You are on page 1of 169


-, A '&-*: + "2 FOUN@ATIOkS
., .;\y
1*:., i
I; ,:

Number,.. 15, vol
.e: >&?a~ :
By Shoshichi Kobayashi 6ad Kabumi~No&zn T r

. ,.a - ‘.

University of California, Berkeley, California


Brown University, Providence, Rhode Island

t 1963 : ?I,
a di~is&n zf John Wilby i So&, New York
a dl~ision of .i& Wiley & Sons, New York +* s


Differential g&etry has a long l&dry as a &ld of mathematics
and yet its rigorous foundaticvn in”&e realm of contemporary
mathematics is:nlatively new. We h& written this book, the
first of the twe;ivolumes of the Fdatiwdf D@nntial Geometry,
with the intition of provid& a”l v&&tic introduction to
differential ‘&ometry w&h @ also, &ve’ ‘a~ a reference book.
Our primary conce~ was to make it ,+l&&ta.@ed as much as
possible and to give compl& ~MO& of ali’&d&d results in the
foundation. We hope that; && pw has been $chieved with
the following arrangtments. In chapter I we have given a brief
survq ;of differentiablei m&nKMs, ,I$& mtips Ad ,fibre bundles.
The readers who-areu&amil& G&h &din.,qay learn thi subjects
from t+e books +f Chevotlley, ppin, Pontrjagin,
and Steenrod, listed in the Bilk aie our standard
references in Chapter I. We a coficise account
of tensor- algti &d MU& fiek& th&“&n&l theme of which
is the notiib;i ef8~h’aFtihc ‘k&ebr& of tensOr fields. In the
Appdiccs, we have @Gn some ‘r&t&s from topology, Lie grbup
theory and othirs which %ve n’eed & .the iain text. With these
preparations, the main t&t of thi book is &l&ontained.
Chapter II -c&ntains&e conilection theory of Ehresmann and
its lateq development. Results in this chapter are applied to
linear and affifie cdnnections in Chapter III atid to Riemannian
connections in Chapter IV. Mat~y basic results on normal
coordinates, convex neighborhoods, dis!ance, cotipleteness and
Monomy groups are’ p&&l here completely, including the di:
Rham decomposition thearem for Ri&mannian m&lds; ‘I -
In Chhpter v, -we ‘int&&ce the sectional curvature of’ a
Rieme lM%&Id.‘a&i tbt Spaces of constant curvattire. A
more complete treMment!&“&$ert& of Riemanni& manifolds
involving Hctional ‘&&+attrS depends on “talculus of iariations
and will tie given II. We’ discuss flat’ affine and
Riemannian connec a*. IV.
In Chapter VI; trqnsfornihtions and infinitesimal
transformations which p&cx+e a given linear connection or A
Eemambn metric. We include here various results concerning
R$ci tensor, holoztomy and infmitesimal isometries. We then’
.’ V


treat the wtension of local, transformations and the so-called
equivaleqC Q problem for a&e and Riemannian connections. i
The resuh in this chap ter are closely related to differential
geometry ,c$ qhomogeaeous spaces .(in particula*$~ symmetric Contents
sPaces) wt\i& ‘are planocd for Volume&I.
In .al!, t& chapters, we have tried& familiarize $b. readers
with various techniques of computatio&phich are curmtly in xi
use in di&rential geometry. These T:,- (1) classical tensor Interdependence of the Chapters and the Sections
calculus with indices ; (2) exterior differew calculus of E. Cartan. C HAP T E R I
and (3) formalism of covariant differentiatiqn a,Y, which is thd
Diffrentiable Manifolds
newest among the the’& -We have also jllu,sfr;tsed as wk see fit 1
1. Differentiable manifolds . . . . .,;‘.L. . . . . . .
the methods of using ,?. suitable
, bundle or &o&&g directly in) . . . . . . . . . . . . 17
the base space. /.,, :.’ 2. Tensor algebras 26
3. Tensor fields . .. . . . . . . . . . . . . .
The Notes include B&F .,histprical facts and supplementary . . . . . . . . . . . . . 38
results Pertinent to thi .&i;&fent of the present, v&me. The 3. Lie groups . . .-.:, 50
. . . . . . . . . . . . .
Bibliogr@!Y at the en&&+* only .tboge booti and papers 5. Fibre bundles
* which we quote thro~$$&uZ~~,~~k. ,+ -_ I ‘. ‘.$ f C HAPTER II
Theoqems, prop~i~~n~~~~l~o~~~~~ ;IKe”&mbered for each Tbeory of.. Connections
section. For example, in e+& ,&@@qs qayr, &apter II, Theorem 63
1. Connections in a principal fibre bundle . . . . . .
3.1 is in S,sMon 3. IrJ.the re&&,+~~~ r; it will be referred . . . . . . 67
2. Existence and extension of connections 68
tQ simply as Theorem 3.l,:~~~iR~ot~~~.~IFUbsequent chapters, I . . . .i . . . . . . .
it is referred to as TheA)rem $.I cQf.(&pta&* 3. Parallelism . . . 71
. . . . . . . . . . . .
‘.Ve originally phmg$ toAte ~@&&&,which would include 4. Holonomy groups . 75
. . . . .
5. Curvature form Bnd structure equation 79
the content o< he present ,vc&@ ~5 .weU.+s &e following topics:
6. Mappings of connections . . . . . . . . . .
submanifolds; variat+ns of’ t& leng# ‘hgral; differential . . . . . . . . . . . 83
7. Reduction theorem ’ 89
geometry of complex and K4&r;unanifblds;,~fferential geometry . . . ., . . . . . .1, .
8. Holonomy theorem
of homogeneo?rs spaces; symme& spaceg;.cbaracteristic classes. . . 92
9. Flat connections . . : . . . . . . .
The considerations of tjple,;an$ spase,h3ve made it desirable to. . . . . . 94
divide the book in two,voluqes. The:top& mentihned above will ’ 10. Local and infinitesimal holonomy groups .. . . 103
11. Invariant ‘connections
. . . . . . .
therefore be included in.V@ume II.
In concluding the preface, we sb+d like, to thank ,Professor CHAPTER III
L. Bers, who invited. US to undertgke this project, a@ Inter- Linear and Affine Connections
science Publishers, a~‘di$sidn of Jc&n Wiley a,nd ‘Sons, for their . . . . . . . . . 1 i3
1. Connections in a vector bundle
patience and kind coaparat$n. We ‘se greqtly indebted tp, Dr. . . . . . . . . . . . 118
2. Linear connections
A. J. Lohwater, Dr. ,H. C)sreki,Mess+!A. Howard,and E. Ruh for . 125
Affine connection 7; . . . . . . . . . .
their kind help which resulted in maqy improvements .of both the . . . . ‘+ 130
4. .Developments
content and the prescnt+#on. .We al& acknowledger;- th.e grants of . ............... 132
5. Curvature and torsion tensors
the National Science Fo&&&n which support&p+rt oift:he work . . . . . . . . . . . . . . 138
.. 6 . Geodesics
included in this book., ;,,, j r: -;f . . . . . . 140
7. Expressions in local coordinate systems
.* .: .’ StiOSHICFiI fbi3AYASHI ? vii

. . 308 Curvature and space Forms Conformal transformationd of a Riemannian manifol& .y T~iWZWblU 1.g * * * * * 236 4 Holonotny anh &finitesimp!iaometries ‘. Affine holonomy groups . . 1.. . . i: 7 ..dgr~*. . 2% 158 3. . . . . . 1. -225 1 2. ix * f . . . .. . . . . +-. : . . . . 267 2. izl N&s k%P~R Iv Connections and holonomy. . . . . . . . 1. . . . . . . -193 2 . . . 1 193 The automorphiim group of a geometric structure . Spaces of constant positive curvature . .’ . .. . . . . i?: Rie-ashdan conneclions Complete affine and Riemannian connections . . .‘i . . : : . . . . . Equivalence problem . Factorization lemma . Flat a&e and Ri &ar&nconrLkt&i . . .groups . . .. . . . dimeiisions . . . .. .. . . .s .. 146 CONT&NTS !a Linear infinitesimal holonomy’groups . ..~tioas . . . . . Extension of local &morphisms . 315 . . . 277 6. . .. 325 CHAPTER vz. . locally compact metric space. differential equati0n. . 281 7. . . Symmetric spaces . .is separable 269 3. . . . . 309 1 . CONTENTS tr* Normal coordinates . . . . . . . . . . . 366 179 6. . 256 IAPPRUDICES 1. . . . . . . Holonomy groups . 292 2. . . . . Infinitesimal affuw tansfbrmatwns 229 3. SpaCeS of constant +rvature Bibliography . . . . . . . . . . . . . Normal coordinates and ~&I neighborhoods : : 1 162. Irreducible subgroups Of o(n) . . . . .: -. 272 4. 172 Parallel displacement of curvature ... 204 4 . Riemannian metrics .. . . .The decomposition theorem of de I&& : : : : 167 L i n e a r c o n n e c t i o n s with. Flat Riemannian manifolds . . . 284 .‘I 1 1 * ’ ’ 2 4 4 5 Ricci tensor and infinitesimal iaomet& ’ . 209 ... . Kemannian connections . . .Lie group . . . : . Green’s theorem .. * ‘252 7. ‘: 1 1 . . . . ~emapp. . .r~urrent. . . . . . Completeness . . . . . . . . . . . . . 275 5.. . . . Groups of &met&s and affine transformations with &APTRRV maximum.. . ‘. . . .~urv~ture . . Sectional curvature . A connected. 154 Ricci tensor and scalar curvature . . . : : : 248 6. . . . 297 4. . . . Algebraic preliminaries . 201 Summary of Basic Notations . . . Ordinary F. . .. . . . . . 366 5. hometries aryl infinitesimal isom& ‘. . . On an arcwise connected subgroup of a . Partition af unity . . . .. . . . . 313 3. . . . . . . . . . . . *?‘. . . . . .

8 requires Section II-IO. A @d&&~~of %@.n F.m 3.. VI-1 ‘$-5.‘tlie identity transfbrmation of Lr -- IVI-3 I n-11 is in r. I.:-Z. . then P(‘“) is a .. the homeomorphism f’ df . if f is con- Chap1t. mean. pseudogroup of transformations ‘of RI.4 requiws Proposition i.4: numbers (xl.r IV: ‘I’hcorcm 4.&&& ‘space S is a set set oikni into Chapter I\‘: Corollby 2. If r < s.class CT of R” -is &he :f*et of! Chapter \ I : Corollary 5. x2 .off) of S 0.6 requires Example 4. their the restriction off to an arbitrary open subset of the domain off is i. I homeomorphism of IJ’ onto V’ and if% V . OsW.I. (5) .. .:. then..1 in Chapter V.8 domain . ” j x it (2) Iffc T’. “ i I (4) -For every cfpen set U iFS.t. continuous. -@ Jj# bsrhe pwofu. 2 . Interdeprndence of the Chapters and the Sections I ui’ 1 Ii-1 to 7 r-7 I I III-l. thenf-’ f I’. .is. I .upla~ of cl&O.-2.*f-“ 5 I: is a p-4 VI-5 cl .6 in Chapter I\‘. in 1‘. . r of transformations sai$y&g the foli~wG3~ a&ms: ? . c’ontb a‘n?open s&df S belongs to I’ if the’restric- tion off .ously i”fiR”) & ca. . homeomocphisms J d.. IO rcq:zilcs Section V-2. .A fitdppingJ of an . : ‘.on&o VTadi.C$ r =+1. -.3 requires Section \. By cla&CX&yz. 4 wquirrs Section 111-7. x”) Chapter lil . ~~th’th~~~~~t~l~g~~~. (4) requires Section III-4 and Proposition 6. hotie@merphism ef an open.I1 : ‘Theorem 11.-.$&h& on’s &I. I ’ .~0 . such that both f an&f? iare .2 requires Section 111-4.another *open set (called the range off) of s . By -class CO l&z mea.t0 Ul’is in I‘ fcir Lverv i:. f’( P n Li’) Exceptions is.?. aa. set (called the of ~~~~-.n U’.: P r o p o s i t i o n 2 .an .+ I-10 .~ is hen-empty.: (1) Each J-e F is-a.. Proposition 6.tkt Lf is real anal& :Tbe Chapter 1’1: ‘1’hcon.2 II-8 r. . A homeo- mkphisn.4 requires Proposition set of S. R”! js sai&+l. b%.2 in C!lnptcr JIJ.. 1* 446) :Iff* 11 is aq bomeamorphism~of U .4 in Chapter III.6 - : .Iffc l’..ijcof. Ii kkgi?#S:qfe~ examples of psixdogroupk~~l&h ai: used in this Chapter ..n:&peaiset +&R” onto: an. own.2 thatf:$ Chapter 1. ~(3) &et Us= uUj where. . each Ui. : .of j-1( K n U’)bofitb. @dagroup 6’(W) . tinuously r tirnq &@&qentiable.~&#@g#iv7natitilPs of. C h a p t e r \ ‘ I : ‘I‘hcorrm 7. set of*w Chapter \‘l : Corollary 6.

) where (Vi. I +anifold.t& ‘VI&p X. a dl&entiable manifold or. X” 0 vi defined on Ui is called a Qj 0 p-l: Sp(U n Vi) + yi(U”i Wi) local coordinate q&m in -Ui. reverse$ o+ntdt$ori where tyi is the’composition of i. if x”. .. defines a dii$erentiable structure of class’O. (a) Each Ut.. ‘a~cliart ‘con&s of an elementfof I’i(R*) and the domain called the dimension of the manifold. We shall iden Cr gives rise to a differenti#e structnre ‘of class $7 uniquely. %’ f: M *‘MI is said to be dif&&iable of class clc.WF shall mean an allowable chart.. .it is possible to find a chart ( Ui. -$I.&w h’ r&ii. . (TW&W! *he bIc Given two man&l& &” *‘MI of class 0. vi>.yqf~iions of C* cm. vi) of.‘:. . . r(v) c r. Then .then the.&‘~aturd’ manner R” is an’oriented manifold of class Cr for fixed complete *atlas compatible with P(R”). a?. l?(O) is a.!‘q a conipIex mar$bld.xs~. ‘8 Under this id&iii. .’ every chart (Vi. . @Ied charts. .a fixed complete atlas com- (i.kis the borhood.& &&e of . x”) + (-&.:-’ . SimilaAy.. Ix”] < a for with l” k compatible with l?. Any open sr&et N of a space compatible with P(R”).g. R*“.~. x”) of R”. q. DIFFERENTIABLE MANIFOLDS 3 subpseudogroup of l?(R”). is non-empty.P(R”). yj) of M’ such that . the mapping qi 0 $’ of transformation (xl. . compatible with r. y’.$ . &) is a chart 0. T h i s subpseudogroup.(afc&&j(&. ti.e. r--AxawmJ vr IJlPP~RaNTIAL ~qyp=. where . . . ’ ‘. for .:-:A complete atlas of Mcompatible with r is an atlas of&f com. IIl. k 5 r. pro$‘o&ris fact to t. The ps of class C’ 1s a Hausdorff spa . If af2ariZly . U.. . is a homeomorphism of If l” is a subpseudogroup of I’. &I cation. we shal15pn~y i$$. . Vi is then called a cub+ tuighbor&od of p. of rq(RsR) for. 1 ” ‘.a$>le with I’(CP)“’ ented differentiable manifold numbers with the usual topology.U. . y”) e R2”. k differentiable mamk~ld of c&s @” is “also called a rcal’ktj& manifold. F orientable man&Id of class 8. patible with r which is not contained in ther atlas of M For any structure under consider%io~ (e. Let C” be”tl&#&e atlas comp.are positive everywhere. . ... vi) of M and every chart (V. . the system of. a diffekentiable structure of~4ass (?fldefinh uniquely a M and y$ is the re&&@ tif ++ . then an SatEas of Mucompatible Vi onto kn?op& set of R” defined by &#nes the manifold with the (b) Each qi isra~hor&&norphism of &i onto an open set of S.( p6. a’ mapping we shall hostly consider differentiable manifolds ofi t&s cm. . . 2.) A complex (anabtic) manifold of QR”).f{mjly of &2&s . when necessary. I. manifold. . is a coordinate neigh- is an element of I’ whenever U A Vi is non&rpty. . .functions x1 0 qi. . structure of class Cr).i” 4&*&o. enlarged to al compkte ad& manifold M of&&~ is a manifiold of class C in a natural manner.~~(R”).. if. If we consider only those f~ P(R”) whose Jacobiang . Not every difherentiable str&&e of&l& Cr is thus dbtained..he reader. s~!y~. Any atlas ofJa RMdorff off. yi) complete atlas containing A. be patible with. .an oriented group r is a fa~~t~&g~ gUj.actly two orientations An atlas of a ti Mcom$Q..) onto qj( Ui n U.$ = xi + &j. .. From now on. ” . .called orientable. . . .any Y. given an atlas A = by a chart . I such that &) i the origin of R” and v. P) E Cn into (x!. . with the j’ (c) Whenever .$. we obtain a sub. Si$@rly?. s’ ly. every complex man&d is oriented as a ma&-‘ . di&ren&ble structure of class C.ov$pt of $ and CJU. l it is obtained from an &kntied Ont$&“is. .iile %v. differentiable. n A!‘. Since gf(Ui A U. Given an allowable chart (Vi. some‘“”yhve“* number a. = M. ‘. For every point p of M.) is an element of r. . by mapping (21. $&&s &. we shall mean a pseudogroup of . if it is connect&L I&$&g the. .bz+n. . an n-dimensional manifold M of class CI. We say then that U. The integer I is’ auy “Y. A dt@xntiable manifold of class c’ is a Hausdorff space with a ’ .(w). An orien&j differentiable structure of class defined and will be den& ii by%+@&). complex analytic) . an allowable chart is a chart which belongs tained in a unique complete to the fixed complete atlas de&&g the structure. for t < s. f&i complex manifolds.. fact. is calied~ the pseudogroup of o$ntatt complex dimension n is a Ha rff space with a fixed complete thy of doSs 0 of R”. ii ‘.Since p(R”) 3 T’(R”) a chart of N is &ve&‘bi (oi.. fold of class C. ~st) d&es anfo$ented manifold. . denoted by ‘differentiable manifold of class T y. Every atlas of M eo e with I’ is con.&l < a.


which proves tll,kt cvcry \‘cctor at p is a linear ccimbination 0.
fir’,, = I;, the n$pping ylj Lf 0 vi I of q),( U,) into wj( V,j ‘is (apfl) ),, . . . 1 (a/&) i,. C:on\,crsely, gi\.cn a lincnr combination
differentiable of c$ass 12”;. If ul, . . . , U” is a local co$inate syst&. c p(aptfl) ,,, ronsidcr the curv~‘&finecl by
in IT, and ,vl., :‘;. . , v r’s is a local coordinate systefn:;\i yj, lhenf I$ -.-= IIJ( p) i- pkp I j = 1, . . . , 72.
Then the vector tnn,g:cllt to 21?iii &rve at t -~ 0 is x S’(a/&‘),,.
may be” YxprcsSed by a set of diffei&tiable r:. ‘: functions, oft~~~s~‘~~~:
II ‘r
,,tal. Efl(d, . . . ,TP), . . . ;.v,‘” ep(d, . . , , li”). i!- ” T o prove the l i n e a r ind&~~ence tJf (aj+‘) xj, . . . , (a/au”),,
.;,. assur11c ?; p(a/allq,, :: 0 . Trig’:’
By a’dQ@rentiable VM/$& or ,simp&, a ‘$.@ing, we shall &$,n a : ~~~*~,::; j :. ;

mapping of class C” .,A &Gxentiable f&&ion 0~ cl+ Ck, on # is 0 Tz x p(auqj3+j; ,;.+k f o r k - l,...,?l.
zi inapping of class C” ?f A4 into R. Thet&fin,it& ‘1 of QT. a h&&@& This complctcs thd *probf of’&r as$crtion. ‘I’hc set of tangent
(or complex anal$k) *ppifig b; function i?si&lar.
vectors at p, denoted by 7’;(%) or 7’,,, ’IS callrcl the tcqenl space 01
By a dferenti$le &e of cl& cn: in M;;ck~~a$~$$$ c@&&$-
M at p: The n-tuple of not&&s El, . . .., :‘I \vill be called the com-
able map@ng,g,f _.“clasd Ck of a closkd intq@3”fgl.b] 6% R.‘into 111,
@one@ of the.vector 2 p(apj, with respect to the local coordi-
namely, tEe’ rcst?iction of a differentiable mapping of claps CLi$f nate system d, . . . . , Y”.
an open inteival containing [q-h] into A4. We shall riow define’s Remark. It is known that if a manifold AI is of class C,‘“, then
tangent vector (or simply’ a @for) ai a point p. of ‘M. Let s(p) be tke T,,(M) coincides with the space of X: 3(p) -+ R satisfking condi-
glgebra of differentiab!e‘functions ofcla’ss C’ defined in a neighbdi- tions (1) and (1) above, where z(p) now denotes the algcbrn of all
good ofp. Let x(t) %e P c&% ‘& &Gs’ Cl, n 2 t S,;fi, such that C” fimctions around p. From now on we shall consider mainly
x(t,,) = p. The vector tansent to_the cufve x(t&$‘p 1s a mapping manifolds of class c,“‘: and mappirqs of class C*‘.
X:. iQ) ,*.R de3ined by+“‘:. , ‘,‘I. .I ’ A&ctorjeld X on a mafiifold 111 is an assignment of n ve:tor XP
., /’ ’ Xf = ~,d&(t))/a)~o. ‘$
‘. to cd point 6 of A/. Iffis a $iffcrentiablc function on IZ~, then
XJis a functioh on .li defined by (.yf) (pi -:: LY,,.J A vcsctor field X
.In other words,. dyf is the ‘&i\i”&%e off in‘ the~directidn ‘of the is rallcd djfhvztiobfe if .x/-is diffcrcntiablc for cbverv differentiable
cllrve x(t)% t’== to. Tlie~~~rror,;~~atisfrdsthe fdllowinq conditiond: function J In terms of a 10~21 coordinatr svstcn; uI, . . . , ~‘1, ii
8. .L )I “:
(,l) X is a linear mapl+&r#$(p) into R; 4. vector field ,Y may be cxl~r~sscd by .Y x IJ(~/&J), whcrc ij arc
functions defined in the coordinat<% r~t~igl~l>nrlloc~d, called the
components of .Y with rcspcct to $, . . . , 11” . .Y is differentiable I if
and only if its componc~rrts E1 arc diff~~l-c:ntiablc.
Let F(M) br th e s,$ of‘ all diffirrntinblc \.cctor fields on .\I. It’
is” a’ real vector spacr undet tllc* ri;ltural atl~lition and scalar
the dimension’of M. Let ul, .,. f , U” be a local coordinate syqqn multiphcatioh. If X and I’ arc ill X( .\f), dcfinc the. bracket
in aLcoordinate neighb&hood G pf p. For ea& j, ($/,@j), is a [X, IrT as’s mapping from the. riii 6 of’ fiitic:tions on .\I into itsctf
mapping of 5Cp) &to, &,, which satisfies condit;ons lQl+vd (2) bY
ah:ove. We shall s&v. that‘?t.he set -of vyct@$ at b is ‘thq.,vqctor [X, Yjf == X( I:/‘; “l’(A4’f’).
space with basis (i318~‘)~, . :,. , (a/a~‘~)~,.‘. c;iy$n any &r% x(t) We shall show that [KI’J is a \.ciLor ficlu. In terlns of a local
with h ‘k x(t,);‘l&‘uj - x’(t)j.,j -- 1, . :.:‘n, ibc its equAtio1~!5 in coordinate system ul;.* . . , II”, \vc
terms of the local coordinate sy2tt:m $, . . . a $!, $ThqqV, .).,” . . \Vrite
(nf(x(t)~~po,” ;I; .xj (@i&q, . (&j(d)~dk\jo~:*, ‘.s
c 2 5,: ,,,,!,,, +‘yf11.14 T-o::,!~~,,;.. ’
* !“lll I?,,: \Illnnr:l!icirl Ill),,,lirri:, -(‘”

‘_ \
Then A l-form w can be defined also as an g(M)-linear mapping of
[X q”f = Zj,E(SP?‘w)~ - ?www>(aflau’). the S(M)-module S(M) into S(M). The two definitions are
related by (cf. Proposition 3.1) ,
This means &t IX, yl is a veeto’$ield whose components 4th .
respect to G, ‘1 . . , ZP iffe given 6 &(~~(a$/&~) - $(w/&.P)), (y(X)), = (wp, X,>> ,& WW, P E M.
j= l,..., n; With respect to this’ b@ket operation, Z@f) is ,a
(of infinite dimcnsioris~. Let AT,(M) be the exterior algebra over T,*(M). An r-form OJ
Lie algebra over the real number
’ is an assignment of an element of degree r in A T,*(M) to each
In particular, we have Jacobi’s id
point p ‘Of M. In terms of a local coordinate system al, . . . , un, ti
can be expressed uniquely as

0= Is d1 <$,< . . . <,.,A *.,. .+, &I A:. * * i dd’.
We may also regard 3s(“Aa) as a module over%& algebra B(M) of .”
The r$hm ‘ii, is - called di$erentiabh if the components f .
are a&differentiable. By an r-fm &q shall mean a differenti*%;
differentiable function on A# aa &Jlows. Iff is a function and X
is a vector field on M, then f X is a vector field on AJ de&cd by r-form. An Cforr& w can be defined a& as a skew-symmetric
(fX), = f(p)X,, for p d M. Then r-linear.mapping ‘.over .5(M) of X(M) x X(M) x . * . x Z(M)
uxdy1 =.m rl +f(Xg) y - 9WjX (I times) into. s(M). The two definitions are related as follows.
Ifo,,..., w, are l-forms and X,, . . . , X,. are vector fields, then
AS c S(M), X,Y e X(M).
( * * * A 0,)(X1, . . . , X,) is l/r! times the determinant of the
For a point p of M, the dual vector space T,*(M) of the tan’genr ztk (w,(Xk))j:,-, ,..., r of de-
spaceT,,(M), .;S called the space of covectors- at p. An assignment of We denote by *b* = W(M) the totality of (differentiable) r-
a covector at each point p is called a l-foti (dzjhential form of forms on M for each r = 0, 1, . . . , n. Then BO(M) = B(M).
degree 1). For e+eh functionf on M, the total di$erential (df), off Each ID’(M) is a real vector space and can be also considered as
at p is defined by an s(M)-module: for f E G(M) and w 6 W(M), fo is an r-form
defined by (fo),, =f@)wa, p,r M. .We set .& = B(M) =
(km,, m = Xf for X E T,(M),
Z~=J)‘(M). With respect to the exterior product, D(M) forms an
where ( , >‘ denotes the value of the first entry on the second algebra over the real @umber field. Exterior di$erentiation d can
entry as a linear functional on T,(M). If uf,‘. . . , Ua is .a local be characterized as follows: \
coordinate system in a neighborhood ofp, then the total drfheren- 1.’ (1) d is an R-linear mapping of D(M) into itself such that
tials (du’)., . . . , (dun), Form a basis for T,(M). In fact, ~they ’ d(T) = w+‘;
form the dual basis of the basis (a/aul),, . . . , (a/8tra), fbr r9(M). I (2) For a function f 6 Do, df is the total differential;
In a neighborhood ofp, every l-form o can be uniqu&’ written a~ (3) If-w e 9’ and r c ID”, then

w = Z, fj duj, d(w A 7r) - doh,r + (-l)‘o~&;

where f, are functions defined in the neighborhood of p and are (4) d2 = 0.
In-terms of a local coordinate system, if o = +. . .<.i,& . . a@ A
called the compowh of w with respect to ul, . . , , u”. The .l:form . .I,
* * . A du”, then dw = pi,<. ..<i, dfil...i, A duil *. - A du’r.
w is called di&wztiahle if fj are differentiable (this conchtron is
independent of the choice of a local coordinate system). We shall It will be later necessary to consider differential forms with
only consider differentiable 1 -forms. values in an arbitrary vector space. Let V be an m-dimensional


real vector space. A V-valued r-form w on 91 is an assignment to (2) If (f,) p is surjective, there exist a local coordinnte .r_sstem ul, . ..,u ‘&
e&h p o i n t p 6 hl a ske\v-symmetric r-linear mapping of in a neighborhood U of p and a local coordinate swtum ~1, . . . , v”’ of
T,,(M) “: ’ - * x T,(M) (r times) into V. Ifwe taka basis e,, . . . , f(p) such that
e,,, for V, we can write CO uniquely &s CO = CJ’!!itiji-+,ej, where
09 are usual r-forms on. M. w is dgerentiabie, by definition, if qf(q)) = ui(d for q E U and i =: 1, . . . , n’.
CO are all differentiable. The exterior derivative du> is defined to
be Cy!Ll dco’ *e,, which is a V-yalued. (Y + I)-form. In particular, the mapping J U -+ M’ is open.
Given a mappingf of a manif&& J4 into another manifold ‘G’, (3];& (f,) D is a linear isomorphism :& T,,( .W) onto T,,,,(M’),
the d$Grential at p off L&e linear mapping. f* of T,,(M) into then f dejines a homeomorphism of a neighbwhoon U ofp onto a neighbor-
T,,,,,(M’) de fine d as follows. For each X l T,(M),.choose a curve hood ‘$’ off(p) and -iF ‘iitversle f -l : V -+’ U is XSO di$erentiabli.
x(f) in M such that X is the vector tangent to x(t) at p = x($). For the pryof,: se% E%‘+lley [ 1; pp. 7%80;1. .’
,:, ‘2 i.,
Thcn,f;(X) is the vector tangent to the curve f(x(t)) at f(p) = ,j.’ ”
1.(x(t,,)). It follows immediately that if g is a function differentiable ,& nqpping f of, M iniq &fT ?i ,.cal&l an immersion if .(f,)@ is
‘in .t neighborhood off(p), then (f,(X))g = X(s of). When it is ,injqc$ve &r every point p of.M. We say then that M is immersql
necessary to Specify the point p, we write (f*),,. When there is no in M’ by f or that ,M is an immersed,,~ubma,+fdd qf M’. When
an imme&n:f is,injective, it is &lled,an. imbedding of M&to M’.
danger of confusion, we may simply write f instead off,. The
transpose of (f,), is a linear mapping of T;,,(M’) inio T,f(IZi). Wesay then that EM (or,the image f (kl)) &an imbeddedsubmanifld
For any r-form w’ on Aii’, we define an r-foim f *o’ on kl by (or, simply, a subman~$~ld) of M,J. A submanifold ,mF.y or-may not
be .a clqsed gubset of, J4’. ThertoRplogy of a subin+fold is in
(f*d)(X,, . . - , X,) = ~‘(f*X,, * *- ,;f;x;), nduced- ii-om
^ M’. 3.n

x1,...; Xr E WW.
Ej;ample 1; 1.. Let f be + “function defined ,on-a,in;l;inifold Mt.
‘I’hr exterior differentiation d commutes with f *: d(f *w') = Let &4 bc the set of Mints p,s .%?$I such that.f((p), = 0. If,(@),; #, 0
J‘* (h'). at every point p of M, then it is possible to introduce the structure
;1 mapping f of M into ‘21’ is said to be of rank r at p E M if the of a manifold..,& M so that ,M is a closed submanif0lk.J # J4’,
tlirncnsion of f,( T,(,M)) is r. If the, rank .of f at p is equal lo called the &perp@e de&d b2 the equation f = 0. More gen&l~y,
II dim i$f, (f,), is itijectite and dim M-6 dim M’. If the rank let-&4pe the.’ zerQs bf fqnctionq;f,;.. . . ,,f;-defined
of:/‘at p is equal to n’ = dim .21’, (.f;) , is su<jective and. dim 111 1: qn M’r.,:!$ the dime=@, say k,Lpf,t&,s.ubspace .of r:(M) spanned
dim .\I’. By the implicit function theorem, \CC haI2 bl::k#il,, .,) ..;, (P~~)...~is.i~pepe;e;e;e;e;e;e;e;e;e;e;e;e;e;ent,pf~,g M, t$n kJ is. a closed
~‘KOPOSI?.ION 1.1. I.ei f be a mappit~~~ of .!I into Al’ and 1) ~2 fwint submanifold of @’ of dimension dirn.44’ - k.
oJ‘ .21. “A ~&j%vfiibr@is~‘~f a‘ Manifold y c%i, ahother mahifold ‘;I!’ is a
(1) !I‘(./;),, is injectiw, there c*.rist (I loccil coordinnte system u’, . . . II” hotieor&P@&& p duch that’ljm 4p and F-’ ai’e difFerentiable.“k
in N nri,~hborhood 1-J of 1) nnd n loc~~l coordinate s_rstem c’, . . . , ~1”’ in (1 diffe&norphisn% 6f?Mk &&“&elf is calie$. a: d@xmtindle ‘f&m+
nc’i,:rlrhr,rhoo(i q/.f‘( p) .wrh that formation (or, simply, a trunsform.ution) of 39. A transf&nation a,
‘ :, of M, indukes -an &it&&phi.& cp* 3 the algebra P(M) of
~~‘(.f(Y)) fI’(Y) jilr q c I' crnd i:-l.,..., n.. 7 differentiak..Qn+ on &&an&,& particular, an automorphism of
r the algebq~,$(M~ of f@cti.wcq.N: ‘. *
IN Jwrticrtlirr, J‘ is II homcomorphi~m of li onto f (CT). j

I 0Ff is .3. equ&to ..oplro. .. 1 .g.&&@ n(M) and Y c r..(N).a local matter.. q.(M) + T. wJ).. If thei%&.respectively. by the &&rule. p.prelated to [X.@b ~&hkk‘i&tq@man&&dbf is. w@ete even if A and.(M x-&f&e tie UC&U tangetit to&e cwve (x(t). S.A. %%ey. If M is defined by an atlas A = ” ((U.. choose ment to each point p ofM’$n +Iimerision&l stibs$a& 8.’ F~OPOS~ON 1 . F .wkMt~b Let local baris foT tlik&kt%b@n S in.p L . c 8. I.: . Similarly.N. -+ Rt*“..y(t))lru):-. .angwlt apace T. q)‘in belong to ‘S if X.& natural di@ke#j& structure on the topologicalspace M 4~: N cd on M’ into a &ffe~q@a&m (p*(w!) Q&M.)). r] belongs to S whenevef ttH0 vector field3.&. is defined in :a nat+ @&%ll $-e N. + by an atlas B = {( Vj.. ‘.~ manifold of N(p). Ike- i\.ri: ‘A ‘vecto$&@ .Ui x V.*U.: t . ‘9. see Chevalley [ 1. By a dismbution MW ‘shall always mean $ differentiable dis. . tb.t’ ~&(~j. Theri ‘[X. Den (X. e vector 2 a T.q.. for any &m&on f dn M 2.). M u(N) c w. -Note. yet)) . .:. DIFFERENTIABLE MANIFOLDS 11.ppiut. which is identifie# &h. observe that w” nt?+@ eB .tx. vi x !v. a vector field !Fdki A4 ijl’+relu&d!t~ a veCtor field . In other words.. . For the p&. I$ X’and Y are qkrelated to X? &$&$ every p&t (p.&-is went to x(t) at p =+ x(&J and I! is c+llid d$f&ttMB~if evefy &Snt p -hWae neighborhood U Y is tangent to. Proposition 11. q). X.. 941. Xi.j(M x N) be the vector [X.Jf. p.LThe’ set X.. called an kc@ manifard of ’ N-M x Mwhich sends q’cUinto (p..yW at (A 41 = (. where f is the becwe.. .. Let f be a difkntiabk mapping.contains N.dM x NJ respectively.for a!1 p c iU. ky. q) and r is the image of Y. for.... ahi related by We now define the product of two manifolds M and N of ” #‘(k+xlf) f X(4p*fl f=‘Is X(M) and f~ 50.. dimension m qd rr.y(t))/dt). Y) c 7’. We $Iso state . the #+ws a u@que .rie: be connected since the differeiitiabiliky . X is the image of X by the kapping ‘M -+ M x N:. 2 . 95. Let W be a submanifold of M z&we integtar manifolds of S.k and Y belong to tangent to&e curve (p. Then’2 ==X+ l’.8ot. XV is call&P’k to the curve $” 4 = W.(.w mt s&d a is define&&y a&t&s C.!~~~~(t~~Y~~jkU)l-b . Any integral man##d t&ugh p is an.J.X’ on m. ... let Y z T. 10 FO~DAT’IONS : .!N. Through eoery . int.a. For h>l& = Xi. ~kPi$%al~&d inooluti~e if M Y N at (p. = xf + Pf.y(t) at q =y( curves x(t) andl(Q such tl+ . L’. S which .$e~ the second ‘axiom of count- >bi&. &ofiM zc. .(N)) %k ‘8. for Up c ML Finally. which is tangent form a basis oPSh’at’*‘q.@$ x .o imbeddirig of N into M.adas is ..4iffer&~~on-El..wi’&t~ the @e&t -8um T&M).#’ Xc T.rV.d&$wtion & Q &ar$$d M. .. For the ‘pr~@~ s& Chevalley fl. N k ci4ii&#J~ nm*indd!iide&l mar&d WS. Let S 6. Yl is .w D##&lWMAL GEOMETRY .s that this .(M x 4).. an ihd&c . Ftor fiekls b y I &% = ~i%l. ’ ’ 4 3’ P+ ” can * be identif$edj. then f is tf$k@a& It induces also an automJ>rphrwn pi* of the Lie algebra T(M) of as a mapping Oj N znto W.Gnal intug4 man#kld N(B) of S. Zf = (df((xCt). B are complete. q).which sends tribution...‘J!I~m4y.&?.+oJa manifold N i .. . The classical theorem of J+obeniuk &k %e ‘formulated as follows. i PROPOS~ON the mapping A connected subman%old~~N4$$# is. ~2 S be an Cwtk M.)) and.q’).(N) in a A distributkk S of ditiensiti I on a manifold ‘M&&i natural marmer. then the Altktigh any mapping up ofA#into k’&i&u d&rent&$.X is said :t.IJ W sati. + $4). .t. We say that v. pi. Yr].. x vector field on Mint0 a vector fkld on M’ iri general. the dh&bntion S iff*( M x N at (p.: : p’ E M into @‘...hcre place analytic$y >“er$ by differentiability throughout and dd -P.(N) amlt.. where qDi Y”Y~: U.

pjs(uj). We now prove the converse.%}. I.G(t) tit%(t. Let X be a vdctor fi&J. defined for u = (ul.L%. If pi.‘.a/au ) 1 in [J. then the preceding prop~$G&rcducrs to the formula:Z := X + ?. I. kt ~17. For each point po of M. anh’ ‘f.: u -+ M. A curve’x(t).j’.J l Z.t(P) -= ClddP)).:. . vl. such thatp. : <y%. a.. . d‘f) = vl*(‘y) and v:*(F)’ = g. then X is i&@ &@&X:is a solution of ‘ . q’) ‘for. u). = As in the case of a l-parameter group of transformations. :3 System of ord48tirgt diReie&&&$&ions : - -.. .n with unknoivn functions f’(t). it follows . . 1.0 such that. Proof. . .*(W + y’2*w:. q’ 6 N. which induces the given X.s. tflerz we say that X is complete..(Y). By the uniqueness of the solution. Th&h. Let ul..some~ e. . 1:.L.t $. U” : Induces X. . . .. which form a solution of the differential equntio? for each fixed IL and satisfy the initial conditions: c: :’ f’(0. x M for . such ‘that ul(p. u) = ui. the Sotbwing * ..21. u).’ Set 97t(r. .i ItI < el. I‘?‘(&.f”(t. H ence. . .j the.!). DIFFEKENTIABLE MANIFOLDS 13 (2’) If t. . ‘I) for 1’/ E 41 sand y&‘) = ~7(p. t v :‘: *\. ai14 let . and if.. and q2. In -fact.that induces a vector field X defined on IT. . then Fl(P’) = 9 (I’. ?. for”every paraxti&te~~#Je . Let X be a vectorjeld on a manijold M. PJf.positive number E Note that if 1’ = M ‘x &i&d 9‘ is fhe ickntity transformation. u”) with lujl . .. ..5. i = l.) z x(0). lu’l F. ‘. theri the functions gi(t) -fi(t d. Since pO is thi identity transformation of C’. u”)(&‘/&“) in W.$?%k We shall say that X generates a local l-parameter group of vector Xx(. . 0 and E :. . .‘JIf It‘}. yS(p) E 11.on a manifkld Al... defined for f (global) l-parameter group of transformations of Al which ItI c : E for some E 5 0. .‘ : Let X = I. . x == x p(.s.*m =. . u) are easily seen to bi%‘sbQtioii of the differential equation for the initial conditions &@j =f’(s. . . for (tl < &I and zd in U =: {u.ii&:llp< is called an intej+” 6t$ve 6f 3T if.f”(t). vt(U) c U..) = .*IsI and. P ROPOSITION 1. This proves that vI(v (u)) =: ~~ >(u).. ’ system of ordinary linear differential equations : - df’/dt = F’(f’(t).‘:Fbr any point pii local transformations pll in a neighborhood of p.. 24” be a local coordinate system in a ?Ilr’ neighborhood W of p. P)~ Proof. By the fundamental theorem for systems of ordinary differential equations (see Appendix l).e kxist 0 1. . t E I. Consider the following du’fdt z+ E+‘(t). If there exists of .).and ps(u)’ &%&+UI.for U = {u...) is tangenT”to’ the &rvk-. It + sl are all less than &I and b:th u. there exist a neighborhood U asp.. there is a unique integral curve -x(t) ‘of X. u). . we haveg’(t) A’f’(t..‘. ” j . there exists a unique set of functions fl(t.. Iu’I < b). i[ . f4)). 7 un(pO) = 0. v*(z) =: 9bt(X) + .(p) is defined on be a local iooidinbte system iti a neighborhtiQd @Of /I. ti(alj . u) f”(t.Qt::“--(y1(t. prb& the definitions ofiT. .” and a local l-parameter group of local transformations qp. Q E D ...&t)). .. .I 6.

?Wehave vector jeld v*X generates y Q q+ 0 v-l. 1.)). On a compact manifold M. Each formula:is valid whenever it makes sense.(p) . UV~ ~~ctmjidd X is The limit on the right hand side is taken with respect to the complete. This proves that qt is a diffeomorphism of u PROPOSITION for I4 < E. Consider f(t. i Proof. x A4 g(o. . take a hJ*XL = P)*(XJ E =&w . x M. natural topology of the tangent vector space 7’. $) =fl&#) .t * g.9. In what follows. .9. Let’ E’ = function g(t&) OR 1. Thenfoq. 1. then the . Gim a function f on y.8.8ii&jk?&& fon M. PROPOSITION 1. QED. ‘QED.(p) is.. I. j it is obvious that pit induces the given mujr’field X in U. p E M} interval ( -e. and E(P) a.). fields or transformations function g. there exists a the domain of definition for vector.RRN'I'LU. p). the v&or X. Hence. then fhnatkm defined on 1. s(p. . d). min k(PJ. It is sufficient to define ‘+! QED. p) =f’(o. p E M. is an open I+) X U(p).U~JJL19 P c MS coincide on U.(M) is tangent @the curve 4) = ad at 4 = x(0).f(p) and apply Lemma PROPOSITION X generates a local I-parameter group of local transfoonnations vt. involved. andg. &r each fixed p c M. For each point p E M. ~-t(q&>> = &Q). p).((Pt)*yJ* Rmark. for some E. We now give a geometric interpretation of the bracket [X.. x U induce the same vector field on U. hence.!) is a @t&n on I. Y-J =ynl[Y . we shall not give explicitly the domain of QED. .I . definition for a given vector field X and the corresponding local l-parameter group of local transformations qt. yt is a local l-parameter group of local tran+ a local l-parameter group of local transformations qr.(t) ” v*X = X. In the course of the preceding proof. ‘. where I.(M). A vector jield X is inzwiant b&! q? that is.*X.&)) = tpo(u) = u for every ItI ’ oE&o vect& fields. p) = 0+&r all p E M. x M such tlrat f(t. Let xg#?@u& p~. yl < E.6. .s on M. p)@h t&f 0 (py = f +. Let 9 be a tranformation. [X. on R x M. Their. b4*Y)f= (r(fo QiHPCS * wL4t. COROLLARY 1. thntfi yolt ~j + t * g.. if and on& if p. i z 1. 7. It is clear that 9 0 qt P 9-l is a local l-parameter group of local transformations. and it is easy to specify. LEMMA 2. E T. and. in ItI < E 1. + t ’ m>. let U(p) be a neighborhood of p prove two lemmas. we showed also that if two local I-parameter groups of local transformations vt More precistly. Proof. DIFFERENTIABLE MANIFOLDS 15 “: 14 FOWuMTIONS OF DIFFJ?.#) = t * g(t.‘ ’ function g..!(P). To show that it induces the vector = Fi i&(P) = go(P)* fie’ld 9. they yt Lx.7. such. Since A4 is compact. positive number such that the vector field X gem-rates a local l-parameter group of local transformations q1 ‘on L EMMA 1. Since q+ induces X.defined on 1. is tangent to the curve y(t) = q o pt(q) = 9 o pt 0 @(& QED. . It follows that the vector 1 Proof of Propos&u 1. = Xf on M. Setp(t) 7 v. =f + t*g. . It is clear that v. If X genera& u (E U and Itl c e. then there exists Q has a finite subcovering (U(p. x U. We first Proof. such thut f (0. The function g(t. and defined on 1. .= g(t. .th f t = aflat. From the construction of vi. k). Thus. let p be an arb&ary point of M and q = (p-‘(P). p) is defined. If -a vectorjield Proof. Let X and Y be vectorjW.. If f (t. if necessary. of M. commutes with yt. the open covering (u(p) . GEOMETRY .. and go = Xf (Lemma 2).. % = fim: CY.

10. Then yt 0 lyS = yS 0 ql for every s and t if and only . V)jN. Suppose X and Y generate local l-param%r bilinear mapsjf?g.11.$ince U.Y 1s d con- . For a fixed value of s. . the free vector space generated by the C OROLLARY 1.I0 (SW Remark.I Corollary 1. The uniqueness of such a cvory Q7.(%+t)*Yl (ru. if [X. Let N be the vector subspace of we have more genera@ ..TU) . 0 pt =.2. U @ V -+ S.8. .e.9.ace and f: U x V -+ S any COROLL\RY 1..y. Define.(Yf) .Y. Tensor algebras and We fix a ground field F which will be the real number field R or the cornpIe? number field C in our applications. Let S be any vector s. &very pt. ja. By the universal fadtorization property of (U @ V. set U x V as a basis. l QED.5’.).f. (u. Let Conversely. [X2 fJY= 0. With the same Gotti&rG as in Proposition 1.@ V. QED.r(u. by Corollary 1. The cbuple (U @ V. Mfe define the tensor product U @ V of two vector spaces U and V as = 0 fey e$$? < by ( W. = pf E(vJ*Y . u) . u) considered as an element of M( U.9. i.v’). extend f to a IQq$+~r&~q mapping f’ : M( U.8. respeciiveb. v) -l -&. All vector spaces we coilsider arc finite dimensional over F unless otherwise = X. Obiiou-.y. $4 v .I&y.f’ induces a linear mapping Proof. Therefore. Since f is bilinear.(u.g.+@$qt . v).(%+J*yl:’ . we can groups q’t and yS. ‘where u& E. (9. V) spanned by elements of the form (u + u’.J) and ( W..v). ’ Prbof’.8.f [X. v) .\I( U. I.. By :. We set U @ V = M(U. y) be Z+ couple hati@$%hi’ universal factorization property Corollary 1 . V) -+ U @ V will be denoted by u @ u. V). above). On the other hand. 91) has the universal factorization property for U X V.for U x V. then (U..+ 0’) .the canonical bilinear mapping 9 of U x V into U @ V by . v) where u f U and u E V.W be a vector space and y : U x V ---* W a bilinear mapping.r(u.3 *X = X by Corollary tion proper0 f&-U x V if for every vector space $ and every bilinear 1. Then we have .9..r : <i. I’] = 0. 3 V is a basis for M(U. f = g 0 cp. Y is invark&bY g:. q) stantvector at each point /J so that Y is invariant by. its and apply Proposition J.+ W.=..(4 4. Since (Q~J* p reServes the bracket. irz tire sense that +&e exi& a linea. By Proposition 1. v) B U x V. Let.v’ E V and r Q F. then d((y. for any value ofs. y) has the universal Remark. Proof.. pairs (u. V) . y) has the universalfactoriza- since 9.. If a coupl~“-( W. image by the natural projection M( U.‘~. Y] = 0. The conclusion of Corollary 1. QED. f ’ vanishes OR ~ .o 0 F.. consider the vector field (PJ * Y For evei+ pair ‘(u. iROPOSITION 2. 0 vs = Y)~ 1 yr for every s and t. V). .. yl) are &omor@&. We say that the couple ( W. isomdiphism (T: U @ V . ZJ) = ‘u @ D for (u. every lys commutes with every vt. DIFFERENTIABLE MANIFOLDS 17 16 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 2.10 can be written as factorization proper& for U x V. f [(q&Y . . stated. = EJ3’l.(% 4 . (&. we obtainIa I mapping f: U x V --+ S there exists a unique linear mapping / g: W+Ssuch that-f =goy.. mapping g follows from tb” fact that 91( U x V) spans U @ V. Let M(U. V) be the vector space which has the proving our assertion.U? v. If q‘. ‘&. b?J *[XT Yl = 1’. f Therefore. .1.

. For a vect?: space U.indW:6W. $e couple (Us .. Hence. : then’ Proof Let fi U X V+V@U be t&e. g is the desired isomorphism. ‘p:. there exists a unique linear mapping 0: u 0 V+ which sends (~1.4.t~(u.. th. ui) into forY~V.:*. I f absfisfor U&&.’ .v6V..~~ u1 Q * * ’ Q up Then.2. V8U. PROPOSITXON 2.c+op by U* the d&al vector space of PROPOSITZON 2.“~tO.V=dLnUdi~. QED. + L& aknotes the direct sum of U. .1.i. + :QV+.:U.1 to the case with more concludk that T 0 (r and 0 0 T are the identity transfbrr&o~ of than two mappings is obvious. @ . and Us respectively.5. :i of V induce a linear mapping iI: ?a @ V + (U.QU&. 0 i.. Thenp. projections..2 can be also generalized. PROPOSITI~?J 2. .*. 3 VI and ps: U.j=l n) is a basis be the multilinear I$*.l iubspqco of V spanned by v. ‘&I ‘Uk l abasisforV. -c V. PR6P&gON 2. 2. linear mapping g: U @ V ~v@vsuchthatg(u@o) Av@u.Usingthtuniquenessofg in the definition of the universal factorization &perty.b~the such that g’(v @ u) = u @ v. The mapping f just given will be U X v and W respectively. u”> denotes the value of the linear there is a unique linqaz yua##g fi.-. @I * * ..ere is a unique isomorphism of F @ U onto U which A O 4 are the identity transfkmations of U.3.hplnrtinrlar.. (Q+ . This proves the first isorkphism~ The proof for the second is similar. $kh set@ (uI. there is a unique linear mapping g’: V @ U + U @ V Proof. is a I-dimensional vector space onto UncI. 0. 8.@I be tb? ai~tmcwi~nal s@pace of U spanned by . u Q (V. The proofs of the following two proposi&ns are similar and mappings.It is meaning@ W write U @ V @ ti Given vector spaces U.+‘u. .. . . . &. . + U. @ V sends 7 @ u into N for aN r c F and u E U.4. i spanned by uf Q v. V) = v @ tl. ‘p) can be charaet&$cd.‘I ( Of. 1 .Q~~~i~X. be the identity transformations of U @I V and V @ U respectively. there is a unique Similar~.~:W-~UOV)suchthaty=uo~(r~p.. tp = ToOoQ)and~=dOToy.dimU~. is defined inductively. ..+U. + U. for U @ F and U.. By l 8~. Vd such that Gnktional u* on u.. ?@r~kaun&etiom4q5&smof(U@ V) @W “ 4.6..U. . DIFFERENTIABLE MANIFOLDS Consider the bilinear mapping U. + u Q v. and vector space ova F. and the identity transformation hence omitted. .~. By Proposition 2. there is a unique isom~rphisro of VI @ * * *I@ V.. can . ontouQ(v@w)whiEh~~uQv)~w~Y~(vQw)fo7 By the the pniversal factorkN$b ~dih~orp.. Similarly. Both ba 0 i. . .. denoted by. QED. . @ V and U. Leti. @ U.@u. 'PROPOSITION 2..~=~~~). -+ &$ = 1.~~sGndruOvifftovQuforollu~~C!dvrV. :. If we regard the ground&&dF qs a l-dimensional Similarly.. which s e n d s &&y.l -.4.. we The generalization of Proposition 2.18 FOUNDATIONS OF DIFFERENTIAL GEOMETRY I ProoK I. &va&h@+~Jf.(%)& all Ul T Ul a?Jd 4 6 u. .Qv+U*Qv.. QED.(uJ and apply Proposition 2. Proposition 2. . + @‘U. @I U. and fil 0 .a are the zero mappings.he~~r prdaucf U. we.U.jJ . p))... For any permu xof(l. + V.U. c For u 6 U and u* 6 U*. vi x * * X‘ ‘. @Y V. By Proposition 1. &b W.. U.)QV=U. t. .:.:U. . x U.Uk --+-. if U. of ( W.~U. we ob#&.4. Evidently. and p1 Q iz are the zero .. QED.Qv* Therefore.l$lhear mapping (u.) @ V. U. .1. and&o i* arethe identity transformations Hence. QED.1. + U. 19 (resp.. The is a unique i.+U.+U. ‘6.... ‘trs iijixtr opositkk 2. u=w 0 * * * 0 %(n* . v. Let pI: U. %) intofi(uI) @f. g’ 0 g and g 0 g’ are the injections..andi.1. . . k). W(resp.. @ V. &et.. Zt follows that II 0 Z. and U. and Jl are defined. + VA = u Q v.x**:XUk’ b .3. defined by f (u. Similarly.. . ’ fh 0 u2) =fX%) Of. By Proposition 2. respectively and & 0 f.1. of U.somorphirm of U @ V onto PROP&ON 2.-+U. ailu6U. (u.~.

‘r (u. . . Tl is nothing.. QED. v*).. . = V* 0 ** * @ V* (S e :rificati@$f t&se .. )$?‘I 0 * .. i = 1.. . . . . u 3E u..j.’ take bys for and PI. . V) e. or tensor space OJ elements covariant tensors of degree S. = V* and&‘& contravariant degree r a& cuearia+ degree s. We shall show that L = x... v) su& th&. .. a basis for V. i.-3 i. y. e.= c-4. u*:v / \ :.1 -to the bilinear mapping Xj $Bj = 8. . T. Since dim U $3 V = dim L(U*. To prove that g k<w* iso. v. I.9j.e. with respect to {e. . e. Ai basis of Vi Let e.g(u. dual ekment of Ti is called a tensor. . can beexpressed uniquely as a linear combination k(u @ 4)u” L (u. = F. element $f T* will be called a ‘contravariant tensor of degree’ Y. u*>v.@ ei. . s). U*. T:=TIoT. ‘ that Proof.9. then For a change 9f basis’of V.. i=l . ... jr.j = 1.n. .matrix of the matrix A .7. T. . vanish.. are the components of i. u*)(u. .follewi. U @ V-+ . (g(u* @ u*))(u @ v). . and Ti = F. .i be two b&es bf V related by a linear transformation and. hence. . . . &... Let .))uf = IS akjvj tb the . v c V and u* E U*. G are the com&nents df R with respect to the basis Then there is a unique linear mapping g.i -.. can expressed uniquely as a linear combination morphism. V). . . u*)(u.pft_ype (r. . @ vj) = 0 where aij E F. . or tensor of contravariant degree r and covariant degree s. every covariant tensor L of degree s such that (g(u 0 u))u* = (u. its components Kil .given by unique z&morphism g of U* @ V* onto (U 0 V)* such that I.) so fir L 411:+. In particti&:~T~ = T’.*~.} ai?d {it} respectively are related U. let ul. f: u* x v* -+ (U @I v)* define&~ b y ff(@~‘ir*f))(u @ u) ~= If K is a contravariant tensor of. as the tensoP product convention. . all ai.. u. ... = cj. Similarly. . = (24. An rel&te&%jy..... we agree that “@is the ground field F itself. ‘the dual basis for u* and ul. a} is linearly indepehdent. For a positive integei’r. u*> ‘.shdl ptfi T’ = V @ . . the components of a covariant tensor L of dcgrcc s m-e times tensor produd) the’c&ai$rian’t’tt?nsor space of degree. . CQnsider the bilinear mapping. . L(u*.ir. C.ymatio&. * ‘ *. To’ prove that‘ g. .n. Zi’= Zj A{ej. 5 n} is a basis for T. .degree r.. .f: L/’ x V -+ L( U*.‘L. . . AIt ..=V’~. I PrOOf. e. A!#. By convention. .is caued the couariant tensor space of degree s an$lb We define the’ ‘[$ixec&‘@zsor s&ace of @c (r.‘. .. U*)V and apply Proposition 2:I: where #I .” = T. - g is an. ’ 8 cir. we . By Proposition 2. If r = 1. ‘Then T.~~~~ (V:r times \ We shall give the exp ksions for these tensors with iesp&.J.* QED. .. . . DIFFERENTIAB LE MANIFOLDS 21 V) be the space of . . 0 *. . (g(~~ @ vj) a V* : s times). . m. i. IfX tra. T.. .OV~V**.. .Blej iz 1 . V..6.isomorphism of U @ V onto L( U*. .’ 2 J s where B = (Bj) is the inverse. e n&e _. 3imilarly. urn be a basis for U. We now define various ttns~~*spa$?e$ ?v& a fix&I vector space JI V.of Vand . .C(.. (e. In terms of a basis e.f?r-m+s is left to the reader * tensor product) . . y. .‘. . and i. . by jpl ’ . where Lj.u* 5 .(r Similarly. 1’ (. V)! k . . . **@ v . u$$p.linear mappings of isomorphism g of U @ V onto basis for V*. . . er be a basis for V and el. of V... ..ei. . .8.v.r defined by (f‘(*... Apply Proposition 2. 1 . V) 5 : K = Xi . and V* and proceed .. . f. . ‘h&omoN 2.. ..Lj . the compo<ents of tensors are subject 0 = (C aijg(ui @ v.!J*..vorpl$m... Given ’ ttio vector spaces U and V. therk is i’ The corresponding‘cha-nge of the dual bases in V* is .. s). is ah’i& the pr?of of Proposition 2. . @ * * * @ ei. . u))u* = :u. . . j. i.w+t I 2. uf. but V. Every contravariant tensor K of degree r for a!!< u l U. .i = C .

e. so ‘that an <element of. .“.~.*eT$.. T: is. . @)w: @*** @ wz) e T x T into u1 @ * * * $0 write K(v.. .11.@. where K: l x are zero except for a @rite number of product that the space of linear mappings ‘of T” = V @ .C. . L) E T: x q by this bilinear mapping will be denoted by K @ L. which sends (ui @ * . Ti is... .~@*‘- ..’ c V*. By generalizing ‘(2.*..~v. .9.. c V. In general. following transformations $compon*ts: Proof.S 2 1. We shall now make T into an associative algebra over F by into F is isomorphic to the space of r-linear niappings of V x .oK:.* .7. in a natural way.. T.. in a natural way.. . .. Let Y.. tensor ‘K h T. @ . @v.5 are called the components of K with respect to the PROPOSITION 2.*....e.. but we @ll.* * @ v. . as an r-linear mapping V x . . Ajf.1.. If. of the tensor product.. the space of linear mappings of T we associate a linear mapping.*~w:~. in a natural way. every tensor K of type (Y.2. @ V them.jl . For .g .+.ei. J. x VintoF. ZJ.not go into it.. ’ ’ x V into F.. if K is given by K./0“1~ . it S e t T..9. .- defining the product of two tensors K C T: and L e-q as follows. a tensor K &+pe (I. By Proposition 2.. . @ V is isomorphic to the space of linear mappings of the dual We now define the notion of contraction. ’ * X V into V such thqt #C(C. . On the other hand.8.. @w. @ a.. 1). of the corresponding endomorphism. v*>.1)’ Rj::::$ &‘2. * @ ejr corresponds to an r-linear mapping the etitence of C follow from the universal factorization property of V 5 ..V arid V* Q V*. In terms of components.d change’of S&is Zi = Ej A$.. . To each space of T. / QED. ir. . r”pI.x V -+ F and Wl @ * * * @w. e.*. where Kj. . (K @ L)f::::w. into V. the contraction C Simllq inkeFpreta$m can be niade for tensors of type (r.x V into V which maps (ui. then u @ v* is a tensor of type (1. space of all r-linear mappingi of V x . we see that T.) = X4 Kj.. From the universal factorization property of the tensor prod- uct...isomorphic. e. s) in maps a tensor K l T: with components q::f% into a ‘tensor general. QED. any tensor K of type (1. The uniqueness and z Kji.. @u: @es* @v: v x *-* x V into V. . an given apY Example 2. is a .=‘Z. . we consider a q i n t o *=. 3 01. mapping of V x * * .. .j. We prove only Proposition 2.:::::::f:. @ ZJ~ @ or 0 . . .~w. . .. . . .9. v.. The contra&on C: ‘T: + F maps v @ U* into (v. u. . DIFFERENTIABLE MANIFOLDS 23 the dual basis el. we. IO. be expressed uniquely as PROPOSITION 2... @ V by Proposition 2. UJ i%o @ vf @ ’ . it folloys that there exists a unique bilinear mapping of q x Following the interpretation in Proposition 2. .. is a basis for V. .. = X. @PI .’ * @ v. . K. .~~~... to the vector p1. (CK)f::-fi-. if e. 1) can be regarded as a linear where the superscript k appears at the *&th position and the sub. by the universal factorization ordered pair of integers (&j) such that 1 5 i d r and 1 5 j 5 S. The image of (K. = K~P:$)+-~. that is T’. v... r) is an r-linear eJ4. . .@ v&1 @$+1 @ *. : : :. . ... property of the tensor product. Proposition 2. A@Q .. to the vector space of all r-linear mappings of V x * * * X V into V.‘: : : :.. In terms of components.) * Of%. CK d rr: whose components. ~K$y. T. then K = where ui ..j.V and vt . which is canonically iso- morphic with T. e. In fact. q 0 WI 8 . If ei...) c V. ej..% and L is given by el*. by definition. q>% 0 * * * @ 2+-l @ v.4 K = xi . into With this interpretation. = V* @ * * * @ V* is the dual vector space of T = V @ ..++a. T’ is isomokphic. endomorphism of V and&e contraction CK of K is then the trace script k appears at the j-th position.. Again. j.-&.ofT:into~S:whichfnapsarl~:*.. ur) c F for ur. .+ then e PROPOSITION 2.*. . * @ V.22 FOUNDATIONS 'OF DIFFERENTIAL CB~IWZTRY I. T is of th follows from the universal factorization property of the tensor r. to the vector basis el. have the space of all T-linear mappings of V* x * * * x V* into F. .. called the contraction and denoted into V can be identified with the space of r-linear mappings of byC..ti s."'.. S) can We shall now interpret tensors as multilinear mappings. is isomorphic. Proof. en of V*. .iiei @ ejl @ * . V @ T.

An inner product g on a real vector space. If A. equal to Xi Ki. provided that f preserves for a E F.‘c’. e\~!.. fix all u E I_ . tensor . . the trace of R and the contraction CK of K are both U*. tion C.e. Let.)~(.2.I of IV onto V. in a natural zq’ a tensor of type (1. and L. for all tensors A and L. U and CT*. say n. DIFFERENTIABLE MANIFOLDS 25 basis for V and K has components Kj with respect to this basis.‘) i. of . ’ ’ ZGclnrj>le 2.ncc of /j it II~. u) = 0 i f a n d o n l y i f u FL 0 (positive d e f i n i t e ) a n d ( 2 ) conditions : g(u. 1 2Thefd.. Since T is generated by F. U a n d V.lii:l -:. U o n t o V. Example 2. On the other hand.tnd I/* F I -*. then its (c) D commutes with every contraction C.D’D for derivations 11 and 1)‘./ /)ar/icnlar. It follows that 11 *Ii is iujccti\. called the extension of A and induces an endomorphism. Since a derivation D maps T:. F T.k we obtain a linear isomorphism of T( U) onto T(V) which maps algebra. ‘/ a natural 1 : 1 correspondence between Hence. .s of T( 1.a vector space. we *‘ ira\x-. we obtain algebra if we set [D. T h e n th(a a~~~onlolphi~rn of T( 1’) induced by an automocphism -4 of I’ maps the tensor A.~nl is from the fact that T(U) is generated by F. Since the tensor algebra T(U) is generated by F. say . 24 FOUNDATIbNS OF DIFFERENTIAL GEOMETRY I..contructions. fu* = id*-'U*. u’) = g(u’. into itself‘ and c. li] = Ilh’ . for u 6 V and v* E V*. ‘k maps T: into itself. Ii and Clearly. u*).ation ofT( 1. it T:(U) onto T:( It). O_ED.algebra isomorphism. with &he group of aulomorphisms of the tensor algebra T( r’) zelhich Since Dv = Bv. f coincides with the extension of -4. we have /he linear isomorphisms of a @&lor space U onto another vector space V and [he cllSqehrtr isomorphisms of T(U) onto T(V) which preserve f@e and commrrle rclilJ1. It is clear thit D 2 11 is a Lie algebra homomorl~hisn~. the* uniqu&esF follows mor$hic with the Lie algebra of endonlorphisms of V. Since f is type-preserving.K’w.f. easy to see that the extension of A commutes w$h . ‘1%~ cGylt. .1 iw- ‘I’( [J) + T(V) which extends A: U -+ V. given an endomorphism 11 of 1. V a n d V*. the group of automorphism of Vi’s isomorphic.ector spaces (b) D(K @L) = DK 0 L + K Q L)Z. the d<G\. extend D to a derivation of T by (b).‘.er:‘c t p crud conmufe rejilh conlractions. T h e extension of . The isomo~j~hi.3.‘. $.* 6 I’* ml& t!‘l>t‘ ai~l commutes \vith contractions. U then the cndomorphism corresponding to K sends ej into xi $ei. of T(U) onto T( l’) is induced Conversely.-. The Lie algebra of derir1alion. v is a Let T be the tensor algebra qver a vector space 1: A linear covariant tensor of degree 2 which satisfies (1) g(u.~~ery: contrac- Proof..c.’ isomorphically onto Ti( V) = V. Deriote the follows from the uhiversal factorization propcrt~ of the tensor IXW r-ic t ion (I[:/‘ 10 I * II). u) (symmetric). = +I’ into itself by (a!. its l’rool: ‘l’hc only thing which has to be proved now is that restriction to F. From (b) it follows easily that I1 mjps every clement of F inio 0. Dv* = . Th’1s isomorphism.. 0’1 = DD’ . transpose A* is a linear isomorphism of V* onto U* and A*--’ is a The set of derivations of T forms . II and I’*. I. 1) ant1 consider it as an cndomorphism of V.I~S T. In’general. l’KOPOSITION 2 .is . is the unique algebra isomorphism PROPOSITION 2. u) + 0 and endomorphism D of T is called a derivation if it satisfies the following g(v. It is also given by assigning to each derivation ifs resfriction to V.. then.*. Let T(U) and T(V) be the ten.I..‘ induced by an endombrphism B of C’ maps K into [I).ommutcuj’\vith every c2ntraktion C. denoted by the same letter A. a linear isomorphkm d @ A*-‘: U @ U* -+ 1’ @ V*. the set of linear endomorphisms of V forms a 1. Dv z-z Uz! for zt E V atid Dv* -= --Ij*.into the -fy+ Q tl*j)r-f(&ll*:) = (u. \VC tlrfinc Da 0 from . By Profiosition 2. 1 ‘(_ (LED. Hence.II~ isomorphism .ll*rl* where I)* is thd transpose of 11.& algebras over \. (a) D is type-preserving.* fiw i. pres.A and f agrees on F.. i. is a linear isomorphism of. It forms a Lie linear isomorphism of U* &to V*. Similarly. 11 is uniquely dctcrminc. Since j: maps every element of the field product.4.

. of T.2). and &X.. x E CT. . Z U. . . r) (reqh Qpe ( 1. X. Then Xi = Ci g’Y. .) at a point x depends only on the-values of& i = 1. . .1’ of x for i = 1.). . stated. . . .cci ?“. .**x X into QED. . which . letfbe a differenti- Jacobian.‘@x. . Example Proof. done by means of the Proof of Lemma. is an r-linear mapping of T. it is sufficient to cross section of a certain fibre bundle over M. . x 3 into $j (resp. let K: I x . we take Xi = a/&l.)). W’e say that K thin we have is of class Ck if its components Kj. . I 1) and hence. . . .. .(x). . ..‘(X) to each point x of N.X. . . . In section 5.. ‘. .*. .(M). T.each x. XT’. X... . and at x.a point x. In other words. s) defined on U is then expressed by first observe that the mapping K can . L ping of X x * * . .: T(Y) = X T. = K. in some neighborhood U of x.1) where the matrix (. so does K(Xr.. s) on a subset N oCA# is & assignment of a tensor r-linear mapping over 3. : : :j: are functions of class C’k.K(Yi. .). X. n L vector fields on M. . 4” :. . trpe (1. = from the. . A’. r)). . Xr) vanishes at x. X. Given a tensor field K.. j n. the com- K. g assigns an (0.. called t. it has to be verified that this notion is indcpendent. Proposition 2. In a coordinate neighborhood U show that the value of the function (resp. > r. course.@.y~~cj~~ OF YIFPCKLN IIAL ULUML’I’KY : . cd =I&‘.) = K(Y. By the lemma. .&. . x E M (cf. . .A.. . .&ld) with a local coordinate system xl. X.he comfionents of h’ with LEMMA. we : :‘+x* = cgl:?l:::k$ 0’.V be a coordinate system around x so that X. I) (resp. Y) for all. . i = 1. r) and (1. . 1(+) in U. *x X -+ 3 (resp.(X.(M) into R (resp.) =f X(X.. *. respect to the local coordinate system xl. a/&&). . .) is to be replaced by the then K(X.(M)) for . .9 (resp.9 and 2.1. IfXi = Yi in a neighborhood 2..) = Ci gi . . . we shall mean by a tensor field that of class C” unless otherwise Then Xi =fX1 and K(X.K(X*. 4). We shall give here show that if Xi vanishes at. For anyy .. . . K(+T. 2iigg. .‘iis the dual basis. . . 3 (resp. Conversely.(M) x . iYe shall interpret a tensor field as a differentiable To complete the proof of Proposition 3. vanishes atp. any such mapping can be considered as a tensor jeld of ppe g(Y. It has been n customary to write dss = I: iid dxi dxj for g.) in U. . 2. A tensor field R T. wh&-e KG3. *: * . have . . YE 3E. . TJ psnems of g are given.1. Iocalized. . another in&zpretation of tensor fields of type (0.. . E T. From now able function on M such that f(r) = 0 . The essential point of the proof i&.(K(X. . . by g&j = g( a/axi. This will imply that K induces an r-iinear mapping of.11.. the vectcu”‘. 3E) satisfying the preceding where T:(x) is the tensor space. @ &. X) = 0 if and only if X = 0 and (2) Conversely. X.). *. . . .) = 0 in Li. It is sufficient to show that if X1 e 0 in ri. . Namely..( (X. . into R (resp.d. A (pos$ive definite) Riemannian metric on M is a K(fA * * * Jx) . X7) . . inner product in each tangent space T. .1. We Of. x T. x * * *x T. 7)) g(f) =f’(x) = 0 for i = 1.). X) such that l$xam$e 3. . Since P ROPOSITION 3. . . R. to K. . r) Let x1.of a local coordinate system. X. X) = g(X.. .u(liQ. . . ^/ by proposition 2. . X.and f = I outside U. . . . Xn.I 3 .w fields . This proves the lemma. In terms of a local coordinate system xr. . @(J’ 0. . K(X. X) 2 0 for all X 4 X. This is easily. . i are functions on U. 5) be an OX tee (I. of K(X. .*‘@. .: A &ensorJeld condition.=‘f1 **-f... viewpoint of Propositions 2. J) over 7’.j: .> . as a ‘basis for each tangent space T.of type (r. . T. w e may take vector fields Yi and differentiable func- algebra of functions (of class Cw) on M and 3 the s-module of tions gi on M such that gi =fi and Yi = (a/8$) for i ‘= 1. WI.matrix between two local coordinate systems. on. type (1.. (X. in U. formula (2. .a for fi 6 5 and Xi E X.. = ?‘i(k) be the tangent space to a manifold 121 at a .) is an r-linear map- point x and T(x) the tensor algebra over T. on M can be considered as an r-linear mapping of X x . Let 5 be the zsifi (a/w). . covariant tensor field‘g of degree’2 which satisfies (1) g(k. .v’. 7) (re. A tensor jield K of t@e (0.of type (0. .“..

. WC define the alternotion A .) = . : \\‘c nolv proceed to dcfinc the notion of’Lie differentiation. r) and c/n) is is not complete. .ll) onto the tangent space T. SK is s)~mmctric and SK = K if and only if h” is (e) L. (S.‘. oZ~(Llf). = K.\I indu& an algebra automorphism of 2(..~). this linear isomorphism can be extended to an isomorphism of the tensor algebra T(g-l(x)) onto the tensor algebra T(x)..n\s: (:lK)(X. is an automorphism of the algebra its sign.I:OVE’I‘KY I. w i t h &pect t o n zmecfor K is ske\\. A cliffcrc~nti~ll form 11) of degree 7’ is nothing but a shall define the Lie derivative L. ‘If L. it is linear a’hd .r(a lhc summation is taken over all pcrmutat. we define a tensor field +K by .2. If 0) and 0)’ are jel4 X sntisJes the following conditions: dilkrcntial forms of degree r and s rcspcctivcly.F(n) f. . if K..28 FOt~NDhTIOSrj 0’~ T)IFFEKENTI. . Let X be a vector field on . .11) lvhich p rcscrvcs I type and ‘commutes with contractions.\I. .\Z) is an algebra over the Ical number field R.5). every transformation p of .12. .‘3 L. The mapping L.K of n tensor Jield K with resject to a coCn..c. 2 .y E . .(. is type-preserving: L. .\f) bc the set of tensor ficlcls of type (r. IA’ is skew-symmetric for any ii and that P PROP~SITIOX 3 . For any tensor fie1d. EYor the sake of simplicity.. commutes with every contraction of a tensor jeld.M -_. r .:‘.K’) K is.\I.Iiave r). Then and set X(.\L (. ZCvcrin/de 3. (1)’ is (a) L. that is. L i e dl$erenfiation L. By Proposition 2...Y and (I) A 0)’ . ? In this way. Proposition 1. Then Z(. which we dcnotc by $ Given a tensor field K. i. 2. 2. !. we set field K on . we assume that .ions 77 of (1. K.W r -. s) clcfined on .. F(T) *K(X. symmetric.’ (NJ = wLl.I&) =.))II~IP/~~--~~~o)I s can bc”de&d is follows. If y is a transformation ‘of -!I. WC .I E -21.(K @ K ’ ) = (L. For any covallant tensor K at * or anv covarianf tensor Z(M). (c) L. ) ‘Yo(r. .K) @ K’ + K @ (L..r(.. f = Xf fir ever_y function f.I covariant tensor. . . the multiplication 12.&i. is a derivation of Z(M). reader will have no difficulty in modifying the definition when X where TT is an arbitrarv permutation of (1. .\I.1s foll. . The . .2(M) . the fAX(l.-symmetric if and only if .-l(cfJ 1. its differential y* gives a linear isomorphism of ‘the tangent space TC. It is easy to verify that .l. DIFFERENTIABLE MANIFOLDS 29 Ewrrl~lle 3. .We.. for all .ct Z:(.11) := XcxL.$es .CI)) c Z:(M). ) XJ . For any K.Y = [X. or tensor field of degree r. It is clear that L.1 co\-ariant tC'JiSOY fic*ld of Q’~ be a local l-param- I. (d) L.11 eter group of local transformations generated by X. is linear. .. Proof. s.L . I. ~~ is a global l-parameter group of transformations of .riant tensor fi&l of tlcgrc:~ P which is ske\v-symmetric : vecforjek? X asmows. . then fOYall K. called the Lie’ dzfirentiation with respct to. . vectdr~field I’.K is \vli(. (b) L.t E 2(A1) ‘then’ (li’ @ L).\I and crI a lb&l I-Parameter group of local transformations gcncratccl by -X (cf. bting clcfined point1 ise. a.K CSZQJ{. . .(. K’ $.(Z:(.‘X.7 * **) . then (:I . .(M). I:or each t . ffJ’j. Y] fof eber_l. ) A-#. -. of 2(M) into itseIf which sends K into L.

so does It follows that if two tensor fields. 5) and x an arbitrary point of ~44. if a tinsor .D’(DK). S.9. For every tensor field K. . Then S is a derivation of Z(M). for any func$ion f.$ l. By Proposition hbod .X)f = of x with a local coordinate system x1. E U. Since D is type-preserving. b hkjPL Let K be a tensor field of type (r. c Set D = D. we showed that a derivation of l(. Iff is a function on M.%I) into itself which satisfies conditions (. Our problem is now to prove that jf a &f)(4 = yz f [. . It fbllows that there is *a iPector field X such that Df = Xf for every f c S(M).( . Ix.u ofx. In fact. then DK Qpe (1. This follotvs from B&.ie a/g&a oj‘derka/ions qf x(. (l.2. 1) forms a subalgebra of the Lit there is a unique tens& field S qf type (1.(M)..Z() and. . DIFFERENTIkBLE MANIFOLDS 31 Since $I preserves type and comFMes~. D . then DK and DK’ coincide on 1-y.. is induted by a tensor field tion iI+ced by S.111).‘. 0 = D(C(Y &G)) = C(D(Y & 6~)) Proof. is a CJ to M and assume that the equaliiy holds in a smaller neighbor- linear endomorphism of the tangent space T. .. thtti 7. where X is iz vector field and S is a tensorjeld of t$e (‘I.D. hence.1). Since K and f vanish at x. then [D. .a).Zlj. the set of Ieruor.. . tensor field of..I) ---+ 3(:VI) PROPOSITION 3. is induced by a Proof.I. the . can be uniquely extended to a derivation of the tensor algebra T(x) over T. Twa derivations D... Let Y be any vector field and C: 21(. If we observe that pl. . so does DK.( l. type (I.:k. it. Since this holds for every vector field Y. let V bc a coordinate neighborhood transformations generated by IX. Since D can be localized. it follows that the f* DY =f* D. and K’ coincide on an open L. jkns a f@]‘~O a n d f = l outside U.K +f* DK.13. we shall mean a mapping of Z (. x” and let X$ Finally (e) is a restateme& of Proposition 1. Then K = f * K and hence subakebm nfthe I.3.::.S] is zero on 3’y(. = S. of 2(M) coincide .21. By Proposition 3.. X E x(:\fj. it maps S(M) into itself and . Every derivation D of Z(M) can be decomposed the obvious contraction so that C(Y & 6J) = (I)(Y) is a function unique@ as follows: (cf. Then we have D = L.C(DY 00) + C(Y (3 DOJ) = C(Y 6 I~J).2.J. To show that a eaincides with the dekva.f!x) -f&‘%)] = -‘. We have But this will follow at once if wc show that I~J = 0 for every l-form w on . letf be a function such that other hand. QED.\f). From Proposition 2.. ’ and D.ret of Lie d{fffrentiations L. D( f Y) = Df * Y + D(D’K) .-f I= .with contractions. It follows immediately L EMMA ..f they that if D is-a .’ for every vector field Y. derivation D is zero on S(M) and X(. For any vector field Y.?.’ ==‘vLtii &local l-parameter g@ot$of local To show that DK vanishes at ?c. ‘* .field K vanishes on an open set U.$@+ion Qf Z (ilr) and S is a tensor field of type coincide on S(M) and g(M). define SK by (SK)..A. for each j. 1) such that DY = SY algebra of derivations of Z(M). then it is zero on Z(M).‘y f f&$%) l-f(x)].. Example 2. The set of all derivations of ‘I(-41) forms a Lie algebra over R We shall show that<an)r derivation D which is zero on S(M) is (of infinite dimensions) with respect to the natural addition and induced by a tensor tieid of type (I. Clearly. In other words. 1) is an ideal of th&Lie algebra of derivations oj’ 1(. .L. we see that L ‘.3. is sufficient to prove the following of type. satisfies’ D( fg) = Dfa g* + f * 02 for f.fie/dr qf that is. WC have D~J = 0. In the proof of Proposition 3. x E M.Y sitice Qf = 0 by assumption.1. On the v&$&es on U. the follow&g. 1). Df . . By a derivation of 2(M). Xi and Let S be a tensor field of type (1. S. U’]K =-I BY is a vector &@+I$. (b) and (cj of Proposition 3.g E s(M). M’e may extend Kj. is a derivation of Z(M) which is zero on s(M).(M). I). QED.Zlj. where Xi = 6’/axi and Q j = dxl. For each x E . = (Dti)j(k’). it sulkcS to ~116~ that 2.h if:and only ifi t is zero on 2 ( Af). set U. We first observ‘e that a derivation D can be localized. + S.l).\I.13. set”of all tensor fields S of type (1.’ . r :. multiplication and the bracket operation defined by [I). 1). 30 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 1.

. As a general formations generated b. .. ..5. . i. For anzl tensor field e.!(a) . Chapter I) : .@Sp. ! /txvqrJield & is @variant.‘we have ps(L. QED. .2 implies Proposition 3.@‘. .<(K) for all tensor fields K. K(Y..14).t:i PROPOSITION 3.Y!>’ E a(M) If 11 = L.Cj K(Y. I. For any vector field X. D(OJ A O J ’ ) = Ih A 0) A.(Y.(1) A BJJ fore. 0 8.K LL +.. .L) and D’ are derir. 1.g1 . DIFFERENTIABLE M..S := X . .tppittg 0 of ‘c(*‘I.> J?l which ho!il for any transformation v of M and from (v.=.. A:$ iF a derivation of de<qrec k -1 k!.f) into itself which satisfies $. degree k’ifjt~maps W(. where C. . ‘..(M) follows -from the by the Jacobi identity. 4 Proof. -. Y]f = L. coincide on s(M) and X(M). Generalizing Corollary 1. l ) Y.1 on 5(A4) and%(M).5.) a(M) is a linecir ttt.xK) ==‘Ls(i. Y]. Y... .2.10. then [D. . 0 V(M) forms an algclxx 0v. [Y.8... we have then . on!>. .J E a'(M). are obvious contractions..6. Let q!:‘t be a local I-parameter group of locat pans- differentiationidC’is’ a skew-derivation of degree 1. . J’:‘.C. By Lemma in the proof of Proposition 3. @ . we have. we have to prove that I.C. 1 (b) If D is a derivation: of degree k and D’ is . Using conditions (a) Let V(. For Our probtetn is therefore to prove that $. . . CoroI‘l.~ric~ covariant tensor fields of degree r. Y.$J = -(d(q. [X.]Z = [X.xjjj G’x(y*f).v and $. .] has th e same effect as ktIg. C. ) YJ) = (DK):(y1: : I.v a vector Jield X. QED.. i .. \Ye have * COROLLARY ?. . DYi. then (e) of Proposition 3.j. .A1) 1~.. By virtue of Lemma above. 111c . (cf.r of degree k and Proof. .L. Z(M) = 6:“.?).e. K(Y. 0 @. ’ . . :. .3. .try 1. then ID. for any derivation D of on A4. Ly]f = XYf... ..v..< I$-‘f = p*f.10 that t.) . By definition. . Let K be a tensor Jield of !@e (1. A derivation (resp.24)..we have result on derivatibns and skew-derivations of 9(.~l 0 L.). Proof. . l)‘t!J A b>’ for f.Kj = --(d(&Kj/dtj... .24).. W). = Dtu A t:’ -1 ( . 0’3 is a&w-derivation of degree k 7 k’. skew-derizxtion) of U(K(Y.1.7. . .[X.\\NIFOLDS 33 32 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Replacing K by p. [X.K).. a skew-derivation D-of a(. 3 “(L. L. . . yh following formulas (cf.).14) into V+“(. . r) which we c$~*((pl.L.8). it is sufficient For Z E E(M). .4.hey coincide [L*y.: I.: .. is a derivation [IdaS. we have c L. We already noted in the proof of Corollary. . of 3 (.. Z]] . It is a straiglltfixxxrcl verification to see that q. @ K). it is sufficient to show that [L.V) for every r. e . For any vectorfields A’ and Y. we obtain A-gerivation or.K. .. . skew-sv ~ltntc.&K) = ‘. .YXf= [X.a skew-derivation of &%ree k’. zjr::i on X(.§l?. J’e I.. . . YJ = C..ation. PROPOSITION 3.]. .Kj/dt). IVC obtain PROPOSITION 3. . we have I.I$ C&K . . &rpret as in Proposition 3. Z]] = [[X. #. The fact that they coincide. 1 F. ’ 9 I. W' E Ia(hf)).IxI( (Q c)l’differential forms of degree r defined and (c) of Proposition 3.[Y.11) is said to be of. Lt. i. Y.s L. .by pf for every t ifand K( Y. k’ respectivel. Y. .on s.c ’ R. The exterior PROPOSITION 3.~K)(Y..V(L. .K : 0 . With respect to tlu esf(*rior product. .@ Y.

&(j. let 21 be-. wehavvet. we consider. Observe first that L. Y. .). . zi. and (c) is straightforward. and cs. QED.. x”).ll)) c a(M) and.e:. @J = (y-‘)*o and. e g(M) (because . Y.3. that C&ie&. . To prove its existence. Since n mapsV(.) for Yi E X(M). !zd F&JND&QNS OF D&ERENTrAL GEOMETRY I. follows easily from the following formula : r-2 .relations among il.’ ivation .(M) associated with the pair (1. U’ can be locabzed. . .8.I) 30(X. such a skew-derivation is . . called PROPOSITION 3. The verification of (a). 7 D. the con- following formula : ’ traction C: Z:(M) -+ 2:-. .~tion y of M.8. .>.. 9 commutes \sith (/. ~.. hence.-. . ) &Y. ) Y. > Y.. .g. is a derivation of the interior Product with respect to X.unique if it defined in Example 3.c~~. The proof of (6) is similar to that of Lemma for Proposition 3. where Q c lo’(M) and o’ E D)“(M).j. . derivation of p(U) such that ulf = 0 for every f 6 Z(-u).Lscc. Set D’. . d(D’g) = 0.J = 0 .X. in ocder to prove D’ = 0. &St as in Lemma for ‘Pr+&ot) 3. (1. . a derivation of degree 0 of a(M) which for the sign’ of the permutation (1. Then II’ is a i$ z 0.-. Y2. . +> L.M) w&h commhtes with tht exterior di$erentiation d.. . commutesrwith Q(veJirst observe that. L*~(Q A 0~‘) = bso) A flJ’ + (.d(~s. . .‘> Y.* 0(X. DIFFERENTIABLE MANIFOLDS 33 (c) If D and D’ are . From D’f = 0 and U’(&) = $cf on Q°CM) = 3(M) and W(M).+ (ji.9. . This follows immediately from the exists.~~(/I~~.ansfi.lm.iu’.3. k) OJ(&. then DD' + DfD is a derivation of degree k + k’. Proof.ct ff t bc alocal I -pattimeter <group of local transformations “‘where the summation is taken over all possible partitions of generated by X. whose proof is the same as that of Proposition 3.Y. . By I ..) = r.1)‘~ A L~OJ’. For each vector field X. y1 fo? any vector&hfs X and Y. is ep&$o Lxfor some~vector$ld X. Let w = fdg. we’define a skew-derivation lay. for any c ~(j.i for every (U l %(nl).lb) {Lx. . 7 L. : . By (c) of Proposition 3.#‘fd~-sf fro+Gtiorl 3. (I-+)) (Y.1 of 3(&U) such de&+ 0 of %(. =.) and (k.. y. . . In other words. *. Proof.. . .X such that Since (&fi$(Y.J > .)) Consider every r-form o as an element of Zp(A4) and define L~LO = C(X 0 0).skew-derivation of degree k and k’ respectively. . cornmu& ‘with the aIternation A By (d) of Proposition 3. 1). I. .-’ Df = Xf for every JE 8(-U). Il. . .. .. . k) s&ids Conversely.l) = s(M) into itself. : * > q. it is As.(a(. . .. w is locally of the form 2. Hence. . .) = r(r . D’(dg) = 0. D is a derivation of 3(M) and there is a vector.8. . k. . ) Y.) and.oJ) and the definition of L. kx4(L * * . ..o it follows that Ley(&‘. k. YJ. (b).5.nley de&vat&y @degree 0 03” B(M) which tommutes with d (a) Lx’ f = 0 for every f c Do(M) “(. . we have s’u@Sent to pro& Q‘(r) rr (j fit t&y i&&n ~3. ) Y. L. ~&J’(Ykl. . of degree .andit is’ sufficient to show PROPO$ITIO~V 3. .-. . D’(w) = (D’f) dg + f. . . Y. . . Ls.2. . to the reader.j. YJ = X(w(Y. Q. . fi dx’ with respect to a local coordinate (d) A denbation or a skewderivutioa is completely determined by its system x1. . o’ < %(-\I). d 0 cx + tx 0 d is a d.-. . for any o.&.c i” Pi i ? ‘10 pro\~c that L. ev&y “&or fieM 2. . .. r -+ S) .(I(+.’ : (b) lays = W(X) for every cu c W(M). . = d 0 lx $ lx 0 dfor every vectorjeld that D’w = 0 when CO is of th& fOrmfdg’*berhf. for each r. . From ~. A @’ + (d A L.] = +i. .). we have The verification that 1x thus defined is a skew-derivation of B(M) iS left. r + s) into (ji.. .field . Y.’ ’ QED. .. commutes with d. we obtain Proof. (w A @J’) (Y.10. .\.i. For.q&&.

Hence.) Jf w is a l-form.9. tfwn . . . . Y) = .w) (Y) = X(w(Y)) .-. we have only .10 we shall prove PROPOSITION 3.vf = (d 0 1-v -!. If r = O. . ... QED..z E 3qAf) . to simplifi the notation. As an application of Proposition 3.. . to show that L. then (d4(&. Y. By Proposition is true for Y = 0. I’)) . By (d) of Proposition 3.([Y. ‘.. it is sufhcient to show that they have the same effect on every l-form CO. .. 2) = f{x(w(r-. X.x t X(M). Assume that the finm$a is true for r . As we noted in the proof of Proposition 3.r.V.To prove that it is actually equal to L. r Since I \-u.o)( Y) = OJ([& Y]) = /. As \ve noted in the proof of Proposition 3. set X == X. ‘.OJ([X.]. x. ly]W = L&o(Y)) .s. . ..(L.8. :....U. DIFFERENTIABLE MANIFOLDS 37 of degree 0. . CjJ([z.induction on r. .OJ([. 1:ormulas in Proposition 3...(x. 2) .c. it is qua1 to the Lie differtntiation Gith respect to some I. where the sym/bol h means that the term ii omitten. .)~jij-C.) w-) w = 47 To P rtive the second assertion (b). that [L.1 is a skew-derivation of degree -1 and that both [ CO be an r-form and.. ) xi. s. . vector-space valued forms..[x. tt. r+l Our Proposition follows immcdiatcly from these three f’orttrulas. (The cases r = 1 and 2 (21-1). 3.5. are part&la+ useful.. If 0) is a 2:form. observe first (l.. Ii .1.9. XT) 1 = .11 are valid also fbr ‘(dOJ)(x.t..L.).10. z)) -i. . .) :L.:) dW(‘YZ x.lOJ. . we have (L.j-(clJ(X)) . . x)) t z(W(x.yW) = X(0(Y)) . {L.~oJ)(X.. It commutes with d &xause dz = 0.9. y(.. z]. .. .slfJ‘: (. .J([x. (a. (. i x. Y)).‘&&i (I) is a function and dw(I\.. we have.. by induction assumption. .(-l)i+j~. or] and lt.36 FOCWDATIONS OF DIFFERENTIAL GEOMETRY I. ..l s do):.. . .* x.c .. . X2Yd(M).m (r 2 ll-form. -\. Proof. -df Proposition 3. If w E’s an r-jirnir. t. . ..Yrj (d 3 I.. ( -l)‘xi(ciJ(XO. which shows that the&rmula above l.(z. f = Xf and (d I’ 1 s i 1 dv 1 d)f = ls(r6J‘) = (Y -:. xi. x) . ) XJ) rfl 1 f. ) X. I ’ ] . then -Kmnrk.l-v 9 d)f for every function j: But this is obvious since I-.rl are zero on z(M). Y]!}.. .x. The proof is by . . By vector ficld. xi-]. sj. is .{X(OJ(Y)) .XOcjj.).11. Y]) which can be proved in the same way as Proposition 3. i (dw.

2.8.usual one.. it is the sum of at most countably GL(n.rnlations of G. 2) and S(X. can be iden&ed with the Lie algebra of O[X) and the exponential plication and In-.1.\. R.) the left (resp. Th e verificati. B] = AR .on of S(X. R) can be identifi.C (). It.‘?: SPXC T..ltcs a (global) I-paramefer group of trans. right invariant) Example 4. The mapping A . hence. We call type (1. right) translation of bracket operation defined by [A. In particular.s E R. is an open subgroup of G. Its identity com- &j. x E G. Its identity component only if the factor group G/Go consists of at most countably many consists of matrices of positive determinant. b) E G x G + ab-1 E G whose +h row andj-th column entry is a. we shall mean a subgroup H (. With respect to this differentiable It follows that G satisfies the second axiom of countability if and structure. R) of . -4s a vector space. (1.. By Proposition 3. Y) = --$(I’. a unique solution of the differential equation a. ‘l’huo CJ is . known G by . is denoted b) able. A left F. Set a.(G) at the identity. = qle by exp A.f’ . is a local l.follows that exp tA = a.j I. elements.~ = nsa l Sol. ponent. We define 111c Lie algebra CJ of G to be the set of all left so(n). tion of a. with initial condition a. at a. the l-parameter subgroup of G generated by A. The group O(n) of all n . for all .x = xa) for every x E G. that the Lie algebra of GL(n.B-4.. : l>c. Another characteriza- QED. (resp.)/Z. consisting of elements of determinant 1. R) be the X Lie group G is a group which is at the same time a differentiable group of all real n x n non-singular matrices A = (aj) (the matrix manifold such that the group operation (a. is called left invariant (resp. 38 FOUNDATIONS OF DIFPERE~JTIAL GEOMETRY I. and that a. GL(n. of the identity e.ic aIgcbra cif \ cc. the multiplication is is a differentiable mapping of G x C into G. The construction of 5 was vector ci. = e.. by Corollary 1. Let GL(n. Z?.ectc)r nt t’ ol’/i e. all n x n real matrices ‘Forms an n2-dimensional Lie algebra with M’e denote b)f I. invariant vector ficlcl on Ii is c~cterrninetl by its value at E and this * f(. a Lie group with respect . right translations orthogonal matrices is a compact Lit group.-‘at = A. The Lie algebra o(n) of all skew-svmmctr-ic n j. In& element o E G: L. Since y. A left or right invariant vector field is always differenti.=. Example 4. X) is easy.) . By a straightforward calculation. In other words. (resp. is that it is a unique curve in G such that its tangent We call S the torsion of A and B. urapping which sends X E $1 into X.A c $3 gc ncr. a. R) coincides (ad a). is equal to LotA. for A = (a:) and B = (b& GO.en by tht. = qte. denoted by (A# = C. R) + GL(n.x ~-2 us ( qi\. The set $(n. ad II is the inner automorphism of G defined by and the exponential mapping nl(n. GO is generated by any neighborhood.1. then ppta can be defined for It] < E for every a E G and is equal to Proof.Lie group G. we see that s is a L. qte is defined for Jtl < E. ..ever).to this differentiable structure. Y) = -S(Y.a bilinear mapping of the S(M)-module X(M) 2: X(M) into the is defined for ]tJ < E for every a E G.1. the connected component of the identity. sib. if p.$arametcr group Of lOCal+ tangent \.(pte) as q?t commutes with every L. vta is defined for ItI < 03 for B(M)-module X(M) . open submanifold of Rn2. X). R).tcket operation. ‘the value By a Lie subgroup of a. Denote a. ( - .vcr). scalar multi. the isomorphism equal to n(n . $1 is isomorphic mapping o(nJ -+-Q(n) is the.’ n matrices invariant vector tiehis on G with the usual addition. S is a tensor field of ‘every a E G. l‘he dimension of O(n) is \yith the tang. GL(n. R) and gl(n. (I E G.t hali invariant vector field on .< n if it is invariant by all left translations L. R) For tz c G. with the usual exponential mapping cxp A = S. as an many compact sets and satisfies the second axiom of countability.. R) can be considered as an open subset and. DIFFERENTIABLE MANIFOLDS 39 Then the mapping S: X(M) x X(-M) .tor fields X(G). it is discovered by Xijenhuis [ 11. O(n) and o(n). tl”/li ! A4 vector field X on subalgebra of dimension * (ti = dim G) which is at the same time a submanifold of G such that Hitself is of the I.t.ed with gl(n. 2). R) is a Lie group. Since G is locally given by connected.x( $f) is a tensor jield of ppe transformations generated by A and. = e. Then atfs = atas for all t.exp A of g into G is called the pjonential mapping.

ad a ofG is called the ndjoint refmsmtcrtion of G group G is a Lie tr&. because the exterior differentiation commutes (that is.1 is left im.J. . its Lie the desired group H (cf. ])*A. Co nversely. .‘ by the Lie algebra of H2. B’E. with do(A.. If B is .x and .denoted by x --+ xa where x f M. and H2 Let E. By Proposition 1. ~*-4 is again a of Maure$%rtti is given by: kft invariant vector field and T. for .U. [E.ariant.4 E 0. Assume that H has ttvo differdntiable structures. we ass&b the space of all#. Set al = . From Proposition 3.$mration group on a manifold -41 or that G ac. Hstltissjics the secondaxiom oj’courrttrbili!~~.. and F. or in other wordh.r distribution and the maximal iritegral submanifbld through P of this distribution is It follows that if H is an invariant Lie subgroup of 6’. T*B] for . p.. For every a 6 G and rl E 9. for every a E C.R. Ry providing H with for every a E G. Let H A differential form co on G is called left invariant if (L.+ect to it.the c$‘s.I= L. E. .u E g. j . 111 = [q*d.) *m = (1) be an arbitrary subgroup of a Lit group G..ts in !I.a left invariant regarded as a Lie subgroup of G possibly in many dcfferent ways form.r. Thtis there is a 1 :l correspondence between connected Lie Conversely. In particular. 109. ‘onto I%. t algebra b is an ideal of g. be a basis for g and set satisfy the second axiom of countability. Chevalley [1 . if -21 6 0.. B)‘= -&o([A. . Since the identity compotieht of Then O’.B E !I. DIFFERENTIABLE MASIFOL. group G.x = R.\ subgroup.if the following conditions are satisfied : because subgroup of G: If both W. then so is do). oiL:n ioi~o/~). H onto itself is a diffeomorphism of H.V into def = --pjl. generated by B.L)S . The vector space g * formed by all left invariant the discrete topology. ‘l% with p*.+[. denoted by H.respect Every automorphism p: of a Lie group G induces an auto. Then ~~iu) = xa.: 1 c.B E n and ql the l-parameter group of transformations of G . tion 1. to thd b&ii E. is differentiable..J. we ha\‘e (ad (1j:1 = (H.. 111 E b.3. we impose t/w cotcdition tht H/Ho. 0’ form a basis for the space of left invariant real H2 is a maximal integral submanifold of the distribution defined l-forms on G. with respect to diflkrent clifkr~ntiable structures). denoted also by ad n..i. A E 9 and ZZ E b imply [B. every subalgebra $ of Q is the Lie algebra of a unique connected Lie subgroup H of G.kej A ek. . . .. . E. B]) for LO E g* and h. so that it is a L. A. We @ of 9 is an invariant subgroup cjf G. .-.k~=l FO~TNDATIO~~ (. This also means that a subgroup IZ of G can be then the function W(A) is constant on 6’.. + Hz is differentiable by Proposi. that is. algebra g. We set . Similaryf-I: Hz --f H.-xp td = vI(e:t. sense. Let H be a subgroup of ‘a Lie ‘. oi”C is a Lie subgroup only il‘it is countable. axa-’ induces an automorphisni of 9. It f$lows that the L-le algebra 1) of H can be identified with a sfiljalgebra of 9. dual space of the Lie algebra g: if 11 c n and WE g*. we have G.L. in fact.. we have the uniqueness of Lie subgroup uniquely determined by ‘_ structure in the following.. make a few remarks about nonconnrctcd Lit subgroups. the connected Lie subgroup H generated by an ideal subgroups of G and Lie subalgebras of the Lit algebra $1. subgroup of 6’. we may regard W as a O-dimensional Lie I-forms is the. Then this is an involuti1.11 we obtain the equation of Maarer- remedy this situation. .x E C. . Ek] = X. is countable. The We now cd&d&$k.. (differentiably) on .“Consid$% the identity mapping f: H.9. transiormation groups. ad a which maps .. . . -+ Hz.. ‘.qy..) The canonical l-&rm 13 on G is the left invariant !I-calued l-form Under this condition.f: H. where HO is Cartan : the id&!>% con~~onenf oJ‘H with re.l. . O(A) 1 A for A e 9.. We say that a Lie representation n .F DIFFEREN'I'IAL GEOlh!E'IXY I I. Let (1) Every element a of G induces a iransformation of -A!. This is proved roughly as follows. ‘_:: where . are calied t&e structu<~ constants of 9 with . It can be easily verified that the cquatinn morphism p‘* of its Lie. the identity mapping of ‘8 = Cfc. ‘I’heorcm 11). a discrete topology. i= l. To each point x of G...

ab-’ is real analytic (cf. a c G. : ...b E G and x E M. = (aA4). . X(M). R. From (3) and from t. G is a real analytic manifold and the linear mapping of g into fi(i%4). R* = aB and a.u and U’ ~wpectively. vanishes on M. the projedon G -+ G/M is real analytic. which differentiable. The tial mapping is one-to-one.&> for xa and say that G acts on M on the right. x) l G x M + xa t M is a differentiable mapping.. 0 L. this implies that a. of M are not congruent modulo G (i.B l g.. . Rat is the identity transformation of M for every Note that Rob = R. The action of the l-parameter able manifold such that the group operatidn (a. -1(c) 2 XU~~‘CU.. for c c G. 0 R.2. on the right. and L. open neighborhood N of 0 in g such that exp is a’ dxeomorphism morphism.(U) n UN is empty for all ..9.x = x fir for every t. 0 exp: N. Then Rat leaves x fixed We say‘ that G acts e$ctiveb (resp. is trivial. see Chevalley [l . Pontrjagin [ 1. we say that This means that the . Let G be a Lie group and H a closed subgroup of G. (&z. let gz be the mapping considered as an open set of R” by an identification of g with R”. If G is effective on M. 0 a. then x and x’ have neighborhoods ‘. there is an m&&g a: g -+ X(M) which sends A into A* is a Lie algebra homo. We shall need later the following the bracket. 1181 or Pontrjagin [l . Proof.. If G acts on M on the right. hence A = 0. . For. If G acts freely on M. free@) on M if R. we assign to each element A l g a Although we defined a Lie group as a group which is a differenti- vector field A* on A4 as follows. t and hence A = 0.. If G. = e for every t and the fact that each R. DIFFERENTIABLE MANIFOLDS 43 FOUNDATIONS OF DIFF~~~~~AL GEOMETRY (2) (a.x # x!#or q&y S:&@). &i& that A.1.a*G. = e for every all x E M (resp. for some x E Mj implies that a = e. every x E M. QED. acts e$ectively on M. by virtue of the formula for [A.4. we obtain &k&tic: In particular.x for ax also.. By Proposition 1.near the origin of rg. There‘ ig’frtither important class of quotient spa&s. If the Lie ‘algebra X(M).exp t-4. To prove that (T commutes with group operation (a.).:The exponen- PROPOSITION 4. a(A) never p. = exp tA on A4 induces a vector field on M. it follows that R. b) .. thus proved that a is a homomorphism of the Lie algebra g into We also write.vanishes at some point x of M.(cf’. It follo\zs that a is a With respect to this atlas. we may replace differentiability by real analyticity will be denoted by A* (cf. :‘Y . (p. 6) . Let C be an abstract group acting on a topological space M 0~1 the right as a group of home&&@&ns.. -+ Ua.) a 6 G + xa c M. WC have PROPOSITION 4. If G acts freely on AM. and I~.nerality for the following reason. Consider the atlas of G which consists of charts. this implies that a. are the identity transformation of M: assume aA.. pp. that is.[ 1. bH) into abH is real Fro. Let a Lie group G act on M. then.l-parameter group of transformations Rat G acts on A4 on the left and we use the notation L.ab-’ is subgroup a. $391).“‘:s d. or I. The action of G is called prop&$ &is- continuous if it satisAee the? following conditions: (1)’ l’f t& @imts~%$@d x’. Then.(U) and N is following manner. (denoting the differential ofh mapping by the same letter) For the proof. let A.. First we observe that a can be defined also in the mapping of R. Chevalley. (Here. Suppose that aA = 0 everywhere on M. that is. B] in g in terms of ad G.Then the quotient s&e G/H admits a structur<‘of real analytic manifoold- in such a ruay that the action of’G on G/H is real ‘analytic. = L. without loss of. we write ax and assume (ab)x = a(bx) instead of (3). We have (3) x(ab) = (x+b for all a.e. (aE)*. 2571). A* = a-4. $1). To prove the last assertion of our proposition. 109-l 111. Ua means R. is one-to-one on A4. that ts. where va: Ua --f N is the inverse.42 I. the fact that R. lhc mapping c x G/H + G/H which maps (a. p. then a is an isomorphism of 9 into of N onto an open neighborhood U of e in 6. for each non-zero A c g.

thalj the distance G’ x .[~:‘. let x andy be points of G/H and (d. 6 L e t C. p. define a weaker notion of discontinuous groups. nectssar).. hns a structure of d~firentiable (resk.O@cn splleres.Y c . arc all mutually distinct). It suficcs to take IT n Ii. the isotropy grow. . I. 208 and p.. Then the quotient spnce (an the l$t) is &continuous. I. 1 -.every sequence of elements (n.l is in bI’F-1. WC may assume (by taking a subsequence if the reader.\ is closed in . it is important to show that a certain given group for c\‘cry . L e t I’ b e a component of ~--l( Vj:0ni0 1’. 6ropGti~n 4. transformntion ofX . E l’ and h. If tllf? a c t i o n o f (.of \t’e s h a l l pro\-c t h e fbllo\ving rcsuit.:lsss. We have thcrelijrc 11 - Remark. QED.\I/c is Hausdorff.. a Lie transformation group by introducing a suitable diflerentiabie x’erge to a point in M. Assuming that the action of D is not discontinuous.ti\.s. empty li)r all n c G. -\i/C.21 is called ~i. tiai geometry.. QED. For the proof of the following theorem.’ ‘1’ lb “1’ if b e pro\xd i n the same w a y . recll analytic) manifold such Proof.. &ax = xj is is always satisfied by a discontinuous action. on the left.ent space .j > K..I1 and. and let each (Q. E H such that d. QED. i.5.) that A. (2 B GrQ@t contained in G. ‘The action of 211 In applying the theory of Lie transformation groups to diffcren- abstrac. any subgroup ol‘G’ acts on the quofient space G/H satisfied. where u. ( tondition (2) . PKOPOSIIION -413. let r be a positive number such that 2r is less than the (3) l*~ach s E .Y E Ci. stable by &. each .pro[~erII~ c~is~onli~uow .II for II > N.) a sequence of distinct elements - Proof’. bet\veerl .Y e M. the or. e 1’.\I< is .-4. (jvhere of differentiable transformations of a manifold can be made into n. is cnlpty for C\:er). Then G is n Lie grou+ and the m%apping let r be a posi.yfj.‘ for finite. .dII. . Introduce a differeniiable (rcsp... real nna!ytic) mnnifld -21.j -sb A’. Since p(bV) is a neighbor- is an admissible chart (I’. i s free.. ‘This means n. wc refer the r&a&r to A%fontgomery-%ippin [ 1.x converges toy.. ::‘>!! each . Given a point x’ outside \he or. thd the jrojertion n: ..+ch that distance between x and the closed set xG . and <r. ‘I‘he verification of details is left-to Since H is compact. Let p: G + G/H be the projection neighborhood T/such that n is a homeomorphism ofeach’6onnected and write x = p(a) a n d _y = p(b) where a.. f o r each . real ana!)*ticj structure in H e n c e . Let G’ be a topological group and H (1 compact dljfirentinble (resp. s$ructure in it.ct li ?hd 1. 0 E (. Let U be a connected component neighborhood of the identity e of G s u c h t h a t b-VVV--lV-lb-’ of x-I( V):Choosing Vsufficien# small. z (a E G. there is an integer N such that d.nH = p-*(d. d.k ::I W. F o r IV/(. DIFFERENTIABLE MANIFOLDS 45 C‘ (2. real anal_ytic). b e a locnl!v con+zct eflective trtm. there exist u. Let G be a topological group and H a closed subgraup of G.d..h. ~ROPOSII-IOX 4.3 ~1 is the composite of ~1: V -+ I: each n B I\.\I...YG F group of a co&@ed mtmifold -41 of class Ck. ~‘oI.~..x c p(bV) for all n > K. \VC tion. AI. of isometrle& of 0 me&ric @uce .. = u. X complex analytic analogur of. Then the action of evey discrete subgroup II oj’C on G/H d~fferentinhle (re.b E G. v)..!.. fvhcre if : l! -+ R”.~~~t~li~~t~~~ if.&’ for 11 ‘) N.can bzl. Consequently...\f has a nci~hbor~~~.bit.. for the manifold hood ofy. hence.rC.bit . the open sphere of radius r and center x as II.a 7 bv. by taking (I’. real clna@ic‘) transform&ions acting fr<ely’ o n a subgrou/’ of G... as an admissible chart. TO prov: (3).e number s u c h that 2r i.u.~brmntion P r o o f .\ does not con. y Ezers discontinuous group c.’ be &c:. \\. Then &(I-) n K i s Dantaig and \xn cicr Waerden [ 11.. THEOREM 4 . contradicting our assump- To give us&l criteria for properly discontinuous Rl‘OtIPS. ‘I’). t h u s p r o v i n g condition ( 1 :I. t h e n cQndition (2j is autoina&lly Then G. cssentiniiy d u r t o v a n radius r ate centers . ()bserve f & t t h a t .r: and the orbit . Let c be n properly discontinuous group of of clnss C’. N. Condition (3) implies that every point of AI/G has a of L) such that d.\I ---+ . group G on a topological space . if i.Y. Condition ( 1 1implies that the quori. =.\ 1 is <of class c”.x E .d. 2121.r a n d XI rcspestively.{x}.. we may assume that’there contains no element of D other than e. converges to an element h E H a n d hence “’ h.u 4.\r of C.. the sequence (K.x) c p-‘(p(bV)) = bVH f o r n :.I/ --f .11/C is d$Jerentiable (resp.\I.d..

..7. of M.K fop every d * P r o o f . . (which we may stil1 deslpte by o). n yZ is non-empty. On the other hand.(bi) (cf... For each x E M. bi) + dhk). Lemma 1 in proving the local compactness of G (cf. ‘ < . locally compact metric space satisfies the If(%M. CE V. Thus the closure-X of U(v. p. E) = LEMMA 3.(x) converges for each x E VI.j. ~db)) . ersume again thaf cp.. be a chain with 6.(x) converges for each x E V. d&x).7. Then there is a subsequence P). - the second axiom of countability.. we may choose a subsequence since rev is an isometry. for i = l... converges for some b .V... for some d(v.(x).~ of Q).V.) is contained in U(qAv(a) .E V.& such that 9+Jbl) converges. the group multiplication for every n = 1. pp. (Such Proof. q.) exist a compact set K and an integer N such thpt cp. ~. V.(a) converges..2.( VJ c .v(a).. M b$ng co+cted.~ such that ~)~. By Z&mma 2. e/4) of a.(. localb com. although it follows from a weaker ‘assumption cp. By assumption q. we change the notation and may assume that v.G and the group’action G x M 3 M’are: continuous every x E V. Kelley I]1 .(x) is in a compact set K for all n > N. Arens [ 11).(x) convergesfor each x f M. ..O).(x). ada>> fact. From Ks. Choose N such that n > N implies”<(v. be a sequence of isometries such. In the notation if Lemma 1. E)) is compact and p. and qn . 6J i a/4. .(b) Lemma 2 allows us to choose a subsequence which converges for . ‘:.. but we Proof. let V. we choose a subsequence Q)~. LEMMA 1. For any finite number of pairs (K.x. let Vi. Since V.)) < 6.. there is an N defined as follows. The diagonal sequence qk.. using the fact that q.) is in K. %M) 5 4%(x)> %W + 4Ynv4.(b)> P.v(~)) + d(b...m > Ni. E) n > N. . 46 FOUNDATIONS OF DIFFERENTIAL GEOMETRY I. by changing the notation) such vAY(U(a. ... Then we have (cf. - topology. This means that For b. ..U.(W 2 4qn(b).(6)) < L choose one. Thus v. We recall that the ‘compact-open topology of G is a {b.. e). v. We * ass& every ljointy of M is in the last term of some chain.. %m . The continuity of the mapping G -+ G converges for each i = 1. We define a chain as a finite sequence ~14. the set coincides with ?compact closure (where d is the distance). Let q. There is an N.) of such that v.(a) con- {x 6 M. has compact closure (this E may vary--from point to point. such that d(vn(bJ. e) = ‘P. In + dk...( V.) c K for n > N.. &) =.(b). it is ‘tggr’t&~ee that the set of such paints y is open and = 4x.(bJ) < 6/4 for n. Every connected. 193). Choose which sends y into + wills ‘be proved using the assumption bi such that d(x. If x E V. converges. and 6 > 0.k( 6. Steenrod [l . This being Said~$btise a countable set (be} which is dense in M. Jet a f M and let E > 0 be such that U(a.. and so on.. .. Then there is a subsequence P). p.(x) &a Cauchy sequence.@+)‘i= U. 158]). DIFFERENTIABLE MANIFOLDS 47 THEOREM 4.(b..Then this subsequence. be a quence v. Let {b.&9) compact and satisfies the second axiom of countability.. such that (1) *each Vi is of the form V. LEMMA 2. < e) . This justifies the use of sequences Thus q”.( V.. Since’M is locally + 4sdbi). q.) ={yeG. closed. u. To prove that ~~. such that pact metric space M is locally compact with respect to the compact-open q. Let p. n > N. that p).} b e a countabld’ietwhich is dense in V. a) < e. then we have ‘of open sets V. .db.(x) The proof is divided into several lemmas. q. G satisfies = 24x. 2. In particular.} exists since M is separable. that p7.) is in K for n > N.&% 6)) q.(2) ~~r‘cont ains a. compact subsets Ki and open subsets r/. and.) By Lemma 1. e(a)) S d(~.+ such that *Denote by V= the open neighborhood ZJ(a. = U(x.~(x) converges for GxG .) converges Since M is regular and locally Cmpact. in Theorem 4. U. Proof.~(b~) the sets W of this form are taken as a base for the open sets t&G.such e for each x) .(b) converges for some point b c V. . second axiom of countability (see Appendix 2). b) + d(pP. But U(vN(@).~}.... The group G of isometries of a connected. let ‘W = M/(X. e/4) such that U(x. (3) Vi and Vi+r have a common point. Let x 4 V.h.. Choose a subsequence cpl. T I! en there used here. are isometrics. verges for some point a E M.. says that q.. converges. (The coonectedness-of Mis essentially of isometries such j that.

. Thus 9. I ysf- element $J &G. y(x)) ~2 d(v.(X)) . to 9’ on every compact sub- that vn(x) converges. Assume that Q). . such that &t(y) converges for every y 6 M.‘j’@~ .Mi and an open . y-’ C OROLLARY 4 . the iso- I exists and is obviously equal to 9. ~(a) E U} of the identity subsequence v. QED. By Lemma 5.(bJ converges for each 6. 8 .(x) x’e ..c&!?e: an . set of M. let a c hl and U an open any a E M. conclude lent to the uniform convergence of F. For each point a c K. Thus the original sequenceYn has a sub- Sequence yl.(x) converges for each’ i e M. thediagonal subsequence qk. which converges to an on every compact subset K of M..) = a yn(x) converges to y(x) for every x c M.. Thus 9~ is an isometiy.. d/4).. d(q(x). Proof. This means that the mapping a.k such that Q.. of G has compact closure..a) +b/4 <6.(a) is contained in the compact set U. U)’ = 1.. G --+ G which maps p7 into p-l is continuous. such that ~J.(x) c K for n > N. $zplete the pioofof Thcorcm 4. Y. a) .. Let 6 > 0 be given.(Vz) C K for n > N by Lemma 1.s a sequence of isometries such that v.. i. q(a) = a) oj‘ G at a is compact for ellev aEM.(x) d(y(y(xh 4 = dcp& Y~YY(4).(a)).(a)) + d(~. integer N.(x) for each x.ll% p. by Lemma 3..7.= d(g?(a). by changing the notation.. and so on. Th en or jnite number of c&&ted components. Let Q)~ b.(x). Proof. If 9. such that n > N.y) for any XJ E M.. and n >. Since y preserves distance and maps M onto M.. For To prove that G is locally compact. c converges for each 10. then Lemma 6 implies that F.. and the convergence is uniform ‘on every compact subset converges for each x e M.) converges. tropy subgroup G. Dejne q(x) = . <2d(x. d(y(x).we have W.. y(r)) = d(. neighborhood of a with compact closure..k such that qk.e a sequence of elements of G. we can prove that ~$1 that 9” + 9 with respect to the compact-open topology is equiva- is a Cauchy sequence.‘WH~ 4 converges for every x z M. maps M onto M.” Denote this diagonal subsequence by P). Thus v.:‘(X) ---f P-‘(X) for ever) LEMMA 4.’ --+ q-” in G. From ~(cD. let 4’ = p(a).. . . p(y)) = d(x. then the convergence is unz$@ for every n. there is a neighborhood W ‘= W(a. in each . T(a)) +8/4. a subsequence vn...(a). Then y preserves Since v. In I’.-’ 0 p(a).9. = {p’ E G. We now want to show that p. Let v)n be a sequence of elements in W. By Lemma 3. = U(a. + d(‘P(a).y) for any x9 z M. observe the second half of the proof for Lemma 2. The mapping 9 defined by v(x) =Z lim vn.. cF!x)) Proof. p... proving that W has compact closure. we obtain.Let vi be a sequence of isometries and 9 an isometry. 4 = I& 4cp. Since p. Then by Lemma Lemmas 3.U.. = 4?w Y(X)) = 09 Y “ r ---f q uniformly on every compact subset of M.7. there is some bi so that there exist an N and a compact set K such that p). N.. 4.J converges. is an isometry. This means that Q. Since 9 . _ kf = lJfzl Mi. closurd of U. y 1PToof. LEMMA 5.. .(a. Define a mapping v by v(y) = $2 p&‘(y). Choose a point a.~(b. we distance.(a).... As in the proof of Lemma 2. there is a subsequence Q. d(yJx).I1 is hll_y compact with re@ct to the compact-oken topolo@. implies d(v. We shall show that the it follows that y. If Al is a loml!~ compact metric space with a f any x E W. QED: it follows that y(9~(x)) = x for each x’..9 in G. q. M. (with respect to the compact-open topology). = ji% +?w. we.‘(a’) converges to a. Under the assumption of Theorem 4. that is. Let COROLLARY 4. :.(x) is an isometry of M by Lemma 4. Clearly. 48 FOUNDATIONS OF DIFFERENTIAL GEOMETRY each x E V. that is. First. Subsequence. From this.~(~. vnp --+ y in G.~. and 5. From can choose.. Decompose’ M into its connected components Mi. we may further choose a subsequence vFa. Proceeding as in JVe shall no. the group G of isometrics oj‘ .a such that 9~t.

.hm T-*(W) ( W..M by (ad (a-l))A 6 g.9.all u f 7r-!(. G acts f$‘d&~ P = M x G on the right as follows. provided with a dif- M the base space. U1.. Q E D .the bdci @ace.9.or simply P. the mapping A --+ (A*).i‘. S} induces a homomorphism gof the Lie algebra g of G into the Lie is a neighborhood of the identity of‘ G with compact closure. q(MJ F Mt for i = 1.. . Let G* = (p1 B G. G) thus obtained is called trivial.Fa each x 6 M. ff w 1s a sub- manifold of M t. generated induced by G. As G acts freely on P. 1 0 . = exp tA. q(u)) where ~. . . Since A* is induced by the l-parameter group of and an action of G on P satisfying the following conditionsi transformations Rat where a. G the structure group and v the pr?jecti@n. Conversely.1.. (10 a)i. Proof.mapping of r-l(‘U) into G satisfying v(W) = (pl(u))a for.‘ to G. The dimension G* is a subgroup of G of finite index.L. in Mi’ with compact closure. Let A* be the fundamental vector jield corre- spondiq to A‘r 9. su& i th?t bundle to the definition and the construction by means of an y(u) = (X(U). G) is a &$&al 8bre bundle.@ . . called the@8 over ~8 If u is a point of S+(X). A* = a(A) is called‘ the fundamental vector JTeld corresponding C OROLLARY 4 . G) . hood U such that kl( U) is isomorphic with q x G in the s&se In order to relate our intrinsic definition of a principal. every point x of M has a neighbor- useful in the theory of connections. g into.. G). If M is &Vnipact in addition to th& {$$$$&i of to A. u) E M x G into (x.e G. tangent to the fib+:‘& +. . DIFFERENTIABLE MANIFOLDS 51 50 FOUNDATIONS OF DIFFERENTIAL GEOMETRY We call it the Portion of P over W or the restriction of P to W and denote it by P 1W.fibre that there is a diffeomorphism ‘y : g-‘(U) . &&&-i_ =. P(M. p&cipal jbre btindle over M with group G consists of a manifold’ P Proof. I. . is differentiable .. Then Given a principal fibre bundle p( M. then W-*(X) is the set of points ua. A (difirentiable) jkld comsponding to (ad (a-‘))A B 9.that is.‘. .U x G. = MtiT Then %* is compact. . We prove 5. $. algebra I(P) of vector f$lds on P by Proposition 4. of 4. . A. . .7. and the canonical projection P: P . the vector field (R. neighborhood Vi of a. Hence. us> == (i.‘. G) we see th@..a& a E G. We call P the total spat2 or. .: i:. ab) E M x G. thtyn G is compact.1. In the proof of Co&l@ry of each fibre being equal to that of g.)*A* (1) G acts freely on P on the right.G sends each fibre into itself. Given a Lie group G and a manifold M.eac ’ h u E P. c P. A* Corollary 4. we have From local triviality of P( M.~a G. For (x. For each a E G. . (p(q) f &for i = 1. ~-r(x) is a closed submanifold of P. . 4 6 f. M = PIG.T.* is _. Every fibi’@ is diffeom’J%.). Our assertion follows (2) M is the quotient space of P by t&e equivalency r&ion from the fact that a-lep is the l-parameter group. . . For each A E 9. . is induced by the l-parameter group of transformations R.&+ by Proposition 1. . 7..rlP 2 -G :. : . The concept of fun d amental vector fields will prove to be (3) P .t’) . The P(M. G is compact. Since the action”@. a. : _.(p)‘ls a linear isomorphism of g onto the tangent space I ‘QEIIf: at u of the fibre through u. .~ and is . let U. (R. *). the action of G on P WC% . a . QED. .called the fibre through u.T&n never vanishes on’P-‘($4 f 0) by Proposition 4.A* is the fundamental vector Let M be a manifold and G a Lie group./: A principal fibre bundle will be deuo@ Jby ‘P(M. I. Fibre bundles PROPOSITION locally trivial.

ac) .iz: U.j. a) where a is induces a mapping of M’ into M.. Suppose first that the structure group G is reducible to able manifold.some c c G. A O&J+.3. x. K G for each index ~1 and let X =.&y and u = b. which willbe also denoted byf. theory of connections. = Ua x G onto-q.2. G. It is easy to see that this definition is independent of the that the structure group G is reducible to G’ if there is a reduced choice of representative (a. the transition functions of P . Since each -?& ‘is a differentiable A homomorphism f: P’(M’.~.. we say (x... n UC. QED.rition. we define an isomorphism of some t( and the identification of (a. We first show that G acts freei)rcan P f: P’( M’. :< G’. M’ = izI and the induced and only if x. Note that we’ do not require in general that the right. This generality is needed in the p-equivalence class of (a. T(~) = x(v) if and only if v = UC f&. y. with another principal fibre bundle P(M. G) is called a reduction of the structure on the right and that P/G = M. a). “Up and b = f(p).l. check that 9. u) .~. each c 4 G maps group G of P(AIF G) to G’. v:(u)) of r’--‘(U) onto U. we first note that. G’) into can construct a (~$&erentiable) grincipal Jbre bund& &A{. the G’ is a closed subgroup of G.(x)-ib E G. We say that ( TX. the definition of x is inde.ence r(v) = T(U). By definition. u -+(n’(u).‘. Let {U.\I be a manifold.M is also an imbeddingi By identifying P’ with (I. I. then. U) with’ (/% % %‘pa(“)“) is “. G’) b e a.I( U. moreover. It is %y. x. Givfn a mclpping ‘y. irr such a wq that the relations (*) 1. DIFFERENTIABLE MANIFOLDS 53 52 FOU~UATIONS O F DIFFERENTIAI.X. U = @a for some u < n’-l(Lf. a) and (B. b) d [/i’).). Proof.a8Q p%3 an isomorphism. we say that P’(M’. G).~~) is an open submanifold of P and that the m.) and u E G and we set Q)=(V) = pA(u able principal fibre bundle. > > u z +( U) is the p-equivalence@class of (a. P(M. The projection r: P -+ M\ maps.Y E ti.(u) = (x n) where anti (.@ eve9 non. x x. Ct.. _. x.} be an open covering of M such that We introduce a differentiable structure in P by requiring . The subbundle P’(M. a) and that G acts freely on P on bundle P’(M.. n Lie gro@. Every v E n-l(Q) may be repretented in the form the action of G on P is differentiable and P(M. n) 1s a dlf%rentl. Let P be the quotient space of X by mapping f: M’ .functions VI. are.E rj. In order to make P into a differenti. let P’(M. if and only if x=yrI/.Ad is the identity transformation of M.$$t each v’-L(&): (a’: the projection of P’ onto M) is provided with ~-r(. each X.P. where 6 = ~-$. b) b e representatives$s 24 ‘andl v respectively. then the induced p in X as follows.n colfering of A4 we define v~: n-l(U. Let . x. G’) is yl. let PROPOSITION 5. = &. If. G’)..(U. this equivalence relation p. by the natural mapping jl G’ and. The structure gro.P( M.s+sJie~. if n(u) = x =y = r(v) E U. as follows.). X is a differentiable 01’ injection if’f: P’ 3 P is an imbedding and if f: G’ -+ G is a manifold in a natural way.~(x) = e for every X. G’) the p-equivalence class of (cc. a) and (cI. a) c {g} x X.P(M. We first observe that the relations (*) imply &($ + e f’(u’)f “(a) for a11 u’ c P’ and a’ E G’. shall denote f’ andf” by the same letterf: Every homomorphism Xx = I-. G) and a Lie subgroup G’ of G. G)‘ is reducible to a Lie subgroup G’ ifund on& if there is an open If v = UC for some c c G.r. In fact. by definition. is equivalent to mapping f: Al . G’) .u E . if PRoPosITIoN 5. reduced bundle. b) are equivalent if a subbundle of P(M.) (n: the projection ofPonto M) onto U x G by extending made by means of differentiable mappings.up G of a princ@lJibre bundle (a. x.) ---L U x G by y. for the same’ covering {U.> are precisely the given vfi. some index. n l’. we for cvcry . X. Consider P’ as a X -+ P = X/p. em/)! 1% 1 7x n l:. x.i $ubmanjfold of2. covering (U.%(x)a. F). For u. CEOJIETRY corresponding to the covering {U. r = YC values in G’. we A homomorphism f of a principal fibre bundle P’(M’. Iff: P’ -+ P is an imbedding.-. a) into the p-equivalence class of is called a reduced bundle. Finally. -i f’: P’ -P and a homomorphismf”: G’ -+ G such thatfJ(u’&) = Proof.?. G’) . then y = X’ and’ h. Given P(A4. G) :is called tin imbedding manifold and X is a disjoint union of Xx. The corresponding transition functions take tzheir values in G’. x E 1~‘~ and 0 E G.~~. x.hebpo- f: P’ -+ P maps each fibre of P’ into a fibre of @ and hence logical sum of X. -yual) for This is possible since every point of P is contained in n: . We remark that (GC. G) consists of a mapping trcln.‘%} an i$l. a) into x.) of M with a set of transition fun’ctions yfll which take their Conversely. We introduce an equivalence relation monomorphism. Now. G’ withf(G’) and A/’ with f(M’). For the sake of simplicity. U. and y~(x)y. x. ---f C. pendent of the choice of representative (a. x. X -+ P induces a diffeomorphism of X. he t. each clement of X is a triple (x. x G is mapped 1: 1 onto 7r-r( Ua).

(v))-’ = F. Bundle .with u.. be introduced in E later and at this moment E is only a set. 0 U. We let H set on G on the right . 1101 tha+iS. I&P d crucial point is that lygd is a differentiable mapping of Ua n U. u 0 u-l ..1. It is proved in Chevalley [ 1. DIFFERENTIABLE M.) and ’ 5. %(u)) is anisomorphism of ~-l( U. $4. can be repre- sented: in the ‘form.. The standard fibre F.. The projection n.a.r G which G acts on the left: . a-it) d P x F. We shall so that any automorphism of Fe can be ~e$&&&d in the form construct a fibre bundle E(M.u E .then czbtain a differentiable tipping F x F -+ M which maps (u.+We call ‘E or more wan = pa (p. p. G into ua.urith P.. denote by ut th.) --+ U.f’ ua for some*c2. onto . 6) + ( . w U X G induces an &.lr . structure of G’. n UB) and that the mapping considered as mappings of F . G) b e ap rincipak fibre bundle and F a manifold on form. the requirement that rril( U) is It is easy to see that qa(u) is independent of the choice ofrepresen- an open submanifold of E which is diffeomorphic with U x F tation v = ua.G. G(G/H.3 . assume that there is a covering {U. b-15) for(x. The PROPOSITION 5. we imbed P’ into P as follows. r(e) in G/H in$. y E M. set of transition functions vPa all taking values in a Lie subgroup .2.) c G’. we see that the action of G on identity of G. In f: P’ ---f P thus defined by { fa} is an injection. we let G act on group of automorphisms of F. . of E onto M. t) E P x F 4~ the natural projection P x F -+ E. pp. a. which is associated &th the &inci#u~ 1:. ‘and F. 6) into q(u) induces a . It follows that the isomorphism n-‘(U)..?of M existence. G.0 a 0 u-l where u is an arbitrarily fix& Point.LDS 54 FOUNDATIONS OF DIFFERENTIAL GEOMETRY : 1 55 i . We can therefore introduce a ’ .. n U. ab. H) : Let G be a Lie group and H a action is denoted by E = P x Q F.$$& that ?r 0 (I is the identity transformation ‘ of the neighborhood. -‘(U. the. G’ with respect to. p.. Let (ua)E=u(aS) forueP.). For each x e MI -with $rt+cture group H. we can construct ‘a principal fibre bundle P’.(M. an automorphism of the fibre FS $“a . X. QED.&to F. local t&&&y follows from the theset nz’(x) is called the fihre of E over X. fhe lej?.(a. i .2.. P) be the j&e. from Proposition 1.. = ri’(x).(M). cf. = fs on n’-i( U.) onto U. This follows bundle associated. We see then that v -+ (r(v). 5) c P x F into (uu. # 4. Every a 4 H maps u f. Every pomt x. By Proposition each u 6 P is a mapping of F dnto follows. -28-331.) of “44 with a G.4. (x. For each u E P and each F E I*‘... Into. We . is hence isomorphic with tfie. A linear f&e u. projection of G onto G/H and e is the Identifying m-l(U) with U x G.. Example 5. x G’ ---f U. G’) from Finally..of -linear frames: Let M be a manifold of :dim+ion n.G..CrF.” By an isomorphism of a fib& @. ihanifold on which G Bets on. <.is 1 le mapping of E onto M.f jbre bundle ‘P. The .xF a n d LEG. x G.a diffeomorphism which. ‘% differentiable stnid&e in E by. of the tangent space 7’. G.‘$& also &eenrod [ 1. we mean .c. Let E(A& F. P) associated with P with u. 1.(U) (~C(JJ))-’ take theirval.&e. E) B G x F 4 a$ E F..ues in G’.another d-1( U. ” ‘: “. the right as follows: an element a 6 G maps (u. 9) the&e bundle over the base M. . e has a neighborhood U such that T+(U) is isomorphic to U x G. called the projection.mapping of the Let P(M. The quotient space of P x F by this group &vample 5.+@r. . of a. Let P( M. note that a Lie subgroup satisfies the image of (II.the. = ail(x) where x = n(u. On the product manifold P k F. . *: .then@here is a mapping ‘a of a neighborhood of n-i(U) x Fontherightisgivenby ’ . &!). For U. G)‘ be a prjp$ai~brc b&d.e)eU x.& d . D Q q-1. differentiable.: structure group G.-prhqipal fibre bundle G(G/H. F. yti is a differentiable mapping of - U. A differentiable structure will closed subgroup of G. _ * ’ I. fibre a point x E M is an ordered basis. Let L(M) be the set of isomorphism 7ri’( U) = U x F.r( U) .) be the composite of the following three The proof is trivial and is left to the reader. ovef the base manifold G/H mqphg Q.@. respectively. The corresponding transition functions hism n. where $ c +(i) ahd u B TF-!(~ are It is easy to see thatf. G’ of G. local cross section. with (standard) I J&e F aqd (structure) qfufi G. into a-Lie group G such that ~~cl(U~ n 17. of n-l(~). x E M.. particular. Then second axiom of countability by definition. x G T T+( U.a. Conversely.’ ..-l(x): En this case.

: . . follows.. . as a linear transformation of R” which maps ej into Ci !Je. . Tenjoy bundles: Let T: be the tensor space of type admits 4 CSO. F. is the origin of G/H. G/H. ua: R” --+ T. where f f G and &.(M) is the composite of the following two mappings: . We can make L(M) into a Coniequentb. set of -M such. let I’(. G/H.4. It is easy to see that the fibre of c(M) over Xc M r(f(u)a) =‘.... Steenrod[l. the linear frame (Y.. .is a linear frame at X.4...12. Then QED. X.O. c M can be. . . nate system in z-‘(U). be accordingly interpreted as follows. In a nattap .) with Xi = C.I?S 57 . G) . e.. with structure group H. H): This follows from local triviality c. P) and G(G}H. then ua is. Consider (I = (a$ E GL(n. Fg group Proof. 4 (t$. A linear frame u = (Xi. let PROPORTION 5. . The proof is straightforward. = (l.. G) is reducible to T(M) over x E M may be considered as T. n...ll fibre L(M) onto . the quotient space G/H on the left. . act? on linear group cL(n. R) acts freely”bn L(M) quotient space of:P by this action of H. Let e. The general bundle and Ha closed subgroup of G’.. R)) is a principal fibre bundle. We call L(M) the bundle of E .u(f(ti>). Now in order to introduce a differehtiable structure in -L(&).) at of G/H such that p-‘(V) & V x H.I ) . GJH.(M) of type (Y. This‘k)leans that . 011 the other hand. correspondence with U x GL(n. we have The tangent bundle T(M) over M is the bundle associated with L(M) . where u is considered as a mapping of the standardjbre G/F into tijibre of GL(n. For P(M.l \\-a~-. wtth standard Jibre G/H can be ident$ed with P/H as fokows.Gt(n.l. bundle M x G (cf. Gj bt a princip.‘. all linear frames u at all points of M and let v be the mapping of’ Returning to the general case. DIFFEKENTIhHLE M. It is clear that GL(n. H) as follows.‘ R” onto T. . . . u(x) =. of Q(M. coset H. . (. p. . where of P/H iepresentkd by ua c P. a linear frame u Proof. then v =~~t&‘for some a c H and hence . . except the local triviality at x. 361)..(M).~reducible to a cl&ed subgroup H and let GL(n : R) can be regarded as a group of linear tr&@rmations. Tangent bundle: Let . Let P/H be the by Yi = Cj u~X. s). P) bc the ’ a f (0. $ame’u at x into x. that nil(U) = U x G/H and let V be an open set O).be the natural basis for R”: e.) E P k G/H is mapped into the element in the form u = (X.ight..u( f(v)) = with L(M). This shows that r-l(U) is in 1: 1 the.) at x defined being a subgroup of G. . . With this stanaardfibre q. .P) Example 5. Let U be an open . R).. a closed su&@&@*@ifaud on& gthc associated bundle E(M. H) ii a principal&e bundle over the base E = P/H differentiable manifold by taking (xj) and (Xq) as a local coordi.ZI.u: P -+ E = P/H is the projection.p(f(‘u)) Ii‘ . The strwture group G of p(M. X. The two definitions are related to each other as E(M. . . The action of GL(n. G. Y. Every frame u at x E 11 can be expressed uni@rely element of E represented by (u. R) on L(hf) can sponds to U x Y under the identification nil(U) M U x G/H.(M). . Sup~~t~. . . It can be easily shown that the fibre of PROPOSITION 5. P) is a mapping (r: M . R) acts on L(&:on the. An hood U in M. 56 FOUNDATIONS OF DIFFERENTIAL GEOMETRY I. where p: G -+ G/H is the x can be given as a linear mapping u: R” -+ T. H acts on P on the. Let E(M. L I... defined as a non-singular linear mapping of of the bundle P(E. P(i. X$(a/a~~). @) be a Tedgad: bundle with injection f: Q --f P. associated bundle with standard fibre G/H. fibre of Q and induces a mapping u: M -+ E. . X.6. A cross section of a bundle E( M.E R” & R” A T. The projection P ---c E maps u e P into u$ B E. . . i:e. s) over the vector space R” as defined in $2. (Xf) is a non-singular matrix.B of is constant on each may be considered as the tensor space’ijvef T. . then p-‘(W) m #V x H. we p: P --f E = P/H be the projection. s) o&r M+dCch is associated 6bre of Q.-R) act on Rn as above. More generally. e. . by definition. 0..U which maps a linear. In view of Proposition 5. :. .) E GL(n. If u and v are in the same obtain the tensor bundle q(M) of gpe (Y. right as follows. a cross sectiona: M + ‘Petits if and only if P is the trivial Example 5.).. by Proposition 2. G. R) and u = (X1. R). 1). a&.. G. .ANIFOI.3. .5: The bundle E = P xG (G/H) associated with P (xl.E = P/H. Then we have and z(u) = n(u) if and only if v = ua for some u c GL(n. R) If .Let the space T. such that VT E 0 u is the identity trdnsformation of M. with standard fibre’ R”.- (Y. Let W be the open set of ril(U) c E which corre- uei=Xifori=. If . x”) be a local coordinate system in a coordinate ne@rbor. linear frames Over A$.(M). itself. G.(M) such that projection.S+~$Q~ a: M -. It is now easy to verify that L(M)(M.

G. n-dimensional manifold M. F.) by an.r. By the very definition of a paracompact space. More generally. then 1 is cleariy differentiable possible by the property possessed by Ui. (cc Appendix 8). For each subset Jof the index to verify that Q is a closed submanifold of P and that Q is a Pi. neighborhood U such that vil( U) is trivial: rir{U) w U x F.:~ u on A n S Remark. has the only when u = ua for some a. Iff is the restriction of a differentiable mapping ‘of an ui = T on S. (see Apb 8).d@t product of any of its maximal com@act subgr$ps H and a’ I section of E dejned on a closed subset contained in U can be extended to. to the orthogonal refinement {Vi} such that Pr 1”” ‘U. A mappingfof a subset (T’. G. argument similar to Chevalley .E = P/H. $361. k“&d extend t& . By Lemma 1. we see that is normal. P) be <a jibre bundle ‘such that . which is open set W containing A into M’.) n Pi contained in open neighborhood U. Assume J # I A if for each point x c A. By the same reasoning as 1 Proof. and 7’ = 7i on a cross section on A we mean a differentiable mapping u of A into pi. i I . a E P ic with R”‘.5 H. It is clear that 0 is a section of E.a cross G(n) is known to be diffeomorphic with a Euclidean space of section dejned on M.of m + . c.the The proof given here was taken from Godement Cl.7). a section on U can be identified such that P(U) = a(*($)> In other words.7. WP( n> . defined on aad$~~t~et~?f It” can LEMMA / bundle over a paraconipact manifold M with group awhich is a be extended to a dijerentiable function on R”.J. J). J”) if J’ C J” and T’ = 7” on SJ.. j. Let A be a closed subset (possibly empty) !$f ‘M.[ 11). 59 given a cross section 0.‘:.E. Let {V. Euclidean space (cf. ! Proof. The. for all . Example 5. let Q be the set of points. Let ‘(Ui}i.. and 7 ~&se$i’~ of E defined on S. Q is the inverse Image efined on U. of a(M) by the projection p: P . ‘Given a fibre bundle E(A4.: M . QED. a E H. s) < (T”.~t@of L(M) can be reduced. G).6.. base manifold M is paracompact and the jbre F i. On the closed’ set (A u S. restriction’ofa to A A Pi to a section on Pi. there is a differentiable mapping f. s&J+& r@&mpty . Iwasawa .point of M has a neighborhood U such that every’ . In the particular case where A is emply. n rd. DIFFERENTIABLE MANIFOLDS .another manifold is called dz$erentiable on a. U. Given a point of M. Let J’ = J u {i} and on A. This follows from the. J> < (T’. which contradicts the maximality of E such that nE 0 u is the identity transformation of A. P) and a subset A of M.. E = P/H and the submanifolds Q is 1 :l. J) where J c I ’ principal fibre bundle Q(M.connectec& Lie group-It. by 7’ be the section on Sp defined by 7’ = 7 on S. The .manifold.‘: (T. R) / Then every cross section CT: A --f E defined on A can be extended @. Extend Ui to a section Ti on pi.7. L I. 15 I]. provided that M is paracompact.- ‘i L EMMA 2. / I . Conversely.ructure’. let P(M.I ‘It is now easy ~5 Vi for all i 6 1. we have a well defined section ui: ui = u on A n Pi and U.} be an open refinement of p(u) = p(u) if and . (7. QED. there is u c Q such that T(U) = x because p-l(+)) is non-empty. k ua’ for some. it suffices to take a coordinate above. fi. there exists a dim~ion &r(n + 1.(~. heorem 5. Ever. admits a cross section if M open covering of M has a locally finite open refinemen. associated with L(M). if u E Q then’v c & ~g& and a locally finitk open covering of M such that each U.‘ [ I .DIiFERENTIAL GEOMETRY where x = n(f(a)). Let (T. Hence. T h e f i b r e b u n d l e E = L(M)/O(n) with fibre cross section of E defined on M. ‘: ‘. by known that G is diffeo&&ic with a . of an and let t l I . the structure group G’qn be reduced to H. J) be A of a manifold M into. Introduce an order in T as follows: defined on a subset of the base. G) be a principal fibre 1.s di@eomorphic with a Example 5.i. su’ch’$h&t . The correspondence between the sections B: M .6. be Given u and v in the same fibre’of P.maximal element (by using Zorn’s Lemma). 1 = J and T is the desired section. For every x c M. group O(n). which is’fissible by r We shall now consider the question of extending a cross section the property possessed by Vi. Let E(M.A dxerentiable function. Let L(M) be the bundle of linear frames over an Euclidean s/&e R”. By Proposition 5. fact that property stated in Lemma 2. every locally finite open covering {Vi} of %+ +r open the st.homogeneous space GL(n . of x in M into M’ such that fz y f on Vi. H) imbedded in P(M. n A.~Theorem 5. :every GW. 58 FOUNDATIONS OF .. J’).only if ?.5.< Eet T be the set. Then ‘(7. T HEOREM 5. p. take’ Ui which me!ets.t:‘:~~e’~ is-paracorn~.

. E and . pp. C) with a homomorphism a jbre bundle _over M with jibre r.) which araorthonormal with respect to g. Then th e covering mantfold E corresponding to H is to an isomorphism) principaljbre bundle Q( N.(M)/H and is called a regular^covering mallifold of M.8. The PROPOSITION 5. Let L(M) be the bundle of linear frames over a with group G and f’ : Q’ i’P a homomorphism which induces manifold A4 cf dimension n. . 0. Y) is independent by p. I. 61 60 FOUNDATIONS OF DIFFERENTIAL GEOMETRY by nq(y. element.7. O(n)) b e a reduced subbundle of L(M). is an isomorphism of the bundle of O(n). We shall show ‘that each reduction of the structure Q’ onto Q which induces the identity transformation of N and group GL(n. (1) G’zven a connected ’ manifold M. u’ E Q’. e. . (unique) structure of manifold such that p is differentiable. .(M).(M) automorphism of G. where nl(M) is the jrst M admits h” Riemann’ian metric. tions acting freely on E.‘. . notion@ induced bundle. u)a = (y. O).(M) logical space M.E’ E R”. there is a M with structure group O(n).product g in T. covering manifolds. unique up associate E = i@IH. Conversely. The bundle Q(N. then E is a regular covering manifold of It is easy to see that G acts freely on Q and&at Q is a principal fibre bundle over N with group G and -iYitb: Projection x0 given I : . 89-971.The result here will be denoted by A?. . X. then has a. A covering space p: E + M is a universal covering space u = (X1. If we if E is simply connected. ua) for k&u) E Q and a E G. unique (unique up to an isomorphism) universal covering mantfold. every covering space regard u E L(M) as a linear isomorphism of R” onto T. ) by. 6’) = g(uf.manz@d with group n. If M is a manifold. u’ . Let Q’ be another principal fibre bundle over $ Example 5. . a Earrespondence is given as follows. It is easy to verify that Q forms a reduqd subbundle of L(M) over PROPOSITION 5. uf’) for all 5 . . locally arcwise.the subset Q M/H is properly discontinuous. To each subgroup H of rl(M).is. a connected space E is called a covering. each u E Q defines an inner. Y) = (u-lx. it is sometimes denoted bYf-lP. R) to O(n) gives rise t&a Riemannian metric g on M. 67-7 l] or HU [ 1. From now on we shall only consider.(M) used later. . u-‘Y) for X.. U such. which An..h&w-‘W)). . Let &CM. QED. see Steenrod [ 1. of O(M) is an orthonormal frame. .-I . cotider . u): -+ (y. The group G manifold and G is a properly discontinuous group of transforma- acts on Q by (y.‘: . Then it is easy to show that the mapping of Q’ onto Q defined orthonormal and which is invariam by O(n) by the very definition by. Y E T. Two covering spaces p: E .. rf H is a normal subgroup of n. we:prove* (3) The isomorphism classes of the core&g spaces over M are in a 1: 1 correspondence with the conjugate classes of the subgroups of nl(M). -The action of nr(M)/H on a regular covering manifold E = Proof. n-rapped homeomorphically onto 0’ The invariance of ( . we mapping f of a manr$old N into M. connected topo- where x = T(U).(M) and projection p: M --f M. pr&ipal jbre bundle P(M. U) =y. ) be the natural inner product f: N + M and which corresponds to.. the identity automorphism of in R” for which e. .p. u) E N x P such thatf(y) -4 w(u).9.. DIFFERENTIABLE MANIFOLLS . When r\ e We recall here some results on covering spaces which-will be regard each II E L(M) as a linear isomorphism of R” onto T. To introduce the. Let ( . Given a connected. G) and. pp. = (1.(M) /H associated with the principal f: Q ---f P which induces3 N + M and 1 which corresponds to the {dent@ jibn bt+e M(M. M with projectiqnp: E -+ M if every point x of M has a connected g(X. if M is given a Riemannian isomorphic if there exists a homeomorphism f: E -+ E’ such that metric g. space over by . O(M). G) is called the bundle induced by ffrm:&M. which corresponds to the identity automorphism of G.(M)). Riemannian manifold is ‘a metric space and hence paracompact. there is a unique (of course.(k!). = (0. r.We shall see later that every homotopy group of M. or simply the induced bundle. E = Ml-H is a principaljbre bundle with group rr. (2) The universal covering manifold M is a principaljbre bundle over hl combined with Example 5:5 implies that every par&ompact. let Q be the subset of L(M) consisting of linear frames p’ 0 f . Conversely.‘O.that each connected component of p-‘:(U) is open in. The bundle Q will be called the bundle of erthonormal frames over M and will be denoted by. open neighborhood. In the direct product N x P. Given ‘a. O(n) implies that g(X.M and p’ : E’ --f M are of the choice of u E Q. 1) are G. u belongs to Q if and only if (5. if E is a connected consisting of (y. G) For the proof.

. .. El. for each q z P.) ‘$. &k&t&) compowt of X:&&$‘i denote 4 by tiX (resp.on ‘4 . :’ (a) 7’. :I’. subspace Q. : II the fibre through u. by &&&ion.(P) consisting of vectors tang&t to I. .’ we$all Y (resp. e.a: !E’ . Condition “(c) means.&!a ?liitrentiable vector field on P so are v$&nd A& ‘.. Theory of Come&ions EXa??lple 5.. and let [r.riri~ c&&e identity transformation of S” and the I transformation oE..8. ..Q.t(#. we showed that every A c g induces a vector field A* . . Q. The quotient G is called the n-dimensional reel 1:.(P) is called verheal +kp. Let G be the subgroup.G be the group consis.. called the fundamental vector field corresponding to A.. :: _ I i -xn+l). . .. . Given a connect@r~ I’ in P.” . I. where &. of R” generated by. . --. . M 1= E/G as follows immediately from the condition (3) in the ~de$&ion of properly discontinuous action in $4. For each Xt L(P). m(&) to be the unique A l g such that 63 . (direct sum).$J& ent manifold R/G is called an n-dimensional w.prop$y.&&.. Let P(M. of T. I. . x”+l) into.. . pro]-eetitpis:spacs. In $5 of Chap& I. . Consider R” as an n+iimensional vector space .&& the.(P): A vector Xc T.. we define.n universal covering manifold.(P) be the tangent space of P ‘ 1 at u and G. (b) ‘Qua = (R. for e&y ‘ic. the subspace of T. . m. G) be a principal fibre bundle over ‘a m&fold & i ’ * with. We call G. a l-form w on >‘wi& :! values in the Lie algebra g of G as follows. .(P) can be uniquely written as i. For each u l P.). ..b. be any. A connection I’ in P is an assignmenr &. Comections in a $‘ncipal jibre bundle ‘# . let r. By (a). is clifferentiable. 1: .). depends differenti&Jl~:..I ik ‘t... is invariant hy G.on P. . (It can be e&lj++ifkd that this is equivalent to say+. .. of 9/G. Let.he transformation of P in&&&@by a 6 G. ‘. basis of:R?.G = {C V. every’$ctor’~Xe T.’ : distribution u + Q.* Condition (b) means thtit $e &stributioq’u ---c Q.@ S” b$ n R”+l with center at the origin: S’. *. i . we define.. nX). .&ample 5. (x’)a = 1). .I’:.I.71 j. 6.(P) to each u c P such that ‘l ..S” whkh’ maps (xl. integers). It. (-x1. and that A + (A*). (j‘-&L&~\ FOUNDATIONS OF DIEFERENTIAL GEOMETRY c . that if X.~tinuo~ and R? isthe universal CHAPTER II covering nianifi$@$ R”/G+.lP and a Q G. horizontal)‘ ifit k&n G. + Q.(P) = G.. Then SC. (c) Q. *. Th+ action of G on R%. is a linear isomorphism~:$$ onto G. 2) the vertical (resp. the vertical sub&ce an4 i&~e~$orizontal sub@uet of & $J4 = ua. a ” . . . (resp.

G is clear from the invariance of the horizontal sub- Proof. to(X) vanish: In the case when X is vertical. we may respectively.?!I= A Uj’ --c: c. . since if u’ = ua..X) = ad (a-‘)A = ad (a’l){w. Conversely. * Xz) = r(X.a g-valued I-form w on P satisfying conditions (a’) zontal component of Y and hence is differentiable.*. f * . Y*]) = [X.connection is We shall’no&ex$-&s a connection form w on P by a family.where x = r(u).X* is the horizontal lzft offX where by Proposition 5. If X is horizontal. .3. Y*] is the horizontal l.a form’ Ed satisfying (al) and (b’).. Let w be the connection form of a connection.vector field. Thus. w(X) = O}. Let {UJ be an open eering of A4 with a family of isomorphisms The horizontal ltft (or simply. is equal to the vertical component o? X. l?is clear that there is a unique horizontal ltft X* of X. For eiiery x z M.... the diMbution u + Q. isomor&ically onto /) :’ forms. w(X) = 0 if a&l only if X is horizontal. . Then X:. . neighborhood-El in M. given.*oj). Then.1 of Chapter I. .(X) = w. further assume that X is a fundamental vector field A*. by G is the lift of a vector-field X on M: PROPOSITION 1.horizontal and wa: n-l(U. Then (1) X* + Y* is the horizontal l$t of X + Y.((R. for every u f P.‘gr%en. .). Yl. we first obtain a differentiable representation of G. defines a connection whose con.) = r(R.2. qED. The mally. The connection form w of a connection satisfies the Proof. it is sufficient to verify (b’) in the of the choice of u such that V(U) = x. Y*]) = r([X*.The vector X. The invariance and (b’). we define have Q.)*X) = ad (a-‘) * o(X) neighborhood Uaf any given point x of M such that n-l(U) R+ for eve9 a E G and every vectorjeld X on P.Xi = a/ax’in Ufor each i. rr(h[X*.(P).(lz4) for each u E P. The verification that u + Q. Y]. that is. so is (R. Let x1.~.. given . in r-l(U). :A (3) The horizontal component of [X*.1.a point u E P such that every vector field of P can be decomposed into a horizontal vector r(u) = x and define X. form The projection x: P -+ M induces a linear mapping n: ‘Tu (P) a local basis for. QED. . both w((R. -+ P be the which projects onto X. Q. every horizontal vectorjeld X* on P invariant connection form of the given gonnection I’. let X* be a horizontal vector field on P condition (a’) follows immediately from the definition of w. vector field Y on n-l(U) such that VY = X.. w((R. The first two assertions are trivial: As for the third. prove that X* is differentiable if X is differentiable. be a local coordinate system in a coordinate nection form is w is easy and is left to the reader.&‘oint U x G. The form w is called the for every a E G. The lrft X* is invariant by R. Then X* is the hori- Conversel_y. rr(X:) =: X&.(M). .(X)). (A*). x.) --f ‘U. condition (b) for a ‘connection. we take a (b’) (R. .(-14). each deSked in an open subset of the base manifold M.. Let Xf be the horizontal lift in n-l(V) of the vector fieid. X+. onto T. The existence and uniqueness of X* is clear from the following conditions : fact that n gives a linear isomorphism of.).j *X is the fundamental vector field corresponding to ad (a-‘) A (2) For every function f on M. Using this isomorphism. When a . QED. let u. THEORY OF CONNECTIONS 65 64 FOUNDATIONS OF DIFFERENTIAL GEOMETRY PROPOSITION 1. . n maps the horizontal‘ subspace Q.)*X) and PROPOSITION 1. Given a connectrIon in P and a vectorjeld X on M. Since invariant by G.: U.G and the corresponding family of transition functions yM: . is independent field and a vertical. For each a. where ad denotes the .ft of (R. = (Xc T.horizontal lift of X and Y ad (d-l) .in g. 6 . ~T. T. take . we Conversely.*).. that is. . Thus we have f * is the function ‘en P dejked by f * = f 0 n’. = n(X:). !R. #‘. . To (a’) &(A*) =‘A for every A z g. Proof. of . lx.. II. then frlliowing two special cases: (1) X is horizontal and (Z&X is r(X.X for every a E C by the lift of the vector field X.)*w = ad (a-‘)o. there is a unique connection I’ in P whose connection$n-m is w. Let X* and Y* be the. the lift) of a vector field X on ii4 is a unique vector field X* on P which is . It is obvious that X* is then the vertical.

-f 2. mapping P x G -+ P. : can be alwcgrs extended to a di&entiablefunction on R” {cf. . of T.) each dejined on U.c . -i. closed.. then u* (r*X). a subspace Q.(x)y~(X)) 7 A: T The converse ca’n. U(X) = (x.(x) means the image of (u. us. e). G) be a principal Jibre bundle and A a ‘I. Let P(M.(X) being considered as a Furthermore.).U x G. with m(u) z A. writtez uniquely in &e . x Q U.(X).. . . 3 .(G).4.. - &OS section on U.) from ! the left ‘irivariant vector field on. . point of M. Given a. g-valued l-form p*o on U.4 (Lelbniz’s formula) f Proof.(x).. 2. sy$e... ’ Ingeed. The proof is a replica of that of Theorem 5. we have’ neighborhood U such that 7~-l( U) is triviali T-~(U) . define . if CT: U -+ v x G is the natural cross section. under the connection dejined bver U. In particular. conditions (a) and (b) for connection (see $1) are satisfied and Q.7 in Chap- and satisfying the preceding condiiions.1.. Taking the is. 66 FOUNDATIONS OF DIFFERENTIAL GEOMETRY II.-. Existence and extension of connections is the identity of C..8.‘Z I ‘t.. depends differentiably on u in the following sense. Appendix 3). .(x) = y. every vector X E T.subset of M (A ma> be empty).. For each non-empty U.&’ u E +(A) is the given space Q.ofyti(X) by the differential of d.. define a g-valued l-form w. n U. for every farnib of &valued l$om {k.. By F. of the fundamental vector field A* at W(X) = ~olxsg*ij~ s w(zfr= (fJ*W)(7r*X) + A.(.(P). bt verified.’ . Hknce %. .* Y. admits a connection . n 4 by ‘I.iGy $$ch is where Y.(P) is given in such a way that 8.. there exist an ogen neighborhood U and a connection in. n U.(P) -I. ( x Q A..: . e) for x z U.. THEORY OF CONNECTIONS 67 . on U. the vector LEMMA 2.. Proof. A.&vi h u. &quaGty.(x)~~(x).. we obtain. n U. hence w(~. every connec- 9 = ad (~2’) W. Proof. if A c.*(o) = ad (a-*)o. same letters.@~?y point of M has a neighborhood U such that every qdX) 6 T. such that the horizontal subspace at every 0. . > * . a connection form w is completely determined by its behavior at the points of U x {e} (e: the where c~JX)y. it suffices to take a coordinate of Chap&‘I. .=&x . QED. is the +&ye. G) be a principal fibre bundle and A a subset of M. Denote -the dif%e&titials df uo. P ConverseZy. y@(X)) E connection dejFntd b a.can. by followi%~ k&k the pr?cewd where A is a uniquieI e&ment of g such that the corresponding rpbtaining {o. 41X. dz&entiable jkzction defined on a closed s&set of Ik’ x E’ U. :. and wa are s@ject ?o the conditio&i THEOREM 2. tion dejined over A can be e&ended to a connection in P. --.f M is paracompact. w ereh u = u@(x). defined by u. then w is completely determined by the’ of o on both sides of the.!& . where e .(X)) and ~7Jx)y~(X) is the image identity of G) because ‘of the property R.(x). where u’ z u$) and a = y:!(x).T. For every point For each a. f+(x) =’ uJx)y&x) f& all LEMMA 1.(a.form &gi) . u = u&)y~(x) and. over A if. The farms 0.LGt P(M.. Since A depends c ‘. ‘U x @j ‘ahd 2 is vertical so that Y = equal to Y&X) at a = y&x) so that: e(y.on P which gives rise to {oc} in t& described & defined. If lJ. n U. Then for every vector Xc T..( U. Let 6 be the (left inv@@&lued) canonical l-form’ on G defined in $4 of Chapter T”(p. is non-empty. t&-e $ a unique connection form w “.. i. Iiideed. ’ PROPOSITION 1. and yap by the .‘(~.&valued l-form ” We say that a conk. ‘) .@) 7 A.ter I. at every point u z P fb on u. that ping of G into P which maps b E G into u=(x) b. ‘)‘f : I..( U x G) . Dn the trivial bundle U x G. fundamental vectdt field A* is equ&to 2 at u(x)...vtion 1. 0~ I$@ U. . closed subjet contained in U can be ekmded to a T. $ tang& to.= ad(y&)-l)w. If M is para+compact. g. be. ” PIU=@(U) .lX) + e.

. The notion of lift of a curve corresponds to the notion of lift of a
only on 2, not on the connection, o is completely determined by
U*O. The equation above shows that, conversely, every g- vector field, Inteed, if X* is the lift ofh vector field Xon 111, then
the integral curve of X * through a point 1~” e P is a lift of the
valued l-form on U determines uniquely a connection form on
integral curve of X through the point .vO = X(Q) E M. N’e now
U x G. Thus Lemma 2 is reduced to the extension problem for
prove ’ i
b-valued l-forms on U. If {A,} is a basis for g, then w = 2 t,J~,~,
where each wj is a usual l-form. Thus it is sufficient to coiisidrl PROPOSITION 3.1. Let. 7 * x,, 0 $ t ;I 1, be a curlIe d class Cl
the extension problem of usual,l-forms or~~U..Let Xl, . . . , ,x” bc a in M. For an arbitray point gb of ,#? &it,+ +,) = x0, fhpre exists c!
local coordinate system in U.: Then“&ery l-form on d is of the unique,kift R* = ut Of 7 which starts fro2 uO,
form Z fi dxi where each fi is a”function on U. Thus otii problem Proof. _ By 10~4 triviality af,the*biindle, there is a curie c1 of
is reduced to the extension problem of functipns on U. Lemma 2 class C’ in P such that v,, = u, and v$u() = xt for 0; 5 t --I 1 . A
now follows from Lepma 1. I lift of T, if it &i&s, must’ be of the..form U( = c,a,, where n, is a
By means of Lemma 2, Theorem 2.1 can be proved exactly in ,curye in the strticture grotip G Such that n, = e. We shall now look
tlie same way as Theorem 5.7 of Chapter I. Let {I/i}ic, be. a for a curve a, ‘in G.irhich makes’u, = v,a, ;L horizontal curve. Just
locally finite open covering of M such that each Ui has. the a? :b the proof of Proposition 114, \ve apply Leibniz’s formula
property stated in Lemma 2. Let {V<} be an open refinement of (Proposition’1.4 of Chapter I) to the mapping P x G --•f P which
(Ui) such that Pi c Ui. For each subset J of 1, set S, = iCJ,. U vi. maps (u, a) into, ua and obtain
Let The the set of pairs (7, J) where J c I and T is a connection ut, = utat i ofa,,
d&n&over ‘SJ which coincides with the given connection over where each dotted italic letter denotes ,the tangent Yector at that
A n Si. Inttioduce afi order in T as bllows: (7', J') < (T", J") -point (‘e.g., ti, is the vec’tor the curve ,* = u1 at the
if J’ c J” and 7’ = 7” on S,,. Let (T, J) be a maximal element of pint u,). Let w be the connection form of I’. Then; as in the proof
T. Then J = I as in the proof of Theorem. 5.7 of Chapter I and of Proposition 1.4, we have’
7 is a desired connection. QED.
It is possible to prove Theorem 2.1 using Lemma 2 o(ri,) = ad(a;‘)o(ti,) + a,lci,,
and a partition of unity {fi} subordinate. to [Vi} (cf. Appendix 3).

where at-‘& is now a curve in the Lie algebra g = T,(G) of G.
Let r,ui be ‘a connectioc form on &( Ui) which extends the given The curve’u, is horizontal if and only if&z, I == --@(tit) for every t.
connection over A n *Vi. Then w’~ Zi giwi is a desired con- The constrnction of U, is thus r$dhced to the fqllowing
nection form on P, where each ii is thl function on P defined by
L EMM A. Let G be a Lie ~group ana’ g its Lie’h&ebra id&$&d w&h
gi = fi 0 7T.
’ : T&G) -.-Let ,Y,, 0 5 t S 1, be a continuous GWVG iin’ T,(G). Then there
3. Parallelism axi.+ in G ,q uniq!e curue a, ~cluss Cl such thut a, = e atld h,a,’ =f Is
. t
j&o 3 t & l.,, e.
Given a connection r in a principal fibre bundle P(M, G), we
shall define the concept of parallel displacement offibres along any Remark. In the case where Y, =.A for all t, rb, ,curve a is
given curve 7 in the base manifold -11. nothing but thti I-p@m&r sub’group of G generated bv 11. C;ur
Let 7 = xf, 2 ;< t :; 6, be a pieccl\-ise digeretitiable c&e of . differential e$tidtion ““;5:’ = Yt is hence a genera&a&n of the
class ~1 in ,9/i. A horizontal li,,i 01‘ simql! % i;ft of T is’8 horizontal differkntial equati&%r +@pqameter subgroups.
curve 7* _- ut. a ;.. t 2: 6, in P such that P(u!)’ = X, for a 2 t 5 b. Proof of Lemma. ac%‘e tia)r assume that Y, is defined and
Here a horizontal curve in I-’ means a. piecewise differentiabie continuous for all t, -QD’ < t,< 00. We define a vector field X on
“’ s .:
curve of class C* whose-tangent \‘ectors are all horizontal. . _ :

I -A
',: .
G X R w ~OI~OWS. The’value of X at (a, t) A G ‘X R is by defini- independent of a specific parametrization xt ssed in the following
tion, eq&d to (I’&, (d/dt)&e Ta(G’) X’ T;(R), ‘wh& z is the i”
natural coordinate system iri R. It is clear-that the infegral curve I
of x starting from (e, 0) is of the form (By, t) and’;rt, is the desired
curve in G. The only .thing we have tq,verifj.& &t’ut is defined
for all t, 0 ‘S; Z S lr Let vb 4 exp tX be a: lo& l+a*meter
group of bkal transfbrmaths ,tif G. x7 R generated by X. For ;
each (e, S) 4 G x R, there is a positive numb& :a, duch that
qt(e; Y) is d&red for Ir + 4.5 8, and I~,<.&;S(PropositiC 4.5: of
Chapter I)., ‘Since the subset (a) x [O, 11 :G x R;is c&p&$ * evident.
we may choose d > 0 such that, for’ r..g [O, 11, q~(e,, r)“&’
~ defined for ItI < 6 (cf. Proof of Proposition .M.‘of Chap&I). -d “~‘i@fgayqpI 3.3. (a) If r is a piecewise dlyerentiable curve of class
, Choose s,,, sip . . . , s, such that 0 = S, < si -2 ‘+ . . ,‘*r #k G.- l’$&d !i c1 3. M, ‘t&ma tMFlle1 displacemeN along 4 is the inverse of the
$1 - si-i < 6 for every i. Then QI$(C, 0) 3, (ut, it)~‘&h9i&I’~~- I parallel di#ac~~t along 7. ’ .
0 I; t 22 $1; gJ,,(e, s,) = (b,, a +s,)isdefinedfarO ~VS% - $1,j ,($I) if 2 is a-cuwefr&n x toy jn M,and p is a.cuwe fromy ,% z,,in M,
where b&l = Yu+,l, and we define a,,& b,+~,~ for s1 s ‘t d So; 1 &par&$ disp@cer& along the composite curve ,u * r is theVcomposF of
. . . ; Qlu(G .s*-*.) ,= (CIU h-1 + u) is defined for 0 S. u .S ‘zl - sbSl, the par&# dispkemznts ‘7 a&#.; ,
where t,c;l = ‘YS+Vlcl, and we% define at 5 c t-&$-l, thus r .:
completing the construction of a,, 0 5 t S 1. QED,

ul S&I that ~(a~) = x1.
obtain a mapping of the
maps a0 into q. We denote this mapping by the same .letter T and
displacement along the curve T. The ‘fact that
is actually an isomorphism comes from the
ll such isomorpbisms of
P~txmpk 3.2. Tire paralkl a?splac&lent along my .“curve r of Proposition 3.3. This
trmuprutcr~~~~~tionofGon~~~~RR,=R,~~~f~~a~G. my group of I? with refmence point x. Let
PI&& This follows &I& the fact that ,evm’ h&zontal curve consisting of loops which are homotopic
is mapped i&to a. horizontal curve by,&,, . ; ;$A$~~ Q&D. ‘to zero. The subgrot@ of the ,holonomy group consisting of the
.. * parallel displacements arising from all T E Co(x) is called the
The parallel displacement along any p&&se differentiable restricted holonomy group of I’ with reference point x. The holonomy
curve-of class 0 can be. defined in an obvious manner. It should group and the restricted holonomy group of l? with reference
be remarked that the parallel displacement along a curve T is point x’ will be denoted by ‘Q(x) and @‘r(x) respectively.

It is convenient to realize these groups as subgroups of the ,
structure group G in the following way. Let u be an arbitrarily (&,p*) . -r* . ru *-l is a horizontal curve in P from v to vb and its
fixed point of the fibre n-l(x). Each T E C(X) determines an element, projection into M is a loop at ,n(u) homotbpic to zero. Thus
say, a, of G. such that T(U) = YM. If a 100p p c C(x) determineS b E O”,(u). Similarly, if b E @O(v), ,&en b E W(u). QED.
b c G, then the composite p * T determines ba because (i’. 7)(~) =
/A(UU) = (p(u))4 = uba by virtue of, Proposition 3.2. The set of i If M is connected, then for every pair of points u and v of P,
elements a Q G determined by all ,r c C(X) forms a subgroup of G there is an element a E G suchthat u - ua. 1.t follows from Propo-
by Proposition 3.3. This subgroup,.denoted by 0(u), is called the sition 4.1 that if M is co&ecfed,,the holonomy groups CD(u), u E P,
llolonomy group of F with reference point U’Q P. The restricted holonomy are all conjugate to each other. in G and hence isomorphic with
group @‘J(u) of r with reference point u can be defined accordingly. each other.
Note that (&&) is a. group of isomorphis’ms of the fibre ,7~-l(x) onto The rest of $is section is devoted to the proof of the fact that
itself and Q(u) is a .subgroup of G, It is-clear that there is a unique the holonomy group is a .Lie group.
isomorphism of 0(x) onto @(a) which makes the following ‘~
diagram commutative : j T HEOREM 4 . 2 . Let P(i, G) be a principuljbre bundle whose base
‘manifold M is connected and puracompact. Let iD(n) @O(U), u Q P, be
the holonomy group and the restricted holonomy group of a connection I? with
reference poin’t u. Then :
(a) @O(u) is u connected Lie subgroup or G;
*, Another way of defining -(o(u) is the following: When two points
(b) @O(u) is a normal subgroup of Q(u) and %Dfu)~,@(u) is countable.
u and v of P cr+p be joined by a,horizdntal-curve, we write u 4 v.
By virtue o&this theorem, G(u) is a Lie sub&cmp of G whose
This is clearly an equivalence relation. Thin @(a) ia equ& to the set
identity component is @( -:: :
of a 4 G such that u N uu. Using, the fact that u N v implies uu N vu
+ Proof. We shall sho that. .ever)r element of @O(u) can be
for any u, v c P and a c G, it is easy $0 verify once more that this
joined to the identity elkint.. by a piecewise,differentiable curve
subset of G forms a subgroup of G.
of class Ck in G which liea in W(Y). By the theorem in Appendix 4,
.’ PL~POSITI~,N 4.1. (a) Ifv = uu,uEG,then@(v) =ad(a-l)(@(u)), it follows then that @O(u) is a connected tie subgroup of G.
that is, the hqlonomy groups F)(v) and Q(u) are conjugate i<,G. Sim{lar&, Let a c @O(u) be obtain& by the parallel displacement ,along a
@O(v) = ad (a-‘)(Q?(u)). ,, _, piecewise differentiable loop T of class Ck whichis homotopic to 0.
(b) If two points u and v of P can bejoined’by a horizontal &rue, then By the factorization lemma (Appendix 7), T is (equivalent to) a
O(u) F O(v) a n d W(u) =,JV(v). * -’ ” product of small lassos of the form T;‘-*“P *‘TV, where or is a
Proof. (a) Let b E’@(U) so’that r&‘nb,Theri ua N (ub)u so that piecewise differentiable curve of class Ck from x = r(u) to a. point,
v N (vu-l) bu = va-lbu. Thus ad (47 (b) E @(ti). It follows -easily say,y, and ~1 is a differentiable loop aty which lies in a coordinate
that O(v) = ad, (a-‘) (O(u)). The proof for v(u) = ad (a-‘)(@(~)) ..neighborhood ofy. -1’t is sufficient to show that the element of
is similar. ed by ev:h 1s~ 6’ c ,+A *‘~~.can bojoined to the identity
(b) The relation u A v implies ub N a& for every b E G. Since-the elem&# ,Tti~.&mer+t. &obviously ,equal to, the element of @O( v)
relation N is transitive,, u 7 u6 if and only if v N ub;:thk IS, defined by,th@~.~, fi9re.u is the point obtained by,the parallel
b E a(u) if and only if b Q (o(u). To prove cl@(u) 3-@@(u), &t p* be displaeement..of u;,&ng 71. It is therefore sufficient co show that
a horizontal curve in P from u to v. If 6 z W(tl), hell’ there is a the element b 6, ~@@): &fined by the differentiable loop fi can be
horizontal curve 7i in P from u to ub such thac’b curve T(T*) in joined to the identie*,element in O’“(v) by a differentiable’c&ve -=
M is a loop at g(u) homotopic to zero. 1 Then the composite of G which lies in @“@&%
.a, L,
Let x1 , . . . t;,x” be a,loqal coordinate system with origin at y

PO(a) is the idehsity-transtbrmation of V for each a Q G.degwe r of( P. . ye shall prove curve a. (f% 4.a representation Y.(s)a. 0 $. To / p prove that ai($). p(a) is a linear described by uJs). In this section. The fact that b(s) is of class 62 : _ to T.S t I. in G sut&’ that’s.bundle and p. s)) = f (t. it is suEient to show that$i”Is). For each jxed. tangent to . of types (p..1. -This implies that @O@] is’:a’-normal ‘subgroup of O(U). then ‘we proof of Proposition 3. awTppins of ckas. 0 9’ t I. is the trivial representation of G on V. 11) such thatf (t.‘4 . we define? for &&fix&l s.0 g i ?%.(s).1. 1) is the trivial curve y.(sj if” a solution’ of the equation a. Clearly. of’ the ’ Example 5.j: . that le. ~‘6..(M). . Curvature form and sirwture equation . . . iI point (e.1. . the. s) for all (b. 4. _ j? (a. .t)l’b tht Wxhi tangent to the curve of G on a finite dimensional vector space V. . curve u~(s)$(s)..W is a V-valued & z&o. modify T within each neighborhood and obtain a element of (PO(U) obtained from.b E G. . V) is nothing but a form 9 . As .i A 1. then 71 * T.. 0 ~2 t 5’ 1.3.&m& and hence the proof of (a) ‘of ~&d&&i . 3:1.(s) irM%low? Such a form Q is called ~i%rwial_fbnn if it ishorizontal in 8the sense the fact that each solution of an ordinary linear’ div1 that 9(X. we saw that.. Set z+(s) t F (t. Since M is con- and uo(s).(u)‘. Par each fixed s. . htf: z )< li .patallel di@lacement of uo(s) along the curve f (t. 0.#oint of P o&&d bility (Appendix 3). Ck. a&)“* 8 vdX@Vm at least tone of. is horizontal. ir. tie defined the parallef @placement along any Proof of Lemma.%~kd.” ” -’ is.Q. If T.. . 0 . then a ‘We n& prove (p).1. by the ... . 0 5 t s 1. r-form on the & 3*‘“t*+/. the tangent equa&n &h parameter s is‘ diffekenti&k ins as ma&ti@# the vectors Xi1 of& is vertical. s). s). a’dz&&fiab& curve of c&ss:cL in -P. Th‘e existence of such an F follows curves of class @.$-I the pr$fofthe’lemma for Proposition valued r-form 9 on P s.whe@ . s) c I x defined the holonomy group a(4) using piecetiise-differentiable I and that F (0. loop f (t.I 74 FQuNDATIONS O F D--L GEOMETRY II.&&&& able of c1a.M b a dti$ientie&. A pseudotensorial form of . % QED. the composite on P which can be %essed.holonomy group thus from local triviality of the bundle P. Let F: I x I-+ P be a differentiabie piecewise differentiable’ curve’ of class ‘0. ‘isa~difkrentiable I. we mapping of class Ck such that rr(F(t. = 1. We may cover 7 by a finite number of coordinate and f (t. 71 and 72 define the same element of @(u)/@(u). ‘1. Then/(& s) = X. s).(M) is countable.j&i that nected and paracompact. THEORY OF CONNECTIONS 75 and let p be defined by xd = x’(t). Set u&) = v. This complke~ ‘th@@W?f’. . i.I = [0. 0 $ s d 1. his. . It follows easily that a.the.f”(t.= p(a-l> .$ector field X(J) throu . curvk of’cla&‘C:i”’ in p:Let “iy &&&e c&mection fornrofF! Set Let P(M. s. i . 11. f LEkMA. homotopic to zero. Thus. equation is (cf: Appendix.. 0 ~2 t 2 1 ands. .of: type (p. ~F~wbcn:s~. for each fixed s. We define a homomorphismf: rl(M) + @(u)/0O(u) as follows. ft and 0 5 t.(s) ‘= -c~(rJ~(s)). cP(u)/@‘J(u) is also countable.uch that v QI‘. 0 s s ‘I 1.Q_ for f E G. it satisfies the second axiom of count- ’ 4uoW =f(O. s) * s + Let Tl(M) be the first homotopy group of ?4 with reference point C1 .ii. k%ce. s d 1.k)q(s)--l & I”. a vector field X(s) on G x R so th& &io. If we denote’by (P*(W) the. G) be a principal fibre. Setf’(t. If 7 and pi tensorial forr& iof de&~ r. let 7 be a continuous loop at x which 1 X 1 into M (where .‘&* differentiable curve of -class : 5. so piecewise differentiable loop pi of class Cr at x which is homotopic that b(0) = b &d b(l)“+ identity. . T~HI t&woe t+(s). In the obtained from piecewise differentiable curves of class C”. 0) is the curve .=. Then as in the“. let q(s) be :&. and 7s are two such loops. . .“whetie &(.JA represents a.1). which is denoted by f (a).‘t) is the integr&kurve of . V) is a V- &(.(s). there is a have obviously @i(u) 2 U+(u)’ 3 . .e. transformation of V for each s rG and p(ab) t p(a)p(b) for p r o o f of’Propc&ion. let b(s) be the neighborhoods. Rimmark: In 53.a fib+.-. If-p. .r C* . a.~ is homotopic to in s (as’s mapping of I into G) follows from the following zero and defines an element of W(U)..’ is a homomorphism of ml(M) onto @(u)/~~(u). 9 = 7r*0]~ where 9.(s).) t G x R: The differentiability of d.w CL. s) ‘= uo(s).(.isfied.@) F e and that the later in $7 that these holonomy groups coincide. J)) is a differentiable mapping of class Ck of For each element 01 of r.S)%‘(t) fi3r.

. Jv c ~ti(~. By Proposition $11 of Chapter A few special cases of Example 5.(P) -+ Q. . * * x T. . The . . . and Q. o(B*) = B and [A*. respectively.+*] = [A. 9(X:. Let G..B Q 0. . can be . 7&Q))). (2) X and Y are vertical.this case.1. The form Dp.l(x). It dw(X.11 of z I ‘7 Chapter I. . . V). v. (1) Xand Y are horizonta$In. is vertical. $) = u(@‘:. .(P). A*]) pseudotensorial form of type (p. choice of u and Xt. which will be also denoted by X. bilinear and skew-symmetric in X 8nd Y. u e P.for X.(M) x . V). We extend X to a horizontal Xi. Let o be a connection form and element of the standard fibre V and u is a linear map$ng .A*(w(X)) . = (dqJ)h is called the exterior covariant derivative of symmetric mapping 9.1.2 will be used in Chapter.nil(x) for. (3) Xis horizontal and Y is vertical. we: have trae (P. Since both sides of the aboi. u g G. k) can be easily verified that this element is. of T. By Proposition 3. 0 hen other hand.... QED. .). be the projectipn.wh c:. ( M ) . hX. . a tensorial O-form of type (p. X:)) is an element of n. .. where u is any point of P with T(U) = x and Xt is anyvector at u such that r(XT) = Xi for each i. independent of the -7 b. Since the right hand side of the equality vanishes.. A. it is sufficient td vkrify’ the.of Y B its curvature form. On the (P. G). I. it. given a skew-symmetric multi- linear ~mapping $=: T. B*) = 0. 0(-Y)] + Q(X. From R. A tensorial form p of degree r of type (p. vector field *on P. = -[A. ’ Proposition 5. By P&L *. . If p is the adjoint representation of G in the Lie algebra 9. A*])..(M) x * * * x T%(M) (i. 1. Let Y = A* (b) dq is a pseudotensoriaZ+ . V) on P can be ho$+&al vectoi.y. X. . Y ) ‘G -8’[w(X). *** . where that is. Namely. we have Let I’ be a connection in P(M.2 (Structure equation). g) is said to be of . V) . Do is a tensorial P-form of type ad G and is called the curvature form of w. (b) follows from R: 0 $ = d 0 R. o(X) = o(Y) = 0 and the equality reduces to the definition of a. Conversely.: connection form% is a pkeudotensorial l-form of type ad G. Then onto nix (x) so that u(v(Xy.. V) can be regarded as an assignment to each x E M a multilinear skew. Let p be a representation of G on V and E the bundle associated with P with standard fibre V qn which G acts if one of Xi’s. =t. III. Here . V). .(P). = (dy)! is atensori~form of it is suffiCieSt4% show that &o(X. Proof. B*j = A*(o(B*)) .t is erLdtdt &het = -&(l....B*(co(A*)) . . .76 FOUNDATIONS OF DIFFERENTIAL GEOMETRY I. cases.rf&lows that $ is a. (c) follows from (a) and (F).(p. Q(A*.’ 77 Example 5. :. A*) = X(w(A*)) . a functionf: P + V suc$ thatf (tra) = &+)f(u). Let 2do(A*. . ehuality in the followi’ng three special. at u. o h = h 0 R. B]*. into the vector space 7ri1 (~1 which is the fibre of E over x. x E M. THEORY OF CONNECTIONS . X:) is then‘ an THEOREM 5. Let X‘= A* and Y = B* at u. T$P). tensoria] fozm 9 of degree r of type (p.) = q~(hX. sponding to A and B respectively.a. since o(A*) = A. 9(x?. (c) The (r + l+jkm Dq d&cd by Dg. . a l G. PR~POSKTION 5.2. w(B*)].A* and B* are the fundamental vector fields corre- identified with a cross section M + E.(M) . . is a tensorial form of trp (p.+ l)+fenn of t%e. In particular. times) ‘p and D is called exterior covariant di$erentiation. . eacli Proof. B * ] ) h: T. B] -[o(A!). . be the t ’ vertical and the horizontal subwaces of T. (a) Theform qh de$ned &y (vhr(X.@e ad G. ) x:) = u-‘(f5. through p.Y E T. a we define (pseudo) tensorial form of type (p. Every vector of-P is a sum of a vertical vector and a.‘X:)). Xi E T .*.(7r(x:). A*) = 0. r pl is a &ndotin&al r-fm on P of typ.:.&’ equality ‘are defined. where x = r(u). . NOW it is suRici& to p&&the following . 2do(X.o([X.o([A*. by .

+: .’ principal fibre’bundle into &nother such ‘as a homomorphism. .t:r$ the vector rP to a vector field pn M a@ .ide. Y) = ~(A+((p(Y)).f g.’ horizontal. q(Y)] = 0 or of the. c&f A cok + @.:. Y> -t..r. then [X. be a .s 4=&c i i: I . . By Proposition 1:9 of Chapter I. ‘: .& 2) = 0 ‘ .d. G’j -+ p (ill. r..rF .denotes the l-parameter subgroup ‘..b ‘= 1 . for the sake of simplicity.the iright ‘hand’ side. The fundamental vector field Ai . +‘I1 + iMx3. where a. e.‘. the structure co&ants of g with respect to. of G gene&edAby. . 2) = 0 whenever X. it suffices.k If both X and Y are horiwntal vectorjieldr on’ P. Cartan”) is sometimes written. Let Then -‘J connectionform and a’ the curvature form @. Apply Proposition’ ‘1. [X. whenever . suffices to consider .A.5. j?.ars via@4 ._ . a di&omorphism’of M’ onto M.&(X) E 0 whenever X is horizontal.. ” I ’ dv(A*.. ‘I Proof.. ::: %’ 1 The structure equation (often called “the structure equation . r.c *<_ I P ROPOS I TION II.>[A*.11 of Chapter I. &morph@ f: G’ + G such that the We apply the exterior differentiation-d to the strutture.i@. . fat X..+(X).. we.. structure equation‘just proved..l.HvC. If a...j Then ?he structure equation In $5 of Chapter I.$o?iwnta~ subspaces of I? b f.= &ejfldr jik. : so that it..DATIONS OF DIFFERENTIAT'GEOMETRY . As in the proof.. we have the three special cases. B ..&en lift it to a~hor&&al vector field on If X. . .element A E g and X is a horizontal vectorficld.jX) .& :. we haye ?Paes of I” .. dL?‘(X. Y. “*a.By t h e n .‘i‘ A* (q(Y)) = . .2. f’. Y).have b o([X. Let X = A% at 3~.(. that is.. ‘as follows : . (We first &. a 6 C... (a) There is a &&&n&tion I? in P. Let o be a connection fmm and ~1 a tensorial l- 79 form of type ad G. $ study the effects of these ’ T HEOREM 5. to provt I that (&l’. so is R. By the definition of D. A*] is horizontal: (@ED.i&. Thus [X. = 0. .is vertrcal.-[A e(Y)].is induced by R. . . A*] = yrj f [R.t . UC P.$ + dW.lwi = --Jr. & A* . Y) = 44x. Y. . G) be a homo- dQ(X.= 1.4 (Bianchi’s ident@... t&n . L Proposition 3.=. yl.X.&A+$) ‘= iA*(cp(Y)).‘e. Y.. :. .].).basis <for. &Q*.XJ..the l-parameter subgroup of G generated by Proof.r>.Let e. .of Theorem 5. shall shqw that . and 2 are all horizontal. ‘. The ‘only non-trivial case is the case where ‘1. . Y]) =‘+(x.. Then L E M M A .--&(A*)) . cjk do’ A & + @ &. which is invariant by R.‘and 2 are at11 horizont&#ectors.j . ..3 of Chapter I to the lefthand. r. . y) = 0. . ‘I Proof. . THEORY OF CONNECTIONS 5. Let w = Bi oie. e.suffices to show A*(q(Y)) + [o(A*).&#ation. i. We ” COROLLARY 5. Then. . 1’ ” I . [e. .&t&‘&at the horizontal sub- Since.. A*] is horizontal. then Lie algebra g and c&? >. ‘: 1 Proof of Lemma.“&noted also by ‘Y. . Q$Q. : 78 FOUN. . If A* is the fundamental vector field corresponding to an Wx. . an do+ = -@.: ’ We extend Y to a horizontal vector field on P. D2 = 0. (PMI . and Q k Cj C’e. we considered certainmap@ngs of one can be expressed as follow$ * ST I : .$ E. of the equality ‘vanishes.Yr T*(P). where A z g. X is vertical and Y’is horizontal. .

QED. then f *o = f *w’ and f *Cl =*f *!A’.e fibre at u. We shJl show that the distrilkion (cl Let 7 be a loop at x ..of r w(fX’) = co@*) = A = f (A’) : f (. 2 . Let I’ be a c&&iion in P. ‘Lqt fzsP’~&4’.. (c) IfB’ 6 P’ and u = f (u’) Q P. 0 R.) = x& of (&. then f *w = f * CO’ andf *a = f +Cl’. Tti~pf (7’*) is the horizonjal lift 6f T starting from u.(P) f *dcu’.. then ub = f (u’)ab and loop .w’. in the case -induces a linear isomorphism 7: Q. 4fX’) =f(4U) for X’ 2 TU. where v’ E P’ and b E G. THEORY OF CONNECTIONS 81 (b) If w and S2 are the connection form and the pyvature form of I? respectively. wl say that f ‘maps the triviality of P. it is sufficient to prove that the projection 7: P -+ M connection I” into the connection I’.‘ -’ ’ ’ Proof.&kc.1. %I@) and every a’ E G’. . Thus V[u’)) are the holonomy grollp and th./et) . tihere O(u)’ and cbo(u) (resp. we have then v’ = u’c’ for some c’ l G’. u’). r is said to be invariant byj’. The invariance of the distribution by G... If u = f (v’) b. (b) If CO’ and Q! irle . X’ = A’* at u’. where x % n(u).= R. then the isomorphism f: G’ + G maps @(u’) into O(u) and @O(u’) into @O(u). Y’) = f (sl’(X’.) b e a hpmo- &irphism’s&h that the c&&onffing hom9morphisti. of T’ starting QUb = R.. G’) is 2 reduced subbundle of P(M. we obtain d(f *o) =~ d(f ‘0’) and f *do = ‘-that u = f @‘)a.J G’ -+ G maps G’ isomorphically onto G. we have f (X’) = A* at f (u’). --+ T.‘.Given a point u E P..‘C. of (Q.bundle P(M. is a connection in P.ij are and Cl tf+&vatulcc form.In the com~+$&~di?g~~m forniation. G) with . Q.(p’).(x’)) ( 1) X’ is horizontal. Cd] +f * w. wt: say that the connection-J’ in P is reducible to the ‘_ connection I” in P’.4 +&(Qd = 4 o R.. where f on the right hand side is ’ every horizontal vector into a liorizontal vector. the homomorphism g’ --+ g induced by f: G’ 4 C.(M).. 80 FOUNDATIONS OF DIFFEREIhIAL GEOMETRY II.o’ or It is sufficient to’tierify the above equality in the two special cases: f * Cl’ means the g’-valuedform on P’ defined by (f.. If u = f (u’)a. of T. of (&A This implies that f *Cl = f *R’. thusprovingthe frofi u’. In particular.. and in this case. . r’) with r@rencipoint U (i@sp. f (v’)b =f (u’c’)b =f (u’)cbtindh encea * cb. respectively.j*w] +f*L2 = -*[f *cd.U’) -+ T.(P) = s’t&men’t ‘(c)‘is now evident. is the tangent space to th. connection form and the curvature form nf I” (b) The equality f *W = f . CO the connection form the mappings n’: Q.where f . If X’ is vertical. W’ can be re. We define the hbrizontal subspace Q.written as ~follows: __.’b$y-is&evident spaces of I” are mappea int&horizontai subspaces of JZ’ by f. Let 7’* be the horizontal lift.. The unicjuene<.that’M = M and f: M’ + M is the ident@ trans- invariant by G. is the horizontal sub&ace of T...(P’) with respect to I” We may assume that u = f (u’) since-the distrib’ution u -+ Q. Set T’ =J-~(T) b that 7’ is a u -+ Q. ‘PROPOSITION 6 . WehaveR.‘(A’*)) = f (d(X’)). of (Q$) = R.(. both sides of the.of (Q”. u’ f P’.: Then linear isomorphisms and hence the remaining ttio n’ihppings (a) There is a unique connection I” in P’ such that the horizorrtal sub- must be also linear isomorphisms. Since f (u’a’) 7 f (u’)f (a’) for O(u) and @O(u’) onto @O(y). thnf: G’ + b maps @ti) onto where d’ Q g’.at x’ = I’. -T.= r(u). 0 f (Q..) = Rb(Qu). equality ‘vanish if-X’ is horizontal. Y’)).2) : by Q.G’)“+ P(M. G) is Qujf-Qu ' called an automorphism of a connection r in P if it maps r into I’. Since f: P’ + P maps or (f* Cl’) (X’. CO’)@‘) =f(~. choose ti’ Q pi arid dk G such From f *w = f . where G. where P’(M’. where Q. We shall now prove T. By the structure equation ‘(Theorem 5. (a). If we set c = f (G’). + G.).(k’) and f: ?“.? restricted ho&m&y ‘&oup. and (2) X’ is vertical. PI and u = f (u!) E P. An autumorphism f of tha. )-. We shall she... which proves our assertion. of I’. . that Qi is independent of the choice of u’ and a. Set 4 = f (A’) c g. then u = -gf*w. By local In the situation ‘as in Propositicm 6. (resp. (c) If u’ c. from its construction. is jection f so. of I&d = R.

THEORY QF CONNECTIONS 83 Proof. G) be a principal fibia&ndle with a connection I’.a-. . over M with group G x H.. and I’* cqwib~. _ _. Then bundle P(M.e form o’ defines a connection (Proposition 1. we say that I” is induced sponding to A. where H is a Lie.?’ the curvature forms of I’.. .J. Set w’ = f-l *fees.1 and. and r.‘I PROPOSITION 6. v) onto (Do(u) (resp. o’(A*) = A. such that g = m + h (direct sum).#‘) and a’ Q G’ and set X = fX’ and a . . and (u. Proof. Then PROPOSITION 6. are We first establish j’ . This proves that th. v z Q. . The homomorphism ‘. mental vector fields corresponding to A and A’ respectively. *-‘&on Q .- .X) = u’(R. ‘F. . The holonomy group @). rp.:i’.’ Then the holonomy group . The connection -defined’ by o in P is reducible to a Q -+ P to P +’ Q. G) and a mapping f: M’ + M. in P induces a connection.= f (al). Let A E b. v) E F + Q. ’.1 and is left to the reader. onto O”(v)).. this larly. i “.1). For ewry connec+n form o~iti P. @O(u) x @O(v)). we consider P + .X) + q(R. ad (a-‘) (W’(X)):‘-& ad p-l) (y(X)) Y . and entiable curve of class C”. given a Q.sub&up of G. fH: G x H + H) maps @(u. Let Q(M. H) be. We define I” by defining its connection form 0’. Let A* and A’* denote the funda.2.and ad (H)(m) = in. we obtain M x M. then o’ =f*o. in the induced bundle f-1P.that we used the fac.4.(uo) will be rQ respectively. For any principal fibre bundles P (M. Let X r ‘T. . where = f-i(ad (a-l)(W(X))) = ad (k-l) (f-l(w(X))) @[u)and @O(u) (resp.6.._ @(u. a curve will mean a piecewise differ- morph fp: P + Q + P andfQ: P + Q -+ Q mi@@in@ I’.2 and is Let A’ B g’ and set A = f(A’). Iff is a bundle map. sia’ted. The proof is similar to those of Propositions 6.J -3 -ii .Q as a principal fibre bundle ad (a-l)(q(X)) is in m. that is. left to the reader.of I’ (resp. Assume that the Lie algebra g of c admits a subspace rn. In particular.3.( Q) and a E H. \ ad (a-r) (o(X)) =. I’ .a subbundle of P(M. I 7:’ in* . Q(v) and @“( v))are the holonomy group and the restricted holonomy group of rp jresp. H). : QED. f: 9’ + g induced from j’: G’ + G.746 -Re&ction t h e o r e m Q( M. Denote by P(uo) the set of points in P .X)). to Q is b-valued.. H) respectiveb. . is a homomorphism with the connection in the subbundk-Q if and only if the restriction of o corresponding natural homomorphism fo: G x H + G. tb THEOREM 7. (b) If o.G). G) and Q (M. The restriction of the projection P x Remark. by f from I’. Let x’ E T. Unless otherwise. where M is connected and paracompact. :.xDt a natural way.^. .. Let I’. Then we have subgfoup of O(u) x O(v) (resp.denoted by fp. II. Observe. v)‘.I <G+I: t!” means w’. onto a(v)) and @O(u. Since AM and M are diffeomorphic with each other in 44%X) = ad tabi) (w/(. every connection -o(R. be corinections in pCJfi’& W . wheref-i: g + g’+is the inverse of theisomorphism (c) Let u i P. (a) There is a tinique cann&ion I’ in P + Q s& W’h home. $imi.1 (Red&i& theorem). -the Ij- The verification of other statements is similar to the proof of j component W’ of w restricted to Q is a connection form iu Q Proposition 6. Let P + Q be“the restriction of P x Q to the diagonal AM of Comparing the &components of the.f WP +f $. G’ = G andf: G’ -* G is* Hence.-. v) ante 9(u) 4&J’) =f-‘(q(f (&x’))) =f-‘(w&q) (resp. forJo: P + Q + Q and f@: G x H-c H. Let 9 be the m-component of w restricted to the identity automorphism. Under the assumption in Proposition 6. The left-hand sides of the preceding two equalities coincide.. op and o9 are the connection formr and@ Zapi &da. =) ad (k-l) (0(X’)).‘$T r. Then &(A*) is the b-component of @(A*) =%A. denoted by (o(h). Then . we have w’(A’*) = f-yo(A*)) =f-‘(A) = A’. Let “0 be an arbitrary point of P. Let P(M. 0 =fti. I .- 82 FOUNDATIONS OF DIFFERENTIAL OEOMETRY ..4.and A* the fundamenta1 vector field corre- In the situation as in Proposition 6.: i+ . where in is the Lie algebra of H. the restricted R&momy group qO(u.. v) of r) is a .. right hand’sides.x). ‘fa: G x H -+ G (resp. P x Qis aprincipalfibre bundleover M x Mwithgroup G x H. +-f&+9 n =.

. then there is an element a ‘k H s& that v = ua . H) be / subbundle of $$‘b) and I? a ” connection in P. differentiable by the very d&&on of a connection. ‘ . (2). .(P) is where S. It.+I. 1 d k I co. : definition the horizontal subspace of T.P(u) is the maximal integral mkifold of S of u along TV.. let T* be the segment from x tog with respect to the. Gcing back to the proof of the.. It is structure in Q in such a way that p::Q ti r-l(U) -+ U x H easy to see that. . for every pair of points u and v of see that and R. we see that Q all isomorphic with each other.1. The horizontal subspace of Tu(Q).1.r *r: L EMMA 1. 0) with respect to this coordinate system. G. QED. ~5 the origin (0. through u by the.n s. Let c(y) be the point of P obtained by the parallel displacement .(P(u)) . x of M has a neighborhood U and a cross sictionta: U -+ P such that we have.. obvious that F is reducible to F’. i. we prove that the holonomy groups aD. u c Q.. It is easy. k . In other words. u E P. .(tl).2) and that the subset P(uO) an’d the Proof. .2.emma l.Nomizu and Ozeki [2].:“P(u) T P(ua) is an iso’mdi-pliism with the corresponding w maps Q n ?T. _.84 FOUNDATIONS OF DIFFERENTIAL GEOMETRY II. H) is a reduced subbundle of P(M.) satisfy conditions (l).$.C Q. Let U be a cubical neighborhood of x defined by ix’] < 6. by a horizontal cuyve. Now (1) of Theorem 7.(u). subgroup of G. (3) $U. a). G): or P(u) n P(v) = empty. L EMMA 1. The distribution S’ is tangent to Q. It isnow evident that Q is a Using Theorem 7. say k. M being paracompact.. either e(u) = P(v) a(U) c Q. 2 :’ i:. To prove the .1 (b) ) . ( 1) We first prove . using Proposition 1. It is evident .holonomy bundle through u.. for every u c-Q. given any u and v. Jet Q be a subset of P(MY G) and H a Lie subgroup of We shall call P(u) the.each u !.(ij S is d$rentiable and involutive.(P) with respect to I’. All the holonomy groups CD.iiand (4) query point v can be joined by a horizontal curve) is an equivalence rekrtion.CONNECTIONS 85 which can be joined to u. + 5. rr:. are as in the proof of Proposition 5. = T.. For each u E n-l(U).introduced in 64 (u .G by setting y(u) = (x. . Since every a e G maps a c G the element determined by u = a(x Define an isomorphism each horizontal curve. .. We know by Theorem 7. .second definition of @(u. It is Proof.(u). . we T HEOREM 7. x”. t h r o u g h u. . . the horizontal suB$ace of T.. We ’ group @((u.‘We first prove coordinate system x1..(P(u))’ = P(ua) y: rr-r( U) -+ U x . Define a distribution S. Let Q(M. becomes a differentiable manifold. Then * Proof of Lemma 2. ) xn be a local coordinate system around x such that x is S. (1) P(uO) is a reduced bundle with structure group Q(~~).P L M-maps Q onto M.‘(1 j We set (2) This is an immediate consequence of the fol1oy. xl. first assertion of Theorem 7. let x = r(u) and the disjoint union ofthe holonomy bundles. (Theorem 4.1 and P(u).Q~P by setting .V E Q andrrju) = zontal curve. Proof of Lemma l’. see that. subbundle of P. To verify condition (4) of Lemma 1. Fix a point u c Q such that r(u) . :n% Li L EMMA 2. Then 0 : U -+ Pis a cross section such that o(U). Assume: (1) the projection.J given before Proposition 4. denote Qm(u) by a(u) and the holonomy bundle.. = s. (2)’ Q is that P(u) = P(v) if and only if u and v can be joined by a hori- stable by H. is su’fficient to show that al(u) = <p.1’follows from T. Since the rela ion. with respect to I” is by (2) The connection r is reducible to a connection in p(uo).l(U) 1 :I onto U x H.1 that P(u) is a subbundle of P with also Proposition 4.. and (3) of Lemma 1 (cf. .e. coincide as was pointed out in Remark of $4. We define a connection I” in Q as follows. 4 Thislresult is due to . Given any 5ince the holonomy bundles have the same dimension. principal fibre bundle-over M with group H and that Q is a 1 5 k’ 5 co.(2) For . &(Q) = &fork&h a Q H.v if u and 6 T(V).? a k-dimensional distrib+og.bbr u z P. let Q(u) as its structure group. . then l? is reducible to a conn&Qp in Q. ‘there is an element a ~‘(3 such becomes a diqFeomorphism. I .3 of Chapter I that p(v) = P(ua). Thus the holonomy bundles P(u). is horizontal and S’ is vertical. into ‘a horizontal curve.3 of Chapter I. If. Introduce a differentiable isomorghism ad (a:‘) : Qt( u ) + @(ua)‘ of the structure groups. u E P. . Then Q(M. the holonomy group @(uO) ip a Lie coincide. S is pointy Q U. THEORY OF. R. P is decomposed into Proof of ‘Lemma 1.

p.E’ < t < t. 8) ir P x F which is mapped into w.. i d n so that ~-l( Uj which form ‘a basis of S” at every :point of n-f(V). ua ‘is a point of P(u). +I.-a-. is. The argument used in proving Theorem 7. lies in the slice through x0 and hence in W. a/ax".&! an arbitrary Qk(u) for every k (cf. 92-J).1. No .%. :. and its tangent vectors the proof of The&em 7. W(i).basrs .(u). To define Q.2 and its proof).let ‘u($).’ * Let a be any element of O. C”-distribution on a C”&$Yd. is by actually the mar&n@ integral manifold of S through II.) . by definition. In particular. G. t. is open in -. This means that u and ua can be joined by a piecewise O-horizontal curve ut. < W(u) of S through U.. on P corr&” ’ 1. we can still define the ho- “.(E) ark defined as follows. . We shall define vector:fi&ls xi. we have dxi/dt = 0 algebra g(u) of Q(u). H’(u&)). the btorizontai subspace Q.a unique d e G such thdrt o = cr(~(@))a. . Then P(u) is an *en submani. Corollary point of lY(@a&l. For each u B I’. Take’ image of the horizontal snbspace Q. Thk*rmplies that S is invoiuti+e..3 follows from Theorem 7. 0 g t 5 1. . C O R O L L A R Y 7.(P(u)) =: : ‘T. as we wanted to show.O g. The vertical subspace F..‘Thus. it sufficq to shoy that of S”.pomt xo such t&tl P + E...for -g(u). and the vertical subspace F. since P(U) is bundle with an analytic connection..We show%hat u(tJ lies in P(u).k. G). Let u. By Emma 1..the choice of .‘“J: ‘.. let U be a neighborhood of x = n(u) with a crosj section’s: W -+ Chevalley [I. It is easy 6 ‘.!+\wk have u(t) Q P(u). v c d(U). be a curve in W(U) 7. of . . a/ad. into vt Q E. For each u(t) Y P(u(t.(u) by using only piecewise analytic horizontal open‘in <W(U). such that . ‘We have thereby proved that P(u) is T. ‘%.)) and P(u(tr)) is open m 7. v curves.) Let -Ji.0. Since P(u) is a subbundle with For each p&k F.’ t = Lemma 2. coincide. x’(t) = xi(O) for k + 1 r. QED. A. Then the hor&ntal subspace Q. ‘&is yill i$l$ :that t.‘. . (I belongs to Q(u). .3. proving that u( 1) =.E < t < t. we recall that we ‘lrave ‘the natural projection P x F -+ E = 1. G) is a real analytic principal !’ oft.(P) by this mapping a local coordinate system ~1. we shall define the notion of parallel displacement in the associated By definition of t. . 0 s t 5 E. Thisimplies that P(u) = P(u(tl)) so that u(t3 l P(i) w E E.(P(o)-) = S.. k = dim S.. The tangent vector 8. W&p&e . F or small values of t. ‘such that t. 0 5 t 5 P x o F.2. Proposition. We see easily that ‘Q. .. on for k + 1 5 i 5 n. (cf. .. an integral manifold of S. 1 $ i g n. THEORY OF CONNECTIONS 87 differentiability of S. . By assumption. structure group a(u). at each point obviously lies in SU1. .that o(x) = u. Theorem 4. we have . Let xt. i. 86 FOUNDATIONS OF DIFFERENTIAL GEOMETRY II. _.i . 1 $ k 5 (2) Let &ic)*-~:th~ maximal integral ‘matiold of% . x* around the. . form a local basis for S (cf. . Proof. t T. . LEMMA 2.@$ T is $r P(u(~. We entire curve xt lies in the maximal iriti&al manifoid fix this [ E F. L&$&e an involutive. q(t) B P(u).there is. 7i-l( U) . the entire curve ut lies in the maximal integral manifold . r.through u co. . ti(1) * u. . is . ‘7 1.ti’~a:basis of the Lie. definition the -gent-space to the fibre of E at w. implies u(t) c P(i(tr) j%et-t &e any value such that t. are given by Xi (dx’/dt) (a/&r’). We set We are now in position to complete the proof of Theorem 7.. 4 e(n) is.Fn isomorphism~ k$ (n-l) (A. . 7 I are eIame* d g(u) and form. be the supremum Remark. By A9 are-differentiable and form -a basis of S” ore Lemma 2.&at-P(&‘= W(U).‘t 5 f&np~. and consider the mapping P + E which maps v E P . The+ . in P. since for. the entire curve u.f.. . TAt rest&ted h+onomy groqs @g(u). lk$bf Chapter I). In the case where P(M.2.3 shows that mu(u) = @r(u) and Q&(u) = Q!(U). the Proof of Lemma 2: We may assume that I xt. The standard continuation argument concludes the proof of Since ad (a-1) : a(u). a Cl-curve.)). Since P(u). F.. . . @i(u) is the connected component of th identity of fold of W(b). I’) with standard fibre F. fibre bundle E(M. 0 5 t $ 1. is a piecewise C%urve whose tang* dLctor5 . vector field. P(U) . every u c P(u). .j. Let t.2 and Corollary lies’in P(a)_ The point. .is posi&e. On the other hand. is independent of . say. QED. xt P(u) sqh. r.’ lonomy group O. Choose a point If-it. such that a($ = tl at&. There &is@. (Such a cross section’%+&+ constructed in can be expressed by xi = x’(t). ‘e ‘s-. . + E Given a connection I’ in a princrpal fibre bundle P(M. lies in P(u). .

(~. A = o([X& Xi*]) = -20)(X:.-L e t at be the lift to P of x.A:. the u E P and hence ifa.. let ut be the lift of x. Then T HEOREM 8 .2. wechoose a point (uO..of’g spanned b’y all elements-of the form !Z$(X. . let S. where v l P(u) and X and Y $e. there is. Lie algebra of G. .‘we may~assum&. that uoE = wO. Proposition 5. Since F is reducible to I”.J in E._I . case of a lift in a principal fibre bundle. Let A. the? u[ = p(u) for every linearly independent everywhere .“J at v z P is equal. . . Conversely. assume that. G) be ‘a principal jibre bundle and are arbitrary horizontal vectors at v.p(uJ is the lift of xt to E start&~from To prove that‘s is involutive. . . OPJ = r(u) E M such that =-‘ru) is. SI the curvafure reduces to the uniqueness of a solution of a system of ordinary form. if U. identify E with P/H (cf. since [Ai. X. 0 I t g 1. parallel if the image of T. ‘t s 1.f or all t. Proposition At each point v E p.. a@ hence of any two of these vector fields belongs to S. . that ‘@ti) = a(xJ along T gives a(xJ. = . be a basis for q’ and Given a curve xt.riaat by R. X. wkere Jf w. G) be a principaljbre bundle.(E) = . XT] is horizontal.‘spanned by. set ug.88 FOUNDATIONS OF DIFFERENTIAL GEOMETRY .* form a local basis for S. where an ideal of g. . . QED.6 of Chapter I). A cross the’subspace of 0.. zontal). Given a curve 7 in M. Finally. . 0 4. 0 d t d 1. starting. F. in M..+%x0 AT. . ” 8. horizontal lifts of Xi. Z$t T* of T is a horizontal curve in E such that nB(+) = 7..uu(u..) = x.&J. F is reducible to a connection vector is horizontal at each point. By virtue of Theorem 7..e. Let a : M -+ E be n GZOSS section and Q ( M.. isomorphic with Ii x G.]. Let at be the lift of 7 = xt starting from uO.) = a(xJ and hewe ut .‘I.. clear that’.“] = 0 as XT is inva. be the subspace of T. where n = dim ‘II and Chapter I). in Mand any point lemma for Theorem 5..] 6-g’ and [Ai.Xr. Holonomy theorem Given a curve T .where v E P(u) = P and X and Y PROPOSITION 7.. Y).because Q is a tensorial form of type ad G (cf. E Q with n(u. a (hori.f&n u. P. .>(P) spanned by 5. I. . coordinate neighborhoocl x = r(u).. . The uniqueness of T* . ^ in Q. H) and hence g’ is invariant by ad C. )I.-we. (direct sum). Proof. the connection form of I’.(M) by a is horizontal for each x E U. By the respect to F).all elements of .the Lie algebra of. ..E 3 Pi&?.on U and XT.s’ :. so that .a unique lift T* = wt starting from w. and a point w0 such that rrB. .6 @f Chapter I). the parallel displacement of “‘. THEORY OF CONNECTIONS 89 (u. .coinci&s with.3. I’). o(u) the kplonomy group witk rsference point u 6 P and P(u). in Q if and only . 6 Q (with respect to I’) is !$rollary 5.x.f d is-parallel with respect to I’. the coset H) of G/H.* *. 4 is p+rallel (with [AZ. . /l(u.]* = [A). Let I’ be a connection in P. . . choose u. c s. is horizontal in P.uJ is a lift of . then a(x) = p(u). Given any curve x1. X.. P) th e associated bundle with standardjibre G/H. . actually.1. u. ‘. Tken. to .‘i. ~1.&(u) is equal to just as in the. that . Let P(M.. 5) in P x F such applying Theorem 7. A. 0 5 t h 1. + Q.E Q and Let v be an arbitrary poimof 2an-d U a. Then a connection r in P is reducible to a connection I” the horizontal subspace Q. AT]. X. the image of Q by the r = dim g’: We shall prove that S is d’iflerentiable and involutive. . By Proposition 7.T starting from ur.E Q for all t. . This is clear for ~(4 F a(4. a(u) = G. We note that if XI. $5) His a closed subgroup of G. To We first prove the following result of Ambrose and Singer [ 1] by prove the existence of T*.5 of The distribution S has dimension % ‘+ r.4. G.xt. BJ shows that every horizontal vector at u.. the linear differential equations satisfying a given initial condition kolonomy bundle through u of I’. for any curve 7 ~. Xz.. X. of Q with vr(uo) = x. 4 El)‘). . so is . be differentiable vector fields on CT which are 5 is the origin (i. . Proof. (with respect to I’). and by the subspace gi = (A:. [) l P x F. (cf? Proposition 5.u: P. it suffices to verify that the bracket a(xo).‘ It is so that a(xO) = . t the reduced subbundle of P(M.. 1 . We leave to the reader the proof that 7’. I-f. for each a c G. We shall prove that g’ = 6. zs connected and paracompact. Since the .. [A:.. XT]) E g. This A = cu([XF. is.A. we have ui E Q. ..@k’$orm section 0 of E defined on an open subset U of M is called S&(X.?) e g’. vertical component of [Xr.) . where A” is the fundamental vector field on P corresponding to A. ***Y A: the corresponding fundamental vector fields. The su‘bspace g’ is actually E(M. in other words. Let P(M. ‘then s&J) . A curve in E is horizontal if its tangent tangent to Q.Let Suppose I’ is reducible to a connection I” in Q.. X’. L e t g’ be the subspace . Since a is parallel.. G/H.e.h w ere 60 is.arbitrary hor&ntal vectors’. P starting from [AT. natural projection . . A. 0 5 t d 1.. G) corresponding to a (cf..

since component of C. We shall then extend F’ to a connection I’ in P 8/8x1 is horizontal at the points with x4 = 0. hplonomy group o&a certuin connection in a trivial bundle P = M x G. TV from Ozeki [l] and then in the general case by Nomizu [5]. that. the connected- ness . Since P is a disjoint for i = 2. . .)) is (t. o by requiring that Let F be a connection in P which coincides with l?’ on n-l(p) that the holonomy group of the symbol).g = g’. .(a. Let P(M.g’. 8) is clearly (0.O. . such that 0 < a1 < as < * * * < a. we can define a connection form connection in v-l(V) coincides with the identity component of G. where P similar computation to that for Proposition 3. G) be a principal jibre bundle. Y.s. consider the rectangle. is a suitable curve with c0 = e in G. . By 0 and are hence omitted. j-1 borhoods of x0 defined by lx”1 < a and lxil < fi respectively. where M is an arbitrurp.*w = ad (a-l)fw). 0) ~0 (0. The where 0 < /? < br. .with dim M 2 2. .) ‘ . of 7J that PI’U = ST. where c. + c. a. .. f $1 * E. 1 5 k _S 1. we s. end point of the lift T * of 7 is (0.l(U) is isomorphic with the trivial bundle U x G. a. there exists a con- nection in Psuch that all the holonomy bundles P(u).. for the moment.:..)) of T: is (t . 0 5 s I dam.O). (0. .E < t < that A. dim g = dim P . The lift . Let U and V be neigh. The above proof which is more direct is due to E.J). 0 5 s ‘9 ‘t.)J. be differentiable functions in -a . + 4 = ad (cy’)o((a/axl)).of P implies that P = P(uo).. . THEORY OF CONNECTIONS 9 1 [x. 0 d We shall construct a connection I” in P 1 U such that the ho. Any connected Lie group G can be realized as the (x. is in the Lie algebra of the holonomy group. c. Fixing t.l” t. exp ( .l * E. = ad (c. 0. . exp ( . . we see letfi(t).2. ..2 was proved for linear connections by Hano and segments 71 from (0. a. coordinate system with origin x0 = T(Q). . exp (-tA. .dz$erentiable mani$old.).1.n = dim . a. Theorem 2. < fi and reference point (0. 0 < t < #?.. Since this is the case for every t. be a basis for the Lie algebra g of G. a. on the x1x2-plane in V formed by the line Theorem 8. Therefore we have ci * c~’ = -A. lift 7. Taking a sufficiently small.) In r-‘(V) = V x G. e) of T: is of the Next we prove form (s.1). proving that Let A i. (Here and in the making use of Theorem 8. 0.. a.‘.3. e).tA. The This implies .) = 8.’ Proof.exp ( -tA. By a ~“HEOREM 8. since its tangent vectors 8/8x2 are horizontal.. This proves our assertion that S is in- volutive. coincide with P. .)f’c. .tA.l) ( i f.tAk) is an element of the holonomy group-of 7+(V) with numbers ai. This shows that the in such a way that I’ coincides with I” on P J r? (cf. (Kronecker’s being valid for any A. and ak. starting from (t. a) of U x G.)). + i. Finally. Ruh (unpublished).s. xn a local = ad (c. Th e 1i ft T: of 71 starting from (0. lonomy group of the bundle P 1 V coincides with the identity exp ( . a&. we have PO = P. the x9 to &coordinates of all the points remain Let P. Its tangent vector is of the form (a/&l). A. XT] belongs to S. both by (4 ak) to (t. QED. s.= U x C. If dim M B 2.2.O. be the maximal integral manifold of S through a..‘)A. we have is connected and M is paracompact. we may assume end point of is is hence (t.. 0. 0 5 5 5 t. On r-‘(U) . we determine the Lemma 2 in the proof of Theorem 7.. 0) to (0.90 FOUNDA-iIONS OF DIFFERENTIAL GEOMETRY II. Choose real exp ( . the preceding conditions determine the values of w at every point COROLLA&V 8. ‘~ union of hqlpnomy bundles each of which is open. u c P... exp ( -tAJ). . QE D . .1. ... ak) to (t.n = dim P. (Note that. e). The result a + E such thatfi(0) = 0 for every i andf*(a. . n. . s 5 ak. i = 1 .. Since the holonomy group @((uo) of l? obviously contains the j=l identity component of G.Tf. 0. of ~z starting from the end point (0. c. = exp ( -sAJ . e).’ by virtue of the property R.tAk)) . . . following argument. 72 frorn’(0.). There- horizontal lift of T = 74 * 71 * 72 * 71 starting from the point fore. the lift T: of 74 starting from the end point (t.. a.. the holonomy bundle P(uo) of l’ has the and that t -@me dimension as P and hence is open in P.3. 4iw’ha. is. Let u0 be an arbitrary point of P and 9. 0) and 74 from (t.

2.wd I? is .. Let I’ be a flat connection in P(M. a principal fibre buslbli over M with structure group @(u. ’ 9..f..) give r&Zx to the same element of @((uJ . an which is mapped into I’ by the homomorphism f: P’ -+ P.. the holonomy group of the induced connection iq b? M starting from x0 defines. u) = x’ and the homomorphism 7. then P is isomorphic with the trivial bundle M x G and F is isomorphic t with the canonicalfEat connection in M x G. 4 M-Letfi P’ -+ . G bf. .1 and since M* is con- U. G) such that the curvature vanishes identicalb. .that. an element of @(u. principal fibr& i br. M* is a covering space of M.-~(U) 4 ur-k G &Inch maps’the horizontal F’ isjut. 2 . By THEOREM 9.along. Assume that the curvatire fol~~. projection . THEORY OF CONNECTIONS 93 92 FOUNDATI-OX% OF DIFFERENTIAL GiOMETRY . The necessity is obvious.) ‘%j%t if and only. group is the holonomy group O(Q). The’ first statement ‘is ‘contained in Proposition .Gi~: . It is easy to verify that w is the connection form of the canonical Thus we o&&r a homomorphism of vr. The hol&nomy bundle through u.8 of Chapter I): jlat if every point x of M has a neighborhood U su& that. In other words. E P and M* = P( The connection isomorphism y: V. a) E M x G as the horizontal subspace at (D(uJ is discrete by Theorems 4. The projection the ‘identity only. A connection‘.6. u) = u so that the corresponding with the canonical flat connection in U x . For each “point i ‘of M.. G).::. . Sirre to M x {af at u 4 (x. Let N flat connection in P. a connection in Pis the canonical flat connection netted. QED. connected and paracompact.1. IYj) In particular.)/N./# consists of where p : M’. where M is a E G.) and set M’ = M*/N..5Fg of Chapter I) f form vanishes identically.1.%xeluiced ‘connection in P 1 U = 7r -l(U).the curv@ure form of F’ vanishes identically curvatureform vanishes identically.’ z& $3. Since the f: M x G ---f G be the natural projection and set rgtricted holonomy group is tnvral by Theorems 4. the set M x (a} is a submanifold of P. N bf. G) by the covering = -gf*e. M* is a principal fibre bundle over canonicalJEnt connection in P is defined by taking the tangent space M whose structure.?fthe the same pr~pos/tioq.+?’ is a covering space of M.). Let I’ be a connection in!P(M.&I . any co =f*e. .p’:is the subset of M’ x P defined as Proof. Applying the Reduction The&em (Theorem 37’: P’ -+ M’ ‘is @&Jx&~ 11)(x’.:G 5. l).~~:Pr~p~~i~o~. Proof. _ .We leca.:in P@& 1.. the M x {e] is a subbundle of P.. a . CC)].P be the natural homomorphism A connection in any principal fibre bundle P(M. $4 of Chapter I). where e is the identity of G. G) is called (cf. II. we see that the induced connection in .2..the induced connection in -P 1 I/ 7: n-l(U) is isoinbrphic with-1 $he PROPOSITION 9.mdle induced from P(M. . Set x0 = V(Q). Let parallel* displacement . ’ two closed cpryp.w. ‘S q&f:k~ the covering projection.flat-. For each connected. Flut connections We shall study the case where ‘M is not necessarily simply Let P = M j( G be a trivial principal fibre bundle. By Theorems‘4. here . . then the holonomy subspace at each u 4 n-‘(tl)‘u~~tf~~~~~~~zo~~~s~bspace at v(u) group a(&) of r’ with reference point tCb is isomorphically mapped onto of the canonictil flat:#‘I connection in U x G. Proposition’ 5. .. The Maurer-Cartan equation of 0 implies be a normal subgroup ef a)(~.2 and 8.3: There exists a unique connection l?’ in P’(M’.‘let X!J:ljff a simply connected open neighborhood ofx and consider tke’. To Prove: that f maps O(ub) isomorphically onto C OROLLARY 9 .). x. x0 e 111. Let’ir#f. G) canonical flat connection in cf X Gi More Precisely. In particular. if and only if it is reducible to a unique connection in M x {e}.f*e]= ‘--&IO. Then M’ is that the canonical flat connection has zero curvature: !. .. Every closed curve.. homomorphism . ‘ d o = d(f*B) =f*(dB) =f*(-@J.. If M is paratotipact and simply connected. Let P’(M’..2 and 8.the structure groups is the identity automorphism.(M. ..B . .is.. G) be the .v is isomorphic f:P’-+P is given by$( x’. x0) onto @(u.(. by means of the Let 0 be the canonical l-form on G -(cf. Ifub is a point of P’ such that f (ub) = u.s +t x0 representing !!n= same element of the first homotopy.

Consequently.-’ 1. Convcrscly.)/Xi$ the covering projection.) c M. dim Q*(u) :< ni> is xvhcre .e’ I. thrrc exists an element a E Q(Q) such that U. U) .uf..) E P’ c M’ x P.M(x. = Y(l4& properties : (1) O*(u) is a connected Lie subgrbup of G which is contained in the and. P(u. = ubb.6 for some -6 E N. for every open l whrrc L”. 1.2). The local holonomy groups haue the following 1’(z41N) == x. g(u) is spanned by all elements of a horizontal curve in n-l(U). Then Q”(u) is generated by all Q*(v).. Since each group O”(u. 4.(X * .II. Since . we denote by @O(u.* .) for such k. (5) For every integer m.. CTk = {x}.. V)for any connected open neighborhood V of x contained @(I(. bc a horizontal curve in P such that u1 = u. = (xi. If v E P(u). is a sequence ‘of connected open neighborhoods of x such that Since /J(.e . By Theorem 8. For every connected THEO.$ == v(u.t$ = ~(u~n).. Since. Let g(u) and g*(u) be the Lie algebras of W(U) open subset U of M.* g(u) is spanned by all g*(v) . Define a horizontal (3) If U is such a neighborhood of x = r(u). . consequently.) --) .. il is thcrcforc sufficient to prove @(uh) = N. For each u l T+(U). c 1. the form Q&X*. 0 5 t I=. let l’. it is defined to ‘be the intersection nV(u. 0 I 1. is horizontal in P and hence is O*(u) = @O(u. restricted holonomy group @O(u) .). 0 . t . open. :\I* = P(u. Write all connected open neighborhoods of the point x = r(u).open neighborhood V of n(v). U. let 6 bc any element of N. G). we have @*(ua) = ad (a-l) (Q*(u)) .) is cbnstant for sufficiently large k and hence that Then the cur\‘c ul.94 FOUNDATIONS OF DIFFEtiENTIAL GEOMETRY II.. QED. For each t. the set {n(u) E . THEORY OF CONNECTIONS 95 ay. we set P-l connected Lie subgroup of G (Theorem 4.). Locnl.where U runs through contained in g*(v).. be the connection in P 1 U = &(‘U) induced from r.v. If (!ij rt. in U. Q”(u. cd injin. v E P(U). = Y(U”U).-: v(u. Let ut. vectors at v.This. Y*) where v E P(u) and X* and Y* arc horizontal. then we have obviously (I)O(u.‘. = (A$.’ then @O(u) = (P”(c) 13 (D*(v) and g(u) = the holonomy bundle through u of the connection I’. U). U) the restricted-holonomy group with reference point u and Proof. shows that a*(u) = W(u. We have PROPOSITION 10.a). 4 = (4. /hen Q*(u) 2 Q*(v) curve uj.H’ = P(u.) 3 . U) and and O*(u) respectively. (2) Everypoint x = n(u) h as a connected open neiCghborhood U such that which means that ~4~ = u.2.1. . x.) -= v’(II. where Af is connected and paracompact.3 @O(U..REM 10. Then (5) means that this integer valued function is upper Let I’ be a connection in a principal fibre bundle P(M. Z’(Z4.1. U. r7. semicontinuous. (4) For every a E G.) is a 77’ (16. and g(u) &spanned by all g*(v). there exists an integer k such that l. we remark that dim Q*(u) is constant on each fibre of 10... be a horizontal curve in P’ such that we have Q*(u) =: fi @‘O(u.6. neighborhood U of x.. P(u. I/) consists of points v of v-l(U) which can be joined to u by g(v) 1 g*(v). V) for every connected . it follows that 16.U. = n’(uL) = n’(uJ = xl.. respectively.). \ve ha1. 0 < t . 13 Ok+l and . As to (5).\I* = P(u. U.) = I = p 0 V(UJ and obvious. U. showing that 6 E @(uh). dim @“(u.‘.:) --: n(u.n) for 0 . v l P(u).J.. T t c 1. Since Q. The following proposition is now contained in . U. in P’ by for evey v f P(u.) =I P=l x. 24”) E P’ c M’ x P .) = V(U”).itesinyl holonomy groups P by (4) and thus can be considered as an integer valtied function on M..) 2 . Then 26 . LCt 7 :. Y*) is contained in the Lie algebra of The local holonomy group Q*(u) with reference point u of 1’ is @!(u. 14..

Proof. We show that m,(u) c g*(uJ by induction on k. The
case k = 0 is obvious. Assume that mk-i(u) c g*(u) for every
point u. It is sufficient to show that, for every horizontal vector
where v E P(u). The first assertion now‘
I.. / follows. from the following field X and for every function f of the form (I+J, we have
lemma. X,,f E g*(u). Let ul, ItI < E for some E > 0, be the integral curve
LE M M A . If the Lie algebra g af d connected .Lie gSujj G is generated of X with u. = u. Since vt is horizontal, we have I* c q*(u)
by a farnib of subspaces {mA}, then every elevent,.of G can be written as a by (3). of Proposition 10.1. Therefore,f (uJ E m,..,ju,) c :I*(&) c
product exp X, - exp X, . . ** * exp ~Xb, where each Xi is contained
in some rnA.
Proof of Lemma. The set’ H of all elements of G of the above
form is clearly a subgroup which is arcwise, connected ; indeed,
every element of H can be joined to the identity by a differennable
curve which lies in H. By the theorem of Freudenthal-Kuranishi-
Yamabe (proved in Appendix 4), H is a connected Lie subgroup L E M M A 1 : Let-f be a g-valued function of ppe ad G 011 P. Then
of G. Its Lie algebra contains all ml and thus coincides with g. -(I) ForayvectorjeZdXonP, wehavev(X). *f = --[wll(X),f (xi],
QED. where v (X) denotes the vertical component of X.
Hence, H = G. .-
(2) For any horizontal vector&lds X and Y on P. we hazle
THEOREM 10.3. If dim @F(u) is constant on P, then Q”(u) =
O*(u) for every u in P. ax9 nJ *f = YxL(K Y),f(u)l.
By (3) of Proposition 10.1, x = rr(u) has an open
P ROOF . (3) If X and Y are vector fields on P which are invariant by all .R,,
neighborhood U such that Q*(u) 2 O*(v) for each u m P(u, U). a E G, then Q(X, Y) and Xf are functions of [up ad G.
Since dim @*(a) = dim Q*(v), we have Q*(u) = O*(v). Bv the Proof of Lemma 1. (1) Let rl. = oU(X) E g and a, = exp t-4.
standard continuation argument, we see that,. d v E P(u),-- Then
CD*(u) = O*(v). By Theorem 10.2, we have * w(!d) = Q*(u).
QED. v(X), *f = AEf = 1’,ty f [f bat)
- f (41
We now define the infinitesimal holonomy group at each point
u 0f.P by means of the curvature form and study its relationship = I,i~f [ad (a;*)(f’(u)) -f(u)]
to the.l&al holonomy group. We first define a series of subspaces
m,(u) of g by induction on k. Let ma(u) be the subspace of g = -r4fy41 = -r%(x~,f(~)l.
spanned by all elements of the form 0,(X, Y), where .X and Y are (2) By virtue of the structure equation (Theorem 5.2), we have
horizontal vectors at u. We consider a g-valued function f on P of
the form 2&(X, Y ) = 2(dw).(X, I’)
(IA ._ f = Vk.. * wxx, WY = x&J(y)) - Y,c(~fJ(X)) - ~~J,t([~~, y])
where Xi Y, V,, . . . , V, are arbitrary horizontal vector fields on = -wu([X, Y]).
P. Let m,(u) be the subspace of g spanned by mk-i(u) and,by Replacing X by [X, Y] in (l), we obtain (2).
the values at u of all functions f of the form (I,J. We then set g (u) (3) Since R is of type ad G (cf. $5 of Chapter II), we have
to be the union of all mk(u), k = 0, 1, 2, . . . .
PROPOSITION 10.4. The subspace g’(u) of g is a subalgebra of g* (u!. fi,,(R,X, R,Y) = ad (a-‘) (12,(X, Y)),
The connected Lie subgroup a’(u) of G generated by g’(u) 1s
called the infinitesimal holonomy group at u.


which shows that n(X, Y) is of type ad G, if X = R,X and Y = where h[X, Y] denotes the horizontal component of [X; Y]. The
R,Y. We have also functions X( Yg) and Y(Xg) are of {he form (I,,,) and the function
h( [X, Y])g is of the form (Ik+J. This proves our assertion for s = 0
WL = Xuaf = VVulf = Uf 0 RJ and for an arbitrary k.
= ad (a-‘)(X,f) = ad.(a-l)(Xf),, Suppose now that our assertion holds for s - 1 and every k.
iff is of type ad G and X is invariant by R,. This completes the Every function of the form (1J can be written as Xf, where f is a
proof of Lemma 1. function of the form (IS-J and X is a horizontal vector field. Let
g be an arbitrary function of the form (I&). Then
Let Xi = a/W, where x1, . . . , x” is a local coordinate system. in
a neighborhood U of x = r(u). Let XT be the horizontal lift of Xi. E&f, &)l = urf, 91) - rf (4, X”‘a
Consider a g-valued function f of the form The function [f, Xg] is a linear, combination of functions of the
f=XZ.. - q(Q(R, X?)); form (I,), r 5 k + s + 1, by the inductive assumption. The
(I;r,) function X[f, g] is a linear combination of functions of the form
where i, 1, jl,, . . . , j, are taken fre$y from
‘C 1II . . . , n. (IJ,‘r I s’+‘k “f 2, aho by th e inductive dssurhption,‘ Thus, the
is a l&&r combination of func$ons -%the form
LEMMA 2. ‘, For each, k, m,(L) {s spanned 6y mkel(u) and by the
values at u of btl functions f of the form (II,). 5 QED.
Proof of Lemma 2. The proof is by induction on k. The case PROPOSITION 10.5. The injnitesimal holonom3 groups have the
k = 0 is obvious. Every horizontal vector field in r-‘(U) is a linear following properties : ’
combination of XT, . . . , X: with real valued functions as (1) W(u) is a connected Lie subgroup of the local .,holonomy group
coefficients. It follows that every function f of the form (Ik) is a @‘*(u>;
linear combination of functions of the form (II,), s g k, with (2) W(ua) = ad (a-‘)(@‘(~)) and g’(ua) = ad (a-‘)($(tf));
real valued functions as coefficients, in a neighborhood of u. It is (3) For each.integer m, the set (r(u) E M; dim W(u) I m} is open;
now clear that, if the assertion holds for k - 1, it holds for k. (4j If W(u) = a*(u) at a point u, then there exists a connected open
i$‘e now prove that g’(u) is a subalgebra of g by establishing the neighborhood U of x = r(u) such that W(v) = <D*(v) = W(u) =
relation [m,(u), m,(u)] c n!k+,,Z(u) for all pairs of integers k and cP* (u) for every v E P(u, U) .
s. In view of Lemma -2, it is su,fficient to prove that, for every func- Proof. (1) is evident from Proposition 10.4. (2) follows from
tion f of the form (I,) and every function g of the form (II,), the
LEMMA' 1. For each k, we have mk(ua) = ad (u-l) (m,(u));
function [f ,g](u) = [f(u), g(u)] is a linear combination of
Proof of Lemma 1. The proof is by induction on k. The c&e,
functions of the form (I,), r I k 1. s + 2, with real valued func-
k = 0 is a consequence of the fact that 0 is of type ad G. Suppose
tions as coefficients. The proof is by induction on 5.
the assertion holds for k - 1. By (3) of Lemma 1 for Proposition
Let s = 0 and let f (u) = fi,,(X, Y), where Xand Y are Kbri- 10.4, every function of the form (I1J is of type ad G. Our lemma
zontal vector fields.. Since g is of type ad G, we have, by (2) of now follovirs from Lemma 2 for Proposition 10.4.
Lemma 1,
(2) means that W(u) can be considered as a function on M. (3)’
“[QI(X, yj, !+)I = v([X Yl)u *if., is a consequence of the fact that, if the values’of a finite number
On the other hand, we have of functions of the form (i,) are linearly independent at a point u,
then they are linearly independent at every point of a neighbor-
u([X, Y]),, . g = [x, Kit, . g - . N-K Yl). . s hcod of u. Note that (3) means that dim Q’(u), Considered as a
= x,,(Yg) - LVi4 - 4X, Yl)u . g> function on M, is lower semicontinuous. To p&e (4), assume

Q’(U) = (I)*(U) at a point 21. Since dim a’(u) is lower semicon-
tinuous and dim O*(uj is upper semicontinuous [cf. (5) of GL(s; R) such that
Propos.ition 10.11, the point x = T(U) has a neighborhood Ii such daij/dt = Xi=1 AikU, and ~z~j(O) = 6ii.
(Note that (Alj(t)) E gl(s; R) cofiesponds to Y, E T,(G) in the
dim O’(v) 2 dim (P’(U) and dim’@*(v) Z< dim @p*(u)
lemma for Proposition 3.1.) Lets (bij(t)) be the inverse matrix of
for L’ E r-l(U). (atj(t)) so that
dbijldt = -X,3-1 bikakj..
On the other hand, O*(u) 2 O’(u) for every‘; E r-l( I/‘). Hence, Then
dim (D*(z)) = dim a’(74 = dim Q*(U) = dim a’(u)
and, consequently, O*(U) = Q’(Z) for every 2 E n-l(I.rj. Applying
Theorem 10.3 to P 1 li, we see that @f,o(u, U) = (D*(U) and
V(v, U) = Q*(c). If 2 6 P(u, r-j, then (l)O(u, U) = iDO(i,, I’) so
QED. Since bij(0) = Bij, we have
that Q*(U) .= @*(zl). l
Ejzl b,,(t)g,(t) = g,(O).
'rHEORE.Zl 10.6. rf c 1im cl)‘j~,‘s i.! constant in a neighborhood of u in P,
thm W(u) = a)*(u). This means that g’(uJ = g’( u ) and, in particular, g’(y) = g’(u).
Proof. We first prove the existence of an open neighborhood Taking CT sufficiently small, we may assume that
lJ of x =- +u) such that ~‘:II: := !I>‘(“) for every c c P(u, U). Let for every v E P(u, U).
g*(u) = Y*(4 = g’(4 = m,(v)
fi> * - . ,l* be a finite number of funciions of the form (II,) such By ‘jTheore!m 8.1, the Lie algebra of @O(u, U) is spanned by all
that f; (U), . . . ,J,(u) form a basis of g’(z4). At r\rery point u of a
small neighborhood of u,fi(c), . . . ? fs( z*) arc linearly independent m,(v), v E P(u, U). A fortiori, g*(u) is spanned by all g’(u),
and, by the assumption, they form a basis of g’(uj. Sincef,, . . . , fs v E P(u, U). Since--g’(v) = g’(u) for every v E P(u, Z;) as we have
are of type ad G, fi(~~o), . . . ,f;(~.‘a) fbrm a basis of g’(va) = j;s; ;hown, we may conclude that g*(u) = g’(u) and @*k@;
ad (~l)(g’(~)). This means that there exists a neighborhood u of ‘U.

x := Z$U) such thatfi(ll), . . . J*(U) form a basis of g’(u) for every C OROLLARY 1 0 . 7 . Ifd im CD’(u) is constant on P, then @O(u) =
point u Q ,r;l(U). Now, let v be an arbitrary point of P(u, c’) and a*(u) = W(u).
let Us, 0 ~2, t 2 1, be a horizontal curbe Tom u to v in r-‘(U) so Proof. This follows from Theorems 10.3 and 10.6. QED.
that u = u; and v = 14~. We may assume that uf is differentiable;
THEOREM 10.8. For a real analytic connection in a real analytic
the case where U, is piecewise differentiable follows easily. Set
~~.jk+ jibre bundle P, we have @O(u) = O*(u) = W(u) for every
~~~a;eJ;(&),~ i = 1, . . . , J, and X = zi,. Since X is horizontal,
Proof. We may assume that P = P(u) and, in particular, P
(6ildt) f * (LK) (‘1) ’ Cl’(‘t)> i = l,...,s.
is connected. It suffices to show that dim a’(u) is locally constant;
Sinceg,(t), . . . , g,<(t) form a basis for 9’(u1), dgJdt can be expressed it then follows that dim a’(u) is constant on P and, by Corollary
by lO.$y that @O(u) = (D*(u) =. CD’(u) for every u c P. Let 9, . . . , xn
(dg,/dt), = q_, ;ljj(t)gj(t), be a real analytic local coordinate system with origin x = n(u).
Let U be a coordinate neighborhood of x given by xi (xi)2 < a2
where &(t) are continuous functions of t. By the lemma for
for some a > 0. We want to show that dim Q’(u) is constant on
Proposition 3.1, there exists a unique curve (aij(t))i,i=l, . . .,$ in n-l(U). Let Xi = a/Jxi and let X: be the horizontal lift of Xi.

qf G (cf $4 of Chapter I) with respect to the decomposition g = @ + m the second power series shows that g’(uJ is contained in g’(uJ. .. .) 11.* * 3 X. Each Xi is of Consider all the functionsf of the form (II. The results in this section are due to Ozeki [l]. 2. j.).)“h(“yt. = (t. h(t) = x. # 0. and fi are real analytic functions are analytic functions oft. f zxj:.(x*f) and so on. then we can write h’(t) = Xu:f. #‘O. . an) with Ci (U~)~ = 1. Invariant connections . k) which is invariant by the left This means that g’(uJ = g’(ut. . h(t) = f (xt). . 6 common -to all h(t) follows from the lemma we prove below. 102 FOIJIQ-JATIGNS OF DIFFERENTIAL GEOMETRY II. .“4J $ (t .fir * a/&i. (1) The proof is straightforward and is similar to that of -common neighborhood ItI < 6 asfollows: h(t) = 8. * qwm 49) they can be expanded into power series of (xl. . .Y z m. . The fact that there exists such a group.1. y> = -Hi q (b-component of -*TX. We skt h(t) = f (uJ. consider Proof. Proposition 6:4. . Let g and b be the Lie algebras of G and H respectively. translations of G (ifit exists) determines such a decomposition g = t. in a neighborhood of x. QED. If X* is the horizontal lift of X. we present an important special case.t..d (4 m. .tph’yt). . . . dejnes a connection in the bundle G(G/H.] < 6 in the Taylor series: expanded into power series in the same neighborhood.h@.“=o$ h(“‘)(O). H) invariant by the left ItI < a. THEOREM 11.( < 6. . (4) Let g(e) . x”) in a common neighborhood ]xi[ < a for some a > 0. xn in a 6 > 0 such that all the functions h(t) can be expanded in a common neigh“h‘orhood [xi] < a. with ]ts] < a. 6 > 0 such that the functions h(t) can be expanded into power series in a Then g(e) is spanned by all elements of the form [X. where X.. For each t. Our lemma then defined on +-i(U). The first and ad (H) m = m. Let G be a connected ‘Lie group and H a closed sub- h”(t)= x:.(G). . Proof. . s. . .). . there exists whit can be expanded into power series of x1. Since f and fij are all real analytic. xn such that X = a/ax1 and x. then the functions f& and afi/axi can be neighborhood It . . (1) If there exists a subspace m of g such that g = IJ + m (direct sum) Now. xl. then the t)-component w of the canonical l-form 6 power series shows that g’(uJ is contained in g’(yJ. + m L E M M A . of G. In a real analytic mantfold.) for It .) = g’(u) for every t with (2) Conversely. 0).--. * * Xi. a jinite number of real analytic vector fields is given by Xl. if It ..toI < 8. proving our theorem. Under the identification g m T. X. continuation argument shows that g’(u. . Then the functions h(t) follows from the fact that iff.t. the form Xi = Z. Ylh. are taken free& from 1. . . . . Then there exists reference point e (identity element) of the invariant connection dejned in ( 1). . Y] E g). we may take a local coordinate system the vector field x = Ci aiXi on U. . then all h(“)(t) belong to g’(ut. .. For any real (3) The curvature form R of the invariant connection dejned by w in ( 1) ’ analytic function g and for. Since X. 0. real analytic vectorjeld X such that x0 = x. consider all the functions of the form Q(X. The preceding expansions of h(t) are nothing We prove that g’(u) = g’(uJ for every t with It] < a. but the expansions bff (x) into power series of x1.) = z. THEORY OF CONNECTIONS 103 For any set of numbers (a’. any connection in G(G/H. the sub- space m corresponds to the horizontal subspace at e. where X and Y are arbitrary left invariant vector jelds on G belonging to f (4 = (-q *. k 2 the L ie algebra of the holonomy group Q(e) with where j. Similarly. Before we treat general invariant connections. Let xt be the ray given by xi(t) = a’t and let ut be the horizontal lift of xt such that u = uO. The standard translations of G. let xt be the integral curve of a and is obtainable in the manner described in (1)..-& (to . .t.

called the isotropy subgroup of K at x. morphism. Proof. of P. we obtain . g1 c g(e) c g. pt(u.j)%(j’)).) of 31 forms a X.i(j))%(. Its e-corn. is horizontal.3. of all elements of K which fix the point x0 = n(u.( pl.. Let v be the form the same 1. Let P = (G/H) x G be the trivial bundle over G/H with = (u..(X) for Xr j. we have r(~l~(u”)) reduces to a single point x0 == VT(U. wd h a v e define a homomorphism R: J + G 3s follows.) a n d h e n c e ~(2~) = w(X.a with some a E G. G) and X the vectorjeld on P induced by ( 1 ) . = j(uo%(j’)) = (ju. For an arbitrary point u. Let m be the set of left invariant vector field? on G such that o>(X) = 0. of K generated by It is easy to verify that..‘&.. and. . .j’) = z+(%(. Hence \VC’ h a v e am = uoi.iissz. Thus we obtain a.(X. = (2) Let w be a connection form on G(G/H. ponent. is a point in the same fibre as u.. is chosen once for all and is fixed throughout this section. Let qt be a I-parameter group of automorphisms where A7 is the vectorjeld on P induced 4)) X.1. s i n c e the connection form q is invariant.(j)) is that ut = Yt(Uo). Now. We imbed the bundle G(G/H.2 to the l-parameter ” Then n. left translations of G. .(X).’ K induces a transformation of . point u.. YE m): Let g2 b e t h e subspace of g s p a n n e d b y t h e s e t closed subgroup of K. is diffeyenti. then QED. H) into P by the mapping Hence. P R O P O S I T I O N 11. We ($2.) = a. For each j . ju. Let K be a group of automorphisms of P(. xt.\f. of Gin g. . we have w(ti.Y) .) a n d uoil(y. If j. (3) follows from Corollary 5.) a t uo. H) invariant by the A. where o is the connection form of F. Y E m and u E G}. (2) h(ad (j)(X)) = ad (n(j)) (A(X)) Jbrj E J and X E 1.).\I in a natural manner. QED. Now. Xt = +4.4. As in the proof of Proposition 3. be an arbitrary (3) A left invariant vector field is horizontal if and only if it is point of P which we choose as a reference point. vl. (1) can be considered as a particular case of Proposi- -tion 6. Note that the geometric meaning of A(J7) is gi\cn by Proposition ut = vtat. vt = the horizontal lift bf xt such that vO = uo. G) tion (Proposition 6. then the curve xt = Proof. vt and a. Since v. j’ c J. Comparing the tangent vectors of the orbits co(&) = ( a d (a~pl))o(tit) + a.Z(X) = n(X). (jj’) u. YE m and u E G. It is also easy to check that 1. 11.Z J. H). Let I3 be a connection in P invariant by qt.1(X) = i. subgroup y. we dejine curves ut.g = b + m is Let K be a Lie group acting on a principal fibre bundle a desired decompositipn. as follows: c where ad (j) is the adjoint representation of J in f and ad (i. Note that 1 depends on the choice of u.)%( j’) Remark. 1 1 . If XE j. (4) follows from (3). we have g 1 = g. A(jj’) = A(j) which shows that 1: J -+ G is a homo- f defined by UEF. uoKij’i =. (1) We apply Proposition 11. $9) of 9. = qt(XIt. Y ) = ju. = u. By The&em ’ 8. THEORY OF CONNECTIONS 105 104 F O U N D A TIONS OF DIFFERE N T I AL G E O M E T RY we h a v e 8. is the l-parameter subgroup of G generated by A. We define %(jj = n. we first prove the’ following proposition which is basic in many applications. The set J (4) ‘Let g1 be the subspace of g spanned by the set {n. defines a connec. PROPOSITION 11. P(M: G) as a group of automorphisms.2..‘e deJine a linear rnn/)~ing Going back to the general case.. Let u.= w. X. and *hence is of the form S&(X. II.‘&.).4) and agrees with the form o in (1). The f(u) = b+)> 43 induced Lie algebra homotiorphism i -* g will be also denoted by where n: G + G/H is the natural projektion.. On the other hand. 2 .‘the reference defining the canonical flat connection (cf. On the oth&r hand. Every element of an element of nt.J = A . group G.(Y.(cp.1. I‘ a ‘connection in P invariant by K.3. restricted to the subbundle G(G/H. Y) for any X.‘ci. Y) . Then of a principalfibre bundle P(A4.

0 Rh. A(X) + A = A(Y) + B. curvature form R of F satisfies the folloving condition: there is a K-invariant connection form o) on P such that A(X) = wU.5 of Chapter I) 0(X*.. 81) + [w(X). Then there is an element of K which maps one fibre into the other. we 2f&(X.(k * R. can write Proof.7 of Chapter I.k-’ E J. U O = kua = k. xv.106 FOUNDATIONS OF DIFFERENTIAL GEOMETRY I.X*) =j c R.eK a n d a.u Proposition 3. j-1 which maps u.Y) = A(X) . It follows that yU’u.Cj-I. We then set Since m is invariant by K. Let X* E T..01(j-1~) = ZUo. By Proposition 1. i . Yl). Y]) for X. THEORY OF CONNECTIONS 107 (2) Let X l f and j E J. it is suffkient to show that.u. . Xc k and A* is the fundamental vector field corresponding to A l g. NV1 . ci(Fji..) = ad (a)(A(X) + A)..2 of Chapter I (cf.A:.if&r. G) and . X -+ X being a Lie algebra homomorphism.11 of Chapter I.O .) = invariant connections in P and the set of linear mappings A: t -+ g idk&. From the structure equation (Theorem 5.3.oJ([F. we have THEOREM 11.3. + “l?o = r.X* = x.. we have n(X . = n(uO) as above.P)) = j&i-p. The following result is due to Wang [l].Y B f.. In view of Proposition 11. Then Ygenerates K at x0 = I. where k E K.(~"O)= j(8. if the action of K on the base M is transitive. + R.3. we have by (c) of Proposition 3. where 2 = ad (j)(X).j-~Juo) 1 = ~j-~uo(&~j-~Juo) where 8 is the vector field on P induced by X.A(Y).= u. We next prove that w(X*) is independent of the choice of k = [NW.4x f7) = k4m = 0. Hence. j-l(u.2) and uo = kua = k 0 R. OJ( X*) is independent of the choice of X and &J(P)) . (k. we have is given by NW = +.l = &o(F)) 1 P(iJ(X)j . From the definition of 1: i . If a connected Lie group K is a Jibre-transitive auto- morphism group of a bundle P( ‘$1. muo( 0 = the k. Let QED.h([X. we obtain k 0 R. On the other hand.X* L jk 0 R~~(j-I. w h e r e YE1 a n d UEg.4. P) = [A(X).o(j(R.(X.k-‘. the correspondence Since the connection form w is invariant by j. then there is a‘ 1 : 1 correspondence between the set of g- the I-parameter subgroup jv.5. We have that is. QED. and a.g.f J is the isotropy subgroup of j ’ R~cj-l.p.Y) = B -~ A. = j? intojq. We say that K actsJibre-transitively on P if. for any two fibres of P.). We first prove that. With the notation of Proposition 11.. A(j) = !a. which satisfies ti two conditions in Proposition 11.(s) for Xc f.).NV. also Proposition 3. If J is the isotropy subgroup of K at x..‘a. By condition (1) of Proposi- tion 11.3.a..j.(P).‘a E G. We set Y = ad (j)(X). We set j = k. A(Y)] .I4). Proof.‘a and k..O + A:. we have so that x.X* homogeneous space Kl J. = j(Ryj-&. Let 27(oJ(X)) . Since K is fibre-transitive.O = &. + A& 2QM 9) = 2dw(Z 9) +. it follows that J(X . 81) = (Lpw)(X) = 0.k-‘u. PROPOSITION 11. A. then M.Y) = A(X . the for every A: f + g satisfying (1) and (2) of Proposition 11.(& for X E t. SO that ‘k.t-G). 4n..

Then k. QED. let A be any element of g and let z+.4.X*) = k 0 R. A connection is flat if and only if its curvature form vanishes identically (Theorem 9. = k. u. We decompose the (zj The curvatureform R of the K-invariant connection dejined by A.4 and Theorem 11. we set k o R.) = B& where B = ad (a-‘) (A). Y). we see immediately that o(k. we define k E K and u.X*) = w(Y*) = ad (b-%z)(il(X) + Aj = ad (b-l)(.u)a.(X) ifXe m. For .X” = (XC!.Y E f.a.r G by if Xc j. As. J).: m --+ g such that To prove that o is differentiable. YIJ for X.(X*)j.A.5 that f admits a subspuce Hence we have m such that f = i + m (direct sum) and ad (J)(m) = m. THEOREM 11. fratifandonbifA:f -+ g is a Lie algebra homomorphism. a) in terms of A. THEORY QF CONNECTIONS 109 arbitrary X* l T. the curvature form so that u.) such that u(n(uJ) = k.*) = R. .X*) = aI(Xi).l 1. = kub(b-la) = kv(b-la). Let b be an arbitrary element of G. Let X* f T. Al = K/J.) . $.(k. = kua.(X*) = x. Y * = R.4 expresses S&(X.6. vector space f into a direct sum of subspaces: f = j + m. Let 0: U. Proposition 11. Then both X and A are uniquely determined and depend dif- ad (flJ(A(Z) -t C) = ad (ar)(A(ad (j)(X)) -1 ad (I. 0 k(A:) = R.1..(P) and = ad (a) (A(X) + A)..K be a local cross section of this bur$le the correspondence is give! via Theorem .(P) and z10 = kua as above.-I.‘(k.4[X. which is a tensorial form of type ad G (cf. we obtain o)(R. We now prove that w is a connection form. Then (1) There is a 1 : 1 correspondence between the set of K-invariant which shows that o satisfies condition (a’) of Prdposition 1... We then have fi.(j))A) ferentiably o n X*.(X*)..(P) and uO = kk. Then Proof. we prove that o is invariant by K.(A.. each u 6 n-‘(U).1. Then both k and a depend differentiably on u. This proves our assertion that w(X*) depends only on X*.7. A(X) = A(X) k = a(+)) and u. = kua.. conrte&ns in P and the Set of linear mappings A.(A.X* E T. For .. = kup. II. P) = PM-V. $5) and i.Y* = k 0 R. We set From the construction of w.~ invariant k 0 R. QED. where v = ub..5 by defined in a neighborhood U of T(u. is completely determined by the values sZUO(x. + A$ where XE m.... a con- and hence sequence of Proposition 11. ad\(J) is the adjoint representation of J iti f. Now.X* = k 0 satisfies the following condition : 2QUO(R. = ad (&)[ad (I(j))(R(X) + A)] Finally. where Mad (i)(X)) = ad (W)(A. C OROLLARY 11. kkrl o R.-I. Thus 0(X*) = ad (a)(A(x) + A) depends differentiably on X*. Assume in Theorem 11. The K-invariant connection in P de&zed by h is which shows that w satisfies condition (b’) of Proposition 1. let ur be an arbitrary point of P and let u.*.([K Yl.. Hence. Consider the fibre bundle K(M.R.(P).**) by K. X. In the case where K is fibre-transitive on P. Ye m. Let X* E T.(X)) forX E m a n d j d J . be an arbitrary element of K. L(Y)1 . k c R.rX* E TkI..(P). where w(Az) = ad (a)(B) = A..u.1). i R..5. Let k.

W. 2. Repeating this process. it is sufficient to prove that mk = mk(uo) for k = 0. we have * QED. . moll + * . . the Lie algebra g(uJ of the holonomy group a(~+.. .). connection in Theorem 11. also Proposition 3. + [A(f).(a4 * * * ww WI) mod mAuoh Theorem 11. Y.) of the K-invariant connection dejned by A: 1 --+ g is given by since [8. W] are horizontal by Lemma 1. . 1. of g. Let Xi = a/&i.with the infinitesimal holonomy group X(cqv.(Yr . Yl E f.1 is the canonical connection just Then defined. mol + CW [A(f). coincides with ntO(u. * ** w-w W)H where m. We defined in $10 an increasing sequence of subspaces The two terms on the right hand side belong to m. Y. X. . ha). Let XT be the horizontal lift of - By Proposition 11. Since f&(2. . We need the following lemmas.A. 1.). be the restriction of A to m. (1) and (3) of Theorem 11. This completes the mk(uo) is the Lie algebra g’(uo) of the infinitesimal holonomy proof of Lemma 2.1 follow from Theorem L EMMA 2. We define a series of sub. By (c) of Proposition 3. where x1.) as [x. With ihe same assumptions and notation as in = (Yr)u. k = 0.. where [X.7. x.2 of Chapter I (2) is a consequence of Proposition 11.fw *. W)) = (LRR)(V. Y].. . where h. Since. By the same argument as in the proof of Lemma 1. we have Proof. the K-invariant connection in p defined by a4YH = WP)(Y) + 4x w A.(r. Remark. x. Since K is fibre-transitivr: on P. m. IfY is a horizontal vector field on P and 8 is the vector of Proposition 11. to see that A . . It is easy jield on P induced by an element X off. Let A. mol. + [A(f).NC& YI). 9) . is t9te subspace of g spanned by mod m. Since L&J = 0. 1. II. Let V. we obtain m.). Y.’ ww W)) THEOREM 11. mol + bV]i: P(f).. . -. + [A(f). W) + a&K fq. (resp. . Y c f}.(!Z!( V. the restricted holonomy group @O(uo) coincides . gives a desired corrtspondence. We . 2. Y]J denotes the m-component (resp.< y. P&of.. = 0 is called the canonical connection -(with respect to the Since o(Y) = 0 and Law = 0. .y and hy . . V] m.the union of these subspaces and [8... x. .4. [X. In Theorem 11.3. A(Y)] . tie have o([x. 2. we have spaces mk. -Lkt i: f -+ g be a linear mapping satisfying ( 1) and (2) L EMMA 1. j-com- ponent) ofrX. The statement Proof of Lemma 1.(u0)- Finally.(u.).7.5 of Chapter I). the subspace m. m2 = m. Ye f!. group @‘(u. W ) + Q(K Emw @‘(uo) by virtue of Corollary 10.(Ll ** * um(K 3)) {CA(X).(u. . we determine the Lie algebra of the holonomy group Proof of Lemma 2.l+. G) = G(GfH. Y]) = 0. y) = !&(/&. . of g as follows: uG’)uO(L1 * ** uw(K W)) ml = m. H) and K = G. = Wu. THEORY OF CONNECTIONS Ill 110 FOUNDATIONS OF DIFFERENTIAL GEOMETRY denote the hdrizontal components of x and p respectively. s = (yJ”o(L1 * YlW[X J% J. x” is a local coordinate system in a neighborhood of x0 = n(u. decomposition t = j + m). %.(u. is’spanned by {fi. then [x. . the invariant jields on P and let 8 be the vectorjeld on P induced by an element X of t.). Y?] is horizontal by Lemma 1 ‘and [x.( Yrel * * * Y.&x.7 if we set P(M.8. .(u.4. (cf. W))) is in m. be arbitrary horizontal vector 11. k = 0. . . *: * Yl&W2 W)> E m. . moJl + mJJ:-1 * * * w4~~ [X WIN* and so on. 9-j is horizontal.).5.have of a: K-invariant connixtion.

-. = m.112 FOUNDATIONS OF DIFBERENTIAL GEOMETRY Xi.-. .7). If 7 = Xt.)’ fdr all r i s.).(u. Let LciL’i?._/ defined by .. c i n4. mod ms-&J.natural Our assertion m.. is spanned : 6e.) and by [A(f). G) be-a principal fibre bundle and p a representation Remark.(u.. X.. ms-J.E. = m.fibre T.f = bw%f 0%>1 :mod nt. = tn.I .(v. we have Linear and Affine C&xnections vau. where X E f and f is a function of the form (Us-J. P) be the associated bbndle (cf. if a = (a. x:)) be a function of the form (II?) as defined in $10. Let f=Xp * q(qxt. We call E a real or complex uector bundle over M according as F = R or F = C.. (4) of Theorem 11.) and the s&t Ofall GL(m. The group CL(m.) is 1. we shall show that m.Em) 6 F’“. QED.2( over F (of infinite dimensions. = m.In field C. On the other hand. from F. = m.m. of: E has the structure of a Vector space such. . it forms a vector space . with standard fibre F’” on which G azts through p.(UX)“of + -Qf = ~c%o(8).. m.(u.! :( Q~F be the set of cross sections 9: hl . then Lemma 1 for Proposition 10. if I is an F-valued function on M.) follows from the congruence (hX)..R or the complex number by m. then at = (Cj (zjllj. is a function of the form (II. every horizontal vector at u.‘Y * . Since K is fibre-transitive on P..= [A(f). if m 2 1) with addition and scalar ’ 7 multiplication p . F”. T(U). Let P(M. is spanned by m.o.[“Y. We may also consider S as a module over the algebra of F-valued functions. F) and 5 = (El.y(x).-. F).. F) the group ofall m x m non-singular matrices with entries [A(X). . a 5 t s b. u%d TJf:. considered as a mapping of ~. DW’).m.f bJ1 + x4J Since xuof E m. where X E f and j 1.G.. q E s. is a curve in M and T* = uf is a horizontal 113 .1 is a corollary to Theorem 11. = s. .) by Lemma 2 ar$ ~ih&‘...) E GL( manner.-.~(x).). f (z&J]. Cj aytj) E F”...(u.E M.-..f. Let E(M. .$Jf~ onto 7f E!(X).. Remark made after the proof of Theorem 11.(u.(u. m.t F be either.-.-. is a liqear isytiorphism of F”’ onto nil(x). is of the form (hX)+ for some Xc f. F’” the vector space of all m-tuples of elements of F and other words.!J?) = A(X).. that every u E P with.... F) acts on F”’ on ‘the left in a. . Let l? be a connection:in P.the real numb& field. .)]i.(u. .f (41. Assuming that m..(q) and the set of all (hT?). If hii? and v8 denote the horizontal and the vertical components of x respec- tively.8 of G into GL(m.). Connections+in a rdctor b u n d l e / ? spanned by m.4 implies I CHAPTER III wu40f = . We recall how l? defined the notion of parallel displacement of fibres of E in $7 of Chapter II.(u. Hence. then (hp)(x) = lb(x) . &&.

for . be a horizontal lift of 7 such that . let X* E T. where A f F. V. Vdtg. III. Proof Of hnma. then ’ r. Let p be a section of E defined on T = SC. Since is defined as follows. . x Q U.J. Then u. ~Gw) =‘u o %-xb(~*))~ hood of x.$ be the vector tangent to T at x.Y E T. E IT.(M) and let Q. a horizontal lift of 7 to E.d41. be a curve such that Uh~ = PM. that X = 5. This vxcp = v*p* implies Tg(y(Xh)) = u 0 + v. then the covariant derivative yayp of p in the direction of (or (2) WP + Y) = VA-v + VxY. Then If 9.l(p)(xJ). then X* -k Y* is a horizontal lift of X + Y. ..(P -G Y) = viy + v*y. Then set displacement of u.+J. to ?c. A the lemma. so that ~~ 0 v(xJ = xt Suppose that a cross section ‘p of E is defined on an open subset for all t.1. (1) will lift of r to P... If I is an F-valued function defined on T. the curve 7’ = u.p. v. for all X E TZ( U) .+.Y* . The following formulas are evident. the covariant derivative V. Then the covariant derivative V. = C.) -+ nsl (x2) denotes the parallel displace. --E 5 t zz E. If v and w are cross sectiops of E defined on T = q. As in Example 5.(x) . ihre 1 is Qn F-valued function dejned in a neighborhood of x. = u(X%f). (3) is obvious.+ -= u so that X* = ~2. Let X. The last formula follows immediately from I u(x*f) = fry ih [u O %?MX*N . by the parallel X = . Then we have h.(M) and X*. an element of the fibre TE’ (x) . The LEMMA. Thus.1$ Wb(xJ) . + (X1) .2. ePw1 Y#?4 = %) * V*f + (44 ’ 4%). of E dejined in a neighborhood of x. fact that.# = u 0 tl. Let .(P) are horizontal lifts of X and Y respectively.y~.E T. Y t T. Sincef is an Fm-valued function.g. Finally.. PROPOSITI?N 1. Let T* = U. X*fis an element of F”’ and u(X*f) is where $h: ril(x. (3) V7. is a cross section of E defined on M and X is a vector held (1) v x+rP = v. Then.l(~)(x~)) along 7 from x. on M. We proved (2) and (4). v(x).t is a horizontal curve in a curve such if and only if Vft~ = 0 for all t.yp = 1 *Ti.each U of M. Let T = xf.J)? thus completing the proof of It is easy to see that V:sq~ is independent of the choice of T..1 follows from the lemma and the if and only if V se. then. We have ment of the fibre 7ri1(xl+J along 7 from xtfh to q.(P) be a horizontal lift of X. .. that is. In order to prove the lemma. 97(x/J is the element of E obtained. -f- .: and y be cross sections \oyED. with respect to) X is defined b) (4) V.(M) and p? a cross section of E defined in a neighbor. for each’fixed f l F”.. --E 5 t d E. the curve v(q) in E is horizontal. we associate with p an fixed t.~(q(x.(M). it is sufficient to prove Let X E T. cross section p7 of E defined on an open subset U of A4 is parallel Now. (1) of Proposition of q~ in the direct&m of X Set [ = u.2 of Chapter II. then x*f =%h and ! uw -f (43 = 1. LINEAR AND AFFINE CONNECTIONS 115 114 FOUNDAtiONS OF DIFFERENTIAL GEOMETRY Proof. cross section q~ is parallel.($) = 1...[ is. Let 7 = Xt.J of q~ in the direction of (or with F”-valued function f on r-‘(U) as follows : respect to) & is defined by Given XE T.’ for every t and defines a cross section of E along T. by follow immediately from the following alternative definition of definition. covariant differentiation. if X.

. entiable in x in the sense that. F)). PROPOSITION 1. ZJ = I.~ E F’” and u. + Y) = VXY + %ts construct a reduced Subbundle Q(M. . a connection in P is called a metric Proposition 1. rlmjh R”. y) ~. we consider the canonical inner (4. . PROPOSITION 1. Proof. G) stood to be hermitian: defined by g. the inner product is under.2 of Chapter (6.) are M admits a jibre metric.r = ($. Then. p(a). If I is reducible to a connection I” in Q.3.cp. Hence. QED.g(u(f). .manifold. P such that. ‘.1. Then covariant differentiation V.y. Let T* = u1 be the horizontal lift of 7 P ROPOSITION 1’. This follows from Theorem 5.) = x0. Then defined fibre metric. for every curve T = kt. . existence of a Riemannian metric on a paracompact . v. Proof. ) defined by Let X be a vector field on M and XT the horizontal lift of X.dg. Then Q is a clased submanifold of P. . LINEAR AND AFFINE CONNECTIONS . @). (3) vnyp = A * v. every vector bundle ~. Set g = Ci . Fm. In Example 5. QED. f”). Conversely. In the standard fibre F” of E. G.5. . F”. If M is paracompact. p(a) E O(m)} if F .C. so that both T’ = ul(f) and T” = ut(t. Appendix 3).(x)h”(Sl. II. Let d%(fL’%J(d) = (f. to P. + v. XEM. a E G (a-’ means p(f-‘) c GL (m. be a curve in M. Let T = x1.i s.: Let g . III. of M.4. { Ui}ifr be a locally finite open covering of M such that rr. .‘(xO) -+ ‘ITCH along T is . .as follows.2. &j for f = (61. Proof. If v: M . if r is a metric connection.g&(v)(x). Let 6. Let X and Y be vectorjelds on M. E.7 of Chapter I just as the then ut Q Q for all t. sections of E.andy cross Sincelki} is locally finite and si vanishes outside pi. _: . corresponds to Lie differ- (f. This is an immediate consequence of the lemma for Given a fibre metric g in E.(1) v x’+pQ) = vsa. p>. is the corresponding function. E nil(~). . F). 0 !Z t 5 1. A connection I’ in P is reducible to a connection I?’ in Q ifand only.5. Ez) for Zr. Let hi be a fibre . .r E Fm. g is a well sections of E on M and 1 an F-valuedfunction on M. to each fibre metiic g. H) of P(M. E(M. which is differ. horizontal lifts oE. if 9 and w are differentiable cross isometric. then X*f is the function corresponding to the * cross section V xv. u E P. 0 g t d 1.. product ( . we saw that there is a 1 :I correspondence between the set of Let’Q ‘be the set of u. in the following sense.. r)) = c. We shall give here another proof using a partition of unity.e U(m)} ifF *- .Q with ~(u. P) an. More precisely. . that is.a jibre m&c in a vector bundle When E is a complex vectcr bundle.l( Ui) proving that I’ is a metric connection.. entiation L. 11 = (71. G) . Givdk a fibre metric g in a vector bundle E(M. where p is the representation of G in GL(m. P). Let {si} be a partition of unity subordinate to {ui} (cf. Y&d = A* VXVJ + (jwpl.= R. y(x)) depends differentiably on x. PROPOSITION 1.= Xi fiqi for 5 = (El. then. q”) E C”. u(q)) = (6. 117 116 FOUNDATIONS OF DIFFERENTIAL GEOMETRY metric in E 1Ui = ngl( Ui). . u*(v)). . It is ‘easy to verify such thatf(ua) = a-l(f(u)). connection if the parallel displacement of fibres of E preserves the A Jibre metric g in a vector bundle E is an assignment.~zr$ to P starting. we have d El. . if I is a metric is isomorphic with Ui x F”* for each i. x E M. .‘(x).. of an inner product go in the fibre r..from u.-G.. as an immediate consequence of Proposition 1. the parallel displacement n. that Q is -a reduced subbundle of P whose structure group H 6 The-correspondence is given by f(u) = u-‘(v(~(u))).7 to E.E is a cross section andf: P -+ F” H = {a E G. the reduced subbundle of P(M. We given by then have H = {a E G. .?. we (2) Y.1.d Q(M. I/) = g(ut(f). 7) for cross sections q: M --+ E and the set of F”-valued functionsfon P f. .

Chapter-IX. to A c 9 and .2.f B(E) is the standard horizontal z!ector field corresponding . Hence.. It corresponds to the identity transformation of the Q by Proposition 7.1:.) and (Q. -Linear connections point u E P.2 of Chapter I). . For each u E P. Given a linear connection P of M. . we associate with each 6 z R” a horizontal vector field B(E) on P as It is easy to verify that the structure equation of the connection follows.j = 1. .Throughout this section. m:“ .hus proving our first assertion. the standard horizontal vector lieids. . T o p rove (3). %(V)) = g(%(aY %blN = (& r). 118 FOUNDATIONS OF DIFFERENTIAL GEOMETRY III.(M) is a. there is a metric connection in P.. m} form a basis ‘of the Lie algebra gl(m.’ take their values in the Lie algebra o(m) or u(m) according as (2) R.(P). (1) Zf OisthecanonicalformofP. j-th row is 1 and%other entries are all zero. R = C. If X is a vertical vector at u E P. then R. following properties : and Proposition 1.F).: R? 7 T. both ((0.. s(4(8.j Q.) = u(i).. then r(X) = 0 and paracompact ma ifold K. P) b e a vector bundle such that G = G = GL(n. thenB(B(Ej) = 6 for i.1. LINEAR AND AFFINE CONNECTIONS 119 connection. be the (R. R”).2 of Proposition 1.‘ = 0 at some 2. tangent space T..*@(X) = 0(R. QED. . . i.(X)) = +F>.5. (2) follows from the fact that if X is a horizontal vector at u... If P is a metric connection.. GL(m. . assume that (B(F)). ‘The ‘second assertion is clear.(M) at each x E M in the sense of Example 5. The conditions B(B(5)) = 6 and ~(B(tj) = 0 (where frames L(. depend on the choice of . E‘ = 0. connections. F). implies that. a &ear connection of M. R). then where u is considered as a linear mapping of R” onto T.X) = (ua)-‘(n(R. hence 0(X) = 0. I dwj L -C. In particular.E. The canonical form 0 of P is a tensorial 1 -form of Hence. . connection form and the curvature form of a connection P in P.1 of Chapter II.j = 1.Cl:.. In other words.(M) (cf.2 of Chapter II. given a fibre metric g in a vector bundle E over a Proof.d. F). together with Theorem 2.. Since u.) = 0. The canonical form 0 of P is the R”-valued l-form on I’ defined PROPOSITION 2. w is the connection form) completely determine B(5) for each n:=dim4i. This means that I is reducible to a connection in ppe (GL(n. then R.). PROPOSWION 2. are skew-symmetric or skew-hermitian according as F = R or Proof. Let g be a fibre metric in 4 and Q the reduced subbundle of P defined by g. F”l. Lie algebra of GL(m. Let 0~ and Q be the QED. R).(X) is a horizontal vector at ua and r(R. The standard horizontal vector Jialds have the TO Q defines a connection in Q by Proposj@on 6..X)) m x rn’? matrix ch that the entry at the i-th column and ihe = a-‘u-l(7r(X)) = a-‘(e(X)).1 of Chapter II . E‘ER”. Set A connection in the bundle of linear’ frames P over M is called w = Z:. Example 5.byG. . $5 of Chapter II) CT be expressed by u such that’r((B(E)). IfA* is the fundamental vectorJield corresponding bY O(X) = u-‘(7(X)) for XE T. restricted to (3) rf 5 # 0.Rn. Then ~(6) = r((B(t)).X is a vector at ua 6 P. both i. Then (E{.V.. .(B(t)) = B(a-lt) for a E G and E E R”> _. then B( 5) ‘never vanishes.wi A f$ $. ut E Q.‘II) by P and the general linear group GL(n. i. we shall denote the bundle of linear Remark. Let je gl(m.5. . PG. and S& restricted’to Q.. Unlike the fondamental vector fields. then the restriction of o PROPOSITION 2. F).. (B(Q)U is the unique horizontal vector at I’ (cf.3. (1) is obvious. R). If Xis any vector at u E P and a is any element of Let E(M. QED. R or F = C.jwjE{. W e call B(t) ‘the standard horizontal vectorjeld correspanding to F.linear isomorphism.

thut is.4 of equation is similar to that of Theorem 5. in Theorem 5. x) e(Y). * * * .Z E T.4 (Structure equations). then [A*. w(Y)]. .2 of Chapter II (see also $1). A. The second structure equation was proved in’Theorem 2nd identity: DSZ = 0. Y) e(z) + s2(Y. we shall give an interpretation of the . i = l . a. (1) dei = ---I& wj A 8’ + O’. On the where A6 denotes the image of 6 by A E g = gI(n. Let o. We define the torsion form 0 of a linear connection I’ by .‘[)) = B.. do&I’.(At).(P) and u E P. Y).4.cpz$ subspace Q.2. QED. The proof of the first identity is similar to that of are three cases which have to be verified and the only non. . covariant clifferen&1.5 (Bianchi’s identities). connection r of M. we have Then the structure equations can be written as I. R*) . o(Y) * 0(X) = 0 and o(X) * 6(Y) = \ w(A*) . 0 = -dw A8 + W A de + de. T HEOREM 2. (2’) do = --w A o + R. This proves the first structure equation.B(a. is a linear isomorphism of Rn onto the horizontal w = &jwjEj.. structure equation de 5 -w A 8 + 0.. Theorem 5. 0 = IZi We.Then 0(X. Y) = 0. With respect to the natural basis e. The proof of the first structure Proof. T/&n /LI torsion form and the first structure equation from the viewpoint of affine connections.120 FOUNDATIONS OF DIFFERENTIAL GEOMETRY III. Let at be the l-parameter subgroup of G generated by 2. . we have d0(x.2 of. QED.3. If we’apply the exterior differentiation d to the first trivial case is the one where X is vertical and Y is horizontal.6(x)) + 0(x. LINEAR AND AFFINE CONNECTIONS 121 to [ E Rn. A. form -of a :lihear In the next section. For a linear connection. then we obtain Choose A E g and [ E R” such that X = A: and Y = B(t). 1st Structure equation: T HEOREM 2. R) (Lie algebra of all other hand. sz = . The second identity was proved. By Proposition 1.1 of Chapter II and Proposition 2. .W(5)) = -4. of R”. Y.:: [B(t) . we also write the structure equations in the following 0 = D8 (exterior. we write 8 7 pi Fe. B(E)]) = -B(B(At)) = -A[ by Proposition Proof. .1. the torsiwa form and the curvature. 0. + C$. Y) = -&(x) i 8(Y) . = exp tA. e.0([A*.. 0 is a (1’) de = +oh e + 0 tensorial P-form on P of type (GL(n. R). WOI = w$L ‘\ .. There Chapter II. Y.~&%~~~~ simplified form : . As in $1.(P).. 2) e(x) + ~(2.: connection foTtR. i. .Izi. + Q(X. -+ (2) dwj = --I& WI A w. 5.. we write Since . j = 1.-‘[)I = Bhirni (E . Y E T. B(5)]) = --8([A*. n x &matrices) which acts on R”.B(t)(e(A*)) .co(Y) . Y) s -$[w(X). . 2nd structure equation : 300(x. since w(A*) = A and B(B(t)) = [. Y) = A*(e(B([))) . Proof.Chapter II. 2de(X.2). . . . .a. n. Considering 8 as a vector valued form and w as a matrix valued form.. Y). and Q be the .9 of Chapter I and (2) of Proposition 2. (3) of Proposition 2. with respect to the basis Ej of gI(n. 2) = hz(x.! + (B/t)). where X.b By Proposition 5. where X. 1st identity: DO = R A 8.. R).. (cf.

(1) V. Y.. Let 7 = x1. B. Let X and Y be vectorjelds on M. E{*. be the standard horizontal vector fields corre. of Chapter I. . Chapter I). we defined covariant *[3) V.Y + (Xf)Yforever_yfej-J(A.. oh-(BJ = $. on Z(M) is completely determined by its operation on the I independent at every point ofP.z. DO(X. Since {Bk} are horizontal and {E$*} algebra of functions g(M) and the module of vector fields are vertical. Let T(x.) be the tensor algebra over .2. s)) associated with the .(ML VJ = IT (cf.(M). . case (3) only.8. {Ei*j are linearly independent By the lemma for Proposition 3. The n2 + n vector fields {B. Hence. By (3) of Proposition 2. of R” and {4*} the funda. Y) --+ V.z.4 of Chapter I). satisfying the four conditions (1) VttK. (3). It is a vector bundle with standard PROPOSITION 2.. although it is valid in cases (1) and (2) with obvious changes.(fY) =f*V. :. Zf X. o&q*) = @j. is easy to verie that {Bi..I @(E{“) = 0. in the sense that (5) V. j.Y.acement along 7 gives an iso- PROPOSITION 2 .K is defined in a neighborhood linear connection.} are dual to each other (4) v x+y = VA. sponding to the natural basis e..Y =y* VXY for every f e s(M) . R).3 of Chapter I and hence is left to on M.. bundle P o_f linear frames.2. . {Be. ofx. the covariant derivative lvith respect to a certain (2) V. . e. that is. X(M). is (3) V. Proof. (4) and (5) were proved in Proposition 1. we state the following proposition in 0 (W k e(X) and dw(kX. (1) v.. Let Z(M) be the algebra of tensor jields on hf. definition of covariant differentiation given in $1. k = morphism of the algebra T(x. 0: 2 t I 1. The parallel . on Q E D .K =f * V. For the sake of simplicity.. 6 .K. J$*} and {Oi. the operation of at every point of P. derivatives of a cross section in a vector bundle in general. when X E T. kZ) = (Q A 0)(X.3 of Chapter I.T. 2). kY) = Q(X.(M) and . . Then co(hX) = 0. kZ) PROPOSITION 2. then fiber T: (tensor space over R” of type (r. section of the tensor bundle r. In $1.(P) (cf.2’.+ v. It (3) V. As an immediate corollary to Proposition 1.z -I. we obtain Proof.K. Y. denoted by (X. Since V. Y and Z are vectorjelds on M.x(M).. Then the covariant dtxeetentiation has QED. $5 ofchapter I).n vectors are linearly independent. f = Xf for every f E S(M).Y + v. Since A + A$ is\a linear isomorphism of g onto the vertical sub. w. ’ Smce the dimension” of P is n2 + n.2 prove that the above n2 i. Z(M) is determined by its operation on . the n2 +Sn vectors preserves type and commutes with every contraction. QED.f = Xf for every function f on M.. s) is a cross (21 v x+yz = v.122 FOUNDATIONS OF DIFFERENTIAL GEOMETRY III. it is sufficient to (1) and (2) by an argument similar to the proof of Proposition 3. As in (4) V. 2) = dO(kX.) onto the algebra T(xr) which n} define an absolute parallelism in I’. Example 5. Y).7. mental vector fields corresponding to the basis (4} of gI(n. of M... space of T. = (da) A O)(kX. the operation of V. due to Kostant [l].(BJ = 0 . kY. we can define covariant derivatives of K in the following three Mie shall prove later in 97 that any mapping x(M) x 3(M) + cases : X(M). The proof of the ‘following proposition. (2) vx commutes with every contraction. i. above is actually.vy. * the following properties : LetB. kY. be a curve in M.fUi (Ej*) } form a basis of T (P) for every u c P. the reader. {B$ are linearly V...: 2(M) --f 2(M) is a type preserving derivation.(Y + Z) f V.l). A tensor field k’ of type (r.displ. we have Let T. when K is defined along a curve -r = x. E{*) are linearly independent at every point of P. when X is a vector field on M and K is a tensor field similar to that of Proposition 3. §3 of w. $1.Kfor every function f on M and K E 2(M) .‘(M) be the tensor bundle “ovqr Al of type (r. From the itB. sj (cf. LINEAR AND AFFINE CONNECTIONS 123 - Denote by hX the horizontal component of X..

.xi=1 K(Xl. l PROPOSITION / -. X. Given a tensor field ‘K of type (r.. we consider a tensor of type (r.E T.j 0 1: (. . R) ----+ 1. for each curve of degree r at x).11 of Chapter I. .Xi r Z(M). * * * 2 XJ. can be uniquely decomposed as follows : L Similarly to Proposition 2. From the viewpoint of fibre kmdles.. the m-th covariant differential V”K is defined ‘inductivkly to be V(V+lK. considered as a cross section of a tensor &ze space.. s) at a point x z M as a multilinear mapping of T. Hence we have . that we enlarge the bundle of linear frames%0 the . As in X .X.(x). t 6 Rn. we have Observe that.(M). is parallel if and only if qxK = 0 for all X E T. ) respect to a vector field.) for (V. If K is a tensorjeld of ape (r. in contrast to Lie differentiation L. .(VK))(. of M. Afine connections T.(M) (s times product) into To(x) (space of contravariant tensors A linear connection of>a manifold M defines. III. . The proof is similar to that Similarly. LINEAR AND AFFINE CONNECTIONS 125’ 124 FOUNDATIONS OF DIFFERENTIAL GEOMETRY that is. X. then Every derivation D (preserving ape and commuting with contractions) of the algebra Z(M) of tensorjelds into the tensor algebra T(x) at x P M (V2K)(X1. 1 VK=O. ..“.. . ‘spaces are regarded as vector spaces and the parallel displacement : PROPOSITION 2.9.(M) and . QED.-J. a tensor field of type the following sequence: (r.. The group A(n.(M) as an a&e space.(M) (s times product) into T. R) of. .X. We set T = x1. we denote it by A”. 0’5 t 5 1. X.. which is. .. . The elemeiit d’maps a -point 7 of A” into ay + t. . represented by the group of al1 matrices of the f$m .V. . X. . For any tensor field K and for any vector jieldr where X E T. . 5 + 1) defined as follows. regarded as an af-hne of Proposition $5 of Chapter I and is left to the reader. covariant differentiation V x makes sense when X is a vector at a point of M.Y l T. . we have D = Vx+S. space. XJ) afhne space at x.. Y) = (V.. .bundle of where X. X and Y.s+2). . .(M) y .12.~-. I commuting with every contraction. .JM) onto the tangent space TsI(M) . R) and 6 = ([*)..). This follows from the fact that Vx is a derivation When we regard R” ‘as an affine space. 2 (V2K)(.(V. X) = (V. as we shall ‘now explain. r Proposition 2.. X. We use thenotation (V*K) (.10.* . s) is defined to be V(VK).X8).(M) and S is a linear endomorphism of T. space T. R) -L GL(n. is a column The second covariant differential V2K of a tensor field K of vector. . the tangent space of M at 5. .11. will be denoted by hI(kf) and will be called he tangunt . Y) = (V. . . We have type (r. . Let R” be the vector space of n-tuples of real numbers as before. .(M) x ..X. I‘called the tangent (VKP..Xi.10. Y). .( z = a E’ PROPOSITION 2. Xl. X. .X. . affine frames.K) . .. x E M (cf.K)(X. Weset I 0 + Rfl CL_ A(n. s).X).(VK))(X.x T.-I. afhne transformations of An is bundle. this means . $1). . X) = b&W’.Xr . where a = (4) 6 GL(n..- 2. . Let M be a manifold with a linear connection. the parallel displacement of the tangent X. PROPOSITION 2. * * * 3 V. . . A tensor field K on M. s).K. the covariant di&rential VK of K is a tensor field of type (r.(M).. Proof.= V.(M). where X. / _ .. In general. .x 3.X.’ We sha 1 now consider each tangent space T. these tangent ( W V . . then is a linear isomorphism between them. with (VaK)(.(?“-‘K))(. .if we regard K as a multilinear mapping of T. A tensor jeld K on M is parallel if and only v .

iden%& with an afflne transformation u’: A” -‘A.) at x can be be the decomposition corres$mding to a(n.. R) into a c GL(n.2 of Chapter I) wAGTC co’ is gI(n. X. X.. . . indeed. Example denoted by M.. . In particular. that is.. * : * .set $$&s kohsi&ng of a. to the image of the preceding one.. X. e. We denote by A(M) the set of all a’ffine ffames’ df M and Gefine (cf. R). we see easily frames at x and the‘ set of affine transformations if A” &to A. * * .2 of Chtipter II. . M .algebr+: wedefineybyy(a) Q A(n.). . R).L(IM) maps (2) The homomor@ism~‘/k. that up? is a tensorial l-form on L(M) of type (GL(tr. We shall call (0. e. R). k) i gI(n. e. MdcJbud&&:+ *. R) identity transformdfion of L(M) .(.1.. $5 of Chapter II) and he&e C&r&ponds to a tensor. .*. . E A. a principal fibre bundle over M with group A(n.). such that b 0 y is the identity automorphi?m of GL(n.) the -canonical frame of A”: Every afline frame (p. valued l-form on L(M). Example 5.wralized aj& connectiops sponding to the homomorphisms /3: A(n. tie have homomorphisms b: A(M ) + tenrorjeld of@@ (l.(P. . L(M) can be which maps 5 E R” into E A(n..X. = identity of considered as a subbundle of A(M).4(n.. ‘By Proposition. linear conhection’ if M and ‘a y: GL(n.. 0’ ‘XJ t) fi gives a 1 :I coriespondence’ between the set of &iiie defines a connection in L(M). R) sponding to the splitting exact sequence 0 + R” + A (n..(M). .6. . . R)-valued l-form on L(M). Evidently. . ah. . it will . R] and’ sh%lI call.$$k@ Q o .) into (X. .). wi ha&‘ the following spi&ing exatt : sequence of Lie . R) is a semidirect product of R” and GL(n.) and 9: E(M) -+ A(M) maps :. I. X. . It can be . R). - (P.l(tz. X. K) is 1: 1. R) (Xl. [) z GL(n. G&(n.GL(n< R) and of M and t&.. .4 of Chapter II.’ t E A(n4j and a” E A(n . el.‘i Therefore. .. beiause (tp. . projection +: A(M) 4 M by setting 7?(c) = x for every type (1.) at x. R). o$M&en by f’ +’ (I’. X. R)-valu+d l-form on L(M) and p is an R% Maps (0. We shall study the relationship exact in the sense that the kernel of each homomorphism is equal between generalizid affine connections and linear connections. R”). . . R)) and j3 is a homomorphism of A(n. Corre- in the sense that there is a homomorphism y: G-I.The group d -+ R” -+ &n. where o. R).. . .) the natural ‘basis for R”. there is a unique pair (a.. R) (I. i.-R) + 1 &g%ups. tirre. . .tnd l$ X bc tire tensorfield oJtrpe (1. . where 1za” is the y*c3 = w + $.ln. R) -+ A(n. R).A(M) the PdOPOSlTION 3. X.’ Let L(:M) be the bundle of linear frames qver M. . R) . ‘X. The sequence is A(M) of affine frames over M. .$ondh~ b&w&n the set of ger. Thenay* is ari a(n. a E GL(n.. ..(@. . X.R) + d. R).proved easily (cf.(M) which so *that o is a S&Z. R).i]. I)ofMasc xp 1 ained in. . . R). R) --f A (n. R} + y (semidirect sum): An afine frame of9 manifold M at x consists of a point p E A. (1) Th e&k. R) x R” such that L? = or(E) a y(a).map f” into r ($ §6 ofchpptiLII). . X. LINEAR AND AFFINE CONNECTIONS 127 C where a is an isomorphism of the vector group R” into A(n. L(M) and y: L(M) --t A(M).+ R”.‘Let ‘0’ be the origin of R” and (e. / I eveiy a”6 A(n. 4 -+ zia”. We shall show that A(M) is a principal fibre bundle over M with group A(n. R) onto GL(n. on a$ne connection % of M and let _ A(M) by (u’. R)-valued and 9 L -R%alued. It is a splitting-exact sequence We denote by R” the Lie algebrti of the vector group R”. R) + . . 126 FOUNDATIONS 0~ DIFFERENTIAL GEOMETRY III. R) .l) o$ . .-G. Namely. . A generaliied ajhe connection of M is a connection in the bundle E . Let . affine frame z? at x. .) into (p. for y.M).‘&. On the other hand. R) e gI(n. . e. b: A(MJ . p 0 y is the GL(n. RI = gI(n. . s composite of the affine transformations a”: A” -+ An and z?: A” - A.P(M) + L(M). R) acts freely on A(M) on the right and that A(M) is conlucfion oJ. We c&fine an action of A(n. Let I’ be the linear that A(n.(M) is the point corresponding:to the origin of ?i”. Let (3 be th connecfion form of a gerier&ed bundle of a&e frames over M. . R). X. field of the.(M) Let & be &e connection form of a geheralized affine comiedtion of and‘ a &ear frame (X. . ..) into (0.

+ &.. the R”-valued l-form 9 is then I . Let a~ be the connection form ‘of P and q the tensorial l-form on c(M) of type (GL(n. yjl. . From the structure equation of ?.‘ (1) It is sufficient to prove that. Proof..(A(m).2. ‘.shall not break with this tradition. t(M). Y) = dw(X. R).W’h dX)l) + D&T Y). . R} and A c R” . 4Y)l + f-W. &] + a. R)-components and. the . we identified Xc T. : QED* the canonical form 6 defined in $2. Since R” is abelian.Y c T. o(X) = o( $‘) = b and hence . /I *. &tie both 9 p(X) and ad (CT) (-A) ape in R* and o(X) is in gt(~.t&.. Y) ‘there is f which gives rise to (I’. Thm field of identity transformations.1 . choose (Here. Y). .$ is horizontal w&r&petit to P.iant &Sren&i& with $&l & c. . [v(X). III.used interchangeably.Y ‘.R) such that u’ = UC? and 8 2 ‘R.:homomorphism .1. Since both X and Y are horizontal.equation of a generalized affine . by C(8) = ad (@)(o(X) + q(X)). c(Y)] + n(X. f is an P ROPOSITION 3. Since X = b(X) = b(X). K).Tj. To provpl tha’t :($$) (X& 2 .c h(n. Assume that 8 is horizontalwith respect :to f dp.respyt to I’.@(M)). + 2. . both sides ’ connection” have been . R) -c G&. ’ g *‘&(X) + &. Y) + dv(X.uation’ to L(M) and by com- (2) Let Xc T. 4X) =.’ (2) both X and Y % Traditionally. = c@I&(X)). . “’ Consider again the structure..q+Dg.@epr @nnptions r of M.G . LINEAR AND AFFINE CONNECTIONS 129 128 FomAfloNS OF DIFFERENTIAL GEOMETRY have Proof.f .men&l :. A(M) -‘Jf(M)/R”.~following of M and ‘the .&jj. There is an element A e a@. Y) = &X. given a pair (I’. we . connection f : It is straightforward to verify that cz defines the desired connection &-j & -. c7.Rn-components w e x.. two cases: (1) at least orie’of X and Y is vertical.1. Dw(X. corresponding to K.+o&$& . This is justified by vanish.vector . ~~orresp&&ng ~ to A. we &iv& : A generalized affine connection p is called an aJne connection if.’ X. let d add $2 ‘Qe the amutirre affine connection if the tensor field K corresponding to v is the forms of f and I? respect&&. iF. q(Y)] = 0.*. Hence. R) such that . with the notation of Proposition 3. .(&l)(c5(X)) -+ ?I. Y) = --j&z(X).proves that if . ‘In the case (2).. a(X) = ad (&)(A) anid’& +-p(X) =.‘ sothat 0 = 6(X). Y) + dq(X. In Proposition 3.Y s T.(L(+q). R”) = -3b(X). of tangent spaces of M. we write - X = &i(X) $ A. Moreover. R) A A(n. We set’u = i(c) and X’= p(X) so that paring the gI(n.Dq.the’~:words -(‘linear connection’? and “affine ’ are horizontal with respect to‘ ‘l$. R) such that u’ = ud sitid Just as in (d)(A). we y(X) and 6(Y) = q(Y). W/W L(M) can be identified with A(M)/R” and dv(X.’ We define * QED. is. there e&t (r E R” c A(r. obtain associated-*&h’ the homomorphism #r”A(n.(X) y(X) with X.. : f. Y) = &(X. The’] ho&omprphism /l: A(M) + L(M) maps where ‘D ii t&i-ekterior c&a.iet ‘ox. /?: A(M) -L(M) can be considered as the natural projection d4K Y) = -3b(X). As an immediate consequence of Proposition 3.. Y). dw(X. Let X.B(<Jb r is horiyp$ .O. In other words. we have ’ I p8 r $z . . Although we.c T&(M))i s irice. considering L(M) as a subbundle of A(M). R) .1. Y. Y).+‘&(A$) = ad . _ This .. the . ‘it is’sufficient to. where A* is the funda. Y) +‘Dq(X.&X)‘. d&Y.8: A(M) --. P). K).$X. y* dc3 = dw + dv and hence is vertical.. . . Y).) On the other hand.(L(M)) and d c A(n.(L(M)).’ between the set of afini cdnnections f i2(2> Y).n the&$ (l). a linear connection I’ of M.yit$. 9ml + qx. = Theorem 3. d&(X.3.=-o/. ”~ T ‘. By restricting both sides of the eq. Given a vector X E T.1 = -t([l+% dJ’)l . Hence. .

) for each t. at so. Let C. set holonomy group) of f or I?.GL(n.(M) -A.hw+Q that is.e. III. be the 6. x0.) in A.8(zi) E L(M).2.tangent shall make a logical distinction between a linear connection and bundle over M is naturally homeomorphic with that of the tangent an affine connection whenever necessary. It is a special case of the parallelism in an associated tion. From bundle which is.= x1 with the trivial point field along T.1 of Chapter II. dg..L(+U) = A(M)/R” ( c f . We shall call @(ti) the a#ne holonomy group of l? or l? and Y(u) the linear holonomy group (or homogeneous PROPOSITION 4. There is a natural PROPOSITION 3. the afhne tangent bundle whose fibres g.(M).(M) in the-usual sense) to Yt for every t.4. Let f. Developments connection.of xt in L(M) with respect to the linear . R) Then C. dw=-cc. AZ. PROPOSITION 3.4+. .ci. the horizontal lift .onto the Chapter II. In particular. the distinction between the two is That Al is a connection in L(M) and an affine connection is a con- the affine tangent bundle is associated with A(M) whereas the nection in A(M). The homomorphism p: A(n.. + ?+a”. Let p. there i+ F curve & in R” c A(n. Let uqbe any. $3) maps from an affine connection of a manifold M.1 of ..1. Let @ and Q be the torsion form and the curvature 1: 1 correspondence between the set of point fields and the set of form of a linear connection . ?b is the parallel displacement y*6 = 0 + R. = utz. zit into ut. in Theorem 2.&me transformation of the affine tangent space.ho&ktal with respect to the affrne connec- A(M). are A. (&I). 0 i t 5 1.defined along 7 so that pFl is an element of ATt( M. be a curve.f. 0 5 t 5 1. that is. curve in AzO(M) starting from the origin (that is. we shail find a necessary and sufficient condition for affine tangent space at x1 defined by the given connection in C& in order that 8. Let ?i be the affine parallel displacement along the:curve’ T from xt to x. be the horizontal lift of xt in A(JI) with respect to the affine connection silch that Zz. with reference point-u’ c A(M). R) such that zi’. From Proposition to x0 and i.NEAR AND AFFINE CONNECTIONS 130 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 131 The total space (i.@(M): maps Q(u) onto Y(u) and @o(e) onto Y’?(u).” . corresponding afine connection. The affine parallel displacement along’~ 1s an and that $ is the M. We shall denote this affie parallelism by +. x Q M. The following proposition allows us to obtain the development Let @(u’) be the holonomy group of an affine connection f of M of a curve by means of the linear parallel displacement.vector field along T. ivhere T: is the linear parallel displacement along 7 from xt groups @O(.) and Y”( u) are-defined accordingly. Let Y(u) be the holonomy group the parallel displacement defined by the corresponding linear of the corresponding linear connection I’ of M with reference connection. Then the development of the curve 7 in M into the affine we rediscover the structure equations of a linear connection proved tangent space AzO(M) is the curve +k(x.5.==--o~p.this section the parallel displacement arising homomorphism p: A(M) . in M. . = uo.‘.point in L(M) such that n(uo) = x0 and u. Then .c point u = .. Given a curve T 2 xt. Co = x0) such that a point field. is parallel (in the’ajhne &ice A. Let T = x#. a linear connection of (vector) bundle over M. Replacing q by the canonical form 8 in the formulas: We Identify the curve T .R) . we obtain of the affine tangent bundle is called a pointfield.’ Proof. be. 0 st t *5 1.) ‘describes a curve in A. the bundle space) of the affine .4.F of M. Let si be the curvature form-of the vector fields. . As in the proof of Proposition 3. LI. we obtain . Since the We shall study in. tangent (vector) bundle is associated with L(M).+Dy. A cross section From Proposition 3. the point field corresponding to the zero. Then #O(pz. The restricted afine and linear holonomy Yt = it.(M).(M) along 7 (in the reversed direction) from xt to where y: L(M) -+ A(M) is the natural injection. in our case. is the vector tangent lo 7 at x~. is the develojment of 7 into A.

R”). and the curvature R can be exbressed as follows: J dC. Yt is independent of f if Z. .(a).- is parallel.(dqdt) = Yt. This form fi of a linear connection.w h ere f is an R%alued function defined We have already defined the torsion form 0 and the curvature by f(u)’ = U-r(YJ.(M) -+ T. I T yj3ec~*)) r ec. QED.(l. Y) = V. R) of Rn.Z = ~v.. Example 5:2 of Chapter II).(M) cii(iJ = ad (Z._..ylz.X . is a parallel vector field along 7. L’ijfi”$j+.(dz. we set = ad (4-i) (w(&) + 19(ti.(M).‘) (e(ti.)) + &-li. where r(u) = x.) h --. we have that R is a tensor field of type (1.‘)(G($)) + &-lb. * ’Proof $f Lemma.iQ’T-. We set . In the lemma for Proposition l.132 FOUNDATIONS OF DIFFERENTIAL GEOMETRY III.Y). Then R(X. horizontal if and only depends only on X.]z.. where X.*(B(y*))).sd thacfis an symmetric bilinear mapping T.l.1. ::. (V. ‘yij j.: u(2O(X*.-: F: ~pwyy .) vectors of L(M) at u with r(X*) = X arid r(Y*) = Y.p. _ and Y* (cf. and is . To prove the second equ&ity.Y E T. R).u.’ _ /’ and only if 1. respectively. Let X*.1. L&MA . f(u) =*$(Y. Proof. X). at every point x of M. since p([i*. torsion) T and the curvature tensor field (or ‘simply. Y and is independent oft. LINEAR AND AFFINE CONNECTIONS 133 which follows from Leibniz’s formula. It follows On the other hqd. Y*))(u-‘Z) denotes the image of u--lZ~ R” Yt = 4%) = by the linear endomorphism 2Q(X*./dt).[X.2 The development of a curve r = xt.’ - T HEOREM 5.i.\?.tN = %(Wt)) Thus. We . curvature transformation of T. and.Z are vectpzjetdson M.(Z~‘(O)). u(X. Similarly. R(X. 3) . into AJM) is a line segment tf and only tf the vectorjetds 1.) for u c L(M). (2fi(X*. %(%-l(~. * T (X. Hence. We now define the tors[on tensorjeld completes the .(M). Hence. In R”-valued function on L(M) of type (GL(n. X*. QED.. Y) the torsion translation in T. We shall call T(X.pioof of the in the proof of Proposition other words.we proved that (YiyY). where G and cu are the connection forms of the affine and the where u.1 of Chapter II.2) such that T(X. ![. is. Y*>) for X. .. ‘Y). Tdefines a skew ‘2. 7 Proof.c’gI(n. In t$e above if 6J(li.. T&L.(M). we obtain . Y and Z.Y .such that R (X. Thus. Thus 1. j X* and Y*. q = fqQr+ (W > where u is any point of L(M) with n(u) = x and:.Y. We have . . Y) isindependent of the choice of u.(Vd)‘a -w> % . Y) = 3. Y* and Z* be the horizontal lifts of X.+ I ’ : z. -\ (or simply. c. not on . definition. 0 5 t 5 ‘. Y*)) (u-‘2) for X. the torsion T Hence. In Proposition 4.‘(xJ) = Uo(q-yU~l(Xt))) = u.. In terms of coyariant d#erentiation. On the other hand. Cwuaiure angj to~ssio~ tensors.(M) determined by X and Y.-‘-TjIx. T (X. is a line segment if and only if Y. We have therefore : curvature) R. Y)Z linear connections respectively. X). Y and.(k. &(Xi Y)‘is an~endomorphism of T. ~ already know-that T(X.Y* are : I ’ ‘. Y] C OROLLARY 4. this fact can be easily verified directly also. Y) = -R(Y.Xz and.&led the = -u&i&-‘) = -+l.x*. = +)(xJ = u.. 5.Y) 2 = u((2Q(X*. determined by X and Y.)) + G. we-set f =f fl(Z*) . We first prove .(M) k T. Tis a tensor field oftype (1.‘& = ad (2.l. Y*}..T(Y.nt. Y..i&-l. =.. X* and ‘Y*. are chosen as above.Z E T. -s. = (V. G ufX:f ). C./dt = -u. along 7 = xt R. &.

I . fir X.lZ . z*)) zsqx. Y*l)J-) = 4(4X”.2. Z)) = (x:f>(% r.Bianchi’s identities generated by A. to the + &&(2o(Y*. .“’ ’ G(R(X.Z. On the other hand. Z)). Y.l(f (4) -.5). Y)z) = @T(T(X. we have = iii . [a.*) = X. thus proving our assertion. thm = u(2Q(X$ Y:)(f(u))) = u(2Q(X$ P:)(u-'2)) ='R(X. R) Using Proposition 5:2. . Let X*.Z f T=(M). vr(Z:) E Z-and n(W. ‘The proof of the = G. we define Y* and Z*. : = -4f (to). Bianchi’s 2nd identi@: where A( f (u)) d enotes the result of the.T)(Y...*. V[X.Y. u-‘(&T)(Y..2o([x*.R and their covarrant denvatrves. bY and Z. L~ QED. -F) Z} ~‘0.> Then by Corollary 5.z*)))(u-1w)j. Z* and W* be the standard horizontal uectpr jielats on L(M) cowes~onding to u-l&. where A* is the fundamental vector field’ corresponding to A. = X. Y*.R)(Y. By Proposition 1.R be #he torsion agd ths cumatuw of a linear connktion of M. Let A be an element of gI(n.([x*. . We consider the On the other hand.2 of Chapter II SO that.134 FOUNDA TIONS OF DIFFERENTIAL GEOMETRY 135 LINEAR AND III. Y*]J = -A. Similarly.rJf whge 6 denotes th cyclic sum with respect to X.We T.f corresponds to VxT. We shall prove only the first formula..2) over R”.‘.: z L = 4t4c W)Uf > = .. Y). z:) =+0(x. + R(T(X.5) in terms of 7’. X. z*))) i \ire have and - !(v&(Y. and u-l W respec.*> = fromr. y). u-l Y. G{( VxR) (Y. L--t T and . Y*]". ‘Let 11 be any point of L(M ) such that T(U). Proof. then = u(-c(y*f) . f”(r. if a. y*l>uf). Z)} z. we have torsion T as a cross section of the tensor bundle T:(M) whose standard fibre is the tensor space T: of type (1. I'.3 of Chapter II. YZ) = -o. if we set NV.~. Let X.2. Y:)e(zf)} Proof. I’*]). linear transformation A: R” -+ R* applied to f (u) E R”.f(4].4 . z:)}.sDs(x:. ux> VYIZ . :. We PROPOSITION 5. we have ‘= x:(fh 0) = x:(2o(Y*. Y)Z. by Proposition 3. Z).V. Y:.T)(Y. Hence.A u(x:(2o(Y*.Jw”f) -vu*. ?I. A:f = lim A [f (zq) -f(u)] I-4 t THEOREM 5. z*)) .(u-l(R(X.11 of Chapter I. Z:) f be the T&valued function ‘on L{M) corresponding.*.3.y(. z)> =. Z . . Bianchi’s 2nd identity:.Yl (4)) In particular. second formula is similar to that of the first. Y:.0. R) such that A: = (u[X*.3.AFFINE CONNECTIONS then torsion T as in Example 5.’ Bianchi’s 1st identity’: G(R(X.*. vertical) component. ) QED. then (Theorem 2. 3 Bianchi’s 1st identity: ... V.We shall derive the first identity tiueb.. 7r(Y:) = Y. v) denotes the horizontalp(resp. .z))w = u((x:(2n(~*. . u+&.‘is the l-parameter subgroup of GL(n.Y. so that P(X. if T = 0.% q = u-1 y and 5 = u-lz. Let . (V. . ?k + (V. 6(&l A 0)(X:. Then. ) . we shall express the. Therefore. Y)Z)}. 5) z U-‘(T(X.(M) and u E L(M) with lift X to a horizontal vector at u and then extend it to a standard 72(u) = x. Z ) ) = O(Y. Q) = :‘&2fi(x:. where h (resp. horizontal vector field X* on L(M) as in Proposition 5. Then s Do & 0~ 0 (Theorem 2. we have G((V. W.X.

A. Y). y .. B’)j= 0. and’ hence is a linear combination of AT. 2.B. VR = 0 implies vectorjelds on L(M ). [B. B’]) = -2de(B. By Proposition 2. . define vector fields on P(uo). B’) is a constant function (with values in The sec. by Proposition 5. Then we have.‘q}. .. p. QED. Just as in Proposition 2. 2)) = 0. .T)(Y. ( 1) If V. = 0. for instance. Y).. .A $DQ(X. Bj] is the standard horizbntal vector field corresponding to . . B’]) =.*)) [AT. The following proposition -2o([x*. . 2)). B’].. coincides with = qX:(R(Y*. .2... B’) = -2i(B.2. Q L(M) 1 . [A:. then [B.T = 0. B.. These vector fields A: .Jin~ horizontal vector field corresponding to the element -2@(B. $3 x$qY*.‘~. Y = n(B. B’) of the Lie algebra of Y”(uo). B’] coincides ‘on Pfu.~. Y*lu. y*lu) = @r&Y*]J) = -U(2&(XZ.. a constant function qn P(uo) (wi. Bjl to P( u 0) coincides with a linear combination of At. be the standard horizontal vector fields corresponding to the basis el. B’]. of R”. B’) = 0. z*)) . A.qnd identity follaws frcq t&& thr$!for&a+ R”) on P(uo). by’ Propo&n 5. . . Let A. ws have . then the vertical component of [B. -2@(B. Hence. Let 4 an4 B’. We observe first that / ‘: they define an absolute parallelism on P(uo).Aiej E R”.B’]) = -2do(B.Y). [B. Y*]. B’-) = -2Q(B. . if VT = 0 and VR = 0.*)) = -T (X.~B’)) = u-‘((V.11). ~ROPQSITION. Let.*. restricted -wcx*> y*l”. (1) rf. 2)). the vertical component of [B. (1) 0([B. 5 ) .vector fields.= 0.. X = +Y*).. .) be . Herice. coincides W e h a v e with a standard horizontal vectorjeld on P(uo). . Aj. Then we have .. 2)) gives some information about [BfI BJ.) and 2 G n(BL). B’).constant Hence we have coefficients. : iJ’ + f j O. A: with . < L .be prdved fr& the formulas. which corresponds to a? ileinent A . ! III. B’] is vertical. values in the Lie. the fundamt%tal vectorjeld A* on P(uo). _ X*(20(B. then ths horiFonta1 component” of &3. Y& 2:) ‘ I (2) If OR = Q. i of R”. e. Let P(uO) be the holonomy subbundle .6. ’ - have’ As in the proof of the first ident$y.h (2) If R. be arbitrary . z. .. Let P(u. Y. .the holoqomy subbundle of L(M) We shall derive the second identity from through uo. Let B.. .-thtiugh -a point u. algebra 1 of yiuo)). (2) w([B. . be a’basis of the Lie algebra of [Bi. It is fields. then-the restriction of reference point uo. B. component of [B.of .B’] is hoiizontal: rli:’ I-:. 2:) = +-y~ (T (x.standard horizpntal . 2:)’ ‘L-y T( T(X.. ... 61. B’] is vertical..qx*.) the &near holonomy group with It follows that. Nomizu [7.1 (see.i[&f). = u-y T( T(X..3 can. to P(uo). . Theorem 5.*.Y*].. .. .a.’ QED. On the other hand. B’j.. : DQ = 0 ( T h e o r e m 2 . B’) = 0.. Y. . This’ mzans th$ R(B. is horizontal. (originally defined on L(M)).of the Lie algebra of the linear holonomy group Y(uo). then [B. have * Proof. : .4.2. !2jji l This means that O(B. A: the corresponding fundamental vector = &((V. B’) Theorem 5. .. (I) Let X* be any horizontal vector at u E L(M ). Since m([B. B’) -is. Since 8([B. P (uo) with the fundamental vector field corresponding to the element -2R(B.~. T . .) = -u-y T(n[x*. .] = -+2@(X. Z)).. B ’ ] ) = Proof. B ana!* B’ !be arbitrao standard horizon& (2) Again. Yqu. LINEAR AND AFFINE CONNECTIONS 137 Y(uo) and A:. ‘b)) = gu-‘((V. ‘he i. AT] is the fundamental vector field corresponding to [Ai.I $*(Q(-B.R)(Y. 2)). y). B’].j with the ‘standard Remark.and Y(u.T)(Y. B’] coincides on -2a(h’. .3. ‘1% -R([x*. the horizontal * QED. We know that 74x*. . By Proposition 5. ‘II ^ . PROPOSITION 5.

QED.: R” + C1 in a manifold M with a linear connection is called a geodesic if TSt(M). there is a unique geodesic with the initial. + ri&. Let u. ing to [.. This is an immediate consequence of Corollary 4. PROPOSITION 6 . ’ tangent to 7 at xt..& Acurver=+a<t<b. . t and &. X) is defined for all values of -co < < co. Since the parallel I As an application of Proposition 6.). .3 is that a geodesic.4 with the! initial straight line. .1.(M).defined in some open zation of the curve in question is < < co. B. Another consequence of Proposition 6.the Lie algebra of ‘4” (tl.. . horizontal and -since t3(ti. respective&).E is parallel along the curve xt (see $7. s = s(t). . LINEAR AND AFFINE CONNECTIONS 139 A:.) = u. Then v~c. rf any. Since xt is a geodesic. In other words. In fact.or. For any point x 4 M andfor any vector X Q T=(9). we have 2. X). A:. 3 . the parametri- Conversely. . A curve r of class Cl through x E M is a geodesic where is an affii& parameter. any parameter t which makes T into a geodesic is a curve of class Ct. ‘be a geodesic’ in M.) = b. that is. t.f its development into. Let b.)) = uclit = E. 138 FOUNDATIONS OF DIFFERENTIAL GEOMETRY III. QED. condition (x.. and ysct) = xt. of class C” (provided that the is called an afine parameter.) = 7r(Bb.. converse& The vectoffields A:.t. be an integral curve of B. is determined up to an a&e xt through uO. .M and af and on& . every standard horizontal geodesic T and X E TJ-II ) a vector in the direction of T. which If T is a geodesic.‘(r(ti. condition (x. where a # 0 and /I are constants. that rs. B. Chapter D. E R”. t PROPOSITION 6.where--03gaa<bsoo. In this definition of geodesics. before Proposition 7. horizontal vector field on L(M) 6. Let r be a curve of class Cl in M. Proof. then g forms a Lie algebra and dim g = dim P(uJ.(M) is (an open interval of) a X E T. unique geodesic xt such that x0 = x and &. be any’ point of L(M) such that PROPOSITION 6.6.denotes the vector lar.field is of class C” and hence its integral curves are all of class @‘. the bundle of linear frames L(M). Aparametriza. QED. We set ht = ~(6. b. Then ir = n(b. We set 5 = u0-‘$. Then s is a function oft.4). let x be a point of a l&rear Connection is ofclass Cm). for any x s. B. The parameter t is called the afhne parameter for T determined by (x. = X. .2. s = at + /I. Let g be the set of all vector fields X on the holonomy bundle P(uO) such thut e(X) and W(X) are constant functions Another useful interpretation of geodesics is given in terms of on P(uO) (with values in R” and in. such that x0 = x a curve of class C” in M. A linear connection of M is said to be complete if every geodesic I. The vector j.).3. can be extended to a geodesic T = xt defined for .2. a basis for g. Hence we have Proof.). . t. .ofclass c where b& denotes the image of [ by the linear mapping b. a of Chapter II). 7r(uo) = x0. .. let x. Let xt and ys be two parametrizations of a curve T integral curve of the standard horizontal vector field ‘B correspond- which make T into a geodesic. the &ode& 7 = xt in Theorem 6..’ $3 THEOREM 6. = X. COROLLARY 5. Tbe. ut is an Proof. is a horizontal lift of xt and 5‘ is independent of the vector field X = R.X exists and equals 0 for all t.. T. r = xt. $ must be a constant different from zero. The projection onto M of any integral curve of a standard horizontal vector field of L(M) is a geodesic and. defined along T is parallel along T.. with constant coefficients on P(u. B. Since b. Hence. In particular.projection onto &f of a curve of class C” in L(M) is t there is a unique affine parameter t for 7. Let u1 be the horizontal lift of tion which makes r into a geodesic.4. is automatically. where . Since ut is transformation t -+ s = at + /I. where a # 0. is equal to d-t particu- if V. we obtain the following ds displacement along 7 is independent of parametrization (cf. defined above form every geodesic is obtained in this way. Let B be the standard. X). Geodesics which corresponds to an element 5 c R”. . If VT = Q and VR = 0... s interval containing 0.$ .

: “0 “I. e .dxj. Let e. ! 7. I. coo~&at~“&gh’6orhoo’&. tL’e recall that a vector field on a manifold is said to be complete if it generates a global-l-parameter group of transformations of the -: manifold.J&t ‘%’ be the wordinate nei&borhood . $f. 1.= &2(X. i = 1. . X’. 10r each x. L :..’ iei’r’b~ iz lineal ion&e&x? of M: ‘Let .~formati”on... Chapter~II. ‘II.” later (in $8) define the exponential map in the case where the Let u be the‘cross.140 FOUNDATIONS OF DIFFERENTIAL GEOMETRY in Theorem 6. . we can define the d(X*j = u+.I .. assigns to ea:h linear connection is not necessarily complete and discuss its x l ‘U the Ii&ear frame ‘((X1).). 2t is valid. k’::s. 2(X. . let 7 = xt be the geodesic with the initial condition (x. Xi) so that u maps es into Z. I% the’ intt?rsectio~ of the : twoi (I’$. X{(X. above formula && Proposition I.% with respct to “local coordinate. We set differentiability and other properties.form of zi linear connecti& I’ of M. . We take (xi.i .(L(M)) PROPOSITION 6..) so that Z. El+wessions in local coordinate s@ms Then wU is a gl(n. .. .rule. Xi) as a local coordinate f-jk bi the components ox. We set Let. . R)-valued l-form defined on ‘d”%e define ‘~9 functions J?&. LINEAR AND//AFFINE CONNkTIONS ‘141 Immediate from Proposition 6. We shall U) = ~~.:. X{Yf = CjY{Xt = 8. : I! ‘i. I. QED.: 8 = Xi Pe. where x = n(u).! .’ .. .‘. then When the linear connection is complete.. L(M) over V which assigns to each x E V the linear frame .:’ concepts in terms of local coordinate systems. For each X E Tz( M). = Zij(Yj 1’) e.. ~. . . . . ?n and’zl. j. : . .over U which.:0f PROPOSITXON 7..(M) into A?. We denote by Xi the sym’bols) ‘of the link connetition’ I’ with . _ 1‘ ‘.. .). .i:.‘...’ . . . ISi Xi(Xi)z). : :.%.. .: . $&iemi system in +-*( U) c L(M). Let M be a manifold and U a coordinate neighborhood in &i The& functions I’$ are &lit& the ckponents (or ‘6hrijtofil’s with a local coordinate system x1. ‘we h&e We shall express first the canonical form 8 in terms of the loc$I coordinate system (xi.. these components are (Ci Xi(Xi). be the natural bdis for R” and set .‘th?loeal~ vector field a/ax!..~#‘(~&j~~~.I’% an/i’ where det (X{) # 0.f and only and if every standard horizontal vector field on L(M) is complete. .. i.eld. Every linear frame coordinate system x1. En fact.L i .). the components’ of a tensor *fi. . It should be noted that they are not’ at a point x of U can bea uniquely expressed by’ . WC jdmive a the. ..j w:: Ed. * ’ .) be the inverse matrix of (Xj. . ? . ‘s - PROPOSITI& f.. Proof.aY&.~espe6tiveEy.). subject to the followings$ect to. we write ” # Thus we have a mapping of T. .1. 1:: -. .‘. ’ exponential map at each point x c Ai as follows.where the the canonical for@ 8 = Xi Bit+ can be expressed as follows: coordinate system 9. ‘on U !by In this section.-. Example 5. .. .section of L(M) . e. . (cf.3 is the following Proof. . . n. . R).. .. . X) . . (X. defined in Cr..5. (. exp X = x1. Let u be a point of L@G) with coordinates (xi. . If X* c T.. a‘. we shall express a linear connection and related . .. ..--:.. In terms of the lo&l ceordirqte $stek (xi5 Xi).2 of Chapter I).. A linear connection is complete ..2. Let% be thwross se&q .4. Let x!. . With respect to the basis (Ei> of gI(n. so that 7r(X*) ._ UIU-&..). .P.).4. . xn.u be the connection .o& 13’ = zj y. x”. .

a VxjXi = Ei l?jkjXk.5: dssumt? t h a t a mtipping .4. A simple calculation shows that Fro& the components I$ we can rec~nstiuct the c&n&ion form w. : . Let x1.. . This implies that the given mapping . X3. n.function ‘on my = (ad I&~)&‘+ Y&Y. xn.Y is the covariant derivative oj‘ Y in -the 9i = Xk Yf(dXj + &m l$. . Then there is a un@ue linear i?‘j. . . It is easy to verify that the’ form o defined by the Proof. Then :’ By a simple calculation. Xi) by ’ . Xj.j. . R) by reversing the process in the proof of Proposition 7. in termsof the coordinate system (xi. Y f e . such a way thnt t&p s&h the tran. . (2). xn are given by .’ This shows that the components of 8 in $4 of Chapter I and in $1 of Chapter II. . .a linear connection I’.. local coordinate ~systi By: Proposition 1... i. X.iY:e.ef M s&that V. .Xi xl. in and -hence _1 VX.CO’ I? j i is given in ter& of the Jocal coordinate sp?teT %(M). (3) and (4) of Proposition 2. If (sj) is the natural the corkection l? defined by w are exactly the functions I& Q E D : coordinate system in GL(n. Leaving the detail to the reader.j.3.X(M) x X ( M ) + form w = z.(MjI :t&t is. X’. r-l(U) cz L(M) which corresponds to Xi.* -. i To apply Proposition l. From Proposition 7.. LINEAR AND AFFINE CONNECTIONS 143 .function corresponding to V. n .1. . R) defined in... i = .I’& dx”. .&ma@-rule of Proposition 7. 9i. this form was denoted by Hence o*wj = &. it follows each.V.e horizontal lift of Xi.XTf is the . .. Y and Y’ arc vector fields the conditions (a’) and (b’) of Proposition Ll of Chapter II.3. there is given d set of”tions I’$.. .‘.I. this can be proved by GL(n..X. . . * I Then there ‘is a t&p@ line&r connection I’ whose components with respect toxl.. $4 of Chapter-I. Y> .. Let x l M..) by (k’i) + (xi. then in terms of covariant derivatives. 142 FouNDAmoNs OF DIFFERENTIAL G EOMETR Y III. . . GL(n. 5..xi = ~. X* kg!venby . QED.turn+) for x E U n V. . section CT: U -+ L(M) used above to define ov is given in terms Let v be the (left invariant gl(n. Then the verify that components I’& of r with respect to x1. = (VXY’). .Y). Let XT be th. i/ PRQPOSITION 7.. the transformation rule of our proposition. . we shall give here an above fwmula’defincs a connection in’ L. .8.$enoted by (X. we see’that this formula is equivalent to -f= I. con’necfion l’ . . then fact that w is independent of the l&al coordinate system used (V. x~ are precisely the given ftitions I?& Moreover..= 1 . .*f = $.E TV ~~ Ed. . is given so as to satisfy the conditions (xi. R)-valued) canonical l-form on of the local coordinate systems (xi) and (xi. .). . . The cross w = 44 . IfX._ . . R) and if ‘(4) denotes the inverse The components of a linear connection can be expressed also matrix of (sj).Y.3. the formula in Proposition 1. and hence Proof. k = 1..2. xn be a local coordinate system in M the proof being similar to that of Proposition 7. rt.). let 3’ be the R”-valued. . (I).: With our notation. 0.. ( a/aZm). .$. rj”i X. The on A4 and if X = X’ and Y = Y’ in a neighborhood of x. PRQPOSITION 7. We define a mapping yov: U n r -+ follows from the transformation rule of I’. o satisfies outline of the proof.. (P/W. direction of X with respect to I’. . Proof.4 of Chapter II can be expressed as follows: ..4 It is easy to with. .3.).2. . y = Zi... the connection 'PROPOSITION 7. Assume t& . a&‘$. Set Xi = a/axi.

v”K These formulas follow immediately from Proposition 7. on L(M) by -q=1R?a ~~eKj$iL. then local coordinate system 9.. F. . Xk) = Xi Tip%?.7.8. if K is a tensor field of type (I. Let 9’ be the components of a vector field Y with respect ro a PROPOSITION 7. From Propositions 7. 3il. Xi)Xj = & l?jktX<. (1) Iff is a function dejined o.::-$.&t”.jiaxk + I& czq = aK$ . EA .. they can be expressed in terms of n l-forms 0’ and functions.8 (where U - is any open set of M).i. R(Xk. .+ The proof of this formula is the same as the one for a vector field. .6.144 FOUNDATIONP OEf DIFFERENTIAL GEOMETRY /I III. k = 1. ./ In particular. calculation. define an If X is a vector field with components p.2. QED. More generally. then the components of and the curvature R as follows.Y are given by Ej $Jj. Since the torsion form’@ and the curvature form B are‘tensorial ponen ts K$ : : : f:.. .f. say.1 bye the transformation rule of Proposition 7. = (art/ax* . By Proposition 7. .q + zz. then the components of VK are given by forms. we (2) If X is a vector jeld on M with components P. then. f r o m The components 7& of thetorsion T and the components qir of the curvature R are defined by T(X5.8.k. . These formulas follow immediately from Theorem 5. .4. . ..4. they define a linear connection. Then they can be expressed in terms of the components I$ of the linear connection r as follows.-“7) We define a set of functions q& and $. we define n8 qk. x”. PROPOSITION 7.arga. . X These functions are related to the components of the torsion T is a vector field with components Fi. if U is a coordinate neighbor- hood with a local oordinate system 9. then the components of absolute parallelism (Proposition 2. Let r~: U’+ L(M) be the cross V. Then I.&’ -Jj. obtain the following formula : ~~j = ?qilaxj + & J&q”.rgrjm). Let 11: be the components of the covariant differential VY so that V. It is dear that VxY is the covariant derivative of Yin the direction of The proof of the following proposition is a straightforward X with respect to l?.6 and has components K$ : : :I:. .Y = . . Let xi = x”(t) be the equations of a cume .J. .. LINEAR AND AFFINE CONNECTIONS 145 I- wfl x wf) . Then Proof. .8.“j.. QED..k 31.K are given by section over U defined at the beginning of this section.3(@ induces a mapping I(U) x 3(U) ---F PROPOSITION 7. . . Y = Eci $(a/&‘). i. L(M) can be expressed in terms of these l-forms and functions. tu2’. Kj: 1: :i. R. II.. We have X(U) satisfying. ... I.. except that Proposition 2.k = zi T~j~i. functions l?j. these functions sat’sfy and Proposition 7.(r.rk .7 has to be used in place of Proposition 2. . -.j#)9 where I takes the place of & and m takes the place ofj. If. u*?+jk = Tky o*i& = Rikp The covariant derivatives of higher order can be defined similarly. For a tensor field K with components K$:: 12. on U the formula given in Proposition 7. I’$K. Kil.4 and 2. Xi r&Xi.n M. where Xi = a/&$. the sfme conditions of Proposition 2. s) with com. Since n2 + n l-forms t!J”.6)) every differential form on V.

3).Y + S(X. exp is defined We shall compare two or more linear connections by their only on a subset of’T..‘. then V.3. is defined for 5 E V and ItI 5 1. respectively.transformations exp B(l) : . 7. on 6.Y As we have seen already in $6. n. 0 5 t 2 1. Q E D . t Proof.10. If E’jk and” l-“ik are the components of linear neighborhood V of 0 in R” such that exp B(t) : U* + L(M) connections r and I” respectively. We set exp tX = xt.1. Normal coordinates Finally. = I’& + Sjk define a linear connection I?‘. Proof. In general.t) I”$ define a linear connection for each t.q* + L(M) is defined.For every point x E M. there is a neighborhood N. P ROPOSITION 8. . connection is complete. Let r be a linear connection of M with com- we consider M as a submani$ld of T(M) = U T. let T = x1 be the geodesic with the initial condition = 0. QED. From ‘the Tormula for the components of V. we set # =bv+l N(x. in T(M) such that the exponential map is geodesics as r.. . = x0. then Sjk = I”. Conversely..8. P ROPOSITION 7. the exponential map exp is defined on the whole of T. there exist a neighborhood U* of I’& and l?’ with components I”fk. where &I and 8s are positive. Let U be a neighborhood of x0 in ponents of a tenorjield of tyje (1. b. connection with vanishing torsion. The exponential map N --f M is dzxerentiable of class C”. components. LINEAR AND AFFINE CONNECTIONS Let A4 be a manifold with a linear’ cormection 1‘..horizontal vector field bn L(M) (cf. 0 5 t 5 1. if we set X = it. they are related to each other as follows : 5 = a(x)-‘X. be any point of M and u. of x in M by the exponential map. By. that ?r(uO). because B(t) depends differentiably The following proposition follows from Proposition 7.6 and 7. Our proposition follows immediately from Propositions Proof.F with components Proposition 1. X) (cf. the set of functions I’*$ = ?CM trjk + (1 . Then there ponents r&. we can choose . 146 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 7 = x.) = U N.(M) for each x c M.) . there exist a neighborhood U* ‘of u0 and a P ROPOSITION 7. In particular. the set of functions tI’& + ~a and a positive number 6 such that the local l-parameter group (1 . III. the zero vector at x) in T.(M ). For each 5 Q R”. X) and’ hence . be the set of X E T. 8. In terms that o(~)-~d E I’ and set: N(x. For each fixed t. if rjk are the components M and (T a cross section of L(M) over U such that u(x. where V and V’ are the covariant di$erentiations with respect to r and:J” exp X = rr((exp B(5)) e u(x)). a point of L(M) such 7. . 2 .. The components of the vector field it along 7 are given by dx”/dt. Given X l N. Identifying each x E &I with the zero vector at x.9. and of a linear connection r and S& are the components of a tensor field S of u(U) c U*. 2)) then I”. Therefore. exp tX is defined in some open given . i = i.X = 0 is interval -&r < < &a. provided that the connection is of class C”. difeomorphically onto a neighborhood Li. QED.. 2). Y) for any vectorjelds X and Y on MS .) = u. Given 147 X E T.5 of+Chapter I. Let x.(M) such gpe ( 1.. . Then n((exp tB(5)) *u(x)) is the geodesic with the V>Y = V.2.(M) ‘for each x E‘M. It is now clear that exp: N(x.(M). of class C 2. for Proposition 718 implies that I’ and I” have’ the same geodesics if ItI <.. If the equivalent to the above equations. In general. for all t E K simultaneously.) + M is differentiable of class C”.t) I?& defines a linear connection I’* which has the same is a neighborhood N of M..l?jk are the com. let N. of local . (x. Given a compact set K of R”. Given x E 0. I?*$ = &(rj& + rij) define a linear defined on N.U* and 6 rjk + rij = r’fk + r$. initial ‘1 condition (x. 0 A In this section we shall prove the existence of normal coordinate P ROPOSITION 8 . set zru of covariant derivatives.(M) which is mapped differentiability of the exponential map. systems and convex coordinate neighborhoods as well as the of x (more precisely. we see that.. Then r is a geodesic if and only .f I L. Theorem 6. we denote by B(E) the corre- b_ sponding standard . given two linear connections . $2).

. lY’.. By the implicit function theorem. x” with morphism u:W + 2”.4 and the We call it the normal coordinate ystem determined by the frame u. p). . say the couariant dz&ential of . be an arbitrary point of M and xi. Consequently.. C OROLLARY 8. p) ordinate system is valid. For any (al. .. definition of a normal coqrdinate system. .* r$(alt.. QED. n.. af 0 < p : c. Let xl.(M) defines a coordinate system in T.9... X” be the normal coordinate system in $7. By’ Corollary 8. * * * . Given a linear connection r on M. . In particular. .e + FL = 0 at x0. If the torsion vanishes. By Proposition 7. xn with the above properg is in Example 3. Let x1.‘p) has a’ ndr%al coordinate neighborhood is the normal coordinate system valid in a neighborhood’ i’df x containing U(x. coordinate system valid in a neighborhood U of x and yl.. .there ea$sts a iositive number c such that. .. !a:. In the above definition of a normal coordinate Then there is a positive number a such that zf 0 < p < . . . Let o be any dz&ential form on M. .5. Therefore. . . is torsion vanishes. x” be a normal coordinate system with origin x. . by Proposition 7. PROPOSITION 8. where X = xi aiXi. then any rjk + l?b = 0 at x0.the sphere deJned by C (xi)2 = p2 components with res#ect to a normal coordinate system with origin xo. X. . Proof. defines a local coordinate system in U. 2” (2) Each point of U(xo. geodesic whiAAis tangent to S(xo. is a geodesic its &. / where Vo. . Let K be a tensor field on M with components Given a linear frame ’ u = (Xi. . . hence. .k r& (xo)aiaE = 0. and if the both are determined by the frame u = (Xi. Proposition 8. Then Then . Then the C OROLLARY 8.. then the covariant derivative Kj: : : : >. . . (1) ‘U(x.a.~&a 1. The second assertion dw = A(Vco) at x0.. i=l. of x in T. . . QED. . . .p) at a point. -+ coincides with the partial derivative aK2 1: $axR at x. From the definition of the exponential map.. singular. . QED. . p) be the neighborhoodof x.... The first assertion is an immediate consequence of the normal coordinate system with origin x. This is immediate from Proposition 8. Let S(xo. This is because if x1.4. xn be a normal coordinate system with tion. if the torsion of F vanishes. ant)aiak = 0.. III.. defined by Ei (xi)2 < p2. Let xi - T x’(t) be the equations of a geodesic i=l . r Proof. . then system.. I2Zj. p) in. X). then . connection has no torsion..uniquely determined by the initial condition (x..) at x. p) denote.8. QED. . . . by Proposition 7. $ U. T HEOREM 8.5.7.. there is a neighbor- Since this holds for every (al.6. X”) at x f M. .4. let F$ . .(M) origin x. of S(x. LINEAR AND AFFINE CONNECTIONS 149 148 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Zj. Proof. . .2 of Chapter I. . x” a Proof. .o and A is the alternation:defined Converse&.a neighborhood ofy. p). Let U(x. a”). Kf’::-i with respect to a normal coordinate system x1. .. follows from the fact that a geodesic is . . .. Proof.. it is evident that the dieerential of the exponential map at x is non.. determined by a linear frame u = (Xl. &I). . . . an) an. d o = A(h).3. . we may assume that the linear then they coincide in a neighborhood of x. the components I&of the linear connection vanish at x0 by origin x0. .. the covariant differential of K in terms of l?jk given P ROPOSITION 8. in a natural manner. the curve defined by xi = a9. necessarily the normal co&dinate system determined by u = (Xl.. . .. n.. we did not specify the neighborhood in which the co. Xn) .(M) which has the property stated above. then I’jk = 0 at x. Let x1. Q E D .I’ k in a natural manner’. If hood N.. formula for. . p) is convex in the sense that /&ty ‘two points of U(x. . then F$ = 0 at xg’ Proof of Lemma 1. . p) lies outside S(x. Since the torsion vanishes by our assump- Proof. rf t h e geodesic r = x1 with the initial condition (x. : . X). . Remark.’ expressed by xi = ait . .6. a local coordinate system 9.. the linear iso. xn is the normal can be joined by a geodesic which lies in IQ. the diffeomorphism exp: N. Let x. the torsion vanishes..

there exists a positive number c such that This means that the geodesi?. These groups can be realized as If the connection .1.. let 0 < p < a. X E V. . where p: = F(t.. LE M M A 2 . exists CL positive number a.8. p) at a point y = (x1(O). the local holonomy for X E p. TY) 0 T. I J ..Xi I’j. 2(t) is tangent to the sphere <the quadratic form with coefficients . For each point in a natural manner..U(x. Linear injinitesimal holonomy groups ’ containing U(x. In gene&. then (d”F/dt2)l=o > 0 asd to. c) . If 0 < p < c. at fhe point x’(to). exp X) defined as in $10 of Chapter II. there exist 1 a nei... This Lemma 1. 91 is a mapping of T(M) into groups of linear transformations of T. a) is a diffeomor- phism. x = w(u).).(M). let x and y be points of U(x.U(x. . F(t) = xi (d(t))? To complete the proof of Theprem 8. Let xi = x’(t).is complete.@acement along an arbitrav. denoted’ by M x M. Set u of L(M). W e consiper A4 as a submanifold of k(M) Let r be a linear connekon en a manifold M.. C.. 3’ (2) Each point of U(x. contradicting hence F(t) > p2 when t & 0 is in a neighborhood of Q.. = V n T. group Y*(U) and the infinitesimal holoilomy group Y’(U) are v(X) =. the holonomy group Y(u). a). xi(t) lies outside Because of the equations of a geodesic given in Proposition 7. let (p will be restricted later).(M).is positive S(x. be the value for which we have F($ attains the maximum. Proof of Lemma 2. Assume that F(t) 2 p2 for. . $4 of Chapter II).. c)‘ for all X F V elements of the form (d?) (X.p). c). . dF = 2Ei o= ( z Xf(tO) $ 1=1 a ) 1-t .. pJ. . By . (1) Any two points of U( x0. Then the geodesic with the initial wise d@eTentiable curve T starting fro?n x. a).. where and ) t 15 1. of the holonomy group Y(x) such that pl: V + U( x. U(+. a) can I be joined by a geodesic which lies in U(x.). MIows from (2) of Lemma 2. < c such that Whitehead [ 11. ’ QED. ‘Then there The existence of convex neighborhoods is due to J. Choose a positive number c as in hemma 1. completes the proof of the lemma. Since the differential of v at x0 is nonsingular. piece- Let X = q-‘(x. Set V. X) joins x and y in U(x. Y E T. Y) = 7-l 0 R(TX. .(M).. a) has a normal coordinate neighborhood 9. 0 5 t 5 1.. H.. x”(O)) condition (x.(6.. we may assume that exp tX E .7.(M) spanned by all and a small.xi) . c). . @ is defined only in a neighborhood of M in Y(z).. The Lie’ alg&a g(x). X. 1). a) is a diffeomorphism. Taking V is equal to the subspace of linear endomor-him of T. T(M).. Let to. vanish at x0.(x.the’choice of definite in U(x. p. This proves (l)* (2).. Let x Then andy be any points of U(x. . ( ) 0 Since I’j. soee t (that is.ghborhood V of x0 in T(M) and a positive number a < c T HEOREM 9.150 FOUNDATIONS OF DIFFERENTIAL GEOMETRY III. Since exp: V. To verify (2). M) and 7 is the paraUd di. the geodisic x’(t) lies inside the sphere S(x. condition (2) is satisfied... a) x U(x. p). LINEAR AND AFFINE CONNECTIONS 151 which is tangent to S(x.. 0 ” .. Y*(X) and Y’(X) respectively (cf. be the’ equations of a geodesic ‘from x toy in U(x. c) (see Lemma 2). Then I.. 0 < to < 1.. We shall show that this geodesic lies in U(A+. To verify condition ( l). p) for some t). Set ’ F ( t ) = Ci (xi(t))2 for 0 I t I 1.

.. . V. V..qe vector Jields on hi anh .. . . . . v. ... Wj. at the point .). Y. WC take’R(X. . . U i .. .) LEMMA 2.2 will follow fi-om Lekima 1 arid.. we Iraq of tensorjelds of type Bi (resp. The rcmaininq terms on the form (G”R)(X.. We have L(M) and the curvature tensor field R (cf. This&lows that every tensor field of type :I.g(. ..h(f(ViR)(u. 1.. .(M).(h4) of the tensor fields of type A j. k = 0. y*... 0. Theorems 10. .... Y) and 211(X*. T HEOREM 9.. 1’. Assume that VI. .V:(2Q(X*.(V’R)(V& V. lli)..f . . V... QED. V.).. Proof. Y)) (resp. the. . 2 . and0 5 k <. 0 :’ j .. * of type . B. W.2..((““-‘R&I-. Wi) The results in this section were obtained by Nijcnhuis [2]. . By a tensor field of &pe 11. Jfk. $5 of Chapter III)..&. . . is a linear combina- tion of tensor fields of type Bj. . . (cf.~ 1.1. . Y)) i s a s u m q&y*.. y. WI. The proof is achieved by the following two lemmas. . i Fr. .. are their horizontal lifis to L(. V. 1’. * * . .‘. (VkR) following lemma. 0 .. . are ar$trary vector is a linear jielis dk $1 ‘?hen ave& tensor jield o f !#e A. . 2. If X.) proceed to the determination of the Lie algebra of ‘l”‘(x). . we mean a tensor Chapter II). . f(ViR)(U. k = 0. Th’is complctcs the proof of Lemma 1. . + (CiR)(U.I?$: (P’Rj(X. . Wj)) By Theorem 10.R)- valued functions of the form (I. The Lie algebra g’(x) of the inJinitesima1 holonomy The first term on the right hand side is a linear combination of group ‘P”(x) is spanned by all linear endomorphisms of T.(V’R)(U.1. I’..f) .1?.s trivial. V. . .. V. Y*)))U 0 U-l (Zj of terms like for Z c T. . V.. .Y’ -/..)) -~ (Tk. Wj) + q .. .lRj(‘C.f combination (with di$krentiable functions as coeficients) of afinite number x*. ‘l’hcorcm 9. 1’ (1. (resp. .2 the restridted = (V.(-w’R)(X. . . V. FOUNDATIONS OF DIFFERENTIAL GEOMETRY III. . (VW)(X.l. QED.3 of Chapter II in terms of Y(x) and Y”*(x). Vh. V. k-f.i : : k. n’(u) is sparmed 6y the values at u of all. 1 . where X. This follows immediately from Proposition where f is a function. Y. Wj).(R(X.nX. ’ By definition. v.8 of Chapter II and Thcorcm 9.) differentials V”‘R.) =m-... W. $0 of L EMMA 1. The case k = 0 i. the form VI.1 of Chapter A s s u m e n o w t h a t every tensor field of type /I. Y. V. gl(n-. .3 of Chapter III. . . Wj) holonomy group ‘PO(x) of a real analytic Ii:lc%. . *.2 and 10.1 ._. .14)..152 .. The proof is by induction on k.. . .. . v. .‘i’..... . LINEAR AND AFFINE CONNEC'I‘~'O)NS 153 Proof. 0 sj :: k -. Vk-.. 0 . is a lineal II and from the relationship between the curvature form Q on combination of tensor fields of type ll.arian + (Vj”R)(U. W..(M) right hand side are linear combinations of tensor fi.).k. where X.. .._. . Y.l) It is easy to reformulate Proposition 10.. . This follows immediately from Theorem 8.(R(X. V. V. . . J&.j k. V. II. . . V& W. Vkc T. . WC shall therefore -. Y. J’. T . * a * LIW(& wjzz Proof of Lemma 1. .5j 5 k . * * . ...3 of Chapter III.-. . V.a~.’ . Proof of Lemma 2.... . Then we have 1.I. . v. W.the T$i. k .. . ) I’. (VR)(C’. Y. WJ.). .tr ‘connection is completely determined by the values of all sue-ccssivc co\.. Y*) as v and f in Proposition 1. V. (res]). V.

Every compact Lie group G admits a Riemannian Riemaxmian Connections metric which is invariant by both right and left translations. a/Z?) = dij (Kronecker’s symbol). we obtain a Riemannian metric on Example 1. .of Riemannian metrics’ v(X. each induced by an where 0 c’ p d n . definite symmetric covariant tensor field of degree 2. q. g(X.1 of Chapter I). thus proving our assertion. Since His compact. f* Y). then G admits the following canonical (cf.7 of Chapter I and Proposition 1. 6) c G x G and x e G.4 gives an indefinite Riemannian metric element a E G such that a(o) = x and define an inner product gz on G invariant by both right and left translations. semisimple Lie group G defined as. We define a . X. . is invariant by the for X. If G is Let A4 be an n-dimensional paracompact manifold. Riemannian metric h on N is defined by h(X.1.? can be given by group and H is a compact subgroup. Another example of an indefinite Rie- element of H which leaves the point o fixed. identified with Chapter III). mannian metric g defines a positive definite inner product in each positive definite inner product g.Y Q g = Z’. When G is frames to a bundle O(M). a-‘Y).e. We discuss this metric in detail in coordinate system 9.Y E 7’.. Example 1. G. The on M and the set of reductions of the bundle L(M) of linear .’ form q~ is bilinear. In . . in T. The T. where G is a Lie coordinate system 9. By Example 1. In fact.3. A homogeneous space G/H. X” is defined by Volume II. An indefinite Riemannian metric on R” with the Example negative definite. where X. is independent of the choice of a Q G such that a(o) = x and that the Riemannian metric g thus obtained is invariant by G. say go. Y) = go(a-‘X. diagonal D. g(X.5.(M) (cf.1 (W2. we write. Let J? be the linear isotropy group at the origin o (i. we have the KilIingCartan form is a 1: 1 correspondence between the set. IV. Riemannian metrics metric invariant by G x G. the point I qc’=l (dxp. . Every Rie.%I. X. Y) = trace (ad X0 ad Y). we. a E G}. gz(X. we shall mean a symmetric covariant N +rto a Riemannian manifold M with metric g.(G/H) which is invariant by J?. so that G = (G x G)/D.that g. Y) = g(f. By Example 1. Y) = tangent space TJM). mannian metric is the carlonical metric on a noncompact.(X. Y) or.5. that is.(G/H). Let f: N -+ A4 be an immersion of a manifold indejinite Riemannian metric. in the Killing-Cartan form Q is indefinite and nondegenerate. We know compact and semisimple.(G). shall always mean a positive g( a/axi. admits an invariant metric. Example 3. By an Example 1.Y E T. The homogeneous space G/H provided with an invariant Riemannian metric is called a Riemannian homogeneous C H A P T E R IY space. G admits .(G) by gb(X. Y) -q(X. 5. For each x E G/H.(N). The Euclidean metric g on R” with the natural G invariant by G x G. . that M admits a Riemannian metric and that there the tangent space T.follows. By a Riemannian metric.of orthonormal frames. a}. . g.4. Y E T. . The induced tensor field g of degree 2 which is nondegenerate at each x f M. 6) * x = ad-l. for (a. . The isotropy subgroup of G x G at the identity e of G is the diagonal D = ((a.a Riemannian 1. represented by the coset H) of G/H. compact and semisimple. symmetric and invariant by ad. Y) = 0 for all Y t T.(G/H) by g. It 154 . It is known that for such a group so is R and there is a positive definite inner product. in T. we take an construction in Example 1. RIEMANNIAN CONNECTIONS 155 - is easy to verify . Since (p is invariant by ad G. Examples 5. 2 is a group of linear trans- formations of the tangent space Z’.J. Example 1. the group G x G acts transitively on G by (a. simply. .(M) implies X = 0.1.2.(G/H).3.4 of invariant Riemannian m&c.the Lie algebra g.(G). Y).3.

*(M) by means of the isomorphism *: %xample 1.. If K and L are tensors at x of type (r. x” such that gi. w(ivad)j = Wij* On a Riemannian manifold M. girl’.l. I be a local coordinate system in M. . B) = . g ij = g(dx’. in the dual space T. gij = g(a/axi. We have then Then :. dxk) =g(dxk.(M) onto its dual T:(M) tensors. i. bf = -b{. . For each x. x” are given by g(AX. * * * .1:: 12. or K” E T?:(x) with components The inner product g. RIEMANNIAN CONNECTIONS 157 156 FOUNDATIONS OF DIFFERENTIAL GEOMETRY then Let A4 be a manifold with a Riemannian metric or an in. . <Y. dxj). The distance The inner product. a/ad) = fa/axj. curve of class Cr in an obvious manner.. of class C1 is defined by On the other hand. i. x”.gJlajbf = Z a# = --z ajb{ = -trace (AB). 4. Given a Riemannian metric g on a-connected manifold M.~.g~l”.&jk:‘f::jf:.yr. 4 = g&T Y) for all YE T. tensors at x of type (1. . Given a tensor K E q(x) with components K. s) with components Kh : : : f: and Lk : ::j: (with respect to x1. for X. In fact. X) and g(BX. it)* dt. . dxj) = z. L) = x&k. L is given by thus proving our assertion. ti = Iz. Y) = -g(AY. fa In terms Of a lOCal coordinate system x1. In The contravariant components gij of g are defined by fact.(M). that is. the inner product gz The isomorphism 1y: T. . = 6. denoted also by g. denoted also by gz. .. . B) is equal to --trace (AB). i.(M). W-V)) for a. by y(a/ax”) = Ejlyir dxj.-:. the infinimum to an inner product.y) on M as follows. define vi. .~) between two points x andy is.(M).. 9. a S t 5 6... S( a ponents Ei of the corresponding covector or the corresponding This definition can be generalized to a piecewise differentiable l-form a = y(X) are related to 5” by.: L zkgj. . .gi~. . Then the inner product g(A. we obtain a tensor K’ E T:. $(K.(M) and in T:(M) can be extended . n . we (space of covectors ‘at x) as follows: To each X E T. . . tensor space ‘I’@) of the lengths of all piecewise differentiable curves of class Cl at x for each type (r. The com- ponents gij of g with respect to xl.z. at x and let aj and 6j be’ the components of A and B respectively. we have * L = og(ft.K~:: :~L$‘. c?$ = iz* giQ*. . . n .(M) + T:(M) can be extended to defines a linear isomorphism y of T. gijg’l” = s:. IV.j = 1 . .. . s: = (a/axi. .:(x) with components assign the I covector a E T.(M).*(M) defined by K’il31:::j. ap). : .-**. B) = I: g. If ti are the components of a vector or a vector field X with L = b respect to x1.gjk. Let A and B be skew-symmetric endomorphisms sda. Then we have since B is skew-symmetric.i(x. that is. Y) = -g(BY. . in the.‘:& definite Riemannian metric g. lu(a/axy) =g(du~. 1) such that Let x1. X) gij = s(alaxi. the arc length of a differentiable curve T = x1. .6. YE T. fl E T:(M). g in T. then the com.j = 1 . X = Zi E’(a/ax(). s). by definition.e. . take a local coordinate system x1. we define the distance function d(”“‘. . g(A.b(V’(a). 2. in Tz(M) defines an inner product. of the tangent space T. . .

Although the results in this section are valid for manifolds with indefinite Riemannian metrics. Then we have The proof of the following lemma is similar to that of Proposition 2. qj A ej + Oi = -cj . On the other hand. definining a metric connection of M. .X ==J’ and a$. w is a connection form.5 of Chapter III.1) forms We shall see later (in $3) that djx. Assume that the connec- definite) Riemannian metrics only for the sake of simplicity.)ining x and-r.. Proof (A).OJ can be expressed in terms of 0’ and OJ~ by the lemma. we represent 0 and UJ by n forms O’. . .4 of Chapter III. Among all possible metric connections. Riemannian connections p.) A ej = cj.r).$n(n . n. Existence. connection in O(M) determines a connection in the bundle L(M) of linear frames. . a metric connection if it is thus determined by a connection in 0 (M).). Fik = -Fi. our proposition follows immediately from. = ckFjk@. both ( iijk . i = 1. Finally.. . Since g is a fibre metric (cf.k F& 8” A 0’. QED. Then v . Since which is given by the following theorem. deJne an absolute parallelism on O(M). Hence.. we and the other using the formalism of covariant differentiation. d(x. 1 d j < k $ n.f and only if g is parallel with . since Proposition 6. . . that the topology defined by the distance function (metric) d is the same as the manifold topology of M. Every 0 = cj ipj 1:. $1 of Chapter III) in the tangent bundle T(M). The n forms O’. e. Then. tions define5 by UJ and Q. It follows that Fj. .w annihilates the vertical vectors.j = 1. . proving the uniqueness. . Let ~0 be the connection form on O(M) zj 7: A fji = -@i. . one using the bundle O(M). d(x. . A linear connection l? of a Riemannian manifold O(M) and. n. have no torsion. n. respect to r. algebra o(n).y) 2 0 .j A p.y) t d(J z) flj? 2)s LEMMA. o is o(n)-valued. that is.o. Since 0 annihilates the vertical connection with vanishing torsion. . i = 1. Every Riemannian manifold admits a unique metric T$ Hence. 2 . .+ T&) and ?“:i are skew-symmetric in i and j. 1 . RIEMANNIAN CONNECTIONS 159 - j. 158 FOUNDATIONS OF DIFFERENTIAL GbOZIETW'. we obtain g and O(M) the bundle of orthonormal frames over M. d(x. . Let 0 . and the . so is T HEOREM 2 . Let p be an arbitrary metric connection form on PROPOSITION 2 . we have restricted to O(M). i. = 0. a linear connection of M by virtue of This implies that F$ = Fflj. . of R” and the basis Ej.1 of Chapter II. and a skew-symmetric matrix of differential forms oi respectively. be an n-dimensional Riemannian manifold with metric’ structure equation of Theorem 2. for every a c O(n). Let q be the connection form defining another metric connection of the canonical form of L(M) Since (?$ + p&) is symmetric in j and k. With respect to the basis and hence e. We write M with metric g is a metric connection .) are skew-symmetric. Proof. . Since y . 0 its torsion form on O(M). ~1) = 0 only when . i < j. verify that the metric connection defined by o has zero torsion. from the first Let . proving our assertior? I 1’. . .*T = ad (a-‘) (7) We shall present here two proofs. . we shall consider (positive where the FjL’s are functions on O(M).6 of Chapter III and hence is left to the reader.y) = d(y. we have 2. of the Lie dV = --I. vectors. Proposition 1. It is easy to show that R. Uniqueness. so does T = (7. A linear connection of n/i is called ($) and (9. the most important is the Riemannian connection (sometimes called the Levi-Civita connection) We shall show that o = (wj) defines the desired connection.

then f is caileil an isometv yof M onto M’.and f: U --+f(ZT) is a diffeo- morphism. QF. More prekseb.1 of Chapter III.f* Y) for all define V x Y by the following equation : ‘B X. Z) + AK VXZ) TZW’) z+ T. Proof. between A4 and M’.FOUNDATIONS OF DIFFERENTIAL GEOMETRY Let M and A!’ be Riemannian manifolds with Riemannian ! metrics g and g’ respectively. then the following diagram is commutative:. satisfies the equation which defined V. that is. an isometric immersion of a Riemannian for all vytorjields X. Vg = 0 (cf. A mapping f: M i M’ is called Proof (B) . In this case. Proposltlon 2. Since f(M) is ppen in M’. in M.4. (AI’). above definition of V.3. With respect to the Riemannian connection. Being a diffeomorphism immediately from the definition of V. Y) = g’( f*X. K m which we‘call an isometric immersion.aY(fj’)). zs an isometry of a @emannian manifold M the four conditions of Proposition 2. f is an isometry of C’ ontof( C’). To . then the dzxereptial off commutes with the parallel displacement. Y and 2. ‘. By Proposition 2. Thus.8 of Chapter III and hence onto another Riemannian manifold M’. satisfies ’ PROPOSITION 2. Thus.5. T&W A T. Y) V. Z) +‘g(V. Given vector fields X and Y on M. for all vector fields X. IV. where x’ = f(x). I’ is the Riemannian connection of 34. uniqueness of the by virtue of Proposition 2.t satisfies V. the differential off commutes with the parallel displacement Proof.1). Y) == 0 for all Y and hence X = 0. It is easy to verify that I has no torsion and is metric G the course of the proof. VG m PROPWTION 2. . a consequence of the. 49 y> + gv.Y. Iff zs. every point x of 121 has an open neighborhood CT such that f(U) is open in M’. show that l? is a metric connection. onto M’. From V7.z-&F Y) proposition. the set of vector fields on M’.Y E T. f* is injective at x.z d-T Y) f which is isometric at every point of M is thus an immersion + g([X Y]. Y] given in Theorem 5. Z) -i’Y *g(X.X . Y) + V.Y L f-‘( P. we obtained ‘the following .Y. In terms of a local coordinate system x1.(fY) with respect to the Ricman- nian connections of M and M’.(M). Proof.mythe with the parallel displacement. . Existence..4 of Chapter III. th. .Y. ‘then the dzjSerentia1 off commutes determines a linear connection I’ of M by Proposition 7.*. we have This means thatf(V. Y = a/ax’ and Z = 8/8xk in Proposi- QED.f r is a curve from x toy x .5. manifold M into another Riemannian mamfold M’ and . Z) .fields on M and.y’ =_f(y) and 7’ I-~(T). of the Riemannian connection are given by is an immersion. f defines a 1: 1 correspondence between the Uniqueness.D. forward verification that the mapping (X.10 of Chapter III. This is.2. dim 124 = dim M’. 2) i.en. tion 1‘ on M by V.6.3 and use Proposition 7. Since f the components I’j.X .Y.fs(M) is open In M’. . Y] = 0. because f*X = 0 implies that g(X. If f maps.. -7. It is a straightforward verification that if VxY set of vector.Y) = V. RIEMANNIAN CONNECTIONS 161 -. . f is 1: 1. COROLLARY 2. If. It is a straight. A mapping 2g(V. it is sufficient to prove f* f* 4 1 x *g(K z> = g(VxY.Y and the formula T(X. Y]. with respect to g. Y and Z of M. x39 Y) + gv.(M) “. 2) = i*:g(Y.g = 0 and VxY . moreover.5 of Chapter III. Let X = a/W. then) it is called an isometric imbedding of M into M’. . we isometric at a point x of M if g(X. where X and Y are vector fields on M. 160 . The fact that I’ has no torsion followsSfr. . PROPOSITION 2. But this follows Riemannian connection in Theorem 2. Iff. the Riemannianconnection l” on M’. tion 2. .[X. Z) + g(lZ. M 1 : 1 which should hold for every vector field 2 on M. x”. c g([X.[X. 2) = X*g(Y. This implies immediately our 2g( z> . we obtain a linear connrc- .

wehave parallel alang T and since the parallel displacement is isometric. By the parallel displacement of the above property which cover 7. Given an arbitrary curve T from x to-y in o(. +..5 and 2. J .pkyc A dt + .. i = 1. together with Proposition 2. it‘is sufficient to prove that Let 7 = x. (1) For a fixed direction (pl. . . QED. . However. . .) = pi. .ala9. r* is a geodesic of M*.) is independent oft. Let 8 = (V) geodesic of M’. . n. . Proof. Every geodesic of M* where ei and W{ are l-forms on U. p”.x* of M* with p(x*) = x. unique curve r* in M* starting from x* such that p(~*) = 7. every geodesic of M is mapped by f into a cross section U -+ L(M) given by a/ax’. Similar state- ments hold with respect to the corresponding affine connections. . The length of geodesic defined by xi = pit. with functions of p’. coordinate system x1. that a/8x1. Since the tangent vectors it are p(x. . .i are linear combinations of dp’.6. and o = (01) be the canonical form and the Riemannian connec- Example 2. . 2.l(d$ + Cjfijpj) A& + . . . X)$. . then the restricted linear holonomy group of M* with reference PROPOSITION 3. (2) ci$’ do not involve dt. . . given a geodesic 7 from explicitly. RIEMANNIAN CONNECTIONS 163 along any curve in I. = -Xc. . . . 162 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV.1 holds also for afhne holonomy groups. It follows immediately that._ introduce a Riemannian metTIc g* on M* in such a way that p: M* -+ M is an isometric ‘immersion. the length of x1 is constant along T.iannian manifold with metric g. tion form on O(M) respectively. 6(x. i. .. .) e 0. . . where the dots . we can find a finite number of open neighborhoods in M with given by (a/ax’). at the origin x) . 1 R$. (a/ax”). . (1) ei = pi dt + vi..6.. we have &(x0) = pi and hence ei(x. Proposition 2. . . we introduce the polar coordinate system (tt.’ 1 = 0 at t = 0 (i. Set ut = cr(x$ To a vector S. dp” and dt argument. to every point of U. If Ili’lil = 1. We set Let M* be a covering manifold of M with projection r’. Since both u. let T = xt be the Let -21 be a Riemannian manifold with metric g. Nol. Bv a normal coordinate @em at x of a Riemannian manrfold ii. . .mul coodinutes and con&x neighborhoods ponents of the curvature tenqrjeld with respect to the framejeld a. = a*& = (t$). t) x toy in M and a point . we On the other hand. n. . . ei and ~. . . & (~9~)~ = 1. indicate terms not involving dt and &.1..pi dt do not involve dt. we have . .e. Since p is a local isometry.6 were stated with respect to Riemannian (3) r$ =OandcT.) = pi always mean a normal coordinate system x1. . From the definition of the canonical form 0. and xt are parallel along T. i = 1. X* at x such for all t. We define a cross section (I of This means that of do not involve dt. Conversely. For the study of Riemannian manifolds. Let M be a Rier. connections which are special linear connections. n. p”. . prove that e’ . (2) Since ut is horizontal by the construction of 0. . p”). It follows that the differential u along the geodesics through x. . point x* is isomorphic by p to the restricted linear holonomy group do not involve dt. . . shows that ifp(x*) = x. . . . a/axn.)~ + s& = u. . xn at x. .M) over U as follows: Let u be the orthonormal frame at x M. The statement concerning linear holonomy groups in Example d6. we attach an orthonormal frame off commutes with the parallel displacement along 7. where vi.. . Let U be a normal coordinate neighborhood of x with a normal 4(%) = f/&j. A similar Then. . . . the Remark. then t is called’ the canonical parameter of the geodesic 7. To compute these forms projects on a geodesic of M. g(X.. (4) d# = .1. i = 1.e. be a geodesic in M. a/+? form an orthonormal fra’me it x. are the com- 3. t as coefhcients. under the assumption of cross section 0: U -+ O(M) thus defined is more useful than the Proposition 2. . .. We can 6 = a*8 = (@) and il. . . -\ @i. .‘(iJ. . there is a bY 2 = pit. of iz! with reference point x. a/@ may not be orthonormal at other points. . . will be denoted by IlXii.

.. Y) = xi rs$qo’(y) for X.dt. r) onto U(x. x” be a normal coordinate system at x. Every line in N(x. QED. be any piccewise differential curve from x to xi P~Q. . (xi)* < r2. r) is the set of points y l Af such that d(x. :m.. Letx’.2.. it is sufhcient to ‘J. = In terms of dt and @. .and since a(y) is an isomc. . n). where Then evev geodesic 7 = xI.y).3.. y 6 [. the expo- nential map is a diffeomorphism of N(x. * indicate: terms not involving dt.tric mapping of R” onto T.(M). we shall show that the length of T is greater than or equal to 6. t = t(s).(n(J))i(S) for every XE T. a”. WC obtain not involve . 3. From xi (p ‘jz = 1: WC obtain 0 = d(2.pi# = 0 by shbwing that Xi pipi is independent of t.. then Proposition 3.a!12 = (dl)” + Ci (@)2 + 2 Cipirpi dt. The metric tensor g can be expressed by PROPOSITION 3.. . b) be . By the very definition of a normal coordinate system.fact that cxp: N(x. In other words. . r). show that d&pi@) does not involve dt. is a di$omorphism. iet T = x. t) introduced at the beginning of the section. g(X.i. r) dejined by Xi (xi)2 = 9. A (pj dt + c$) From Proposition 3. a 5 s 5 p. r) and such a geodesic is unique.. the point where t = 0). r )can bejoined to x (origin of the coordinate we haye system) by a geodesic bing in U(x. + clj. On the other hand. . -. To prove (2). N(x..2. because ($) is skew-symmetric. d(p” dt + cp’) = -Cj 6. . let ((I’. by Proposition Let p’ =p”(s). U(x.1. PROPOSITION 3. we set and hence (4).ijpj = 0. we have vi(&) = 0 and Gi(&. r) through x by the exponential map and By Proposition 3... . (3) li(x. ..(M).kL(p” dt + @) A (p” dt + q-9) perpendicular to the sphere S(x.1.. . . QED. bc the geodesic with the initial condition .1.J.@ A fir = C. Then we have (1) Ever)tpointy in U( x* . Let r be a positive number such that ds2 = (dt)R -t C. This prolcs that d(Ciipiq’) does (4) From the structure equations. we obtain d$ A -& c. we can cxprcss the metric tensor g as lJ(x. we shall prove that ($l. a 5 s 6 /I. r) = the neighborhood of x in M dcfincd by X. the classical expression ds2 = C gij dx’ dx’ for g as explained in Example 3. RIEMANNIAN CONNECTIONS 165 (3) Given any unit vector X at x (i. . .x. exp: N(x.(x.(M) defined by ]/XII < r. (2) The length of th e geo d esic in (1) is equal to the distance d(x.dinate system Since # = 0 at t... r) Proof.2 implies the following inequalities: where the dots . We have. y) -=z r. the coordinates ofy with respect to the polar coo. @. (pa)z). .1. . xi = ait (i = 1.. r) the* neighborhood of 0 in T.) = 0. $7 of Chapter III) For each small positive number r. (cp’)‘.Y z T. \ 164 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV. follows (cf. 6. we have vice versa. . Since . . be &c equation of the curve T.jpih.Rj.e. . Let C. If we denote by L( T ) the iength of 7. r) -+ U(x. By (1) and (2). through x is “I$ = C. Since O(Xj -. = 0 by Proposition 3. r) --) .4. X) so Ci. = 2 Xi pi dp’. .p” = p’(S).(M) and y E U. (cf.’ does not involve dt by Proposition 3.. . . r) through the origin 0 is mapped into a geodesic in t’(x. that X . p”. PROPOSITION 3. (1) follows from the . ds” = & (fi)“. Proof.1 of Chapter I). r) is a diffeomorphism. Now.

From Proposition 3. .4. p).5. By taking an open refinement if necessary. d(y. For each y in U(x. (2) Identifying every pointy of M with the zero vector at y. Then the family B = U B3. r). P) . (2) {Vi} is Zocally~ nit e in the sense that every point of M has a neighbor- We now proceed to prove the existence of a convex neighborhood hood which meets only aJinite number of V.j) + d(y. The second assertion follows from (3) of As an application of Theorem 3. . p) first point on T which belongs to the closure of U(x. y) < r. Take any (1) Any two points of U(x.. There exists a positive n&ber a such that.. z) 1 d(x. and it is the unique geodesic joining the two points and lying in neighborhood ofx (in the sense ofTheorem 3. we. in some UL. Z NV into (y. p). around each point of a Riemannian manifold in the following (3) any nonemptyjnite intersection of J:‘s is di$eomorphic with an open form. . If_y and z are points of U(x. Set U(X.) of M has an open reJinement {Vi} such that \ A geodesic joining two points x and y of a Riemannian manifold (1) each Vi has compact closure.(M) such that exp: NV -+ U(x. closure. r) but not to is a diffeomorphism (cf.} is looally finite and that each $!. r). Hence. which sends Y .‘s. then. z) = PROPOSITION 3. Then every open covering { 17. clearly. covers 0:. z) is a di$krentiabZe function ofy and z. Suppose T does not lie in U(x. p) x U(x. it follows that if x fy. For each x E M. then index set) such that 0: c U. It follows that the intersection is diffeomorphic geodesic joining y and z in M. Let y’ be the the neighborhood of y in T. The length of T from v= u N. M is called minimizing if its length is equal to the distance d(x. we may THEOREM 3.f n V. If V. As we remarked earlier (cf.y) < r and let T be a curve from x toy such consider y as a point of T(h4). then d(x. If z = exp Y. For each M.y’) = r by (1) and (2). (2) of Theorem 8. Let (Ui} b e an open refinement of {U. metric) on A4 ((Y/l. Thus d is a metric. p) and that the length of r is equal to the distance. n . . &y) = d(y. I ‘i . has compact of a lkemannian manifold M. p). Theorem 8. s ince IIY/12 is a differentiable function on d(x. we Conversely. y) 2 6. for all cn (cf.. d(x. Then. p) x U(x. It is clear that 7 is the UnkpK minimizing V.. cell of R”. Proposition 8. QED. Ify is in U(x. let IV. we have WC P) x wf .7 Since 0: is compact. which is a contradiction.1 of Chapter III). I/ YII is the function on V which corresponds and d&es the same topology as the manifold topology of M.y) > 0. r). are members of ‘B and if x and neighborhood of-v (cf. . y). QED.6. In other words. . Then the mapping v--f U(x. (1) Let a be the positive number given in Theorem 8. Since the lying in U(x. k. Appendix 3). r). V. . 4.4. .. I’.7 of Chapter III).6. IVZ n GG is non-empty}. i = 1. let (‘2) In U(x. d(x.7. QED. p). p) by the same ment of {UJ. is a desired open refine- they can be joined by a geodesic 7 lying in U(x. QED. T H E O R E M 3. . .y) is a distance function (i. we see from y are points of the intersection VI n *. it lies in the intersection that is.. then d(y. %I& = {IV. let NV be that L( T ) < r. . Let M be a paracompact dlyerentiable manifold. it is clear From the construction that % theorem.. geodesic lies in each Fi. then there is a Proposition 3. let d(x. L(T) 2 r.6) which is contained U(x. d(x. to the distance function d(y. p) is contained in a normal coordiwte satisfies (1) and (2). p) can be joined by a unique minimizing Riemannian metric on M. PI. z) under the diffeomorphism V --f Proof. there is a finite subfamily %. of 2Ba which of Chapter III and‘let 0 < p < a. exp Y) is a diffeomorphism (cf. 7 is minimizing. z)” is a differentiable function .e. d(x. p ) x toy’ is at least r. r) and hence y is in U(x. In fact. .4 that 7 is a unique geodesic joining? and z and unique minimizing geodesic joining x and y in M. RIEMANNIAN CONNECTIONS 167 We shall now prove (3). with an open cell of R”. Proof. 2).obtain the following Proposition 3. the end of $l). be a convex geodesic. 166 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV.7 of Chapter III). Since U(x. X” be a normal coordinate system at x assume that (U. the square of the distance d(y. lKZ’kGP) Thus -r lies entirely in U(x.on U(x. * n V.) (with the same $0 -=c p < a. Proof. Let x1.

by xi..). 3. such thilt. x. Finally. isomety. Proof.?) 1 /.f(j*)) fur nfl x. respectively. Let x be an arbitrary Remark.c whcrc d is the distance function. I.(1 is constant hcncc is a g:ro&sic in 1. If X.-llj /. ~(0 4. #(a + e)).j. ‘l’hus \v(‘ ha\.. I d(. -.!. x.. Ict T be COROL. lvith the iriitial condition (x.I.. 7. If.y) -: d’(. 6.J is parametrized by the arc 2.-) c I. is a geodesic with arc length t as hood li’ of x’ let I. -t.._lj). QED.. a gcotlcsic in 1. and t. moreover. X(C)) + 444. we obtain The parametrizatitin of a segment may not be affmc. Then ever_v segment is a geodesic d(x. RIEMANNIAN (:ONNE(:‘I‘IONS 169 168 FOUNDATIONS OF DIFFERENTIAL GEOMETRY arcs into which T is divided by these points by T. x2) i. t.f’(.. In any metric space Af.) L. nix*.... is a continuous curvr such that d(x. ~f(x.r”g.c. Let T = x.ric including the pnmmetri2ntion.} satisfying (l). x3 = L(r.8. of A4 (cf. g b. for a 5 t.f( l..\I and .t 5 b. we see that x(t) is a gcodcsic for n 1 : 6..9. .J + d(x. ~\‘e see that g und d the distance function dejined by g.Y).“.. is a segment. Le. Proof. z4(72) : L(T:$ :.. Using Proposition 3. A covering (I’.10.’ CI -t F. x&J) -’ L(p) -1.(T. a . J‘(T) = f(x. For a normal coordinate nrighbor- It. f oj‘ -II onto itself which preserves d is an this argument. wc lla\. be a normal coordinntc ncighborhoorl of x parameter. XJ = L(T. u : . x1) L(T.u. 5 t. This Al onto AI’ ~ehich nrt~ps the tensorjield g into the tensorJield g’. ~3).)) < d@(a). Let T be dctcrminrs the Kicmannian metric.&!. u + E. a < c < a + E. on T. Its usefulness lies in the fact that the etch d(x. that x(t) is a gcodcsic for a r: t 2: a + E foisome positive E. -.g c j 21. Since 7 = x.~ is param- cllot!q 7. Weil [ 11).) 4x(h).. 3.. the minimizing geodesic from X(U) to x(n + c. IV. \ etrizcd by the arc length s a n d since d’( f(xS. x3) = d(x.<. x(b)) + Wt2)9 x(h)) = 4x(h). Ckarly. be a piecewise dtferen.J‘is a homeomorphism. we shall show that the distance fknction There exists E > 0 such that x(t) c Ufor a 2: t I. x:J L ( 73) 9 cohomology of M can be computed by means of a simple covering d(s. As an application of Theorem d(. then f is a di$eomorphism of contradicting the fact that x(t). Let A4 be a Riemannian manifold with metric Similarly.).(T2j.6.c a simple covering..j‘(T) is a segment with respect to d’ and d(a.sfl.\I ant1 set x’ = J(X)..l for all t.J).. IJ f is a mopping (which is not assumed to be corrtirruous or dtJerentiabluj of zVl onto AP’ such that Ma). f(xSL)) = Proof.) + djx. 2. then 7 is 0 geodr. We shall show 'rtIEORES1 3. i = 1.8. xi) -i /f(x. lf(. Riemnnniun n2ftrir. such that X(C) is not.~) = point of . n Hz t . a segment is defined to be a continuous If we did not ha\rc the cclliality (*\cry~h~. a :. (as a point set).6.d(x. which is a contradiction.7:1). Let x(t). By continuing In particular.-1. (2) and (3) is called d and d’ be the distance Then jiunctions of . t. that is. x2) = L(7..s g and g’.. 7:) and TV Remark. d(x...: t 5’ a $ E.\I’ be Riemcmnian manijolds with that T and x(t). x.L(7j) :=: L(7) 2: d(S. 6.” with oriain x’. X)...AKY 3. coincide as a point set. and the .11 end . d(x.r. WC: ~oultl have image x(t) of a closed interval a 2 :I: b such that Il d(x. . Tlren 7 is II geodesic 0s n point set.r t L’ n + E. I. 5 t. shows that x(t) is a geodesic for a r* t . 1/1.t.. x. We first show .\I’ res/)Pctice!y. Suppose there is a number c. It suiliccs to show that 7 is a scgmcnt. I”or anv unit tangent \rcctor A’ at x../ t. x. er*erl’ tr. Let U be a convex neighborhood of x(a) in the sense of Theorem 3.).).t n . then X. Denoting the points x. we have a PROPOSITION 3. bc a segment in. Since 7 is a linbfe curz’e oJcfuss C’ from x toy such that its Imgth L(r) is equal to scgmc’nt with rcspcct to d.).

we have proportion.. . L EMMA . sin $a = sin :a’ Hence we have and hence / . Y) respectively. . Both exp. the lemma implies that .( M’) .at x’. yJ.. is reversed can be treated in a similar manner. (M’) mapping. . Set we shall obtain a contradiction. form an orthonormal basis for T.. X) and such that (x. RIEMANNIAN I I length s also.. of T’(M) into T.(M). . yJ > c sin +a.with the initial conditions sense that (x’. F( Y)). = exp 2. Sincefhas an inverse which also preserves the distance g’(F(X + Y).(M) onto = g’(W). sin &a. Set s-o 4. X.. Let h be the Riemannian metric in U given that F is a linear isometric mapping of T.. yJ > sin 1% vectors..) Choose c > 1 Let x. Then both F(X) and F(Y) are unit vectors at x’. may assume that k h < g < ch on U in the Then xi and y: are the geodesics .. Given X and Y in Tz(M). Y) . . (The case where the inequality cos a = g(X. we. Thus. n.(M’).Xi) = g(X. =f(xJ and A =f(r& Taking U small. F is a mapping of the set of unit tangent vectors at x into Let X. exp. we may assume that both X and Y are unit iij $ d(x.yJ and sin $a’ = lim . h is a Euclidean metric and hence We shall now prove that F is lineal. 4 -=I d(y.(M) and z z U. x:) T... . we shall complete thhroofofour theorem.(M)’ and for any constant. andy. (resp. It is also clear that foexp. are diffeomorphisms.) + g’(V). . T. F(Y)).. by xi (dx i ) 2 and let 6(y. Lemma. Y c 7’. Supposing that X.2 sin2 ta = 1 .. Since F(cX) = cF(X) for any X E 7’. r WNIJA 1 IUNS OF DIFFERENTIAL GEOMETRY 171 CONNECTIONS 1 I” IV-. . 0 F and IIF(X)II = (IXJI for X E T.Xi) + g(Y. X:) = g(X + Y.). 5 where exp.) functions. .2 sin2 $a’ = cos a’ =.YJ > ?- 2$ d(x.) is the exponential map of a neighborhood of 0 in r++M) (resp. 2) < g(Z. U’). . it is therefore sufficient to show system x1. i = 1..(M). and hence F(X + Y) = F(X) + F(Y).yJ > c sin $a for small s. denoted by the same F. for every i. X. . 2) < 4 h(Z.J d(x:. and Proof of. F(Y)) for all respect to h. F(Y)).g’(F(X). !.(M’)) onto U (resp. be an orthonormal basis for T. be the geodesics with the initial conditions (x. c.l’ which maps g into g’. To prove thatfis a diffeomorphism of Let U be a coordinate neighborhood with a normal coordinate M onto h. Assuming the From the definition of the distances d and 6. X.= g’(F (X). 2) for Z E T. we obtain lemma for . F(X)) and (x’. .(M). Y) = cos a = -1 . exp. sin 4 a = li-i ix d(x. Then the set of unit tangent vectors . Let F(X) be the unit vector tangent tof(~) at x’.(M) and any constant c.(M’) by as we have just proved.. x..the moment. it is clear that F is a 1 :I mapping of T. . Since f preserves distance. respectively. ’ C s-o We ‘shall give the proof of the lemma shortly. It is sufficient to prove the first formula. We already observed that F(cX) = cF(X) fbr any Xr T. Y) and cos a’ = g’(F (X). X:) = g’(F(x) + F(Y). both parametrized by their arc length from ii21 i d(x.%. It can be extended to a ‘Xi = F(X.(M) onto T. g(X. 4 < c ICY> 4. xn at x. On the other hand. h(Z. z) be the distance between y and z with We first prove that g(X. KY.

Let x. For a connected Riemannian maniJi. We first prove that E(r*) = S(r*)..X.. Let r* be the supremum of r... 0 < I C E ... (3) Every bounded subset of M (with respect to d) ir rel&e!l.. RIEMANNIAN CONNECTIONS 173 This is a contradiction.Xo.. let is said to be con$iete if the Riemannian connection is complete. we have ing conditions are mutually equivalent: 1 ( 1) M is a xomplete Riemannian man$old. s i I. the proof is and adopted from Palais 121. .l. y) as we wish. we can extend the geodesic for the values of s such that L 5 s 5 L + E for some Similarly. (The existence of such a sequence {yi} follows from the fact that x and y can be joined by a curve whose so that (x8”} is a Cauchy sequence in . is defined. Let Is.ll)) skirting from the origin. 7( be.. there is a curve r in hl starting from x which is It follows that {yi} converges to y0 and hence that d(x. assume where s is the canonical parameter. For each l&r .. This proves the compactness of E(r).2. we obtain & > 0. Let x be any point of M.ld A\i.. = e x p r. existence of a normal coordinate..for some E Y 0. d(x. where Y < r*.) = I. The limit point x is independent some E(Y ).’ Is. xs. $6 of Chapter III).) converges to a non-negative number Y. Since M is a complete Riemannian manifold. be a sequence of points of E(r) and.. Conzpbctencss S(Y) for every Y > 0. 1 .y.I. 1 (ii) Proof of Theorem 4.x andy of A4 can be-joined by a minimizing geodesic.JJ = sin $a.61.. .. T HEOREM 4 . Since ri is minimizing.10 is C~UC to Myers and Stecnrod [1] . This implies that the geodesic exp .rX. . Theorem 3. that d(x. of the choice of a sequence {s. 5 h d(x. WC show that this geodesic “. we know that E(r) = S(r) for steps. Let Xi be the unit vector that is..y. we set QED.) Then each yi belongs to hence converges to a point. is minimizing THEOREM 4 .(. tile afine tangent sbace .a minimizing geodesic from x toyi..v i= 1. be an infinite scouc’ncc such point ofS(r*) and let {yi} bc a sequence of points with d(x. yi = exp (d(x. Then 4x... exp r. we divide the proofs of these theorems into several hood around x (cf. yi) d Y for all i.Yi)Xi)* (2) M is a complete metric space with respect to the distance frtnction d. We set (4) For an arbitrary point x of M and for an arbitrq curvy C’ in the y. > 0 (i) The implication (i) ---+ (1). By taking a subsequence if necessary. To show that r* = 00.E(r) = S(rJ for r < r.. assume that {Xi} converges to a unit vector X0 in T. /fM is a connected coniplete Riemannian manzfold. We set xL = x. tangent space T. . Le! y be a can be extended beyond L. By using a normal coordinate system at x.. t L. y) I r} Theorem 3. E(i) = {Y E S(y) . * -4 2 s Y > 0. and that y0 is in E(r).(M) ( or more precisely. ” A Riemannian manifold A4 or ri Riemannian metric . 2 . if every geodesic of A4 can be extcnrlccl for arbitrarily tangent to 7i at X.. say x.. compact.} converging to L. We are going to prove that E (r) is compact and coincides with 4.y.. 0 ..r.system an&a convex neighbor- Proof. such thpt . we may large values of its canonical parameter (cf. bc a gcodcsic. Now’we shall prove that E(r) = S’(r) for $1 r > 0. the follow.To prove the compactness of E(r). and hence each yi belongs to E(Y*). let yi.. Since We shall prove the following two imp&ant theorems.g on .) .. we may assume. again by taking a subsequence if necessary. S(r) = {y c M. Hence.Y can be joined to x by a minimizing geodesic}. By the then any two points .. yi) < r* tFiat s.’ s 2 Y.. 172 FOUNDATIONS OF DIFFERENTIAL GEO$fETRY IV.J... 0 .d(x.’ that r**Y -00.AI with respect to d and length is as close to d(x.) = sin f~.(M). d(x. . which converges to y.2. developed upon the given curve C. for each i.

This contradicts the Since E(r*) is compact.. It is easy to see that limt b x t=y. then obviously Ir(z) .) = r* and d(x.y’) +‘d(y’.y) I 6 5 r(y’). Hence r(y) 2 r(z) - We want to SLOW that b = a.y) 2 r(z). completing theproofof the that lim x1 in M. let T(Z) be the con.y)..d(z. Then the curve xt.(M) ( or more precisely. in M. Theorem 3.. which contradicts the definition of r*. 0 5 t 2 b + 6.9. (iv) The implication (3) -+ (2) is evident.y).yJ + d(yk. there x of a connected Riemannian manifold M are in@&@ extendable.y’) + d(y’. Let Cit. z E M. I_ t d t. along the curve xt. tangent space. there exists a positive point x can be extended’infinitely. . . tamedmS. S(r* + S) = E(r* + S)..y).. y) = d(x. We know that Y(Z) > r(y). 0 5 t 2 b.y).3..y). Let 6 be the supremum of such E > 0. by the affine (not linear!) parallel displacement of the curve C. the distance E(Y*).2. To complete the proof of Theorem 4x t .} is a Cauchy sequence in M. 7 is a geodesic.(M) starting from the origin. If d(z. If T(Z) = co for some point t. where L(T~) is the length of TV. serves the a ‘c length.2.> r* and that d(x.(M). Let t. Since we know show first that there exists a point y’ in S(r*) the.r(y)/ < there is E > 0 /such that C.y) < r(z). Let yk be the last point on 7k which belongs to E(r*) = S(r*). On the other hand.(r(z)).y) + k.y) d L(7J < d(x. then SJr’) = {y’. Lety &S(r* + 6) Thii im”plies that {x~. it . Then L( T ) = L(T’) + L(T”) = d(x. then M is a minimizing geodesic T” fromy’ toy.(r(z)) = (iii) The implication (1) --+ (3) in Theorem 4.y) I r and d(z. y) j $. such that any two points of S. On a Riemannian manifold M. thatis. For each 2 6 M.d(y’. 0 5 t 5 b + 6. ) is less than or equal to the length of xt. let r(z) be’the supremum of Every bounded subset of M is contained m S(r) for some Y. if necessary. it is obvious that every geodesic can be extended We shali prove the continuity of r(z) by showing that [r(i) . in A. Since the development pre- lemma. say y’.is sufficient to assume that every geodesic issuing from the particular LEMMA. Hence S(r*) = 7”. Since a geodesic is a curve in of M and any positive continuous function on h satisfies the M which is developed upon a straight line (or a segment) in the condition of the lemma. Then C.. of &(r*): We have d(x. some 6 > 0. Hence y 6 E(r* + 6)) completing the proof of Theorem 4. By E(r*). RIEMANNIAN CONNECTIONS 175 174 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Since E(r*) is compact. a minimizing geodesic from x toy. There exist 6 > 0 and a curve xt. If d(z. of a convex neighborhood (cf. QED.Let this end. To prove that E(r) = S(r) is compact for every r. t-4 Going back to the proof of Theorem 4. 5 t $ t. is the origin of T. f 6. is the development of a d(y). d(y. (vi) The implication (1) + (4). where r’ = T(Z) .y’) s r’] is con- curve xt. that (yk} converges to a point. T > 0 such that any two points y and y’ with d(z. Let 7’ be a COROLLARY 4. proved that (1) implies that E(r) = S(r) is @pact for every r. which is Thend(x. b 5 t g b + 6.(M)) obtained y such that L(TJ < d(x. the length of xt. be an 5 d(t. t. The existence regardless of the point x we choose in the proof of (ii). implication (1) -+ (3) by (iii) and (iv). developed upon Cl. t. Let 7 be the join of T’ and is complete. in fact. We need the following Remark.y’) A.y’) s r can be joined by a minimizing geodesic..6) implies that ~(2) > 0. we may assume. x. If all geodesics starling from any particular point minimizing geodesic from x toi”.r(y)1 5 d(z. To converged to a point. y).y) = d(x. y belongs to E(r*). is developed upon C. 0 I t I a. Next we shall show that S(r) = I for r < r* + 6 for Corollary 3. First we show d(z.1. In (ii) we can be joined by a minimizing geodesic.y). Ir(z) . continuousfunction r(z).qr.y) = d(x. b 4 t 5 b + 6.r(y) 1 infinitely. we see that {xt. say y. is equal to the tinuous function given in the lemma and let 6 be the minimum of length of 8 t. Assume that T(Z) < 03 for every z E M. positive integer’& choose a curve TV from x to Cl be the curve in T. Since d(-y’. t. IV. Without any loss of generality. then r(y) = co for every pointy (v) The implication (4) + (1). by taking a subsequence definition of b. 5 t I t.y’) = r* and d(x. 0 g t I E. Assume that 6 < a. 0 5 1 5 E. for every. we may assume that arbitrary curvein T.

. jection. can be Choose r > 0 such that exp: N(x. 2~) % U(x. 27) % U(x. r) = (X E 7’.y*). Chapter III). In particular. for every unit vector Xat x. There exists T > 0 such that.i unit vector of M” at x* such that p(X*) = X. then M* is complete. Let x be a point of a homogeneous Riemannian mani. }. Proposition 8.1. p maps . QED. 2r). Example 1. we have isometry of M which maps x into x. The proof is divided into several steps. Every compact Riemannian manifold is complete.k 1 Xa t s = v(exp sX) for 0 d s 2 r. so is exp sX. r) = {X” E T. For a given x E M and for each positive number r.y) < r}.*. (ii) If M* is complete. fold M. that p(Y*) = Y. be the set p-‘(x). Let a. complete. u(x* . x.. 0 2 s 5 a.U* onto M. Now. For each XT.2. Y).. where Y is a. by Corollary 4. r). Then exp sX = p(exp sX*) is the geodesic \ p(exp aY *) = x and hence exp aY* = XT for some xr. be a minimizing geodesic from y to x. 11X/l < r}. for x* E M”. -co < s < 03. Let p: M* -+ M be an isometric immersion. Sety* . transformations preserving the metric tensor g. r) is empty if xf # XT. Let Y* be the unit vector aty* such (i) If M* is complete so is M. be a COROLLARY 4. i. where X is a unit vector at x. IjXll < r).e. N(x. Since exp sX is a geodesic through 5 cp(exp sX) is a geodesic through N(x. Let exp sY. 2r) is a diffeo- extended to a geodesic defined for 0 I s 5 a + r.. be any geodesic with canonical parameter s in M. r) is a diffeomorphism for every i. On . d(x*.5 follows also from the gerieral fact that every locally compact homogeneous metric space is complete. . (a) follows from the fact that both p: N(xf . x. Such a geodesic Proof. 0 5 s 2 a. r j = {y* c M*. 2r) --f N(x.fp: M* -+ M is a covering projection and . RIEMANNIAN CONNECTIONS 177 Proof. Since exp sX* is defined dently. E(r) = S( Y ) is compact for every r.3. of isometries. (1) If M* is complete.(M*). We shall show that T = x. To prove (b). we set. we set Chapter VI.3 and Theorem 3. such that p(X*) = X. As we remarked at the end of (ii) in the proof of Theorem 4. be an morphism..y* l U(x. into a unit vector X at x: X = pl-l(&). . the geodesic exp sX is defined for IsI $ r (cf. Proof. subset of M is contained in S(r) for some r and hence is relatively complete... r).a at x. This follows from the implication (3) -+ ( 1) in Theorem existst by Theorem 4. r)) = u U(xf. . 6 .. Proof..i. It is sufficient to prove the following three statements: Theorem 4. Given a point y of M. Let {xr. . 2r) . (c) U(xa . then p: M* -+ M is a covering pro- of M is transitive on M.r) n U(xj. then M* is a covering space of M with pro. T HEOREM 4.. Then exp sY*.4. A Riemannian manifold M is said to be homogeneous if the group Thenp(y*) = exp aX =y. * compact. M is.*.4. Let x* E M* and x = p(. Every homogeneous Riemannian manifold is complete. (a) p: U(xT. . geod sic from x toy. and exp: N(x.) U(x.y*) < r}. QED.. . r) = { y E M . r)) and sety = p(y*). . 27) -+ U(x. 0 5 s 5 a. of the same dimension. proving that p-l( U(x. let y* up-l(U(x. X). . QED. 2r) ----t U(x. Let M and M* be connected Riemannian manifolds (b) p+(U(x. Y)) c U U(xf . Evi- in Al with the initial condition (x. (iii) If M* is complete. r) + U(x. Let x* E M* and set x = P(X*).2 since M is complete by (i).(A4). (Cf. Let T = x. Let X* be the 4. 0 d s d a + r. 2~) are diffeomorphisms in the above (2) Conversely. Then v-1 maps the unit the following commutative diagram : vector i. let exp sX. is a geodesic in M* Let X be any unit vector of M at x and choose a unit vector X* at X* starting from y* such that p(exp sY*) = exp sY. Similarly. Then T = x. is a geodesic.5. 2r) jection p and M is also co?plete.. 0 5 s d a.1 of N(x* . Every bounded for all s.unit vector aty. We set ?. .exp ax*. N(x. d(x. T HEOREM 4 . 176 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV.f M is diagram.

.x* and y* of diagram. QED. Y) . of M.4. Sincef maps diffeomorphism.178 FOUNDATIONS OF DIFFERENTAL GEOMETRY IV. 2r) -+ U(x.. Take any Riemannian metric g on M. prolongeable. Let d be. Then x8’. Using the above connected component of G(x). the distance function defined by the Rieman- nian ‘metric of ti and d* the . RIEMANNIAN CONNECTIONS 179 L the other hand.distance ‘function defined by the COROLLARY 4. let M be a connected Riemannian plane. To show so is M. Using a normal coordinate system at x. Then x = lim x. there is a unique curve x:..prolo. y* E U(xf . It is easy to see d* (xs. H ence M’ is a homogeneous Rieman- (iv) Proof of (2). (QED. From the above observation.4. QED. Example 4.. Set X = p(X*).at G be a group isometrics of a connected we shall study Y(x) and local structures of a reducible Riemannian Riemannian manifold M. every connected component of G(x) onto a connected component of’G(x). be. we have shown that p: U(xf. Let-M* be a. 2~). As an open submanifold of the Euclidean Throughout this section.1).ngeable. Let s. in M and given a point x$ in M* such that p($) = x0. b 5 s < a. 0 5 s s a. Theorem that this geodesic can be extended to a geodesic x. I or irreducible as a linear group acting on T. is defined. r) for every open set of M. {x3.T. Since p(xf) = p(xT). Since d(x. be the supremum of s such that x. The converse is not true. it is obvious that p(U(xt. is a Zongeable if it cannot be isometrically imbedded into another geodesic in’ M *. let M be the 5 . QED. If the orbit G(x) point x of M contains an manifold. X8. M is i and hence p-‘( U(x. say x. . and let a. It is easy to see that G(x) is opei in M. M has a natural Riemannian metric which is obviously not manifold with metric g and :Y’(x) the linear or homogeneous complete. manifold * Hence. in MT such that \ M* = M .x. l. For example. 0 5 t s a. This shows t/ at M* is complete. With respect to the natural Riemannian metric on M* holonomy group of the Riemannian connection with reference (cf. is closed in I!. Let x* e M* and let X* be any unit PROPOSITION 4. M* Js not complete by Theorem 4. Example 2. that a = cc. ‘be a geodesic in M* same dimension and let p : M* + M be an immersion. In this section.*. that is. Observe first that. 0 5 t g a. It can be point x E M (cf. If M is complete. Thin xi* E U($. suppose homogeneous.> is a Cauchy sequence in . f (M*) = M*. r)) 1 &l U(xt . r)) c U(x. M is said to be reducible or irreducible according as Y’(x) is reducible . Let M be a Riemannian manifold and M* a vector at X*. %f has a neighborhood Usuch that every connected!omponent of U 0 M* we see that there is a nique curve x. Since the connected compact. the point XI belongs to M *.. Let x.7.6.2f and that there is a unique Riemannian metric g* on M* such that p hence converges to a point./j n. Then shown that M* is non-.. we must have x: = x7. assume a < cc. Let U is an isometric immersion.. For any two points..9. (with respect to the topocogy of M*) is closed in U. so such that xz = x* an that p(e) = exp sX. If M* is compact. the property stated :n Proposi- p is a covering projection by Theorem 4. exp sX*.6 shows that every complete Riemannian manifold is non.(M). T a. a neighborhood of x in M with. = x. x. 5. Let M and M* be connected manifolds of the induced Riemannian metric of A@. $4 of Chapter II and $3 of Chapter III). Then . b g s < a. 0 ZII s 2 a $ 6. p(xf) = x1 for 0 s t 5 4. To prove (c). flolonom~~ group Euclidean plane with -origin removed and M* the universai covering space of A4. 7) n U(xj’ .. then the orbit G(x) coincides with M. there is an3elementf of G such that/(x*) =y*. in M* . r). and p is a covering projection. the geodesic submanifold of M which is loca@ closed in the sense that every $oint x of exp sXis defined for -co < s < ao. Proof.) 5 Is. lies in Ufor some 6. Proof. Set . x: = is M* with respect to the induced metric.8.. A Riemannian manifold is said to be non-pro.. Since M is complete. we see Riemannian manifold as a proper open submanifold.) 5 Proof. -co < s < 4. 2~) is a M*.1. Since M* is complete by Corollary 4.. Eyidently. given a curve nian manifold isometrically imbedded into M as an open iub- xt. for some 6 5 (0. componknt of M* r\ U containing x.6 and hence M is ti. Assume that p : M* + M is a covering pro- jection and that M is complete. COROLLARY 4.

x.1 of Chapter III). . to x.%.Y .~ . j = respect to the induced metric of N. let T: be a non-trivial subspace of T. T. then N has a naturally and Y arc vector fields belonging to T’.Y. Lemma 2 for Theorem 7. so that d(x. that T is a minimizing geodesic from x = x. then the naturally induced linear \ integral curve of X starting from an arbitrary pointy. By Corollary 3. with a linear connection whose torsion vanishes and if N is a 1’0 prove that T’ is involutive. . (a’. Then M’ is a totally geodesic submanifold of M.. RIEMANNIAN CONNECTIONS 181 180 FOUNDATIONS OF DIFFERENTIAL GEOMETRY the point y = x0.) belongs to Ti. = rxi for 2 E U. so are 0.1. Let 7 = xt displacement along T. Let x. Considering only small values of t. where L(T) is the arc length of 7.) >= d(x. (1) To prove that T’ is differentiable. (XF). M’. IV. It now suffices to show that involutive. -we see that & belongs to T& for every t and hence the parallel displacement along .N be a total& geodesic submanifold of a Riemannian T = f. . T( TJ = T.9. We thus obtain a distribution T’ which proves that M’ is a totally geodesic submanifold of M. so is A submanifold N of a Riemannian manifold (or more generally. induced linear connection such that every geodesic of N is a because the Riemannian connection has no torsion and [X. let y be any L(T). be connection of N is the Riemannian connection with respect to the parallel displacement along this curve from the point xt to . it is metric of N is a geodesic in M.y. Let x E N and Xr T. The subspace Ti of T.Y) = ‘Ii? f (T~Y. X). the geodesic L E M M A . From the assigns to each pointy of M the subspace Ti of T. Given a pointy B M. . Since Y. 1 ’ *> n. (1) The distribution T’ is di$erentiable and Since N is totally geodesic. Assuming that Mis reducible. X) lies in N for small values of manifold M. we may conclude that. (1) If M !s a manifold form a basis of Ti for every point z of U. be a basis for T. we obtain d’(x. xJ = Proof. X ). The arc length of 7 point of ikl and x1. a”) being the coordinates of z. T lies in N. X. Xz facts which will be proved in Volume II. .X (cf. it is’sufhcient to prove that if X totally geodesic submanifold of M. Hence. From the definition of the For each i. Y ] = geodesic of M. Theorem 5. 7 is a geodesic with where 7 is the geodesic from y to z given by xj = ait.(M) is independent be a geodesic of M with the initial condition (y. we define a vector field XT in U by distance functions d and d’. Every geodesic of N with respect to the induced Riemannian the parameter t. The lemma is a consequence of the following two a difIerentiable vector field Xi* in U. measured by the metric of M is the same as the one measured with valid in a neighborhood U ofy. where of the choice of T. . QED. then if!&fk$nd XE TJM’) = 7$ Since the tangent vectors i’t are P -r . . --+ curve from x toy and TI the image of TL by the (linear) parallel (2) Let M’ be a maximal integral manifold of T’. called a totally geodesic submanifold of n/i. Since the parallel dtsplaccment 7 depends differentrably on (a’. a=).(M) which is invariant by ‘F(x). . x. T is a closed curve at x and the subspace TL is invariant by parallel along T. . let T be a (V. be the geodesic of M with the initial condition (x.) = L(T). 1 ~5 i : k. In fact. If M is complete. . Let d and d’ be the distance functions of M and N M. . x” a normal coordinate system at y. Let of. belong to T’ for eve+ t. . . respectively. . . 0 d t s a. that is. .. if . This and hence T( Ti) = p( Ti).(N). .zu-’ . p-l . and YE. of M determined by (x.(N). d’(x.. be the Riemannian manifold M. PROPOSITION 5.U is any other curve from x toy. T is a geodesic of N with respect to the induced Riemannian (2) Let M’ be the maximal integral manifold of T’ through a point of metric of N. Let . for every X E T. .(M). . we obtain Remark. .. following lemma. if M is complete. we may assume so is M’ with respect to the induced metric..) = L(T). .2 of Chapter II). a manifold with a linear connection) M is said to be totally geodesic at a point x of N if. respect to the induced metric of N. Proof of Lemma. . Let X. Y and VrX. Let T = xt. x. T lies in M’ (cf. It is clear that XT. If N is totally geodesic at every point of N. (2) If N is a totally geodesic submanifold of a C .

k vectors lie in T:.. A Vx. .” and that Y”(x) is 1 5 i 5 ki define an integral manifold of T”. a/i3xn) form local bases for T’ and T” respectively. .(M) which coincides with a. In other words. xk+l. k + 1 5 j I n). In other words. and T” are orthogonal to each other at every point. there existis a local = g(V. Xj) +-g(Xi. we show that each other. belong to T” and since T’ and where V’ (resp. T’)J. to T’ and X. yk. 1 2 i 5 k.“) ‘and which is trivial on Tz (resp.] = 0. . Let M’ and M” be i. xn is a local coordinate system with the Proof. .(M) such that the first k vectors lie in TL and the remaining n . It is easy to see that trivial on Ti. a2) be the restriction of a to TL‘ (resp. we see that VXnXi belongs to T’ and that I.1 holds under the weaker assumption that M is a manifold with a linear connec. . Y’( X) andX”(x) such that Y’(X) is trivial on T. To prove that the Riemannian metric of distributions T’ and T” are complementary and orthogonal to V is the direct product of those in V’ and V”. . Hence. there exists a local coordinate system x1. . 9 = 0. independent of 9. . Note that Proposition 5. k + 1 d j 5 n. x” thus transformation of T. . 1 $ i 5 k. . the equations xj = cj. .” hood of y in M’ (resp. We set WE obtain=a distribution T” just as we obtained T’ from T$ The xi = a/ad.obvious since Xi.[Xi. tn with origin y such that ( a/azk+l._ n.c’.2. the equations xi . TJ. Let V be the neighbor.3..mXi. . . then the homogeneous holonomy Zk+l -*. gii = gCxi9 xi) are independent of xk+l. . It is clear that v’ (resp. V.(M). . . i s t* s_ n.ixm .] = 0. . VAymXj) = 0. X. As in the proof of (1) of Proof. a/ax”) and ( i3/axk+l. . . Let a’ (resp. . a/6’yk) form a local basis for T’. TI). that (a/ax’. We now use the fact (resp. metrics in V’ and V”. X. . Given. M”). RIEMAXNIAN CONNECTIONS 183 the induced metric of element a z Y(x). x” with origin y such that thus proving our assertion.‘. .. let a. V”) be the set ofpoints in V defined by 191 < c.” be the orthogonal complement lx’1 < c. then. . -. Since T’ is involutive. 1 I i d n. Then y has an open neighborhood V such that V = V’ x V”.1 basis for T”. on T: (resp. QED.+X. Let y be any point of M. 1 5 i s k. . . n) define an integral manifold of T”‘(resp. the equations xi = ci.EMMA. .1. a2 on T. V”) is an of Ti in T. nnd transformations of T. Since g % parallel. . .#X. Proposition 5. small positive number so that the coordinate system x1. .vzy Xi . From the subspace T. Xi) are. Let T’ and T” be the distributions on M used in T’..2. k + 1 <_ *J I. M”) and that V = V’ x 6’“.v&Xi = VxfXm . we have M.jrnandthatgi. . . we have Ck. If T’ and T” are two involutive distributions on a manifold 0. . k + 1 5 j 5 n (resp. ‘. . We first prove the following Proposition 5. . Let T. 1 I i s k. j s k that gij = g(Xi.2. that the. Similarly. coordinate systemyl. j s k. (resp. for each pointy of [Xi. Since the torsion is zero and since M which are complementary at every point of M. x” tion whose torsion vanishes. V“) is an open neighborhood ofy i? M’ (resp. gives a homeomorphism of V onto the cube lx”1 < c in R”.Xi = 0. If we take an hood ofy defined by lx’\ < c.182 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV. ck+1 P). We now prove that the Riemannian metric in V is the direct ‘product oithe Riemannian the first assertion. Let V’ Let T’ be a distribution defined as before. Then T. xL. (resp. . . where c is a sufficiently orthonormal basis for T. Let 1 d i 5 k and k + 1 5 m I n. x6 fork + I 2 PROPOSITION 5. x’ = c9. w91. k + 1 2. . . . with e shall prove Proposition 5.(M) is the direct sum of two subspaces integral manifold of T’ (resp. define an integral manifold of PROPOSITION 5. a”) be the orthogonal the local coordinate system x1.(M). x” with origin at y such v.. . . then these linear .=g(X. . T”) through y and is a neighbor- T’ and T” which are invariant by Y(x). 1 s i. ln other words. belong above. belongs to ‘T”. . .at every point of M. there exists a local coordinate systemxl. Proof of Lemma. If M is sirnfib connected. a/%?) form group Y(X) is decomposed into the direct product of two normal subgroups a Id. .Xi> =Oforl Sidkandk+l 2 the maximal integral manifolds of the distributions T’ and T” defined j I n. . for any set of constants (cl. homogeneous holonomy group consists of orthogonal and xi = 0. . xn for 1 I i. . and the proof of the second assertion is similar. The last assertion is . .

and Ye(x) consists of the identio or&l. . x neighborhood V = I.n. a”) the image of (T by the that the Riemannian metric in V is the direct product of the Riemur&iun natural projection Y ---* v’ (resp.V’).2.. we denote by cr’ (resp. (TV . Tc’ a canoniral decomposition of T. In this’ v1 k * a.1. for each i = 0. small lajsos at x. IV. Hence the parallel displacement along T (3) If dl is simply connected. it follows that (cf. where no = dim l\l. V --+. Ty’ is unique up to an order. Tl” of nlutually orthogonal.. Lety be a point of .. Let T be a shall call T. TFr is the direct sum of. . 7”) is trivial on T. where Y. Then closed curve at y which is small enough to be contained in a (1) The point +y ...” . those along u’ and u”.~ Y(X)>. \\*c shall show that both a’ and a” are elements of’(x). .(. k.\) = is a small lasso in the following sense. x V.(M) which are left fixed by Y’(x’).. I-dimen- Proof of Lemma. First we consider the special case where T T HEOREM 5. T’ Y(x) is the’direct product Y. . then the metric is of the form dxl dxl We now proceed to define a most natural decomposition of Hence V. It follows that a’ (resp.ariant by Y(x) and can be decomposed into a direct sum 7:: . 1. Since the homogeneous holonomy group of V. u Q Y(X)). and special case. normal coordinate system.(Jr) on which Y’(x) acts tri\%tllv. 7’ = /J-l .(. RIEMANNIAN CONNECTIONS 185 18-1 FOPNDATIONS OF DIFFERENTIAL GEOMETRY elements in T. Ti. 7”) is trivial on T: (resp.U’ * .11) = Cf. It is’in\. .(x) x Yi(x) x . This follows from the factorization lemma sional subspaces.. is the direct product of the Riemannian general case. where each V8. We set metrics on these laimensional submanifolds.2. Let M be a Riemannian manifold.. consists of the identity only..hus an open neighborhtiod V such that V = V. on any l-dimensional manifold with a local coordinate svstem x1 QED. u.(M) and T”’ the involutive called a small lasso if it can. X. thus proving our Tc’foreachi= l. las . given element a c Y(x)..’ is an open neighborhood ofy in decomposed into three curves as distribution on . .. every Riemannian metric is of the form grr. . then a canonical decomposition T. Jt’e . Vu is a direct product 6’f l-dimensional submanifolds and that the It is now clear that both a’ and a” belong to Y(x) in the Riemannian metric on V.. .(U) = Tn the general case. k.’ x V” ofy given in Proposition 5. . On the other hand.x YJx) of normal (rrsp. P (2) The maximal integral manifold Mo is 1oculJy Enclidean in the The parallel displacement along 7’ (resp.1/). TL).4. yyx) = Ia’. It is the maximal transformations can be expressed by matrices as follows: linear subspacc of r. Let Tg”’ be the set of . Let Ti be the orthogonal complcmcnt of T:‘) in 7.. The parallel displacement along u is the product of open set of an n.” sense that every point of A4. we decompose T into a product of small Cf:~.21 obtained by parallel displacement of Tgj jar each follows: 7 f p-1 * u + . (1) This is a generalization of Proposition 5. has a neighborhood which is isometric with an (resp. From the prbofof Proposition 5. Tp’ a canonical decombosition (or de Rham closed curve at x such that the parallel displacement along T is the decomposition) of rZ( M) . a”) IS subgroups. k.-dimensional Euclidean space.os as follows. If A1 is simply connected. T”). assertion in the’case where 7 is a small lasso. it is sufficient to prove L E M M A . We set metrics in the V. A closed curve r at x is Cf=. then the homogeneous holonomy group is the product of those along T’ and 7”.(x) is trivial on TLj’ rf i # j and is irreducible on the parallel displacement -along T’ (resp. (2) Since y is an arbitrary point of M.-dimensional along 7 is the product of the parallel displacements along a finite number of Euclidean space.21 and let. _. where p is a curve from x to a pointy i = 0. then the parallel displacement that V0 is isometric to an open subset of an n. invariant and irreducible subspaccs. dx? dxl: If xr is z Then Y(x) = Y’(x) x Y”(x). ypj = {d.(. Tz(Al) and derive its consequences. (4) If A4 is simply connected. .. .. Proof. On the other hand.‘s. is isometric to an open set of a Euclidean space.u. Appendix 7). T.2.& T* 7 p -1 . (so hat p-1 is a curve from y to x going backward) and ~7 is a Mi be the maximal integral manifold of T(l) through y.

RIEMANNIAN CONNECTIONS 187 (3) This is clear from the definition of a canonical decomposition without loss of generality. only on the end point zr of th e curve zt if .(X . . Tz. to X.a curve m-h4 with x0 = o such that the T HEOREM 5. The purpose of this section is to prove is left fixed by every element of Y $(x). In fact. - The proof of this result is given in Appendix 5. Let X = Et=. contains T$).\I”. Proof. let T. is orthogonal to T$). where Xj E T$. in the affine tangent space> . either 5.-.. simply connected and complete element a. The curve x.1)) we may assume.) Since To(M) is ‘the direct product of the two It is therefore sufficient to prove. 65 6 ‘ri(X)I is in Tz) n S.(x). curve it).li). our assertion follows fr:om the foIlowing (4) First we prove result in the theory of Lie groups: Let G be a connected Lie subgroup of SO(n) which acts irreducibly on L E M M A .(x) of $5. 1 Yi(x) is irreducible in Ti).+ ai + . By the lemma.(M). the irreducibility of QED. Euclidean spaces Ti and. we shall indicate the main idea of the covering space of M is isomorphic with the restricted homo. placement of the T’-component of ii from z. %S follows. and B t are curves in TL and Ti respectively either it is orthogonal. we must have Xi = 0. Let C. let T ( II) = q(X) = X0 + Xl + . Since the homogeneous holonomy group of the umversal Before proceeding further. respectively.I4 is simply connected. Then G is closed in SO(n). may be represented bv a pair coincides with some T$). bet.(M) and from the proof of Proposition 5. are left 6. This shows that X is integral manifolds of T and T” through o. .3. Proposition 5.1.. (i) As sume that all vectors of S. be any subspace of T.Y for some ai l Yi(x) and for some submanifolds of M.‘Then S. Ti + TE be a decomposition into subspaces invariant Idy the linear holonomy group Y(x) ‘and let . for each i = 1. Then. respectively. Since this holds for every a. For each ti let ‘X. it must be contained in the - unique curve xt in M ’ wli’lc hIS * deve I oped upon the curve I In orthogonal complement Fz) of EF=i Tz) in T. is orthogonal to T$’ or S.j’ be any simplicity we identify the affine tangent space with the tangent other canonical decomposition. and spans Tz’. SL’). This is view of Proposition 4. In the first case.1. Consider. If ui(X) = X.186 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV. . to T. we have Riemannian manifold. thus proving the lemma.(M) invariant 6y Y(x). proof. let X = Ejjk=o Xj be any element of S. totally geodesic (ii) Assume that ui(X) # . . (3) of Theorem 5. ai # 0 is a vector in T:) as well as in S..(M) = CiEo Sl. Let S.5. C. We fix a point o c M and let M’ and M’ be the maximal is irreducible in T$‘.1 to M’.* . (For the sake of Going back to the proof of (4). M is isometric to the direct product 31’ x . since shall define its projection on M’ to be the curve x (lo. that each S$). of Y.). Since.(. k. B. Proof of Lemma. .. By orthogonal to T$).example. X E for every i L 1 or it contains Tz) By applying (4) of Theorem Example 2. The restricted homogeneous holonomy group of a result of the parallel drsp’lacement’ of i’t along itself to o is’equa] Riemannian manifold M is’s closed subgroup of SO(n).* * + X. Xi. then distributions defined by TL and T”+. . for each t. First it is clear that TL”’ = Sk’)< (vector) space..4..1 of Chapter III we may define the -c:‘rve obviously a contradiction. where n = dim M.rt depends geneous holonomy group of M (cf.. i . ‘ (A. Assuming that M is reducible.t :.he develop: S. where A. (along the The following result is due to Bore1 and Lichnerowicz [l]. The subset T HEOREM = Xi - 6. where Xj 6 Tg). both M’ and M” are complete. each Xj.T’ and T” be the parslIe since ai acts trivially on T!!) for j # i. This implies that Tc) is contained in with x0 = o which is obtained as foIIows. 0 5 t 5 1. in M with z = o we {6. We show that the end point ir of the projection . for . be the result of the parallel dis- Sf) implies that Sin actually coincides with Tz’. 1 5 j 5 1. X. the n-dimensional vector space R”. ment of z. In the second case. as’in the beginning ai = Xi. E Yi(x) and since Y. For an arbitrary Let M be a connected. j # i. we see that there exists a for some i B 1. X . to o = z. The decomposition theorem qj’& Rhccnl fixed by Yi(x). In vic\v of of T. For any curve zl. that M is * simply connected.

0 5 s 5 sl..$O of the curve with parameter t for the fixed value of s (resp. Their tangent vectors Proof of Lemma 4. and let r2 be the curve z. Theorem 4.) of T’ thr.. the parallel displacement jection of 72 in the mnximal integral manifold M’(z. For. = o (2) For @Y woe 7” =J+‘. in u”(x.2. V”) wrth x0 = z by a finite number of neighborhoods of the form V = V’ x V” (resp. whrch 1s de. and let a be any number with 0 5 a 5 1. 0 I t ::. in 0 5 s 5 i. be a T’-curve. r) (resp.. Thus p is an I. -=c r (resp. j differentiable) curve z1 is called a. has all the Proof of Lemma 2. M”(z) of T”) through z. Let -rl be the curve z.. Thus xt is the projection of the curve z. =y. 0 5 t S t. from xt to x0 = z LE M M A 4 . r) c {x1} ‘x v” for Al“(z) respectively. (b) and (c) of Lemma 3.%4 3 izI”. r) onto C”(x. u. kaspqberties (a). be a T-curve and let T” =y9 along the curve xt. is the result of &Led by z(.be denoted by iy) and it. r). along the “parallelogram” formed t/a the curve xl. /’ implies that p is a covering projection of . This means that z.. Let 7’ = x..Al’ and. the inverse of the curvle3p. 1s the same a+ the paraliel displacement of f.!. For any curve z. a pro. with parameter s for the fixed value of t). d”) denote the distance function on the maximal parallel to Y. the inverse of the curve zl(Q. The mapping p = (P’. w E nZV(z)) such that d’(z.= (x.. For the existence of a neighborhood properties (1) and (2). ~1. set z. be a curve in 64 z&h z. th Then the projection of T = r2 * r1 in . 0 5 s i sr.y) .. it follows that i. * Let T = tt. Then 0 5 t 5 a. so] 4 M. Let 7... = z in V. In fact. 0 /r t i 1. ?roof of Lemma 3. the curve geodesic rN along the curve T’.. to z. d”(z. L EMMA 1. is trivial. = z along the curve z. the projection of 0 5 t :> r. where V’ and V” are open neighborhoods of z in hi’(z) and sufficiently small so that x1 E V’ and U”(xt. Let 7’ = x1.Zi onto M’ x h4”. s 2 so. the T-component. is clear that the family of isometries f. be the pro. where s is the arc length: M’(z). *7-l. 1 J x {o.‘“(x. z 6 M. A (piecewise fobtking p r o p e r t i e s : : (i) zj??~* x.6 then (resp. Similarly. U”(z. Given a (piecewise (2). Thus. S) = z. T’-curve (resp. then LEMMA 3.. is uniquely determined...~) for-r E U”(x. T”-curve)+ rf ‘f belongs to T:. a 2 t I=: 1..) the parallel displacement ‘of’ the initial tangent vector Y. see Proposition 5. It zt in M’(z) is given by the natural projection of V onto V’ = x0.ith the We introduce the following terminologies. respectively.2. r)) d enote the set of points W:E Al’(z) AZ/i” will be shown to be isometric at every point.. 0 s s 2 so. 2. ‘Proof of Lemma 1. Let V be a neighborhood of x0 of the form 2 . Since V = Y’ x V”. nnd z&. then zi = exp sY. along the curve z&. of i. if Y.ough z. b e a P-geodesic withy0 = x0. 0 i t ‘5 1. = ft(y). tangent vectors..(x. Choose a number r > 0 of z in M. r) I. T:J for every f. the parallel displacement of and using the above argument for each neighborhood. (resp.z. with z. The @ail now follows. 5 1 and 0 5 s.’ We first prove the uniqueness. of simply counected. (a) For any 0 5 t. If h is a homotopy in M from a curve of iz4’ to another curve of iZ4’.. any pomt (c) and by the fact that ‘7” is a geodesic. 188 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV. from z. 5 so. yt) where xt (resp. similarly.. Let. We define ft byf..A4’ coincides with the projectron of curve zt.. (1) The di$erential offt at xti coincides with the parallel displacement along the curue #from x0 to xt. The homotofiy z. is. 0 d t 5 t. and 7’ = 7. differentiable) homotopy z: [0. The family ft can be extended easily for V = V’ x V”. 0 5 t d 1. 0 5.. By (a) and will . Then there P’(h) is a homotopy between the two curves in r2il’. we shall denote by zj’) (resp. Then there exists a homotopy z. P”) of h4 into &4’ x Let U’(z. r... M” is simply connected. z$. let d’ (resp.31~). i& is parallel to js along the curve ~1’). r) with the following properties: isometry of M onto 44’ x M”. 0 I t i 1.. This is obvious from the second definition (b) For any s and t. . 0 5 t 5 1.() is a .. of the projection by means of the (linear) ‘parallel displacement of (c) For any s and t. jection p’: . integral manifold M’(z) of T’ (resp. = z). Let z Q M and let. = . is a curve in V’ (resp. ij”’ is parallel to xt along the curre $. V = I” x V” be an open neighborhood LE M M A V = V” X V” as in Proposition 5. w) < r). be given by the pair 0 5 t F 1 and for a suitable r > 0 by covering the curve T’ = x1 (x. 1Z4’ is exist a number r > 0 and a farnib of isometries f.i.?‘. RIEMANNIAN CONNECTIONS : 189 Thus we obtain a projection p’: h4 .

t: the following assertion.Thisprovesthatz’. the projection of the Vi x V:‘. of the cur\‘c 11. AS a neighborhood Jr. a curve from 0 to zl. :-?j by Ixrnnla I ..1. zj: tion similar tb (2). suprcmum of such 0. = csp sIVI. and K’ . it follows that zi = *p* t m”forO:lt. . V’ .: t. = f. M.I?-II LEMMA 6.e*. As befcrrv... V bv a cur\‘c I(‘..‘11 Proof of JAemma 5. the ~K$YII ~1 along TV * T ‘-1 to -yxo is the same as the parallel dishlacement of of K’ * . ICC shall al.zy) and the Y-geodesic see that the projection of K.u .u* a n d the T-curve -K.11 n e i g h b o r h o o d Vj of the f.[iOll in .I. Lrt j(* I!(> .-. Ct’e first remark that 7. Let n bc the s m a l l neighborhood J’ := J-’ x J ” ’ and K a cur. b e c a u s e p” a n d I(* . Let v be anotl1c. Keeping the notation of Lemma 4. ‘I’hc p.. 7k curre 7’ ... lye m a y assume t h a t 7 ’ is a difXerentiable curve SO of the form t” V” as in Lemma 2. .lsimaI intc.-..x and (b) imply that. 7?. s (.. 7”-l in M’(y’o) coin&h with &) = . T and K . lvinq sntisfics conditions (1) and (2).. be the filmily of isometries as in Lemma 3. T h e homritop!. lies in a sma. Hcrc . a point z in M are homotopy was construcirxl b y the parallcl displnc-cmr-.< 1 andn. Thus if wc~rcplacc . assume a < so. It is obvious that there exists a number B > 0 It suffices therefore to pro1. the projection 0: . . z. \\c‘ we apply Lemma 3 to the T’-curve T f(1) = .c 2. Lemma for Theorem 4. lvhcre 11 = s . Since tii” ” depends only on p* and K and not on y. Ict -.\f’(z. =p..and let T: bc t h e gmtlcsic in 1.) 2)“~) aloQg zjso) toyso.emma (3.I’ is equal to v’ followed by K’. SimilarIy for the projection o’f K’ .-. and the end point of /A’. . the same parallel displacement for T’-vectors. Combined with the above limit argument. the parallel displacekent of i. Condition (1) is obvious..II’ ha1.) is EL’..I snl~ll deformation of a curve Z~ means that. . \it’e now dividr 7 into a finite nu-mbcrof’&rcs.. We sho\v that the projections of the curves K’ .ins in . v’ * TV have the same end point in the m.U in . cur1. jciir. f. actuall) hood J’. satisfies condition (2) for 0 5 t d 1 and 0 (= s 5 a + r.Cjt. the curve 2:“) is the limit of the parallel displacement along the TO pr-ovr this.: A”~ ‘-I L-= lim $I).“’ a s s t n (cf..2 of Chapter II). = esp sI’.. \~e replace :i portion z. If two curves TV and.. for each t. contra- dicting the assumption that a < sO. = iii it. ~\~shall a s s u m e t h a t K i s CL 7”- ‘I’hu~ condition (a) is satisfied. = 2.r (:ur\~e from z1 to z2 \\rhich lies i n I*.u-’ * p*. displaccmcnt of Y-vectors at 2z along p-1 is the same as tllc parallel displaccmcnt along .-e f<om zz t o :.u’. p a curve f r o m z1 t o z. I t remains t h e r e f o r e t o p r o v e t h a t :.t . p in . its projection in manifold of T’ through the initial point of TV. Let r be s u c h t h a t z.i.lm LEhlMA 5. w h i c h lies i n a satisfies condition (2) for 0 CL t <> I and 0 .~1. By I.t of the homotopic to each other. it’ rn:t) first reJ&cc thr curve K by its J)l. t -I 1 and 0 2 s 5 8. 0 5 t 5 1.‘= Vi x C’. that is. ‘I’hcn r11(* First KC: show that 21” satisfies (2) for 0 5 t 5 1 and 0 :: s 5 n. 0 5 t 5 1. s). is obtninctl honl gcodcsic i\. To .c. satisfying a condi- from the Y-geodesic . ~h(t p a i r (ill’. is differentiable in (t. ive see that t h e projcctitrn of K . ‘I‘~IIIG.. projections of K .-rsssa+r.c the same v*I(I Since 2. T h u s . T2 from Q to. T. . s u c h t h a t rach T. E. Let J. the end point of 7k to the end point of 7.U b>’ is parallel to-f” along the curve u$“) = zia). 18 .I* \Ve now c‘ome to the main step for the proof of Theorem 6. -rl b y a f i n i t e succession o f s m a l l dcformatidns. is differentiable in (t. Y in the abo1. and c u r v e . 6. reduces to the point ~“0. J-’ I . :\g:iin.. for a ccr:aill small ncG$lbor- proving the uniqucncss. theu their projections in M’ = M’(o) have the i n i t i a l t a n g e n t v e c t o r Gf the .ti~c.. T1.. 71) in V.e.\I’(. s). and wt. t h a t -1. (8. p”) i n J’ =.ayi take a Gcighb~rl~ootl prnvt’ (2). the parallel displacement along point. In order to prove a = s.-joining z:.I* construct+sd and a homotopy xj’. i\c In ordrr to show that z: has property (2) beyond the value a. T’-curve 2 . t. Since 7’-l is a T”-cur\. \vherc v 5 fv’.!4’ * TV is the curve 7:. . 7 i n . homoiopy \vhic% is cohstructcd from th& T”-geodesic -: ail’! ~I. A/‘( -1’“~) is trivial. We see then that the!e exist a number r > 0 by the ‘curve K’ obtained by using the homotopy . -r 2 u $ r. Jo’ . 2.‘T. TV-.I’.\I’.ith i n i t i a l t a n g e n t vector Y. IV. KIEMANNIAN CONNECI‘IONS 191 Proof of Lemma 6. We have also ib. L in J’. By Lemma 2.licl 81 I gi\cs conditiois (b) and (c) for 0 s t 5 1 and s = a.. . Let /l b e rcprc~scntc~d !:.gSrodcsic 7: aIonS k’ p’ o r nlony same end point.) is the curve !A’ tollo~\~~l J”.( _I+) for 0 . say. followed by the cur\c obtained bl.e arqumcnt.ith w. v”) :in‘V I=’ J” ‘\ J“‘.zi ‘o). This means that 7” * $-I pro-Jects o n T . this ?“‘-geodesic in J. s u c h t h a t z$‘) = zp) a n d ZEN. t : f. = zt’. Conditions (a) (T.

Theorem 5.'I. where hf. . Since Y’J is isometric at 2.. . T. TIjEQREM.(M).’ . ” .is discrete. Let z be any point of 121 and let T be a K. Since !i is a normal are differentiable. the parallel displacement Thcorcm 6. Although we shall hfk. rf+' ' thert either ) (1) Q” contains all translations of T.b is a group of We continue to the next stage of projecting the curves Euclidean motions of the affine (or rather Euclidean) tangent space . a result of the above remark. This means that the parallel displacements of the initial tangent vector of 7: along CL’ and Y’ are the same so that the two homotopies produce the curves pr and v.. proving”that p is T’(M). along ~‘-r . p”( 2) and Y have the same length. IV. let 2 = where K is not discrete.Since ~‘-1 . Consider next the case where K.(M)). RIEMANNIAN CONNECTIONS 193 192 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Theorems 6. In particular. a-‘Ka = K for every d E ‘PO. is a closed-subgroup of SO(n). the form to show that the mapping p = (p’. and Let 1%1 be a connected Riemannian manifold. . chat Y” normalizes K”.u” is trivial for T-vectors. and ending at the same point. we have only subgroup of CD0 and since every element a of (1)O is of. ii where d E ‘F” and 5 is a pure translation.(M). 7. It follows that 2 and have V = T.\fJ parallel displacement of X from z to o along T and then from o to invariant by ‘F”. starting at the end point of T. . By definition of the projection.(.(M). 1”’ . where a curve K~ S t a r t s in such a way that K~ * pk * 7. it was inspired by the work of Reinhart [ 11. ((0) = 0 for every t z K.h ere fore.p”) of M into M’ x M” is a = 6. identity component K” of K.(M) defined by f(a) = 40) for a c CDO. same point.. We know’ T”-vector along . complete. Let K be the kernel of the homomorphism of Go onto Similarly. we denote the affine holonomy group Q(x) and the linear / tively. of -41. QED. we have two curves ending at the We first prove the following result.1 and 6. QO THEOREM 6.For any tangent vector 2 z 7’.4 an*& Theorem 6. Since Y” is connected. A connected.. this indeed follows from (cf. Y” normalizes isometric at each point. M. we define a mapping p” of M into M”. WC Similarly. irreducible Riemannian a direct proof of this fact. the invariance by ‘F” is a consequence of the fact p’(z) along p’(T). 01: Now we are in position to complete the proof-of Theorem 6.(Al).(M) . Now it is obvious that this process can be continued. . t(O) is invariant by Y” (where 0 denotes the origin of Combining Theorem 5.(M) has a fixed point. Proof. are all simpb connected. thus completing the proof of Lemma 6. On the other hand. . respectively... for every 6 z K.5 of Chapter’III). (2) O”jxes a point of T. Lemma 6 allows us to define a mapping p’ of M into M’. we obtain the and vk are the same.2. means that K consists of the identity element only and hence that @O is isomorphic to Y” in a natural manner. is compact.(. 1~’ is a curve in F’. ~ : 1 . is irredunble. . are the projections of K’ . .ait normalizes the X + Y.p”(Z)) have the same length. As we indicated before Lemma 1.. any compact group of affine nian manifoti M is isometric to the direct product M.1. where n = dim :hl. Let f be the mapping from a0 into manifolds. that is. Proposition 3. The proof of K’ * V’. is a Euclidean space (possibly of dimension 0) and prove a more general statement in Volume II. where X z Ti and Ye Tf. that p’(Z) is the same as the vector obtained by the T. Y” commutes with K elemantwise. 7k and K’ + v’ * r3( rcspec. These mappings ‘+“O ‘(cf. we shall give here hf.1. simply connected and complete Rieman. T. Since YO’is irreducible. This means that Q” contains all translations of p(Z) = (p’(Z). Fixing a point x Kk ’ Vk * T. This decomposition theorem of de Rham. x -11. Consider first the case curve from o to z. . It is a non-trivial linear subspace of ‘J.. We also remark that the parallel displacements of every holonomv group ‘F(x) simply by @ and Y.11) by K”.1 is new.. Since ‘1’O is irreducible by assumption. Kk’ pk. connected.‘< transformations of T. 7k--1 a n d Kk . * 7k--1 by the same method.2 arc due to de Rham [l].vk * 7. As i-‘..5) that the restricted linear holonomy.l. p’(Z) and X have the same length. Hence. Such a decomposition is unique up to an order. Let V be the orbit of the origin of it is. clear. x * .group ‘I’0 property (a) of the homotopjr in Lemma 4.

for any closed precisely ?I( Y) = -r{. Assume that two geodesics 7 and p L EMMA 2.. is complete. of T. Since M. = {i 5 T. denotes the origin of T.(Y) + C.(M) + M is surjective.(Y) = d)(Y) 4 c. q) be the point x jixes a point of the Euclidean tangent space T. depends only on t) if and only if p is parallel. to x. Y (resp. Let p be a point jeld along a curve T = xt. Since a0 fixes the origin of T.X + it = 0 for 0 $ t 5 1. YE T&W. Let r = xt. r). Let M be a connected. 194 FOUNDATIONS OF DIFFERENTIAL GEOMETRY IV. that is. Our assertion follows therefore from . then M is point field along T defined by the affine parallel displacement of isometric to a Euclidean space. contradicting the assumption that where C. and let x1. the exponential mapping T. we have Assume that exp. curve . it is sufficient to Lemma 2. Since ?:‘“(p. x” be a normal coordinate system on c’(x. We observe that the affine holonomy and end point p at each point of 7.(M). Then p is a parallel pointJield .P-w.(M) . issuing from x meet at a pointy # x. If the (restricted) ajine holonomy group @O at a Proof of Lemma 1. Assuming that X0 (F T. the origin of Tzl(n4)) and let Y* be the vector field along Proof.. displacement along r from x. 195 Let da be a bi-invariant Haar measure on @O and define prove that b is parallel along each geodesic through x. r) = {r E M. 0 5 t i 1. r). f . manifold h4 and let $ (resp. Thus 7 = . .u * T Going back to the proof of Theorem 7. QED. let T defined by the linear parallel displacement of Y. lix’i] < r) onto C(x.+.(M).( M) .y) < r}.(M ) is a point fixed by CDO. is the development of r = x. 0 5 t I 1. 0 s t 5 1.1 (without 1 assuming the irreducibility of M).. We set X = -Cr. these developments which are line segments coincide with each other. linear) parallel . is a diffeomorphism of N(x.. where 0. that is. we may assume that CD0 fixes the origin of T.(M). let p (resp. in a Riemannian manifold M and let X be the corresponding Vector jcld mcnt p-1 * T maps the origin 0.2. s LEMMA 1.f and only . Riemannian manifld. .) and p-‘(0. Then We now investigate the second case of Theorem 7. we assert that maps X0 into itself.(M) into itself and hence along r. xi (c?/axj) and let p be the corresponding point field (ef. (for a fixed t and with parameter h) is the development x +. RIEMANNIAN CONNECTIONS . 2 . in T.. d(x.M of T into T. 9J denote the afine (resp.) is independent of h (and is 1:l.. X0 = f(u) da. We show that p is a parallel for 0 2 t 2 1 if and only ifp is parallel.T. (7-l * p * 7)X0 = X0. simply connected and complete where C..) = 7tlfh(Xt+J + G..f we have C.(M).hY respectively. Hence the origin of T. This shows that -7. At the point x0.r. The afBne parallel displace. Since T-‘(0. In fact. 0 5 t j 1. Given Y f T.*(M). Y* is the vector with initial point q line segment tX. completing the proof of point field.(A4). Then p = q + 7 be the geodesic from x to a pointy which is developed upon the Y* at each point of 7.(M)...u. be a curve in a Riemannian It is l&y to verify that X0 is a fixed point of CPO. this means group @O(y) aty fixes the origin of T. $4 of Chapter III)..(M) ~:+“(P.) Proof of Lemma 2.U at JJ. +(o”) ..(M) given by ~(3’~) is left fixed by y. This proves that the exponential mapping T. This follows from We show that it is 1: 1.. . T HEO R EM 7 .x + v = 0 for any vector field I’.. +f. into TJM). we obtain are the end points of the developments of T and p. the affine parallel displacement along 7-r * . From Lemma 1.

2 f..J. :Zs a consequence WC obtain the following corollary due to C-oto and Sasaki [ 11. every point of M has a neighborhood which is isometric VayY = ‘-CL.. that is. RIEMANNIAN CONNECTIONS 197 ‘76 FOUNDATIONS OF DIFFEREXTIAL GEOMETRY COROLLARY 7 .J o n t o U(X. x” on the \vhole . Z] = . QED. A1 is not locally Euclidean. Since the boundary of n-(. COROLLARY 7 . But vt is of the form Thus the ftmctions g. csp. r. Z) + g([X. . I f M is a complete Riemannian manifold of X(P(K Z)) = g(~. Apply Theorem 7.. Y]. r0 + F).\1. it follows that (esp. assertion now follows from Theorem 7. is a cliffcomorphism. Z) -t g(Y. From G. is non-singular at every point of ‘Vjx.z = T. at every point of AII.. r). is an isometric mapping of N(x. [X. By choosing a normal coordinate system x1. QED.K Z) + g(K T’SZ) dimension > 1 and .) and hence nonsingular in a neighborhood of such-y. r) onto U(x.Y) = (I.. translations of T. /I. Hence. r). r. 111 is a Euclidean space.(M).. are constant along each geodesic through x. 21. 21). we see that there exists 8 > 0 such that exps is a difli~omorphism of X(. contradicting the tlcfinition of r.[X. r& is compact. t!f2”. Then we have Proof. by the argument above.. . Y] to an open subset qf a Euclidean space).u. be the supremum of r > 0 such that esp. md Proof. then the restricted ajine holonomy group W’(x) contains all = -2g(Y. we obtain (cf.2 to the universal covering space of v. Theorem 5. r) with Euclidean metric onto li(x. Let Y =‘ a/&j and Z = 8jaxk for any tixedj and k.I-( Y. at Fvery point of C’(x.k := glk(. of S(x. . ThiT s h o w s t h a t expl..J. Let r..\I.? + [X. . we conclude that gjk = dj. ‘:‘his shows that rxpr: is a dif%eomorphism of T. If the restricted aj’ine holonomy group @O(x) jixes a point of the 0 and y-.X + Y = manifold. then M is locally Euclidean (that is.1 and Corollary 7. This means that the functions gjk arc invariant by the local l- parameter group of transformations p’l generated by X. .f the restricted linear holonomy group yPO(x) is irreducible. is isometric at (‘very point 3’ of the boundary of . If r0 < CC.3. is a diffeomorphism of lV(x. .X + Z -= 0.Y + [X.(2Fl) for some x E M.nd Z be arbitrary vector fields on Ail.1 of Chapter III) Euclidean tangent space T.). hcncc an isometry.(A1l) onto .1. Let Al be a connected and complete Riemannian Let I’ . 3 . Our. Y] =: . r0 + E) onto U(x. Since the differential (rsp.X + [X. Hence.). Since Y-““(x) is irreducible. T. IV. T. QED. M. 4 .


Proof. We may assume that T = 0 ; consider R - T and 0
instead ofR and T. We assume therefore that R(v,, ~3, vl, u2) = 0
for all zr, v2 E V. \Yc have
CHAPTER V 0 = R(v,, v2 -:- zj4, vl, u2 + c4j
y= R(zjl, u2, Do, u4) + R(rl, z’$ iSl, c2)
Curvature and Space Forms = 2R(v,, vz, z’~, v4).
(1) WV,, 7.~2, 2’1, up) = 0 for all z’r, v2, v4 E K
1. Algebraic prelirninnries From (1) we obtain
Let V be an n-dimensional real vector space and R: V :< V x 0 = R(u, + ~3, v2, u1 i 03, 4
V x V -+ R a quadriiinear mapping with the following three
= R(zl!,, u2, 2’3, ~4) + R(u3, ~12, 2’1, 4~
Now, by applying (d) and then (b), we obtain
(4 Wl, u2, u3, u4) = -Nu,, vl, u3, 21~)
(b) R(u,, ~2, i3, 4 = -W,, v2, 1’4, v3) 0 = R(v,, c2, c3, up) $ R@,, ~1, 23, u2)

(4 WI, u2, u3, 04) + R(% us, u4, u2i i- R(Ul, u,, u2, 03) = 0. -
- - Rjv,, ~2, ~3, ~4) - R(v,, ~4, ~2, 03).

PROPOSITION 1.1. If R p ossesses the above three properties, then it Hence,
possesses also the following fourth proper9 : (2) Rh, u2, u3, u4) = W,, K,, zj2, 4 for all vr, z2, v3, v4 E V.
(4 W,, ~2, 7.~3, ~4) = R(v3, ~4, ~1, ~2).
Proof. We denote by S(v,, up, v3, v.,) the left hand side .of (c). Replacing v2, v3, v, by vD, zf4, v2, respectively, we obtain
By a straightforward computation, we obtain . ( 3 ) WV,, 02, ~‘3, 4 = R(z:l, 7~3, c.1, ~‘2) for all
,, vr, v2, v3, up E 17.
0 = Gh u2, u3, UP) - S(7J2, u3, u4, %) - S(zr3, u4, Vl, u2) From (2) and (3), we obtain
+ +4> Ul, 02, u3) W ul, u2, v3, ~4) L R(v,? v2, v3, z!,j -t- R(v,, v3, v4, v2)

= W,, u2,v3, ~4) - R(u2, ~1; ~3, ~4) - WV,, ~4, ul,vp) + R(u,, zi, vz, u3),

+ R(v4, ~3, ~1, ~2). where the right hand side vanishes by (c). Hence,
By applying (a) and (b), we see that
4 Vl, u2, u3, u4) - 0 for all vr, v2, v3, zj4 E T’.
2R(v,, US, u3, v4) - 2R(v,, v,, vl, up) = 0 . ” QED.
QED. Besides a quadrilincar mapping R, we consider an inner
PROPOSITION 1.2. I.& R and 2” be two quadrilinear mappings with ,product (i.e., a positive dcfinitc symmetric bilinear form) on V,
65e &beproperties (a), (b) and (c). rf which will be denoted by ( , ). Let p be a plane, that is, a 2-
dimensional subspace, in T ’ and let v1 and v2 be an orthonormalb’
4~1, ~2, a;, 4 = T(vl, vi, ul,u2) fbr all ul, v, E V, I. :
basis for p. We set
then R = T. . AI:(/)\ R(zl,, c2, vl, z12).



As the notation suggests, K(P) is independent of the choice of an

properties (a), (b) and (c), we associate a symmetric bilinear

orthonormal basis for p. In f?%ct, if WI and w2 form another form S on V as follows:
orthonormal basis ofp, then
S(u,, u2) = R(e,, ul, e,, u2)‘ t R(e,, ul, e2, u2) t - - .
wl = au, + bvz, wz = -bvl + au2 (or bv, - au,),
+ We,, vl, e,,, c2), ul, p2 E V.
where a and b are real numbers such that a2 + b2 = 1. Using (a) It can be easily verified that S is independent of.the choice of an
and (b), we easily obtain R(%, u29 UlY u2) = R(w,, w*, WI, w2). orthonormal basis e,, . . . , P,,. From the definition of S, we obtain
PROPOSITION 1.3. Ifvl, v2 is a basis (not necessarily orthonormal) PROPOSITION 1.5. Let v c V be a unit vector and let v, e2, . . . , e, be
of a plune p in V, then an orthonormai basis for V. Then
R(5) u2, Ul, 02)
K(P) = G01) (u2, u2) - (Ul, 42 qv, 0) = &q + * * .* + K(p,),
where pz is the plarie spanned 4~ v and e,.
Proof. We obtain the formula making use of the following
orthonormal basis for p:
2. Sectional curvature
Let M be an n-dimensional Riemannian manifold with metric
tensor g. Let R(X, Y) denote the curvature transformation of
where a = [(vl, vl)((vl, vl)(02, u2) - (vl, b2)2)1' QED. T,(M) determined by X, Y 4 T,(M) (cf. .55 of Chapter III).
We set The Riemannian curvature tensor (Jield) of M, denoted also by R, is
the tensor field of rovariant degree 4’defined by
R,(v,, v2, us, VP) = (% u2)(% 4 - b29 d(% 4
R(X,, 472, &, 4) = g(R(X,, &)X2, Xl),
for ul, u2, v3, v4 E V.
Xi E T,(M), i = 1, . . . , 4.
It is a trivial matter to verify that Rl is a quadrilinear mapping PROPOSITION 2.1. The Riemannian curvature tensor, considered as a
having the properties (aj, ( b ) and (c) and that, for any-plane p quadrilinear mapping T,(M) x T,(M) x TJM) x T,(M) + R
in ?‘, we have at each x E M, possesses properties (a), (b), (c) and hence (d) of tj 1.
K,(p) = Rl(zJl9 u2, Ul, u2) = 1, ,
Proof. Let u be any point of the bundle O(M) of orthonormal
where zjl, cz is an orthnnorm al basis for p. frames such that r(uj = x. Let X;, XT E Tt(O(Af)) with rr(Xt) =
X3 a n d rr(.XT) = X4. ,F rom the definition of the. curvature
PROPOSITION 1 .a. Let .q he a quadrilinear mapping with properties
transformation R(X,, X,) given in $5 of Chapter III, we obtain
(3), (b) and (c). VA-(P) = c?r allPlapes P, then R = CR,.
Proof. By Proposition 1.3, we have g(W,, &)X2, Xl) = g(uW(~$, -W (~-‘&)I, X,?
Rl(v 1, v2, ul, 21~) = cR,(z'l, u27 ul.l, v2) for all ul, v2 z V. = ((2fqX,*) .Q) (u-‘-Y,), u-IX,),
Applying Proposition 1.2 to R and cRl, we conclude R = CR,. where ( , ) is the natural inner product in R”. Now we see that
QED. property (a) is a consequence of the fact that n(X:, X:) E o(n)
is a skew-symmetric matrix. (b) follows from R(X,, X4) =
Let e,, . . . ) e” be an orthonOrmal basis for V with respect to the -R(X,, X:,). Finally, (c) is a consequence of Rianchi’s first
inner product ( , ). To ea ch quadrilinear mapping R having identity given in Theorem 5.3 of Chapter III. QED.

For each plane p in the tangent space T,(M), the sectional
curvature K(p) for p is defined by For an arbitrary X, we choose Y, 2 and U in such a way that
X, Y and 2 are mutually orthogonal and that U = 2 with
K(P) = R(Xl> X2, Xl, XA = g(R(X1, X,)X*, Xl), g(Z, 2) = 1. This is possible since dim M 2 3. Then we obtain
where Xl, X, is an orthonormal basis for p. As we saw in $1, (Xk)Y - (Yk)X = 0. _
K(p) is independent of the choice of an orthonormal basis
Xi, X,. Proposition 1.2 implies that the set of values of K(p) for all Since X and Y are linearly independent, we have Xk = Yk ,= 0.
planes p in T,(M) determines the Riemannian curvature tensor This shows that k is a constant. QED.
at x.
If K(p) is a constant for all planes p in T,(M) and for all C OROLLARY 2.3. For a space of constant curvature k, we have
points x E M, then M is called a space of constant curvature. The
following theorem is due to F. Schur [l]. NX, Y) Z = Ug(Z, x,X - g(Z, 4 Y).
This was established .&thS>ourse of proof for Theorem 2.2.
T HEOREM 2 . 2 . Let M be a connected Riemannian manifold of
If k is a positive (‘i&p. negative) constant, M is called a space
dimension 2 3. If the sectional curvature K(p), where p is a plane in
of constantpsitive (resp. negative) curvature.
T,(M), depends only on x, then M is a space of constant curvature.
If R( &id gij are the components of the curvature tensor and
Proof. We define a covariant tensor field R, of degree 4 as
the If?etric tensor with respect to a local coordinate system (cf.
follows : $7 of Chapter III), then the components Rijkl of the Riemannian
R,P’> Z X, Y> = g(W XkU, Y) - cd3 WY, W, curvature tensor are given by

W, 2, X, YE T,(M). Rijkl = zm SimQ
By Proposition 1.4, we have : i :*
If M is a space of constant curvature with K(P) = k, then
R = kR,, .
Rijkl = kkikgjl - gj1SglJ, Rihl = k(6;gjl - gikSf). J
where k is a function on M. Since g is parallel, so is R,. Hence,
(V.R)(W Z> 4 Y) = (W)R,(K Z, 4 Y) As in $7 of Chapter III, we define a set of function; kj,, on L(M)
for any Ur T,(M).
This means that, for any X, Y, 2, UE T,(M), we have
where B = (fij) is the curvature form of the Riemannian connec-
[(V,R)(K y)lZ = WkV, Y)X - g(Z, XP’). tion. For an arbitrary point u of O(M), we choose a local co-
ordinate system x1, . . . , x” with origin x = n(u) such that u is the
Consider the cyclic sum of the above identity with respect to frame given by (a/ax’),, . . . , (6’/axn),. With respect to this
(U, X, Y). The left hand side vanished by Bianchi’s second identity
coordinate q&tern, we have
(Theorem 5.3 of Chapter III). Thus we have
gi, = dij at x,
0 F (uk)(g(Z> Y>X - g(Z W’) and hence
Rj,, = Rijkl = k(Sikajl - d jkSli) at x.
+ (X4 (s(Z, u) Y - s(Z Y) uj
Let o be the local cross section of L(M) given by the field of
+ (Yk)(g(Z, X)u - g(Z WV.
linear frames a/&i,. . . , a/ax”. As we have shown in $7 of

\I’ (rcsp. for example. which may be stated as f o r m (xl)” f . * . then t2 2 1 at every point of . .1 of Chapter I) ..n.. Let g be the Riemannian metric of M obtained by . .* * -k (~~tl)~ = r. set of points of A4 with t 2 1 (resp. t e. Let (x1. curvature .A 0’ o n O(:\l). First we observe that II4 is a closed submanifold of &(x1). If . y” as follows: THEOREM 3. In fact..* .y”) defined by yi = Xi/t. A4. This means that M is isometric with a sphere of raclms rl. Iff is a linear mapping of a subspace U of’ I. + (x”)2 + rt2 invariant acts transitively on M as a follows. into V such that (3) If r > 0.OhIE’i‘K\ V. then we set xn+l = r?t. where 0 = (0’) is the canonical form on O(. consists of transformations leaving the point o with coordinates . we leave the verification to ’ leaves Q invariant and which maps x.24. if x0 and x1 are elements of V with &(x0) = Proof. . .rt” invariant. . . x”. &U(x)) = . Namely. .Cci Y’ h’j21 .4rtin We begin with the proof of (3). The mappmg Since 21 is an arbitrary point of 0(-U). c (dx”)) + . see. Y < 0.1). t) 4 (y’. Example 1. x”. the-teader. . . u A4 is a space of constant curvature with secttonal curvature k. we shall prove (1). The transitivity of G (=i]-M 1s a. and the metric g is the restriction of (dx’)? -i .n... G is a.. then f can be extended to a linear isomorphism of V onto itself such that Q(f(x)) = Q(x) for all which is d$eomorphic with R”. . . Let H be the subgroup of G which (xl)” + .(do” ‘I)’ to M. For the proof of Witt’s theorem.1. s ace p ofconst an tcurvature with sectional curvature I/r. . Hcncc. CURVATURE AND SPACE FORMS 205 Chapter III.b. Let Q be a nondegeneraie quadratic form on a vector space V. p. Let .QH f or all x c U. Then M is given by Finally.. group of isometries of AC?. we first consider G as a group acting on R’+f~l. . . it leaves the form (dx1)2 + .21”) be rhe q = x-01 A 0’at Il. we prove A straightforward computation shows that the metric g is expressed in terms ofyl.. If r > 0.ijkl = k(d.. (2) The group G of linear transformations of R” -1 leauing the quadratic on A4 is a consequence of Witt’s theorem.+ (dxn)2 + r dt2. .. [l. t)’ be . .restricting the following To prove (2)... form to Al: Since’ G is a linear group leaving (xl)” + * . there is a linear isomorphism f of V onto itself which R”+l (cf. i = l. PROPOSITION of isometries of the Riemannian manifold Al. In ‘particular. Thkn variant. into x1. i= l. then the curvature fornz (2 = (Qj) is given by is a diffeomorphism of h4’ (and M”) onto the open subset of R” given by ilj = kei.the coordinate system of R”f-l and M the kypersurface o~‘R”+l dejined by T[(T + zi (J’)2)(ct (dyi)2) . \ve have CT*%$. . rfr < 0 - then M consists of two mutually isometric connected manifolds eack 0. .. (r + Xi (yij2j2 W” + -** + (x”)2 + rt2 = r (r : a nonzero constant).. 1211. the inverse mapping is given by xi = yt..~ = Rj. Thus. x E V. .c5js81i) a t !l.. considered as a group acting on.: . .+ (dx”)2 $ r dt2 In-. * + (x”)” -t . Clrl (Y’)~ + r < 0 . we have (xl. L. then M is isometric to a sphere of a radius-r!.4. 20-i F O U N D A T I O N S O F DIFFERENTL-IL GL’.

.lf) i s a n isomorf&iJm oJ‘ the principal jibre bundle G(G/H.(G) (d a real number. maps eac’h orthonormal fi-ame of -21 into an orthonormal frame. composite mapping n ofi G -+ O(bZ) a n-l maps an element of where X4s an n x n-matrix. We have a. n E G..tll\ isomorphic with O(n) : H = be the (left invariant) canonical l-form on G. then H i s nntur. .zw = 0. . . .1f at the point-a(o). a . The transitivity of G on M implies that the induced mapping f: G/H . -r YC = 0.* .(G).‘.~“)~.I. . . i ‘2 w i (a/&~‘)~ so that the frame u.\f) bc the frame at the point o = (0. 206 FOCNDATIONS OF DIFFERENTIAI~ GEOMETRY 207 V. . (3ja. In particular.j = 1. . . i ‘c 61 normal frame of . .. . . + aj I 0. Set Xi = a = .i +. Let u0 E O(. O.-4 = 0 and 0 and t are f(Q) = &o). d/9’ = -& c& A . . .). Then the condition for a to be in G is expressed her tXX + r . this transformation ( ri corresponds to the left translation by a in G under the isomorphism fi G -+ O(M). It is easy to verify that f: G -+ O(‘+i) cornmules with the right translation R. . (p) is invariant by the left translation -3 .*. On the other hand. IVc define a mapping f: G 3 O(J1) as follows. 1) It follou*s that the Lie algebra of G is formed by the matrices of E M given by (a/W). i. ‘Xy + T .. If we consider G as a group of‘ \ri 71: < (n + 1)-matrices in a natural manner.4”.. ) n. i. . (n + 1) x (n + I)-matrix: Proof of Lemma 2. . The mapping f: G -+ O(. for every a E H = O(n) : The Maurer-Cartan equation of G is expressed by J-V4 = f’(b) * Q for b E G and a 6 H = O(n).\f. .. To prove f *V = fl’. H) onto the bundle ()(.z ‘z = I. FORMS (0. . it is therefore sufficient to XY . . every element Qq O. we see easily that An (n + 1) x (n + I). n.. pi + ry. . n(~. 0.pi A yj. show that (f*P)(X*) = pi(x*) for all X* E T.(cf. L e t r3 = (f?) andto = ((II:) be the canonical form and is a bundle isomorphism. j = 1.M is a diffeomorphism and hence that J:: G -+ O(M) LEMMA 2. = 0. da: 1 -& cc. being the form . Let a E G. A a: . is given by (X.. Let LEMMA 1. n. * . E G induces a transformation of O(iZI) . 0..matrix a is an element of G if and only if 8 = (et) is invariant by the transformation induced by each ta&a = &. CITKVATvRE ASD SPXCF. Proof of Lemma 1.. X. . the connection form on O(M).. where ta is the tra:pspose of a. .\ is an ortllo.1/‘j . Then The quadratic form defining A1 is given by the following f *oi = /?i and f *co: = c$.I!. From the definition of 8. !t’c define where A is an n Y n-matris with ‘.j= l. As we remarked earlier. . ’ i. . . 1) fixed. The .~I ‘:. . elements of R” satisfying .? and z are elements of R” and 2~’ is i e d with the Lie algebra of G) of the form i en tfi T. A b an isometric transformation of . .. $’ + r * It’? : r. Every element n 6 G. . $4 of Chapter I). by each a E G. .

QED. .* + (x”)2 < u2: It is easy to verify that m is stable under ad H. if I. the bundle of afljpe frames over %M.1. \IC . respcctivkly. paracompact given by i Oi 0’. dx: = --Zka~Aaj --p’~y A Riemannian -manifoib of constant curvature is said to be and i elliptic. we obtamed (f*o*(x*)> *.rJEat (or locally Eucli<ean) according as the sec- tional curvature is positive. A manifold with a flat affine connection is . R) (cf. To see this.’ ’ facts: (1) (p’) corresponds to (19~) under the isomorphismf: G + M is a space of constant cu&ure urith sectional curvature >. (2) tl 11 ’ I’. (!y’)“j -.5 satisfies the equality: dp = -Xn ai A pk. complete space of together with zero curvature.7(i\f) --t O(M) in the same way as we y: A(M) --t 0 )<’ G ‘$hich maps’ the horizontal space at each definedf: G -+ O(M).‘(n q-(x*)) THEOR. . (3) the connection form (~j) connected. hjlperbolic . with respect to the Riemarinian metric given by m for every a c H. A n a@ne ‘connection of M is said to be fiat it M.++ (dx”+1)2 to M. These. let 3(i\l) be the group of isometrics of M and every point c#M has an open neighborhood U and an isomorphism define a mapping f: . 43 Remark. Gow a2[(a2 . . homogeneous and hence cdmpl&e’ by Theorem 4. ~5 A Ok.) dcfincs a connection in the bundle G(G/H.2 of Chapter The spaces M constructed in The&em 3. By ‘Proposition 2.l$mdle with structure group G -= A(n.. In the course of the proof of Theorem 3. Let g be the restriction of (dx1)2 + * . Applying Theorem 11.lla2.a space of constant curvature wzth sectional curvature lla2.1 of Chapter II we see that (3. . Therefore.. Let g and f~ be the Lie algt:bras of G and H. . M will be a connected.. that is ad (a)(m) = Then. . and Riemannian connections showing that the curvature form of the Riemannian connection is Throughout this section. . mannian connection form (0):) is characterized by the proper1). (2) Let M be the open set in R” dejined by (xl)” + . do’ = --C. . Then G c 3(M) andf: S(M) -+ O(M) u E A(U) into the h&fzontal space at v(u) of the canonical flat connection of U x G. .(Ci yi &“)2] the second asscrtinn of Lemma 2 follows from the following thred (u2’+q (jq2)2. Theorem 7. principal fibre .* E ‘Ik(G). + (d$‘)2 gives a simply connected.(M) where b’. (1) Let M be the sphere in R”+l dejined by = (P(n*). Then.10 of Chapter \i I). with space of g consisting of matrices of the form respect to this Riemannian metric g. and hence O(M) and si’ncef(G) = O(M). which proves the first assertion of thq. 6* are the components of 6. that the torsion is zero (Theorem 2. . Lemma 2. negative or zero. M is a space of constant A manifojd of dimension n. (9)” + .2.f*o’yx*)) = u.1.lemm. 4.>2 are’ ali simply IV).EM 3. we must have G = 3!nl).4. p-(x*)).. O(M). spaces are implies also called space forms (cf. Flat afine. then r of = is an extension df f: G -+ 0(-W). it is 1a curvature with sectional curvature l/y. of Chapter IV.. CURVATURE AND SPACE FORMS 209 into the vector Xi 6$X.ISi fyi)‘) (C. * ** . . * + (x*+1)2 = $. M is. Since f maps 3(M) 1: 1 into Xi p’(X. The space R” with the Euclidean metric (!x’)~ + We shall now complete the proof of Theorem 3.: “L& A(M) be. that is. The group G is actually the group of all ikoriktries of of Chapter $$I). 208 FOUNDATIONS OF DIFFERENTIAL GEOMETRY V. . E T. . H).*)A’. ‘Let m be the’linear sub.

compact. ’ QED. with. . Pmof. Let u. k I n. The equivalence of (1) and (2) follows from Proposition 3. u. CURVATURE AND SPACE FORMS PlU FOUNDATIONS OF DIFFERENTIAL GEOMETRY said to be local& a@ne.) are discrete. of Chapter III. Euclidean) if the Riemannian connection is a flat at&e connection. of the flat affine manifold is discrete in O(n) (cf. the following the group pf affine ing our assertion. the linear holonomy group Y (a. . form a basis of R”. Let hf be a manifold with a complete.e. 6. . . sional Lie groups.2 and connection and M’ the holonomy bundle through u0 of: the car- P the remark following Theorem 4. group R” and its invariant Riemannian metric 1s given by ditions are mutually equivalent: . where a.metric Chapter III). manifolds of M. of R2 with coordinate sytem (x. . .1. the connection in L(M) is flat. On the other hand.y) forms a discrete subgroup of Remark. The homomorphism /?: 11(M) -L(Alt) defined The action of G on Rn is properly discontinuous and R” is the in $3 of Chapter III maps M*.2) we shall see that the since the metric must be parallel. Jat &ine flat Riemannian metric on R”/G:‘ The manifold an/G with the Riemannian metric thus defined will be called a Euclidean cylinder. (2) Its .2 and 8.2). then. Kuiper [l]. . Theorem 4. (dx’)2 + + . . f2..1 of Chapter II. tk be linearly independent elements cipal fibre bundle over M whose structure group is the affine of R”.+ (dx”) 2 of R” is invariant by G and hence induces a THEOREM 4:2. Hence M * is a covering manifold of hf’. . For an a&e connection of M. we mean that they are O-dimen. Let hf* be the holonomy bundle It is called a Euclidean torus if tr. . Then M* (resp. both M* and M’ are covering G = {C mi[i. By Theorem 9.y+n). responding flat linear connection. It&II be’shown Let M be locally affine and choose a linear frame u0 f L(M) t that the linear holonomy group of a corn act flat Riemannian A(M) . mutually equivalent: uously oq R2 and the quotient space R2/G is diffeomorphlc with a (1) It is flat. Let C be thesubgroup of R” generated by [i. In fact. The equivalence of (2) and (3) follows from n.m = 0. mo$over.. it is a Euclidean torus. (3) The linear (or homogeneous) nection on R2/G. Our (aY)2 (1) It is Aat.y)-+(x+ny+m. The flat affine connection of R2 induces a flat affine con- curvature vanishes identically. 0nto. . E L(hl) c A(M). . the foliowing con. . .. Let 5i.M (cf. connection which is not Riemannian.2. In fact. . thus prov- discrete in the affine group A(n.orus. M’) 1s . i e through u0 of theflat ajine connection and -11’ the holonomy bundle through k = n. it acts properly d. The‘ Eu&dean . .3). i. This was taken from (3) The aJi<e ?zolonomz group is discrete. Riemannian. Example 4. Let M* be the holonomy bundle through u. R) (cf. if it is Riemannian.)). The set G of transformations flat if and only if its curvature form fi on A(M) vanishes identi- cally. mi integers). a dx dx + 26 dx dy + c dy dy. Every connected abelian Lie group with an invariant * *. 3 need not be discrete in GL(n. Later (cf.5 of universal covering manifold of R”/C. the umversal covering group of such a Lie group is isomorphic. Then . b and c are copstants. for a . Proposrtlon 3.a prm- Example 4.4 (x. Theorems 4.linear connection of M. A Riemannian manifold isJIat (or locally &mannian metric is a Euclidean cylinder. . R). 211 V. of the corresponding Jlat linear connection.rscontm- conditions are.* + (dx”)2 by a proper choice of basis in R”. . + * . Since @(u. This flat affine connection of R:/G 1s not holonomy group is discrete. Al.a vector THEOREM 4. Similarly... the proofof 4) ofTheorem 4. tk: holonomy group (D(uO) (resp. G 1s not a affine holonomy group of a complete flat affine connection is group of isometries of R2 with respect to this metric.) and Y(u. connection..e affine holonomy group and the linear metric on the universal covering space R2 must be of the form: holonomy group are discrete. (2) The torsion and the curvature of the corresbonding linear connection The following example shows that a torus can admit a flat affine vanish identically . But the linear holonomy. and if it is. ” _ assertion is now clear. . an affine connection is Example 4. the induced Rlemanman When we say that th.1.1 of Chapter II.

). nian manifold. . then it 1 LET&A 1. and the first homotopy group of M’ is isomorphic to metric on the abelian Lie group M’. . . . . . Z T. A”). . w. . .(c) Let X be a vector . an) E R”. 0 5 t 5 1. be the image of 7* by the projection R” . . (3) If M” is a Euclidean cylinder and is a covering space of M.f M is a compact Jat Rieman. such that p(X) = c’ (ci: morphic with N. n. Let 7 * =x:.. . . The affine parallel displacement from 0 to ‘connection on M’ is naturally isomorphic with N (cf. . G = (l$=i nili. proving (1). By Lemma 1 . (P. we can write M” = M*/H.(R”). . constant) for i = 1.5 of and hence is contained in the kernel N of the homomorphism . P) E R”. ) An) . A &. Proposition 6. P(i3[W). . . The group N consists of pure translations only the tangent affine space A. CURVATURE AND SPACE FORMS 213 (1) M* is the univer Cal covering space of M and. . . The affine holonomy group of M” is H by ci. .). . a(M’) is horizontal. . the affine holonomy group of (a) O’. 6% form a basis for the space of covectors at every point of M’. n .j .j= I . . Note that (b) implies that the group is aJne space A”. that is. . This completes the proof of (2). Proposition a = (a’. ) an) sends (Al. i. . Then the.. i. Set and let T = x. We assert that. m. R”/G is a group of translations isomorphic with G.3 of Chapter II). . . Let CT: M’ -+ A. Since X* is complete (cf. . Then the afline parallel displacement along T yields the translation The structure equations imply that dei ~‘0.. We identify the tangent space T. . (11. . be the structure equations on L(M’) of the flat affine connection The linear parallel displacement from 0 6 R” to a c R” sends of M’.j = 1 . ’ \ 3’ Cf=. n. Our assertion follows I ‘ Being a covering space of M’.(R”). X Proposition 9. Proof of Lemma 1. M * is also a Euclidean cylinder.3 of Chapter II. If M” is a Euclidean cylinder. be a Euclidean is a couering space of M’. (b) d@ = 0 for i F 1. 6. . 212 FOUNDATIONS OF DIFFERENTIAL GEOMETRY V.t. The completeness of the connection implies (c) as follows. . Let R”/G. Then 7 is a closed curve in Rn/G.R”/G. the affine holonomy group H consists of translations only (cf. so is X. TJR”) ia C.0 G t (. . cylinder as dejned in Example 4. . . . . As we have seen in Example the kernel of the homomorphism @(uO) -+ Y’(uJ. . Lemma 1) corresponds to X*.i= I .bealinefromOtoYZ~~. M’ is n It *clear that t?fil + * * * + 8”&’ is an invariant Riemannian Euclidean glinder. the first homotopy group of M’ is iso- . miEi = (a!..field on M’. n. . . M* is trivial. .here H is a X* be the horizontal vector field on L(M) defined by V(X*) = subgroup of @(t(. from the following three facts: By Proposition 9. . it is isomorphic to the ordinary Chapter III). Let each point a E R” with R* by the following correspondence: da” = -Zj 0. . l. . . In +.M* must be the ordinary affine space A”. 4. into (A1 + a’. considered as ah element of 9. n. for an arbitrarily chosen point o of M’. = -& w.(R”) at Proof. Let @ =b*fji. An) -+ (A1 + al.mi~icC L(M’) be a globally defined parallel field of linear frames. .. &. Let N be the kernel of the homomorphism @(us) -+ Y(u. . . with respect to the JEat afine connection induced on M*. Let covering space of M. we have G$ = 0. Under the diffeomorphism CT: M’ + a(M’).. = a*w! i. . M’ is a Euclidean cylinder. there exists a unique abelian (A’.‘jj. element and that the forms e’ are invariant. . (2) With respect to the flat afine connection induced on M’. Since M’ = M*/N. integers}.a%): group structure on M’ such that the point o is the identity This completes the proof of Lemma 1. A”) c To(R:) into the vector with the same components Since M’ = M*IN. Then X is complete in the sense that Let M” be a covering space of Al. Since h/i* is the universal it generates a l-parameter group of global transformations of M’. . the affine holonomy group of the fl& affine (Al.I” + an) E and the linear holonomy group of M’ is trivial. . .==. abelian.1. . . Since g is horizontal.(R”). . @ne holonomy group of (4) M’ is a Euclidean torus if and only .3 of Chapter II.1. . .

we In other words. I. flat’oithogonal matrix (called the rotation part of the motion) and p is an element of R” (called the translation part of Wemayalsoassumethat~. . = n(+J. and is to prove that R”/N is a Euclidean torus. Let L4 and B be unitary. fact. we may . . A. q)+ sin (Bj . This lemmas.‘ Y) we set commutes with ABA-‘B-l.pi) = 0 for i = 1. q/ real parts. .l<&1+r=***= the motion).&. . S. L EMMA 2. . we choose an inner product in TJM).pi) > 0 . . the subgroup of G B. k = 1. The Al 0 flat affine connection of M is the Riemannian connection with A2 respect to the Riemannian metric thus constructed.-. . The latter follows from the inequality of :Schwarz.x-+Ax -k-p.tPa -=c*** 5 /&.. It follows that M is compact Similarly. we have.j”=l (laj2 + kq12) *sin (pi . where A is. Proof of Lemma 2. X<Rn. where A = (a. . LE M M A 3. we may conclude that M” is a covering space of M’. This motion will be denoted by (A. . matrices of degree n such that A For any matrix A = (a. Since A-’ = “A and B-l = tB = i?..) of type (1. . a”3 = a’2 = (I for i 5 pr and j > pl. V. ELery Euclidean motion of R” is given by . xc. dAB) 5 d4 : d4. (A. . we have I . shows clearly that A and B commute. . p) and (B. Comparing the imaginary parts. p2. and then we extend it to a Riemannian metric by parallel displacement. = bJ. Without loss of generality.). thus proving (3). set for i 5 p1 and j > pr. p).. . This implies that the flat affrne connection of Al is Riemannian.) -4 ‘l’(~~. P)-‘(B. n.’ J-1 sin flk. Pd = (4 P) (By q) (A. Suppose . We first prove several I. consisting of pure translations. .. where G is a discrete subgroup of the group of 0 .* * are the identity matrices of degree p1. v(CA) = q(C). we obtain . . Given any two..i)-th entries. 12. . we have Comparing the (i. . = v(B). Let N be. connected. then A commutes with B.) of hl is a finite group.* q(AB). ‘5 A = Conversely. pp. . we may vector with rs components.zI*/N.!lk + q(A +B) I y(A) +@). = cos .1 assume that B is diagonal with diagonal elements b.=@2=-*~=&.i 5 PI +pz and j > p1 + p2. we have and the linear holonomy group @(u. = bp. In 0. . . Since all the bk)s have positive real parts. n. In view of (2) and (3) our problem where A. suppose M is a compact.j”=l ujbj?$hi = Cyz. If the characteristic roots of B have positive vim = (&j j@)~. Euclidean motions acting on R”. . y(A) is the length of A when A is considered as a have ABA-‘B-1 = Bd-‘B-‘A.. CURVATURE AND SPACE FORMS 215 2 14 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Hence we must have a(~. We have . .. .. Since &i’ = . write M = R”/G.). If A is an orthogonal matrix.. Since AABA-‘B-2’ = ABA-IB-lA. .@+z{.<p1 +n.M’ is a Euclidean torus. are unitary matrices of degree pl. By virtue of (1)) identifying 31* with R’“.p2. = ai = 0 f0r. q). we have ABt/@ zzz ABA-1B-1 = Brl-‘B-‘/l = BtJ&j. b. . invariant Continuing this argument we have by the linear holonomy group with reference point x0. nian manifold. B.Euclidean motions (A. .i.

. (AB .B) ..n.) < b . Let G be a discrete subgroup of the group of Euclidean LEMMA motions oJR”.. Since pk = 0. Set ’ Then.I) < 2kakb. . ii 5.(B .BA)d-‘B-l p.I)“-‘p. q(B -I) + q(B . (B .-l~-~?-~ %3. up= . fi2 = (. A.(1 _ /j)q) = 0.II2 + * * * + la.I)“-lp. R). . Thus -4 and B commute.I) < a and which proves our assertion. Q~(A. for some integer m.B)p .I) < ‘Jab. of an orthogonal matrix A with Then.4. . we obtain (4 q)(A. . v(Ak . p1 = A(Z . By applying Lemma 2 to 4.< by@) + add. Let a < 4 and G(a) = ((A.R 3 &I .“4 * cp(d . Proof of Lemma 4.(1 _ A)~) = 0.I)“‘-‘PI = (B .‘I . Lemma 3 establishes the inequalities.Q. . 2.)s.p. Proof of Lemma 5.(1 .q.~)-‘(A. By Lemma .). we obtain q(A . we have (lj-q(A. . = (A ..I) * p(A . 3. q) and noting that By the same reasoning as above.z)y(Z . Hence.F(P) + dZ .I) < + have positive real parts. we obtain Al = I. = ((A .Z)(B .. If U .. we have a unitary matrix such that I/AU-l is diagonal. (2) q(pA < V .A-l)B-lq. .Z)rN-l((Z . = Zandp.I) = c@(A .I)).i(q).I) < a).I) .l)k-l{(l .(P) + ak . = 0. Proof of Lemma 3.~. (B .p.I))A-‘B-l. Continuing this argum& . . from = (Ia1 .1 12)* < 4. UA U-l and UBU -1 are both diagonal. .I) < b. then we haue AA. rn there exists an integer k such that Al.I) .I)P~-~ = (A . Since A-‘B-l is an orthogonal matrix. = (A -.A)q.I) I_ ?(A’. LEMMA 4. = ( A . v(A . from (A . we see that 11 k-l = I. = 0. p) and {B. . (1) ~(‘4~ .-. we obtain @(PA I q(Z . . . .&-l)(A. we have By a simple calculation..p) l G.l)a’-‘b . = (A . we obtain (A .Z = ABA-‘B-1 . there exists a unitary matrix IJsuch that (A. .p)-l = (I. .I) and q(pk) approach zero as k tends to infinity. .216 FOUNDATIONS OF DIFFERENTIAL GEOMETRY V. Changing the roles of (.A-‘8 k-l. . = q(B .I)“-yl.._. .f) < 2ab.Z)‘“-IPi = 0 .Z)U-‘) = &VAU-’ . then (ai . We show that the characteristic roots al. .I) (A .I) Similarly.Z =.4.. for k = 1. If p(A .(P) + a * dd.(1 . A simple inductron using. With the same notation as in Lemma 3.B)A-‘Jo + AB(Z . Since G is discrete m A(n. a. z 0: i==l. .+. set Since 4 and B commute.) = (A.p) and (B.~)p .~)(A.4P . . (23 dp..I)?pl. Then any two elements (A. We have PI = (1 . P)V4 d-YA. CURVATURE AND SPACE FORMS 217 Let I be the identity matrix of degree n. q) of G(a) commute.

I)B-1) = fp(A . . We define a vector t E ‘Rn by L E M M A 6 . . AiOi + Pi = BiPd + Qi fy(BAB-1 . : ponents. q) (A. Hence we have group of G. I f H is an abelian normal subgroup of an irreducible L group G of Euclidean motions of R”. q) commute.a&. Indeed. where (4 P) c H. In other words.I) = y(BfA . then P.I) < a.$}=O. -t) A = . This shows that the group generated by G(a) is an invariant sub.Zk. A subset V of R” is called zi Euclidean subspuce if there exist an ” (B. there exists an element (A. & . A = I. {X + x.12)ti = qi for all (B....218 FOyNDATIONS OF DIFFERENTIAL GEOMETRY V.A)q = 0. . (A-l”)t= .-Iz)-‘(Bi .. for all (A. where I. . . we may assume.(1 . (bi .Ia)(A. . . is non-singular. so that Ai . Moreover. (4. P = k (A.IJqi = (Bi .Is>‘* = (B.In)t). p) Q H such that Ai . By Lemma 5. which completes the proof of Lemma 5.--2k is th:: identity matrix of degree ‘n . 4 Pl PI ‘0 L-21 b I> >I’ .n.I)“-‘((1 -bJr* -(l -a.(A.)r. ” . . i. We say that a group G of Euclideatrmotions of R” is irreducible if R” is the only Euclidean subspace invariant by G.2k com. Assuming k 2 1.)pi.e. we have IW>f) eG( a ) and (4 q) E G.. q)-’ l G(a). thus proving our assertion. For each i. choose (A. (A.Is)pi element x0 E R” and a vector subspace S of R” such that V = = (Ai . we have We shall show that p1 = (I .). . p) of H are simultaneously 1 reduced to the following form : A. i = 1...B)p . is non-singular Hence we have and define a vector li with 2 components by (1 . or (Ai . (Bf .Is)-‘pi = (Ai . a = l. that the elements (A. each pi is a vector with 2 components and p * is a vector with n .:. we shall derive a contradiction. then H contains pure translations only. q) l H.(I . 0 We have now . for each i.I..I. p) E H. x E S}.. p) of’H Pk’ Pk such that Ai is different from the identity matrix I. Since (A.p) and (B. P) (In. . Proof of Lemma 6. t) (A. t> (4 P) (4. ) n. I iifP+ 0) . Since H is abelian.&)-l(A. by applying an orthogonal change of basis of R” if necessary.I. CURVATURE AND SPACE FORMS 219 we obtain Our task is now to prove that k = 0. it is moreover abelian if a < 8.I& = pi.Ii)q. = (A.b. . t1-I.. P) (B. = 0. = qi.

is non- singular. Let N be the subgroup of G co@sting of all pure translations of G. We shall show that V is also invariant by G. we -b&e N . Let G and prove that the first 2k components of r are zero. we. Since G(x.2k. . is compact. p) of H are of the form element (C.. hand. translations. we may now assume that the Since (Ai . it is an abefian normal subgroup of G. ular to V. L&WA 7. . On the other. the distance lies in V. This proves that R”/N is a Euclidean torus. we laihemphasis on affine holonomy groups.. (B. (4) was originally ’ . Assuming that G is not irreducible. and X. G/H is finite by construction of G(a).) and = yi. Y) = (C. Then V is invariant by all ducibility of G. is equal to m. it denote the orbit of G through x. . p) (C. This means that \0 C” 1 R”/G is not compact. be a sequence of points on L such that . [ 11. C = C’O ‘. discrete.f where c’ and C” are of degree 2k and R . thus contradicting the irre- whose first 26 components are zero. ’ (4) means that the linear holonomy group of a zero vector of Rfi and comparing the (2i . Applying the equality (A. Q E D . for each (A. Although u_(2) and (‘3) are essentially in Auslander-Markus ” A. Let H be the group generated by G(a) .. A = . Let xl.220 FOUNDATIONS OF DIFFERENTIAL GEOMETRY V.c =. then ACv = CBv = CU.& let (A.of J&+&an m@ons of R* is irreducible if First we observe that I/’ is precisely the set of all vectors which are R%/G. in R”jG a . Let (C. . CURVATURE AND SPACE FORMS 221 Al 0 By translating the origin of R” to t.P= : * o c”. xa.. .It) is non-singular.the distance between x.T j left fixed by all A where (A. p) of H. compact flat Riemannian manifold M = R”/G is isomorphic to nents of the both sides. . . Since G(a) contains N.) is in F’.. have G/N and hence is finite.. we obtain yi = 0. . q) to the Remark. let *V normal subgroup of G. To We are now in position to compltie Ic proof of (4). Hence R”/H is also compact and hence is a Euclidean torus. p) c If. W. ’ elements (A. p) be an element of fl such that Ai .I. r= ( 1 . T) of G is of the form 0 . Lemmas 6: and 7 imply that H contains nothing but pure.r H. be any point of V and let L be a line through x0 perpendic- (A. P) G 4 = CC. and X~ is equal to m. Thus . is at least. Hence C is of the form bettieen G(x.every clemcnts (A. . r) E G. . A group G . respectively. m and.I* ‘i i. Assume that R”/G is compact. /C’ 0 \ Therefore the distance between the im#ges of x. there exists an element be a proper Euclidean subspace of R” which is invariant by G. Ak 0 0 La* ‘P* Let V be the vector subspace*of R” consisting of all vectors This shows that V is invariant by G. since G is Z= . Sinp H is a Proof of Lemma 7. Since Cv is left fixed by all A. hence. This completes the proof of Lemma 6. x. d. t) (B. q) E H such that j Let x. . Let G(xO) If v l V. by the projection R” + R”/G is equ to m. For each. p) E H. write G(a) be as in Lemma 5 and assume jr c +.I)-th and Pi-th compo.

.n-1’:. JI) --f (x + ma + ?7. Euclidean cylinder and hence is a Euclideae first prove that eyery clcmcnt of G \\. il. . . . fix an arbitrary (2) Ordinal torus (orientablr) irrational real number 1..j). memannian . m = 0. M” is a covering space of M’: Thus. 111 -= 0. CURVATURE AND SPACE FORMS 223 C proved by Bieberbach [ 11. \Ve M’ is a compact.1 is COROLLARY 4. R2/Nin this. &l.c .xl. A &n&old J4 wit+ a&t a&e connection admit.3. By (3) of Theorem 4. We shall present here their results &ith an Endication’of the assumption that G acts freely on R2. Iffis an element of G lvhich rc\Tcrscs the orientation There are four types of two-dimensional complete flat Rieman. . QED. By Theorem 4. .k itcrate the transformation z -+ EZ -A-.l h).2. The proof goes roughly as follows.E) is left fixed nian’ manifolds is due to IGlling-[1. 4~2. - 0 0 Then we SC..+& rn‘? O. J. m = 0. Esample 4. &2. Riemannian m&old. \\-hich proves proof.Jt is easy I (4) Klein bottle or histed torus ‘in6-biiiAt:. .Let. . transformation of the type describing its action on the universal covering space R2 in terms of the Cartesian coordinate system (x.If “6 ’an o US b dw1z ‘tl z zn ’ nlj e‘twidth or twistedc~~Iin~/er (non-oricntablcj (*%JJ) . If ~.p(m)) .case:* transformation and nian manifolds other than the Euclidean plane. . Indeed.. . Let G be such a discrete group. By (3). complex number.2. Riemannian manifold may not be finite. We thus proved that &. .‘r 1.)I) -+ (x -:.. if E + 1. in contrcldiction to the [ 11. up to a constant factor. The linear holonomy group of a non-compact flat 12 = 0. isometric.hich prcscrvrs the orientation . . .(x + n.jl...21 and H. to see that G is a discrete subgroup of the group of Euclidean motions of R3 and acts freely on R3.’t (X. (x. b .3.. then we obtain the transformation . ’ by the transformation z + FZ . 6: non-zero real number. :eopf. . 0. Hcncc F = 1. Any two-dimensional complete flat Riemtinnian manifold . The converse is contained in (4) ofThorem 4-2. + 1 )w* direct product of a circle with a’ line. . FOI* each integer m..f h/t is a compact f?aL gf surfaceS. Then every manifold. . torus. }. ’ z --f &‘Z + (&-I + pi? . ..}. 9(m) = -sin ’ilm& cos Amn (3) . N be the subgroup ocG consisting’of trafislatidiis: I “ ’^ where E is a complex number of absolute value 1 and w is a (x. We see easily that.(X -+ n. n = 0. G = {(d(mj. It: 1) &2.2 is given by times. 12. w r Then the covering space M’ defined in Theoiem 4. (-l)‘. that is.!)I. orientation preserving motion of R’ is of the form Example 4. cos Lmn s i n Am77 n.) .cry clement of G is a fundamental group (the first homotopy group) for each type. we set (x. to one of the above flour types a E$jdean torus 4s a covering space t$ and on@ . +2. The proof given here was taken from (1) Oxdinar_y r_ldi?lder (orientablc) Frobenius [I] and Zassenhaus [‘l]. of Theorem 412. Let M” . f 1. b: real numbers. ’ . Set z = . the Proof. We give the hence is a translation.In Exapple 4. ... then the point x~/(I . of R2 is necessarily a translation. linear holonomy group n. of R2. the _. ~ > of R3/G is isomorphic to the group {A(m) . set M = Rz/G.4. . (-l)y). Klein [I. .2].i.. .J’ -” mb) nz. . 222 FOUNDATIONS OF D I FF E R E N T IA L G E O ME T RY \‘.. . M is a compact flat.. motions acting freely on R”.y) --+:(x 47 tp. ’ / The determintition of the P-dimensional complete flak Ridman&.be a Euclidean torus which is .a cover&space problem reduces to the determination of the discrete groups of of M. then J” ‘is an orientation preser\. $2. .‘. h4’ is Itn ordinary cylinder.x -.2. our assertion.

R). Proposition 1. Throughout this section.M’ be an a@ne mapping. where z is the complex conjugate of z. that M must be one of the above four types of surfaces.My of class C* induces a continuous mapping f: T(M) -+ T( M’).) = A’&. that is. that is. where n = dim hl and n’ . It is now easy to conclude . Let X. Consequent&.1. fo exp X = exp of (A?). A differentiable mapping J: Al . . We recall that a vector field X of M is f-related to a vector field X’ of M’ iff(X.lf) and L(n4’) of M and M’.224 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 24zfw o r z+Z+w. A@ne mctppings and afine transfomations Let M and M’ be &iifoIds provided with linear connections I’ and I” respectively. respectively. respectively. we denote by’ P(M.1 implies that an affine mapping is necessarily of class C” provided that the connections r and F’ are of class C”. R) and G’ = GL(n’. respectively. G) and P’(M’.M’ an afine happing if the induced mapping f: T(M) -+ T(n4’) maps every horizontal curve into a horizontal curve.2. for all x c M (cf. where T(M) and T(n4’) are the tangent bundles of M and M’.dim M’. The CHAPTER VI detail is left to the reader. We call f: M .(M). f commutes with the exponential mappings. iff maps each parallel vector field along each curve 7 of M into a parallel vector field along the curve f (7). P ROPOSITION 1. An ajne mapping f f 121 --+ M’ maps evel. Proof. Y and Z be vectorjelds on M which are f-related to vectorjelds X’.&t/z M .y geodesic of M into a geodesic of M’ (together with its a&e parameter). so that G -= GL(n. Y’ and ‘2’ on M’. G’) the bundles of linear frames’L(. Then 225 ‘ . QED. $1 of Chapter I). This is obvious from the definition of an affine mapping. Xc T. . PROPOSITION 1. 1.

. . Y)Z isf-related to R’(X’. the induced _ _ of M such: that F = fy Since f-maps every horizontal curve of P automorphism f-of the bundle P of linear frames leaves the canonicaljorm into a :horizontal curve of P. curve of T(M).(M). then the induced automorphism thus completing the proof. The automorphism fof P maps the connection I’ into a connection. torsion tensor fields of M and M’. Proposition 6. Jinduces the identity transformation on the base M. respectively. respectively. Go) and inzariani . Then J is a fibre-preserving transformation of P leaving 8 invariant. VI.flu)-l( fX).(P) P(M. we have I .(P). Hence. (2) T(X.. ‘X. .‘ . . every jibre-preserving transformation of P leaving T(M) maps every horizontal curve of T(M) into a horizontal 0 invariant is indticed by a transformation of M.. . Conversely. 92 of Chapter III) tioi both in M and M’. TRANSFORMATIONS 226 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 227 Proof. Moreover.(P) andsetX=rr(X*) sothatX~ T. QED: f-of P leaves both the canonical form 0 and the connection form co invariant. T). (2) Let f be an affine transformation of M. where R and R’ are the R” onto T. (1) For every transformation f of M.(M).) at’ x E-Y into the onto the horizontal subspace of TzU. say P+(M.. . . We prove that f”= F. that This implies that isif” = F(u). we have induced by F. the following diagram is commutative: Proof.1 of Chapter III.’ Since f commutes with parallel leaving 8 invariant. fX.lYhis means thatf: M -+ M is an affine mapping (2) Iff is an afine transfokation of M.) and let 7:” be the parallel displacement along this Conversely. Since f 1s an auto- I’ and hence f-*co = w. definition offthat. Therefore. Let f be the transfor’mation of the base M displacement. respectzvely. for each u d P. Then the bundle P’ Concersell! eve7yJibre-preserving transformation of P leaving both 8 and (0 consists of two principal fibre bundles. let F be a fibre-preserving transformation of P field of P invariant. there exists. (1) Let xt be an integral curve of X such that x = x0 and I let 7. and the A. (2) and (3) f 11 J(u) = U. a transformation f P R O P OS I T I O N 1.- U-‘(X) = 0(X*) = e(JX*) = J(u)-‘(X) for X* Q T. the horizontal subspace of T. to x’ = x0. diffeomorphism f of M onto itself is called an afine trunsforma- form f**o is the connection form of fir) (cf. Assume that M is orientable. J = fzl 0 F. where x = n(u). . we induces in a natural manner an automorphism f-of the bundle see that f-maps.s induced by an a#ne transformation of ni. y’. the transformation f: T(M) + 0 invariant.Y isf-related to V. th u s p roving that 8 is invariant byf: Set xj = f(x. We set. From the definition of an affine transformation..(P). By (l). G) . u-‘(X) . Any transformation f of M Chapter II). f maps a frame u = (Xi. QED. be the parallel displacement along this curve from xt to x = x0.(M) and T. Then (cf... o ow from (1) and Theorem 5. say. morphism of the bundle P. where V denotes covariant diferentia. Remark. $1. Y’)Z’.. This means thatf(I’) =’ frame j(u) = (fX1.) at f(x) l M. where the frames i andf”( u ) are considered as linear mappings of (3) R(X.f”(I’). It follows from the curvature tensorjelds of M and M’..Y. where T and T’ are the e(x*) = u-l(x) and O(fk*) = f”(u)-l(fX).3. of Chapter III) fiu) TdW -j+ Tf(*) (Ml * (V. leaving 8 and w invariant. Y) is f-related to T’(X’. = ynf (TkY. Then (cf. let F be a fibre-preserving transformation of P image curve from xi.Y). it leaves every fundamental vector Conversely.). (1) V. .1 of tion of M if it is an affine mapping. . (1) Let X* E T.

. that C z GL(n. For a trans. Then any transformation F of Pf or P. there exists a implies that fZ is parallel along f (7). for ecep a c G. Y* d T”(P).in a fibre by therefore jinite number of connected components. where w is the connection form of I’ and f-is the trans- g(X*. every vector field X on M (v) (4) -+ (1). . PROPOSITION 1. The group of isometries of P is a Lie (ii) (2) -+ (3). where ? = intd a horizontal subspace. . the space of covectors at every point u of P -(cf.1f. Then the group !ll( A-l) of ajine mapped into a curve. Let r be a linear connection on M. This means thatf(I’) = I’ and hence d i m . unique vector Jield ‘. We set (1) f is an a$ne transformation of M..(fZ) for any vectorjelds Y and Z on M. -This follows from Proposition 1. for every u t P. Injnitesimnl ct@ne transformtithns (iii) (3) -+ (2). (4) PROPOSIT&:$ 1 For each vectc r jiekd A’ on . ‘u( I’). We prove that leaving 0 invariant is fibre-preserving and hence is induced by a B(M) is a Lie group by establishing that VI(P) is a Lie group with transformation f of the basegYM. tranfforniations of M is a Lie transformation group with respect to the are connected. By Proposition 1. also Theorem. Let Y be the vector field along r tangent to 7. topology by Theorem 4. .4. Proof. R).228 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI.. Let 8 = (ei) and rrj = (u{) be the canonical form and formation f of M.. wj(X*)w~(Y*). k = 1.@. denoted by 9t(M) or (3) x’ is n-related to X. Let Z be a parallel vector ‘field along a curve induces a vector field X on Pin. R(E) = B(E). which is canonically isomorphic with ?I(M). denoted by 91(P). b’denotes the bundle of &car .‘-B(5)).:.3. <!J!. the following conditions are mutually equivalent: the connectioq form on P.t+j @anner.I&.ph that transformation. This means thatF1.:M induces a transformation of p in (iv) (1) * (4).Riemannian metric on P which is invariant Proposition 1.. Eycry transformation Q of.?O = 0. The set of affk transformations of M.9 of Chapter I (cf. We extend Yand Z to vector fields defined on M. “‘:J?. 2 . Let r be a linear connection ori a manifold M with a 0(X*) = 0. F is. TRANSFORMATIONS 229 P-(M. ~(~r”) = .. In fact. a natural manner. GO). This means thatf is an. (2) L.‘. Th e horizontal subspace at u is given by the set of B(t). The set of all fibre-preserving transformations . forms a group of G = GL(n. (2) implies * closed subgroup of the group of isometries of P. compact-open top0100 in P. fibre-preserving. formation of P induced by f. (2) f**w = o. (i) The equivalence of (1) and (2) is contained in Chapter III). Y*) = ISi eyx*)ei(. R). forms a group. we have transformation group of P with respect to the campact-open 5 = 8(B@)) = (pep(t)) = e(j-‘B(t)). of P+ or P. that is. 3.-.10 of Chapter IV). QED. 0 = c@(E)) = (fl*o)(B([)) = w.respondin$y.fkamcs over a m$fold J1..: Y = it.. so.P(#.~ j*. which 1 w% be denoted by the same letters Y and Z respectively. (1) Ar is invariant by ‘R.6 and Corollary 4. . i..2. ii n. s . j. Proposition 2. . that is.. (3) f-leaves every standard horizontal vectorjeid B(E) invariant. g is a.. we prove i .is mapped into a vertical vector by F since O(FX*) = T HEOREM 1.y*) + Cj. Any curve in any fibre of Pf ot P. More precisely.3. a(P) is also a Lie transkmation group of P. Since the n* + g l-forms 8’. every vertical vector X* respect. by ‘u(P) by Proposition 1.5. form a’basis of (4) f(V. Since 41(P) is cleariy a Since o(B(t)) = 0. A. 7 = x. Sin& the fibres of Pf or P.6 of Proof. . Hence (3) implies that f-maps every horizontal subspace Throughout this section. to the compact-open topology in P. C?.? & AP . where Go is the connected component of the identity of P leaving 0 and (~1 invariant.3.kl..Z) = U. = 0.

Finally. Par a vector ‘. The uniqueness of X is evident. Because of (3). 0 L. (2) and (3). the above formula is (1) X is an injnitesimal a$ne transformation of @. Let qt be a_ local l-parameter group (iii) (3) -+ (1).. if each q??t: U ‘+ M is an affine mapping. TRANSFOkMATIONS 231 Conversely. the vector field X on M. natural lift of X. Proof.::We fix a point x E M.Z . :.3. A vector field X on M is called an intnitesimal a#& trans$ormation of M if.a vector field. let Xi be another vector field on . will be denoted by X. manner. . For each t. If Lgo = 0. then gt leaves B(5) invariant of local transformations generated by Xi. S ince X satisfies (l). F or every vector field X.K. and preserves the canonical form 8.y f [V. dVYZ) = V. we have I implies that X = 8.(U)) in a natural.I.4. By the lemma.Z = ‘. Let r be a’ linear connection on Ak.2.true. vt induces a trans. for each‘t. X satisfies (2).1) &: 71 -l(U) -+ P and hence the induced vector field on P. let “qt formation of P induced by pt. B(l)] = 0 for evev l 6 R”..*YhZ)~ for any vector fields ‘Y and 2 on ‘M. Corollary 1. which { (iv) (1) -+ (4). If K is a function. and Wt) = (v. ’ . F and X is an infinitesimal affine transformation group of local transformations yt of M. P satisfying (l). more precisely. a local 1Lparameter group of local transformations pt of a neighbor- hood U of x into M preserves the connection I’. where U is provided with the affine connection I’ 1U which is the restriciion’of I’ to ti. = Vt~r. 230 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI. . let X be a vector field on P satisfying (1) and (2).L. B(5)] = 0. have (2). we have (Proposition Thus we obtain a local l-parameter group of local transformations 2.. X..for Proposition 3. ihere B(6) is the stan$rd horizontal vectorjeld corresponding to [.:- evidently. 0 = &J@(E))) = (Lpwv)~ + 4Lc W)I) = 4cx mm To prove the uniqueness of X.3. a f G... Thus vYt =‘T~ and hence qt = gt. yt induces of M. > field X on M. By Proposition . ’ 2 I PROPOSITION 2. By Proposition 1. formation $t of n-l(U) onto r-‘(v. . (v) (4) -+Xl). set X.z)).cpi(VyZ)]. the (2) Lgcc.4. 0 L.V.which means that [X. which 0 = Jiwm = kww) + w-X m)~) = e(cZ wm. B(t)] is horizontal. which means that [X. the following conditions are mutually equivalent: ‘when K is. L. By (1) of subspace at h(y). . is independent of the 4 we obtain I choice of u and thus we obtain a vector field X which satisfies (3). where cr) is the connection form of r’ and X is the : formula holds’for any tensor field K. B(E)] = 0. $j6 preserves o. V(t) =i(&(v.. 0 V.K = ‘V. Conversely. 1 ‘>. _. a unique vectorjeld X on M satisfying (3).. . = 0.v.. whenever $5$(u) is defined. be a local l-parameter group of local transformations generated (i) (1) + (2).. 0 $. choose a point u e Psuch that n(u) = x.of Chapter . QED.K . (. : <.. B(E)] is vertical. there exists (4) Lx o v. given a vectorjeld X on P satisfying (1) and (2). (ii) (2) + (3). If [X. 7~ 0 qt = ~J?J~ 0 7~ implies (3). for every a E G. We set (3) [X.8 of Chapter 1). Let F be a linear connection on M. Hence we by X in a neighborhood U of x. For each x E M. [X.Hence. for each x c M..1. = 7r(XJ. Then qt commutes with and thus maps the horizontal subspace at u into. qt be a local trans- Proof. Since $t preserves the form 8.. then X satisfies (1) (cf. Given a vector field X on M and a point x E M. Let vt be a local l-parameter group of locai transforma- We shall call 8 the natural lift of X. Therefore PI preserves Proposition 1. tions of M generated by X and let. We’tlien From the definition of*Lie-differentiation given in $3 of Chapter I. s:We thus verified the formula: .1. the horizontal every R. it follows that $7t is induced by a local l-parameter the connection.rl for every vector jeld Y on M.&twv . Since gt commutes with R.

s. d X(O(B.l-. (LA. (cf. is open. Proof of Lcmina. It follows that a(M) we obtain is a subalgebra of X(M) isomorphic with a(P) under the corre- spondence X --+ 8 defined in PFoposition 2. Then vector fields on M into the Lie algebra X(P) of vector fields on P.‘.*. TH E O R E M 2. the Lie algebra a(M)0f tn’ jinz ‘tesimal a&e transformations of M is of differentiation d ahplied to the first structure equation yields dimension at most n2 + n.o% that P.Y]Z i. is im. On the other dim a(P) = n2 + n.EMhfA. then xun = 0 for every a E G as 8 A*(O(B. Then F is closed in M. + td.’ a(P) LX@ j: LsY(&9 + w A 0) = 0 and from (3) of Proposition 2. by x = n(u) c M such that . If $.. B’) = -A . given any element A E 5.4 of Chapter If.(P). then it Therefore.1 is an isomorphism of the Lie algebra X(M) of respectively. we obtain by b. in Proposition 2. The following lemma = lA8 o d 0 = L.. ‘Since M Proposition 2. . B’)) = A. Proposition 3.(@(& B’)). then. we have evidently V(0) = W(0).opo&@n 1. B’]). where. This dW/dt = v4(V[. In partic- QED. Let F be the set of points If we take as A the identity matrix of 9 ‘= gl(n.1 and also (2) of Proposition 2. we first observe that the exterior I?. Let b. $8 bf Chapter III).1. &et u be an arbitraw point of P.z.n = dim M. o LJQ.ariant by R.. B].mapping f: a(P) --+ T.’ there exists a -(unique) element Let a(M) be the set of infinitesimal affine transformations of 13 E a(P) such that & = AZ. X 1s mvarlant (iv). H’] = B’. maps a(P) onto T. .O = (d 0 ld. Hence. ular.ndard horizontal vector fields corresponding to f and E . it sufices to show that F is open.4. p. that is..).v. B] = B and [A*. Since [x. B’) + O([X. V(t) = W(t).O) (B. .). B’) + O(B. B’)) = (L. [X.. (cf.3. To prove the first statement it is sufficient to shoti that L.(Q A 0 - implies ttat the linear . linear mapping J f (3) = X. R). 0 L. Since L. = 0.(P ) defined by and f(J) = d\. B’]) = 0. Let a(P) be the set ofvector fields X on P satisfying (1) and (2) of Proposition 2. In fact.(M). where A* denotes the fundamental M.A. Let u bc an arbitrary point of P. B].2.l = W e coinpute both sides of the equality separately. If M is Q connected manifoold with an a@ne connection To compute the right hand side.O(B. TRANSFORMATIONS 233 232 FOUNDATIONS OF DIFFERENTIAL GEOMETRY For each t.. . since a(Ai) is isomorphic with OJ A 0) = -tu(A*) *@ a(P). we have . Assume xU = 0. be a local l-parameter group of local transformations [A*.. both the t&ion and the curvature of I’ vanish identicalb. If an elemtnt. all vector fields on M. In. As in hood of U.2. the correspondence X --+ x defined the sta. Then the hand. B’). both V(t) and W(t) are elements of T. 0 d)O a(P) is of dimension at most n2 + n. O(B.. In the definition of a normal coordinate system (cf.(P) = n2 +.(O(B. X. To prove the second statement. Let B = B(6) and B’ = B(5’) b. is injcctive so that dima 2 dim T.. we assume that dim a(M) = From our assumption we obtain N(t)/& = dW(t)/dt. Hence we have Proof. 0 L. 4bru) cover a neighborhood of x = T(U) when 5 and t vary. = 0.( = 0.. it is sufficient to prove that V(t) = W(t). Proposition 2. forms a subalgebra of the Lie algebra X(P). B’) + O(B.3 of Chapter III.. ..2.view of generated by a standard horizontal vector field B( 5) ic f Feighk- s Pr.O)(B. 0 = -i-J/i O. vanishes identically on P. Then a(M) farms a subalgebra of the Lie algebra X(M) of vector field corresponding to R.?.J? -of a(P) vanishes at some Point of P. ’ Froin L. If dim a(M) = n2 + n. [A*..f if+3 + do. then r is jut.1). I. B’)) = -A . we saw that the points of the form w)/dt = %((Lx o V&q’F. B(t)1 = 0 by Proposition 2. B’) + @([A*. and hence &.’ is connected.L. VI.

t.5.1 = -@. the existence of normal coordinate PROPOSITION 2. . 234 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI.(B. Y)> This means that by taking suitable L3(. which lies in the same fibre as U. t . the Lie that the curvature form vanishes identically.Y = -V’. 1) because it is zero on the Proof.X.1 of Chapter III. consisting of complete vector fields.. Then where R derwtes the curvature.y] = [Ad. .* . Then it follows that X is complete. Then algebra of a(M) . In other words. to the point x0 = n(us) by a finite succession of geodesics. It] < 6.40 bl .(VrX + [X. . Y] + 7-(X. B(tk) and an element a l G such that = -V. . comparing the both sides of the equality: . It suffices to show that every element X of a(P) is function algebra S(M) (cf. that is. a local l- parameter group of local transformations generated by X (cf.). .(U) is defined A.2 implies that b.(u) by- w. A [x.(u) = Qt 0 b. every infinitesimal a&e transformation X of M is complete. u = (bjl 0 bl”. . . . Proof.3 of complete under the assumption that M is connected. In general. vector field B(l) is complete since the connection is complete.5 of Chapter I). B’) = O.4 means that if the connection is complete. bi. By Theorem 5.’ a 0 = X.“u. QED. R).%LYI .T (X.1 of Chapter III. and conversely.T(X. We then Proof. . 0 btz 0 * * *0 b. %I . By Proposition 6. thus proving our assertion.(O(B. we have Proposition 1. We shall prove that F.[Ax> VY..5 of Chapter III.6. B’) + @. where each bf is the l-parameter group of transformations of P We conclude this section by generated by B(ti).‘B’)) of affine transformations of M gives rise to an element of a(M) and letting A equal the identity matrix of g = gI(n. Similarly.VxY = [X. . then o(M ) can be considered as the Lie T HEOREM 2. H/l. the derivation A. u = vu for a smtable a 6 G.Vx is generates a global l-parameter group of a&ne transformations of M. I?‘). then of an integral curve of a certain standard horizontal vector field. I?‘)) and that qt is generated by 8 follows from (1) of Proposition 2. algebra of 2l(M) can be identified with the subalgebra of a(M) We now prove the following result due to Kobayashi [Z]. For any vectorjelds X and Y on M.(u) whenever they are both def$ecD.(I?. By Proposition 6. o . induced by a tensor. Let u. TRANSFORMATIONS 235 Thus we have The fact that qt(u) is independent of the choice of bi.2 of Chapter III) and the a&c transformation if and or+ if connectedness of M imply that the point x = T(U) can be joined V.) = R(X. field of . This fact. we have Given any point u of P. (1) By Theorem 5. there exist standard horizontal vector AxY = LxY . be. (1) A vector field X on M is an injnitesimal neighborhoods (cf. bl L VLLYI = [Lx . and (3) of Proposition 2. . Proposition 8.0 b~u. the proof of Proposition 3. 0 g. B’)) = A:(O(B. B’)) = A:(R(B. = Lx . every geodesic is the projection (2) If both X and Y are injkitesimal afine transformations of M. every element of the Lie algebra of the group ‘U(M) X. Y) = Fbf. .tyPe (1. be an Chapter I). may be derived alao from the following arbitrary point of P and let pt: U --f P. every standard horizontal where T is the torsion. PROPOSITION 2.%w = CL.2. Let I7 be a complete linear connection on M. showing that the torsion form vanishes. Y] .(k.). we have define p?. we obtain a point v = b:.X .). . we’ see which is complete.]. .(!qB. Y). In fact. Y) for every vector jeld Y on M.4. for every u z P and It1 < 6. Y). QED. For any vector field X on M.. Theorem 2. Y)) fields B(t.1 + WX.3 of Chapter III. note that (3) of Proposition 2.)a.(B. 23’) + O&3.

. a . (1) This follows from the fact that a transformation f ? ‘(iii) Since V.7 canonical form 8 on 0 (M) invariant is induced by an isometry of M. Isometrics rend in.1541y. into itself. fi$+. QED. By (l). the the ipte&al onme of integral curve oj ‘8 through each point of O(M) is contamed PROPOSITION 3.g = 0.(V. QED. 236 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI.. . If X is an infinitesrmal Consider the bundle O(. J induces the identity for Y.afie transformation.2.L[. maps every function into zero.1.y. transformation on the base M. Z) = -g(A.VYI . .v A-.rthe local I-parameter groups of local transformations is necessarily an affrne transformation of .(J. Then J is a fibre-preserving mapping of O(M) A. We Proof.1. $$ ‘be +. We have Conve.Z) . (1) A trans formation f of Ai is an isometr_y . let &and know from Proposition 2.. (2) The natural lift $ of X to L(M) is tangent to O(M) at every QED.&. We set J =j-r 0 F.0 for all Y and Z. and X respectively. IIif --2 --is tangent to O(M) ‘at every point of O(M).&! is a for qrbitrary’pe@orjFel& Y and Z. we consists of local’isometries.dl with respect to I’. formations generated by X in a neighborhood of each point of M (2) BY Theorem 5.] for every Y.U) of orthonormal frames over A1li isometry. x = 7(X*).(A. Y).Y] = [Lx .&)):= (ksg)(Y. thus proving the equivalence of (3) and (4). the following coiditions are mutually equivq~ent : (1) X is an injnitesimal isometry..g = 0. (i) To prove the equivalence of (1) and (2).ye have (2) Letf be the transformation of the base M induced by F.I. This’means. For a vec@ij&ld X on a Riemannian manifold M. Therefore we have Since A. f is an and only if H(X. + EV.2. A. g(A. generated by X. Conversely. The tensorjeld Ax = Ls .sK Z) + gv. Z) + g( Y.Y..11) of linear frames over into itself. Hence A. VI-l-.pdint of O(M). f V[S. that is. Thus X is tangent to O(M) at kvery. Y)Z = V.. of Chapter I.1 .Z) .equivalence of (I-) and (3)~follows from Corollary 3.] = R(X. if and isometry of = 0 for any vector field X. x* E T”(O(M)). CA. An isometry of . where g is the metric tenswjieM:$:M. 41 .ni.tesimnl isometrics (3) Lxg (3) US6 = 0.))Z Killing vector field) if the local l-parameter group of local trans- for all Y and Z. . u .. Proof. Z)) k 0. that is. 1) G skew-symmetric (4) 7-h Let M be a manifold with a Riemanniarr metric g and the with respect to g everywhere on M.ys.A. transformation of M which leaves the metric g invariant.Z. only if ‘.V s of tyj%? . that each rjl is a local isometxy and hence X itself. V. Moreover.i / . An infinitesimal isometry is necessarily have an infinitesimai.‘ ( X ) = 0(X*) = e(JX*) = J ( u ) . and only . then qt are local isometries and hence Qt map O(M) which is a subbundle of the bundle L(. TRANSFORMATIONS 237 By Proposition 2.if and only if g(A.2. A. Z) + g(4. Y) = -[A. X is an infinitesimal affine transformation if This implies that J(u) = u. & . Lsg = 0 is equiva- of -41 is an isometry if and only if it maps each orthonormal frame lent to A.I? . is’an ir&nitesimal isometry.I-. Y) corresponding Riemannian connection l’. / r.+l.Z) = (V.Y]) PROPOSITION 3.Z E F(M).l. point 3.R(X.. M.‘ ( X ) .(g(Y.5 of Chapter IV that an isomctry of M $1 ‘bc. that is.1 of Chapter III and. AXZ) into L (M ) which preserves 0.‘A [X.’ (2) A jbre-preserving transformation F of O(M) which leaves the (ii) The. Since Ax is a derivation of the algebra of tensor at an arbitrary point x into an orthonormal frame at f(x). j(u) = F(u).. Proposition 2. A vector field X on M is called an injinitesimal isometry (or.f the induced transfoormatiq f of L(M) maps O(M) into bv Prooosition P 3.. point ofO(M).f in 01O(M) and hence each q1 maps O(M) into itself.

2.). then the..‘e T.10.d of a connected Riemannian manzfoold M is of dimension at most -@(n + 1)) We thus proved . Given an_y vector constant curvature.(M) of 3(M) at an arbitrary point x is is given by corn&act. ’ (l)‘For a Riemannian manifold M with’ a j&e E’ = U-‘(X’). the for any point u of O(M).5. Given any vertical vector X* c of M. Bi).gr@ 3(M) {s-compact.M’.Y. We where n = dim M. w h e r e B ( 5 ) a n d number of connected components.. Proof. this Let a be an element of SO(n)+ch that a.that K(p) depends only on the point x. R. al) (5) This follows fi-om Corollary 4. B’) + SZ(B.). 8. = Y*. If X. In fact. Y*.L.Rj(B.. vector’ fields corresponding to 6 and [‘. Y] is an infini- tesimal isometry of M. From the “is naturally isomorphic with the Lie (2) The Lie algebra oJf 3(M) definition of the curvature transformationgiven in $5 of Chapter algebra of all complete kfinitesimal kometries..10 of Chapter I...t Let X. there exists an element X E i(M) such L. = ((a *2!A(B. Hence. Bh. x(Q(B. To prove the first assertion. By the same proposition. is certainly an This means that K(p). another plane q in T. The Lie algebra i(M) of inJinitesima1 isometries This implies that Q(B. the linear mapping X -+ 8.. let .3. J THEOREM 3.. respectively.. Clearly. To prove that with the Lie algebra i(M) of all injnitesima[ isometries of M. TRANSFORMATIONS The’set of all infinitesimal isometries of M.. Bh.. X. we have TheoremA 3. Y’ be an orthoaormal basis of (5) Zf M is compact.) I’ . Proof. B(at’).:.. ) denotes the natural inner product in R”.9 of Chapter I and Proposition 2. QED.. Bh) = n(B.(O(M))..g = 0 that x7. B = B(t) a n d B ’ = B(5’).then the proof of Theorem.. complete infinitesimal isometry X generates a l-parameter sub- Hence the sectional curvature K(q) is given by group of 3(M). let ti be a point of O(M) such that ‘v(u) 4 x. B]. = O.) for every a e SO(n).-We now prove the second assertion.a?‘) th’> ‘7) (4) This follows from (2) and Theorem 2. [X. ’ THEOREM 3.4.2 of Chapter V. it is sufficient to show that.) = SZ(B. B&J) = a * R(B. aA’)a[‘.) 2 W(aE).plane. since the mapping X-+X. We have by Proposition 3.(O(A4)) with n(v) = x. the fact that K(p) is indepe.23 constant.. If dim i(M) = &z(n + l). X’ be an orthonormal basis of a. BL. B’) + h([X. 1 :I into T. \tTe shall see that. subgroup 3.. (1) ‘ks we indicated in. (If dim M 2 3..(M) and set 17 = u-i(Y) and 7’ = U-l(Y’). = X* if dim i(M) = bn(n +. the Lie algebra of g(k) is naturally isomorphic where ( . 92 of’chapter V) (3) The isotropy.(p) = K(q). By follows from Theorem-4.6 and.2. the proof of Theorem 1. then M is a space of prove that K(p) does not depend even on x. let Y be an element of i(M) such that Yz. = W(B. B’]) = 0. ’ (4) If M is com&Jete.$1).) = Q(Rd-~(4. 1). ..$ = g-and &’ = q’.. if X and Y are infinitesimal isometrieH Q(a.I 238 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 239 VI.p in T.“is locally element of T..a-‘.‘ maps’:(M) function ((222(B.of Chapter IV: (9) Every l-parameter subgroup of 3(hl) induces an inhnitesi- Q(B(v). the group 3(M) -of isometries of M is a B(t’) are the restrictions to O(a) of the standard horizontal Lie transformation group with respect to the compact-open topology rn M. By Proposition 3. Bk) . (3) This follows from Corolliry 4..4.‘(O(M)).(O(M)).2 of ChaptePIII. then r.f%J) ma1 %ometry X which is complete on’ n/r and..t. considered ‘as a function on M. sectional curvature K(j) (cf. We have again P(Q(& B’)) = 0.8 of Chapter I.(O(M)). it is sufficient to show that forms a Lie algebra. III we see that the. 3(M) is a closed subgroup of %(. . for. B:. conversely. constant on M. maps i(M) onto T. 0 L.p L.. We set E = u-l(x). B’)) = (L.fixed 5 and J’. B(r’). [X.. denoted by i(M) To prove that K..g . = ad (a)jQ(B. K(g) is independent of p. Since it is continuous and M is connected. B’))k“ 5)is’ constant in a’neighborhood of u.-. ” 0) = t Pwb?) . every. C+-?lary 4.g = i.(M) and of x follows also from Theorem 2. it must be shows that X = O.

where T: denotes the parallel displacement from x0 transformation 9$: Mi + Mi by to xt ‘along 7.. .) x * * e x Yt( Mk) isomorphically into LEMMA 1.. i Proof. Let T be any loop at x and set T’ = 9(~) ~(4 = tplo(xo). ~1 6 M. 9dxl). TRANSFORMATIONS 241 .4 of Chapter IV. xi. P’(~P) = Tpctj (O) . . Xl(t). If we denote by the same letter 7 and 71 rallel displacements along r and r1 respectively. Proof of Lemma 1.) x Y(M. .(t). . . z - It follows easily that p(T$?) is invariant elementwise and every i + l .(t).(M): 09 * . We need the following two lemmas. .. generated by l-parameter subgroups. pli: Mi . Then then (1) p is an a$‘ine transformation of M-if and only . xr. for some j. . . thus proving our . x(t)’ = (X.5. then We prove that. To coriiplete rhe proof of’ Theorem 3. . Lemma 1 is due to Nomizu [3]. and. in fact. tion Tp(z) = Zfzo 9( Ti)) which should coincide with the canoni- cal decomposition Tqcz. is a basis for Tcij Clearly. . ~~(4. . . Let pt be an arbitray transformation of Mi for ever_y i. This meansprecisely the statement T’O’ + ..trivial.--f Mi. .* * * 2 Xk(t))> 0 5 t 6 l. . . 0 rl i h k. Let T. Q?&. . Xi+l(t). x1. X. . . t 7’ is a loop at 9(x). y 7 p 0 7(X) = 7’ 0 9(X) . Let 9: be the transformation of M = hl. . Thus we obtain the canonical ddcomposi.~-~l(x~).assertion. show that.sothatx. x * . -V = of t (cf. * x M. be a curve from xi toy.-. . We then define a 6 > 0.6 Cue(M). there exist transformations 9GT)) = T$$. .LetT =x(t). $9(x)) depends only on xi = pi(x). dejined by first prove a result by Hano [l]. . such that (2) If ‘9 Q /IO(M). for each] = . d-4 = (~o(xo).) x * * ’ x WyM. . x 121. . Let 9t be a. we have g(9t(X). $X.tW Sinceg(90(X). * + T(“). This means that 9. ‘.~ i s k . then -. a$ne transform&ion of Mi. Hence$. it suffices to (1) If Q l A(M). . * * . 1 $ j 4 k.(9(x)) = p. .ByLemma 1.(9(y)). Hence. b e Y(~(x)).J) for x 2 (x0. 9&))~ for x = (x0.. 3”(M) m 3O(M. + T(i-1) + Ttiffb ‘+ . -The corresponclence. x M. In particular. we have then g(9t(Xj). Let T = x c 7 9.xk) e M. in 9( TF)). if Xi. 9i. 1 < i I k. the identity component Y(M) of 3(M) .“‘) c T$L. then q( Tz’) = T& f o r every i . $X) # 0 for ItI < 6 for some $. simply connected Riemannian manifoold M. then p( TL’)) = T$$. The proof+ .5. .). . their dimensions being the the curve from x toy in M defined by same. We hf. $ I and ItI < 6’ for so& positive number 6’ and hence 9JXj) E Tgizj for ItI < S’. VI. . * * * 9 Xi-. .1~a T. . . implies 9. ’ g(X.t( Tg’) = TgAz. the tangent vector i(t) to 7 at x(t) is in the distribution Theorem 5.( Tc)) = T:b’& for ItI < 8’ because of the linearity of QED. Xi. = ZfEQ Z& up to an ‘order by (4) of For each t.J90. The following result is due to Kobayashi [4]. for ItI < 8’ for 9*txi) = pi(9iX))’ some positive number 8’.(O) =xjandxj(l) =yj.(9(x(t))) is independent non-zero element of T$). 7j = xj(t). q-ii(t)) (1). since %O(M) is It is therefore important to study (U(M) when M is irreducible. 0. T HEOREM 3.* * x coincides with the identity component %0(M) of g(M). .) # 0 for 1 s j. 240 FOUNDATIONS OF DIFFERENTIAL GEOMETRY L E M M A 2 . 0 $ i s k. . let. in many instances.). which’ 9(xj k (~)~(x~).Yi+l.) x * . Lemma 2 for Theorem 7. Given in:’ Point. 1 I i s k . x . 2 maps ‘zl(Mo) x B(M.. ifpi: M -+ Mi ’ denotes the naturalprojection.) 2 WyM. andfor each i. .(M) = IZtEo Pi’ be the canonical decomposition: 11(M). . 0 5 t $ 1. If M = &I.2 of Chapter Hj. 9i(x.‘. for X. l-parameter subg&qof ?IIO(M) and let X be a lies also in the same distribution.9J -+ 9 defined in Lemma -7. is irreducible by the linear hoionomy group M. .* X’ Y(M. for every 9.‘0 5 t 2 l . 9t. is the de Rham decomposition of a complete. * * * > yk). This concludes the proof of the statement (2). . X) # 0. k. . 9( ?y.)..f every yi is an ‘aO(M) * WyM. * 23-1.

j =p*(g). Clearly. Y) Y. g*(X.5 and 3. X E a(B). an integer m. Y.. Let @~ = ‘MO x M... . Q?Y)(cJPY). Consider the Riemannian metric g * defined by g*(K Y) = ddXL v. if the distance of M. Let w be Prqofof Lemma 2..5.converges to some a(M. 7 .) + ..6. v(Y)) = c"g(X.242 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI. . cpnstant c associated with v is tesimal isometry of fi if and only if X is an infinitesimal isometry less than 1. From the proof (B) of Theorem 2.(Y)>. Let K* Theorem 3. the Lie algebra a(a) is isomorphic with 1. IfM is an irreducible Riemannian manifld. It is easy to see that the point element of a(MJ + a(M. there exists a positive constant c. X is an infini- necessary. whose restricted linear holonomy group ‘P(x) leaves no non-zero vectar at x curvature tensor field.2 implies that then aI(M) = 3(M) except when M is a I-dimensional Euclidean space.. Y) for all X.2. ‘By. Since p is’ an affine Proof. X. Let Y(x) be the linear holonomy group of r with reference point x. c. Proof. Assume that p is a nap-isometric homothetic the vector field on a induced by X. Since it is irreducible and leaves Letting m tend to infinity. complete Riemannian manifold any unit vectors Y.) + * * * + a(M. timK* 2 Ig(R(q+X.(M).* = cz .)Y. the two Thus we obtain Riemannian metrics g and g* (defined above) determine the same CONK* h IK(q)I for m 2 m.6..( LEMMA 2. X.g(X. g.. Take. are all infinitesimal isometries by Let U be a neighborhood of x* such that 0 is compact. E M and q any. Since both g * and g’ are parallel Riemannian manifold M. find a number of sufficient conditions for X to be an infinitesimal . Let 2. ’ Let X. TRANSFORMATIONS 243 T HEOREM 3 . at y E C: where R denotes the .*. is constant.every afine transformation cp of M is homothetic. then x0(M) = 3O(M).Y be an orthonormal basis for.. we see that the Riemannian connection defined by g* coincides with the one On the othea iand. Theorem 3. since M is complete. . p = A?. x) = c2”ff(q).“(x) and v”+‘(x) is less th an cm& It fo!lows that (e”(x) . + a(M. Proof of Lemma 1. J? is p-related to X. and Y. . QED.een position of the complete simply connected Riemannian manifold q. where p: fi --+ M is the natural projection. gz Let X be an infinitesimal affine transformation on a complete (cf. Theorem 1 of Appendix 5). Y) = cz . x * * *x M. Using Theorems 3. The linear holonomy group of M is naturally isomorphic . say x*. an arbitrary pbint x of M. Since X. YE TJM) and x E M. If M is a complete. we shall tensor fields with respect to I’.l. RWX ‘p”Y) (q”Y) = y”(R(K Y) Y). Let (X. be a positive number such that Ig(R( Y..} is a Cauchy sequence and hence. be the de Rham decom- betw?en x aud g(x) is less than 6.q. + . that is. let hi be the universal local& Euclidean. then every homothetic transformation v of M is an covering manifold with the naturally induced Riemannian metric isometry.pldne in T. . YJI < K* for C OROLLAR Y 3 . Y) for all X. A transformation p of a Riemannian manifold is said Hence we have to be homothetic if there’is a positive constant c such that g(v(X). (3) of Proposition 1. transformation. . Riemannian connection.) be the point. = +g(R(X. If M is a connected. such that v”(z) E U for every m 1 m. t.hen the distance betw.. Considering the inverse transformation if an infinitesimal affine transformation of a. such that locally Euclidean. X. . X is an infinitesimal isometry if and only if X0 is. This shows that M is both g ar# g* invariant. Assuming that M is connected. If A4 is a complete Riemannian mant$old which is not isome&. . then . we have K(q) = 0. This means that every homothetic transformation of approaches 0 as m tends to infinity. we have LEMMA 1. q~X)j for m 2 m.(M). Since cp is an affine transformation. Then x is transformation of M. say I’. In other wordi. Since the lengths of the vectors ~F’X and rp”Y are equal to cm.).) corresponding to x* is left fixed by q. there exists a Riemannian manifold M is an affine transformation of M.j’+ a(M. we may assume that the.2 of Chapter III.. . fixed. irreducible Riemannian man$old. the distance between x* = v”(x*) and V(Z) defined by g. . YE T. > . 6 .

13 = 1. we have Applying the formula (L.” 0 V.YJ. -For an infinitesimal affine transformation X of M.= 0 fbr 9. for small values . we show that 5-r -5 I (‘::I . . By (:orollary 3.rlZ (cf.). we have (L. E T.0.(M). On a compact manifold M.Y.. A of T. If the length of X formations of T. l-parameter group of local transformations leaving x fised and the introduced in $2. consider the case where X.). C. If X is an ininitesimal a&e transformation of a For each t. is compatible’with parallel x0 = c~=l~z(a/a~). .a ZQ& l-parameter group of linear trun.8..(M). Holonomy ancl injinitesiaml isometrics = ~itn~(C. every vector field is of -KsLL = (VSY).Ly). We may assume M to be connected. with the restricted linear holonomy group yO(x) of A4 (cf.) . . = exp tA. . by T. I QED. the length of X0 is also bounded on MO.11 which was originally proved by a completely the curve 7 = x1 is given by 9 = t. = c). .(M) such that C. consider the case where X. a% =. Since qt” maps the portion of T from x0 t@xl into the xl. This means that MJ reduces to a point and neighborhood 0f. In Lhis case.+ Proof. -+ 4. . then X0 is’an infinitesimal isomctry of nfO.) = fiy f (Th o v. values oft.1 of Chapter IV). we give a Second. ___ LIil ax'1 ax. .2) to Y = a/axiJ and 2 = a/&. x2 = ** * = x” = 0 for small different method. Then x has a coordinate Corollary 3. then X T.(M) :. = L.8.' This ‘proves the first assertion: Thus there is a linear endomor- This means that X0 is of the form phism. 0 (P)J* of * comfjlete Riemannian manifold and . ..V. On . jc” such that result of Yano 1. . is. xp -t y) (a/ax=).x. is a2 injnitesimal isome@.‘.. Evidently.r. Cl+s . TRANSFO~IATIONS 245 244 FOUNDATIONS OF DIFFERENTIAL GEOMETRY r Let x be an arbitrary point of M and let vt be a local l-pa. the parallel displacement along the curve T from x8 to xt.Y.. ’ PROPOSITION 4. T I is a local geometric interpretation’ of the tensor ‘field A.Y. . Example rameter group of affine transformations generated by X in a 2. say. We denote hence X0 = 0 in the above notations.9.1. We may therefore extend Y.Y. VI. we consider a linear transformation C. = 0.of t.f the length of X is bounded. = T:. To prove this. 0 LAyO)Z = V. obtained by Hano [l].\I). implies the following neighborhood with local coordinate system x1. is bounded on M. = (vsy)z! ‘I bounded length. = !iy f (. Let T be the orbit xf = v. We have W( Al) =: 3y . and C. we see that a2c-z z 0. First. 0 C. and since’?. ) x’ be the Euclidean coordinate system in MO and set phOn Of TfrOIn Xt to'&+.a compact ‘Riemannian marifold M’.Y).‘7. Let Proof. Thus if X0 is of bounded length. . . = exp ( -t(A. Let M be a differentiable manifold with a linear connection I’. . COROLLARY 3.(x) of x. . It is easy to see that length of X0 is bounded on MO if and only if for Y. . a. * . ever infinitesimal affrne v transformation X is an infinitesimal isometiy. . r. Proposition 2. displacement.‘.Y.= C. COROLLARY 3. to a vector field Y-on JI in such a way that TV = Y$. Proof. QED. The second assertion says that A = -(Ax)+. QED.. . # 0.

tained in iV(Y(x)) as well as in N(YO(x)). 3 . = 0.ui 0 pt. By Green’s We recall that N(g(x)) is by definition the set of linear endo.). C.(S.2) and. B) = j -trace (AB) .of Chapter II and can be derived from it: respect this inner product.u at x. Let T be an arbitrary curve from a . then. Propo- identity component of N(Y(x).an injnitesimal a&e transformation of M. E(x) and.) = V. / . be as before.) belongs to --g(x) also. 9.). If X is an infinitesimal isometry of M.pojnt x & the parallel displacements along .(M).(S.u 0 C. Proposition E(T) arc.2. is con. for each t. 0 . concluding the proof of the lemma. we see that V.ii. For the given infinitesimal isometry X PROPOSITION 4. The tensor jeld B. TRANSFORMATIONS 241 246 F&JNDATIONS OF DIFFERENTIAL GEOMETRY The following theorem is due to Kostant [ 11. Y) (cf.(B.2. This shows that C.Y).(BJ is. (A. (A. of Lemma. Then C.’ is an element of Y(x). we set LEMMA. B] E g(x) for every B F g(x).(M). on M. Let pt and 7.) = 0. Proof.A.u and Proof. . +B. If X is an infinitesimal isometry of a compact Riemannian manifold M.) = lim 1 (C. We have T HEO REM 4. E B(x).maps c.S. is an orthogonal transformation of T. Then to another pointy. B.(B. We denote by .YJ. If X is an infinitesimal isometry of a Riemannian manifold M. = -(&A algebra g(x) of the linear holvnomy grou$ Y(x). we introduce a. ~. (Note that N(Y(x)) c N(YO(x)) since Ye(x) is the On the other hand. T&W We shall show that B. is skew-symmetric (cf. ) ..l~~~=To’~~~~P)to~. For any loop .(M) such that [ setting This completes the proof of.ply. for each = -trace (B. E(x) of skew-symmetric linear endomorphisms of T.field. we have (assuming that M is orientable for the moment) . Appendix 6).) = -trace’(BXB.(B.) is in e(x) whereas. Y. is in the normalizer N (g(x)) of g(x)’ in the Lie algebra = -trace (B. If X is.X. theorem (cf. parallel displacement TI. The paralle! displacement. We have then T HEOREM 4.).- p is in YO(x). at each point x E M. we set *i’ of the Linear holonomy group Y(x) and the restricted linear holonomy group A.(B. T gives an ‘isomor- cpt 0 p = .) x c M. . 1) is parallel. In the Lie algebra E(x) of skew-symmetric endomor- = .) 2 0. div Y du = 0 (dv: the volume element). =s. We have phism of E(x) onto . Proposition 3. Hence the Lie algebra g(x) of Y( x ) in the Lie algebra of endomorphisms of V.(A. Theorem..(y’.X)) = trace (V --+ B.WXI’).4.) + then.(M). c g(x). Let M be a Riemannian manifld and g(x) the Lie 1 Since div Z!is equal.. =~q&u)~o tha t. to the trace of the linear mapping V-+ V. . sw then A.Y at each point x. QED. for any vector. Let N(Y(x)) ‘and N (Y”(x)) *be the normalizers of M.the second assertion. where S. respectively.4(y) . This means that. the formula V. we .(M).in B(x) at each point x.) sition 2. (A. and the B(x)-component of the equality V. positive definite inner product ( .) belongs ta g(x) at each x E M (cf. Y”(x) in the group of linear transformations of T.(M). for each x E M.).u at&p:. QED. Proof.B.LJ~ is a loop at xt = pt(x). of type ( 1. = 0.which maps -g(x) onto g(y). belongs to the Lie ~(AA!Yz! = (VxY)z . VI. Remark. E M. f (Y. We set Y = B.-. Thus (V.V. Since the inner products. x E M. (A. in.1”7~=7~O~~o70 I’ B(x) onto B(p).trace (B.1O~~c~l=T~oQ)tO~O~.I’.(A& = R(X. presefved by T.5) div Y = trace (V + V.1 of Chapter III). reduces to the identity transformation of T. morphisms A of T.5.6) implies that V. belongs to the~normalizor N(g(x)) of V. It is in Y!(x) if V. t-o t phisms of T.(A.X)) algebra of Y(x).1 is indeed a special’ case of Proposition Let B(x) be the orthogonal complement of g(x) in E(x) with 11. By comparing the g(x)-component C OROLLARY 4 .) .have (Lemma and Proposition 2.u. then.

X) Y.W).. tensor algebra over T=(M).~.Y.Y) C. S(X$ Y)’ = C.(M). In the proof of . .4. Y .Y).‘. X) = -g(A.@)~&v (AXX) Z 0 at any pokzt where f attains a relative m&num.(x) =-x. Since PROPOSITION_5.-) ( 1) Evev parallel tensor field K on M is invariant ty the identity component So(M) of the group of isometrics of M.... ’ “F~RO. i n v a r i a n t ..G. Y) = trace of the map V + R(V. Hence g. -. X) = g(V. for every vector field Y on M. \ I On the other (3) Vnf: = g( V.Ye T.=. the l-parameter group C.(.)I trace (BxB.trace (~~s.X.(A.X) = -S(X. If A[ is a Riemannian manifold and if VI. . _. for every t.6. we obtain Vf = Q(V. . is extendedi& a l-parameter group of automorphisms of the i = . L Ve(A. Y) .‘/ belongs to Y(x). .‘A. . If M ’1s not orientable.’ the linear isotropy group of 30(M) is contained in div (A. $1 of Chapter V) implies S(X. we saw that C.=O.v = 0 . is an orthonormal basis of T. ‘ Proof.Y) is’the trace of V --f V.. Then B.. which implies trace B.: .vector field V on M. In particular. S(Y. TRANSFORMATIONS 249 2.(1) I’j = g( V.‘(A. V.ZI). -5 .trace (risA. S is symmetric. it leaves P invariant. we have Let M be a manifold with a linear connection r. we prove a result of H.. since vt.(K. = Kz.8 FOUNDATIONS OF DIFFERENTIAL GEOMETRY where X.~Rienuinnian manifold fl. By Proposition 2.6.) = $K. X) Y and that div (A. the linear holonomy group Y(x) . $2 of Chapter V). X) an M. = 0.) dv = 0.1 and Theorem 4. that is. \” ‘< ‘!~Q&~v (AiX) S 0 at any point where f attains a relative maximum. for some’ to.X) fm evqy tungent vector V. there exists a l-param- mansfold M. I’) = By Proposition 4. A. Then eter subgroup vr such that 9 = ql. Rick te&sot and infinitesimal isometrics P&f. curvature tensor (cf.FOSITI~N 5. V.Y) = +(X. I’ E T. Property (d) of the Riemannian B. Proof. If M is a compact Riemannian manrfold.k. . For an injinitesimat isometry X of a Riemanr&n 3O(M) is a compact connected I$e group.(JI). The Ricci Z(g(X u))i~ v&x. . T: is also in -k(x). X. X ) . Vi) = Cy=.B. where R in the last equation denotes the Riemannian curvature fold orientable covering space M* of M.5.Y) . Y) = QED. Y and 2 on M. Since P(M) is connected.:. X)Y = -(V&I-y))Y = -+(AsY) + A-y(V. If X is an injnitesimal afine transformation of a Wang [I]. consider tkjfrnction f = +g(X. then I . Vi. (1) Let X be an arbitrary infinitesimal isometry of M. (obtained from d. Thus v.+ls).5. X) Y of T. When R(V. V&. (I). then THEOREM 4. then div (A.) for any . Y)) = g(V.5. we have R( V.Y) tensor field S is the covariant tensor field of degree 2 defined as for arbitrary vector fields X. (2) At each point x..@ = 7: 0 CtO v. X) . QED.(M) is contained in Y(x).. that hand..X I’) + SC& f. As an application of Theorem 4.V. X). where q1 is the l-parameter group of &r proposition follckvs from the fact that S(X. X) A -R(X. we lift X to an infinitesimal isometry X* of the two.X)) for every vectorJicld V su&. g(R( Vj. ‘He&e 6 of linear transformations of T. = 0 and hence B.Since ~g is parallel.(. VI. it leaves K I’ + R( V. = 0 implies tensor (cf.X) ’ f(X. Y) is’the trace of isometries generated by X. R(V.1. Applying this formula follows : to the case where :X = Y and 2 = V.) is in Y(x). Q E D .element i$f 30(M) . C. (2) Let 9 %e any. i .such that v(x) = x.

it is.Proposition . following.. ’ C O R O L L A R Y 5 .of an .\l) reduces to the identity. the isotrop! definite. X) = 0 and trace (A. (JJ3l.5.)] dv = 0.) so that Vi = ii (0).< R&cl.(M).l).= 0 everywhere ‘on .v.(t) for small values of Proposition 5.11 of Chapter .\I is a (resp. we obtain --S(X. in a neighborhood of u.r. X) + trace (-4. then (Vtf)a 2:’ 0 (rep. we have div (1 !xX) I 0 at x. maximum) at x. we prove the every vector field V on AI.every point -21” reduces to the ibentit) maximum at x.p. (cf. .5 and the skew-symmetry of A.6.tensor.we have X .X) + g(Vip vJ7i(AxX)) i.\li CIJ irometrirs o/^ .1.= --* Tr. we obtain the following result of Ii&- T HEOREM 5 .. A.(r) be the geodesic with the We may assume that M is orientable. negative definite. we x. . X) 22 0. By Theorem 5.) 5 O. Since div (4.u = P.. From initial condition (x.250 FOUNDATIONS OF DIFFERENTIAL GEOMETRY . V. at Proof. X vanishes in a neighborhood X.(>. . . But 8(X. COROL. Proof.1f. Thk Y(M)... LA M b e a cokzected Riekannian manifold:whese n e r o w i c z [l). we must.2).V = 0.x==0 ‘- b by assumption. we prove . Theorem 3. ‘111 is a of C(. VI. COROLLARY then X vanishes identically on M.. X =-0 at x.Zd) such that T(U) = x.X) is thetrace of-the linear mapping V + V. If the length.O = ASP’ .I.A. manifold with Now. . Proof. is’ a compact Thus we have S(X. X) = 0 and A.3.1 can be derived from Proposition 5.of M. . then isjnite. we obtain t..t:. As we have seen in Example 4. then the natural lift x’ of Xvanishes Euclidean torus.. Rick tensorjield S is negative de&ite evuywhere on M. /hen M is a Euclidean”?oru.2.ARS 5. man$old~.(3):follo%s from the fact that.X).3. .1 by means of Green’s theorem in the following way. We extend each Yi to a vector field which coincides with $(t) at x. ’ Ij .since As is skew7symmetric. By Proposition 2.l..K for As an application of these two propositions. QED. Since 3O(-a1) acts effectively on M.5.. Corollary 5. we have .5. then every injnitesimat? i. let 7i = x. Since the length of X attains a relative subgroup of 3Q(Jt):.trace8(A+fX) 5 0. 5 .a compact Riemdnn!nn. As another application of Proposition 5. To prove (3)) let Vi. QEIji parbllel vectorj?eld. X vanishes identically on M.1 and Green’s theorem. TRANSFORMATIONS 251 by virtue of Proposition 2. . VVt(AXX))* ‘S&e S(X. If V is a vector field such that negative deJinite Ricci tensorjield.X) . JO(. for any infinitesimal. -+ 1 1. . rf .(i”.If u is any point element.A.finit. M has &ro.ometr_v of .5. iffattains a relative minimum vanishing Ricci tensar~jield.. From Corollary 5. Y ] = V. isometries X. be an orthonormal basis for T.II. bve I S(X.4.) = 0 everywhere on .at.A. proving (2).~. For each i.V. coniected compact homogeneous Ri~rn~nnlqn infnitesimdl isometry X attains a relative maximum at some point . As we have seen in the proof of Theorem 3. . Remark. Since S is negative abelian group. I f M is? . Sinc. result of Bochner [l]. V. (vff)5. X) .\I v. By l’heorem 5. otherwise. 3 .) 5 O. we have only to consider the orientable twofold covering space of .Y . (lE. This proves (1). . 5.\I.Y.e.and trace (A.1 of Chapter III and Corollary 5. Proof.g 0 [X.X). = 0 at x. Since J(M) is compact (cf.A. By Proposition have 5. This proof we have gives also div (4. then the group 3( .:‘S’.] s’O)..ivi’i. d2f(Xi(t))/dt2 = vff= g(v. = ET==. . QED. and trace ’ (A. By. Assume the length of X attains a relative maximu?. Then we have r [S(X.\I is n conrpct Riemnnnian manifold with Proposition 3.

X = 0 along the orbit Qua. X)*. . such that Z. Let M be a ~connected.+1. . . . t&en they coincide on M. Let S and S’ be anabtio distributions on analytic manifidtls M dnd M’. . . . From tht well known properties of power series. U.V at x.J{(x> areanaly’tic functions of x1. Since M is connected. y’ = f i(X1. Let I '1 .) = Xqltz) by (1) of Proposi. m. x9). Since X&E N and 1~) + x0. . By (1) of Proposition 5.If M is cqnnected. we of points x c M suchiihat jr(x) A g(x) and ‘that the power series have g( V. . manifold with a cdmplete ‘analytic linear connection.X) = 0 for every vector . . Let f be an analytic mapping of M into M’ Jlcch that 6. . . 2. to each point u E L(M) that (*) is satisfied in a neighborhood of x. i = l . Iff andg coincide on a non-empty words. say. Since qt(X. Let U be a con- The main object of this section .is to prove the following f (x0) and hence in a neighborhood v’ off (‘x0). Let N be the set the function f = $g(X. . Si@larli for ‘the mapping g. consisting of analytic vector fields defined on U. .. . of some xb such @at LEMMA 1. . .oof is preceded by several lemmas. Then N is clearly a closed sub- x. that is. . .A vector fields.7. . Zh+1.. The mapping f can be expressed by linear connections and f ah analjtii happing of M into A&‘. . then the orbit qua is a geodesic. Let M and M” ‘be a%a&ic manifolds with analytic Y. . X)*. ***r-z. If the . Hence A. . j generated by an injnitesimal isometry X of a Riemannian manifold M. there exists a neighborhood. .X = 0 at expansions off and g at x coincide. X. Let M’ be an analytic basis X. This shows that follows that F is open .1. . Let N be the set of all poipts x l M &C% ! ? which assigns the h&ontal subspace Q. .y” be an fields.. tion 1. LEMMA: 2. Let f and g be analytic mappings of a connected analytic U. it suffices to local basis for the distribution Q consisting of analytic vector show that Nis closed.. be an analytic local coordinate system in a neighborhood of x. in8 M.any pointof M and let x1. x”. k. Since M is connected. . X) also. .: !_ follows that f f = 0 on Ufor the&same i and j. It open’subset of M. Let pt be the l-parameter group of isometrics I a set of analytic functions h.. x” such that f(U) so V’ and that?2 has a local manifold with an analytic linear connection. . . N = M. These functions can ‘be’ &iax&ed at x into convergent power Proof. nected neighborhood of x0 with an analytic local coordinate THEOREM 6. . be-a local basis for the distribution S’ in K From bundle T(M) (for the notion of horizontal sqbspaces in an alay.. Extension of local isomorphisms (*I Let M be a real analytic manifold with an analytic linear or eumy point x of un open subset of M. 2.. we have mapping fv of a connected open subset U of M into M’ can be uniquely fKL = qLf3x) * 2. T(N) which defines the notion of parallel displacement in the ’ Let Z. . . This proves that (*) Proof of Lemma 1. . . LEMMA 3. Then N is clearly a IS analytic iti the sense that each point U has a neighvbrhood and a non-empty open subset of M. . In . . Let x be. Th!% same is true for the distribution on the tangent bundle analytic local coordinate system in a neighborhood V off (x0). . . V. .. QED. . manifold and the connection form o is analytic. . . Z. are linearly indeptn&nt . A. Z. * * * . Let$. the orbit 4pl(x) is a-geodesic.. see $7 of Chapter II).2. choose m . The bundle L(M) of linear frames is an analytic sutisjed at every point x of M. simply connected analytic system x1. . U1 for 1 %-:i d k and h + 1 $ j I m. . If x is a criticai point ofg(X.2. . Then every afine Since f is analytic. we have V. then (*) is connection l’. . x”. . set of M. i = 1. al@*.X = 0 at x. wheie. 8. . extended to an afine mapping f of M into 121’.y” be an analytic local coordinate system in a neighbor. c W: and that (*) 4s b&isfleQ: at every point x of U1.m associated fibre bundle. x” is satisfied at every point x of U. 252 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 253 PROPOSITION 5.&er marlifold M into an anajytic manifold M’. f{(x) = 0 on.-* hood of the point f(x).. it is a critical point of series of x1. Let xI E N and xI + x0. it . . The p.. The distribution Proof of Lemma 2. : If x is a critical point of the length function g(X. .

mapping of the We extend’ftl to an. given mapping fK can be extended to an Is .a neighborhood x(tJ into :I!‘. By Lktinia 3.V’: . We have thus proved the exist’ence of Proof of Lemma 4.h.exp X’ is well defined and \YC set . and fs coincides with ft in a neighbor. .\I and &I. Thus N is open.. Since the expc$.kt ‘M a. have alrea&. we have the affke . Set X’ = f. The Ricmannian Oonnectlkm of an anali.) and from the the ptint x.Qs~morphism between connected open subsets of . ti.. thatJ)(S) = a(X) for X e T&(M).:..(M) at . let x(t). be a cur\.3. then f is an afine mapping of Ad into M’.\I’.~ >-0 such that an analytic uniquely extended to an @ne isomorphism between . L of Chapter taining k1/.piinty E Y determinq.1J and . frorri x tcidy.ff isan &Fine mapping of:.&a~ mapping gi\-cs a @lko~orphism L’ ma ps the horizontal subspkce at each p@ni of n-l(u) into a of an open neighbqrhoo$m. then thc’nn5lytic Continuation ft of such t. By assumption. Then N is’ clearly a non?empty To complete the proof ofTYtkorem 6. . X E V* c TJ(. N iinpiies that a normal co. 11.-z exp X’. Let 2 and IIF be di$erentiable mariifoldj -with linear e g p o n e al” mbppingk Sinte ‘the exeonential m a p p i n g s a r e connect&w and let f and g be a&e mappings~ of 44 into M’.e mappings (Proposition 1 . :d NN(f $’ .I). /by definition.) as in Theorem 8. pf from the factorization lemma (cf. We &n easily define the continuatio$. 1Let . (2) For each t. Let F be the analytic . the satisfying the follo+g conditions: curte x(t) ‘is homotopic to zero and our assertjpn follows readily (is) For each t. r.’ Of& for 0 5 t S t. tiith fLJ in a neighborhood x. let x(t).il) such 7’( . s u c h t h a t t. we have N = M. be connected and sinzp[y -connected It follows easily from Lemma 4 that an anaIytic continuation of wal_vtic maniflrls with complete anabtic linear conn+ions.m evey f I-.(..(X) so that X’is a vector atJ. neighborhood of_y int@ M’: We Show. I.s<&)). Let Ti b e .Appendix 7.. say 8. then ever_y .if x[tj is‘.~i~niciue. i COROLLARY 6~2.‘it is‘suficicnt to otierve assumptions..\f’ is rompfete... nkighborhbod li. .f is an affine rkappirig of M’iAto M’.t\ -e 6.. it aj%e.. 2 is an affine m.’ can be exists.hction off to an open subset U of Al is an afine mapping and ifA4 is co~ltlPc-~ad. “’ ’ I’int6 .. For each point? $f M. be a curve in M that.which c o n t a i n s W. (3)h =fu..iS alsd’anafytic. FOUNDATIONS OF DIFFERENTIAL GEOMETRY \‘I.EXI& 4. each horizontal subspace in T(M’y (herb.t’pping of g{.I.. Let W be a eontrex.some point x c M and . Z. then x(s) Q U.) ($14) onto. Applying &m&a 2 to the rhapping &‘.” j$ along x(t) g’~2s ’ risk to the affinC maPpingfl which coincides i s . TRANSFORM. If f (X) = .1.yj-j. 1. exists for 0 5 1 5.. along the curve . we see that thaty = exp X.j .ng the whole curve x(t).. x E. ThG &&i$si&ti’Q oflf. 0 3 t .1. b) fherr J and g Eoincide on M.i:~n mctric.$I’.in Theorem 6. th$t g(v) ‘is inclependent of W e %re liow in position to prove Theorem 6. S&e there exists an analytic cqntinuationf. An analytic continuation off. Let N be the set of all points x E M such ?-“’ an &&tic coatinuation ft alo. . ?‘HE~REV an afl&e mapping’<? of a connected. of 1’ into Ji’ as tan~cnt bundle T(hf) into T(M’) induced by f. 0 ‘2 t g 1. ‘0 5 t :G 1 . a f a m i l y o f afTine m a p p i n g s ft.tic Riernan- every point of W h a s a normal coordinate neighborhood con. let x bear5 arbitrarily fixed closed subset of M Since f and g commute with the exponential point of U.Il! /w rrnrr!rtic Riema~lnian maniflds..) e*W.Y) for everll X E x(O) .K* Qf the origin in T.Tp’.. Since M is simpli connecied. Lemma 4. 1Ve have the corresponding results for analytic Ricmannian hood of the point x(t.. uniqueness of .is an$ytic. t h u s defi‘ned commutes with tllc I. I ‘.f M is connected.a closed cukvk. Take t. Since M is ‘along the ciir+e k(t)’ aria <gives rik.x(t) is unique if it exists: We now show tlkat.\I! onto 1M). continuationf. Theuniqueness off follows from hood <if x(s) .il~r’~:. 2 j-1. be the supremum of t. Since M’ is complete. QED. using this affine mapping g. there exists a positive number 6 such that if ! It f o l l o w s that the.. a n d that. .et f. 72 dendtes the projection of -.1..7 of Chapter III such that manifolds. hwf of Lemma 3.4 . alodg the curve:%(t) .mappingjti’ of.\TIONS 2 5 5 :analytic continuation we. element.. Under the the choice of a cdryc fi%m x t@:F$r~this.\I’.(t) into 121’.. The afl%ie ni%@bing fi.\I and . .. this ti)llo).C lY. can be &alyti&ll~ continued ordinate neighborhood of x is in N.k$lytic. -Th.proved. a normal coordinate neighborhood of x(tJ . If ‘\f is Connected and Sbtt//!l.f. af%jne mappingJ:of 21% into .x[r. 0 5 4 I 1. beyond 1. I.nei&bor. analytic mapping.\~s frqm Corollary 2. < t.

. we have isometry between M and M’. (resp. coordinates (al. xn. i. 7. ‘ii I \ Riemannian metrics g and g’. QED. . The&em 6. Aj(ta)aj = ai.i”:: = a%$ = Zck. Let ut = a(xJ. ” Since an isometric immersion maps every geodesic Proof. . well as from Theorem 6. then .m. Proof of Lemma 1.j = 1.$. . . . x” (resp. . n.. by x‘=ta*. . (1) p = a*Oi = . In the present case. For a fixed i = (a<). ’ Y’.L(M:)) be the cross Proof of Lemma 3’. . Lemma 2 was necessary only to derive system x1.en f as an ~rrometra~ . . .. . IV): Hence the following corollary-folloislr from Corollary 6. Let M and M’ be analytic manifolds with analytic . rcspectiuely. y.the horizontal lifi of xt starting from uo. f: M + M’ is an affine mapping (cf. O S td 1. through without any other change.v af ~?4 w y Let U (resp. ’ . Let M and ML be manifolds with linear connections.-. . If the resti’ction ‘off to an open subset . .2 as where ta stands for (tal. OSSl. t d 1.1. Remark. .L(iiQ (re. . Equivalence problem i.. then every isometric immersion i = l.1. a/ax”). . n. . . Let M and M’ be connected and simply to prove the following theorem. i = k. . be theimw . . th. Wd define a isometric immersion fv of a connected open subset U of M into M’ can be cross section C: U + L(M) as follows.& M be a manifold with a linear connection.. .. i = 1. . LEMMA 3’. . . .1. the proof of Lemma 4 is still valid.4.n. We indicate only the necessary changes. (1) f maps the frame a(x) into the frami&( f(x)) fbr eachpoiit x c U. . x. .C. . Proposition 2. ‘of M into M’.. Lemma 1 can be used 0 S. . y”): A da@omorphism f of on U. . Th&-eni 6:l. . TRANSFORMATIONS 257 256 FOUNDATIONS OF DIFFERENTIAL GEOMETRY - frame at the origin x0 given by (a/ax’. consider the geodesic C OROLLARY 6. i. we replace “affine mappings” by ‘. .x” (resp. . coincide section adapted to x1. We set ’ ” mappings. and let U be a nFig% borhood of x0 given by Ix”\ < 6. immersion of M into M’. . VI. The proof is quite similar to that of Theorem 6. V) be a normal coo&tate nt&hborhood of a point x0 c M isom&c immersion and if M . * * * n. yl .j = . . In Lemma 4. If x is a point of U with uni. . . . . ostrr. . . If’ dim M & dim. a”) + (0 0) with Ia’] < 6. Lemma 3. . .**a. parallel displacement of uO alo@ the geodesic given by xi = ta’. . . a?: V .quely extended to an isometric immersion f of M into M’.Bjk dxk. n. . .. . we prove the following Lemma 3’ directly.6 of Chapter (4) x:k&(ta)a” = 0. Let 19 = (U) and cu = (co. tan). i= I .is qmnected.y”) andlet a: U . ‘Since t&e. gw. l : LEMMA 1. Since ‘an isometric immersion f: M + M’ is not we have necessarily an affine mapping. Aj d#. Then every isome~ betwezn which-is. The first objective of this section: . .. 6 M’) with a* normal L. and letf be an analytic matping T HEOREM 7.a. . a”).” Then the. * .. . proof goes (2) .3.oordinate system x1.’ . n. .. We call (T the crbss se&on adapted to the normal coordinate without any change. Let x1.M’. . Since the frames connected open ‘subsets of M and M’ can be uniquely ex& to M ut are paraikl dong xt. . Let u. x” be a normal coordinate system at a point x. the argument similar to the onq us&d in the prNfc$Lemma Uonto V is an a#ne isomor~his+n ifit satisjies thefol&ow&g two conditions: 1 shows that they coincide on the whole of M..c(x> is the frame obtained by the Proof. . .) be the canonical f&m and NJ into a geodesic and hence commutes with: the exponential the connedtion form on L(M) iespectively. For any (a’. complete analytic Riemannian manifolds. Compare g and f * (g’) . ‘In the rest of the proof of Theorem 6.T(isometric (2) f priserves the torsidh and ncrvatur&~~tensorjie& I ’ immersions.3 does not follow from (sj IZ. we replace “a&e z mapping” by “isometric immersion.

differential equations (cf. p*@. .not . a”). . (ll).i 9. we (10) ? = Sj $(tn) dui + ai dt.?REM. .. . . R at& 9 (resj: p.&a) d&.. . R and CS.Eemma 2. .e * . ) a’“) = (&I’.(ta’l. (17)~ and the secmd structure equation.!.. ordina?>) linear dtyerential equations: (8) Q = ((t. t~...the case afabreal analytic&near connection. . . _.“the terms . . .. = -Ek [E. .2.‘ curvatke tensor fields and their successive FovariaRt derivatives From Lemma 1. :..* ?n-i> sii0lW . . . %.5Rj. I’KANSIWKMA’~‘IUIYS LJJ 258 FOUNDATIONS OF DIFFERENTIAL GEOMETRY From (lo).@ M and M’ be apalytic manifolds with analytic From (5) and (6). a’... it w&a (9) A. normal coordi&ty$&eti in $ nelghbo&d ‘V ofy<!su& that . .Sj 1 dajLA dt +.L ~. . .(t) -are utiiquely determined by the torsion tensop. we obtain ‘&ear connections. If a linlar isomorphism F: TCO[&l) 7 T. . Appendix. 1 q&). . IJet @. . 13’1 C $$e:a I.Q Oj A .v”.. defined by :-I and the curvature tensor fields R and also by the cross sectim.I prog L&.(t) and i&(t).‘t~(ta)‘(tA.. = -~j [C. tB~(ta)U’ + ~~ ~~(t~)(tA~(t~))~‘l da’ A dt + .!. a”) .t(.Simi@ly..Zf yi such that f (xi) = ys’ p& ‘thcit the (1-k) dcL == --Xj 1 (tAj(tej) . i .&&. n>. q = p*Qj. < 6 for i = 1. cl$sr@necf by T. the torsion and .. _’ i. On the other han&3he fun&ons. we obtain . iet T. ~ ) LEMMA & For an arbitraribfixed (a’. for Frotn ( 10) and [ 11). ikil ‘34.fTk ( Fjk = O*qk). I&(t) and defind .) q = a*q = c.(g) = 0 . . $7 of Chapter III) (5) gi = g*@i = xj. (7) &f. &.. L. respectively.QE”- 81 = p*fj( q = p*(.!* . .. More precisely.O into thlk tensors (V’m’p)yO and (v’mR’)y. . . a”))... . (12) and (13). ” _’ . 7.(tn))a”l:da” h dt + ’ .j(t)d + &. . &(t) = tBj. ‘: .J'HIZ.t &“..I. -... there if .(t)/& = l$+ Zz 8. have ’ i: .qte.. This proves.U. we set (17) -xk 3 A q + q! Aj(t) = tdj(ta).j A 51 + G* it.: -m =?JJ.I is uniquely. Let jJ. VI. . . . coordinate where the dots denqte tlrc te-r+Fls n@jnvolvFg dt... (9) is obvious from the definition of isomorphism f of a neighborhood 0 of x0 or$“u Si$$b@$i . . . * “a i.&(t) = Rj&z).. : ( 1 I-) .riormal . . ..(ta).. . .(t) tire uniquely deter&ed Proof of Lemma 2..(ta).(i)A~(t)& Finally.fn a$. A. . ‘tie obtain \Vc set [cf. = !q. then_. into 1. and ( VmR). (16) and the first structure equation.f. y”.. . l?. .).d” A 6” (& = cJ*l&). ...: . (16) -xi .-(for p(t. l&(O) = 0. ai.j. Now (7) Then the functions A:(t) and &(t) satisfy the following system of follows from (14).: . M’).Fram s( 1) we’ see th& the connec‘tion We set form .(M! maps the% t&hors (Vm T). 5) _-.tuy. R’ and ‘9’) ‘be ‘the to’$& the curvature and the covariant d&t&t&ion ofM (r&p. . . From Lemma 2 and from the uniqueness theorem oRsystems with the initial conditions: of prdinary linear . . 7 Fpp h&i -e I. each fixed (&. where the dots dFD.Gelds T p be the mapping of.: _ qm.. we obtain ’ at a point determine the connection uniquely. dij%e&i[ off&i ‘$3 ‘.k . fOllows that the futrc$ions A&J) .j = C. i:. We consider the opepr set @ of R”-l by p&(t) and i&(t). be’ a c ilormal :coordiurate gysteti’iti B &igM&%ood U 6f x0.from (15).involkng dt.Tp.

. i=l ._o. once the‘ coordinate system . . = ((a/&l).in:Proposition 1.2 and . . This proves (i). connec. For each X 6 T. Xl. Let u.. . . (dm. 2. Let M and M’ be di$erentiable manifolds with linear _ the proof goes through when Q is defined. Asume .. . and the horizontal lift u1 of x.:onl g :curve in M (cf. . (iv) The forms 0’ determine the’ cross section the proof of Theorem 7. ’ proof of (ii) is similar. . then u = a’. Proof of Lemma 2. it is determined by (VT).o. connected. Let T: be & tmsm spade of gpe !(r. THEORE& 7. coincides wilt? F. b.$.. . T’. .T with respect to (i) YIJ$ ten&rs”(V*T). . .T& I.2.’ . determine t-he func. The tensor field K and the function i here corre. m = . Proof. i = 1. . and (5. we obtain K(4) = ~r’uL().@oof of ‘Lemma 1 -“This is a special case of Proposition I.ct given by. C&OLLARY ‘7..f M and . (aj8xn) 1 of Chapter JII& :> connections. . d $ t I ... .’ let if & the T+alued where C(X) Q L(M) is considered as a linear isomorphism of R” functioic’ de&d along u. we obtain u(x) =--‘(.u(x)-1x. . . . We shall show (Q’ . . . (v)’ The cross section o and the forms ~j determine the... s) ovq .Y). .~.2. . have 0 I t 5 1. i. (Actually. . nloreowr. of To prove (i) and (ii) we need the following lemma. Let T. Propositioa.. where x E (J.q (r.. . &m&y connected analytic manifolds with complete anal_rtic &ear cyuztions.4. a(x)-‘X = a’(x)-‘X. we apply Lemma 1 to the torsion T.1.t~$$!?. m = 0.. (&q(X) = lY(uX) = a(x)-++JX)) =. u. coincides with F. .r . x” is fixed. t = 0. . ... 1. . an) are fixed. z-. we a L EMMA 1.x’ = fa’. of Chapter III is defined . This is an immediate consequence of Corollary 6. TRANSFORMATIONS 261 260 FOUNDATIONS OF DIFFERENTIAL GEOMETRY with u. . be h ho&&Z fzjit if a curvk’kt. / subset U M. If u*O = u’*tJ on U. we see that. . . 1. . Now (iii) follows < .e ui .alobg xt.. i: 1 * : onto T. n. 0 I t $ 1. . .3 of M’ whose difuential at x.)m T) is the element of the of U onto I’ defined by tensor space. Using the same equation for u’. . It suffices to prove the following five ‘statements.2. then where .’ TV. it is not hard to see that -P We use the same notation as in the proof of Theorem 7. the geodesic curvature and the covhriant $$retitiation o f . .(i) determine the forms ‘ei k&(f) determine the functions 4(t) and &t).thc cross section v and the fuqctionf there.O.Tk with components’&+ pjk(t)/dt”. determine the func. . the To prove (i) . .*(t) is an analytic function of t. R and V (resp. 4 from the formula (1) and (Z). . m’= 0.. LEMMA 2. In particular. VI.1. QED. Although v Theorem 7.‘. . s) . setting * yi of = xi.A 9’.f’ ’ (iii) The functions ?&{l) and $... s) . m = 0. I!‘\. .A/’ are.are the components of V*. The .2..) Since each f’. Letf be the analytic homeomorphism times) implies that. Then we have Finally. In Theorem 7. (iv) follows from the following lemma.e!s$s’ ($J&i .(M). a! yr pf &. . co&de&d C+ a &neG mopping of** onto the tensor’ space Since this holds for every X in T. are all that f is an affine isomorphism of U onto K determined by (VmT).(M).f&/dta). 1.l.R”. the-krmma for.(M).1 implies that the functions f$(t) and tions &i(t)..‘( ( Vt.~t& whole of M. QED. If the normal coordinate system$$ . .2. Chapter III.3. x” and Clearly the differential off at x.. tu L(M). spend to .1. . Given d’tensor jield K of t@e’ (r. fhp there exists a z@ue a$ine &morphism f of M o?lto .x1.3. for each t. 1 Lemma 2 for Theorem 7.\? :--:h. Let (T and (T’ be two cross sections of L(M) over an open tion form 0. n.. OltSl.” ‘whe. R’ and C’) be the torsion.. Then Lemma 1 (applied m system exists and is unique. . . (v) is evident from the definition of (5. 1.

I’(cf functions 3&(t)’ and &(t) are constant functions and hence are Example 5. .. then the/e zs an afine consisting of normal coordinate gstems. 7 . Xi). We use the same notation as in the proof the tensors T. . . . .. .(M) + T.(M) + T. onto a neighbor od V ofyO such connectioti i s anal_ytic. X$i.L. TzO(M) onto T. then the connecti?n is invariant by (20) Ui = +j”. we see that the cross section 0: U -+ L(M) r: T. and RbO aty. . respectively..+ = Sk A. . i. Proof. n . qrom (1) in the (unique) local al&e isomorphism f such that f(x) = y and that the differential off at x. connection such that VT = 0 and VR = 0.~RY 7. 2. . an) by virtue of the assumption that VT = 0 and VR = 0.i. y. Appendix 1) that the functions . From (19) and (20). By Theorem 7. Namely. . Let M be a di$erentiable manifold with a linear isomorphism of.1 and 7.j =” 1 . Hence the of M and for an arbitrary curve T from x to.5.o’ and RX0 connection such that VT ‘= 0 and VR = 0. Let x1. In the proof of Corollary 7. QED. . xn. n.the components I’& of the connection are analytic in x1./axm + x:1 rf$q.. III) : &ROL. and R. coincides tiith F.obtain andonlyifVT=OandVR =O.1. QED. Since the functions ?$(t) and l&(t) are &nstants f(x) = y and that the differential off at x coincides with T. and R. we obtain 7 maps T. . coincides with the parallel displacement proof of Theorem 7. ~’ ($g*wj -c . we saw that is given by if YT = 0 . and RzO (and the coordinate system x1. The connection implies (cf. which do nof. (iv) and (v) in the proof of . . there exists a local affine isomorphism f such that ofTheorem 7.and VR = 0. . i = 1. and Rz at x.. the parallel displacement T: T. respectively. . can be expressed as follows (cf. .1.. into the tensors TL.. an). and R. Xi).2. Y that f (x. .By Lemma 1 in the proof of Theorem 7.inverse rnatL. 2”). . Then f maps T. parallelism. 1.1. into T.(M).‘(t) and zikl(t) are I’ is said to be invariant by parallelism if.j = 1 . matrix of (Xi).= Cj (Xi CfBj. .j.CXj/&“+lj.depend on (a’.be the.2 of Chapter I). . and (22) I&F & k~(ac. . the in n-l-l(U) c L(M) in a natural way as in $7 of Chapter II. Letfbe a local affine By con$aring (21) with (2) in the proof of Theorem 7. r. We show that a neighborhood ofy.5. OR = 0. andy of &. respectivily. .2 of Chapter Theorem 7. . Let 7 be an arbitrary curve from x toy. We introduce a coordinate system (xf. we C OROLLARY 7 . .y.2.4. and that the di$erential off at x.1 Let M be a manifold with a linear connection I’.x of (A)). the forms o~j are also analytic with respect to (xi. Propositions 7. and & a’re analytic in x1..% 262 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI. and R. i. into T. Let M be a di$erentiable manifold with a linear (18) Oi = Cj Ul’dxj.2. for arbitrary points x and y analytic in t and depend analytically on (al. Assuming that the connection is invariant by parallel- is&n.O(M’) and maps the tensors T. then the canonical form and the connection form Our theorem now follows from (iii).) = y.(&: + )& l-&j’ dx$ Proof. for any two points x (19) w. Then. . Hence the parallel displacement Transforming (22). R. QED.. If F is a ‘<inear T HEOREM 7 . -TRANSFORMATIONS 263 // VT -= 0. Ui(dXj” + C. there exists a functions 2. . V’ T’ = 0 and V’R’ = 0. Clearly it Proof. we have Ii Let-(Cj). I . . This means that T and R are parallel tensor fields. .&into the tensors T. . With respect to the atlas at x. = C. If we denote by (Ui) the inverse determined by Tz.(M) maps is sufficient to show that. The converse is also true. .. there exists an k&e isomorphisin of a na’ghboihood of x onto The forms ei are analytic with respect to (xi.. Lemma 2 in the proof of Theorem 7. Xj” dxm). A line& co&e&on ii invariant -by parallelism if . . M is an anabtic manifold and the isomorphismf of a neighborhood U of x. .. . ... let T be an arbitrary curve from x toy.. Since VT = C and VR = 0. we obtain isomorphism such that f (x) = y and that the differential off at x coincides with the parallel displacement T. (23) lT. 6 .k=l of Theorem 7. We follow the notation and the argument in the proof open set U.. xn be a normal coordinate system in an Proof.

at x. . respectiveb. . . . .2. . we have a solution of the following system of linear partial differential THEOREM 7. . . b. . . QED. xl.\I and M’ have the 111. . ) t(m). x”. . .5 and Theorem 7.8 and from the fact that. . . simply connected complete Riemannian manifold of where the functions a.(v) da’. TRANSFORhfATIOi%S 265 which shows that the components I$ are analytic functions of T HEOREM 7 . . . follows from Corollary 7. they are analytic functions of x1. a 09 = Ej a:(u) duj = B. Xj) induced from the normal coordinate C OROLLARY 7.M and M’ are. 9.7. connected. . Proposition 1. Let (xi. ) wnl are analytic with respect to both ul. . i = 1.4ppendix 1). If F LEMMA. are analytic with respect to (xi. The second assertion which dejine the coordinate change are analytic. . and RvO. unr). Proof. Any two connected. We write In 93 of Chapter V. X’. . . . moreover.264 FOUNDATIONS OF DIFFERENTIAL GEOMETRY VI.f” are analytic functionsof xi and Xi. . simply connected complete equations : Riemannian manifolds of constant curvature k are isometric to each other. y” are clearly *independent of Xi. . . . . the following lemma implies that L(M) is an Proof. Let M be a connected. for each real number k.5 of dixerential off at x0 coincides with F. we constructed. m. u’” constant curvature k. QED.. This proves the lemma. then the functions In particular.. . simply connected complete (resp. Let ul. . then there is a unique a&e a local coordinate @em valid in an open set U (resp. into Let xl. . Proof of Lemma. . simply connected and complete then there exists a unique Since the n2 + n l-forms 8’ and wi are analytic with respect to afine isomorphism f of M onto M’ such that f (x0) = yyo and that the’ (xi. . urn) are analytic same sectional curvature k. urn. . O* be 1 -fohns dejning an absolute parallelism is a linear isomorphism of TJ M) onto TJM) which maps the tensors on a manifold P of dimension m. . Y$ . i. vm) be Tz. * * * >.b. we have VR = 0. Xi).7. . Since y’. analytic manifold with respect to the atlas consisting of the Theorem 7. . the functions f i(ul. the lemma just proved.(v)) are analytic in ul. In Theorem 7. Our assertion now follows from Since the functions c:(v) and af(u) are analytic in vl. . If the form transformation f of M such that f (x0) = y0 and that the di$rential of 01. at x0 into the curvature tensor Rio sty. . Namely.) and define an absolute parallelism (Proposition 2.10. Xi) and (Jo. . at y. fatx. the group %(A!) of ajine transformations of Al is vi =fi(d. . . . if . Let (c:(d)) be the inverse matrix of. . if both . urn and Theorem 7. vl. n. . By Chapter V). . is strutted. Urn). Let 09. transitive on the local coordinate Rz.4. This proves the first assertion of Theorem 7.4. . . . coordinate system (x’. .(u) (resP. urn (cf. . x” and pi. (b:(v)). By Corollary 2. Since we have already proved that the form&w. . . connected. . W/W = C. and Theorem 7. space of constant curvature k is isometric to the model we con- Then the system offunctions vi =f’(u’.j=l. v). . then any linear isomorphism in u1 >***. . .p” be two normal coordinate the metric tensor gi. x”) of M. . Chapter III). . necessarily maps the curvature tensor systems in M. . QED. . simply. . (cf. i=l . 8 . ci(v)af(u). . . . . .7 we have “’ .3 of Chapter V. . This is an immediate consequence of Corollary 6. Q E D . Any connected.. .. . . The first assertion is clear.8.. F : TJM) -+TJM’) mapping the metric tensor gz.. As an application of Theorem 7.2 of systems in L(M)in d uced by these normal coordinate systems. for a space of constant curvature. . y’.isF. with a complete linear connection such that V 7’ = 0 and VR = 0. . and RTO into Ty. m. . .9. * * * . connected manifold systems (xl. urn respectively. x”.. . um (resp. . . . urn and ~9. . . the connection is analytic. Proof.

_ . . where D is an open set in R”~?‘“. s) is dz@entiable of class C”‘.. s). . s) is di$erentiable of class Cp+l . Let f (t. Iff(t. rl = (Tl’.W. s) is continuous ‘. 71. 9. yl. ) y”). 0 . then g(t. . b’ < d and D’ is an open subset of D.‘. in t and of class C”. . . s) is di$erentiable ofclass Cl in t and 1.y. where I. > 17”). Then y = q(t. .xm).1 W.s) = 9. ) P). 7. . 267 . Iff(t. For the sake of simplicity.y. s) which depend on the parameters s.r the system of differential equations : (+t = f (t. sj E D. The proof will be found in various text books on differential equations. . x = (xl.. in y and s.. . . in t’ and of class C” in q and s. s = (9.t.1. APPENDIX 1 The ‘purpose of this appendix is to state the fundamental theorem on ordinary linear differential equations in the form needed in the text. s) be a family OJ n functions dejined in ItI < 6 /(’ and (y. Conside.. . .. s). . s) E D’. S) is called the solution satisfying the initial condition: y = q when t = 0. 0 p :. . 77.y. . /. s) of n Jitnctions dejned in ItI < b’ and (q. s!/at =f(t. 2 CI’ = (f/l. 1 . .. f = (f’._ 1’1.Y”). such that (1) Q (ti q. 7 (4 ‘I. _’ (2) %(t. in t ad dzj?erentiable of class Cl in y. I Then we have THEOREM. ‘1. ‘:.y. ..f”). . q 5. then there exists a unique farnib . . . we use the following abbreviated notation : 3’ =: ( . (3) Y(O.

@st a sep&abfe’ U(x. locally separable metric space M. . We see easily that y z 5% if 1 . m. X& for every x E M. that is. Kelley [I.\We have also xRy if and only ifyRx. Proof of Lemma3. A compact metric space is separable.!?‘A = &P-l& n = . 1201). Y) bi: the Interior of the sphere of _ center x. then every solution y = v(x) is differentiable of class Cr+l.o. . we write Sx for S(r)... The following lemma is due to Sierpinski [ 11.j=l. It is called ZocaUy @arable if every point of M has a neighborhood which is separable. for a metric space. If {x> is the set consisting of a single point x. It follows from the theorem that if the functions fj(x. U(X. I ’ L EMMA 2. n. i=l.. Set . yR.y). locally sefirable metric space is #arable. are R-related to a point of A: SA . Evidently.a separable .ob. we denote by SA the set of points which . Y) = (3 c M. ‘Let d be the metric of a connected. ... LEMMA 3. p. For every subset A of M. 0. the separability is equivalent to tjre second axiom of countability (cf.7 of Chapter VI. A connected. y) -C Y>.2.$@ radius r. For every point X’ c M and every positive number r. let U(x.3. . d(x. This fact is used in the proof of Theorem 7.~ r $ .there. This is a trivial consequence of Lemma 1.-r’) ‘containing x. . 1381. For the proof. A locally compact metric space is LocaJy separable. A connected.. see Kelley [l.= (y Q M. p. We Say that t$vo p&it4 ‘x and y of M are R-related and write xRy. ‘$‘. r) coritaining y and .. ..268 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Consider now a system of partial differential equations: aypw =fj(x.. locally compuct metric sp&e is separable We recall that a topological space M is separable if there exists a dense subset D which contains at most countably many points.Note that. . The proof of the statement in the title is divided into the following three lemmas.y) are APPENDIX 2 differentiable of class C’.x for some x E A}.

J .. Proof. m d(w.4 of Chapter III. . thus proving (6). then (b) If d is separable. z ) r. r. rO). This shows countably many such spheres and the union of these spheres is that (3) implies’ (4). i m p l i e s S”x c Sk+m+ny f o r e v e r y k a n d h e n c e C’(x) c U ( y ) . Let or. Kelley [I. 1561).there is z E D such that d(z. .D a separability:The implication (2) --f (3) is therefore a consequence c o u n t a b l e d e n s e s u b s e t o f . Since D is dense in . r -rro}+<.y) + d(y. (4) M is paracompact. w) < d(y. p. Hence x E S”Z c S”+“y.5 of Chapter IV. (c) implies M = U(x) for every x E M. U(x) is r.g) < ru < r. From* d(x. r.p) < m i n {rO .y) < r’.y) -r d(y. separable sphere [‘(j. r. z ) + d(z. d(x.: -c .y) c r. thus completing the proof of the statement in’the title. By (b). r’). because there are only hence. 2) -2 d(x. To prove that C’(z. rlj so that (2) M is metrizable. rr:. rO). To conditions are mutual@ equivalent : prove that c(z.t D and whose radius is a rational number. This proves’that zRx for eve* 2 t U(y. r’) which is separable. The implication (1) --f (2) was proved in Proposition Hence w E E(y.c ^s. Proof of (c).J. . r) which is separable and x is in U(z. 2) + d(t.c..-1. As we stated at the beginning. Hence M is separable. ’ 2.y) < r-1 < r’. Let 12 b e a s e p a r a b l e s u b s e t o f .I. 7) which is separable. w E c’(z. 6) + rl < rt.\f a n d .y). ’ -then . ‘It suffices to show that U(y. M is paracompact (cf. The implication (4) + (1) follows from separable. for a metric that is.d(x. be a ‘pdsitive rational number such that d(. for every*?6 M. Then there isJ E d such that XRJ~ and ‘there Propiition 1. APPENDIX 2 271 270 FOUNDATIONS OF DIFFERENTIXL GEOMETRY it follows that x E U(z. r’). It suffices to”. a separable sphere’rirhose center is a point of Urysohn’s metrization theorem (cf.. 3. we shall show that l_-(z. If (a) Sx is open for every x E ?tii. space. It suffices to show that /7(z. For a connected di$Erentiable manifold M. r) and x E V(y.u. is a. rO) is separable. Let y be a point of Sx. r) a n d u(y.d(x. -c rl . eithm U(x) n U(y).li. This separable and that y E U(X. (c) Set U(X) = U S”X for each x E M. .y). r. rO) contains x and is separable. T h e n (3) M sat$es the second axiom of countabilip.Y. QED. Kelley [l.Prdbf Qf (B). Hence U(x) is separable.y) < r. d(y. Assume that U(x) n U(y) is non-empty and let Proof of (a).provre that every of the statement in the title..rl. Hence U(x) Let rO be any positive number such that is open. w) 12 d(y. U(y. . Since d(x. S”x is separable for every n. we and only if x E Sty. r). E I-(z. e? r’. there is a positive number rr such that Similarly. contained in U(y. THEOREM.+ rl < r‘. If z E Ii(-y. r’) are From z E Smx. W) < rl. we have U(y) c U(x)-.y) -+ TO’ < m i n [r.). 1251) a n d . Let x E S/f.. Then. positive numbers r a n d r’ s u c h t h a t U(. ‘CVe prove the following three statements: show that c(z. Let ze. Since XRy. then -41 is metrizable by . rl) is separable. . the following Hence z is in U(X. By (a).J is contained in Sx. is empty or U ( X ) A= U(T). zj : d(x.. there exist z E U(x) n U(y).d(x. the second axiom of countability is equivalent to the . We are now in position to prove -. a-1 and hence w E C’(y. If (3) holds. I~) is (1) There exists a Riemannian metric on M. tr). . r) containing X. 2) $ d(x. < r . Since A4 is connected and since each. SA k U SX is open for any subset A of Al.-j’ E .y).y) 2 d(w.x E S:l is contained in. we obtain x E S*z. T hen z E S”‘x and z E S’y for some m and n. . + d(z. r.. rO) is’ contained in 1r(y. for any x. so is S/1. r. r. p. open. d(x. d(t.

taken over all a such that m: c Ui..‘f.. compact manafold M such that each Ui has compact closure Ui. ... for each point x B M. For each that Eifi(x) is a finite sum for each SC. . We set. and (3) gi = 0 outside Wi. Then every dtxerentiable function f dejned on F can be extended to a hood U of K. . if the following conditions are satisfied: neighborhood of x such that wz is contained in some Ui. We first prove the following three lemmas.L(x) # 01.the proof of Theorem 1. for each say that f is differentiable on F if. Then we set Partition of unitj f =fz. exists a partition of unity .For every compact subset K of M and for every neighbor... {Vi} is an open covering of M. there exists. and (3) f = 0 outside U. . there exists j. let W. 272 . .. Ix E M. intersects only finitely many Ui’s. xk of K such that vz.j = 1. . We first prove Let {Vi} be as in Lemma 3. fl(xl) . This completes the proof of Lemma 2.onPn V. QED. Then there a differentiable function gi on M such that (1) gi . We are now in position to complete .fi = SilS* the:=: b di$erentiable function f (of class Cm) on M such that (1) Then {fi} is a partition of unity subordinate to (U. letfi be a different- iable function on hl with the property off in Lemma 1. Then.(O) = 1 and thatfj(xj) = 0 for lxil 2 a. . Since K is compact. for each i.(fi) subordinate to {Vi). the same index set) of { U.}. x”) = thatf =f. i.. every point of M has a neighborhood which such that Vi c Ui for every i. fn(xn). . n. of x such f.‘let Wi be an open set such T HEOREM 1. (2) gi > 0 on Vi. lx’1 < u}.2 M. for each point x f F. : Proof of Lemma 1. is contained in the corresponding Ui. Since two are valid without the assumption that M is paracompact whereas the third holds’for any paracompact topological space. Since {IV:} is 1ocalIy finite. finite number ofi’s. x = 0 and U = {(xl. fZOonM. . Then there Let {U& be a locally finite open covering of a differentiable exists a locally jnite open refinement {Vi} (with. Let (1) 0 5fi 5 1 on M for every iE I. x”) . The first each gi contains Vi and is contained in Ui and since {Ui} is locally finite.e. . we let fj(xj) be a differentiable function such that a differentiable functionf. g > 0 on M.. We set f(xl.. . (3) i-J&) = -1. we have Vi = lJ l&r. Let F be a closed subset of a paracompact manifold M. APPENDIX 3 there exist a finite number’of points xi. be the neighborhood of x defined by fz > Q.} manifold M. where the union is .}. Let {Vi} be a locally finite open covering of a para. that Vi c Wi c wi c Ui. For each point x of K. B y Lemma 2. Let I’. This can be easily reduced to the case where Let f be a function defined on a subset F of a manifold M. LEMMA 2. We thus obtained ‘an open Note that in (31. Let {Ui} be a local&jr&e open covering of M. This proves Lemma 1. the closure of the set Vi be the union of all IV: whose closures are contained in Ui. i.e. and (3) f = 0 outside U.r 0 on M. APPENDIX 3 273 Proof of Lemma 2. v *** u Vzk 1 K. the sum g = Zi gi is defined and differentiable on M. Since the support of Proof. .I. 0 5 f d 1 on M. . let (2) The support of each fi. {IV:> be a locally finite refinement of {W. For each i. on an open neighborhood V. + ..(2)f>O on K. T HE O REM 2 . (2) f(x) = 1.(x) = 0 except for a covering {Vi} with the required property. for each i. there exists a dzjerentiable function f on M such that (1) di$erentiable function on M. For each point x E M and for each neighborhood U of x. We M = R”. x E M}.* * +&Ii. LEMMA 3. be an open functions (fi) on M is called a partition of unity subordinate to the covering {U. A family of differentiable Proof of Lemma 3. . Forieach point x. : .

pp.j. is involutive (where L.S. the distribution I x -+ L. i. This result is essentially due to Freudenthal [I]. we denote by ill. then X + YES. If we set v. is really a finite sum. S is a subalgebra of g. (cf. denoted by K.274 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Proof. This shows that if > “i X 6 S. Let a be any point of H and : = xt. Y E S. This shows that if X. Chevalley [ 1.lemma. T. We claim . = [&. then ir.(G) which are tangent \ the sheaf of germs of differentiable functions on a paracompact to differentiable curves of class Ci contained in H. letfi be a differentiable function on an open neighborhood V. Then .. Let r be any real number and set zt = qt.(G) = g is a subalgebra of g. = r 5. Given a curve x. 120-1221). Theorem 2 means that Proof. such that wtp = x. We define subgroup of a Lie group . THEOREM Let G be a Lie group and H a subgroup of G such that QED.] (cf. x E G.. of x such thatf.. which is sufficient g = &f&&. If there is no k’. and set On arcwise connected subgroUps of a Lie group gi = restriction off. 1. Y] E S.. Let .F and V. Then H is a Lie subgroup of G. Let xt and yt be curves. in G.. We shall prove here the following weaker theorem. to Vi. Kuranishi and Yamabe proved that every arcwise connected Let { fi} be a partition of unity subordinate to {UJ. In the terminologies of the sheaf theory.. Y c S. Since (UJ is locally finite. is the Lie subgroup of G corresponding to the subalgebra S. = f on F A V.?O + B Lie subgroup(see Yamabe [I]).. = .ytx.. then rX E S. 375 . Then i. for our purpose (cf. ? Proof of Lemma.2 of Chapter II). We first prove that K 2 H. curve of class C1 which is contained in H. Theorem 4. then [X. We shall show that H = K. We identify manifold M is soft (“mou”. Let S be the set of vectors X E T. For each i.Ui be a locally finite open refinement of the covering of 111 consistmg of M . is the left translation by x) and its maximal integral manifold through e. Chevalley [l. Thus g is differentiable on M. every point of M has a neighborhood in which xi fig. we choose such a I’. the ‘... If Ui is contained in some V. For each x E F. There exists a curve w. in G such that x0 = y. which contains Ui. Since S c T. vector tangent to the curve at the point x. pp. This shows that if X. J : LEMMA. we define a differentiable APPENDIX 4 function gi on Ui as follows. every element of H can bejoined to the identity e by a piecewise d@erentiabb It is easy to see that g is an extension off. then we set gi = 0. 2381). 0 5 t s 1. 120-1221 or Pontrjagin [l. a curve from e to n so that e = x. = Codement [ 11). x e F.’ and we have &.. = xtyt.(G) with the Lie algebra g of G.3. and a = x1. : p. thus com- pleting the proof of the .

$. In the second case.. 0 5 t 5 1. A is non-singtilarand A”-‘.K at the origin is non-singular.. (direct sum).* THEOREM 2.fori=l. k be the smallest integer such that Ak = O.. then A = 0. (h.** Xfk.that k 2 2. 1. for each fixed t.. . . or A = al. . contradicting the assumption of irreducibility. the differential off: U . L emma 2 for Theorem 7.S for all t. derive a contradiction. . let e. mce w-is invariant by G. . E L. . L. Let G be a subgroup of O(n) which acts irreducibly on U onto an open subset f( U) of K. x1. k.. (1) If A is nilpotent.. .2 of Chapter APPENDIX 5 II). J is a linear transformation such that JY T (cf. : the identity transformation of R”). I. (1) Let. (0).‘(i. We begin with the following lemmas.. contained in H. .*2 We prove the following two theorems.. Since every element of G commutes with A and hence with fi(A). Since K is connected. . Let W & (x c R”. Let G be a subgroup of GL(n.for all t and x. Hence a E K. h the first sa&. Consider the mapping f of a neighbor- hood u of the origin in R” into K defined by f(ti. we may assume that f is a diffeomorphism of THEOREM 1. $. : .. where ’ Wi = {x l R”. = 4:’ * AK = 0. . . Let G be a connected Lie subgroup of so(n) which acts iweducibly on R”. fi) -=‘li. be curves in H such that xh = e and ib=e. L EMMA 1. Proof.. From the defimtlon off. we. where a and b are real numbers.‘(T) at e and hence lies in S. we have either W = R” or &‘a=.J(A)x = 01. K C. . To prove that H 1 K. .. thus proving our assertion. we the n-a’imensional real vector space R”. . Since & . In fact.) is the vector tangent to the curve L. . b f 6. Let A be a linear transfmmation of R” which commu’tes with every element of G. Assuming . showing that K 1 H. $j form a basis for S... it follows that Wi are both invariant by G.jA F 0. Hi . ek be a basis for s and Irreducible subgroups of O(n) Xt’1 * .s. ( a : real number. R) which acts iweducibly on R”. (2) The minimal pobnomial of A is irreducible over R. + b J. . As = O>.) = 1 tk) E U. and n is even.‘Then . (2) If the minimal polynomial f (x) of A is a product jr(x) -f a(x) : wrth (... = e. Then G is closed in SO(n). (3) Either A F al. Thus 277 . T&n every symmetric bilinear form have f(U) c H...tI+zn RF F WI + IV. the curve x1 lies in the maximal integral mamfold K of the distribution x -+ L. Since i.. Taking u sufficiently small. . This shows that a neighborhood of e in K is on R” which is invariant by G is a multiple of the standard inner product QED.276 FOUNDATIONS OF DIFFERENTIAL GEOMETRY that the vector it is in L.

+ block being with respect to a certain orthonormal basis b J.x ). 278 FOUNDA TIONS OF DIFFERENTIAL GEOMETRY APPENDIX 5 279 f(x) = g(x)“. g. If b f 0.. Jx = Kxj' and W. we have ad (c)g c g.1)” = det J2 = (det -y. and A = al. In the second case. A = 0. Assuming that G is not closed in SO(n). + M’.y). on the complex vector space C” with standard hermitian inner Proof of Theorem 2.)/b. Jxj = (X.0)” +. y) + (x. of there is a synime’tric linear trarisformation A such that f(x. P r o o f . = {x E R”. His proof.. Proof. We K commute with tvery eleinent of G. (2) If A is skew-symmetric. Applying (1) to g(A). Since G is irreducible. have already seen that the connected Lie subgroup generqted by b f 0.. Proof of Theorem 1. By . c real:. morphic with the circle group. then 3 = (c$. w ic h is h a contradiction for x # 0. Now consider SO(n) and hence its subgroup G as acling of G. Thus< = g.JKx)/2 and representation of a semisimple Lie algebra is completely reducible. so is-b J. the minimal polynomial of A is either x or x2 + b2. and ?Y2 are invaiiant by G. Jis a certain skew-symmetric linear transformation with 52 = -I. then A commutes with every element g = [g. poss’ibly with b = 0. (. and hence is iso- Let A. then B =cA. (Ax.. yj . generated by 5 is closed in so(n). . R). sl. . Let A + 0 and B f 3. B are skew-symmetric linear transformations.u) Now J is representable by a matrix which is a block form. It follows that the orthog- (3) I$ . g has a positive definite inner product (A. and z E W. it follows that (Ax. Since ad (C)g c g. g].y) ‘= (x. which implies that g is . TV1 n W. In the first case.. Iff is invariant by G. 1. = (0).. B. (3) of Lemma 2 implies that B = CA for be its closure: Since G is a connected closed subgroup of SO(n).I) 1 dimension > 1 and since gi = [gi. = R”. let G with every element df G. We show that R” = W. A =‘--bJ 3 is closed in SO(n). gi] for each i. W. and n = 2m.aI. based on a theorem of Weyl that any Clearly. . which commute irreducible by G. where g( x ) is irreducible. J is symmetric so that contains a proper ideal. (3) By (2)) the minimal polynomialf(x) of A is either (x . . = Rn or W. Obviously. . either K -b-J or now prove the following result by the same method as Yosida’s. W. (direct sum).r)” > 0 so that n is even.4 # 0 and B are skew-symmetric and AB = BA..( 1) of Lemma 2. that is.l) By (3) of Lemma 1. each or (x . z = (x + JKx)/2. V. as we see by setting y = (x . we see thatf(A) = g(A)L = 0 implies g(A) = 0. Every connected semisimple Lie subgroup G of GL(n. A = a& + bJ. Thus dim 3 5. A connected semisimple Lie subgroup G of SO(n) is closed in SO(n). say. We imaginary. then A = al. (3) Let A = bJ and B = /‘K. Ay). This will finish the proof of Theorem 2. ... where J” = iY2 = -I. each block being I \ . If z. We have (.. then A = 0 Since g is the subalgebra of the Lie algebra of all skew-sym- metric matrices. We first show that thi: center 3 of the Lie product which is left invariant by SO(n). i. . Then C” is the direct sum algebra g of G is at most l-dimensional. where.. y)’ = a(x. In the case of a subgroup G of SO(n). d = aIn and henctf(x.. y). = (K E R”. Thus we (Jx. A = ~1%.. cos t -sin t ! block form...and s = s1 + 5’. In ‘the second case. \sin t cos tj’ (l)‘lf A ii symmetric. let J = (A . (Ax.b2 with b # 0. Then J2 = -I. C) is closed in GL(n. B) = or A = bJ. Ay) = 0.e. We have JK = KJ. -trace (AB) whichisinvariant by ad (G). . onal complement I of the center 3 in g with respect to this inner product is an ideal of g and g = 3 + 5 is the direct sum. which are invariant and 4. In the first case. Let G be a subgroup of O(n) which acts irreducibly on R”. be linear trunsfwnutions ‘of R” which commute with G. some c. If A is symmetric. For any symmetric bilinear form f(x. We first remark that Yosida [ 1 J proved the following result. Since g is a direct sum of simple ideals gl. (2) Since the eigenvalues of skew-symmetric A are 0 6r purely see that 4 is a direct sum of simple ideals: s = s1 + . C). . Let g be its Lie algebra. where it is a Lie subgroup. i I of R”.$X) = -(x . Jx = --Kx). i. because J and also works when we replace GL(n. . we need not use the Weyl theorem.+ 5. where J2 7 -I. g c 3. C) by GL(n. Since of complex subspaces V’. If dim 3 = 1.e. K = -Ji This means th?? B = CA for some c. We nowshow that the connected Lie subgroup with J2 = -I. Every x E’R” is of the f&m -y + z with y t W. we have either ‘Mt. The l-parameter subgroup exp tJ consists of matrices of a LEMMAS. . then the orthogonal comp4cment s’ of sL in 5 is an ideal of 5 (in fact of g) . .

. Vi commutes with the actions of {Ai. which is a/&“) > 0 for each oriented local coordinate system 9. 161-1671%.Since :1 + A. is a constant. (which is an obvious consequence of Lemma 1. [I. we -w s have H = 0. Proof. By Schur’s Lemma An n-form o on M is called a volume element. considered as a function oft. that is. is the Lie differentiation in the direction of X. G is closed in so(n). (2). Since the Lie algebra of SO(n) has a positive definite inner product invariant by ad (SO(n)) as WC already noted. Chevalley [l.] = O.6 of Chapter I). has determinant 1. is a representation of g on 1: and since g == [g.l. on Let M be an oriented n-dimensional differentiable manifold. we have Ui .Q. Now let U c it. GREEN’S T HEOREM. By continuity. For any . that is. . Since we have (cf. . . For every vekorjield x on M. where L. we have s M div X dv = 0. we have obviously trace [-di. the integral f dv of any continuous that hi = 0. This means that trace :4? = 0 for every -4 6 (7 and for each i. This means that u = (0) and ij =. B. the restriction of a E 6 on each I’. the dimgence of& denoted by div X.-1 E g. By definition 281 - . Its action B. Proposition 1. = 0. p. pp. . valid for any field instead of R). Thus the Green’s theorem restrktion of c1 E G on each V’i has determinant 1 (cf. S]). 1651) IM 9P w.~ For any A.. This proves function f with compact support can b:defined (cf. . 91. A E $1). where I is With a fixed volume element w (which will be also denoted by a the identity transformation of Vi. p. Each summand Vi of C” is also’ invariant by c and hence by ?j acting on 0.280 FOUSDATIOSS O F D I F F E R E N T I A L GEOMETICY an idcal of ij. B t $1. is a function on M defined by (div X) o = zxo. Since trace B. Let qt be the ‘l-parameter group of transform+ions generated by X (cf. B. denote by Ji its APPENDIX 6 action on Vi for each i. . . it follows that R is the direct sum of (1 and the orthogonal comljle- rnent u of 9 in fi. we have trace Ai = 0 for every A E g. Chevalley that 6 = G.b. if w( a/a& . Let &f be an oknted compact manifold with a jixed volume element o = dv. For eachvector field X on M with a fixed volume element w.. = 0.9. . it follows more intui ’ notation dv). This being the case for each i. Corollary 1 of Chevafley [I.

* 9 ‘X) = (A. . . we P ROPOSITION . Hence assumption T = 0. (div X). . . In fact.. CfX. . Let A. chosen. Since Vxw = 0 and since Ax. APPENDIX 6 2 8 3 282 . . Hence. . The above formula follows also from Stokes’ formula.. we have [$ (~II*w)]~=~= -Lsn. . X.s C> dv(X<. . Xn) proposition as well as Green’s theorem is . It is easy to verify that dv is defined independently of the frame X i. .) = 1.. holds without the of L. . w h e r e G=det(g. . . dv=z/Gdxl~dxaA***Adrn. X.. we have Remark 2. = Lx .. At an arbitrary point x of A$ let X1.T (X.Xi. since ds = 0. X. . we have = det (Cf) = $%. have the same orientation. so that ggj = Z& CtCt and G = l. n) . Let M be an oriented Riemannian manifold. div X dv... The above formula is valid for a non-compact system x1.. as a derivation. .) w is parallel with respect to F. . QED. . d+$.. V 6 T. of the metric tensor g.. .)) . we have det (Cj) = fi > 0. . ..Z -(trace:Ax). a/ax” and X1. . . . We then have In fact. . . . i. Since 8/8x’. X..o)(X. = trace of the endomorphism V --t V. . .Ci 0(X*. Y) and from the assumption that T = 0. . every orthonormal frame into an orthonormal frame and preserves maps every function into zero. X. ..X.4. L.(M).o(X.. E C? .. . ..X.‘. * * * .(M). e a mp A. * .. If I? is an a&e connection with no torsion on M such that dv(a/axl. the volume element du is parallel.) Z.p. . . = 0: . we have Lscu = d 0 ls II) + I-~ 0 do) = d 0 15co.V. . In terms of an allowable local coordinate Remark 1. . . manifold M as long as X has a compact support. at x.u sJf QED.. let (a/&$). Thus the k&4(&. * s. in) is an even or odd Proof. .m = . . . . be a Since the parallel displacement along any curve T of g maps basis 0: T.valid for the volume element du of a Riemannian manifold.A. Let X.(OJ(Xl. . as in $2 of Chapter VI.Xi.... X. xn and the components g. = A. where E is 1 or -1 according as (i 1. . . .Y = -V. .(M) compatible with the orientation of M. det (Q2 at x..). * *s y X. We define an n-form dv by .. = Z.’ This shows that div X = -trace A. be an orthonormal basis of T. 1) defined by permutation of ( 1. . .‘alaxn) = l?Zil .5 of Chapter VI) : A. = ..). FOUNDATIONS OF DIFFEKENTlhL GEOMETRY Remark 3. We define a natural volume element du on M.. we have the orientation. be the tensor field of type (1. =j-$o. X. for every vector field X on M. . then. .. Let M be an oriented manifold with a fixed volume have element 0 = dv.. Proposition 2. Our assertion follows from the formula (cf. . * s *y Xn) = -cio(Xl. The formula div X = -trace A. *.X .

Two curves T and T’ are said.x(t). Now let U be an arbitrary open covering. be equivalent.j/m) -+ chosen in su’ch a way that it is piecewise Ck on Z x Z. With these definitions. 5) Q A4 of integers (i.‘When x(t) andy(t) are the following order: piecewise differentiable CUIV~S of &ss Ck (briefly. j) has the required property. m) ” ’ T(m.l>/m. (j . ’ Proof. mapping f: (t.(I .by f lies in some open set of the covering U. we may assume that the homo- topy mapping f is piecewise Ck. into A4 for each (i. 1) =y(t>. where p is a curve from x to a point y and (I is a closed curve aty which is contained in an open set of lf. if the mapping. --.u-1 .f(O. 9) . Letf be a homotopy Z x I-+ M such that f(t. 0) = x(t). A(i. we prove 284 . 1) Let A4 be a differentiable manifold. they are piecewise CL-homo&$ic. We shall say that a closed curve T at a point x is a U-&JO if it can be decomposed into three curves 7 = p-1 . QED. m) ’ ” T(1. 0 I t S 1.f(O. (j’. where p is a piecewise Ck-curve and Q is a Ck-curve. Then T is equivalent to the product of LEMMA. x Z. j) be the image of A(i. J) =f(L 4 = x for every t and s in 1. We . YW The factorization lemma is taken from Lichnerowicz [2. j) be the closed curve in the such thatf(t.) is contained in some coordinate neighborhood for each (b) By the preceding lemma. Geometrically.T(1. f is a differentiable mapping of class (6 . pair (i. S) 6 Z x Z -+f(t. f can be (O.1)/m) --t (i/m. (a) Let 7 = x(t). By choosing m larger if necessary.can take a suitable subdivision Z = i$1Zi so that 2) 7(1. let n(i. then they are piecewise Ck-homotopic. 4 = ~(0) =yP) and square Z x Z consisting of line segments joining lattice points in f(b) = x ( l ) +y(l)for every t and s in Z. 1) .u-1 * . Two continuous curves = x.f(t.O> + W/m) -+ (6 . (a) Any closed curve which is homotopic to zero is equivalent to a product APPENDIX 7 of a jnite number of II-lmsos. By modifying the mapping f in the small squares we may also assume that f is Ck on each of the ma small squares.1)/m) + (i/m&) -L certain subdivision Z = $r Ii. 1 d i. Let 2I be an arbitrary open coking of M. j) looks like a lasso. 2) T(m. if T’ can be obtained from 7 by replacing a finite number of times a portion of the curve of the form . we can obtain a piecewise Ck-homotopy between ‘x(t) and Then each lasso T(i. piecewise Ck- curves). j $ m. f(l.u * p.f(t. 1). p. We divide the x(t) andy(t) defined on the unit interval Z = [0. T(m. j). for a (6 . so that x = x(O) = x(l). Let T(i. j) by the mapping f. j). APPENDIX 7 285 FACTORIZATION L EMMA. For each pair there exists a continuous.U by a trivial curve consisting of a single point or vice versa.. If two piecewise Ck-curves x(t) andy(t) are homotopic to each U-lassos other. (b) Zf the curve is moreover piecewise Ck. j). Z* x Z. 5 13. l] with x(O) = square Z x Z into ma equal squares so that the image of each small y(0) and x(l) =y(l) are said to be homotopic to each other if square . Ck of Zi x Z.1)/m&d -c (W/m) + V-4 0).p. that is. then each U-lasso in the Factorization lemma product can be chosen to be of the form ..1)/m.

Tensor calculus. Kobayashi [6]. The Christoffel symbols were introduced by Christoffel [l] in 1869. a connection in an arbitrary fibre bundle as a field of horizontal subspaces and proved the existence of connections in any bundle. We have . 21. NOTES Note 1. Chern [ 1. where (U. Lichnerowicz [2] and others. 2. projective’ and conformal connections.9] to the introduction of affine. 8. Ehresmann [2] also defined the notion of a Cartan connec- tion. one and reveals best ‘the geometric nature of the connections. Nomizu [7]. Connections und holonomy groups 1. projective and conformal connections.4. The definition of s1 given in this book is due to Ambrose and Singer [l] who proved also the structure equation (Theorem 5. whose examples include affine. 21 defined a connection by means of a set of differential forms o.: founded and developed in a series of papers by Ricci. itwas not until 1950 that Ehresmann [2] clarified the general notion of connections from the point of view of contemporary mathematics. Ambrose-Singer [ 11.} is an open covering of the base manifold (see Proposition 1. 2] and E.5. Although the approach of Cartan is the most’ natural. Covariant differentiation which’was formally introduced in this tensor calculus was given a geometric inter- pretation by Levi-Civita [l] who introduced in 1917 the notion of parallel displacement for the surfaces.2 of Chapter . Cartan [ 1.. His paper was followed by Chern [ 1. was given a systematic account in Levi-Civita and Ricci [l] in 1901. for the first time. the notion of a Rieman- nian space originates with Riemann’s Habilitationsschrift [l] in 1854. Ehresmann [2] defined.4 of Chapter II). Although differential geometry of surfaces in the 3-dimen- sional Euclidean space goes back to Gauss.ructure group.II). He introduced also a connection form w and defined the curvature form Q by means of the structure equation. on U. See also Kobayashi [6] and Takizawa [ 11. with values in the Lie algebra of the st. This discovery led Weyl cl.

Walker [3] proved that an r-dimensional bundle e(M. I-I. a G-structure on M is a tions on homogeneous spaces. any G-structure on M corresponds to LichneIowicz [I] proved it explicitly.C* * g’. ‘In our terminologies. there is a unique torsion-free connection in P. p. of Hano ancI.&eorem Let G be the subgroup of GL(n. The results in $10 of the holonomy group by itselfdoes not give any information other Chapter II were obtained by him in the case of a linear connection than thosebbtainable by topological methods. 1011). natural manner.. combined ( $9 of Chapter III).bundle.2 of The fact that the holonomy group is a Lie group was taken for Chapter IV (in which g can be an indefinite Riemannian metric) . Chevalley also obtained Corollary 10. . [7] and Kobayashi [6] by 6&t proving the. II in the case of a linear connection (unpublished) and his result able manifold M. 2. reductios.ekuft. The notion of holonomy group is due. F(rJ.1 . a result of Narasimhan and Ramanan [I] which is closely related Weyl (9 and E.which. This will be treated in of G. to the . G) and a co.of.. This structure 8.whm. by Nomizu trivial. can be quadmti4form of any signature: obtained as the inverse image of I’o by a certain homomorphism of P into E . Wang [l] as in $11 of Chapter II. See also the holonomy group is of considerable interest. Chern [9) defmed the notion of a Gstructure on a differenti. to E.. We shall here state mention some other special cases. Cartan [SJ proved the following. For a general theory classes of any principal fibre . See Chem [2].is essentially due to. granted even for a Riemannian connection until Bore1 and in fact. Steenrod [l. Bernard [l] and Fujimoto [ 11. I. showed that Theorem 8.7 of Chapter 3. Given a compact Lie group G and a positive integer n.1 of Chapter II) of R. However. See also Klingenberg [ 11. . Cartan was rigorously proved I&ample 5. R). where Q.ia . G). see Chein f33. Nijenhuis’ results were generalized by Ozeki with other.or u closed to the notion of a universal bundle (cf. will be treated in Volume II. n z 3. 7.2 of Chapter II) to the e&t that the structure group G of P(M. A G- and Ehresmaule Kobayashi &i$.t$eorem an indefinite Riemannian metric on M in a similar way to (Theotim 8. G) caa be reduced to a subgroup H if and only if there exists a 4. of Chapter II). The proof was simplified. the following two conditions are T H E O R E M .’ where G is a certain Lie subgroup of GL(n. We Volume II.fiisFcowing fact. dim M I n. Cartan [2. 21. conditions (such as a “torsion-@ linear connection”). C) regarded as a subgroup of GL(2n . epivutent : there exists a principal bundle E( N.T(p)~ whot)c .Qz+i [l] and Non&u [5].structures.wth group T(G). connection if and only if the distribution is integrable. P of L(M) with group G. The notions of local and infinitesimal holonomy groups were connecti& in P whose holonomy group is exactly H means that introduced systematically by Nijenhuis [2]. For = G’ O(~). See also For any connection I’ in?. if G is such a group.m. there& a naturally induced connection Willmore [I. The implication (1) -+ (2) is clear from Theorem 2.288..7 of Chapter I.~ The holonomy .&tan which leave the r-dimensional subspace Rc of R” invariant. His results were generalized by reduction of the bundle of linear frames t(M) to the subgroup G. G).h&q~yL group is T(a)‘. FOUNDATIONS OF DIFPERENTLAL GEOMETRY'/ NOTES 289 given in the text a detailed account of the relationship between (2) For every n-dimensional manifold M andfor every reduced subbundie linear and afline connections. R) in a hg+oiiiy flpup.7 of Chapter I). R) was used by Nomizu [2] who discussed invariant linear connec- with n = dim &f.nection ITO on E such that ( 1) G is the group of all matrices which $reserve a certain non-degenerate any connection I? in any principal bundle P(A4. co+ier t$principal fibre bundle T(P) over distribution is parallel with respect to a certain torsion-free linear T{M).1.~:T( ~) denotes the tangent bundle. [I] to the general case as presented in $10 of Chapter II. one can define characteristic given on M (see Example 5. subgr?fi d of GL(n. 31. 61. A G-structure on a 2ndimensional manifold M :The I. The implication (2) -+ (1) is non- first by-Ambrose-Singer [ 1 f.~ a G-structure isnothing but a Riemannian metric 5: Ry making use of a connection. Cartan [ 1. R) consisting of all matrices (Th+re.(Theorem’ is nothing but an almost complex ‘structure on M. For a principal fibre dimensional ‘distribution.aS @+ucture ‘on an n-dimensional manifold M is nothing but an Y- essentially equivalent. is the Let 0 be GL(n . Nijenhuis [33.

[5] who gave an interpretation of the iRiemannian connections of manifolds imbedded in Euclidean spaices (see Volume II). The holonomy groups of linear and Riemannian connections conrlections were studied in detail by Berger [ 11. it suffices to Walker [Z]. a direct I’jk = 0. the canonical connection in a Stiefel mani.Hopf and Rinow [1] proved Theorem 4.4 of Chapter IV.-bc=l) group. For our present purpose. (xl. connected solvable group. The linear holonomy group respectively. that every pair-of points can be joined by a geodesic. Matsushima [l]). Wolf [S] proved that this is also the case for a compact canonical connections in the Stiefel manifolds.2 of Chapter IV) has been treated by a number of authors. Saito [ 1] the direct sum of invariant irreducible subspaces) but the linear gave a necessary and sufficient condition in terms of the Lie connection is not a direct product even locally. Complete ufine wad Riemunniu~ 6. Lie group G such that the geodesics emanating from the identity The global decomposition (Theorem 6.2 of Chapter V). Such connections will proved by de Rham [l]..[Z] and Hermann [I]. Condition (4) of Theorem 4. Th eorem 4. see Proposition 6. For instance. is compact (well known) or if G is nilpotent (cf. see also Kobayashi locally symmetric Riemannian manifold. See Note 7-(symmetric spaces). the same problem was.. The restricted linear holonomy group of an arbitrary Riemannian manifold is a closed subgroup of the orthogonal (ad./ he obtained a list of groups which can be (I). (2) and (3)). Hano and Ozeki [I] gave an example of a torsion-free linear connection whose restricted linear holonomy group is not of X(2. are the connection forms of 1‘ and I?. computation shows that an elemefit 8. tion given by the Christoffel symbols F]. To construct 7.. a >O). x2&= x2 and other this question was studied by Sib. fold with structure group O(k) is universal for O(k). take the linear connec. His example is as algebra of G for the answer to be affirmative when G is a simply follows: On R2 with coordinates (xl. his proof can be also found manifolds with non-parallel curvature tensor. A more general situation than the direct product has consider the affine connection which makes every left invariant been studied by’Reinhart 111. The answer-is yes. For example. Cartan’s book [8].uya [ 11. Then the question is whether every strongly metric property. The linear holo. nomy group is completely reducible (that is. closed in the general linear group. Even for a simple group. it is compact (Theorem 4. we followed the with the list of connected orthogonal groups acting transitively on appendix of de Rham [ 11. His list coincides in E. The local decomposition of a Riemannian manifold (Proposi.1 (the equivalence of curvature tensor. counterexamples. the tangent space is For a solvable group G. R) lies on some l-parameter subgroup if and only if 4 . it does not hold-3 in general space. The connection F0 is therefore called a universal connection for G as Example 4. curvature tensor is~ transitive on the unit sphere in the tangent For a complete affine connection. restricted linear lholonomy groups of irreducible Riemannian Thearem 4.2 of Chapter II). if G linear connection with the following property. the answer is not affirmative in general. Nagano. this is no longer true in general. of an arbitrary Riemannian manifold is not in general compact. By a careful examina$ion of the . where OJ and w. the existence and the uniqueness of such a It is worthwhile noting that even the local decomposition is a connection is easy to see. x2). Ozeki gave an exampleof a torsion-free element of G is on a l-parameter subgroup.2 of Chapter IV) was are precisely the l-parameter groups of G. For a compact flat Rieman- (and n). vector field parallel. dote 2. In $4 of Chapter IV. Simons [l] proved directly that the linear holonomy Chapter IV was given as the definition of completeness by group of an irreducible Riemannian manifold with non-parallel Ehresmann [1.2 goes back to Hilbert [l] ..1 of spheres.also . The holonomy group is ((l i).3 of Chapter V shows. w = f *q. For the Recently.treated by be studied in Volume II. consider an affine connection on a connected tion 5. 21.290 FOUNDATIONS OF DIFFERENTIAL GEOMETRY NOTES 291 (that is. nian manifold.2 and Theorem 4. For some linear real algebraic groups.

= 0. R) and is affirmative for SL(2. &mm + 4. Rd. The simplest proof is pr&ahly by means of the .’ ’ x + x + 1 is an affine transformation as it sends ‘the metric 8 dxa into e p &a... the Riemannian curvature tensor R and the either u + d > -2 or a = d = -l and 6 = c = 0. 4) = wd + Qhs) uniform structure (compatible with the topology) (Dieudonne.. 4) = MAll) = 2w.a manifold it4. ‘fhe . Proof. constant provided that n = dim M 2 3. and that and finally using the following formulas not every pair of points can be connected by a geode&&It is not known whether every pair of poin’ts of a compact. 1. XJ . If M is a 3-dimensional Einsbin man@d.(M).. Xs. = &m fold with a complete affine connection can be joined by a geodesic. . Markus [1] (n .ws. QED: defined by the metric aY = 8 d.(y). shows. k and 1 fold such that the (Riemannian) connection is complete.) -denotes the sectional curvature determined by the plane PC.$&. has a PROPOSII%ON 2.2) A.. R). then K(p) < 0 for all planesp in T.(M). . q QED. Let g‘.e. R) modulo its center. Consider the Riemannian connection on the real line R1 Hence 1 is a constant.(M) such that p is spanned by Xi. Nom&u and Ozeki [I] proved that.9.(M). gik& . 1.a. Analogous to the theorem of Schur (Theorem 2. (i # j) so thatp.+. It is not complete as the length of the where J is a constarrf. is a space of constant cuwature. This furnishes anon-complete. i. there exists a positive functionfon M such that f-g is a complete Riemannian metric. we have) K(p). Smith [ 1] also constructed a Lorentx metric. Let p be any plane in T. if 0 < c < S(X. for every element A of $42. on a 2dimenskmal mani- Multiplying by g’” and ggl. Xs be an of a non-complete affine connection on a simply connected compact orthonormsil basis for T. summing with respect to i. . an indefinite Riemannia n metric. Similarly. is a function on M.-A (possibly system x1. [I])..RlfLIs -R. connected mani. Wl. Q. . respectively. Hence we have Note 3. compact. it is flat. If S = Ag. xR. Ricci tensor and scalar curvature y&t Xl) + w* %> . is necessarily a. we obtain An affine connection on a compact manifold is not necessarily complete as the following example of Auslander and.3 of both) lies on a ‘l-parameter subgroup.t.= -&. a circle..equator is a geodesic.. X8) = mbl) + W**).j~ + Rijmw = 0 ..e.. X. An example Proof. given a Riemannian metric g wd. Since . we have the following classical result. if 2c < 5+(X4 X) < c c 0 for all tensor g and Ricci tensor S. Then Bianchi’s second identity (Theorem 5. = krs. Xl) = mh) + Wld It is known that every met&able space admits a complete w. j. with respect to a local coordinate that.R. then K(p) > 0 for THEOREM 1. th#n if non-complete flat afiine connection. y i.metric tensor g. system in R1. Let M be a connected Riemannian manifold with metric all planes p in T. The above formula implies also that. NOTES 293 292 FOUND&IONS OF DWFERENTIAL OEOMETRY c classical tensor calculus.. the equator of a sphere and’ extending it on the whole sphere so Then that the.translation The following proposition is due to Schouten and Struik [ 13.V). homogeneous affineiy connected manifold. 0n. Thus the real line module 1. = &A. and Ri. This means Ricci tensor S. . where K(p.(M) and let Xi. geodesic from x = 0 to x = -co is equal to 2. manifold is obtained by defining the above affine connection on Let& be the plane spanned by Xi and X. where x is the natural coordinate A Riemannian manifold is called an Einstein manifold if S = Ag... either A or . .2 of Chapter . then 1 unit vectors X l T. be the components of the . Thus the answer to our Chapter III) is expressed by question is negative for SL(2. X) <‘$c for all ‘unit vectors X 6 7’. Remurk. where 3..

AOZPx = fix for x c R”+r. we have orthonormal basis and is called the scalar curvature at x. and every element of G can obtained be written uniquely as A’S”.1. We thus where 1 < t < q and tP = 1 mod q. (b) f(k) ‘= 1 mod Q.1 and 0 2 k 5 p . we have (cf. .:. we have T(Cy:. complete Riemannian 0 = Cf’. we have In the case where n is even. The Jc) (AiS)P = A”.1. there exists an. 2771) which is obviously a contradiction. Assuming that G is non-abelian. most general result in this direction is due to Zassenhaus /2]. of A are & 1. which . let Gi be the cyclic sub- freeb on s”. Then. Hu [l. the determination of these groups G 0 = C.1. Hence. define an integer f(k) by f(k) = T HEOREM 1. Then. Thus. consider the case G is of order pa. p . ‘= Ajs and. Hall 11. C. . the scalar curvature is given by xi. Then G is either cyclic % 2 = )p) = #4) x order of G (if n is even). Let G be a Jinite subgroup of O(n + 1) which acts SAS-’ f Al.p-1. = Gj. Hence.. by Theorem 3. The case where n is odd has not been sol??d completely.. of order p (cf.. M = P/G.l . G is either cyclic or a G.’ Tp) = zr$ Ty for each y c R”+r. In terms of the components Rit and gij of S and g. G consists of either the identity I only or Z and another [I> P. and G. We prove that Ci n G. This implies k = 1 ’ direct product of two cyclic groups G. not necessarily distinct) is cyclic.. For eacIi i. respectively. (AiB)‘x = 0 for x E R”+l a n d i=O. Hall Hence. A = -I. p. and (c) follows from T HEOREM 2.. Indeed. Second. in Rn+l given by (x’)” + * * * -t (x~+~)~ = a2. Note 4. where G is a finite subgroup of O(n + 1) which acts freely on S”. be an orthonormal basis for T.(M). manifold of constant curvature l/a2. of order p.o’ A”‘Bjx = Cy:. Sk = S.yk. Since (A’S)” = A”‘@)9 = Z. the eigen-values SAS-’ = At.:o’ A’jBjx for x E Rn+r. where 0 5 i 5 q . p. let. Let x(M) denote the Euler number of M. we have Ai . Henfe we have either Gi n Gj = (I} or Gi = G j for Proof.integer k such that (A(S)& = AjS.qr are group of G’ge’nePated by A’S. Every connected.l (A’B)‘x = ET:.l. . Since A can not have any fixed point on S”. If First. that is.* giiRi. It suffices to prove that if G is order pq. Then. A’jBjx = 0 forxER”+l a n d j=l. if k = 1 mod p.. ti {Z) if i # j.. complete Riemannian manifold M of 1 + t + t2 + *** + P-l. Gi is prime numbers. Then we have (cf. Then. Then p.2 . thenG is cyclic. We then have even dimension n with constant curvature l/a2 is isometric either to the sphere S” of radius a or to the real projective space S”/{-&I). By setting T = A’B and y = x. we obtain ZT:. 491). consider the case where p < q. Clearly. the eigen- values of A are all equal to . let S and A be elements of order p and q. connected. By (c). Gr and G. we have A’S” = A’S. respectively. X. (a) f(p) = . For each integer k. (a) follows from tP = 1 mod q. .O mod q.“e’P. or non-abeiian. Since every element T # I of G is bf order let X.. 0 5 i. Spaces of constant positive curvature and hence Let M be an n-dimensional. we obtain covering manifold of M is isometric to the sphere S” of radius a crz. by setting T = A’ and y = Bjx..fW.2 of Chapter V and Theorem 7. respectively.1. 294 FOUNDATI b NS OF DIFFERENTIAL GEOMETRY NOTES 295 Going back to the general case where n = dim M is arbitrary. hence.10 of Chapter VI. j 5 q . any subgroup of G of order pq (where p mui8.. Assuming the latter.l Cj’=. is extremely simple. A and B be generators of mod p and f(k) e 1 mod q.. Cf..O 5 i I q . Hence. the universal On the other hand. 511) element A of O(n + 1) such that A2 = I. Since T SGL 4) + * * * + S(Xm Xn) is independent of the choice of has no fixed point on S”. G must be cyclic..

where 1 B C with Euclidean motions has a fixed point (cf. for n = 3. . then R”/G and R”/G’ spaces. we have Gi n N J= acting freely on Sn. results : . Let M = P/G be a homogeneous Riemannian manifold containing N.{I}. is the m x m identity matrix. ‘R”/K is a compact of constant curvature l/as.. Finally.2 of follows. the group of affi.’ .G’ are conjugate in a. .296 FOUNDATIONS OF DIFFERENTIAL GEOMETRY NOTES 297 implies i = j. . TX = &‘& TX ‘+ qx for x c transformatrons of R. Z”) E Cm. .-. Since G/K is also finite by (2). 0 5 i I q . To prove (?). tion v&such that Q?G~-’ = G’. following papers: Vincent [l].2. . q”) l ct”. any finite group of and G is a Jnitc group of matrices of the form AZ. and G is ajnite group of matrices of of R” such that IPIG is a compactflat Riemannian manifold.Hence.. * + lq”l” = as> (2).1 111 = 1 and I. Chapter V. of Chapter \7. His result may be stated as Indeed. flat Riemannian manifold. that is. 1. Let N be the normal subgroup of G generated by f. G. In fact. the proof of Theorem 7. Wolf [l] classified the homogeneous Riemannian (4)‘. K contains nothing but translations by . the case Theorem 4. manifolds of constant curvature l/as. . Flat Riemannlan manifolds On the other hand. . Conversely. See also Hel- Gp. By the same Let N be the subgroup of G consisting of pure translations. Hopf [l] I (b) For each n.. Calabi and Markus il ] and Wolf [3. Wolf 151. -if G and .Qis the jeld of quaternions) . q .1. . Let R/G and R”/G’ be two compact flat Riemannian manifolds. is obviously a contradiction. Therefore we have Biemanman manifold. Q N.. Milnor [I] partially. for every T. where p E Q with I p( = 1. we have XrrN TX = 0.‘. i. Then G can be real&d as a-group of Euclidean motions -. Iq’lZ + *-.’ The reader interested in the classification problem of elliptic (a) If G aid C’ qr+somor-hit as abstract group. of Chapter IV) and hence G has no finite subgroup. By counting the orders of N. .e. Then reasoning. the form pZ. Note 5. QED* . (where. we do not have to consider . which generators. . is referred to the are equitfaknt. .. Theorem 2 to the case vvhere G is a group of homeomorphisms Since N is of order q and since Gi is of order p. spaces of constant positive curvature.+ IPI = a2}. G. then We say that they are equival’ent. For the study of spaces covered by a homogeneous GG:’ I-. (1) and (3) have been proved in (4) of Theorem 4. * L. . In addition to (4) of In view of Theorem 1. only~ajinite number of equivalence classes of and Seifert and Threlfall [ 11. Since G acts freely on S”.. see FVolf [2]. (2) N is a maximal abelian subgroup of G. G.1 and (1) N is an abelian normal subgroup of G and.+ we see that G is a disjoint union of N..z manifolds of constant positive curvature.‘(I). gason [I]. 1. . . from the fact that G’acts freely on R”. Auslander and Kuranishi [I] proved the converse: Let G be a group with a subgroup N satisfying the’above conditions -(l). we have zrtCi TX = 0 for i = 0. Bieberbach [l] obtained the following where n is even. we have qx = 0 for each x E R”+l. G is a discrete subgroup of the group of Euclidean motions of R”. H. ii}.-.Tx = XTcN T X for x E R”+l.?h& aie.. 2I TcN TX + && TX + * ’ ’ + &.. Since K is an abelian normal subgroup (1) Zfn +‘l = 2m (but not divisible b 4): then of K. where TO(CT~N T X ) = CT& T. lz112 + * * . generalized cbrnpactfi7at Riemannian manifolds R”/G.Lemma 6 for Theorem 4. then .G has no finite subgroup.. let H be any abelian subgroup of G THEOREM 3. we have Let M = R”/G be a compact flat Riemannian manifold. $G is ajnite group of the type described in (1) or (2). K = N. 41 studied {I} for each i. (4) follows P = ((9. . . . (2) If n + 1 = 4m. S” = {(q’.orent.2 of Chapter V.. Recently. (3) and (4). Hence. . if there exists an affine transforma- M = P/G is homogeneous. .is free on n zz rcO TX = 0. \ (3) G/N is finite.. .

Since (A. Let u l R” be a point left fixed by G*. Z). T’ = (t. Since F is finite.) be a basis of N.!) (A’. &). and translation part p. empty) is a finite set. As manifolds. . T-‘A T is unimodular. H2(K. p’) (Z. Let N’ be the subgroup of on H. for example. . The. We denote by The mapping which sends t c N into St E N is an automorphism of (A. Although (b) does not hold for non-compact fiat Riemannian manifolds. Biebcrbach 123. . to each other.O)-l(A’. consisting of reduced matrices in the sense of’Minkowski. To prove (b). Each G determines a group manifolds are equivalent if and only if they are homeomorphic extension 0-t N-+G-+K+l. p) Q G.1 is the element of H2(K’. there are only a finite number of for all (A. theorem of . H. Then (T. .P) an affine transformation of R’” with linear part A and N. Most of the results for flat Riemannian manifolds cannot lattice points of R”. . N) ‘s finite.p)(T.).= G’ are isomorphic. where Z is ingtoanelementO+N--+G-K+10fH2(K. QED. number of mutually non-isomorphic groups G such that R”/G are Note that (a) implies that two compact flat Riemannian compact flat Riemannian manifolds.. Thus our problem is reduced to the following (Z. let Xi E H. let (A’. .T’-lp’) . Let H.N).. A’$) corresponds to (Z.thenGand the identity matrix and ti 6 R”.. Z) be such that S[X. This theorem of Jordan follows from the theory of Minkowski- singular and integral together with its inverse. Then (a{) = GL(n. . ti). T-lp . t. t:). 7’~) = (T’. N). is finite. t. we have seen in the proof of (a). if we set U) s~cuz. ti) (A. This compietes the proof of (a). as follows: G’ corresponding to N by rhe isomorphism G’ w G. p) B G}.tj.first propetty of R implies that any finite subgroup K of Then G* is a group which contains no pure translations and hence GL(n. p) E G. it suffices to show that there are only a finite Bieberbach and Schur [I] and Siegel [l]. Then Since (A’. (Z. Z). l of degree n which are conjugate in ’ the group GL(n. Given such a finite group X. H2(K. (Z. . S[R]‘n R =: non- G* = { ( T-IAT. = I: ACR *AA). Z). then K = G/N is given by unimodular be generalized to flat affine connections. and ScGL(n. Let K and K’ be two finite groups of unimodc!ar matrices Auslander [I]. the dimension (Bieberbach [3]). . We shall sketch here an outline of the proof. matrices. Let R be the subs& of. . see Hantzche and Wendt [ 1] and Nowacki [ 11. N) correspond- translations and let (Z. set of group extensions 0 -+N-+G-K+1isgivenbyH2(K.N). Z) is conjugate to a subgroup of GL(n. Let T be an (n x n)-matrix whose There are &1. Z) contained in F. In other words. Z) . p) c G by the isomorphism G ’ w G. p)-1 . there are only a finite number of homeo- where the’ finite group K = G/N acts linearly on N when N is morphism classes of complete flat Rkmannian manifolds for each considered as a subgroup of R”. (A. Z) of all . u$. 0) S-‘KS c F. conjugate classes of finite subgroups of GL(n. then T-IA’ T’ = T-1 A T.. Tu)-‘(A. p) (I. t. be a fixed point of K (for instance. that is.) Siegel. t{) correspond to (Z. Let S E GL(n. Z). Denote ‘SXS by S[X]. p’) c G’ be the element corresponding to matrices of degree n. For the classification of 3-dimensional complete flat Riemannian Since K is finite and N is finitely generated. Then we have Indeed. p’)-’ = (Z.] E R.wl = 4. N) = H2(K’. Let N be the subgroup of G consisting of pure ifO-+N-+G’-+K’ . Hence S induces an isomorphism. Z) acts properly discontinuously (A. we can write Ati = C&i a3. Then GL(n. if we identify N with c ’ e integral . NOTES 299 298 FOUNDATIONS OF DIFFERENTIAL GEOMETRY unimoduhr matrices so that SKS-1 = K’ for some S E GL(n. a jnite number of conjugate classes ofjnite subgroups of i-th column is given by ti. A’ti). Hence N’ is the subgroup of G’ consisting of pure translations. set X. Ati) E N for any (A. Set (ii) The set F defined by F = (S E GL(n. T = (ti . see. As references we mention Minkowski [I]. tl). Hence we have A’ti = Cy=. Then N’ is x + tSxs for XcH. (A matrix is called unimodular if it is non.Tordan : where each u{ is an integer.~‘)(T. Let (Z. normal and maximal abenan in G’. be the space of all real symmetric positive definite To prove (a).

(G/H) Although the theory of symmetric spaces.1 of Chapter II. q’) the plane in T. The origin hemispheres and their neighborhoods.1 of Chapter II. (3) follows from the fact that a$ for any with f. call this connection the canonical connection in G(G/H. to each end of the right circular cylinder S1 x [0. where Si is the unit circle. Denoting by the same letter onto M’. We have obviously [h. y’. let M be a compact Riemannian manifold (r the involutive automorphism of the Lie algebra g induced obtained by attaching a unit hemisphere to each end of the right by u. NOTES 301 Let G be a connected Lie group with an involutive auto- Note 6.(G/H) at the origin o. H)) . let M’ be a compact m = (X @. QED. Now.e.canonical projection of G onto G/H. = exp tX and let xt F n(a. We shall then say that G/H is a symmetric general. u # 1). we define a mapping f of G into the Riemannian symmetric spaces. Letfi T. (1) follows easily from m” = m. the linear transformation of T. Parallel displacement of curvature morphism (T (u” = 1. and then X” = X} coincides with the subalgebra corresponding to H and smoothing out the corners. by a).(M) be a fixed orthogonal r in G(G/H.300 \ FOUNDATIONS OF DIFFERENTIAL GEOMETRY .(M) (resp.c broberties: -1 1 transformation. ever. Symmetric spaces linear isotropy h. which generalizes an invariant connection F determined by the subspace m.. ml = b. Y] c h? through parallel displacement. Let H be a closed subgroup which Let M and M’ be Riemannian manifolds and rp: M + M’ a lies between the (closed) subgroup of all fixed points of u and its diffeomorphism which preserves the curvature tensor fields. Hicks [I] obtained a similar result in the case of affine The projection r gives a linear isomorphism of the horizontal connection. denoting G/H by M. i.(M) -+ T. . m] C m and Riemannian manifold obtained by attaching a unit hemisphere h.. subspace m at e of F onto the tangent space T. circular cylinder S1 x [O.4 of Chapter VI. For a symmetric space G/H. By Theorem 11. fixed h Q His the horizontal lift through h of the curve x1. .. fixed point of A/‘. How. tensor fields. (2) is contained in Theo- a unique isometry F: M -+ M’ whose differential at x coincides rem 11. coincides with K(q) is equal to the sectional curvature K’(q’) for all srmply the left translation 4 + a. induced by the transformation h of G/H which fixes o. (3) For any XE m. g. p preserves the curvature. p’) be a plane in T. let a.’ where h = {Xc 9. h E H. the symmetry around o.(1/2)[X. T. H) has the foL1owin. Ifh E H. we have (t:= m + $ (direct sum). X” = ‘--X}. we shall give here its definition and basrc propertres. x an arbitrarily fixed point of M and x’ an arbitrarily . broken geodesics T and all planes p in T. the correspondence being given by f (2) The curvature form is. then ad (h) on m corresponds by this isomorphism to the IVote 7.4 of Chapter VI in the Riemannian case.(M).. his result generalizes Theorem 7. then there exists Proof.(M’)) and q (resp. -11’ from x’ to a point. T H E O R E M 1. Let a0 be an arbitrarily Volume II. this does not imply the existence of an isometry of M homogeneous space (defined. We-call gn Al and M’ are flat.(M)) to m.h. Let p (resp. of G/H such that c*(rx) = n(x”) for every x E G. Assume that p’ corresponds top byf. Since the cylinder parts of o = -r(e) of G/H is then an-isolated fixed point of a. given by R(. simply connected Riemannian 3 manifolds. H) admits Ambrose [l] obtained the following result. in particular. In identity component. M and ill’ cannot be isometric with each other. 7’). the canonical connection . We Theorem 7.. m bundle of frames L(M) over M as follows..LJ. PC diffeomorphism which induces an isometry on the attached where 7r is the.(M) where X and Y are arbitrary left invariant vector Jields belonging (resp.2] and then The automorphism a of G also induces an involutive diffco- smoothing out the corners in the same way. If the sectional curvature The parallel displacement of thejihre H along the curve x. T.) = at(o). H). Let 7 be a simply broken geodesic -of M from x (1) r & invariant by -the automorphism a of G (which is a bundle to a pointy and T’ the corresponding simply broken geodesic of automorphism of G(G/H. p’) by parallel displacement along T (resp.. For instance. Let M and M’ be complete. Y) = . Similarly. obtained from p (resp. will be taken up *in detarl. 11. Let pl: M + M’ be a morphism a. the bundle G(G/H.

A linear connection r on M is local5 symmetric if and on& if T = 0 and VR = 0. it is the Riemannian connection by the uniqueness for h E H. In particular.0. Remark. let a.e. It is easy to see thatfis a morphism A E SO(n + 1) -t&U-r E SO(n + 1) where S is the bundle homomorphism of G into L(M) corresponding to the homomorphism 91 of Hinto GL(n. Ae. be the G/H has the following properties : standard orthonormal basis in Rn+l.. . Proof.2.. Hence T = 0 and VR = 0 on hf. equivalently. since the torsion tensor to be localb symmetric at x E M.. frame at the point Ae.. f(h) = h . let (T be the involutive auto- induced by h with respect to the basis u. The canonical linear connection on a symmetric space with the unit sphere S” in R”+l. since there exists X E m such that X * -X in T.. if it is locally xt ii (3) has the initial tangent vector X. Y) = (T(XOo. where v(h) E GL(n. can be considered as an orthonormal the linear isotropy group I?. VR =. at o. Thus T = 0 at o and hence everywhere. of degree n.(o). matrix which represents the. R) is the (Theorem 2. .(M). i. xt is a geodesic. let e. This justifies the name of canonical linear connection. For any a E G. .. (2) and (3) follow from the corresponding prop. which can be identified with a certain basis of m. For any such metric g.(M). Ae. In fact. then system with origin x. (l).l)/SO(n) onto the bundle of parallel displacement of vectors along xt is the same as‘ the transformation orthonormal frames over S”. ES” induces a diffeomorphism of SO(n + l)/SO(n) onto S”.. linear transformation of T. e. . torsion. u0 = u. of S”.302 FOUNDATIONS OF DIFFERENTIAL GEOMETRY NOTES 303 fixed frame X1.. if there exists an involutive afine field T is invariant by a.l) over SO(n -t. the canonical connection I’ 1 in G(M. The (2) The-restricted homogeneous holonomy group of Y at o is contained in set of vectors Ae r.(M) Exampte. *v(h). Chapter II).1. we have T(X.T ( X .4 of Chapter VI. The con\(erse follows the Riemannian connection coincides with I’. The bundle SO(n +. Y ) and hence T(X. Zi = 0 for any X. shall write SO(n) for Ho with this understanding. . (4) follows from (1) . . Y) = 0 for any isolated fixed point. A linear connection I’ on a differentiable manifold iz2 is said erties in Theorem 1. if K is a G-invariant tensor field. This local symmetry at x. with respect to the Riemannian metric of S”’ as imbedded (5) Every G-invariant tensorjeld on G/H is parallel with respect to I’. property (1). In fact. x is 0 at x. Since F has zero sisting of the images of -Xi by the differential of a. In particular. We _ _ the canonical linear connection on G/H and denote still by F. then any tensor field of type Let G/H be a symmetric space with compact H. . . must be X and Y in T. This gives an isomorphism of the (3) For any XE m. E T. . We say that r is locally symmetric. The symmetric homogeneous space SO(n + l)/. SO(n + l)/SO( n) coincides with the Riemannian connection of Sn (4) The torsion tensorjeld is 0. If I’ is locally symmetric. if H contains no non-trivial invariant subgroup of G)..f(o) is the frame at u(o) con.xt = 7~(uJ = a. In particular. In G = SO(n + I).. Proof. e. r is the unique linear connection on G/H which has T HEOREM 3 . of Chapter II. . QED. since it induces the linear transformation 0. X. . which we shall call u consists of all matrices of the form where B l SO(n) . There exists a (r. = exp tX and ‘. The mapping A 6 SO(n + 1) + (1) Y is invariant by G as well as the symmetry u.. ‘I’& identity component Ho of the subgroup H of fixed points of By Proposition 6. (5) of the form (xi) --+ (-xi) with respect to any normal coordinate follows from (3). around o. tensor field g is parallel with respect to I’ by (5). S) with odd r + s which is invariant by the local svmmetry at G-invariant Riemannian metric on G/H. R). then f and v are isomorphisms.. H) induces a connection in L(M). if it exists. the curvature tensorjeld R is parallel.Y ) = .T ( -X. symmetric at every point x of M. . The canonical linear connection on by a. In fact. . If G is effective on G/H (or matrix of the form with identity matrix I. submanifold of -Rn+i. .(M). the metric from Theorem 7. QED.!@=(n) is naturally diffeomorphic TH E O R E M 2. Y”o))“o = transformation of an open neighborhood U of x which has x as an . .

Let V R = R @ a’and let Rj. If dim M 2 3 and if the Note 8. letf: L(llif) * III. _ .. the curvature extensively by Cartan [ 71.l (RjiJ2 is constant on contained in the linear isotropy group at every point. Ricci tensor field..~. A non-zero tensor field K of type (r. Let J manifold M.~ = Cj. then M is each fibre of O(M). Nomizu [4. Let G = (UO(M).l). Let a. we have Cj. By Bianchi’s second identity (Theorem 5. E L(M).(R”) be the mapping which corresponds to a given tensorjeld K of OPe (r. Hence 9 itself is constant on each fibre of P(uJ. see Nomizu [2] and Kobayabhi [3].. simply connected and complete.g. curvature tensor R.constant and g is the metric tensor (cf. s). for any complete Riemannian negative. The components of R and a with .. Y) . As a special case..3 of Chapter T HEOREM 1. the condition VmR = 0 for some m -> 1 implies be the function on M defined’ by’ A(X) = l/v(u). a.( is a constant function by curvature . 41 when M n(u) 4 M. where c is a.. then . By Corollary form and X and Y are horizontal vectors at u. Then AR is a parallel tensor field.. Define an involutive automorphism of G by a” = o0 0 a o IT. We apply Theorem 1 to R. Then H lies between the subgroup of all fixed & &A application ofTheorem 1.9 of Chapter is spanned by all elements of the form a. and a. Then the Lie algebra of its linear holonomy group Y (u. for the holonomy bundle P(uo) through any u. For a Riemannian manifold M with recurrent curvature The Riemannian symmetric spaces were introduced and studied tensor whose restricted linear holonomy group is irreducible. Wong obtained M is connected. The first assertion follows from Corollary 7. In the notation of $5 of Chapter III. Then K is recurrent if and only if. rf Chapter II). hbwh3 The Riemannian versions of Theorems 3 and 4 are obvious.-. where aj = Zn. xl .respect to a local coordinate system following result is due to Wong [ 11.2.t = 0.) Proof. Let r be a localb symmetric linear connection on M. be the local symmetry at a point o of M.) They remarked later that the assumption of com. Since Et. Simons [l] has a similar theorem. see also Note 3).k. QED. Then M = THEOREM 2. where 0 is the curvature VI. Since Q locally symmetric. we have 6. s) on M is said to be vanishes identically. we _ have T. 1. Wo) * . p.(R”)-vaiued proved the converse of (2) of Theorem 2 that if the restricted linear function on O(M) which corresponds to the curvature tensor field holonomy group of a complete Riemannian manifold M is 8. which is involutive.1 of ' THEOREM 4. f(u) = v(u) -f (uo) for u Q P(u& Hence.. Let R&t be the components of the T. Since the Ricci zero on P(uo) such that tensor vanishes identically.6] Proof. we shall sketch the proof of the elements of (T and its identity component. For the canonical linear connection on tensor is necessarily parallel provided that dim M 1 3. where x = VR = 0. If we denote by S the is compact. NOTES 305 304 FOUNDATIONS OF DIFFERENTIAL GEOMETRY -+- Using this result and the holonomy theorem (Theorem 8. Theorem 1 of Appendix 5). R is parallel. can be extended to an affine transformation of M onto itself dim Y(uJ I= &z(n . Linear connections with recurrent Ricci tensor S is non&trivial. '-'THEOREM 3.~gj”‘Rj~~. M implies that 1s = c . Let M be an n-dimensional manifold with a linear connection Next we shall consider the case where the Ricci tensor 5’ I’.gjnlRj. Since JR is parallel and since 1 is a constant. there exists a dzjkentiable function q(u) Wzth no Multiply by gJm and sum with respect to j and m. In particular. . then the group ‘II(M) of all afine transformations is transitive on M. (This was known by Lichnerowicz [3. K is parallel if and only if f (u) is constant on EjRjktaj = 0. be the recurrent’ if there exists a l-form a such that VK = K 8 a.gjmam..Theorem 1 of Note 3.vanishes on P(uo). The irreducibility of pleteness is not necessary.. it is either always positive or always Nomizu and Ozeki {3] proved that. q+ is constanton each fibre of P(uo). never .*a*. then fS is also parallel. Let r be a linear ‘connection on M with recurrent G/H is a symmetric space for which r is the canonical linear connection.. symmetric G/H.

lI Myers and Steenrod [I] proved that the group of all isometrics of a .. autbmorphisms o f a n ahsolutc parallrlism i s ..III .‘i11m *1111 f ‘t ~5IIIlai a.I. . g 1 9’. $1 la +t &. Hy virtue of l’hco~cm 2. Propositic n 2.d by $1’. iinbedding it into the manifold.ic !yc.i. II’. then there are only three possibilities: (1) Y7 R = 0 and stsucture (such that the mapping G x Al --. Let x be an Riemannian manifold is a Lie group. by the Holonomy Theorem group of diffi~rcntinblr transformations \vhich is IOWA I. I.Y. 21 pro\.n +ul’ (.\rily fisrcl point of hl and let X and _I’ be any vectors at x.IIo\\ II IO 1~. If c is the Note 9. On the other hand. that the group of all complcs :trl.m:ln- analytic transformations of a bounded domainin C” is a Lie grotip.(A4) -+ 7’. Riemannian) manifold ill.c (I I. (‘l’hcorcm 8. If the curvature tensor of M x M’ is T H E O R E M .t dimensional Riemannian manifold.IY~~IIII~)I~OII f<(. the group of Theorem 2.I J.rl)l~lit~tl 10 I 11.[i].~ .c.11 I III‘ \\‘ong [ l]).6.5 of Chapter \‘I). This method can 1)~ .IS stl:dictl in. then G nhril.I.21 (cf. Palais [I]. infinitesimal isometrics) Given a differentiable manifold M. Chapt(*r 1. then the formations of a compact complcs manifold is a con~plcs 1 . The first result of this nature was B-y . this assumption was lalcr rcnio\-(~1 bv t 1. group. recurrent.ip grolij.lllill )I(1 is spanned by the set of all endomorphisms of 7’. (2) R =. every non-flat 2-dimensional Riemannian connected manifold .\ol’llizu 1I’L.3 (r&. If M is a complete R~emanniart manifold with recurrent structure will be discussed in 1’olumc COROLLARY.(. If gomcry [ I. then’the universal covering manzjiild II? of M is either a A global theory of Lie transformation groups \V. bearing on us.14) maps I’ into the they made use of a general theorem concerning 3 10~111~~ C‘OIIIIJ~IC I zero \‘cctor. We shall here state one theorem \\hich has. nian~~nianifdld M JV. The clutomorphism group qf group of all affine transformations (rcsp.3) of Chapter VI.) We have the following applications of this result. 61 prcrvccl that the. difl’erentiable transformations of M leaving a certain geometric Theorem 3. the Lie algebra of the linear holonomy grou Y(x) group ofall affine transformations ofan aflincly connc~I~~cI Ill. Kobayashi (1.6 of CA. Let M and M’ be group of transformations which belong to the giv1.. the group of all differeniiable which . end QED. it follows that !I is fillitc-dirnension~l.I c1irc. Use the decomposition theorem of de Rham (Theorem formations acting on a differentiable manifold :I/.are globally integrable (note that if Al’is complete.tbl~* trails- Proof.111() (I I I( l and hcncc is zero by the irreducibility of Y(x).sct 6. ture tensors.(. isometric.& the fbrm of’l’hcorcm .\1’1 of . Jf !I is finite-dimensiona/. respectively. It follows that I/ is invariant by Y’(x) Of C@r$eteness.4 of Chapter VI).iptcr 111 . these infinitesimal transfor’mations are . sti-ucture is often a Lie group.2 of Chapter IV) and Theorem 3 above together with the of all vector fields X on ‘11 which gcneratc a global I-paramcrcr following fact which can be verified easily. G is ri Lie. 1’ E T.4. M i s diffb0/linhle) y7 R’ = 0.:3r. .~il 1. R is parallel. The Lie algeb?a of G is naturai[~~ isomorphic rc:itfi $1. I~ochnc*r antI 1 IOIII- arl)itr.c.1 of Chapter II) and Theorem 1 of this Note e also valid.: manifold is of recurrent curvature if the.(ni). not vanish anywhere.ttc. Since x is an arbitrary point of izf.ct manifolds with linear connections and let R and R’ be their curva. 9 be the Lie subalgebra of the Lie algebra X(&Z) gcnc*r.’ absolute parallelism of the bundle of frames 1. : given by H.l:j.sectional curvature does Thcorcm 1.ti FOI!NDATIONS OF DIFFERENTIAL GEOI\lETRY NOTES 307 ‘I’llis ccluation has the following geometric implication. Let !I’ bc thc*. VR Motitioto.I Ili~t~~l\ On the other E’) : 7’.tf~~t. Automorphisms of a complcs structure and .sj of’an afJincly connected (resp..!!IS- \VI’ clc*notc by I’ the vector at x with components &(x).: linc’ar translbrmation R(X. (3) VR # 0 and R’ = 0. However. 1’) with X. Cartan [l] who proved that the group of all complex The Lit algebra i(:\Z) 0f. 0 and VR’ # 0.I KiillI(. then $1’ is the set of all n geometric structure infinitesimal affine transformations (rcsp.. ’ isonncztrics of‘s I<ic.(A4) given by is aLiegroup ~~asfirst proved by Nomizu 111 untl~~r tllc. ‘1’11~ t1lcoI~*111 t1t.IS studic(I in sl*mnretric space or a direct product of the Euclidean space W-2 and a 2. (See also Walker [I].always globally integrable bv transformations of A4 is a very large group.the theorem above. Let G be a certain group of difl?rc~nti.i:a grort~) I!\ QED. &oul~ 01’ Sill vanishes at x. curvature tensor.

The local the projective space Sn/( f I].e. Let M be a connected. n-dimensional manifold M with an affine connection is of dimension at most point x of M. other words the connection is complete. If p is identically equal to i. p is nothing but any antipodal points x and x ’ .p at a single point of ii?.3 and 3.8 of Chapter VI. (d) subject in the book of Yano [2].10 of Chapter VI).3 of Chapter VI. then M is an ordinary ajine space zvith the natural flat a&ne connection. This ‘tmphes that conformal if L -vg = ag. we proved that the group . there exists an infinitesimal.formations with maximum dimensions L(M) leaving the canonical form and the connection form invariant (cf.3 are classical (see. it follows that go(M) is transitive on each M is of dimension at most &z(n + 1) and that if dim 3(M) = connected component of L(M).manifolds and affine connections admitting very Proof. Theorems 2. Finally. W e see easily that if M is isometric to if 0 is a constant function. The reader will find references on thk covering space i@ of M is isometric to one of (a). By Theorem 4. Riemanm&. An infinitesimal transformation X of M is said to be 3 of fi = S” which vanishes only at x and SC’. If M is flat. iVote 11. (b) and. then M is isometric to one of M is an ordinary affine space. M is simply connected in case the is a positive function on A4. If $f is isometric to a sphere S” for thetic transformation. Conformal transformations of $$(i + 1) = dim 3(M) $ dim 3(M) ~-&. Every mfiutewmal i&&try ‘x of M induces an infinitesjmal isometry X of M. n-dimensional Riemannian manifold as $ n -= dim L(M). (b) An n-dimensional sphere S”. v is a homo- curvature is nonpositive. Proof. From the fact that a(M) acts In Theorem 3. then i(M) is naturally isomorphic with i(x). for instance (c) An n-dimensional real projective space S”/{+I}.3’ of Chapter VI. NOTES 309 308 . (a) An n-dimensional Euclidean space R". then 3(M) is isomorphic to l-Parameter group of local transformations generated by an O(n + 1) modulo its center and hence of dimension $n(n + QED. If a is isometr5 to (a) or (d). Note 10. where 0 is i function on M.4 of Chapter VI. T HEOREM 2. Hence.2 of Chapter V ok by T HEOREM 1. It is homothetic M = Sn or M = S”/( & I>. n-dimensional Riemannian Theorem 7. From the proof of Theorem 3. to that of Theorem 2. Yano and others. Hence. we proved that the group s(M) freely: on L(M) and from the assumption that dim g(M) = of isometries of a connected. Eisenhart r 111. (d) An n-dimensional. infinitesimal transformation X is conformal if and only if X is iconformal. Theorem 7. a Riemannian mun$old ’ tesimal isometry X of M. isometry an isometry. then Let M be a Riemannian manifold with metric tensor g A there exists an infinitesimal isometry x of M which vanishes only transformation p of M‘is said to be conformal if p*g = pg. Every element of ‘u(M) induces a transformation of . we see large groups of automorphisms have been studied by Egorov that M is homogeneous and hence is complete. 1. above (cf. We prove simply connected and analytic together with the metric. QED. If dim 3(M) = $z(n + l). $1 of Chapter VI). and it is isometric if 0 = 0. then the connection is curvature tensor field and its covariant differentials.n(p + l). simpb connected hyperbolic space. We shall horizontal vector field on L(M) is complete* the proof is similar outline the proof of the following theorem. In Theorem 2. then M is a space of constant curvature. the universal covering space ~2 of manijold.3 of Chapter VI. . where x is an arbitrary point. The universal Wang. Groups of isometries and afine trans. 1). whichrmphes that every infinitesimal isometry 8 of M IS induced by an mfim. Th’is ’ implies that every standard Jp(n + l). whe. a certain Lie algebra i(x) is constructed by using-the n2 + n and that if dim a(M) = n2 + n.FOUNDATIoNS OF DIFFERENTIAL GEOMETRY \ B(M) of affine transformations of a connected. If p ’is a constant function. If dim X(M) = n2 + n. the fact that fi = M can the following spaces of constant curvature: be proved in the same way as Theorem 1 above.

I 3 for which CO(M) # 3”(M). . Nagano [I] made use of a result of Tanaka [l]. For the largest connected group (.1..* * + (Yy .310 FOUNDATIONS OF DIFFERENTIAY.. dimension n.. system of the real projective space P.. P. conformal transformations.I1 is compact.+l ontb itself.“(M) = a( n . wherep is the natural projection from Rnt2 . then M is isometric to a sphere.+l. Then (1) If .0 imply that the Lie algebra of infini. 'I'IIEOREM 2. considered as a transforma- (3) If Al is a complete Riemannian manifotd of dimension n 4 3 with tion group acting cn M. *o =-# +Y"+% xl=yl.(0) onto P. Indeed. xn+l be a homogeneous coordinate dimension at most -$(n + 1) (n F 2).X0(M) of Let h be the Riemannian metric on P.2) (cf. Then.. Tl& integrability M be the n-dimensional sphere in R” 1-l defined bv (J*)~ $- conditions of L. p.. Provided n 2 3. We imbed M into P71+1 by ‘means of tesimal conformal transformation X is of dimension at most the mapping d.-p -: n (Goldberg and Kobayashi [l]). I@ M be a Ricmannian manifold-with metric tensor g which admits a l-parameter group of isometries.(M) (xl)” + ..3 (Goldberg and Kobayashi [2]). G is a group of conformal transformations parallel Ricci tensor. * Then G maps the image of M in provided n 3-. then M is isometric to a sphere + (x”)2 . Let A4 be a connected n-dimensional Riemannian manifbld for which Co(M) # Y(54). The case n = 2 is exceptional in most of the problems concer- positirle scalar curvature (Yano [2.. In $3 of Chapter 1’1 we showed that.. Eisenhart [l. the group of conformal trans. the largest connected group ‘?lO(M) of affine trans- formations of . (other than spheres) for which &O(M) # DO(A4). To show that dim 6. complex manifold of complex dimension 1 with a local coordinate (3) is an improvement of the result of Nagano and Yano [l] to system z = x + ZJJ.2x?F+l = 0. GEOMETRY NOTES 311 THEOREM 1.A4 coincides with the largest connected group . Then.2xOx”+i invariant. The group of confohal transformations of a connected.$ l)(a +2)forasphereMof . Let p be a positive functioneon M which is not invariant by this l-parameter group of isometnes. Let g be a Riemannian metric on M which is the effect that if M is a.7.IS *isometric.e =y". . 1341). where f is a positive function on M. with respect to the new metric pg. It is easy to verify that. By the 1 theorem of Palais cited in Note . of dimension &(n + 1) (n + 2): (.lf is compact and homogeneous. * Let G be the group of linear 1 . Let M be a p. Let This can be proved along the following line. .+l of dimension n -i. for almost all Riemannian The image of M in P.+l is given by manifolds M.9. ning conformal transformations for the following reason. . Then every complex analytic On the other hand. this group is a lrparameter ‘group of non-isometric. . (2) If . Let x0. we have the following several results in the same direction. transformations of Rn+2 leaving the quadratic form (xi)” + . for instance. 2851). Xn+l =-l&l -y"+l)~ formations is a Lit transformation group. .g = a.21. .I) 34 cannot be a compact Riemannian mantfold with constant non. then . it is easy to construct Riemannian manifolds transformation of M is conformal. * . there is no harmonic p-form of constant length for the imbedding M -+ P. p. + (JJ”+~)~ = 1.24 is isometric to a sphere (Nagano [ 11) .+l given by conformal transformations of M. of isometries of . x1. f(dx2 + &2) =f dz dz. we imbed M into the real projective space of di- n-dimensional Riemannian manifold M is a Lie transformation group of mension n + 1.complete Einstein space of dimension of the form .efiiled by i(n + 1) (n i. 2791 and Lichnerowicz [3.

.. ISi.j.Y’ in R” (X. rcspec- tively.(M) : tangent space of Al at x S(M) : algebra of differentiable functions on . C.2f 2(M) : algebra of tensor fields on Al B(M) : algebra of differential forms on AI T(M) : tangent bundle of M L(M) : bundle of linear frames of . .. xl?’ in C”) CL@.II 313 .j. . R) : Lie algebra of A(n. 2”) (x.. C) O(n) : orthogonal group o(n): Lie algebra of O(n) U(n): unitary group u(n) : Lie algebra of U(n) TI( I’) : tensor space of type (r. . . A(n. . .) where the range of indices is generally clear from the context. 2.u. . M denotes an n-dimensional differentiable manifold. R) : general linear group acting on R” gl(n. Sl-XMhRY O F B A S I C NOTATIOSS We summarize only those basic notations which are used most frequently throughout the book. . 7’. stand for the summation taken o\er i or i.. 1. x”) C”: vector space of n-tuples of complex numbers (zl. R) : group of affine transformations of A” .V T(V) : tensor algebra over V A”: space R” regarded as an affine space . C) : general linear group acting on C” gl(n.. . s) over a vector space .. Rn: vector space of n-tuples of real numbers (xi.y) : standard inner product C. . .\. C) : Lie algebra of CL(n. R) 3. R and C denote the real and complex number fields. R) GL(n. . R) : Lie algebra of G&(n._ a(n. etc.A I(M) : Lie algebra of veltor fields on .

rn c~trv. . 593610. . group of Al affinc transformations . of .~443. G) : principal fibre bundle over Al with structure group [1] Examples of locally affine spaces. 0 ‘1 t 5 1. M.: tangent vector of a curve x.5 (I!153 . Interscicnce. Y’(x) : linear holonomy group at x E M [1] Sur la geometric differentielle des G-structures. of hlath. II. For a Lie group G. 4.: covariant differentiation with respect to a vector (field) BIEBERBACH. . 52 (1946\. Sot. T: torsion ‘tensor field (with components Ti. 151-270. M. Go denotes the identity component and 9 Axrm. New l-ark. 72 (3912). Ann. Bull.: Lie differentiation with respect to a vector field X [lJ A thcorcm on holonomy. 207-216. R. S. France 83 (19X). G AUSLANDER. 0 = (@j) : torsion form [l] Sur les groupes d’holonomie des varietts a connexion alline et des I’jk : Christoffel’s symbols varittes riematiniennes.(M) or O(M) A(M) : bundle of affrnc frames of . 64 (1956)..: left translation by a E G ARENS. Amer.tturc. Inst. P) : bundle associated to P(M.\I.Ilath. a( .v. BIEIERRA~. Sot. IV. of Lath. (1912).41 I-415. I.\nn. Sitzungsberichte d. :\nn.428. . E: [I] Geometric . and KURANISHI. 11‘1th. tion in g AU~LANJJER. Berlin 33 i(. 62 E(M. &lath.) 70 (191 l).~. L. c? = (szf): curvature form [l] Holonomy of flat affinely connected manifolds. 337-363. Ann. ad a: inner automorphism by a E G. L. G) with fibre F (1955). 1957. . Sot. 6-l fan: the transform of a differential form m by f (1956). 5.tionen. Bocns. Math. L. Ann.\III~. L.) [Z] ijber die Minkowskische Reduktion der positiven quadratischen S: Ricci tensor field (with components Rij) Formen und die endlichc Gruppen linearcr ganzzahlige Substitu- VI(M) . L. k.: right translation by a E G [1] Topologies for homeomorphism groups. For an affine (linear) connection I’ on n/r BERGER. W. AUSLANDER. of Math. G.and MARKUS.\lgebra. I. R. F. 1397151. 3 1 5 . A*: fundamental vector field corresp b ding to A E g [l] On the holonomy group of locally -Euclidean spaces.21) : Lir algebra of all infinitesimal affmr transformations [I] ijber die Minkowskischc Reduktiontheoric der positiven quad- 3(AII) : group of all isomriries ratischen Formen. and SINGER. Trans. : differential of a differentiable mappingf AMBROSE. 68 (1946). Fourier Q(x) : aflinc holonomy group at x E M (Grenoble) 10 (1960). Giittingen Nachr. the Lie algebra of G.. at the point xt AMBROSE.ic algebra of all infinitesimal isometries (1928).21 c(M) : tensor bundle of type (T. Wiss. Preuss. 297-336. [1] Parallel translation of Ricmanni. L. II.. co = (mj): connection form 65 (1957). L. also adjoint representa. 400-412.314 FOUNDATIONS OF DIFFERENTIAL GEOMETRY O(M) : bundle of orthonormal frames of M (with~respect to a given Riemannian metric) 0 = (Ol): canonical l-form on /. of Math. 7.4nn. 510-535. C. :lkad. . BERNARD. R: curvature tensor field (with components Ri.+rct ‘i‘r:~t\ #:i. :\nn. P(M. L:and SCHUR. [I] \‘crtor fields and Ricci curvature. Intcrbcl(. s) of AI f. 2555259.‘ll) : I. 279-330. Math. [l] iiber die Bewegungsgruppen der Euklidischen Riume I. 7767!17. Bull.

C. 99-109. [I] Sur lcs espaces uniformes complets. #194. 01 [ 1 ] Theory of Lie Groups. [4] Sur les varietes a connexion projective. [5] La Gdometrie des Espaces de Riemann. Math.2lath. 17-88. [1] Differential geometry of liber bundles. [ 1] Sur les groupes de transformations analytiques. J. Int. R. 15-27. 70 (1869). [I] Thiorie des. Math. Nem. Sup. 47 (1946!. B. . EC.\mer. Sem. 639-6. 1946. . S. 68 (1962). J.des espaces gcneralises. [I] Groupes d’holonomie des varietes ricmanniennes. Colloque deGeomCtrie Differentielle. Pures Xppl. 2 (1923). Akad. 9 (1955). Bruxelles 8 4 (1962). 40 (1923).. C. Preuss. GOTO. 2nd edition. Abh.53. Xcad. Nat. #1252. Ann. 48 FROBENIUS. Rcinc i\ngew. s [ 1] On affine transformations of a Riemannian manifold. I. Acta Math. [I] Uber metrisch homogencn Raume. [3] Pseudo-groupcs continus infinis. et Ind. et Ind. 1946.1 7 4 . 3 7 4 . 18 (1960). Amer. DIEUDONN~. L. Weyl. C. Sot.EY. 55 (1927). [1] Sur les varietes a connexion affine et la theorie de la rclativite general- ~(’ 277-291.I. Paris. CHERN. E. (19501. Bruxelles (1950).Memorial des Sciences Math. 114-134. H. Sitzungs- [7] Sur une class remarquable d’espaces de Ricmann. Hermann.. and M ONTGOMERY . Proc. GOLDBERG. S. [ 1] The Theory of Groups.. \Viss. [1] Relativistic space forms. 1935. 214-264. L. Ann. ull. HALL. 5 7 .\cacl. et Ind. varietes oh opere un groupe dc Lie. 25-26. Math. and S ASAKI . 48 (19473. l-25. 41 (1924). . [6] Les groupes d’holonomie . and LICHSEROWICZ. VAN DAN-~-ZIG. Sci. FREUDENTHAL.3 7 6 . 2 (1923). Colloque de Topologie. Sci. Colloque de Topologie. Bull. 10. [8] Lecons sur la Geomttrie des Espaces de Riemann. New York. D. applications aux groupes de Lie et aux fold. Ann. Ann. Trans.(1950). [2] The conformal transformation group of a compact homogeneous [3] La transgression dans un groupe de Lie et dans un espace fib& Riemannian manifold. 148-158. 6X 665. D. Strasbourg (1953). Math. S. Paris. [1] Locally compact groups of differentiable transformatjons. 325-412. 80 (1955). FUJIMOTQ. France. L. GODEMENT. Math. J. 378-381. for Adv. [ 1] Sur la notion d’espace complet en geometric differentielle. Paris. Paris 202 (1936). France. CARTAN. [2] Groups onanalytic manifolds. H. EISENHART . EC. [ 1] Some theorems on holonomy groups of Riemannian manifolds. S. [ 1] Uber die Transformation der homogcnen Dilferential-ausdricke B OREL. l-42. II. CHRISTOPIW. [1] Uber die unzerlegbarcn diskreten Bewegungsgruppen. is&. B. J. Hamburg 6 CALABI . 1928.\nn. (19281. HANO. 48 (19621.-Math. 167-192. S. Princeton University Press. CHEVAI. (1924). Ann. principal. Sci. Gauthiers-Villars. Proc. H. 9.S. Vol.A. [2] Les espaces a connexion conformc. Macmillan. 194Y. 397-411. 2033. A.51-1074. 2nd edition. Bull. Hermann. Colloque de Topologic. Hermann. 65966!1. Acad. Inst. 1925. (1926). 205-241. . 52 B . J. P. Sagoya Math. Paris. A. Sot. Math. Alath. [1] Riemannian Geometry. 46-70. Study. R. 42 (1941). and VAN I>ER WAERDEN. of Math. la Theorie des Groupes Continus et Ser. [I] Die Topologie der Lieschrn GruppCh al3 algcbraischrs Phanomcn I. Actualitds Sci. Sorm. 56 (1939).. ‘inn. 75 (1962). 157-169. S. 1958.. A. Amer. 1951. 1935. [2] Les connexions infinitesimales dans un espace librd diffcrentiablc. Sot. Princeton University Press.. Sci. 42 ‘EHRESMANN. (1925). Coll. E. J. 119-136. les Espaces Generalist%. [1] The conformal transformation group of a compact Riemannian mani- [2] Notion d’algebre diffcrentielle. C.. U. berichte d. Paris 234 (1952). R. Bruxelles ( 1950)) 2Y-55. 63-76. 1 7 0 . Faisceaux: Ac_tualitis Sci. . A. of Xlath. Sot. [2] Topics in Differential Geometry. Princeton. and MARKUS. 1835-1837. 1’01. Sup. Math.7 1 . G. of Math. 54 (1926). Norm. Berlin 2’3 (1911 I. 171-221. Kyoto [9] La MCthode du Rep&e Mobile. zweittm Grades. M. E. 1959. Actualites Sci. Vol. M. #lY8.316 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 317 BIBLIOGRAPHY BOCHNER. CARTAN. Sot. K&rig. [l] On the structure tensor of G-structure. Univ. and KOBA~ASHI. [3] Sur un theoreme fondamental de Xl. Congress .

Sot. 129 (1965). Nagoya Math. HI C K S . 507. Math. KO B A Y A S H I . Math. 623-630. [4] h theorem on the affine transformation group of a Riemannian Math. i . Y. Princeton. 97-100. Math. . Sci. W. Academic Press. Rome. J. Math.. 3. Rend. 632100. [5] Induced connections and imbedded Riemannian spaces. J. 10 (1956). Reine Angew. Ann. specificazione geometrica delta curvature Riemanniana. Math. L. Dunod. W. (1954). j [I] On the holonomy groups of linear connections. Comm.. KUIPER. [3] &paces a connexions affines et riemanniennes symetriques. 110 (1934). 242 [I] Holonomy and the Lie algebra in infinitesimal motions of a Ricman- 254.. R. S. Math. Ann. New York.~TSUSHIMA. [1] Note on the group of affine transformations of an a&rely connected [1] Eine Kennzeichnung der Riemannschen sowie der Hermitschen manifold. 9 (1955). B. Nagoya Math. [2] Vorlesungen tiber nicht-euklidische Geometrie. M. 528. H. KO S T A N T . W . 8 (1955). R. Leibzig. T. [2] ‘I’heorie Globale des Connexions et des Groupes d’Holonomic. (1957) . and OZEKI. 125-201. S. J. 10 [l] Le gtiupe de transformations qui laissent invariant un paralltlisme. [ 1] General Topology. H O P F. III%ois J. nian manifold. 25-37. 43 (1893). 1958. Nagoya Math J. 9 (1955). Van Nostrand. 129-146. [I] Ober den Begriff der vollstiindigen differentialgeometrischen Flachen. H. 1928. [1] Sur les surfaces localement affines. K. [l J Mtthodes de calcul differentiel absolu et leur applications. 257-278. J./ 318 FOUNDATIONS OF DIFFERENTIAL GEOMETRY BIBLIOGRAPHY 319. N. 72 (1960). . KLiNGENBERG. [ 1] Espaces homogenes kahltriens. MILNOR. 445-457. Nagoya Math. 70 (1959)) 300-309. 3 (1959). A. J. [1] Groups which act on . HERMANN.. Z. 1885. tielle. IWASAWA. 593-6 11. -hlath.\nn. [1] Nozione di parallelism0 in una varieta qualunque c conscqucntc 313-339. MINKOWSKI. T. . 129 (1905). G. T. Math. Hu. J. W. J. LEVI-CNITA. J. Ed. [ 1] Homotopy Theory. (1956). H O P F. A. Annali di Mat. 119-194. 71-8 1. Amer. Palermo 42 (1917). 1955. 220-274. Nagoya [1] On the differential geometry of foliations. and Rrccr. KI L L I N G . 80 (1955). 9 (1961). Mannigfaitigkeiten. Math Ann. Stand. 709-710. Math.[i) On the discrete sul. J. Colloque de GComCtrie Diffcrcntirlle. Math. N. Proc. LEVI-CIVITA. [2] Espaces a connexion de Cartan complete. S. and WEIVDT. Cremonesc.. J. HANO. 73-205. Ann. 79-88. Trans. and RINOW . D. Amer. 2 (1951). Ann. Helv. HILBERT. Reinc :Ingew. of Math. Paris. [3] GComCtrie des Groupes de Transformations. H. Strasbourg (1953). 43 (1957). [S] Theory of connections. J. [1] Vergleichende Betrachtungen uber neuere geometrische Forchungen. [ 1] On some types of topological groups. [1] Die nicht-euklidischen Raumformen in analytische Behandlung. 1959.542. 54 (1901). and MOMORXMOTO.~oups and homogeneous spaces of nilpotent Lie Teubner. [I] A theorem on affine conncxions. Math. 95-l 10. ~groups. i I Math. manifold. Ann. 50 (1949). Spriuger. [I] ijber das Dirichletsche Prinzip. KELLEY. LICHNEROWCZ. H. 39 ‘( 1891). Math. 39-41. ’ [ 1] DiskontStiuititsbereich fur arithmetische Aquivalenz. 558. H ANTZSCHE . 3 (1931). 30 [I] Drei dimensionale euklidische Raumformcn. Nagoya HELCZASON. Colloque de Topologie. Berlin. [I] Zum Clifford-Kleinschen Raumproblem. Strasbourg (1954). Strasbourg (19531. [2] aber die Clifford-Kleinschen Raumformen. 15-25.S” without fixed points. 171-201. 95 (1926). I HANO. [1] Some remarks on the exponential mapping for an affine connection.. F. Colloque de GComCtrie Differrn- 63-67. 1955. 79 KLEIN . 209-225. Japan Acad. M’. of Math.

107 (1933). Sot. [4] Reduction throrem for connections and its application to the problem SAITC). Nagoya [ 1] Sur certains groupes de Lie &solubles I. [I ] Einstrin spaces admitting a one-parameter group of conformal trans. 33-65. 101-103. [ 1] A Global Formulation of the Lie Theory of Transformation Groups. 104 (1930). %J”“HIX. Ann. 12 (1961). [ 1] Topologische Untersuchung der Diskontinuitatsbereiche endlicher [9] Sur les algebres de Lie de generateurs de Killing et l’homogcneite Bewegungsgruppen des dreidmensionalen sphirischen Raumes. Plead. of Math. A. Univ. 572. [2] Invariant affine connections on homogeneous spaces. NOMIX. Japan. K . Amer. Papers. ibid. Amer. K. T. 1 5 (1957). Amer. 10 (1956). Bull. d’une varidte riemannienne. ‘T. Univ. [l] @. New No~rzu. Amer. Sot. N ACANO. Nagoya Math.. 76 (1954 . 1 9 5 5 . [I] On the group of affine transformations of an affinely connected 119-132. Amer. 69 (1959). 17-25. H. 400-416. of Math. of Tokyo 7 i957). Amer.. DE RHAM. Interscience Tracts #I. Y o r k . of Math. 816-823. Indag. N. S. [2] On the differentiability of isometries. 563. a n d YANO. Math. M YERS. Math. Ann. 328-344. 537-567. L. 81-93. Sci. J. and ZIPPIN.~'. n oca an g o a existence of Killing vector fields. Math. S. Sot. Mem. PALAIS.S. :\.. [3] . Princeton University Press. [1] Existence of universal connections. Helv. D. [ 11 The group of isometrics of a Riemannian manifold. M. Nagoya Math. 10 (1956). Math. Amer.’ &HUR. Sot. Prod.of. 83 (196l). l-l 1. J. J. Japan 11 (1959). 16 (1954). 68 (1962). [II Uber den Zusammenhang der Riume konstanter Kriimmungsmasses ‘* [81 fy5. Indag. J. d . 233. Werke. SOC. “. _ II [2] On flbered Riemannian manifolds. 10-14. W . of isotropy and holonomy groups of a Riemannian manifold.\ note on infinitesimal holonomy groups. Nagoya Math. 805-807.431 Studies on Riemannian homogeneous spaces. and S TEENROD. 26 (1952). J. ’ [l] Foliated manifolds with bundle-like metrics. _ OZEKI. M.and THRELFALL.A. Publ. . [1] The existence of complete Riemannian metrics. 1 1 . B. 111-114. 69 (1959i. F: [7] Lie Groups and Differential Geometry. [ 1] Die euklidischen. 43 56.. D. . 13 (1951). [ 1] Infinitesimal holonomy groups of bundle connections. J. Sot. J. Nagoya Math. Proc. B. Math . . Interscience.gravitation. Helv. Proc.Xnn. 1956. 105-120. VIath.-249. dreidimensionalen gescholssenen und offenen NACANO. Nagoya Alath. [3] A theorem on curvature tensor fields.Math. 1-70. welche de Geometrie zugrunde liegen. and STRUIK. manifold. Sci. .. S. J. 157-168. 4 (1953). College General Educ. Math. J. P ONTRJAGIN. Educ. [I] X. 9 [l] Uber die Hypothesen.-t-forming sets of eigenvectors. Comm. and RA>IANAN. Xfath. 889-891. J. G. Math. Ges. 206-207. Raumformen. 48 J. J. 8 (1957)) [2] On the holonomy group of linear connections. 27 (1886). 200-212. 1957. 72 (19601. Math. K.: Papers College General hlath. SCHOUTEN. H. and OZEKI. Sot. Nat. Osaka Math.51 . 15 (1953). H. 14 (1962!. 40 [2] On the degree of differentiability of curves used in the defini- (1939). [ 1] Transformation Groups. 45-51. 14. I.. R. 17-27. P [ 1] The conformal transformation on a space with parallel Ricci tensor. tion of the holonomy group. :imer. (195-i). 213-216. S. ’ 74-75. ’ R I E M A N N . Math.WJ FOUNDATIONS OF DIFFERENTIAL GEOMETRY BIBLIOGRAPHY 321 M ONTGOMERY . 543-586.5 -146. Comm. . 105-123. Sci. ( 19551. b l mit den projektiven Raumen. :inn. L. N OWACKI . 43 (1921). ( 1962). 9 119551.Un thcoreme sur les groupes d’holonomie. 451-461. N.\nn. 7 (1934). II. 57-66. F~INHART.\R:\SIYIIAN. Math. #22.some properties of general manifolds relating to Einstein’s theory [6] On infinitesimal holonomy and isotropy groups. U. Math. J. of Math. formations. &III. SEIFERT. Math. ofTokyo 10 (1960). B. Proc. [I] Topological Groups. [ 1] Sur la reductibilite d’un espace de Riemann. Math..

325-341. Abh. London Math. Hclv. (3) [ 1] On an arcwise connected subgroup of a Lie group. 38-45. Fund. (1947). W. 195 1. 1!118. Univ. C. K. hlatcrir. of Math. [I] Sur les thcoremes de de Rham. (Inal. SIEGEL. Proc. c. J. [1] Einhcitcn quadratische Formen. Hamburg. J. Math. (1962). [l] Beweis eines Satzes tiber diskrete Gruppen. 3443-3445. 81-84. Comm. Acad. (1933). (1956). BIBLIOGRAPHY 323 322 FOUNDATIONS OF DIFFERENTIAL GEOMETRY SIBUYA. TANAKA. SOC. Kyoto. Berlin. [I] Note on real matrices and linear dynamical systems with periodic [I] Convex regions in the geometry of paths. K. 99 (1961). Mem. 43 (Part II) (1937). Sot. Proc. Hamburg 11 (1936). J. Math. ’ WONC. Math. Einornung drr lmjc’ktiven und kon- formcn /\uffassung. Comm. [5] Vincent’s conjecture of Clifford translations of the sphere. Math. WALKER. ZASSENHAUS.. WHITEHEAD. &it.. 26 (1952). :\. ser A. 1 (19601. W. 36 (1961). Paris 250 (1960). 209-239. H. 13-14. -. H. L. . H. SIMONS. (2) 52 (1950). J. Math. [I] On invariant connections’&ver a principal fibre bundle. Sot. H. J. [1] Les groupcs lineaires finis saris points fixes. Amer. A. J* Math. Nagoya Math.N. VINCENT. symmetric spaces. [. Amsterdam. Abh. coefficients. 84 (1962). 1 9 5 7 . 289-312. Sem. London Math. (2) [1] Sur les espaces metriques localement separables. WANG. Helv. Coll. Ann. Helv. Comm. [2] The Theory of Lie Derivatives and Its Applications. TAKIZAWA. Princeton University Press. . 213-234. Sem. 33-42. Sci. Trans. (Oxford) (2) 6 (1955). 21 7 (1956). 95 (1969). J. STEENROD. 77-80. (1950). Amer. 33-41. y. constante. J. Trans. 3 (1932). Tohoku Math. WEIL. Math. N. 76 (1962). 527-542. [4] The Clifford-Klein space forms of indefinite metric. Appl. J. [3] Homogeneous manifolds of constant curvature. 82 (1960). Amer. Comm. 384-411. Quart. YOSIDA. [1] On Cartan connexions and their torsions. H. J. YAMABE. 75 [ 1] The Topology of Fibre Bundles. Z. 112-147. 20 YANO. Sot. Y. Math. of Math. WOLF. Math. Springer. [1] Lorentz structures on the plane. 363-372. Quart. [3] Raum. % 187-220. 36-64. Quart. Gottingen‘ Nachrichtcn (19ll’!:rtl). 13 (1958). London Math. [2] Z u r Infinitesimalgeomctrie.G. Hamburg 13 [I] Parallel distributions on manifolds. G. I-19. Math. SMITH. 226-237. 221-231. 92 (1959). [l] Recurrent tensors on a linearly connected differentiable manifold. [l] A theorem concerning the semi-simple Lie groups. 12 (1938). C. Sci. Helv.I] RI-ine infinitesimalgeometrie. of blath. hlath. [I] On harmonic and Killing vectors. 269-276. C. Scm. Sot. II. R. (3) 6 (194Oi. Amer. Osaka Math. Math. iLlath. 191-204. WEYL. Math. 9 (1958). Proc. Sot. 301-308. T. A. l-19. 661-688. Trans. [I] Conformal connections and conformal transformations. Ann. [6] Discrete groups. 168-190. [2] Faber endliche Fastkorper. and global holonomy. Ann.N. Math. J. [3] Connexions for parallel distribution in the large. 36 (1961). 2 (1918\. 107-I 13. Amer. 29 (1955). WILLMORE. Abh. 117-171.J. J. 119-145. SIERPIN~KI. [2] The manifolds covered by a Riemannian homogeneous manifold. Publishing Co. 55 (1952). Math. J.C. Math.112. Math. [ 1] Sur la classification des varittts riemanniennes homogtnes a courbure [1] On transitivity of holonomy systems. [2] Connexions for systems of parallel distributions. [2] The fibring of Riemannian manifolds. North Holland [1] On Ruse’s spaces of recurrent curvature. 2 3 (1953). 199-217.

162 transformation. 92 principal fibre. 125 parameter of a grodcsic. 126 canonical flat. 110. 302 parameter. 46 Analytic continuation. 118 frame. 40 decomposition ( = dc Rham decompd- Affine sition). 155 space. 127 form on L(. 13 of a connection. 125 l-form on a group. 2 infinitesimal. 2 Riemannian manifold. 92 generalized. 2 Riemannian metric. INDEX . 230 Christoffel’s symbols (I’:. 172 Automorphism vector field. 60 canonical. 26 Bianchi’s identities. 192 connection. homomorphism of. 50 flat affine. 41 tangent. 64 sub-. 92 of linear frames. 139 Atlas.. 138 metric. 141 Alternation. 130 invariant Riemannian . 301 of affine frames. 50 by parallelism. 262 vector. 122 Canonical Adjoint rcprescntation. 60 . 56 induced. 129 induced.). 53 form. 5 Bundle Conformal transformation. 40 of a l-form. 125. 209 reduced. 309 associated. 81. 6 of a tensor (field). 127 tangent. 126 invariant connection. 40 of a linear connection. 28 Compact-open topalogy. 302 of orthonormal frames. 141 of a Lie group. 110. 85 Connection. 82 tensor. 53 affine. 129 flat connection. 309 holonomy. 113 Levi-Civita. 301 holonomy group.M). 63 . flat. 21. 130 linear connection. 81 Components of a Lie algebra. 78. 135 of a vector (field). Absolute parallelism. 172 complete. 155 parallel displacement. * Arc-length. mapping. 254 Complete. 226 Chart. 55 infinitesimal. 56 canonical linear. 121. 56 invariant. 53 generalized affine. 103 trivial. 158 L 325 . 225 metric. 185. 157 linear connection.

176 Isotropv ’ Exterior Cross section. 210 vector. 39. 262 Uniwtd. 117. 119 manifold. 2 10 Riemannian manifold’. 209 Contravariant tensor (space). i5 affinr. 49 covariant differentiation. 43 curve. 35 involutive. 301 subgroup. 88 Lit connection. 133 canonical. 184. 179 metric. 73. 55 of rl function. 309 Killing vector field. 10 Elliptic. 68. 9 horizontal.ift. 257 Homomorphism of fibre bundles. 75 Devclopmtnt. 202 component. 68. 10 minim&ng. 158 totally. 218 locally. 57 space (qubtient space). 130 Coordinate neighborhood. 30 Immersion. 126 connection. 154 i