.,.)

e M. The

defined as follows. If v e M". • then tbp(v) is to be a tangent vector at tp(m). so we describe how it operates on functions. Let g be a CrJO function on a neighborhood of tp(m). Define d,,(v)(g) by setting
(2) tbp(v)(g) - v(g


It is easily checked that dtp is a linear map of speaking. this map should be denoted dtp ,or omit the subscript m when there is no possibility is called non-singular at m if dtp. is non-singular, cons~sts of 0 alone. The dual map


M". into NyC.,.). Strictly simply dtp.,.. However, we of confusion. The map tp that is, if the kernel of (I)




(3) 6tp: is defined as usual by requiring that (4)


I ,

1.,.) -


6tp(co)(v) - co(dtp(v»

1 I·

I: M

whenever coe N;I.) and ve M".. In the special case of a ClIO function -IR, if v e M.,. and/(m) - ro, then
df(v) -

v(f).i...I. r, dr

I t

Tangent Vectors and Differentials In this case, we usually take df to mean the element of M! defined by (6) df(v) - v(j).


That is, we identify dfwith 6f(0), where 0) is the basis of the l-dimensional space IR: dual to (d/dr)lr . Particular usage will be clear from the context.

1.23 Remarks on 1.22

(a) Let (U, Xl, •.• ,X.) and (V'YI' ... ,Yt) be coordinate systems about m and ,,(m) respectively. Then it follows from 1.22(2) and 1.20(b) that '" aYI "c",,' The matrix {a(y, ,,)/ax,} is called the Jacobian of the map " (with respect to the given coordinate system). For maps between Euclidean spaces, the Jacobian will always be taken with respect to the canonical coordinate systems.

d,,(.!. 1'" ) - Ia(YIax, ax,

0 ,,)

1 .1..1

(b) If (U, Xl" •• ,X.) is a coordinate system on M, and {dx,I",} is the basis of M! dual to {a/ax,I",}. If f: M function, then df", (c)

mE U,

• I g[_

then IR is a CtJO



Chain Rule. Let ,,: M - Nand IP: N - X be CtJO maps. Then

or simply d(IP dIP dtp. It is a useful exercise to check the form that this equation takes when the maps are expressed in terms of the matrices obtained by choosing coordinate systems. (d) If ,,: M - Nand f: N - IR are CtJO,then 1J"(df..(,,,,) - d(/o ,,)"" for 1J"(df.,c,,,,)(v) - df(d,,(v» - d(/o "),,,(v) whenever v E M",. (e) A C.., mapping 0': (a, b) - M is called a smooth curve in M. Let t E (a, b). Then the tangent vector to the curt'e 0' at t is the vector
0 ,,) 0

d( IP 0



d"", ,



E M~(t). 0'

We shall denote the tangent vector to

at t by aCt).



Now. if v is any element of M_. then v is the tanlent vector to a smooth curve in M. For one can simply choose a coordinate system (U,,). centered at m, for which


Then v is the tangent vector at 0 to the curve t ....,-I(t. O••••• 0). One should observe that many curves can have the same tanpt vector. and that two smooth curves a and .,. in M for which a(t.> 'r'{t.) - m have the same tanaent vector at t. if and oaly if d(f
a) , _ d(f

T) ,



for all functions / which are COO on a neipborhood of m. If a happens to be a curve in the Euclidean space IR·. then
a(t) -

~I!.\ ar
I .M ) V\t -

1 .U)

+ ... +

da·l..!.l u; ar.

If we identify this tangent vector with the element

(da11 •... , da·l) dr dr


of IR·, then we have I' a(t 1m ....

+ h)

- a(t)


Thus with this identification our notion of tangent vector coincides, in this special case. with the geometric notion of a tangent to a curve in Euclidean space. 1.24 Theorem Ltt '" be a Coo mapping 0/ the connected manifold M into the manifold N. Suppose that for each m eM, ehp_ • O. Then '" is a constat map. Let n e ",(M). y-l(n) is closed. We need oaly show that it is open. For this, let m e y-l(n). Choose coordinate systems (U, Xl>' •• ,x.) and (V.y" ... ,Yo) about m and " respectively. so . that ",(U) c: V. Then on U, .

0- d",(..!.)_ :i

8(y( 0 "') 8 8X1 8y(

() _ 1•... , I),

which implies that 8(y(o "').0

8xI (I -

I, ...• c; ) - 1. . . . • I) .

Tangent Vector, tmd Diff"..",u,/s


Thus the functions I. •" are coastant on U. 1bia implies that ,,(U) - 11; hence ,,-1(11) is open and consequently M.

r<n) -

We shall now see that in a natural way the collection of all tangent vectors to a differentiable manifold itself forms a differentiable manifold called the tangent bundle. We have a similar dual object called the cotangent bundle formed from the linear functionals on the tangent spaces. 1.25 Tangent and Cotangent Bundles with differentiable Let


Let M be a CO manifold



U M••

•• Jl

There are natural projections: (2) 1r: T(M)-M.

Let (U.,) with coordinate functioas IRId and ~o: ("O)"'"l(U)_1R1d by (3)
~v) (Xl(w(V», ..•• x,(w(v»,


,,0: TO(M) M. _

1r{v) - m if veM •• "·(T) - m if T eM!.

Xl ••••• X,.

Define ~: ,,-1(U)dx,(v»

dXl(V) ••.••

~O(T) (X.(".(T», ... , X,("O(T»,T(a:).···. -


for all v e ,,-1(U) and T e ("O)-l(U). Note that, and ~. are both one-to-one maps onto open subsets of 1R1d. The following steps outline the construction of a topology and a differentiable structure on T(M). The construction for T*(M) goes similarly. The proofs are left as exercises. (a) If (U,!p) and (V,,)

e', then.;p


is Coo.

, •i




The collection {~-1(W): W open in 1R1d, (U,!p) e.F} forms a basis for a topology on T(M) which makes T(M) into a 2ddimensional, second countable, locally Euclidean space. Let!F be the maximal collection, with respect to 1.4(b), containing
((,,-1(U),~): (U,,)




!F is a differentiable structure on T(M).

I ,

T(M) and T*(M) with these differentiable structures are called respectively the tangent bundle and the cotangent bundle. It will sometimes be convenient to write the points of T(M) as pairs (m,v) where m e M and v eM. (and similarly for T*(M».




If ,,: M - N it a of the taDgeDt buDdlea (4)

c- map, &Ileathe
il¥: T(M)

dift'enmtial of " deftDea a mappinl

- T(N),

where il¥<m,v) - il¥.(v) wheDever ve M.. It i. euily checked that (4) haC- map.

l.26t IDgber Order Taageat Vecton and Dfft'erentIaJa It is useful to look at M. as (F./F.I)*, for this point of view allows an immediate leneralizapon to hiaher order taDgeDt vectors. We dipess for a moment to live these definitions. Recall that F. i. the alpbra of lerma of functions at m. F. c: F. is the ideal of germs vanishing at m, and F.· (k an intepl' ~ 1) is the ideal of F. consisting of all finite linear combinations of k-fold products of elements of F•. The vector space F.!F!+l is called the space of kth order dffferentillis at m, and we denote it by • M.. As before, f denotes the germ of lat m, and { } will denote cosets in F .!F!+l. Letlbe a differentiable function on a neiahborhood of m. We define the kth order differentilll ,pI 011at m by


{f - f(DI}}.

A kth orthr tangent vector at m is a realliDear function on F. vanishinl on J?!+l and vanishing also on the set of germs of functions constant on a neiahborhood of m. The real linear space of kth order tangent vectors at m will be denoted by M••. We have a natural identification of M." with ("M.)* since any kth order tangent vector restricted to F. yields a linear function on F. vanishing on J?!+l, and hence yields an element of {"M,J*; and conversely an element of (AM.)* uniquely determines a linear function on F. vanishing on F!+1, and this extends uniquely to a kth order tangent vector by requirina it to annihilate pnna of constant functions. We can tie up this notion of hiaher order tangent vector with the usual notion of hiaher order derivative in Euclidean space by lookinl at the forms that these tangent vectors and dift'erentials take in a coordinate system. Let (U,qI) be a coordinate system about m with coordinate functions Xl, •.• ,X. such. that ,( U) is a convex open set in Euclidean space IR·. Let at - (~ ••••• Gt,J be a list of non-neptive integers. In addition to our conventions of 1.1, we let

(x - x(m»- - (Xl - xl(m»-I ••. (X. - x.(m)r.

Letl be a C'" function on U. Then it follows from the lemma of 1.17. that

t The material or this section will Dot be UIec:I elsewhere in the IkippecI without loss or continuity.

book. and 10 it may


Tangent Vectors and Differentials


where the Ia. are Crt> functions on U and where (3)

(4) Thus the collection (5)

[{(x -s(m»"}: 1 ~ [at] ~ k]

spans tM",. The proof that these elements are linearly independent in t M", is the obvious generalization of the proof for the case k - 2 which was treated in 1.17. Thus the collection (5) forms a basis oftM",. Consequently t M", is finite dimensional with dimension equal to the binomial coefficient




1). As the dual space of M",.


M ",t is also finite dimensional

with the same dimension. Since M",t is identified with (tM",)·, and these spaces are finite dimensional, we have a canonical isomorphism ofkM", with (M",~·, under which the element of dkj'etM"" considered as an element of (M",t)., satisfies

dkf(v) _


· .• ,

Let (7)

PL ax· '"





. ·.

Since the derivative is linear, and since the value of 01l0x- at m depends only on the germ off at m and vanishes if I is constant on a neighborhood of m or if lis an lot] + l-fold product of functions which vanish at m, then is an [ot]th order tangent vector at m. It follows that



.1)1: {(ott ax· I...: 1 s lot] s k\


is the basis of M ",t dual to the basis (5) of t M "'. If v is a kth order tangent vector at m, then (9) where




ax· ...


In terms of the basis (8), equation (3) becomes (11)

a _-- 1 0-[


ax· .,'

As in the case of tint order tangent vectoR, we customarily think of tanlent vectors as operatinl on the functions themselves rather than their lerms; indeed, we define

whenever f order at m. Finally, just as there are natural mappinp oftanaent vecton and differentials associated with a differentiable map ,: AI _ N, 10 are there linear mappinp (13) dJ:,: M.J:_N~.), ~,: J:N.(.) _J:M.

v(j) - v(f) is ClIO on a neipborhood of m and v is a tanlent vector of any

defined by

dJ:,(v)(g) _ v(g lJi,(tPg) _ dt(g 0,)

0 ,),

whenever v e M.J: and g is a CrJO function on a neipborhood of ,(m). It is easily checked that (14) does indeed define the mappinp (13) and that the mappinp dJ:, and lJi, are dual. Our definition of a tint order tanpnt vector in this section aarees with Definition 1.14 in view of Lemma 1.16. Moreover, we have seen three interpretations of the tint order differential df of a function f; the interpretation (13) agrees with our original definition 1.22(1). the interpretation (6) aarees with 1.22(6). and we have the additional interpretation (1).


(a) (c)



M _ N be ClIO.

is an immersion if ehp. is non-singular for each m e M.


(b) The pair (M.


is a subnumifold of N if tp is a one-to-one immersion.

is an imbedding if tp is a one-to-one immenion which is also a homeomorphism into; that is. tp is open as a map into y(M) with the relative topolegy,
tp is a diffeomorphism is CrJO.




maps M one-to-one onto N and y-l

1.21 Remarks on 1.27 One can. for example. immerse the real line IR into the plane. as illustrated in the followinl figure. so that the tint case is an immersion which is not a submanifold. the second is a submanifold which is not an imbeddinl. and the third is an imbeddinl.

Sub11llllJilolds, iffeomorphisms,lnverse Function Theorem D


-I I
Immersion, but Dot. SUbmaniCold


0 -- ()

, - lim • .(1)



Submanifold, but Dot an Imbedding

Observe that if (U,q» is a coordinate system, then q>: U - q>(U) is a diffeomorphism. The composition of diffeomorphisms is again a diffeomorphism. Thus the relation of being diffeomorphic is an equivalence relation on the collection of differentiable manifolds. It is quite possible for a locally Euclidean space to possess distinct differentiable structures which are diffeomorphic. (See Exercise 2.) In a remarkable paper, Milnor showed the existence of locally Euclidean spaces ($I is an example) which possess non-diffeomorphic differentiable structures [19). There are also locally Euclidean spaces which possess no differentiable structures at all [14). . If ." is a diffeomorphism, then d'Pm is an isomorphism since both (d'P d'P-1)1,,(m) and (d'P-1 d1p)lm are the identity transformations. The inverse function theorem gives us a local converse of this-whenever d'Pm is an isomorphism, 'P is a diffeomorphism on a neighborhood of m. Before we recall the precise statement of the inverse function theorem, we give a definition which will be needed in the corollaries.
0 0

1.29 Deflnidon A set Yl, ... 'Yi of Ceo functions defined on some neighborhood of m in M is called an independent set at m if the differentials dYl' ... , dy, form an independent set in M! .


1.30 Inverse Funcdon Theorem



U c IRII be open, and let

IRII be Ceo. IJthe Jacobian matrix

is non-singular at E U, then there exists an open set V with roeVe U such that V maps V one-to-one onto the open set fey), and V)-1 is ceo.



{~} 0'1

i.i-I .....II


This is one of the results we shall assume from advanced calculus. For a proof. we refer the reader, for example, to (31) or [6).


Corollary (a) AISII1f'N t/un ,,: M _ N 18 CfIO, that m eM, tmd t/un dyJ: M. - N.,(., 18 an Isomorphism. Then there 18 a Mighborhood U of m 8UCh that ,,: U - ,,(U) 18a di,ff.omorphism onto the o~n .t ,,(U) In N .. Observe that dim M - dim N, say d. Choose coordinate systems (V,III) about m and (W,,.) about ,,(m) with ¥<V) c: W. Let lII(m) - p 1 and ,-(tp(m» -IJ. The differential ()f the map 'r 0 ,,0 111-IIII(V) is non-singular at]. ThUi the inverse function theorem yields a diffeomorphism IX: 0 - «(0) on a neighborhood 0 of p with 0 c: III(Y). Then ,.-1 « III is the required diffeomorphism on the neighborhood U - 1II-1 of m.


Corollary (b) Suppou that dim M - d and that Yl>'" independent ut of functions at m. e M. Then the functions Ylf a coordiruzte system 011 a neighborhood of me.


18 an ,y, form



Suppose that the Yi are defined on the open set U containing Define ,,: U - IR' by ,,(m) - (Yl(m) , ••• ,y,,(m» (m e U).

Then " is C", Now ~ is an isomorphism on (IR:'. ,). since .which implies that talces a basis to a basis. Consequently, the differential ehp•• (which is the dual of is an isomorphism. So the inverse function theorem implies that" is a diffeomorphism on a neighborhood V c: U of me, and consequently the functions Yl' ... ,y, yield a coordinate system when restricted to V.

6"1,,,•• ,


- d(ri

0 ,,)


6"1,,,•• ,)

Corollary (c:) Suppou that dim M - d and that Yl' ..• 'Y'f with I < d, is an independent set of functions at m. Then they form part of a coordiruzte system on a Mighborhood of m.

Submanifolds, Diffeomorphisms, Inverse Function TMorem 25

. ..;.

Let (U, Xl' ••• ,X.) be a coordinate system about m. Then , dy" dxl, •••• u.} spans M!. Choose d - / of the X, so that {dYlt ... , dYIt dx, •... ,dxc._,} is a basis of M:'. Then apply

{dYl' ...

Corollary (b). Corollary (d) Let '1': M - N be ceo, tmd assume that d'l': Mm - N¥(ml is surjective. Let Xl' ...• XI form a coordinate system on some neighborhood of'l'(m). Then Xl 0 '1', ••• , XI 0'1' form part of a coordinate system on some neighborhood of m.


The fact that d'l'm is surjective implies that the dual map 6tp1..(ml is injective. Thus the functions {x, 0 '1': i -= 1, ... ,/} are independent at m since 6'1'{dx,) - d{x, '1'). The claim now follows from Corollary (c).

Corollary (e) Suppose that Yl' ... ' y" is a set of Ceo functions on a neighborhood of m such that their differentials span M!. Then a subset of the y,forms a coordinate system on a neighborhood ofm.


Simply choose a subset whose differentials form a basis of and apply Corollary (b).

Corollary (f)

Let '1': M - N be ceo. and assume that d'l': Mm - N¥(ml is injective. Let Xl' ... , X" form a coordinate system on a neighborhood of 'I'{m). Then a subset of the functions {x, 0 'I'} forms a coordinate system on a neighborhood of m. In particular, 'I' is one-to-one on a neighborhood of m.
PROOF The fact that d'l'm is injective implies that 6tp1,,(ml is surjective. This implies that {d(x, 0 '1') == 6'1'(dxi): i == 1, ... , k} spans M! . This corollary then follows from Corollary (e).

1.31 The situation often arises that one has a Ceo mapping, say '1', of a manifold N into a manifold M factoring through a submanifold (P,q» of M. That is, 'I'(N) C q>(P), whence there is a uniquely defined mapping '1'0 of N into P such that q>0 '1'0 == '1" The problem is: When is '1'0 of class ceo? This is certainly not always the case. As an example, let Nand P both be the real line, and let M be the plane. Let (IR,'I') and (1R,q» both be figure-8 submanifolds with precisely the same image sets, but with the difference that as t - ± 00, 'I'(t) approaches the intersection along the horizontal direction, but q>(t) approaches along the vertical. Suppose also that '1'(0) =: q>(0)= O. Then '1'0 is not even continuous since 'I'o-l( -1,1) consists of the origin plus two open sets of the form (<<,+ 00), (-00,-«) for some « > o.


Suppou that ,,: N M 18C-, that (P,,)
. "

-··-··--l··~· . -.~.
18Q 6UbmJmiJoltJ

of M. tmd that" factDrS tlttwgh (P,,). that 18, ¥<N) c: ,(P). Since, 18 InjectiH, tMre 18 Q UIIitJw mtlppl"g '" of N '~, P",ch that , 0 ". - " • q,to

1.32 Theorem




". is

c- iflt

18 contlltutnu.

(b) '" 18 continuou.r if , 18 an imlJedding. Another important case in which ". is continuoUi occun when (P,,) is an integral manifold of an involutive distribution on M, as we shall _ in


Result (b) is obvioUi. So assume that '" is continuoUi. We prove that it is C-. It suffices to show that P can be covered by eoordinate systems (U•.,.)such that the map.,. restricted to the open set ",-1(U) is C", Let peP, and let (Y.,,) be a coordinate system ODa neighborhood or ,(p) in~. Then by Corollary (f) of 1.30 there exists a projection.". of IR' onto a suitable subspace (obtained by setting eertaiD of the coordinate functions equal to 0) such that the map.,. - .".0 " 0 , yields a coordinate system ODa neighborhood U of p. Then


.,. 0

".1 ",-1(U)

which is C-.

_ .".0" 0" I",-1(U).

.".0" 0, 0 '"

I ",-1(U)

1.33 Further Remarks on Submanlfoldl Submanifolds (Nl.fl) and (N ••,.) of M will be called etpdVtllent if there exists a diffeomorphism IX: Nl - N. such that '1 -


». ........


.... ...

........ ...

, r~
.... "lIo.


This is an equivalence relation on the collection of all submanifolds of M. Each equivalence class E has a unique representative of the form (A./) where A is a subset of M with a manifold structure such that the inclusion map I: A - M is a COO immersion. Namely, if (N.,) is any representative

Stlbmllllifoids. DiffeomorphiRns. Inverse Function TMorem


'..,. ;r.

'¥(N). We induce a manifold Itructure on A by requiring '¥: N - A to be a diffeomorphism. With this manifold structure, (A,i) is a submanifold of Ai equivalent to (N.,¥). This is the only manifold structure on A with the property that (A.I) is equivalent to (N.,¥); thus this is the unique such representative of E. The conclusion ofsome theorems in the following sections state that there exist unique submanifolds satisfying certain conditions. Uniqueness means up to equivalence as defiDed above .. In particular. if the submanifolds of M are viewed as tubsets A c: M with manifold structures for which the inclusion maps are CfI> immersions. then uniqueness means unique subset with unique second countable locally Euclidean topology and unique differentiable structure. In the· case of a submanifold (A,I) of M where i is the inclusion map. we shall often drop the I and simply speak of the submanifold A c: M. Let A be a subset of M. Then generally there is not a unique manifold structure on A aucb that (A,I) is a submanifold of M. if there is one at all. For example, the diagrams in l.3l illustrate two distinct manifold structures on the fi~8 in the plane, each of which makes the figu~8 together with the inclusion map a submanifold of IRI. However. we have the following two uniqueness theorems which involve conditions on the topology on A.

or E. then the subset A of M must be

(a) Let M be a differentiable manifold and A a subset of M. Fix a topology on A. Then there is at most one differentiable structure on A such that (A.i) is a submanifold of M, where i is the inclusion map.

Again let A be a subset of M. /fin the relative topology, A has a differentiable structure such that (A,i) is a submanifold of M, then A. has a uniquemanifold structure (that is, uniquesecond countable locally Euclidean topology together with a unique differentiable structure) such that (A.,i) is a submanifold of M.

We leave these to the reader as exercises. Result (a) follows from an application of Theorem l.32. Result (b) depends strongly on our assumption that manifolds are second countable, and for its proof you will need to use the proposition in Exercise 6 in addition to Theorem 1.32. 1.34 SUces Suppose that (U,'¥) is a coordinate system on M with coordinate functions Xl' ••• , X., and that c is an integer, 0 ::;: c ::;:d. Let a e '¥(U), and let


l, ... ,d}.

The subspace S of M together with the coordinate system





forms a manifold which is a submanifold of M called a slice of the coordinate system (U.'¥).





-...-' ----~,-.


1.35 Proposition LIt ,,: Mo - N' I¥ 1111 ilnlMrsion, tmd let m E M. Then there exists II cubic-centered coordilulte system (Y,III) about ,,(m) tmd II neighborhood U of m such tlult " U is 1:1 f;!wJ ,,(U) is II slice of (Y,III) •

... OOP Let (W,.,.) be a centered coordinate system about tp(m) with coordinate functions Yl"" ,y,. By Corollary (f) of 1.30 we can renumber the coordinate functions so that


T _ ".

° .,.° "

is a coordinate map on a neighborhood Y' of m where ".: IR'-IR· ia projection on the tint c coordinates. Define fUnctions {x,} on (". ° .,.)-l(f(Y'» by setting (2)


(I - 1, .•• , c)



1, ... ,d).

The functions {x,} are independent at tp(m), since at ,,(m),

(i - 1, ... , c) (3) dx - ( , dy,

+,_I dYI I,lIil

(i - c

+ 1•...

, d)

for some constants au. By Corollary (b) of 1.30the {x,} form a coordinate system on a neighborhood of ,,(m). ~t Y be a neighborhood of ,,(m) on which the Xl, •••• x, form a cubic coordinate system. Denote the corresponding coordinate map by III. Let U - .,--1(Y) ('\ V'. Then U and (Y.III) are the required neighborhood and coordinate system.

I -...

SuJmumi/olds, Diffeomorphisms,Inoerse Function Theorem


We emphasize that this proposition only says that there is a neighborhood U of m such that y(U) is a slice of the coordinate system (Y,IP). Even if (M,y) is a submanifold of N, it may well be that y(M) " Y is far from being a slice or even a union of slices. For an example, consider again the figure-8 submanifold of the plane:

• 1ft



However, in the case that (M,y) is an imbedded submanifold, the coordinate system (V,IP) can be chosen so that all of y(M) ("\ V is a single slice of V. Let us now consider the question of the extent to which the set of Ceo functions on a manifold determines the set of Ceo functions on a submanifold. Let (M,'P) be a submanifold of N. Then, of course, if fe C""(N), then f M is a Ceo function on M. (More precisely, f v is a Ceo function on M.) In general, however, the converse does not hold; that is, not all Ceo functions on M arise as the restrictions to M of Ceo functions on N. For the converse to hold, it is necessary and sufficient to assume that y is an imbedding and that y(M) is closed. We prove the sufficiency in the following proposition, and leave the necessity as Exercise 11 below.

1.36 Proposldon Let y: M - N be an imbedding such that y(M) is closed in N. If g e CIIO(M),then there exists fe Ceo(N) such that f y - g.

To simplify notation, we shall suppress the map 'P and consider MeN.

p and an extension of g from 0" ("\M to a Ceo function gIl on 0".

One simply has to take 0" to be a cubic-centered coordinate neighborhood of p for which M ("\ 0" is a single slice, and then define ill to be the composition of the natural projection of 0" onto the slice followed by g. The collection {OIl: p e M} together with N - M forms an open cover of N. By Theorem 1.11, there exists a partition of unity {IPs}, with j - 1,2, ... , subordinate to this cover. Take the subsequence (which we shall continue to denote by {IPs}) such that supp IPs ("\ M ~ 0. For each suchj, we can choose a point Ps such that supp IPs c OIlS. Then f == I IPsg", is a Ceo function on N, andfl M - g.


For each point p e M there exists an open set 0" in N containing





From the inverse function theorem we mall obtain two theorems which will provicle us with an extremely useful way g{.provJp, that c:ertain subleta or manifolds are submanifolds. Uncler suitable conditioaa on a dift'erentiable map. the inverse image of a submanifold of its will be a submanifokl of its domain. We first recall the statement of the classical implicit function theorem. This is limply a local (but somewhat more explicit) version or the first "jmplicit function" theorem (1.38) that we sha1I prove for manifolds. We suggest that the reader supply a proof of 1.37 after readina 1.38.



1.37 JmpUdt PUIldIOD Theorem Let U c: IR..... x IR' I¥ ~'" tmd let f: U - IR' I¥ CfIO. We denote the CImD"ica/ coordiMte system on IR.... X 1R4 by (rl ••..• r..... ~ •...• s,). Suppose that at the point (r, ,.r,) e U

fer, .So) tmd that the matrix


{~L..... ,}

I-l ..... '

is non-singular. Then theJy exists an open neighborhood Y of r. In IR..... tmd an open nt!ighborhood W of s. in IR' such that Y x W c: U. tmd there exisu a C'" map g: Y - W such that for tlGCh (P4) e Y x W



I •
Implicit Function TMorems 31

1.38 Theorem
that P - "l(n) is me P. ~n P has of M, where i is imbedding, and the

Assume that ,,: Me - N' is C-, tltat n is a point of N, noIHmpt)', iI1Id that dtp: M". - N¥(m) is surjective for all a unique manifold structure such that (P,i) is a submanifold the inclusion map. Moreover, i: P - M is actually an dimension of Pis c - d.

According to result (b) of 1.33, it is sufficient to prove that in the relative topology, P has a differentiable structure such that (P,i) is a submanifold of M of dimension c - d. For this it is sufficient to prove that if m e P, then there exists a coordinate system on a neighborhood U of m in M for which P ("\ U is a single slice of the correct dimension. Let Xl, ••• ,X. be a coordinate system centered at n in N. Then since dtp: Mm - N. is surjective, it follows from Corollary (d) of l.30 that the collection of functions {y( - x(

tp: i-I,

... ,d},

forms part of a coordinate system about m e M. Complete to a coordinate system )'1' ••• ,)'. ')'11+1' ••• ,)'e on a neighborhood U of m. Then P ("\ U is precisely the slice of this coordinate system given by
)'1 - )'. •••

= )'.-


In this theorem the inverse image of a point is shown to be a submanifold as long as the differential is surjective at each point of the inverse image. A point can be thought of as a O-dimensional submanifold. We now generalize this theorem by proving that under suitable conditions the inverse images of higher dimensional submanifolds are themselves submanifolds.

1.39 Theorem Assume that tp: M - N· is Ceo and that (oe,cp) is a submanifold of N. Suppose that whenever m e tp-l( cp(O»), then
(I) N.,(m) - dtp(Mm)

+ dcp(O.,-I(.,(m)))

(not necessarily a direct sum). Then if P - 'l'l( cp(O») and is non-empty P can be given a manifold structure so that (P,i) is a submanifold of M, where i is the inclusion map, with (2)

dim M - dimP - dimN - dim O.

Moreover, if (O,cp) is an imbedded submanifold, then so is (P,i), and in this case there is a unique manifold structure on P such that (P,i) is a submanifold ofM.

In general, if (O,cp) is not an imbedding, P need not have the relative topology, and there is no unique manifold structure on P such that (P,i) is a submanifold. We leave it to the reader to supply examples.


The proof will consist of locally reducing this case to the case of Theorem 1.38. Let P EO. By Proposition 1.35. we can pick a neighborhood W of p aDd a centered coo.rdinate system (Y.,,) with coordinate functions Xl ••••• X, about "CbJ'llCh that IP(W) is the slice


(3) Let (4)



(}-1•...,d- c).

R' - R'_·_

w(a) - (ao+l •••••

(5) Let (6) and let (7) U - ",1(Y).
ft' 0

1{IP(W» - {O}.

"1- "

0" 0"

I U: U _



Now, (3) through (7) imply that (8)

1(0). d"l For each point m E M. is surjective since d(" is surjective and since by (1), (5). and (7)


"1-1(0) - ",l(IP(W».




T)(N"(,,.,) - t:/vJl(M".)

+ {O}.

Thus by 1.38. ,,-l( 9(W» has a unique manifold structure such that (.,-l( IP(W»,i) is a submanifold of M; and in this manifold structure,


Implicit Function Theorems


vr1( ,(W» has the relative topology and has dimension equal to dim M - dim N + dim O. Cover 0 by a countable collection {W~: i -= 1,2,3, ... } of such sets. Then (9) p -=


u ,,-1( fP{lf,».

If i ~ j, the submanifolds vr and vrl( intersect in open subsets of each other. It follows that the union of the topologies on the various vrl( forms a basis for a topology onP. With this topology, P is a locally Euclidean space with dimension equal to dim M dim N + dim 0; and, moreover, P is second countable since (9) expresses P as a countable union of second countable open subsets. The manifold structures on the various vr1( are compatible on overlaps because of the uniqueness of the manifold structure on vr1( ,(W,» (or any open subset ofvr1( ,(W,») such that (vr1(,(WJ),i) is a submanifold of M (by result (b) of 1.33). Thus the collection of coordinate systems on P containing the coordinate systems on the various vr1( and maximal with respect to 1.4(b) forms a differentiable structure on P. That (P,i) is a submanifold of M now follows immediately from the fact that the (,,-I(,(W,»,i) are submanifolds. If (0,,) is an imbedding, the coordinate neighborhood V can be chosen small enough so that ,(0) " V consists only of the single slice ,( W), and thus U "p -= ,,-1( ,(W». It follows in this case that P has the relative topology; hence (P,i) is an imbedding. The uniqueness of the manifold structure in this case is guaranteed by result (b) of 1.33.






1.40 Examples
(a) The differential of the function 1(P) -=

except at the origin. Thus the sphere I-l(r"), for a constant r > 0, has a unique manifold structure for which it is a submanifold of IRd under the inclusion map. In particular, this is the same manifold structure as the one defined in Example 1.5(d). (b) We define a map" from the general linear group G/(d,lR) (Example 1.5(f» to the vector space of all real symmetric d X d matrices by
,,(A) -= AA',

'-1 it follows from 1.38 that

I r,(p)"


on IRII is surjective

where A t is the transpose of the matrix A. Let

Oed) - ,,-1(1)

where I is the d x d identity matrix. O(d) is a subgroup of GI(d,lR) under matrix multiplication called the orthogonal group. To apply 1.38 to conclude that O(d) has a unique manifold structure such that (O(d),i) is a submanifold of Gl(d,IR), and that in this manifold structure i is an imbedding and O(d) has dimension i(d(d - I», one


need only check that ilyI. is surjective at each fI e O(tI). For this, it sufth:es to check that dtp, is surjective, since whenever fI e O(tI),

where '. (right trII1I8lt1tlon by 0') is the dift'eomorphism of G/(d,lR) defined by r.( T) -= Ta. We leave the details to the reader u an exercise. VEcrOR PlELDS

1.41 De8DldoDl Smooth curves 0': (a,b) - M and their tangent vecton 6(t) were defined in l.23(e). We say that a mapping 0': [a,b] _ M is a smooth clIIVe In M if 0' extends to be a CO mapping of (a - ., b + s) into M for some, > O. The curve 0': [a,b] - M is said to be piecrNise smooth if there exists a partition a - OCt < «1 < ... < at. -= b such that 0' [«, '«'+1] is smooth for each 1-= 0, •.. , n - 1. Observe that piecewise smooth curves are necessarily continuous. If 0': [a,b] - M is a smooth curve in M, then its tangent vector

6(1) -=

dO'(:,D eM.w

is well-defined for each t e [a,b]. 1.42 De8DidODI A Hctor /kId X along a CUIW fI: [a,b] - M is a mapping X: [a,b] - T(M) which lifts 0'; that is, " X -= 0'. A vector field




[• I





Yector Flcrllb


X is called a smooth <CO) vector field along a if the mapping X: [a,b]T(M) is C", A vector field X 011 till open .t U In M is a lifting of U into T(M), that is, a map X: U - T(M) such that
" 0

X - identity map on U.

Again, for the vector field Xto be smooth (C.,) means that X e c«'( U,T(M». The set of smooth vector fields over U forms in the obvious way a vector space over IR and a module over the ring CO(U) of c«' functions on U. If X is a vector field on U and m e U, then X(m) (often denoted X.) is an element of M",. Iffis a C., function on U, then X(j) is the function on U whose value at m is X",(j).

1.43 Proposldon
equivalent: <a) X Is (b)

IAt X be a vector field on M. Then the following are

{a,} is the


(U, Xl' ••• , X,) is a coordinate system on M, tmd collection of functions on U defined by

•a x] U-!a,;-,
then a, (c)


e C"(U).
C"(V), then X(j) e C"(V).

Whenever V is open in M tmdfe


<a) => (b) The fact that X is smooth implies that U is smooth; and since the composition of differentiable maps is again differentiable, it follows that dx, X U is smooth. (Recall that the dx, are coordinate functions on .".-1(U) c T(M), 1.25(3).) But


dx, (b) => (c) It suffices to



X(j) U e C"(U) where system on M for which

Hence the a, are C., fimctions on U. prove that

(U, Xl" •• , X,) is an arbitrary coordinate U c V. But, by (b),

and the right-hand side is a C., function on U.

(c) => (a) To prove that X is C." it suffices to prove that X U is c«' where (U, Xl" •• , xtl) is an arbitrary coordinate system on M. To prove that X U is c«' , we need only check that X U composed with the canonical coordinate functions 1.25(3) on 71'~l(U) are C., functions. Now, X, 71' X U - x, and the, X U X(x,), all of which are c«' functions on U.
0 0





1.44 LIe Bracket If X and Yare smooth vector fields on M. a vector field [X.YJ called the LIe bnIck.t of X and Y by setting (I) 1.45 (b) (c) Proposition (a) [X.YJ i6",..d" Tff,ge mrooth Hclor!kld on M. COO(M). IMII fIX,gYJ -/g[X,f] [X.l1.(f)- X.( Yf)- Y.(Xf).

we define

+ f(Xg)Y

- g(Yj)X.'

[X.YJ - -[Y,zJ. !kids X, Y, II1Ul Z on M.

(d) ([X,YJ.Z]

+ ([Y.ZJ,x] + ([Z,X],Y] -

O/or allmrooth vector

We leave the proof as an exercise. Part (d) is known as the Jacobi Itknlily. A vector space with a bilinear operation "Satisfying(c) and (d) is called a Li. algebra. 1.46 Deflnidon Let X be a smooth vector field on M. A smooth curve in M is an inlegral CII1W 0/ X if
6(t) I



(I) for each in the domain of



1.47 Let X be a COO vector field on M, and let m e M. Let us now consider the question: Does there exist an integral curve of X through m, and if so, is there a unique one? A eurve v: (a,6) - M is an integral curve of X if and only if (1)

d"(:r I) - X(,,(t»

(t e(a,b».

Let us interpret this in coordinates. Suppose that 0 e (a,b) and ,,(0) - m. Choose a coordinate system (U,.:p) with coordinate functions Xl' ••• , X, about m. By 1.43(b),
(2) where the" are C'" functions on U. Moreover, for each I such that ,,(1) (3)

d (d

" dr,I) - fd(X1o")1 ,ox, I dr 0




Thus, in view of (2) and (3), equation (I) becomes (4)

Vector Fields 37

Thus " is an integral curve of X on ,,-1( U) if and only if

(i -= 1, .•. , d and t E ,,-1(U»,

where", -= x, Equation (5) is a system of first order ordinary differential equations for which there exist fundamental existence and uniqueness theorems [11]. These theorems, when translated into manifold terminology, give the following. 1.48 Theorem Let X be a C'" vector field On a differentiable manifold M. For each m E M there exists a(m) and b(m) in IR U {± co}, tIIIIi a smooth curve
(1) such that " ...: (a(m),b(m» M


(a) 0 E (a(m),b(m»

and " ...(0) -= m.

" ... is an integral curve of



If p:

(c,d)M Is a smooth curve satisfying conditions (a) and (b), then (c,d) c (a(m),b(m» and p -= " ... 1 (c,d).

We continue with the statement of the theorem after the following. Definition

For each


IR, we define a transformation X, with domain


!iJc -= {m

M: t


by setting
(3) (d) Xt(m) -= " ...(1). For each m E M, there exists an open neighborhood an e > 0 such that the map (4) (t,p)

V of m and


is defined and is C'3J from (-e,e) (e) (f) (g) (h)

V into M.

!?it is open for each t.



-= M.
is a diffeomorphism with inverse X_t.

X,: ~t -!?i_t

Let sand t-be real numbers. Then the domain of X, 0 X, is contained ill but generally not equal to ~Ht. However, the domain of X, 0 X, is ~,+t in the case in which sand t both hare the same Sign. Moreover, on the domain of X, 0 Xt we haoe (5) X.

X, = XHt .


t -----38 Mtlllifoids We let (a(m),b(m» be the UDion of aU the open intervals which contain the origin and which are domains of integral curves of X satisfying the initial condition that the origin maps to m. That (a(m),b(m» " f1J (and hence part (f) holds) follows from an application of the fundamental existence theorem [11, THEOREM 4, p. 28] to the system 1.47(5). Now if at and {J are integral curves of X with domains the open intervals A and B (with A n B " f1J), and if at and {J have the same initial conditions at(I,) - {J(I,) at some point t, E A n B, then the subset of A n B on which at and {J agree is nonempty, open by the basic uniqueness theorem [11, THEOltEM 3, p. 28], and closed by continuity; and hence this subset is equal to A n B by the connectedness of A n B. It follows that there exists a curve )'. defined on (a(m),b(m» and satisfYing parts (a), (b), and (c). The existence of an e > 0 and a neighborhood Y of m such that the map (4) is defined on (-e,e) x Vis the content of THEOREM 7 on p. 29 of [11]. That the map (4) is smooth (and hence part (d) holds) follows from THEOREM 9 on p. 29 of [11] on the differentiability of the solutions of 1.47(5) with respect to their initial values. Next We prove part (h). Let IE (a(m),b(m». Then 8 .... )'",(1 + 8) . is an integral curve of X with the initial condition 0 .... )'",(1) and with maximal domain (a(m) - I, b(m) - I). It follows from part (c) that



(a(m) - I, b(m) - I) - (a()'",(I»,b()'",(I»),

and for s in the interval (6),


" -


+ 8).

Now let m belong to the domain of X, X,. Then IE (a(m),b(m» and s E (a()' .. (I»,b()' .. (t»), so by (6).8+ IE (a(m),b(m». Thus mE !iI.... " and (5) follows from (7). It is easy to construct examples to show that the domain of X,O X, is generally not equal to !iI..... (Consider, for , example, the vector field alari on IRa - to} with s - -1 and I - 1.) If, however, s and I both have the same sign, and if m E!iI...." that is, if s + IE (a(m),b(m», then it follows that IE (a(m),b(m» and, by (6), s E (a()' .. (I».b()'",(I»); hence m is in the domain of X, X,. Parts (e) and (g) are trivial if I - 0, so assume that I > 0 and that m E!iI,. (A similar argument will prove (e) and (g) if 1<0.) It . follows from part (d) and the compactness of [0,1] that there exists a neighborhood W of ),,,,([0,1» and an e > 0 such that the map (4) is defined and CO on (-e.e) X W. Choose a positive integer" large enough so that tln E (-e,e). Let ~ - X". W, and let Wi - ~-1(W). Then for i - 2, ...• n we inductively define


at, -

X". I W'_i

W, - ate-i( Wc-J.




is a COO map on the open set WC-l c:: W. It follows that W. is an open subset of W. that W. contains m (since if X,/. composed with itself" times is applied to m, we obtain y",(t), which Jies in W), and that by part (h),


atl° eta

0 •••


I W.

- X,

I W•.

Consequently, W. c:: g,; hence g, is open, which proves part (e). Finally, X, is a I: I map of ~, onto g_,with inverse X_,. That X, is C'" (similarly for X_,) follows from (8), which locally expresses X, as a composition of C'" maps. Hence X, is a diffeomorphism from g, to e.; which proves part (g) and finishes Theorem 1.48.

1.49 DeflD1doDS A smooth vector field X on M is complete if g, ...M for aU t (that is, the domain of y", is (- eo.eo for each m « M). In this case, the transformations X, form a group of transformations of M parametrized by the real numbers called the I-parameter group of X. If X is not complete, the transformations X, do not form a group since their domains depend on I. In this case, we shall refer to the coJ1ection of transformations X, as the loeall-parameter group of X. I.SO Remarks A simple example of a non-complete vector field is obtained by considering the vector field a/arl on the plane with the origin removed.. If a > 0, the domain of the maximal integral curve through (a,O) is (-a,+CX». In the case in which the manifold M is compact, .any COO vector field on M is complete. We leave the proof as an exercise.
'P (that is, X

Let 'P: M - N be COO. A smooth vector field X along C"'(M,T(N» and 'II' 0 x ...'P) has local C'" extensions in N if given m E M there exist a neighborhood U of m and a neighborhood Y of 'P(m) such that 'P(U) c:: Y, and there also exists a C'" vector field ~ on Y such that

1.51 DefinfdoD


1.52 Remark It is easy to prove that a C'" vector field X along an immersion 'P: M - N always has local C'" extensions in N. However, if 'P is not an immersion, such extensions generally do not exist. Consider the following example. We shall first define a smooth vector field X along a smooth curve at: IR - IR in the real line. Let
(I) at(t) ... la,

(that is, at ... ra where r is the canonical coordinate function on IR), and let (2) X(t) - i(t) ... dat(:r



T(1R1) ----


... ..


,.. l,i

-+1R1 I


Since at is a homeomorphism, there is induced a vector field X on 1R1 so that the above diagram commutes. Now, X is a smooth vector field along at, but X is not a smooth vector field on 1R1. To show this, let u - t·. Then (3)

X. - X.e.) --

X(I) -

1 -

dati dr



- 31







dr •

Thus (4)

X - 3rl/'


and the function rll• is not differentiable at the origin. If we extend a mapping into the plane by setting (5)

to be

at(l) -


and again we let X(t) - Ii(t), then X is a smooth vector field along the one-to-one CfID curve at in the plane, which admits no local CfID extension to a neighborhood of (0,0).

1.53 Proposldon Let m e Md, and let X be a smoolh vector field on M such that X(m) ~ O. Then there exists a coordinate system (U,,) with coordinatefunctions Xl, ••. , Xd on a neighborhood of m such that

IU -

..!..I u.

PROOF Choose a coordinate system (V,T) centered at m with coordinate functions Yl' ... ,y" such that (2)

"Yl ,. It follows from 1.48(d) that there exists an e > 0 and a neighborhood W of the origin in 1R'-l such that the map aCt, a., ... , ad) - Xt(

X,. -



(0, a., •.. , a,»

is well-defined and smooth for (t, aI' ... , a,) e (-e,e) x WeIR'.

Distributions and the Frobenius Theorem



a is non-singular
arl X",

da(.!.I) -


at the origin since and

aYl '"


aYI '"

(i ~ 2).

Thus by Corollary (a) of 1.30, qI - crl is a coordinate map on some neighborhood U of m. Let Xl ••••• X, denote the coordinate functions of the coordinate system (U.qI). Then since da(aa

. .

(t •••••••••• ,




we have



-..!.I u.
vector fields X on M

1.54 Definition Let qI: M - N be CO. Smooth and Y on N are called qI-related if dql 0 X Yo qI.

1.55 Proposition Let qI: M - N be CfID. Let X and Xl be smooth vector fields on M, and let Y and YI be smooth vector fields on N. If X is .:p-related to Y, and if Xl is qI-related to YI, then [X,Xl] is .:p-related to [Y. Yl].

let m

We must show that dql 0 [X,Xl] = [Y, YI] Mandie C:"'(N). Then we must show that dql( [X,xl]m)(f)




(2) dql([X,XI]m)(f)

= [Y, Yl];>(",,(f).

We simply unwind the definitions:

= [X,XI]m(f = Xm{Xl(fo = Xm{(dql

- Xm{YI(f) 0


qI» XI)(f)

Xllm (X(fo Xllm{Y(f) YII;>(m'(Y(f)


- Xllm {(dql


qI) -


- dql(Xm){YI(f) Y;>(m,{YI(f) [Y, YI];>(m,(f).

- dql(Xllm>{Y(j»




1.56 Definitions Let c be an integer, I S; c S; d. A e-dimensional distribution 9 on a d-dimensional manifold M is a choice of a c-dimensional subspace 9(m) of M m for each m in M. !!} is smooth if for each m in M there is a neighborhood U of m and there are c vector fields Xl' ... , Xe of class C'" on U which span 9 at each point of U. A vector field X on M is said to belong to (or lie in) the distribution !i) (X E 9) if X me gem) for each



m eM. A smooth distribution II is called bwolutive (or compktely Integrabk) if [X, Y) e II whenever X and Yare smooth vector fields lying in !II. 1.57 De6nition A submanifold (N,,,) of M is an 1nt.".,,1 mtI1Iifold of a distribution II on M if (I) dyl(N,J - 1I(,,(n» for each"

e N.

! ,

Our object in this section is to prove that a necessary and sufficient condition for there to exist integral manifolds of II through each point of M is that II be involutive. Perhaps a word of explanation is in order about the expression "completely Integrabk" sometimes used in place of "involutive." We have required integral manifolds to be submanifolds whose tangent spaces coincide with the subspaces determined by the distribution. One could define a weaker notion of integral manifold by requiring only that the tangent spaces of the submanifold be contained in but not necessarily equal to the distribution at each point. It is possible for a distribution !II to be "integrable" in the sense that it hu low-dim.ensional "integral manifolds," but not completely integrable in the IeDIe that II does not have integral manifolds of the maximal dimension. For us, unless specified otherwise. integral manifolds of distributions will always be taken to mean integral manifolds of maximal dimension. that is, u defined in 1.57. 1.59 Proposition each point of M there

1.58 Remarks

~ ,~


Let !II ~ a smooth distributio" on M IUCh that through Q1I Integral manifold of II. Then !II i6 /nfJolutive.

Let X and Y be smooth vector fields lying in !II. and let m e M. We must prove that [X. y)", e !II(m). Let (N,,,) be an integral manifold of !II through m, and suppose that f1{1Io) - m. Since tJ.p: N.- !II(,,(n» is an isomorphism at each n in N, there exist vector fields~. on N such that


tJ.p tJ.p

It is easily checked that ~ [X, Y) are tp-related. Hence [X, y)", -

r _ Yo". and r are smooth.



0 ".

dyJ([~.l]..) e !II(m).

By 1.55. [~, 1] and

1.60 Theorem (FrobeDfus) Let !II N a c-dimmsional. inrJOlutive. ell) distribution on M~. Let m e M. Then there exists Q1I Integral mQIIifold of !II passing through m. Indeed. there exists a cubic coordinate system (U,'P) which is centered at m, with coordinate .functions x" ••• ,x, IUCh that the slices
(1) x, - constant

for all

t e {c

+ 1, ...•


are integrall'1lllnij'olds of !II; ad if (N,,,) i6 a COlfMcted Integral mQIIifold of!ll such that ,,(N) c:: U, then f1{N) lin In OM of these slices.


.~ -~I~·--


Distributions tmd the Frol¥"iu1 Theorem



We sball prove the existence part of the theorem by induction on e. For the case e - I, choose a vector field X lying in g, defined on an open neighborhood of m, such that X(m) "0. Then Proposition 1.S3 yields a coordinate system (U,cp) about m, which can be taken to be cubic centered, for which X U - a/ax1• Hence the theorem holds for e - 1. Now assume that the theorem holds for e - 1; we prove it for a distribution g of dimension e. Since g is smooth, there exist smooth vector fields Xl' ... , Xc spanning g on a neighborhood. P of m. By 1.S3, there exists a coordinate system (V,]u ... ,].) centered at m, with V c: p. such that


a XdV-a-. Y1

On V, let (3) Y1Xl'

(i -

Y, - X, - X.(Y1)X1

2, ... ,e).

Then the vector fields Y1,..., Yc are independent C spanning !I in V. Let S be the slice ]1 - 0, and let (4) (S)


vector fields

z, ==

Yel S



2, ... , e).

Then since (2) and (3) imply that Y'(Y1)

== 0


== 2, ... , e),

q e S. The Z, span a smooth c -

the Z. are actually vector fields on S; that is, Z.(q) e S" whenever 1 dimensional distribution on S.


MfIIIl/olib We claim this is involutive. Indeed, the Z, are I.related (inclusion map of S in M) to the Y" and therefore, by 1.55, their Lie brackets are also i.related to the corresponding ~~~ets of the But [Y., Y/1, (1,1 ~ 2), has no component in the Y1 t:liieetiOD (apply to 11 and get 0). Therefore, there exist functions cu. such that




[Y,'Y/1 -

on Y, and thus
(7) [Z,,z/1 -




__ I





This proves that the distribution on S is involutive. By the induction hypothesis, there exists a centered coordinate system w..... , w. on some neighborhood ofm in S such that the slices defined by w, - con. stant for aU i E {c + 1, ... , d} are precisely the integral manifolds of the distribution spanned by Z., ... , Z. on this neighborhood. The functions (8)

"I -

w, 0 ff'

(j- 2, ..• , d),

where ff': Y - S is the natural projection in the y coordinate system, are defined on some neighborhood of m in M, are independent at m, and they all vanish at m. Thus there is a cubic-centered coordinate system (U, 'P) with the coordinate functions "1' ... , ". on a suitable neighborhood U of m. We now prove that (9) y,("e+9') • 0 on U (I - 1, ... , c; r - 1, ... , d - c). From this it follows that the vector fields form a basis for !I at each point of U, and thus the slices (1) are integral manifolds To prove (9), fil'lt observe that (8) implies that

a/a"I,' .. , a/a"e



~ _ {I aYl 0

(J - 1) (J - 2, ... , d)

on U; and thus (2), (3), and (10) imply that (11) ~-

a aXl

on U, Now let IE {2, ••• , c} and

so certainly (9) holds for i-I. r E {I, ... , d - c}. By (11), ._ .(12)

Distributio1lS tmd the Probenius Theorem

of !I implies that there are C«> functions


The involutivity (13)

C't such that

[li.Yc] - t-l IeuYt·


Using (13), (12) becomes (14)


• (Y-<xe+r» -= IcuYt(x.+t')



2, ...•


r -= 1, ...• d - c).

Slices x. -



JC. -


Fix a slice of U of the form X. -= constant, ... , x, -= constant. On such a slice, Y,(x.+t') is a function of Xl alone, and (14) becomes a system of c - 1 homogeneous linear differential equations with respect to Xl. Such a system has a unique solution with given initial values [II]. Since the system is homogeneous, the 0 functions give a solution. But each such slice has a unique point in S f"'I U, and on S f"'I U,



-= Z,(we+r) -= 0


-= 2,

•.. , c).

The first equality follows from (4) and (8), and the second from the fact that the integral manifolds of the distribution on S determined by the Z, are given by suitable slices in the w coordinate system. It follows from (14) and (15) that the functions Y,(xc+t') must be identically zero on U. Thus (9) holds, and the induction step is completed. Finatty, suppose that (N,,,,) is a connected integral manifold of !I such that ",(N) c U. Let 'IT be the projection of IR' onto the last d - c coordinates. Then vectors in !I are annihilated by d('IT 0 'P). Thus


'P ",)1.


for each n E N. By 1.24, 'IT 0 'P 0 '" is a constant map since N is connected. Thus ",(N) is contained in one of the slices (I). 1.61 Remarks The classical version of the Frobenius theorem appears quite different from our version in 1.60. The classical Frobenius theorem can be formulated as follows.



Let U and V be open sets in IR· and IR" respectively. We use coordinates '" ... ,r. on IR· and I" ••• ,1" on IR". Let (1)
b: U


V -M(tt,m)

be a CO map of U x V into the set of all n x m real matrices. and let (r.,1.) e U x V. If

ab" I' (2) ab., ab., ..- b1,-- b) - 0 ---+ ~ (ab" a" ar, aS aS
1_1 I I

(I - 1•...•

,,; y.




on U x V. then there exist neighborhoods U. of and a unique C«J map (3) such that if ar.,(r) - «(r.l) then (4) «a(r.) I.

r. in

U and V. of

in V

11.: U.x V.- V


d«,lr - b(r.«(r.s»

for all (r,l) e U. x V•. Equation (4) is a so-called total differential equation. We specify in (I) what the differential of a map should be as a function of the graph, and in (2) we have a necessary and sufficient condition for the existence of such a map with the specified initial conditions. It can be shown that this version is equivalent to 1.60. For example, if we start with a c-dimensional, involutive, C«J distribution ~ on M" and a point m e M. then we can obtain 1.60 from the classical version as follows. We can first choose a coordinate system (W ..,) about m with coordinate functions Ylt ...• y" and with T(W) - U X V c: IRc x lRtl-c. for which there exist (:tie functions III on U x V (I - 1•..•• c; j - I•...•d - c) such that the vector fields (5) span~on (6)

y, - -

"-c + III' ay,



T --


(I - 1•...•


W. Then we define a map b as in (I) by setting b(r.l) - Ui,(r ,I)}.

It turns out that the involutivity of ~ implies that (2) is satisfied. and from the map« one can obtain the desired coordinate system 1.60(1). Conversely. one can obtain the classical version from 1.60 in a similar way. We shall give in Chapter 2 yet another version of the Frobenius theorem in terms of differential forms and differential ideals.

Distributions tmd the Frobenius Theorem In 1.32 we factors through 1p.: N - P, and tant case occurs tion on M.


considered the situation in which a C«J map 1p: N - M a submanifold (P,'P) of M so that 1p -= 'P 01pt where we sought sufficient conditions for 1pt to be C", An imporwhen (P,'P) is an integral manifold of an involutive distribu-

1.62 Theorem Suppose that 1p: N - M" is C«J, thllt (pe,'P) is Il1I integrlll manifold of lin involutive distribution !I on M, tl1Id that 1p fllctors through (P,'P), that is, 1p(N) C 'P(P). Let 1p.: N - P be the (unique) mapping such that 'P 0 1p. -= 1p.




is continuous (tmd hence C" by 1.32(a».


Let P belong to an open set U in P, and let n e 1p.-1(P). Using 1.60, we can obtain an open set 0 with P e 0 C U and a cubic coordinate system (V,T) centered at 'P(p) with coordinate functions X1' ••• ' X" such that the slices
(2) x.

-= constant

for all i e {c

+ I, ...

, d}

are the integral manifolds of !I in V, and such that 'P( 0) is the slice (3)

-= . . . -= x" -= o.

",l(V) is open in N. Let W be the component of ,,-l(V) containing n.

W is open. To prove that 1p. is continuous, we need only prove that C 0 C U. By the commutativity of (1) and the injectivity of 'P, it suffices to prove that 1p( W) lies in the slice (3) of V. Now, 1p is continuous and W connected; hence 1p(W) is connected. Moreover, 1p(W) has at least the point 1p(n) in common with the slice (3). So since 1p(W) lies in a component of 'P(P) f"'I V, it is sufficient to prove that components of 'P(P) f"'I V are contained in slices of the form (2). Let C be a component of 'P(P) f"'I V, and let 71": V - IR"-e be defined by


7I"(m)= (xH1(m), ... , x,,(m». -

Then since P is second countable, and since 'P(P) f"'I V is a union of the slices (2) due to the fact that (P,'P) is an integral manifold of !I, it follows that 71"( 'P(P) f"'I V) consists of a countable number of points in IR"-c. Thus 7I"(C)is a connected countable subset of IRcI-e; hence 7I"(C)is a single point, and C lies in a single slice.



1.63 DeflDidon A maximal integral manifold (N.,,) of a distribution tI on a manifold M is a connected integral manifold of!6 whose image in M is not a proper subset of any other connected integral manifold of!6. 'Jl?at is. there does not exist a connected integral ~fold (NlO",) of!6 such that V'(N) is a proper subset of 'heN,).

1.64 Theorem

Let tI H a c-dimensional. involutive. Coo distribution on

i8 contained in the maximllione.


M". Let me M. TMn through m there passe8 a uniqw maximlll connected integral manifold 01!6. and every connected integral manifold oltl through m

Existence Let K be the set of all those points p in M for which there is a piecewise smooth curve joining m to p whose smooth portions

IDtep cune of •

are l-dimensional integral curves of tI. that is. their tangent vectors belong to!6. By 1.60 and the second countability of M. there is a countable covering of M by cubic coordinate systems {(U,. Xl' ••••• xl): i - O. 1.2 •... } such that the integral manifolds of!6 in U, are the slices
X~+I -


for all) e {I, •.•• d - c},


We shall assume that me U•. Now let p e K. Then there exists an index ill such that p e U'•• and there is a slice S'. of U'. of the form (I) containingp. Observe that S'. c K. It follows from 1.60 that the collection of all open subsets of all such S'. as p runs over K forms a basis for a locally Euclidean topolOgy on K. and that we obtain a differentiable structure on K if we take the maximal family of coordinate systems (with respect to 1.4(b» containing the collection (2) {(S'.'

1 S'.' ...•

Xc'· 1 S,):

p e K}.

We claim that K with this topology and differentiable structure is a connected differentiable manifold of dimension c. K is clearly connected since it is pathwise connected by construction. We have only to prove that the topology on K is second countable. For this. fix an i e (0. 1.2 •... ). We need only show that there are at most countably many slices of U, in K. Each point of U, which lies in K is joinable to m by a piecewise smooth curve whose range also lies in K. To each such curve from m to points in U, there corresponds (although not uniquely) afinite sequence (3) U•• U'a' •.•• U'.' U.

( ,



•' .

Distributions tmd the Frobenius Theorem


t r

t. i:

of the coordinate neighborhoods through which the curve passes in order. The curve thus begins in the slice of U. containing m, passes through some slice of U'l' then through some slice of U.. , and so on, until in a finite number of steps it reaches a slice in U,. Since there are at most countably many such sequences (3) from Uo to U" we need only show that for each such sequence there are at most countably many slices of U, reachable in the above manner. For this, we need only observe that for any J, k e (0, I, ... ) a single slice of U1 can intersect at most countably many slices of Ut; for if S is a slice of U1, then S f"'I Ut is an open submanifold of .~ and therefore consists of at most countably many components, each such component being ~ connected integral manifold of [I in Ut, and hence lying in a slice of Ut. This proves the '---+----__. u. second countability of X. (X,i), where i: X - M is the inclusion map, is now a submanifold of M and is a connected integral manifold of [I passing through m. Moreover, (X,i) is a maximal connected integral manifold of [I. For let (N,V') be any connected integral manifold of [I passing through m, and let p e V'(N). There is a piecewise smooth curve c: (0,1] - N joining V'-l(m) to .,-1(p). (Connected manifolds are pathwise connected.) Then V' c is a piecewise smooth l-dimensional integral curve of !!d . connecting m to p. Thus p e X, and so V'(N) c: X, which proves that X is maximal. Thus we have proved the existence of a maximal connected integral manifold (X,i) of [I passing through m, and have proved that every connected integral manifold of [I through m has its image inX. Uniqueness (cJ. 1.33) Let (N,tp) be any other maximal connected integral manifold of [I passing through m. As we have observed above, V'(N) c: X; thus V' factors as follows:

11'11 is Coo by 1.62, and is 1: 1 and non-singular since V' is 1: 1 and nonsingular. 11'0 is onto since the fact that (N,V') is a maximal integral manifold of [I through m implies that V'(N) cannot be a proper subset of X. It follows from Corollary <a>of 1.30 that 11'0 is a diffeomorphism. Thus (N,V') and (X,i) are equivalent, and the maximal connected integral manifold of [I through m is unique .



BXEltCISES 1 Prow that in Example I.S(d) one ~~,i~deed

structure on ~." " obtain a differentiable

2 The usual 4ifferentiable structure on the real line R was obtained by taking , to be the maximal collection containins the identity map. Let'l be the maximal collection (with respect to l.4(b» containins the map I ....II. Prove that , " but that (IR-') and (IR-' J are 4iffeomorph.ic.


3 Let {U.} be an open cover of a manifold M. Prove that there exists a refinement {V.} such that V. c: U. for each IX. 4 Use the fact that manifolds are regular and paracompact to prove that manifolds are normal topological spaces. Prove that if ,,: M _ N is Coo, one-to-one. onto, and everywhere non-singular. then" is a diffeomorphism. (This proposition depends strongly on the IecODd countability of M. Here is an outline for a proof. This map " is a diffeomorphism if and only if dtp is surjective everywhere. If dtp is not surjective at some point, then the dimension of M is less than the dimension of N. Let dim M - P and dim N - d. Assuming that p < d. one arrives at a contradiction u follows. Let (U,'P) be a coordinate system on N such that 'P(U) - IR·. Since" maps M onto N. the.range of 'P is all of IR·. Now we show that this yields a contradiction. One way is to use 1.35 to observe that the .range of 'P is a countable union of nowhere dense seta in IR·. and therefore, by the Baire category theorem, could not possibly be all of IR·. Another way is to use the fact that p < d to prove that the range of 'P 0 " has measure 0 in IR·. which also is a contradiction (where a set in IR· has measure 0 if it can be covered by a sequence of balls. the union of whose volumes is arbitrarily small). That the range of 'P has measure 0 in IR· follows from the second countability of M and the fact that a CS map IRlJ -IR· has image of measure 0, and this, in tum, follows from the fact that a CS map IR· _ IR· takes sets of measure 0 to sets of measure O. To prove the latter, observe that if f: IR· _ IR· is CI, and if A is a compact set in IR·, then/has a Lipschitz constant KonA,

5 Prove 1.25(a), (b), and (c). ,







, I


e A).

so that/magnifies the volume of balls in A by at most X·, and hence takes sets of measure 0 in A into sets of measure 0.) 7 Prove 1.33(a) and (b).



8 Obtain the classical implicit function theorem 1.37 from 1.38. , Let I: IRI - IR be defined by. I(x.y) - xl + xy

+ ,. + 1.

For which points p - (0.0). p ... (l.l). p ... (-1.-1) is f-l(J(p» an imbedded submanifold in IRI?

10 Let M be a compact manifold of dimension n, and let f: M - IR" be CZ'. Prove that f cannot everywhere be non-lingular. 11 Find counterexamples to show that Proposition 1.36 would fail if either of the hypotheses closed or imbedded were deleted. In fact. one can prove more; namely. if (M.V') is a submanifold of N such that whenever g e C"'(M) there is a CZ' function I . on N such that f V' ... g. then V' is an imbedding and V'(M) is closed inN.


12 Supply the details for 1.4O(a) and (b). 13 Prove Proposition 1.45. 14 Is every vector field on the real line complete? 15 Prove that if (U. XIo •••• (iJ/iJx.,iJ/iJx/l ... 0 on U. x.) is a coordinate system on M. then

16 Let N c M be a submanifold. Let v: (lI,b) - M be a C'" curve such that ,,(a,b) c N. Show that it is not necessarily true that r(/) e N,,(,> for each I e (a,b). 17 Prove that any C'" vector field on a compact manifold is complete. 18 Prove that a C'" mapf: IRI _IRI l' cannot be one-to-one. Supply the details of the equivalence of the two versions 1.60 and 1.61 of the Frobenius theorem.

20 Let cp: N - M be C ..., and let X be a C ...vector field on N. Suppose that dcp(X(P» - dcp(X(q» whenever cp(p)... cp(q). Is there a smooth vector field Y on M which is cp-related to X? 21 The torus is the manifold Sl x Sl. Consider Sl as the unit circle in the complex plane. We define a mapping cp: IR - Sl x Sl by setting cp(/) ... (el .. " .e.... where IX is an irrational number. Prove that (lR,cp) is a dense submanifold of Sl x Sl. This submanifold is known as the skew line on Ihe lOTUS.

22 Let ,,(/) be an integral curve of a vector field X on M. Suppose that r(/) ... 0 for some I. Prove that" is a constant map, that is, its range consists of one point.

23 A Riemannian structure on a differentiable manifold M is a smooth choice of a positive definite inner product ( , ). on each tansent space M., lDlooth in the sense that whenever X and Y are CO vector fields on M, then (X, y) i. a ClIO function on M. Prove that there exists a Riemannian structure on every differentiable manifold. You will need to use a partition of unity argument. A RJemaruaitm ma"ifold is a differentiable manifold together with a Riemannian structure. 24 Consider the product manifold M x N with the canonical projections "'I: M x N-M and "'I: M x N-N. (a> Prove that at: "'1 at are CO.

ll- M x N is CO if and only if




(b) Prove that the map V .... (tbr1(V),tbrl(v» is an isomorphism of (M x N)(•. ,., with M. E9N". . (c)

Let X and Y be CO vector fields on M and N respectiveJy. Then, by (b), X and Y canonically determine vector fields Z - (X,O) and , - (0, Y) on M x N. Prove that [Z, Y] - o.

(d) Let ("'0.".) e M x N, and define injections i.. : M - M x N and i.. : N - M x N by setting i.. (m) == (mA), i.. (,,) -= ("'0,,,).

Let ve (M x N)(..... ,. and let VI -= h1(v) e M.. and hl(v) eN ... Letfe CIIO(M x N). Prove that v(f) - v1(jo i,.)



+ VI(jo



JTlEW§(Q)IR§ IIDIIlFlFlE~W1rllAl1 IF(Q)m.M[§


There are a number of vector spaces and algebras naturally associated with the tangent space M",. Suitably smooth assignments of elements of these spaces to the points in M yield tensor fields and differential forms of various types. We shaU first develop some of the pertinent facts from multilinear algebra, and then beginning with 2.14 we shall apply these concepts to manifolds.


Throughout 2.1-2.13, V, W, and U will denote finite dimensional real vector spaces. As usual, V· wiD denote the dual space of V consisting of all real-valued linear functions on V. 2.1 DeftnldoDl Let F(V,W) be the free vector space over IR whose generators are the points of V x W. Thus F( V, W) consists of all finite linear combinations of pairs (v,w) with v E V and WE W. Let R(V,W) be the subspace of F( V, W) generated by the set of all elements of F( V, W) of the foUowing forms:




+ VI' w) + W.) -





(av,w) - a(v,w) (v,aw) - a(v,w)

v, VI'




The quotient space F( V, W)/ R( V, W) is called the tensor product of V and Wand is denoted by V ® W. The coset of V ® W containing the element (v,w) of F(V,W) is denoted by V ® w. It foUows from (1) that we have the foUowing identities in V ® W:

+ VI) ® W (WI + W.) -

® W + VI ® W ® WI + V ® WI a{v ® w) - av ® w - V ® aw.

I ,

Tensor and Exterior A.lgebras


2.2 The following properties of the tensor product are easily established, and are left to the reader as exercises, (a) Universal Mapping Property, Let cp denote the bilinear map (v,w) ..... v ® w of V x W into V ® W. Then whenever U is a vector space and 1: V x W - U is a bilinear map, there exists a unique linear map


i: V ®


U such that the following diagram commutes:

V® W


I i 1






The pair consisting of V ® Wand cp is said to solve the universal mapping problem for bilinear maps with domain V x W, Moreover, V ® Wand cp are unique with this property in the sense that if X is a vector space and qi: V X W - X a bilinear map with the above universal mapping property, then there exists an isomorphism lit: V ® W X such that lit cp .. qi, (b) (c) (d)

e W is canonically isomorphic with W e V, V e (W e U) is canonically isomorphic with (V e W) e u.

V By property (a), the bilinear map of V* x W into the vector space Hom (V,W) of linear transformations from V to W defined by (j,w)(v) == f(v)' w for fE V*, V E V, and wE W determines uniquely a linear map lit: V* ® W - Hom (V, W). Cl is an isomorphism. As a consequence,

dim V ® W == (dim V)(dim W).


Let {eo: i == 1,. , . ,c} and U:i == 1" " ,d} be bases for V and W respectively, Then {e. ®J;: i == 1" .' ,c and i == 1,., , ,d} is a basis of V e w. The tensor space Vr •• of type (r,s) associated with V V 0 .. , ® V ® V* ® . , , ® V*, '"-~
r copies , cciPlea

2.3 Definitions is the vector space


The direct sum


T(V) ==

I Vr,.

(r, s ~ 0),

where Vo.o == lR, is called the tensor algebra of V. Elements of T(V) are finite linear combinations over lR of elements of the various Vr •• and are called tensors. T(V) is a non-commutative, associative, graded algebra under ® multiplication, where if u == ® ' , , ® ® "~ ® ' , . ® belongs to Vr1"1 and v == VI ® ' , , ® t'r. ® v: ® ' .. ® v~ belongs to Vr •••• ,





TelllOrsand Dffferentia/ Fann.r


then their product"


v is deJined by



® ••• ®


® • •• ®



® ••• ®



® ••• ®


tf~i;;;';;'L' ;\

and belongs to Vri+r•••i+•• , Tensors in a particular tensor space Vr•• are called homogeneous of degree (r,s). A homogeneous tensor (of degree (r,8), say) is called decomposable if it can be written in the form

e ... ® e, ® vi e ... e V:




(j-I, ...,s).

2.4 DefinidoDl

We let C(V) denote the subalgebra

Let J( V) be the two-sided ideal in C( V) generated by the set of elements of the form V ® V for v e Y, and set .
(1) 1~(V) - leV) f"'I Y~ ••.


I V~.oof


T(V) .

It follows that (2) and is a graded ideal in C(V). The exterior algebra A(V) of V is the graded algebra C(V)/I(V). If we set
(3) A.(Y) Y~.oI1~(V) (k ~ 2), Ao(V) IR, AI(V) Y,

then (4) We shall denote multiplication in the algebra A(V) by A. This is called the wedge or exterior product. In particular, the residue class containing VI ® ••• ® v~ is VI A ••• A v~. 2.5


A multilinear map
h: Y

x ... x


is called alternating if (2)'

h(v,,(u, ••• , v,,(r) -

(sgn 1r)h(VI' ••• , vr)

(VI' •.. , Vr

e V),

for all permutations 1r in the permutation group Sr on r letters. Sgn 1r is the sign of the permutation 1r (+ 1 if 1r is even, -I if 1r is odd). The vector space of all alternating multilinear functions

will be denoted by Ar(V), and for convenience we set A.(V) - IR.

Tensor and Exterior Algebras


2.6 The following properties of the exterior algebra are left to the reader as exerCises: ;' (a) Ifu e A.,(V) and v e A,(V). thenu A v eAI:+I(V)andu A v - (-I)"lv A u. (b) If e1, ••• , ed is a basis of V, then

is a basis of A(V), where CI> runs over all subsets of {I•.••• d}. including the empty set; where e. == e~A ••• A e'r with il < ... < ir when CI> is the subset {i1 ••••• ir} of {I•..•• d}; and where e. - 1 when CI> == 0. In particular. ~(V)=: fR. (2) ~/(V) == to} (j> 0). Moreover. it follows that dim A(V) _ 2d. (3) dim A,,(V) _

(d\k _

d! k!(d - k)!

(0 ~ k ~ d).

(Hint: Observe that the elements {e.} span A(V). To prove that they also are linearly independent. fint prove that el A ••• A ed is not zero in ~(V). For this. one must show that el ® ••• ® ed does not belong to I(V). Express an arbitrary element of I(V) in terms of the basis vectors eh ...• ed' and show that it could not equal el ® ••• ® ed. Then for linear independence of the entire set {e.}, multiply the equation I a.e. - 0 by suitable products of the e, to land in ~(V), and conclude that the various a. are aU zero.) (c) Universal Mapping Property. Let cp denote the mapping (Vt ••••• VI") ..... A v., of V x ... x V (k copies) into A.,(V). Then cp is an alternating multilinear map. Now to each alternating multilinear map h of V x ... x V (k copies) into a vector space W, there corresponds uniquely a linear map h: A,,(V) - W such that h cp == h.
VI A •••

A.,(V) (4)


- ...-- ....... -

....... h

-_ -_.,

t copiea

The pair consisting of A,,( V) and cp is said to solve the universal wrpping problem for alternating multilinear maps with domain V x . .• x V; and this is the unique solution in the sense that if X is a vector space and ~: V x ... x V - X an alternating multilinear map also possessing the universal mapping property for alternating multilinear maps


TtlU011l111d D!fftrt"tlDl Fo"",

with domain V x ••• x V, then there is an isomorphism at: A.,( V) - X such that at In the special case in which W _ IR, the diagram (4) establishes a natural isomorphism
0, _ ,.


of A(VJ· with the vector space ..4.,(V) of all alternating multilinear functions on V x ... x V (k copies). It follows from property (b) that ..4.,(V)_ to} for k > dim V. We shall now consider various dualities between the spaces Vr•• , A.,(V), A(V) and the corresponding spaces (V·)r .•, A.,(V·), A(V·) built on the dual space v· of V.

2.7 De8nldOD Let V and W be real finite dimensional vector spaces. A pairing of V and W is a bilinear map ( , ): V X W - IR. A pairing is called non-singular if whenever w " 0 in W, there exists an element v e V
such that (v,w) "0, and whenever v" we W such that (v,w) " O.
(1) ,: V-

0 in V, there exists an element

Let V and W be non-singularly paired by ( , ), and define W* by

tp(v)(w) _ (v,w) (ve V; we W).

It follows that , is 1: I. Similarly, there is a 1: 1 map w-- V·. Therefore V and W have the same dimension, and hence , is an isomorphism of V with W·. Thus a non-singular pairing of V and W in a canonical way yields an isomorphism ,: V - w· and similarly an isomorphism W - V·. 2.8 De8nldon ..4 non-singular pairing of (V·)r .• with Vr.•• This pairing is to be the bilinear map (V·)r .• x Vr.• - IR which on decomposable clements . and 17._".®···®".®" "r "1 .. _ "1 ® ... ® yields.the following:

® ... ® ..r+. e(ylh Jr •• ® V~. e Yr.•

"r ® V:+l ® ...


(v·...) _ vt(..J


It is easily checked that there is a unique such bilinear map and that it is a non-singular pairing. This pairing establishes an isomorphism (2) (V·)r .•~ (Vr•• )·. On the other hand, the obvious extension of the univcrsal mapping property 2.2(a) shows that there is a natural isomorphism
(3) (Vr ••)·

Mr ••(V)

• I

Tensor tmd Exterior Algebras where Mr •• (Y) is the vector space of all multilinear functions V


x ... x V x V* x ... x Y· - IR• ....


Under the isomorphism (3). if (V,..•)·. then the corresponding multilinear function h in M,.••(V) aatisfies


• eoP'lea



v,.. v~ •••••

V:) -

,(&\ ® ••• ® v,. ® ~ ® ••• ®


Finally. from (2) and (3) we obtain an isomorphism


(V·)r ••~ M,.••(V).

A raon-singular pairing 0/ A.. V·) with A..(V). This ( pairing is to be the bilinear map of A.. V·) x A.,(v) - IR which on decom( posable elements A ••• A in A..CV·) and .. - ~ A ••• A.... in A..( V) yields

2.9 Defln1doD

v· - vt



(v·...) - det(v~("i»' Again it is easily checked that there is a unique such bilinear map and that it is a non-singular pairing. For k - O. the pairing is simply multiplication of real numbers. This pairing establishes an isomorphism
(2) A..(V·) ~ A..(V)·.

Composing (2) with the natural isomorphism


A.,( V).

S!: A ..( V)

of 2.6(5) yields an isomorphism


A.,(V·) ~A ..(V).

Using the isomorphism (2). and observing that the dual space of a finite direct sum is canonically isomorphic to the direct sum of the dual spaces, we obtain isomorphisms (5) A(V*) - IA .. Y*) S!: IA ..(Y)· ~ A(Y)·, (



and from (3) we obtain an isomorphism (6) A(Y)· ~A(Y)

== IAt(Y).

... "_0

Henceforth we shall make use of the identification (5) of A(V·) with A(V)· via the pairing (1) without further comment.

2.10 Remarks OD 2.9

(a) If {el, ...• e"l is a basis of V with dual basis {rh ...• r,,} in V·. then the bases {e.} and {r.} defined in 2.6(b) are dual bases of A(V) and A(V·) under the isomorphism 2.9(5).



Te1l.J01's Ild DiffernJllaI Fomu G

(b) In 2.9(5) and (6), the following isomorphisms were established:


A(y)* ~ A(Y).

The second of these is the natural isomorphism arising from the universal mapping property 2.6(c). The first, call it at for the moment, arose from our choice of the pairing 2.9(1). There is another pairing in common use, this one obtained by replacing 2.9(1) by (1) This pairing gives a different isomorphism of A( V*) with A( V)* , call it fJ. Now A(V*) is an algebra under wedge multiplication. Via the above isomorphisms we obtain two algebra structures fl.. and fl., on A(V). It is not difficult to see that if IE A.(V) and g E Af(V). then these induced algebra structures on A(V) take the form

I fl.. g(Vl



-I ".f
and (3)

(s,n ")/(v.I1), ... , 11-«,,)g(V.I

.... 1I ••••

, V.Ilt+e)


I fl., g(Vl'

•••• 1


I (s,n + q)! •• 8" ...

")/(···. V.I ..)g(V.I



Here a permutation "E S..... is called a "p, IJ shu1lle" if ,,(I) < ,,(2) < ... < "ep) and ,,(p + I) < ... < ,,(p + q). It follows from (2) and (3) that (4)
IfI..,_(p+q)!/fI. p!q!


For example. consider the case in whichp == q == 1. Let rand /) belong to V* == A1( V). and let e, WE V. Then r fl../) and r fl., /) E AI(V) and


r fl../) (v.w) - rev) 6(w) - y(w) 6(v).

whereas (6)
r fl., /) (v,w) == HY(v) 6(w) - r(w) 6(v».

We shall use only the pairing 2.9(1). It has the advantage of avoiding factors such as the I in (6).

• 1
:'f '.1'.',~

Tensor and Exterior Algebras


2.11 Unear Transformations or A(Y) End(A(V» will denote the vector space of all endomorphisms of A(V) (i.e. linear transformations from A(V) into A(V». Let U E A(V). Left multiplication by u is the endomorphism e(u) of A(V) defined by

e(u)v - u A v



The transpose ;(u) of e(u) is an endomorphism of A(V)*. Under our identification of A(V)* with A(V*), the transpose ;(u) can be considered as an endomorphism of A(V*), and this endomorphism is called interior multiplication by u. In terms of our pairing of A(V) with A(V*), i(u) is defined by

(/(u)v*,w) == (V*.B(U)W) == (v*.







If u E V. then ;(u) maps At(V*) v* E V*. then i(u)v* E IR and

into At-I(V*) for each k. In particular. if

;(u)v* == (/(u)v*.I) == (v*. B(U). 1) - (v* .u) == v*(u). An endomorphism I of A(V) (or in general of any graded algebra) is (a) a derivation if I(u

(b) an anti-derivation

+ U A I(v) (u, v E A(V», if I(u A v) == I(u) A v + (-1)· U A I(v)

v) == I(u)


E Ap(V-);

E A(V»,


of degree k if I: Aj(V) - AjH(V) for ail}. (We assume that ~(V) == {O} if 1< 0.)

It is easy to see that I E End A( V) is an anti-derivation if and only if on decomposable elements we have
(3) l(vl A'"

Vi) ==


I (_l)'+lVI A'"






1/ u E V,

then i(u) is an anti-derivation of degree -1.

That the degree of i(u) is -I if u E V is clear from the definition 2.11 (2). To prove that i(u) is an anti-derivation, we check that 2.11 (3) holds. For this, it is sufficient to see that both sides of 2.11(3) give the same result when paired with an element of A( V) of the form A ••• A wi' That is, we must show that




A' •• A


WI A •••


== (t(-1)4+lv~










tmd Differential


Now. the left-hand side of (1) equals

(ot A ••• A 01. II A
WI A •••

A WI) -


where we have put

WI -


The right~~acn4.~~ of (1) is 0:



I (_1)C+I0:,(uxor


01. WI A'"


k-I •...• (I - 2••..• - det«(w,»

T, .... J)

'.i.l •...• J) ( 1- 1 •...• J

(Here the circumflex over a term means that it is to be omitted.)

2.13 Effect or Linear TraDSformadoDS

(1) I: A(V)-A(W),

Let I: V.- W be a, linear Then I extends to an algebra homomorphism




A •••

A Ot) -

1(t1J) A




1(1) - 1.

The transpose 6/: W· - V· also extends to an algebra homomorphism

6/: A(W*) - A(V·).

Our isomorphisms 2.9(5) of A(V·) with A(V)· and of A(W·) with A(W)· are natural in the sense that (2) is the transpose of (1); that is.

(6/(w·),0) - (w·./(v»

(w· e A(W*);




2.14 DeflDldoDS

Let M be a differentiable manifold. We define

(I) Tr.~(M)(2)
(3) A:(M) A·(M) -

U (M .Jr.• U At(M:'} U A(M:')

tensor bundle of type (r,s) over M exterior k bundle over M exterior algebra bundle over M .


... JI

In the cases in which k .. 0 and (r.s) - (0,0), the unions in (I) and (2) are disjoint unions of copies of the real line-one copy for each point in M. Tr••(M), A:(M), and A *(M) have natural manifold structures such that the canonical projection maps to !rI are Coo. If (U,tp) is a coordinate system on M with coordinate functions Yl' ... ,Y4' then the bases {alay,} of M .. and {dy,} of M!, for me U. yield bases of (M ..)r •• ' At(M!), and A(M!). For. example, the basis of At(M!) is {dY'l A ••• A dy,,,: II < ... < It}·

Tensor Fields and Differential Forms


Using these bases, one can define maps of the inverse images of U in Tr.,(M), A: (M), and A*(M) under the respective projection maps to cp(U) x Euclidean spaces of the proper dimensions. By requiring these maps to be coordinate systems, one obtains the natural manifold structures on Tr.,(M), A:(M), and A *(M) just as we did previously in 1.25 for T(M) and T*(M) (which incidentally is simply At(M).

2.15 Definfdon A C<IOmapping of Minto Tr.,(M), A:(M), or A *(M) whose composition with the canonical projection is the identity map is called a (smooth) tensorfield 01 type (r,s) on M, a (differential) k-form on M, or a (differential) form on M respectively. Since our tensor fields and differential forms will always be smooth in this sense, we shall drop the adjectives "smooth" and "differential" unless needed for emphasis. 2.16 Remarks A lifting at: M - Tr.,(M) is a smooth tensor field of type (r ,s) if and only if for each coordinate system (U, Y1>... ,Y.) on M, Y4. where the a,I ..... .:II ....,1. e C<IO(U). (In the classical tensor notation one 4 uses lower indices on tangent vectors, upper indices on functions and differentials, and the reverse on coefficients; thus the individual terms of (1) would appear as



I U - Iah ..... ..... I. -aa e ... e -aa e dYI e ... e dYI., ~II



'. 1.

all ®



al' ® dyll

® ...

® dyl «, )

A lifting p: M - A:(M) is a differential k-form if and only if for each coordinate system (U,Ylt ... ,Y..) on M, (2) where the b,I ..... are C<IO functions on U. 2.17 Deflnidons -E1r(M) shall denote the set of all smooth k-forms on M, and E*(M) the set of all differential forms. £G(M) can be identified with C<IO(M); indeed, the differentiable manifold A:(M) is simply M x IR, and smooth liftings of Minto M x IR are simply graphs of C<IO functions on M. Forms can be added, multiplied by scalars, and given a product (A). If co, cp e E*(M) and c e IR, then co + cp, cco, and co A cp are the forms which at m have the values co", + cp"" cco"" and COm A CPm respectively. In the case in which I is a O-form and co E E*(M), we write I A co simply as fi». E*(M) thus has the structure both of a module over the ring C<IO(M) and of a graded algebra over IR with wedge multiplication.



Tensors and Differential FDmfJ

l.18 Let co e I!'(M). Then co. e ~(M!>, and can be considered (via the duality 2.9(4» as an alternating multilinear function on M.. So if Xl> ••• ,Xa are vector fields on M, co(XI, ••. ,Xt) makes sense-it is the function whose value at m is . ., (I) Thus, if we let X(M) on M, then (2) denote the Coo(M) module of smooth vector fields x ••• x X(M) _ Coo(M)


co: X(M)

and is an alternating multilinear map of the module X(M) We stress that co is multilinear over the Coo(M) module X(M);
(3) co(XI, ... , X'_l

into COO (M). that is


,IX + gY, Xf+l,


.•. , Xt)

- fco(X1,·

,X,_to X~ Xf+l' .•• ,Xt)

+ gco(X

1, •••

,X'-b Y, Xf+l' .•• , Xt)

whenever f, g e Coo(M) and Xl' •.• , XI-I, X, Y, Xf+l' ..• , Xt e X(M). Conversely, it is useful to observe that any alternating Coo(M) multilinear map (2) of the module X(M) into Coo(M) defines a form; for we claim that if co is such a map, then co(X1, ••• , Xt)(m) depends only on the values of the vector fields X, at m. Assuming this for the moment, it follows that co defines an alternating multilinear function co.. on M .. , and hence defines an element of At(M!); namely, given (Vi' .•. ,Vt) e M .. x x M .. , choose Vl, .•. , Vt e X(M) such that V,(m) - V, (i - I, ,k), and define (4) By our claim, CO... (t'l, ••• ,val is well-defined, independent of the choice of the extensions V,. Thus co gives a lifting m H co", of M into A ·(M), and this is easily seen to be smooth; hence co is a form. For simplicity of notation, we illustrate the claim with the case in which co is a linear map of the module X(M) into C<I:(M). Let Xe X(M). We wish to show that co(X)(m) depends only on X(m). It suffices to show that co(X)(m) ... 0 if X(m) - o. Let (U, Xl' ••• ,Xci) be a coordinate system about m. Then on U, we have X - I a,(a,ax,) where the a,(m) - o. Now, let cp be a Coo function which takes the value 1 on a' neighborhood V c:: U of m and is zero on a neighborhood of M - U (see 1.10). Then the vector field X, which is cp(a,ax,) on U and 0 elsewhere is a COO vector field on M; the function 4, which is cpa, on U and 0 elsewhere belongs to coo.(M); and




I '.x,+ (1 -




Tensor Fields ond Differential Forms



cu(X)(m) ...

I I,(m)cu(X,)(m) + «(1 -


... O.


So cu(X)(m) ... 0 if X(m) ... 0, as was to be shown. Finally, we remark that via the duality 2.8(S), tensor fields can be given a similar interpretation. If Tis a tensor field of type (r,s), then we can consider Tas a map


T: £l(M) x •..
r copies

x EI(M) x X(M) x .•.

x X(M) -


• coplfll

which is C;c(M) multilinear with respect to the C«'(M) modules £l(M) and X(M). Observe the simple form which formula (2) of 2.10(b) takes when cu, cpe £l(M) and X, Ye X(M). In this case, we have



cu /I. cp(X, Y) - cu(X)cp( Y) - cu( Y)cp(X).

Deflnidon If f e C«' (M), the differential of T(M) into IR which is linear on each tangent considered as a l-form, df: M - At(M). The exterior derivative of the O-formf, and this exterior has an important extension to E*(M) given by the


df is a smooth mapping space. Thus df can be l-form df is called the differentiation operator d

2.20 Theorem (Exterior Dift'erentiadon) There exists a unique antiderivation d: E*(M) - E*(M) of degree + I such that
(1) (2)

dI ... O. Whenever f e C«'(M) - £G(M), df is the differential off.


Existence. Let p e M. Let E*(p) be the set of all smooth forms defined on open subsets of M containingp, with E"(P) the corresponding set of k-forms. We fix a coordinate system (U, Xl>' •• ,x,,) about p. If cu e E*(p), then (3)

ao)"t· -

I a. h.

where the a. e C«'«domain cu) n U), where tJ) runs over all subsets of {I, , d}, and where the dx. are either h'l /I. ••• /I. dx, when tJ) ... {il < < ir} or the constant function I when tJ) - 0. We define dcu at p by setting


We will have to show that the definition of dO). is independent of the choice of coordinates. But fint, we give the following properties: (a) (c)

e £rf.p) => dO).e A,...1(M;).

(b) dO).depends only on the genii


where the domain of al~>I. -

til' at p.
0) 0),

al(dcoJl. + a.(druJl., (a, e IR;

(d) d(~"

ruJl. -


+ asru. is (domain ~ " "COal. + (-IYeua!. "deual.




• ;





In view of properties (b) and (c), it suffices to check (d) for ~-Idx'l " . . . "dx'r and eua- g dxll " • . • "dxl, on some neighborhood of p. For the case r - s - 0, property (d) is simply d(j' g)l. - dl.· g(p) + 1(P)' dg.; and the case in which only one of r or s is 0 is similar. Now suppose that r > 0 and s > O. If {/1, ••• ,Ir} "Ul'" . ,j,} " e, both sides are O. So assume that this intersection is empty. Then


dx'l " ... " dx,)

" (g dxll " ••. " dxl)

- a/' g dx"

" ••• " dx .....

where 11 < ... < Ir+. and a is the sign of the permutation that has been carried out. So we have that

d(O)l" ruJl. - deaf' g dx'l " ... " dx'r+)I. - a(dl•• g(p) + f(P) dg.) " dx,ll. " ... "dxIr+J. - (dl. " dx'll. " ... " dx,r1.)" (g(P) dxlll.. " .•• " dxIJ.) + (- W(j(p) dx'il. " ... " dx'rl.) " (dg. " dxlil. " ... " dxl.I.) - dWll." COal. + (-IYwll." dCOal.·
(e) If lis a C<IO function on a neighborhood of p, then d(dJ)l. - O. For on (domain/) " U, dl- I (aI/ax,) dx, so that

d(df)l. -

I d(!l.iJa x,


dx,l. -


I iJxiJ~ ldXII." I ox,


But (iJlj/iJxl Ox,)(P) - (iJlj/iJx, axl)(p), whereas dXII. " dx4. hll.. So d(dJ)j. - O.

-h,l. "

Now we claim that the definition of d at p is independent of the coordinates chosen. For let d' be defined on E*(P) relative to another coordinate system, and let 0) e E·(p). Then 0) on (domain 0) " U is given by (3). and since d' must also satisfy properties (a)-(e), it follows that


Tensor ~Ids ond Dlfferentiol




a :.,:.



d'(cu)l~ - d'<I a. dx'l A ••• A dxJI~ -I d' (ae dx'l A ••• A dxJI~ -I d'(ae)l. A dx .. l~ A ••• A dx'rl~ + I (-It-Ia.I~ dx'll~ A ••• A d'(dx't)l~

(by (b» (by (e)

A ••• A


(by (d» (by (e)

- dcu~. Now. if cu e £.(M). we define dcu to be the form which as a lifting of M into A .(M) sends P to dcu~. It follows that tIl - O. since if cu e £·(M) and p e M, dcu has the form I da. A on a coordinate neighborhood of p. and thus


d(dcu)l~ J

I d(da.

dx.)I~ - 0


by properties (e) and (d). It follows that d is an anti-derivation of £.(M) of degree +1 satisfying (1) and (2).

vanishes on a neighborhood W of p, then d' cui. - O. Choose a C"" function cpwhich is 0 on a neighborhood of p and 1 on a neighborhood of M - W. Then cpcu- cu and d'(cu)l~ - d'(cpcu)l. - d'(cp)l.

+ 1 satisfying

Uniqueness Let d' also be an anti-derivation of £·(M) of degree (1) and (2). We first show that if cu e £*(M) and cu

cu~ + cp(p) d'cul~ -


Now d' is defined only on elements of £.(M), that is, on globally defined forms on M. We wish to define d' on £.(p) for each p e M. If cu e £.(p), we can extend cu to a form on M having the same germ at p as does cu. Simply let cpbe a C"" function which is 1 on a neighborhood of p and has support in the domain of cu. Then cpcue £*(M) (cpcuis defined to be 0 outside of the domain of cu), and cpcuagrees with cu on a neighborhood of p. Thus we can define d'(cu)l~ - d'(cpcu)I~. and by the above remarks this definition is independent of the extension chosen. Thus d'(cu)l~ is defined for all cu in £*(p) and clearly satisfies properties (a)-(e). (In (d), extend CUI A cu. to a form on M by using CPWI A cpcu. for a suitable cp; and in (e). observe that d'(d!)I.d'(cp dj)l •... d'(d(cpj»I .... 0, since dcp(P) ... 0 and since d' satisfies (1) and (2).) The equations (5) now imply that whenever cu e £*(p). in particular whenever cu e £*(M), d'(cu)l~ ... d(cu)I.. This proves uniqueness. Observe that it is clear from the above proof that dcu U ... d(cu U) whenever U is an open set in M.

--------68 T..,., _ Di.ffnatlill Fomu

2.21 Interior MuldpllcatloD by vector Fielda Let X be a smooth vector field on M, and let cu e E·(M). Interior multiplication of cu by X is the form 1(X)cu whose value at m is the interior multiple of cu. by X. (see 2.11(2»: ""_ .



'!bat 1(X)cu is smooth follows easily from 2.16(2); and from 2.12 it follows that I(X): E·(M) - E·(M) is an anti-derivation of depee -1.

2.22 Meet of MappfDp Let ,,: M - N be a smooth map, and let me M. Then we have the difl'erential dtp: M. - N",." its transpose 6y1: N;c.) - M!, and the induced algebra homomorphism 6y1: A(~c.».A(M!>. If cu is a form on N, then we can pull cu back to a form on M by setting

6y1(cu)l. - eJ"(cul,,c.,).

This is one of the particularly nice features of differential forms. Under a smooth mapping, they can be pulled back from the range to the domain of the map. Vector fields, on the other hand, do not display such pleasing behavior under mappings. 2023 Proposition (a) (c) Let ,,: M _ N'" a smooth map. TMn

eJ,,: E·(N) - E·(M) and iJ an algebra homomorphism. (b) 6y1 commutes witA d; tlurJ is, d(6y1(w» - 6y1(dcu) (cu e E·(N».
6,,(w)(Xl, ... , XJ(m) - w"c.,(d,,(Xl(m», ... , dtp(X~(m») for cu e E~(N) and for vector fields Xl' ... , X~ on M.

PROOF Result (c) is clear from 2.13(3); and 2.13(2) and the definition 2.22(1) imply that eJ" is an algebra homomorphism. Before we check that 6y1(w) is ac:tually a smooth form for cu e E·(N), observe the following special case of (b): If fe C""(N), then eJ,,(f) - f e C"'(M), and l.23(d) implies that



6y1(df) - d(fo ,,) ... d(6tp(f).

Now let cu e E·(N), and let m e M. Choose a coordinate system (U, Xl' •••• x.) about ,,(m) and a neighborhood V of m such that .,,( V) c: U. Then there are C'" functions a. on U such that

cu I U 6y1(w)

I a. dx'l

A ••

A dxlyo

It follows that

I V - I a.


0 ,,)


0 ,,),

which is a smooth form on V. Hence eJ,,(w) eE·(M), and thus (a) is proved. To complete (b), we use (3):


(4) .


d(~(Q))I. - d(I a. 0 "d(X'l

0 ,,)

-I (d(a.

/I. •••
0 ,,)

0 ,,)





doe /I. dx'

/I. •••


dx .. )I.

d(x •• 0 ••• /I. d(x ..


,,»1. ,,»1.

- ~(dQ)I-.
Thus (b) is proved, and the proof is complete.


2.24 Deflnidon Tensor fields and differential forms can be differentiated with respect to a vector field. The resulting derivative is known as the Lie derivative and is defined as follows. Fix a smooth vector field X on a manifold M. Recall (see 1.48) that we use X, to denote the local I-parameter group of transformations associated with X. Let Y be another smooth vector field on M. We shall define the derivative of Y with respect to X at the point m E M. First we follow the integral curve of X through m out to the point X,(m) and evaluate Y there. Then we transfer YX,( .. , back to M .. via the differential dX_, of the diffeomorphism X_,. In M .. we take the difference of the vectors dX_,(Yx,( ..,) and Y.. , divide the difference by I, and then take the limit as 1- O. In other words, we consider the smooth M..-valued function t .... dX_,( YX,( .. ,), and we take its derivative at 1 - o. The result is a vector In M .. which is called the Lie derivative of Y with respect 10 X at m and which is denoted by (LxY) ... Thus we define
(1) (LxY) .. -lim



dX_,(Yx,( ..,) - Y.. d -t dt


( ) _, Yx,(.., .



Teuorl _



In a similar way we define the Lie derivallve of a diffcrelltilll form cu with respecI 101M DeClOrield X, except that in this case we evaluate cu at X,(m) f and then pull back via 6X, to A(M!) where we take the difference with cu(m), divide by t, and take the limit as t - 0... Thus we define (2)
(Lxcu) ..


6X,(cux,c .. ,) - cu..


- -d

dt ,..

(.aX (cu.z.c ,» . u, ..

Smoothness of L.zcu and of LxY is asserted in Proposition 2.25. The Lie derivative Lx can be extended to arbitrary tensor fields in the obvious way. If T is a tensor field of type (r,s). then (LxT) .. is the derivative at t - 0 of the (M..)(r .•,-valued function whose value at I is dX_'('" ® ••• ® "r) ® 6X,(~ if

® ••• ® ":)

(a) (e)

Let X be a CO Hetor field Oil M.


wlantnB fe


(b) LxY


each CO DeCIorfield Yon M.


Lx: E·(M)-E·(M), and Is a deriViltion which commJIIes with d. d

(d) Oil £.(M), Lx - I(X)

Theil Ly,(cu(Y" •.••

+ do i(X).
, Y~ be CO DeCIOrfields Y~) Y(-1 • Ly.Y"

(e) . Let cu e ~(M), and let Yo ••••

Oil M.

Y~» - (Ly,cu)(Y1 •...•

+ I cu(Y1 •••••

Y.+1 ,

••• ,



Assumption as in (e). Theil

dcu(Yo,···. Y~) -


~ I (-l)(Y,c.u(Yo •••••


Y" ••••


~ .•,[v +.( - l)(+I,"".I(.YI•] y.0.· .. ' -Y, ..... «I


-YI.· .. ,YV• ,\

We leave result (a) as an exercise. For (b), we need only show that LxY(j) - [X,Y]ffor eachfe C""(M). Let m e M. Then
(LxY)(f) -


(lim dX ,Yx,c"" - Y"')u)


- -dd [(dX_,(Yx,c t t-o


The Lie DeriViltive


Define a real-valued function H on a neighborhood of (0,0) in IRa by setting (2) Then by (a), H(t,u) - /(X_,(Y.(X,(m»».

.. ..

(3) and (1) becomes (4)

YXc(.,Uo X_,) -

aa r.



To evaluate this derivative, we set (5) K(t,u,,) - j(X,(Y.(X,(m)))).

Then H(t,u) - K(t,u,-t). (6)

art ara


It follows from the chain rule that



aaK I art


a·K art art



Now, K(t,u,O) - /(Y.(X,(m»).



Also K(O,u,s) - /(X,(Y.(m»). Hence


I art art

aKI a r,


- Xf(Y ..(m».
- Y",(Xf).


Part (b) now follows from (4), (6), (7), and (8). One immediate consequence of (b) is that Lx Y is a smooth vector field. For result (c), first consider Lx as a mapping of £·(M) into forms which are not apriori smooth, and observe that the derivation property follows immediately from adding and subtracting suitable terms before taking the limit in the definition 2.24(1). Next we check that Lx commutes with d when applied to functions; that is,





ti1Ul Diffemttllll Forms



(I e CtO(M); m EM). (Lg{d/» .. - d(Lzf) .. Since both sides of (9) are elements of M:'. we need only prove that they have the same effect when applied to .~ arbitrary vector Y... in M... The right-hand side gives' .. .
d(Lxf) ..(Y..) - Y..(Lxf) - y..


(.!..f....(fo x.»). dt

where loX. can be considered as a ClIO function on (-e.e) x W. for some e > 0 and some neighborhood W of m in M (1.48(d». The left-hand s(de of (9) gives (11) (Lx(df) ..(Y..) -


dt ....


-.!..I .... (6X.(dlx,c .. ,XY..» dt


dt ,_.

(d/(dX,(Y .. ») -


dt ....

(Y.. (lo


Now let Y be an extension of Y.. to a vector field on W. Then. according to Exercise 24(c) of Chapter 1. dldt and Y have canonical extensions to vector fields on (-e.e) X W where they satisfy [did,. y] • o. This fact together with (10) and (11) implies (9). Finally. to see that the form Lxw is actually smooth and to check that Lx commutes with d on all of E·(M), simply express an arbitrary form w in local coordinates as in 2.16(2), and compute using equation (9). result (a). and the fact that Lx is a derivation. For result (d). observe that both Lx and I(X) 0 d + do i(X) are derivations on £·(M) which commute with d, and both have the same effect on functions. Then (d) follows from a simple computation in local coordinates. We shall leave (e) as an exercise. The proof is not difficult-one simply has to be persistent enough in unwinding the definitions. Again. one checks the identity by restricting to a local coordinate system. Try it first for the case w - f dx1 A. dx. in a coordinate neighborhood. Finally. result (f) follows from (e) by using (d) and induction on p. For the case p - 1. we have from (e) that (12) Ll",(w(YJ) - (Ll",w)(YJ

+ w[Yo.YJ.

Applying (a) and (d) to (12). one obtains Yow(YJ - «i(Yo) d - dW(Yo.YJ

+ do I(Yo»wKYJ + w[Yo. YJ + Y1(w(Y + w[Yo,YJ.


p - 1. Then (f) is obtained for p by again starting with (e) and applying
(d) and the induction hypothesis.

which is result (f) in the case p - 1. Now assume that (f) holds for

DljferentillJ Itkals




Our objective here is to give a version of the Frobenius theorem in terms of differential forms, and then to describe E. Cartan's useful method of obtaining maps by looking for their graphs. 2.26 DeflnidODS Let 9 be a p-dimensional ce distribution on M. A q-form 0) is said to annihilate !j if for each m E M

0) 0)


v.) - 0


VI' •••

, 11. E


A form parts of

e £*(M) is said to annihilate 9 if each of the homogeneous annihilates 9. We let

J(9) - {O)e £*(M):

annihilates 9}.

2.27 DeflnidOD A collection 0)1" •• ,0).. of J-forms on M is called independent if they form an independent set in M! for each m E M.



2.28 ProposidoD
(a) (b)

Let 9 be a smooth p-dimeTlSionaldistribution on M.

J(!!;) is an ideal In £*(M). J(!!;) is locally generated by d - P independent k-forms. (Thai is, to each m E M there corresponds a neighborhood U of m and a set of independent l-forms 0)10' •• ,O),,_p on U such that: (i) If 0) E J(!!;) , then 0) U belongs to the ideal in £*(U) generated by 0)1' ••• , O)... p. . (ii) If 0) E £*(M), and if there is a cover of M by sets U (as above) such that for each U in the cover, 0) U belongs to the ideal generated by 0)1> ••• , O),,_p, then 0) E J(!!;).)


c: £*(M) is an ideal locally generated by d -: p independent l-forms, then there exists a llII/que Coo distribution!!; of dimension p on M for which J - J(!!;).

If J


Part (a) follows from the definition of J(!!;) and the definition of multiplication in £*(M). Let m E M. Since 9 is smooth and p-dimensional, there exist Coo vector fields X....P+I' ••• ,X" defined and spanning 9 at each point of a neighborhood of m. This collection can be completed to a collection XI, ... ,X" of smooth vector fields forming a basis of M.. for each n in a neighborhood U of m. Let 0)1' ••• ,0)" be the dual I-forms; that is, (I(roneckerindex) for each n in U. Then
0)1'" • , O).... P

are the desired J-forms on U.


T",.", tIIId

Diff."",iIIl Form.r

I GeO»'. A •••

They are independent smooth I-forms on U. If 01 e J(!iI). 01 A Q1,P' here 4) runs over nonempty subsets {il' .•.• ip} c: w {I ••.•• d}. and where the ... must be identically zero unless {Is ••••• /p} "{I ••••• d - p} " (6. ,,:,~, ... belongs to the ideal in E*(U) generated b1 011 ••••• 01...... Conversely. if 01 is form such that for each such U in some covering of M. 01 U belongs to the ideal in E*(U) generated by 01...... 01...... then clearly 01 e J(9). This

I u-


For part (c).1etm e M. and let tbeindependent I-forms 0Ia ••••• Q1 ....~ generate J on a neighborhood U of m. Define !iI(m) to be the subspace of M. whose annihilator is the subspace of AI! spanned by the collection (Q1,(m):1 - 1••..• d - pl. It follows that!il is a smoothp-dimensionaJ distribution on M and that J - J(9). Uniqueness of!il follows from the fact that !iI " !ill implies J(9) " J(91). 2.29 De8Dldon An ideal J c: E*(M) is called a diJIerentiai ideal if it is closed under exterior differentiation d; that is. (1) d(J) c: J.

proves (b).

J •,

•• •·

2.30 Proposition

A C" distribution on M is lnvolutiH 1/ and only !iI tM ideal J(9) is a differential ideal.


PllOOF Let 01 be a ,-form in J(9). and let X••..•• X. be smooth vector fields lying in!il. Then 2.2S(f) together with the involutiveness of 9 implies that dco(X •••.•• X.) liE O. Hence dOl e J(9). and J(9) is a differential ideal. Conversely. suppose that J(9) is a differential ideal. Let Y. and YI be vector fields lying in 9. and let m e M. By 2.28(b). there are independent I-forms GIa ••••• 01..... generating J(9) on a neighborhood U of m. Extend these forms to M by multiplying by a C" function which is 1 on a neighborhood of m and has support in U. We shall denote the extended forms similarly by GIa ••••• 01...... By 2.2S(f).


Q1,[Y I] - -dQl,(Y••YI) + Yoeu,(YI) YIQI,(Y,) ••Y -

for 1 - 1, ..•• d - p. The right-hand side of (1) is identically zero on M since J(9) is a differential ideal and since 01, e J(9). Thus Q1,([Y YID(m) - 0 for i- 1, ...• d - p. Now 9(m) is the subspace •• of M. whose annihilator is the subspace of M! spanned by the collection (Q1,(m):1- 1•..•• d - pl. Thus [Y••YI](m)e 9(m). and 9isinvolutive. 2.31 De8Dldon A submanifold (N.",) of M is an integral mtmi/old of till ideal J c: E*(M) if for every 01 e J. c5tp(Q1) O.A connected integral liE manifold of an ideal J is maximal if its image is not a proper subset of the image of any other connected integral manifold of the ideal. L

DiJlerentiid Itkals


It now follows immediately that we have the following version of the Frobenius theorem 1.64 in terms of differential ideals.
Let J c: E*(M) be a differential ideal locally generated by d - P independent l-forms. Let me M. Then there exists a unique maximal, connected, integral manifold 0/ J through m, and this Integral manifold has dimension p.

2.32 Theorem

2.33 We shall now consider a technique which in certain situations enables one to find a map by looking for its graph as an integral manifold of a differential ideal. This will have important applications in the theory of Lie groups in Chapter 3, and also has important applications in such areas as isometric imbeddings in Riemannian geometry. Suppose that f: Ne - Md is ClIO and that {cu,} is some collection of forms on M. Let"l and". denote the natural projections of N x M onto Nand M respectively. For each I, we define a form p, on N x M by setting
(1) p, .. lnrlJf(cu,) - lnr.(cu,).

Let J be the ideal in E*(N x M) generated by the p,. Now the graph of/is the submanifold (N,g) of N x M where
(2) g(n) .. (n,/(n».

, .

We claim that the graph is an integral manifold of the ideal J. For this, it suffices to show that c5g(p,) .. 0 for each i. Now'''1 g .. id, and g .. /, and thus it follows that
0 ". 0

6g(p,) .. 6("10 g)lJ/(cu,) - 6(".0

g)(cui) .. lJ/(CUi) -




Thus starting with a map f: N - M and a collection of forms on M, we have observed that the graph of/is an integral manifold of a certain ideal of forms on N x M. Now, suppose that we start with the manifold Md, and suppose that there exists a basis CUI' .•• , CUdof I-forms on M, that is, {cu,(m)} is a basis of AI! for each me M. (Such a basis does not generally exist, but it does exist in a number of interesting situations for which the following technique is quite useful.) Suppose also that we have a manifold Ne and a collection CX1, ... 'CXd of l-forms on N and that we wish to find a map f: N - M such that

for I .. I, ... ,d. If the map / exists, then, as we have observed, its graph will be an integral manifold of a certain ideal of forms. So we attempt to find the graph from the ideal. We define forms p., on N x M by setting
(4) p, - lnr1(cx,) lnr.(cu,),

and we let J be the ideal in E*(N x M) which they generate. If J happens to be a differential ideal, then we can obtain the desired map / (at least locally) from an integral manifold of J. That is, one gets at/by looking for



TtI'UOrItIIId Dlfflmltilll Fomu

its graph as an .integral manifold of a suitable differential ideal. So suppose that J is a differential ideal. and let ( eN x M. Then since J is locally (in fact globally) generated b)' d independent I-forms. the Frobenius theorem guarantees that there is a maxim~.rfRq~~~! integral manifold I of J of dimension c through (ne,me). Lef IJ I. We claim that dtrl I, is 1: 1. For suppose that vel, and that dtrl(v) .. o. Then since ,..,(v) - 0, it follows from (4) that roi_t:br.(v» .. 0 for 1- I •...• d; and this implies that dtr.(v) .. 0 since the ro, form a basis of I-forms on M. But now if both t:brl(v) - 0 and dtr.(v) .. 0, then v .. O. Hence dtrl I, is I: 1. Thus "'I II: 1- N is locally a diffeomorphism. So there exist neighborhoods Y of (n,.me) in I and U of n, such that "'I Y: Y - U is a diffeomorphism. We define/: U - M by setting

1 I


/ - "'.- ("'I YJI.

Then/(n,) (6)



the graph of/is an open submanifold of I. and moreover, 6/«(J)J- ex, U.


For let v e U. for some n e U. Then by (4),


,..i_d("'1 y)-I(V» - «t(v) - ro,(dtr.o d("'1

, , - ex,(v) -

I y)-l(V»


which proves (6). Thus we have obtained the desired map locally. That is, given n, e N and an arbitrary choice of me eM, then under the assumption that the ideal J .generated by (4) is a differential ideal, we have found an open neighborhood U of n, and a Ct/O map f: U - M such that /(raJ .. me and such that lJf(ro,) - «t U. Moreover. there is a unique such map. More precisely, if me eM, and if U is an)' connected open neighborhood of no in N for which there exists a Ct/O map f: U- M such that /(ne) - m, and such that lJf(ro,) .. ex, U for t .. I, ... , d, then there is a unique such map on U. For let/be any other such map. Let (U,!) and (U,g) be the graphs of/and/over Urespectively. Thus


;(n) - (n,/(n»


g(n) .. (n./(n»

for n e U. Then. according to our remarks near the beginning of this section. not only is (U,g) an integral manifold of J through (n,.me). but so is (U.!). Now. the subset of U on which g and; agree is non-empty since it contains ne. and is closed by continuity. and is moreover open. For let i(n) - g(n). Then it follows from the uniqueness of integral manifolds that there exist sufficientl)' small neighborhoods Wand JJ' of n so that
(8) g(W) -


It follows then from (7) that W (9)

JJ' and

gl W-il W.

DilferentW ldals


Hence, since U is connected, g .. I on U. wbich implies that I .. on U. ! This proves uniqueness. . In suitable situations one can conclude that trIll is a covering of N; and then if N is simply connected, one can conclude that there exists a unique ClIO map I: N - M such that I(no) .. mo and such that (3) holds. In Chapter 3 we shall make several concrete applications of this method for proving existence and uniqueness of certain maps. This technique was originally used by E. Cartan for the problem of finding local isometric imbeddings of Riemannian manifolds in Euclidean space. We summarize the results of this section in the following.

2.34 Theorem Let Ne and Md be differentiable manifolds, and let "'I and "'. be the canonical projections 01 N x M onto N and M respectively. Suppose that there exists a basis {cu,: I .. 1••.•• 4} lor the 1-forms on M.

If I: N - M is

ClIO. then the graph 01I is an integral manifold the Ideal of forms on N X M generated by


(61r161(cu,) - 61r.(cu,): I .. 1••.••




If {ex,: I .. 1 •••••
N (2)

II} are 1-forms on N. and if the idealollorms on II}

M generated by . (61rI(cxJ - 61r.(cu,): i .. 1•••.•

is a differential ideal. then given no e N and mo e M there exists a neighborhood U 01no and a ClIO map I: U - M such that I(no) .. mo and such that (3) (i .. l•...• d). Moreover. if U is any connected open set containing no lor which there exists a ClIO map I: U - M satislying both I(no) .. me and equation (3), then there exists a unique such map on U.

EXERCISES 1 Supply the proofs of 2.2(a}-(e). 2 (a> Show that homogeneous tensors are generally not decomposable. (b) (e) (d) Show that if dim V ~ 3, then every homogeneous element in A(V) is decomposable. Let dim V> 3. Give an example of an indecomposable homogeneous element of A(V). Let ex be a differential form. Is ex " ex



Supply the proofs of 2.6(a)-(c). Derive the formulas (2), (3), and (4) of 2.10. 1(/ whenever

S Prove that Lz/field on M.

Ie, C-O(M)

and X is a



6 Let X and Y be ce vector fields on M with corresponding local I-parameter groups X. and Y.. Let m eM, and let {J(t) Y-V;X-v7Yv7XVi(m) for a sufficiently small •.


for t in (-.,.), Prove that


1({J(t»-I({J(O». ,.... t If {J(t) were a smooth curve at t - 0, then the right-hand side' of (1) would simply be the eft'ect of the tangent vector to {J at t - 0 applied to the function f. However, because of the ..(i. {J is not generally smooth at t - O. Thus the existence of the limit in (I) is part of the problem, and the problem asserts that this curve {J, even though it is not smooth at t - 0, defines a tangent vector in M. in the usual way, and this vector is precisely lX,



I i



7 Prove 2.2S(e). 8 Let M be connected, and let ,..: M x N - N be the natural projection. Prove that a p-form cu on M x N is 67r(ex)or some p-form f exon N if and only if i(X)cu -= 0 and Lxcu - 0 for every vector field X on M x N for which tbr(X(m,n» - 0 at each point (m,n) e MxN. , Prove that the elements VI' • • • independent if and only if VI "
• Vp of ••• " Vp "

the vector space V are linearly O.

10 Prove that linearly independent sets {VI' •••• vp} and {WI ••••• wp} are bases of the same r-ciimensional subspace of a vector space V if and only if VI " ••• " V,. -= CWI " ••• "wp where necessarily C ,,0; and in this case. C - det A where A - (a'l) and v,-

(a) dcu, -


11 Let J be an ideal of forms on M locally generated by r independent J-forms. Say J is generated by CUI' •••• CUpon U. Then the condition that J be a differential ideal is equivalent to each of:

1:cu'l "cuI

for some J-forms

CU'I (for each such for some I-form ex

(U, CUI' ••••


(b) If cu - cut" •.• "cup. then dcu - ex"cu (for each such (U, GIlt ••• , cup».



i •

12 Let V be an n-dimensional vector space, and let I be a linear transformation on V. Since A..( V) is one-dimensional, the linear transformation which I induces on A ..(V) is simply multiplication by a constant. Define the .terminant of I to be this constant. If A is an n X n matrix, let VI' ••• ,v .. be a basis of V, and let 1 be the linear transformation on V whose matrix with respect to this basis is A. Then define the determinant of A to be the determinant of I. Prove that the determinant of A does not depend on the basis VI •••• ,V .. chosen. Using this definition. derive the standard properties of determinants. For example, derive the expansion det A ..


I (sgn w)al.(l)



where A .. (aH), sgn 'fr is the sign of the permutation 'fr, and w runs over all permutations on n letters. Prove also that the determinant of the product of two matrices is the product of their determinants. 13 Let V be an n-dimensional real inner product space. We extend the inner product from V to all of A(V) by setting the inner product of elements which are homogeneous of different degrees equal to zero, and by setting
(WI" ••• " Wp, VI " ••• "

. -.
~ ~




and then extending bilinearly to all of Ap(V). Prove that if el, ••• , e.. is an orthonormal basis of V. then the corresponding basis 2.6(1) of A(V) is an orthonormal basis for A(V). SinceA ..(V) is one-dimensional. A..(V) - {O} has two components. An orientation on V is a choice of a component of A ..(V) - {O}. If V is an oriented inner product space. then there is a linear transformation

called star, which is well-defined by the requirement that for any orthonormal basis el •••• , e.. of V (in particular, for any re-ordering of a given basis),


ep)" if el



A..(V) - {O} determined by the orientation and ,,_to otherwise.

Observe that

where one takes


" •••

s e;

lies in the component of


T61IIDrIand Di6lm1tifll Fomu

Prove that on A.(y). (5) •• -

(-1)-'-"). v. we A,,(J1.
their inner product is

Also prove that for arbitrary given by (6) (v,w) -

.(w A. .v) -

.(v A. .w).

14 Let V be a real inner product

y: A'*I(Y)

by 1 e Y.

space. as in Exercise 13. Let A,,(Y) be the adjoint of left exterior multiplication That is. (y(v),w) - (v. 1 A. w)
Prove that
A. (.v».

for v e Ap+I(Y) and we A,,(Y). y(v) 15

(-I)"" • (I


1e V.

Prove that the composition A,,(Y) _.






of left exterior multiplication by 1 with itself is an exact sequence; that is. the image of the first map is the kernel of the second. 16 Cartan Lemma Let P ~ d. and let 0)1•••• ,0)" be I-forms on M' which are linearly independent pointwise. Let 61••••• 6" be l-forms on M such that




Prove that there exist ClIO functions A.u on M with A'I - All such that (I - 1•...• p).



t ~ I ,

~ -----~ -..-.'".,-.," -- ........ ....


.. ..



Lie groups are without doubt the most important special class of differentiable manifolds. Lie groups are differentiable manifolds which are also groups and in which the group operations are smooth. Well-known examples include the general linear group, the unitary group, the orthogonal group, and the special linear group. In this chapter we shall set the foundations for the study of Lie groups. Of central importance for Lie theory is the relationship between a Lie group and its Lie algebra of left invariant vector fields. We shall study the correspondence between subgroups and subalgebras and between homomorphisms of Lie groups and homomorphisms of their Lie algebras. We shall study the properties of the exponential mapping which is a generalization to arbitrary Lie groups of the exponentiation of matrices and which provides a key link between a Lie group and its Lie algebra. We shall investigate the adjoint representation, shall prove the closed subgroup theorem, and shall consider basic properties and examples of homogeneous manifolds. Along the way we shall derive many properties of the classical linear groups.

~ ~

1 •

1 ;



A Lie group G is a differentiable manifold which is also endowed with a group structure such that the map G x G - G defined by (a,T) ..... aT-I is Coo. Throughout this chapter, G and H will denote Lie groups, and we shall universally use e to denote the identity element of a Lie group.

3.1 DeflnidOD


3.2 Remarks
(a) Let G be a Lie group. Then the map T ..... T-I is C'D since it is the composition T ..... (e,T) .... T-I of C'D maps. Also, the map (a,T) aT of G x G - G is Coo since it is the composition (a,T) .... (a,T-I) aT' of C'D maps.


1k Group' tmd 'l'Mir 1k AlgebrQl


(b) The identity component of a Lie group is itself a Lie group; and the components of a Lie group are mutually diffeomorphic. (c) Since we have included second countability in the definition 1.4 of differentiable manifold, Lie groups for us will alwa.ysbe second countable. In particular, they can have at most countabl,- many components. It should be observed, however, that if second countability is dropped from the definition of Lie group, then connected Lie groups (hence also those with countably man.y components) would still be second countable. We leave the proof of this as an exercise. You will need to use the fact that a connected Lie group is the union of powers of any neighborhood of its identity (see 3.18).


3.3 Examples of Ue Groups

(a) The Euclidean space IR" is a Lie group- under vector addition. (b) The non-zero complex numbers C· form a Lie group under multiplication . (c) (d) The unit circle $1 c:: C· is a Lie group with the multiplication induced from C~ . The product G x H of two Lie groups is itself a Lie group with the product manifold structure and the direct product group structure; that is, (0'1 ,'7'1)(0'1,'7'.) .. (alai ,'7'1'7'.). The n-torus T" (n an integer> 0) is the Lie group which is the product of the Lie group $1 with itself n times. The manifold G/(n,lR) of all n X n non-singular real matrices is a Lie group under matrix multiplication. The set of all super-triangular n x n real matrices (all entries below the diagonal are zero) is a Lie group under matrix multiplication. Let IR· denote the non-zero real numbers, and let K be the product manifold IR· x IR. With a group structure on K defined by (',t)(SI ,t1)


.. ...




(f) (g) (h)

(SSI , st1 + t),

K becomes a Lie group. This Lie group is the group of affine motions of IR, for if we identify the element (s,t) of K with the affine motion x + t, then the multiplication in K is composition of affine motions. (i)

Let K be the product manifold G/(n,lR) x IR". We define a group structure on K by setting (A,V)(AI' VI) "'" (AA1, AV1 + v). With this group structure, K becomes a Lie group. This Lie group is the group of affine motions of IR", for if we identify the element (A,V) of K with the affine motion x ..... AX + v of IR", then the multiplication in K is composition of affine motions.

3.4 DefInition A Lie aI,ebra 9 over IR is a real vector space 9 together with a bilinear operator [ t ]: 9 X 9 - 9 (called the brllClut) such that for all x.y. z e g• . (a)

[x.yl - -(),,xl. ([x.y),z] + [()',z),x] + [(z,x).y] - O.

(llIIIi-comnuItativity) (Jacobi ilkntity)

The importaDce of the concept of Lie algebra is that there is a special finite dimensional Lie algebra intimately associated with each Lie group. and that properties of the. Lie group are reftected in properties of its· Lie algebra. We shall see. for example. that the connected. simply connected Lie groups are' completely determined (up to isomorphism) by their Lie algebras. The stud.y of these Lie groups then reduces in large part to a study of their Lie algebras.

3.5 Exam", of IJe Algebras

(a) (b) (c) The vector space of all smooth vector fields on the manifold M forms a Lie algebra under the Lie bracket operation on vector fields. Any vector space becomes a Lie algebra if aU brackets are set equal to O. Such a Lie algebra is called abelilln. The vector space gl(n.lR) of all n X n real matrices forms a Lie algebra if we set [.4,B) - All - B.4. A 2-climensional vector space with basis x, y becomes a Lie algebra if we set [x,x) - ()'.y) - 0 and [x.y) _ y. :' v:: - - " and extend bilinearly. (e)
IRa with the bilinear operation X X Yof the vector cross product is a Lie algebra.


The proof that the above are Lie algebras is left



the reader as an

3.6 DeftDidcma Let a e G. Left translation by a and rilht tran.rlation by a are respectively the dift'eomorphisms I. and r. of G defined by

1.(.,.) -


r.(.,.) - .,.a

for all.,. e G. If Vis a subset of G. we denote r.(J') and I.(J') by Va and (1V respectively. A vector field X (not assumed apriori to be smooth) on G is called left inVtlri""t if for each a e G. X is I.-related to itself; that is. (2) tlI. 0 X _ XO I••

1M Groups and '£Mir Lie Algebras


The set of all left invariant vector fields on a Lie group G will be denoted by the corresponding lowercase German letter 9. 3.7 Proposldon tector fields. (e) ut G be a Lie group and 9 its let of left Invariant

9 is a real vector space, and the map «: 9 - G. defined by «(X) .. X(e) is an isomorphism of 9 with the tangent space G. to G at the identity. Consequently, dim 9" dim G•.. dim G.



invariant vector fields are ImDDth.


The Lie bracket of two left invariant vector fields is Itself a left invariant vector field.

(d) 9 forms a Lie algebra under the Lie bracket operation fields.


PROOF That 9 is a real vector space and that « is linear is clear. injective, for if «(X) .. «( y), then for each a e G


X(a) .. dl,,(X(e» .. dl,,(Y(e» .. Yea);

hence X"'" Y. Moreover,« is surjective; for if x e G., we let X(a) "'" dl.(x) for each a e G. Then «(X) .. x, and X is left invariant since X('Ta) "'" dl.,,,lx) .. dl, dll1(x)- dl.,(X(a» for all a and 'T in G. This proves part (a). For (b), let X e 9, and let f e C«>(G). We need only show that Xfe C«>(G). Now,

Xf(a) "'" Xl1f .. dll1(X,)f .. X,(f 0/11).

Thus we need to show that a t-+ X,(f 0/11) is a C«>function on G. We do this by exhibiting this function as a suitable composition of C«> maps. Let cp: G x G - G denote group multiplication, cp(a,'T)"'" a'T. And let il and ill be the maps of G - G x G defined by I

il('T) .. ('T,e), il1l(T)- (a,'T). '

0 0

Let Y be any C«>vector field on G such that Y(e) - X(e). Then (0, Y) is a smooth vector field on G x G, and [(0, Y)(f cp)] Il is a C«> function on G. Using part (d) of Exercise 24, Chapter 1, we obtain [(0, y)<f

cp)] il(a) "'" (0, Y)(".•,(f cp) .. OI1(fo cp il) + Y,(fo cp il1l) .. X,(f cp illI) "'" X,(f 111).
0 0 0 0 0 0 0

Thus a t-+ X.(f 0/11) is a smooth function on G, which proves part (b) .



1M GTDIIIM Since. by (b). left invariant vector fields are smooth. their Lie brackets are defined. Now. if X is I.-related to itself. and Y is I.-related to itself. then according to 1.55. [X. Y] is I,,-related to [X. Y). Thus the Lie bracket of two left invariant vector fields is again a left invariant vector field. Part (d) is now an immediate consequence of 1.45.

3.8 Deftnldon We define the Lie algebra of the Lie ".DllP G to be the Lie algebra 9 of left invariant vector fields on G. Alternatively. we could take as the Lie algebra of G the tangent space G. at the identity with Lie algebra structure induced b)' requiring the vector space isomorphism 3.7(a) of 9 with G. to be an isomorphism of Lie algebras. It will be convenient at times to consider the Lie algebra from this alternate point of view. In particular. we shall see (3.10 and 3.37) that the Lie algebras of the classical groups admit particularly nice interpretations in terms of their tangent spaces at the iden~ty.

3.9 Remarks on Vector Spaces a Manifolds In l.S(b) we observe that any finite dimensional real vector space V is. in a natural way, a di1ferentiable manifold. Let {e,} be a besis of V, and let {r,} be the dual basis. Now let p e V. Then there is a natural identification of the tangent space V. to Vat p with V itself. given by

Ia'-a a • r,


I a,.,.

It is easily seen that this identification is independent of the basis chosen. We shall often make use of this identification in describing tangent vectors to a vector space as elements of the vector space itself. In particular. if a(t) is a smooth curve in V, then (2) 6(t.) _ lim a(t) - a(to) • ''''. t - to

3.10 Eumples

of Ue Groups and Their LIe Algebra

The real line IR is a Lie group under addition. The left invariant vector fields are simply the constant vector fields {)'(dldr):)' e IR}. The bracket of any two such vector fields is O.

(b) . The Genn-al Linear Group. The set gl(n.lR) of all n x n real matrices is a real vector space of dimension nl. Matrices are added and multiplied by scalars componentwise. As we have already remarked. gl(n.lR) becomes a Lie algebra if we set [A.B] - AB - BA. The general linear group GI(n,lR) inherits its manifold structure as the open subset of gl(n.lR) where the determinant function does not vanish. and is a Lie group under matrix multiplication. Let X'I be the global coordinate function on gl(n.lR) which assigns to each matrix

1M Group' tmd 1'Mir 1M Algebru


its ijth entry. Then if (I, .,.e GI(n,IR), . X'I(O'.,.-I) is a rational function of {XI:'(O')} and {XI:,("')} with non-zero denominator, which proves that the map (0',,,,)- 0'7'-1 is C"'. Now let 9 be the Lie algebra of GI(n,IR). Let ex:gl(n,IR). - gl(n,lR) be the canonical identification of the tangent space to gl(n,lR) at the identity matrix e with gl(n,lR) itself. Thus for v e gl(n,IR).,

Then since GI(n,IR). - gl(n,IR)., we have a natural map gl(n,lR) defined by

(2) P(X) ex(X(e».



We claim that P is a Lie algebra isomorphism, and in this way we consider gl(n,lR) to be the Lie algebra of GI(n,IR). P is clearly a vector space isomorphism. We need only show that (3)
P([X,1']) [P(X),P(


whenever X, Y e g. Now, (4) 1.)(.,.) .. X,~O'7') "'" I x.t(O')XI:~"')' I: Since Y is a left invariant vector field,


(Y(X;I»(O') - dl.(Y.)(x'l)

"'" Y.(X;lo I.)

.. I)

.. X;I:(O')Y.(XI: I

I X(t(O')«(YJI:I

I: Using (5), we now compute the ijth component of (J([X,y]): (6)


"'" I X,I:(O'){J(Y)I:I'



- Y.(X(XH»

"'" I {X.(X.J{J(Y)I:I "'" I {{J(X)J(Y)I:I

I: .. [{J(X),{J(Y»)'I'


- {J(Y),t{J(X)I:I}

Hence (J is a Lie algebra isomorphism. (c) Let V be an n-dimensional real vector space. Let End(Y) denote the of all linear operators on V (the set of endomorphisms of y), and let Aut(V) c End(Y) denote the subset of non-singular operators (the automorphisms). End(Y) is a real vector space of dimension nl, and it becomes a Lie algebra if we set



LMGroup' A basis of V determines a diffeomorphism of End(V) with gl(n,lR) sending Aut(V) onto GI(n,IR). It follows that Aut(V) inherits a manifold structure as an open subset of End( V) and is a Lie group under composition. Under the nat~~, i<i~ntification of End( V) with End(V). - Aut(V). (where e denotes IIlHaeiidty transformation on V), End( V) inherits a Lie algebra structure from the Lie algebra of Aut( V). This induced Lie algebra structure is precisely the one described in (7).


denote the set of all n x n complex matrices, and let be the subset of non-singular ones. GI(n,C) is known as the complex general lineargroup. gl(n ,C) is a lnl-dimensional real vector space, with a basis consisting of the matrices 6~1 and ..!=i6iJ (i,j - 1, ... , n) where 611 is the matrix all of whose entries are zero except for a I in the ijth spot; and gl(n,C) forms a Lie algebra if we set [A,B1 - AB - BA. GI(n,C) inherits a manifold structure as an open subset of gl(n, C) and is a Lie group under matrix multiplication. With considerations entirely analogous to those in Example (b), one sees that the natural identification of the Lie algebra of GI(n,C) with gl(n,C) is a Lie algebra isomorphism. Thus we may consider gl(n,C) as the Lie algebra of GI(n,C). Let gl(n,C)
GI(n,C) c:: gl(n,C)


Similarly, in analogy with Example (c), if Vis a complex n-dimensional vector space, and if End(V) denotes the set of complex linear transformations of V, and Aul(V) c End(V) denotes the non-singular ones, then Aut( V) is a lnl-dimensional Lie group with Lie algebra End( V). Definition A form on G is called left invariant if 6/olJ) (0

3.11 (1)


for each (1 E G. As in the case orleft invariant vector fields, it is not necessary to assume that left invariant forms are smooth, for smoothness is a consequence of the left invariance. We shall denote the vector space of left invariant p-forms on G by ErIDV(G), and we let


E'*lDV(G) -


E~lnv(G) .

Left invariant f-forms are also known as Maurer-Cartanforms. The following proposition is the analog for left invariant forms of 3.7. The proofs are straightforward, and we leave them to the reader as an exercise.

t. .
i ...

HomomorphimU 3.12 (a) ProposldOD uft invariant forms tire smooth •




(b) E~DV(G) is a subalgebrtl of the algebrtl E*(G) of all smooth forms on G• tIIId the map w - w(e) is an algebrtl isomorphism of E'~DV(G) onto A(G!'). In particular. this map gives a naturtll isomorphism of EflDv(G) with G! tIIId hence with g*. (In this way we shall consider EflDv(G) as the dual space of the Lie algebra of G.) (c)


If w is a left inVtlriant I-/orm and X a left invariant fJectorfield. then w(X) is a constant function on G. tIIIdthis constant is precisely the effect w has on X when w is considered as till element of the dual space of 9 as in part (b). (We shall consider w(X) either as a constant function on G or as the corresponding real number, the particular choice depending on the context.)
If w e EflDV(G) tIIId x, Ye g, then it follows from 2.2S(f) that (I) dw(X,Y) .. -w[X.Y]. ut {XI"" , X,} be a basis of 9 with dual basis {WI' ... , w,} for Etmv(G). Then there exist constants Cillo such that (2)
[X"XI]" IC'I~I:'





(The Cmo are called the structural constants of G with respect to the basis {X,} of g.) They satisfy

I (CH,.crl:.

+ CI,I:" o. + CI.".cne + c.,.,.crle)




The exterior deriVtltives of the w, are given by the Maurer-Carton equations (4) HOMOMORPIUSMS 3.13 DeftDidons A map .:p: G - H is a (Lie group) homomorphism if .:p is both Coo and a group homomorphism of the abstract groups. We call .:p an isomorphism if, in addition, .:p is a diffeomorphism. An isomorphism of a Lie group with itself is called an automorphism. If H - Aut(V) for some vector space V, or if H "'" GI(n,C) or GI(n,IR), then a homomorphism .:p: G - H is called a representation of the Lie group G. dw, ..

I cl...wl:




If 9 and I) are Lie algebras, a map ,,: 9 - I) is a (Llt a1,ebffl) Mmomorphism if it is linear and preserves brackets (,,[X, Y] - ["(X),,,(Y)] for all X, Ye g). If, in addition, " is 1: 1 and onto, -then" is an isomorphism. An isomorphism of 9 with itself is calIe4·,AA>tIItOlPJ() - End(V) If for some vector space V, or if I) - gl(n,C) or gl(n,IR), then a homomorphism ,,: 9 - I) is called a rtplY#ntation of tht Lit a1gtbra g. Let ~: G - H be a homomorphism. Then lince ~ maps the identity of G to the identity of H, the differential t/fp of ~ is a linear transformation of G. into H.. By means of the natural identifications of the tangent spaces at the identities with the Lie algebras, this linear transformation d, of G. into H. induces a linear transformation of 9 into I) which we shall denote also by Thus




d~: g-I),

where if X e g, then d,(X) is the unique left invariant vector field on H iucb that

d~(X)(e) .. d,(X(t».

3.14 Theorem ut G mul H be Lit groups with Lit a1gebl'fl89 mull) respectioely, and Itt ,: G - H I¥ a Mmomorphism. Then
(a) (b)

X and d,(X) art f9-relatedfor tach X e g. d,: 9 -I)

u a Lit

algebffl Mmomorphism.


Let Z ... d,(X). Then Z and X are f9-related. For since , is a homomorphism, 1.(.) 0 , , 0 I.; hence


1'('(0'» ... dl.(.)Z(t)


d1.(.) d,(X(t» ... d(l.(I1) ,)X(e) ... de, 1.)X(e) ... d,(X(O'».

This proves part (a). Now let X, Ye g. Then for part (b), we must show that
(2) [X, Y] -


[Z, Y). [Z, Y).

By 1.55, [X, Y] is f9-related to the left invariant vector field In particular, (3)

• i

i i

But [X, y], by Definition 3.13(2), is the unique left invariant vector field on H whose value at the identity is d,([X, Y](t». Thus (2) holds, and the theorem is proved.


[1', Y)(t) - d,([X, Y](e».


3.15 Meet of Homomorphisms on Left Invariant Forms



,: G - H be a homomorphism. Then 6, pulls left invariant forms on H back to Jeft invariant forms on G, since whenever Q) is a left invariant form

· t




61.6.:p(w) - 6(.:p 0 I.)w _ 6(1.(.)


- 6.:p~.(.)(w) - 6.:p(w). Moreover. the mapping 6.:p: EllDv(ll) - EflDv(G). considered as a mapping of the dual space of I) to the dual space of g. is precisely the transpose of d.:p: 9 -I); that is.




Recall that if {w1 ••••• wJ is a basis of EI1DV(Jl). then there are structural constants {cm,} (see 3.12(4» such that



dw, -



It. WI'

Thus since d and 6.:p commute.


Now let W1 and W. be the canonical projections of G x H onto G and H respectively. Then the ideal J of forms on G x H generated by the collection of independent l-forms

{67T16.:p(w,) - 67T.(w,): i"'" 1 ••••

, d}

is a differential ideal; for using (4). we obtain for each i, (6) d(6wI &p(w,) - 6w.(w,»

- I cll:I(67T -I CII:I[{6wI
1<1: 1<1:

&p(w.,)It. 67T16.:p(w/) - 67T.(w.,) It. 67T.(wi»

It. 6WI &p(wi)

&P(WI:)- 6w.(wl:)}

+ 6w.(w.,)


{6WI &p(wi)


which belongs to the ideal J. Mpreover, observe that the L-forms (5) generating the differential ideal J are themselves left invariant l-forms on G x H. This follows immediately from applying 6/(".f) to the forms (5), where (O'.E) e G x H, and from using the fact that WI 0 I(".f) == I" 0 WI and I(".f) "'" If W•• Note also that the basis {w,} of Eflnv(H) is a basis of the l-forms on H in the sense of 2.33.
~. 0 0

To carry out the above construction of a differential ideal on G x H. it is sufficient to start simply with a homomorphism of the Lie algebras rather than a homomorphism of the Lie groups. Let G and H be Lie groups, and let '1': 9 - 1) be a homomorphism of their Lie algebras. 'I' has a transpose '1'*: Eflnv(H)- EIlDv(G), namely,



"'" w(tp{X»




Let {CI)Jbe a basis of Ei.,,(If). Then we claim that the ideal J of forms on G x H generated by the collection of independent I·forms


{~"'l(y*(CI)J) - ~"',<CI)~,b,t:=

1:.... d}

is a differential ideal. This follows from a computation similar to (6) together with the observation that (9) To prove (9). it suffices to prove that both sides have the same effect on an arbitrary pair X. Y of left invariant vector fields on G:
(10) d(,,*(CI),»(X.Y) -

J f


- -CI),(tp{X),tp{Y»)

i .

- dCl),(tp{X),tp{Y»

- l<lI ICI~II"


Here the first and third equalities follow from 3.12(1).

Finally, observe that, in this case also, the forms (8) are left invariant on G x H. We shall make use of these ideals on G x H in proving existence and uniqueness of homomorphisms in various situations.

3.16 Theorem G be connected, and let .:p and" I¥ homomorphisms of G into H such that the Lie algebra homomorphisms d.:p and tltp of 9 into I) are identical. Then .:p - ".


Since d.:p - tltp, we have

(1) 6.:p - 6,,: E!IIl..{H)-E!lllv(G). Thus we have two C<» maps .:p and" of the connected manifold G into H, both agreeing at e E G, and both having the same effect of pulling back a basis of I-forms from H. Thus since the ideal of forms on G x H generated by the I-forms 3.15(5) is a differential ideal, it follows from 2.34(b) that .:p - y.

Lm SUBGROUPS 3.17 Deflnitious

(a) (b) (e)
(H,.:p) is a Lie subgroup of the Lie group G if

H is a Lie group;
(H •.:p)is a submanifold of G;

.:p: H - G is a group homomorphism.