Obs) Kies 14) FOUNDATIONS OF DIFFERENTIAL GEOMETRY VOLUME I SHOSHICHI KOBAYASHI University of California, Berkeley, California and KATSUMI NOMIZU Brown University, Providence, Rhode Island 4700030260873 AP!NOMAO LEIA INTERSCIENCE PUBLISHERSPREFACE This is a continuation of Volume I of the Foundations of Differential Geometry. The chapter numbers are continued from Volume I and the same notations are preserved as much as possible. The main text, Chapters VII-XII, deals with the topics that have been promised in the Preface of Volume I. The Notes include material supplementary to Volume I as well. The Bibliography duplicates, for the sake of convenience of readers, all the references in the Bibliography of Volume I in the same numbering and continues to references for Volume TI. The content of each chapter is now briefly described. Chapter VII gives the fundamental results and some classical theorems concerning geometry of an n-dimensional submanifold M immersed in an (x + p)-dimensional Riemannian manifold N, in particular, R**?. In §1, the natural connections in the orthogonal bundle and the normal bundle over M are derived from the Riemannian connection in the orthogonal bundle over N. In §2, where N = R"*, we show that these connections are induced from the canonical connections in the Stiefel manifolds Vin, p) and V(p,n), both over the Grassmann manifold G(n, p), respectively, by means of the bundle maps associated to the generalized Gauss map of M into G(n, p). In §§3 and 4, we use _ the formalism of covariant differentiation Vx¥ to study the relationship between the invariants of M and N and obtain the classical formulas of Weingarten, Gauss and Codazzi. We prove a result of Chern-Kuiper which generalizes the theorem of ‘Tompkins. §§5, 6, and 7 are concerned with the classical notions and theorems on hypersurfaces in a Euclidean space, including a result of Thomas-Cartan-Fialkow on Einstein hypersurfaces and results on the type number and the so-called fundamental theorem. In the last §8, we discuss auto-parallel submanifolds and totally geodesic submanifolds of a manifold with an affine connection and prove, in particular, that these two notions coin- cide in the case where the connection in the ambient space has no torsion. The content of Chapter VIT is supplemented by Notes 14,15, 16, 17, 18, 21, and 27. Chapter VIII is devoted to the study of variational problemsvi PREFACE fon geodesics. In §1, we define Jacobi fields and conjugate points for a manifold with an affine connection and discuss their geo- metric meaning. In §2 and 3, we make a further study of these notions in a Riemannian manifold and prove the classical result, fon the distance between consecutive conjugate points on a geodesic when the sectional curvature (or, more generally, each of the eigenvalues of the Ricci tensor) is greater than a certain positive number everywhere. In §4, we prove Rauch’s comparison theorem. In §5, we study the first and second variations of the length integral, considered as a function on the space of all piece- wise differentiable curves, and obtain, among others, a proof of ‘Myers’s Theorem. The Index Theorem of Morse is proved in §6. In §7, we prove basic properties of cut loci. Although the results of §7 are not used elsewhere in this book, they are basic in the study of manifolds with positive curvature. In §8, we prove a theorem of Hadamard and Cartan which says that for a complete Riemannian space with non-positive curvature the exponential map is a covering map. Applications are made to a homogeneous Riemannian manifold with non-positive sectional curvature and negative definite Ricci tensor. In §9, we prove a theorem to the effect, that on a simply connected complete Riemannian manifold with non-positive sectional curvature every compact group of isometries has a fixed point. Applications are given to the case of a homogeneous Riemannian space. Results of §§8 and 9 are used in §11 of Chapter XI. Note 22 supplements the content of this chapter. In Chapter IX, we provide differential geometric foundations for almost complex manifolds and Hermitian metrics, in particular, complex manifolds and Kaehler metrics. The results in this chapter are essentially of local character. After purely algebraic preliminaries in §1, we discuss in §2 the notion of an almost complex structure, its torsion and integrability as well as complex tangent spaces, operators @ and 4 for complex differential forms ‘onan almost’ complex manifold. Many examples are given, including complex Lie groups, complex parallelizable spaces, complex Grassmann manifolds, Hopf manifolds and their gen- eralizations, and a result of Kirchhoff on almost complex structures on spheres. In §3, we discuss connections in the bundle of complex linear frames of an almost complex manifold and relate their PREFACE, vil torsions with the torsion of the almost complex structure. In$4, Hermitian metrics and the bundle of unitary frames are discussed. ‘The most interesting case is that of a Kachler metric, whose basic properties are proved here. In §5, we build a bridge between intrinsic notations and complex tensor notations for Kachlerian geometry, In §6, many examples of Kachler manifolds are dis- cussed, including the Fubini-Study metric in the complex projective space and the Bergman metric in the open unit ball in C*, In §7, we give basic local properties of holomorphic sectional curvature and prove that a simply connected and complete Kaehler manifold of constant holomorphic sectional curvature ¢ is a complex projective space, a complex Euclidean space or an open unit ball in a complex Euclidean space according ase > 0, = or < 0..n §8, we discuss the de Rham decomposition of a Kachler manifold and the notion of non-degeneracy. §9 is concerned with holomorphic sectional curvature and the Ricci tensor of a complex submanifold of a Kaehler manifold. In the last §10, we study the existence and properties of Hermitian connection in a Hermitian vector bundle following Chern, Nakano, and Singer. This chapter is supplemented by §6 of Chapter X, §10 of Chapter XI (where examples are discussed from the viewpoint of symmetric spaces), and Notes 13, 18, 23, 24, and 26. In Chapter X, we discuss the existence and properties of invariant affine connections and invariant almost complex structures on homogeneous spaces (especially, reductive homo- geneous spaces). In §1, the results of Wang in §11 of Chapter II are specialized to the situation where P is a K-invariant G-structure on a homogeneous space M = K/H, and K-invariant connections in P are studied, In §2, we specialize further to the case where K/H is reductive and obtain the canonical connection and the natural torsion-free connection of Nomizu. In §3, we study homogeneous spaces with invariant (possibly indefinite) Ri mannian metrics. As an example we provide a differential geometric proof of Weyl’s theorem that a Lie group G is compact if the Killing-Cartan form of its Lie algebra is negative-definite. In §§4 and 5, results of Nomizu and Kostant on the holonomy group and reducibility of an invariant affine connection are proved. In §6, following Koszul we give algebraic formulationsPREFACE for an invariant almost complex structure on a homogeneous space and for its integrability. This chapter-serves as a basis for Chapter XI and is supplemented by Notes 24 and 25. In Chapter XI, we present the basic results in the theory of affine, Riemannian, and Hermitian symmetric spaces. We lay ‘emphasis on the affine case a little more than the standard treat- ment of the subject. In §1, we consider affine symmetric spaces, thus giving a geometric motivation to the group-theoretic notion of symmetric space which is introduced in §2, In §3, we reverse the process in §1; thus we begin with a symmetric space G/H and introduce the canonical affine connection on G/H, making GJH an affine symmetric space. The curvature of the canonical connection is given an algebraic expression. In §4, we study totally geodesic submanifolds of a symmetric space G/H (with canonical connection) from both geometric and algebraic view- points, The symmetric Lie algebra introduced in §3 is to a symmetric space what the Lie algebra is to a Lie group. In §5, two results on Lie algebras, namely, Levi's theorem and the decomposition of a semi-simple Lie algebra into a direct sum of simple ideals, are extended to the case of symmetric Lie algebras, The global versions of these results are also given. In §6, we consider Riemannian symmetric spaces and the corresponding symmetric spaces. The symmetric Lie algebra ‘corresponding to a Riemannian symmetric space is called an orthogonal symmetric Lie algebra. In §7, where orthogonal symmetric Lie algebras are studied, the decomposition theorems proved in §5 are made mote precise. In §8, the duality between the orthogonal sym- metric Lie algebras of compact type and those of non-compact type are studied together with geometric interpretations. In §9, wwe discuss geometric properties and an algebraic characterization of Hermitian symmetric spaces. Many examples of classical spaces are studied in §10 from viewpoints of symmetric spaces, including real space forms originally defined in Chapter V and complex space forms discussed in Chapter IX. In the last §11, we show, assuming Weyl’s existence theorem of a compact real form of a complex simple Lie algebra, that the classification of irreducible orthogonal symmetric Lie algebras is equivalent to the classifi- cation of real simple Lie algebras. In Chapter XII, we present differential geometric aspects of PREFACE ix characteristic classes. If G is the structure group of a principal bundle P over M, then using the curvature of a connection in P we can associate to each Ad(G)-invariant homogeneous poly- nomial f of degree k on the Lie algebra of G a closed 2k-form on the base space M in a natural manner. The cohomology class represented by this closed 2k-form is independent of the choice of connection and is called the characteristic class determined by f. In §1, following Chern we prove this basic result of Weil. In §2, we study the algebra of Ad(G)-invariant polynomials on the Lie algebra of G and determine the algebra explicitly when Gis a classical group. In §3, adopting the axiomatic definition of Chern classes by Hirzebruch, we express the Chern classes of a complex vector bundle in terms of the curvature form of a connection in the bundle. The formula for the Chern character in terms of the curvature form is also given. In §4, using Hirze- bruch’s definition of the Pontrjagin classes of a real vector bundle, we derive differential geometric formulas for the Pontrjagin classes. In §5, we characterize the real Euler class of a vector bundle in a simple axiomatic manner and derive the general Gauss-Bonnet formula, This chapter, particularly §5, is supplemented by Notes 20 and 21. We wish to note specifically that we do not go into the following subjects: the theory of (2-dimensional) minimal surfaces; the theory of global convex surfaces developed by A. D. Aleksandrov and his school; Finsler geometry and its generalizations; the general theory of conformal and projective connections; a deeper study of differential systems. On the subjects of complex manifolds, homogeneous spaces (especially symmetric homogeneous spaccs), vector bundles, G-structures and so on, our treatment is limited to the foundational material in differential geometric aspects that does not require deeper knowledge from algebra, analysis or topology. Neither do we treat the harmonic theory nor a gener- alized Morse theory, although these theories have many important applications to Riemannian geometry. The Bibliography of Volume IT contains some basic references in these areas. In particular, for the global theory of compact Kaehler manifolds Which requires the theory of harmonic integrals, the reader is advised to read Weil's book: Introduction @ Etude des Varietés Kahlériennes.x PREFACE During. the preparations of this volume, we have been most encouraged by the reactions to Volume I of many readers who wanted to find self-contained and complete proofs of the standard results in the field, We sincerely hope that the present volume will continue to meet the needs of these readers. We should also like to acknowledge the grants of the National Science Foundation which supported part of the work included in this book. Suosmicur Komavasit September, 1968 Karsumt Nomi CONTENTS OF VOLUME IL Cuarrer VIL Submanifolds Frame bundles ofasubmanifold. =... The Gauss map... : Covariant differentiation and second fundamental form Equations of Gauss and Codazzi. . se Hypersurfaces ina Euclidean space. 0. ‘Type number and rigidity : Fundamental theorem for hypersurfaces. | Aworparaile braids and stall gendee submani- folds. soe Cxaprer VITT Variations of the Length Integral Jacobi fields ‘Jacobi fields in a Riemannian manifold. Conjugate points Comparison theorem The frst and second variations of the length integral. Index theorem of Morse para Cut loci. ae Center of gravity and fixed points of isometries | Cuarrer IX Complex Manifolds Algebraic preliminaries, ‘Almost complex manifolds and complex manifolds Connections in almost complex manifolds. Hermitian metrics and Kaehler metrics... Kaehler metrics in local coordinate systems Examples of Kaehler manifolds : Holomorphic sectional curvature - . De Rham decomposition of Kachler manifolds |, 10 22 29 2 47 33 63 7 76 79 102 108 14 121 41 146 155 159 165 imLX 2 3. 4 5. CONTENTS OF VOLUME It Curvature of Kaehler submanifolds. Hermitian connections in Hermitian vector bundles. (Cuaprer X Homogeneous Spaces Invariant affine connections . Tnvariant connections on reductive homogeneous spaces Invariant indefinite Riemannian metrics =. Holonomy groups of invariant connections ‘The de Rham decomposition and irreducibility Invariant almost complex structures... Cuarrer XT Symmetric Spaces ‘Affine locally symmeti spaces ae Symmetric spaces. ‘The canonical connection on a symmetric space. Totally geodesic submanifolds. : Structure of symmetric Lie algebras... Riemannian symmetric spaces : Structure of orthogonal symmetric Lie siete Duality : : Hermitian symmetric spaces... Examples. ‘An outline ofthe clasification theory Cuapren XIT Characteristic Classes Weil homomorphism . 5 ee Invariant polynomials . eae Chern clases. Sees Pontrjagin classes. toe Eulerclases ee Appenpices Integrable real analytic almost complex structures. ‘Some definitions and facts on Lie algebras 175 178 186 190 200 204 210 216 222 225 230 234 238 243 246 253 259 264 289 293 298 305, 312 3i¢ 32 325 2! 13. 1. 15. 16. 7. 18, 19. 20. CONTENTS OF VOLUME 11 Nores Connections and holonomy groups (Supplement to Note 1) ‘The automorphism group of a geometric structure (Supple- ment to Note 9) aie erect ‘The Laplacian Foe Surfaces of constant curvature in RE}, Index of nullity. soe ‘Type mumber and rigidity of imbedding Isometric imbeddings : Equivalence problems for Riemannian manifolds. Gauss-Bonnet theorem. Total curvature . aaa Topology of Riemannian manifolds with positive curvature Topology of Kachler manifolds with positive curvature Structure theorems on homogeneous complex manifolds . Invariant connections on homogeneous spaces Complex submanifolds Minimal submanifolds foe Contact structure and related structures 0. Bibliography Summary of Basic Notations. 5. Index for VolumestandIE Errata for Foundations of Differential Geometry, Volume 1 331 392 337 343 347 349 354 357 358 361 368 373 375 378 379 381 387 455 459 469CONTENTS OF VOLUME 1 CHAPTER VIL DirrerenTiabe MaNirous Submanifolds ‘Tueory oF Connections Linear AND AFFINE CONNECTIONS 1, Frame bundles of a submanifold Let ey, «++ eas» be the natural basis for R"*?. We shall denote RIEMANNIAN CONNECTIONS GuRvATURE AND Space FoRus by Re and R the subspaces of R'+? spanned by é, ..., ¢ and fasts «+> Enyp Tespectively. Similarly, we identify O(n) (resp. ‘TRANSFORMATIONS : O(p)) with the subgroup of O(n + p) consisting of all elements f which induce the identity transformation on the subspace R? (resp. R°) of R°+?. In other words, ( O(n) 0 0 ) 0 ow) | where /, and J, denote the identity matrices of order m and , respectively. Let o(n + p), 0(n) and 0(p) be the Lie algebras of O(n +p), O(n) and O(p), respectively, and let g(x, p) be the orthogonal complement to o(n) + 0(p) in o(n +p) with respect to the Killing-Cartan form of o(n + p) (cf Volume I, p. 155 and also Appendix 9). Then q(x, p) consists of matrices of the form (24 4) where 4 is a matrix with n rows and p columns and ‘A denotes the transpose of A. Let be a Riemannian manifold of dimension n + p and let f be an immersion of an n-dimensional differentiable manifold M into N. We denote by g the metric of IV as well as the metric induced on M (cf. Example 1.2 of Chapter IV). For any point x of M we shall denote f(x) € N by the same letter x if there is no danger of confusion. Thus the tangent space T.(M) is a subspace 1 O(n) ~ ) and O(p) ~(2 FOUNDATIONS OF DIFFERENTIAL GEOMETRY of the tangent space 7,(N). Let T,(M)+ be the orthogonal complement of 7,(M) in T,(/N); it is called the normal space to M atx. Let 0(.M) and O() be the bundles of orthonormal frames over M and N, respectively. Then O(N) | M = {9 € O(N); m(a) € Mf}, where : O(V) + N is the projection, is a principal fibre bundle over M with structure group O(n + p). A frame v € O(N) | Mat € Missaid to be adapted if vis of the form (Yay. » Vou Yayy-+ +5 Yuss) with Y,,..., Yq tangent to M (and hence ¥uyy, +3 Yar normal to M). Thus, considered as a linear isomorphism R"*? "+ T,(N), 2 is adapted if and only if » maps the subspace R" onto T_(M) (and hence the subspace R? onto T,(M)+). It is easy to verify that the set of adapted frames forms a principal fibre bundle over M with group O(n) x O(p); it is a subbundle of O(N) | M in a natural manner. We shall denote the bundle of adapted frames by O(N, M). We define a homomorphism A’: O(N, Mf) — 0(M) corresponding to the natural homomorphism O(n) x 0(p) > O(n) as follows: HQ) =(Yyeeey Ya) ford = (Ys, 00., Yoys) € O(N, M). If we consider v as a linear transformation R'+? -> T,(N), then W'(0) is the restriction of v to the subspace R*. Hence, O(M) is naturally isomorphic to O(N, M)/O(p). Similarly, denoting by (0) the restriction of » ¢ O(N, Mf) to the subspace R? of R™7, we obtain a homomorphism fi": O(N, M) -+ O(N, M)/O(n) cor- responding to the natural homomorphism O(n) x’ O() -» O(p). By a normal frame at x¢M, we mean an orthonormal basis (Zy +++, Z,) for the normal space 7,(M)*. If (Yi, .--, Yay Yysiy +++) Yagg) i8 an adapted frame at x, then (Yy.y ++» Yay.) is'a normal frame at x. Since every normal frame is thus obtained and since two adapted frames give rise to the same normal frame if and only if they are congruent modulo O(n), the bundle O(N, M)/O(n) can be considered as the bundle of normal frames over M. Then A": O(N, M) + O(N, M)/O(n) maps an adapted frame » = (Ya, «-., Yass) upon the normal frame (Fay, «+5 Yuss). We denote by T(M)+ the set Ure Te(M)+. It is then a vector bundle over M associated to the bundle of normal frames O(N, M)/O(n) by letting the structure group 0(p) act naturally on the standard fibre R? (cf. §1 of Chapter III). We shall call VI, SUBMANIFOLDS 3 this vector bundle the normal bundle of M (for the given immersion Ff into N). The following diagrams illustrate these bundles: O(a) = OW, 81/019) «OWN, at) “> O(N, a4)/0(0) Oem |e 0) x ow |= owe M - M _ M O(N, M) ++ o(¥) | M2 0.) 20) x oun |r Om tnle ol +n)|x ay where both i and j are injections. Let @ and ¢ be the canonical forms of M and N, respectively (cf. §2 of Chapter III); @ is an R*-valued I-form on O(M) and g is an R"+-valued I-form on O(N). Then we have Propostrion 1.1. A'*(0) coincides with the restriction of @ to O(N, M). In particular, the restriction of the Re+*-valued form g to O(N, M) is R-valued. Proof. By definition of g we have oY) =eMa(¥)) for Ye TO(N, M)). Since »(¥) € T,(M), where x = x(v), and since x! maps T,(M) ontoR®, 9(¥) isin R’, Since A’(v) = 0 | R* and since n’ © H’(Y) = (¥), we have a PMa(¥)) = H(A AY) = (K(X) (H*())(X). QED. Let y be the Riemannian connection form on O(N). Its restric- tion to OV) | M, that is, j*y, defines a connection in the bundle O(N) | M. But its restriction to O(N, M), that is, ##/*y, is not, in general, a connection form on O(N, Mf). Propostrion 1.2. Let y be the Riemannian connection form on O(N) and let w be the o(n) + 0(p)-component of itj*y with respect to the decomposition o(n + p) = o(n) + 0(p) + g(n, fp). Then w defines a connection in the bundle O(N, M).4 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Proof, Since ad(O(n) x O()) maps (x, #) onto itself, we see from Proposition 6.4 of Chapter IT that the form @ defines a connection in O(N, M). QED. Prorostrion 1.3. The homomorphism h':-O(N, M) -» O(M) = O(N, M)O(p) maps the connection in O(N, M) defined by co into the Riemannian connection of M. The Riemannian connection form «! on O(M) is determined by H*(0') = 9), where wyty) denotes the o(n)+component of the o(n) + 0(p)-valued form a. Proof. By Proposition 6.1 of Chapter IT we know that A’ maps the connection defined by w into the connection in O(M) defined by a form «’ such that h'*(w’) = @4,). To show that w’ defines the Riemannian connection of M we have only to show that the torsion form of a” is zero. Restricting the first structure equation of y to O(N, M), we obtain di*i*9) = —(i**y) A (i). Since i*j*p is equal to A'*(6) and is R*-valued by Proposition 1.1, comparing the R'-components of the both sides we obtain d(n'*(0)) = —A'*(a’) AK'*(0). Since h' maps O(N, M) onto O(M), this implies d0 = —o' 8. QED. Similarly, by Proposition 6.1 of Chapter II we see that there is a unique connection form o on the bundle O(N, 4)/0() such u = Wo") = eu where @,:,) denotes the o{p)-component of the o(n) + 0(p)-valued form ow. Geometrically speaking, «" defines the parallel displace- ment of the normal space T,(M) onto the normal space T,(M)~ along any curve + in M from x to y. The bundles O(N, M), O(M) = O(N, M)/O(p), and O(N, M)/ O(n), and their connection form w, a’, and w” are related as follows: Proposition 1.4. The mapping (A, H"): O(N, M) — O(M) x (O(N, M)/O(n)) induces a bundle isomorphism O(N, M) ~ O(M) + acne The connection form w coincides with h'*(a') + The proof is trivial (see p. 82 of Volume 1). VIL. SUBMANIFOLDS 5 Finally, we say a few words about the special case of a hyper- sypface. By a hypersurface in an (n 4. 1)-dimensional manifold we mean a (generally connected) n-dimensional manifold M with an immersion f. For each x M, there is a coordinate neighbor- hood U of x in M and a differentiable field, say, &, of unit normal vectors defined on U. Such a & can be, easily constructed by choosing a coordinate system s!,..., x” around x in U and a coordinate system j3, ...,.)"*# around x (= f(x)) in Nj in fact, a unit normal vector field on U is determined uniquely up to sign. For a fixed choice of € on U, it is obvious that ¢ is parallel along all closed curves in U (with respect to the connection in the normal bundle). Assume that NV is orientable and is oriented. Then we can choose a differentiable field of unit normal vectors over M if and only if M is also orientable. Indeed, for a fixed orientation on M, there is a unique choice of the field of unit normal vectors & such that, for an oriented basis {X;,... » X4} of T,(M) at each x eM, {€,, X,,...,X,}is an oriented basis of T,(N). Conversely, if a field § of unit normal vectors exists globally on M, then a basis {X,, «.., X,} of T,(M) such that (8, Xy, ++, Xq} is an oriented basis of T7,(N) determines an orientation of M. If we forget about the particular orientations of N and M, then again a differentiable field of unit vectors on is unique up to sign, For a choice of é, it is obvious that ¢ is parallel along all curves in M. Without assuming that V and M are orientable, let us choose a unit normal vector &, at a point x,on M. The parallel displacement along all closed curves at x, on M will map &, either upon & or upon —f,. In other words, the holonomy group of the linear connection in the normal bundle is a subgroup of the group {1,-1}. (This is also clear from the fact that the bundle O(N, M)/O(n) of normal frames over M has structure group (1) = {1, =1}.) If the holonomy group is trivial (and this is the case if M is simply connected), then é is invariant by parallel displacement along all closed curves at xy. In this case, ‘we may define a differentiable unit normal vector field ¢ on M by translating & parallelly to each point x of M, the result being independent of the choice of a curve from x to x in M, We may thus conclude that if M is a simply connected, connected hypersurface immersed in a Riemannian manifold N, then M admits a differentiable Jfeld of unit normal vectors defined on M.6 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 2. The Gauss map We consider R"+? as an (n + p)-dimensional vector space with an inner product such that the natural basis is orthonormal. Let G(n, p) be the set of n-dimensional subspaces of R'*, We shall make G(x, p) into a manifold as follows. The group O(n + p) acting on R"? acts transitively on G(n, p) in a natural manner. The elements of O(n + p) which map the particular subspace R* onto itself form the subgroup O(n) x O(p). Thus we have G(x, p) = O(n + p)/0(f) x O(p). ‘The manifold Gn, p) is called the Grassmann manifold of n-planes in RN, By an n-frame in R+? we mean an ordered set of n orthonormal vectors in R*+?, The group O(n + p) acts transitively also on the set Vn, p) of n-frames in R'* in a natural manner. The elements of O(n + p) which fix the n-frame (¢,,... ¢) form the subgroup 0(#). Thus we have V(n, p) = O(n + p)/0(9). The manifold Vin, p) is called the Stiefel manifold of n-frames in Re, Similarly, the isotropy subgroup of O(n +f) at the point of ¥(p,n) represented by the p-frame (ég415- ++ nes) is O(n). Hence P(p, n) may be identified with O(n + p)/O(n). ‘We have now the following three principal fibre bundles over G(r, p): E= O(n +9) over G(x, p) with group O(n) x O(2), E’=V(n,p) over G(n, p) with group O(n), E'=V(p,n) over G(n,p) with group 0(p). ‘The projection in the bundle £’ has the following geometric mean- ing. An n-frame projects upon the n-plane spanned by it. Similarly, the projection in the bundle E” maps a p-frame into the n-plane normal to the p-plane spanned by it. Let y be the canonical I-form of O(n +p) with values in o(n +p) (cf §4 of Chapter 1), Let wg: be the o(n) + 0(2)- component of y with respect to the decomposition o(n + p) = o(n) + 0(f) + 9(n, p) defined in §1. By Theorem 11.1 of Chapter Vit, SUBMANIFOLDS 7 II, the form wg defines a connection in E, which will be called the cangnical connection in E and will be denoted by Pp. he natural projection of O(n + p) onto V(n, p) (resp. onto V(p,n)) defines a bundle homomorphism of E onto E’ (resp. E") which will be denoted by J” (resp. f"). By Proposition 6.1 of Chapter II, there exists a unique connection in £’, denoted by Py. (resp. a unique connection in E", denoted by T's.) such that f” maps I’, into ['y. (resp. f” maps l'yinto I’s-). By the same propo: tion, the connection forms wy, and wp. of gy and Pp. are deter- mined by *(ox) = Yam and f*(o) where yoq) and Yay) are the o(n)- and o(p)-components of y, respectively. We call I'g. and T'g- the canonical comnecioas in E’ and E’, respectively. In §6 of Chapter II, we defined E’ + E” to be the restriction of E’ x E’to the diagonal of G(x, p) x G(x, p)- We have (ef. Propo- sition 6.3 of Chapter IT). PRopostrion 2.1. The mapping (f',f"): E> E’ x E" induces a bundle isomorphism E = E' +E", The canonical connection forms are oe op = f*(oe) + foe). Proof. The first statement is trivial. The second statement follows from hob» © = Yam + Yetor QED. Let M be an n-dimensional manifold immersed in the (n + p)- dimensional Euclidean space R"+*, Considering R'? as a flat Riemannian manifold we apply results of §1. In particular, we have the following three principal fibre bundles over M: P = 0(R™, M) over M with group O(n) x O(f), P’ = 0(M) = O(R™, M)/0(p) over M with group O(n), P* = O(R"?, M)/O(n) over M with group O(f). ‘The connections in P, P’, and P” defined by o, w, and a” in §1 will be denoted by I'p, Fp, and I’p-, respectively. We now define a bundle map g: P -» E. The bundle 0(R*) of orthonormal frames over, R'is trivial in a natural manner, thatis, O(R™+) = Re? x O(n +p). Letp: O(R™*) + O(n + 2)8 FOUNDATIONS OF DIFFERENTIAL GEOMETRY be the natural projection. Since P = O(R"?, M) < O(R™), every adapted frame v ¢ P can be considered as an element of O(R?). We define gl) = p(t) forveP. Evidently, g is a bundle map of P into E = O(n +f), that is, g commutes with the right translations by O(n) x O( #). The bundle map g induces bundles maps g': P’ + E’ and g": P’ + E" ina “natural manner. It induces also a mapping g: M —> G(x, p). We ‘summarize various bundles and maps in the following commuta- tive diagra | | \ plo We are now in position to prove the main result of this section. Tueorem 2.2. The bundle maps g, g', and g" map the connections Tp, Dp, and Pp upon the sanonical connections Pp, Tg, and Tp, respectively. Proof. Since f’ (resp. f*) maps I’, upon I’, (resp. T'z.) and since A’ (resp. A”) maps Tp upon T'p. (resp. I'p.), it suffices to prove that g maps T'p upon I'p. The flat Riemannian connection of R"? is given by the form p*(y) on O(R™) (cf. §9 of Chapter II), where p: O(R"+*) = R'+? x O(n +p) + O(n + ) is the natural projection. The restriction of p*(y) to P is given by g*(y). The connection form w on P is the o(n).+ 0(f)-component of g*(y) and hence is equal to g*(wg) since wg is the o(n) + 0()= component of y. QED. ‘We shall give a geometric interpretation of g, ¢’, g”, and g. Let Pi, 5.9%? be the coordinate system in R%? and let wy be the ‘orthonormal frame at the origin given by 2/dy!, ..., /@)"*?. To each element a ¢ O(n + p) we assign the frame aw, at the origin of RY, This gives a one-to-one correspondence between O(n + ?) and the set of orthonormal frames at the origin of R'#?. For each adapted frame ve P at x ¢ M, g(0) is the frame at the origin of £209 VIL. SUBMANIFOLDS 9 Re? parallel to v; this follows from the fact that the mapping 2 Q(R™) = RY? x O(n +p) + O(n +f) is the parallel dis- plagement of O(R*#) into the frames at the origin. This inter- pretation of g implies that g’ maps a frame u ¢ O(M) upon the neframe in R*” parallel to u. Similarly, g” maps a normal frame wat x €M upon the p-frame in R"+? parallel to u. Finally, g maps x € Minto the element of G(n, p) represented by the n-dimensional subspace of R"” parallel to T.(M). When we consider the orientation in R"”, we take the following three principal fibre bundles over G(n, p) = SO(n + p)/SO(n) x E = SO(n +p) over G(a, p) with group SO(n) x SO(p), E’ = SO(n + p)/SO(p) over Gin, p) with group SO(n), E* = SO(n + p)/8O(n) over G(n, p) with group SO(/). ‘The base space G(n, p) is the Grassmann manifold of oriented n-planes in Rv, If M is orientable and oriented, taking those adapted frames which are compatible with the orientations of M and Rr? we obtain a subbundle P of P with group SO(n) x SO(p). We set Piso(p) and P* SO(n) and we define mappings WHS, Ff", & Wand Z" analogously. Then we obtain a theorem similar to Theorem 2.2. Moreover, if M is an oriented hyper- surface in R**, then the bundle map g: P > E induces a map ¢ from the base space M of P into the base space G(n, 1) = S* (n-sphere) of E, called the spherical map of Gauss. Geometrically speaking, $ assigns to each point x ¢ M the unit vector at the origin of R'*! which is parallel to the unit normal vector , of M atx chosen in such a way that &, is compatible with the orientations of Mand Rv4, We shall often identify 4(x) with &,. If Mis not orientable, we can assign to each x € M the normal line, but we cannot obtain a continuous field of unit normal vectors, and thus we obtain the mapping of M into the real projective space G(x, 1) induced by the bundle map g: P + E, Example 2.1. Let M be an oriented hypersurface in R'+ as above, The bundle £ = £" = SO(n + 1) over G(n, 1) = S* with group SO(n) can be identified with the bundle of oriented ortho- normal frames over $* in a natural manner and the canonical10 FOUNDATIONS OF DIFFERENTIAL GEOMETRY connection in £ (or B’) can be identified with the Riemannian connection of S* (cf. the proof of Theorem 3.1 of Chapter V). Then Theorem 2.2 means that the spherical map $: M —> S* together with the bundle map g': P’ + & sends the Riemannian connection of M into the Riemannian connection of S* in the following sense. Let + be a curve from xeM to yeM. The parallel displacement 7: T,(M) + 7,(M) along + corresponds to the parallel displacement $(r): Tgie)(S") > Tyin(S*) along the curve (7) as follows. Since both T,(M) and Tya)(S*) are per- pendicular to £,, they are parallel to each other in R™4 and can be identified by a linear isomorphism y,: T,(M) + Tya(S")- Similarly, we have y,: T,(M) + Tyy(S*)- Then +: T..M) > T,(M) coincides with yz1° $+) © ye 3. Covariant differentiation and second ‘fundamental form In this section we shall discuss the Riemannian connection of a submanifold by using the formalism of covariant differentiation ‘Vc¥. We shall also define the second fundamental form. ‘Let M be an n-dimensional manifold immersed in a Riemannian manifold N of dimension n +p. We denote by V’ covariant differentiation in N. Since the discussion is local, we may assuine, if we want, that M is imbedded in NV and that we can choose p cross sections &,..., €, of the normal bundle 7(M)+, namely, differentiable fields of normal vectors, that are linearly independent at each point of M. They may be further assumed to be orthonormal at each point. Let Xand ¥ be vector fields on M. Since (Vic¥),is defined for each x € M, we shall denote by (VF), its tangential component, and by &,(, ¥) its normal component so that (Vy ¥)e = (Va¥ a + 10(% ¥), where (Vx¥),€7,(M) and 4,(X, ¥) © T,(M)+. Here (V,x¥), is introduced just as a symbol for the tangential component; the point is to show that it is in fact covariant differentiation for the Riemannian connection of M. a in gna Vi. SUBMANIFOLDS u It is easily verified that the vector field Vxc¥ which assigns the vector (Vx¥), to each point x ¢M is differentiable and that a(%, Y) is a differentiable field of normal vectors to M. We prove Propostrion 3,1. Vx¥ is the covariant differentiation for the Riemannian connection of M. Proof. We verify the properties (1), (2), (3), and (4) in Propo- sition 2.8 of Chapter ITI. (1), (2), and (3) are obvious from the corresponding properties of V’ on N and linearity of the projection T,(N) — T.(M). To verify (4), let f be a differentiable function on M, Then Vx(f¥) = (XAY +F(¥4¥), where (X/)¥ is tangential to M. Thus taking the tangential components of both sides, we obtain Vx J¥) = (XY +F(Vx¥), proving property (4) for V. By Proposition 7.5 of Chapter III we see that there is a unique linear connection P on M for which V.x¥ is the covariant differentiation. To show that I’ is the Rie- ‘mannian connection for the induced metric on M, itis sufficient to show: (a) the torsion tensor of I's 0, that is, Vx¥ — VyX =[X,¥], (b) Vg = 0. In order to prove (a), let us write VyY = Vy¥ + a(X, ¥) and VpX = VpX + al ¥, X). If'we extend X and ¥ to vector fields X’ and Y’ on N (as we may do locally), then the restriction of [X’, ¥'] to M is tangent to M and coincides with [X, Y]. Thus (X.Y, =[%, Yl, where xe M. Of course, we also have Ve¥' =Vg¥ and Vy.X"=VyX on12 FOUNDATIONS OF DIFFERENTIAL GEOMETRY From the equations above we obtain Viel! — VpX" — [X,Y] = Vx¥ — VyX — (X,Y) +a(X, ¥) —a(¥, X). Since the left hand side is 0 (because the torsion tensor of the Riemannian connection V' of N is 0), we see that Vx¥ — VX — [X,Y] =0, proving (a). Furthermore we have a(X, ¥) =a(¥, X). To prove (b}, we start from V'g = 0, which imy XY, Z) = e(Vx¥, Z) +a(¥, VxZ) on M for any vector fields X, ¥, and Z on M. We have, however, 8(Vx¥, Z) = a(Vx¥ + a(X, ¥),Z) = a(Vx¥, Z), because «(X, ¥) is normal to M. Similarly, we have &(Y, VeZ) = a(¥, VxZ). ‘Thus we obtain X-a(¥, Z) = g(Vx¥, Z) + g(¥, VxZ), which means Vg = 0. We have thus proved Proposition 3.1, QED. ‘We now prove the basic properties concerning the normal com- ponent a(X, Y). We denote by ¥()+ the set of all differentiable fields of normal vectors to M; it is a real vector space and a module over the algebra §(M) of differentiable functions on MM. Provosrrion 3.2. The mapping a: X(M) x X(M) ~» X(M)+ is symmetric (ist. «(X, ¥) = a(¥, X)) and bilinear over §(M). Con- sequently, a,(X, ¥) depends only on Xp sand ¥, and there is induced @ symmetric Bilinear mapping a, of T,(M) x T,(M) into T,(M)+. Proof. The symmetry of « has been proved in the proof of Proposition 3.1. Additivity in X or ¥ (when the other is fixed) is obvious. For any f'€ §(M), we have Vex¥ + a(fX, ¥) x =f Ve¥ (Wx¥ + a(%, ¥)), 1-20 > a. Vit. SUBMANIFOLDS 13 which implies : a(fX, ¥) = fe aX, ¥)- By shmmetry, we have a(X, fY) = f-a(X, ¥), thus proving that is bilinear over §(M). The rest of Proposition 3.2 is similar to the situation in Proposition 3.1 of Chapter I. QED. We define a: #(M) x ¥(M)—3(M)+ as the second funda- ‘mental form of M (for the given immersion in NV). For each x € M, a,: T,(M) x T,(M) + T,(M)+ is called the second fundamental form of M at x. In the case where M is a hypersurface immersed in N, choose a field of unit normal vectors & in a neighborhood U’ of a point Xp € M. For any vector fields X and ¥ on U, we may write a(X, ¥) = A(X, ¥)E, where A(X, ¥) is a symmetric mapping of X(U) x X(U) into ‘B(U) which is bilinear over §(U). At each x U, A, is a sym- metric bilinear function on T,(M) x T_(M). In classical litera- ture, A is called the second fundamental form of M for the choice of £. If it is possible to choose a field of unit normgl veetors € globally on M, then we can define h globally as a mapping of X(M) x 4(M) > B(M). ‘More generally, if M has codimension p, then we may locally choose p fields of unit normal vectors £,,...., & that are orthog- onal at each point. We may then express « by. a(X, ¥) = $4(%, YE thus getting p second fundamental forms in the classical sense. Next, let ¥ €X(M) and ¢ €X(M)+ and write (We = (ADs + (Dab where, for the moment, —(4,(X)) and D.x€ are just symbols for the tangential and normal components that depend on X and It is easily verified that the vector field x + (4,(X))_ and the field of normal vectors x + (Dé), are differentiable on M. About A, we prove4 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Proosrrion 3.3. (1) The mapping (X, 8) €X(M) x %(M)+ — AMX) €X(M) is bilinear coer %(M); consequently, (A,(X))e depends only on X, and Eq, and there is induced a bilinear mapping of T.(M) x T,{M)+ into T,(M), where x is an arbitrary point of M. (2) For each € € T,(M)4, we have 8(A(X), ¥) = g(a(X, Y), €) “for all X, ¥ © TM) ; consequently, Ay isa symmetric linear transforma- tion of TAM) with respect to ge Proof. (1): Additivity in X or & (when the other is fixed) is obvious. For any fe §(M), we have Vig FE) =f Uxe + (XAE = “f+ (A(X) +f: Dak + (ANE, Comparing this with Vx fe) = —Ayn(X) + Dx(FE), we obtain 4,,(X) = f,(X) for the tangential components and Dx(fé) = (XE +f: Dxé for the normal components. (The second identity is used in the next proposition.) On the other hand, a similar argument for V;x(¢) implies that 4,(fX) = f+ A(X) ‘This shows that A,(X) is bilinear in X and & over §(M). (We have also D,x =,f* Dé, which will be used for the next propo- sition.) (2): For any Ye X(M), we have g(¥, ) = 0. Differentiating covariantly with respect to X (for the Riemannian connection V’), we have 8(VXY, &) + @(¥, Vek) =0 so that &VX¥ + o(%, ¥), 8) = a(Y, A(X) + Dud) = 0. Since g(Vx¥, #) = g(¥, Dx#) =0, we get (0X, ¥), 8) = a4), Y)- This shows that A, is the linear transformation of T,(M) which corresponds to the symmetric bilinear function r. on T,(M) x T{M). Thus 4, is symmetric: g(4,(X), Y) = g(%4,(¥)), proving Proposition 3.3, QED. Vi, SUBMANIFOLDS 15 As for Dx, we have Brorosrion 34. The mapping (X,£) €X(M) x X(M) Dg eX(M)+ coincides with covariant diferentiation of the cross section & of the normal bundle T(M)* in the direction of X with respect to the connection in T(M) + defined in §1. Proof. In the preceding proof we have verified the same formal properties of D.x as those in Proposition 1.1 of Chapter ITI. Thus Def is actually covariant differentiation of a certain linear con- nection in the normal bundle, Moreover, for §, 7 €X(M)* we have = ~4,(X) + Dan so that (Db 0) + (8 Dyn) = 8 Vie, n) + (6 Von) Xgl 7), which shows that our connection Dx is metric for the fibre metric in T(M)+, namely, the restriction of g to the normal spaces. It remains to show that the metric connection Dx coincides with the connection defined in §1. However, we shall omit the Proof. QED. We have thus developed the first set of basic formulas for sub- manifolds, namely, () VyY = Va + 2(X, Y) m) ef = A(X) + Dyk (1) is called Gauss's formula, and (IT) is called Weingarten’s formula, Tn the case of a hypersurface M, (I) takes a simpler form. In fact, if we take a field of unit normal vector fields & then dift ferentiating g(&, €) = 1, we obtain aVxe, 8) =0, and hence g(Dx#, ) =0. Since Dx§ is normal and is therefore a scalar multiple of & we musthave Dx = Oateach point. Thus D ¢@ = 0 (when g(&, &) = 1) In the following examples, we shall discuss some special cases of (1) and (II) and also give some geometric consequences of Propositions 3.1-3.3.16 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Example 3.1. Let M, and M, be submanifolds, both of dimen- sion n, in a Riemannian manifold N of dimension n+p. Let 7 =a()), 0 S05 1, be a differentiable curve in M, A M,. We shall say that M, and M, are tangent to each other along 7 if Ty(M;) = Tauy(M;) for each 1, 0S #3 1. In this case, the parallel displacement along 7 in M, coincides with the parallel displacement along + in M,. In fact, if X = %, then for any vector field ¥ along 7, we have VY = VOY + 0f(X, ¥) = VRY + 0!(X, ¥), where V4 (resp. V) is the covariant differentiation for My (resp. M,) and a” (resp. a‘) is the second fundamental form for M, (resp. M,). Now if we assume that ¥ is parallel along z in M,, then vey which implies that Vjc¥ is normal to M, (and to M,). This in turn means that VPY = 0, that is, ¥ is parallel along 7 in My. In particular, if + is a geodesic in M,, 7 is a geodesic in Mz. Example 3.2. Let M be a submanifold of N. Let + 0 51 1, be acurve in M. Then 7 is a geodesic in M if and only if VipX, where X = &, is normal to M. In particular, if + is a geodesic of IV contained in M, it is a geodesic in M. (A geodesic in M is not in general a geodesic in N; we shall discuss this question, in detail in §8.) Example 3.3. Let M be a submanifold of dimension n in a Riemannian manifold N of dimension n + p. Let x» be a point of ‘M. It is possible to take a system of normal coordinates »#, . y'*? with origin x5 such that (3/8)%)...--» , (2/@")., span T,,(M). In fact, let Yay. Yur Yasin +++s Yary be an orthonormal basis of T,,(N) such that Y¥y,..., Y, form a basis of T_,(M). We may choose a system of normal coordinates *, ..., 9" such that (8/8y%)., = Yu 1s is m+ p (ch. §3 of Chapter IV). Note that Yyun +++» Futy form a basis of T,,(M) 4. Let s4, ...")x" be an arbitrary coordinate system in a.neighbor- hood U of x, in M and let HAH ae), ae VIL. SUBMANIFOLDS W be the system of equations that defines the imbedding of U into N. We shall show that a {( 8/8") (8/8%"), 2X (Gy/Ox Ox"), Fes namely, the coefficients of «,, with respect to the basis (0/@x*).,. +++; (0/Oe")xqin T,,(M) and the basis Yayi,.++s Yaapin T,,(M)* are those of the Hessian p*/@x*Ax" at x. In order to prove this we compute: Vian ( 3] 2 ) Fan ($(4/84(87874)) = Zaria (ay 2 Ving( A) +S (aye (a1) =F rey arene) + ¥ ayyaray (ala, where It are the Cristoffel symbols for the Riemannian connei tion in N with respect to y*, ...,9°**. Note that [7 are 0 at the origin x» of the normal coordinate system, Taking the normal components of both sides at x5 of the equation above, we have 'S (ayr/ataxy Te Example3.4.. Let NV = R*, the Euclidean space with standard Euclidear metric, and let .M be an n-dimensional submanifold of Re, that is, a hypersurface in R"*1, We may represent M locally by 54 (8/8x*)) (8/2")x,) PaHH yey 1 whore y!, ..., 9%1 is the standard rectangular coordinate system and 2',..., #* is an arbitrary local coordinate system of M. Or we may consider y = (j!,..., 3") as the position vector of a point with coordinates (7!,..., y"*#) and represent M locally by the equation for the vector-valued fanction: DAH ee) sn+l,18 FOUNDATIONS OF DIFFERENTIAL GEOMETRY For each i, 1S i $1, the vector field @/ax! can be expressed by the vector-valued function ¢, = 3y/@x'. The induced metric g on ‘Mis given by Bes = B(9/ Ax, 88x") = (ey €4), where (, ) denotes the standard inner product in the vector space R"+4, Choose a field of unit normal vectors on M which is represented by the vector-valued function £(2!,..., x") on the coordinate neighborhood of M. Thus (B8)=1 and (54) =0, Isisn Since the Riemannian connection of R"*" is flat, we have Vis = Bes/Ox, which is the partial derivative of the vector-valued function ¢;. Thus the formula of Gauss can be written as a) aes! = Fhe + huss where we know that T¥, are the Christoffel symbols for the Riemannian connection of the hypersurface M, that is, Vana(@/ae) = ¥ 1502/24) and that fy, are the coefficients of the second fundamental form. Similarly, the formula of Weingarten takes the form ay aglax! = — ‘where (at) is the matrix representing A = A, with respect to ¢ = (2/@x), 1 = i = n. Asa special case of Proposition 3.3 (2) (or as can be directly verified), we have g(de,, ¢)) = h(ey ¢), that is, where (g) is the inverse of the matrix (gy,). Example 3.5. Continuing Example 3.4, we shall reconsider the spherical map $: x € M — é, €$" defined in §2. For each point xe M, the differential (4,), is a linear mapping of T,(M) into Vil. SUBMANIFOLDS 19 (S*). Let us denote by p, the natural linear isomorphism of (BR) onto Ty(R'); observe that p, maps T.(M) onto T,c(S*). We show that the linear transformation p;!+ (6%), of T/{M) into itself coincides with —A in Example 3.4. In fact, each ¢, = (8/@x") is mapped by (J4)» upon the vector in Tyn(S*) represented by the vector (@f/@x'), in R", Thus p."> ($y) aty is also represented by (8/2), By (II) in Example 3.4, this means that 9! ($e). maps ¢ upon —Ef., afe. Thus our mapping coincides with —A. We may therefore say that, through the identification by fo —A, is nothing but the Jacobian of the spherical map. We may also relate the spherical map to parallel displacement on M (cf. Theorem 2.3). Let x(t) be a curve on M and let X, = 2(t) be the field of tangent vectors of the curve. Suppose that ¥Y, is a field of vectors that arg parallel along x(¢) on M. Then p,q) * Y; is parallel along the curve 4(x(t)) on S*. To prove this, we write aY,Jdt = Vx,¥, + WXu Yi) Say = WX ¥,) Sec since Vx,¥, = 0. This means that dY,/dt is normal to M along x(t). Since fu) * Y; is represented by the same R"-valued vector function of tas that for ¥,, and since faa * Exe i8 equal to the unit normal vector to S* at $(x(t)); it follows that d(py * Y,)/dt is normal to $* along $(x(¢)). This proves that pg * Y, is parallel along ¢(x(1)) on S*. We shall conclude this section by expressing the second funda- ‘mental form of M in terms of the canonical form and the connec- tion form of the bundle O(N, M) Let M be an n-dimensional manifold immersed in an (n + p)- dimensional Riemannian manifold M. Let and ye the canonical form and the Riemannian connection form on O(N), respectively. We define an R?-valued quadratic form & on O(N, M) as follows: &(X, Y) = the Rr-component of »(X)9(¥), where X and ¥ are tangent vectors of O(N, M) at a point ve O(N, M). Take the natural basis for R"+? so that the first n vectors of the basis span R" and the last p vectors span R?, Then we can write and y in matrix forms (p) and (pd), respectively. We shall use the convention that indices 4, B, ... run from 1 ton + p, indices20 FOUNDATIONS OF DIFFERENTIAL GEOMETRY i,j, --., run from 1 ton, and indices r, 5, ..., run from n + 1 to n+ p. We know by Proposition 1.1 that, restricted to O(N, M), the forms g* vanish identically. The first structure equation of O(N) restricted to OW, M) yields 0 = der = —Z ving Itfollows that, restricted to O(WV, M), yjcan be written as follows: w= ZAig! with Ay = Ae Hence @ = (a) can be written as follows: # = Aig'p. Proposrrton 3.5. The second fundamental form a of an immersed manifold M in a Riemannian manifold N is related to the form & on O(N, M) by a(nX, n¥) = 0(&(X,¥)) for X, Ye TAO, M)), where 1 denotes the projection O(N, M) — M. Proof. Since the statement to be proved is local, we may assume that O(N, M) admits a cross section. We extend the given vector Ye T.(O(N, M)) to a vector field ¥ on O(W) | M such that it is invariant by the action of the structure group O(n + p) and that it is tangent to O(N, M) at each point of O(N, M). To construct such a vector field ¥, take a cross section @ of O(N, M) and extend first the given vector to a vector field on o(Af) and then extend it further to a vector field on O(N) | M by translating it with the action of O(n + p). The restriction of Y 10 O(N, M) will be denoted by the same letter ¥. Then g(Y) may be considered as an R'+¥-valued function defined on O(N) | M cor on O(N, M) according as ¥ is considered as a vector field defined on 0() | M or on O(N, M). From the definition of the canonical form g, we have AY) a(Y,)) for we OL) | M. (We denote by = the projection of O(N) | M onto M as well as the projection O(N, M) —> M). We are now in a position to apply the Lemma in the proof of Proposition 1.1 of Chapter III (p. 115, Volume I); @(Y) will play the role of f there. VI, SUBMANIFOLDS al Extend 7X € Typ(M) to a vector field on M and denote by X's horizontal lift to O(.V) | M with respect to the connection defihed by the restriction of y to O() | M. Then (Vex) ater = V(X P(L)))e)- Similarly, denote by X* the horizontal lift of rX’ to O(N, M) with respect to the connection defined by the (0(n) + 0( p))- ‘component of the restriction of y to O(N, M). Then (yet Pet = 2X @(E) 0) Since #X = m(X/) = 2(Xf), we have Vigal — Viger ¥ = o(((X" — X*)(0(¥))))- Since p(X"), = 0, we have y(X" — X*), = —p(X*)., We set A= y(X" — X4), = —v(X*), € ol + A) ‘The fundamental vector field A* on O(N) | M corresponding to A coincides with the vertical vector field X’— X* at v, To evaluate (X’ — X*)(p(¥)) at » we consider therefore 4*(p(¥))- By Proposition 3.11 of Chapter I, we have A*(p(¥)) = ¥(@(4*)) + @(L4*, YI) + 2de(A®, ¥)- Since A® is vertical, we have @(4*) = 0. Since Y is invariant by O(n + p), we have [4*, ¥] = 0. Hence A*(g(¥)) = 2dp(A*, Y) = —¥(A*)9(¥)- We evaluate the equality above at v. Since A? = (X" — X'*), and y(d*) = A = —y(X*),, we have (X= XN). = WEN = GE MN) Hence Vign¥ = Vygn¥ = o(((¥" = X*)(@(Y)e) = R(X, Ye). Aigig’, it follows that (&(X*, Y)). = ((X, YJ oY) + v(V)e(A*) Since # = E,,, By definition, a(nX, nY) = Vigr¥ — Vixr¥e Hence, a(nX, n¥) = 0(8(X, ¥))- QED.22 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 4, Equations of Gauss and Codazzi Let M be an mdimensional Riemannian manifold which is isometrically immersed in an (n + p)-dimensional Riemannian manifold N, We shall first find a relationship between the curva- ture tensor fields of M and N. Since the discussion is local, we choose # orthonormal fields of normal vectors Gy ..., & to M, Let # be the corresponding second fundamental forms and let 4, = 4,,. Using the formulas of Gauss and Weingarten we obtain for any vector fields X, ¥, and Z tangent to M Vig(Vj-Z) = Ve(VeZ + DAY, Z)E) = Vx(¥rZ) + FAX, Vy Z)E FDNY, ZB + THY, Z){—4(X) + Dxbdy Vx(VrZ) — SHY, Z)A(X) + LX MY, Z) +A, Ve Z)VE + ZAM, Z)Dxby where the summation extends from I to p. In the last expression, the first two terms give the tangential component, and the last ‘two terms the normal component. For Vi-(VxZ) we may simply interchange X and ¥ in the equation above. We have also VixniZ = YoxnZ + ZAG ¥] Z)& = VexaiZ + L(V x, Z) — (VX, Z)}is by virtue of [X, ¥] = Vx¥ — Vy-X on M. Using these equations, we find that the tangential component of BUX, Y)Z = Vx(Vp2) — Ve (Vx2) = Vix is equal to R(X, Y)Z + SMX, Z)AAY) — HY, Z)4(X)}- If Wis tangent to M, then we get 8 (R(X, Y)Z, W) = g(R(X, ¥)Z,W) + MX, ZY, W) — HCY, Z)A(X, W)} = (R(X, ¥)Z, W) + a(a(X, Z),2(¥, W)) —a(al¥, Z), a(X, W)), VI, SUBMANIFOLDS 23, since, for example, $ SAX, Z)e(A(Y), W) = SAX, ZY, W) = a(e(X, Z), a(¥, W)) because of orthonormality of &, ..., & Thus the relationship between the Riemannian curvature tensors of N and M (ch. §2 of Chapter V) is given by Prorostrion 4.1 (Equation of Gauss.) RW, Z, X,Y) = RW, Z, X,Y) + g((X, Z), a(¥, W)) — ala(¥, Z), a(X, W)), where X,Y, Z, and W are arbitrary tangent vectors to M. If we wish, we may state the equation of Gauss in terms of the curvature transformations as follows. For X, ¥, Ze T,(M), there is a unique element in T,(M), which we denote by B(X, Y)Z, such that 8(B(X, ¥)Z, W) = g(a(X, Z), a( ¥, W)) — gla(¥, Z), a(X, W)) for every W € T.(M). It is obvious that B(X, ¥)Z is trilinear in X, ¥, and Z, and that B(Y, X) = —B(X, ¥). The equation of Gauss says that the curoature transformation R'(X, Y) followed by the projection: T,(N) — T,(M) is equal to R(X, ¥) + B(X, ¥)- Coroutary 4.2. If N i of constant sectional curvature k, then “R(X, YZ = Kel, Z)X — g(X, Z)¥} — BUX, Y)Z. In particular, if N = R°*® (with flat metric), then R(X, ¥) = —B(X,¥). Proof, This follows from the expression of R’ given in Corol- lary 2.3 of Chapter V. QED. Example 4.1. Ifthe codimension p is 1, that is, if M is a hyper- surface of N, then we have BUX, Y)Z = A(X, Z)AY — A(Y, Z)AX = (AX, Z)AY — g( AY, Z)AX,By FOUNDATIONS OF DIFFERENTIAL GEOMETRY In particular, if N = R", then we have R(X, Y)Z = g(AY, Z)AX — g(AX, Z)AY. This is the classical equation of Gauss (for n = 2). Note that the right hand side is independent of the choice of a unit normal field. Example 4.2. Let M be a hypersurface of R'*!, For a choice of a unit normal field &, A = 4, is a symmetric transformation of T,(M). Thus there exists an orthonormal basis Xy,..., X, in T,{M) such that AX, = 2,X,, 1S i Sn, where Ay» ‘are the eigenvalues of 4. For any pair (i,j), where i and with origin 0 in R"+4. If we choose the outward unit normal , = x/||x] for x ¢ M, then ~ (1/7) by Example 3.5, where I denotes the identity transformation of 7,,(M). By the classical equation of Gauss, we have RX, Y)Z = (Ifr*)(g(¥, Z)X — aX, Z)¥), which shows that $*(r) has constant sectional curvature I/r*, (This result was established in Theorem 3.2 of Chapter V by a different method; our present method is classical and more elementary.) Let us now look at the normal component of R'(X, Y)Z for an arbitrary submanifold M in N. Using the expression for Vie(V-Z), V,(Vi-Z), and Vix; Z and observing that X-H(Y, Z) — VAY, Z) — HUY, VxZ) = (WANN 2), Vil. SUBMANIFOLDS 25 we see that the normal component of R’(X, Y)Z is equal to Fvsh WZ) ~ (WAN ZB + LAY, Z)Dxk (% Z)Dv ed. Actually, we may give a simpler expression to the normal vector above. For the second fundamental form a, we define the co- variant derivative, denoted by V2, to be (¥x0)(¥, Z) = Dx(al¥, Z)) — a(Vy¥, Z) —a(¥, VyZ). (This is the covariant derivative of 2 with respect to the connection in T(M) + T(M)+ obtained by combining the connections Vx in T(M) and Dx in T(M)*; see Proposition 6.3 of Chapter If and Proposition 1.2 of the present chapter, although we shall not elaborate on this matter.) Using $1, ... &,, we have (Wxa)(¥ Z) = Do( EM, Z)B) —Z (VY, Z) + AY, PZ), = DX-w(Y, Z)8, + THY, Z)Dxk LW RL, Z) + HUT RZ)E =Z(VekUY, Zi + THY, Zeke It follows that the normal component of R'(X, ¥)Z is expressed by (Vx2)(¥, Z) ~ (y2)(X, Z). We have thus Prorostrion 4.3 (Equation of Codazzi). The normal component of R(X, ¥)Z is given by (¥xx)(¥, Z) — (Vra}(¥, Z) = TUT HI Z) — Crh ZY, + TM (Y, Z)Dxé, — A(X, Z)Dy by. Coroutary 4.4. If N is of constant sectional curvature, then we have (Wxa)(¥, Z) = (Wy2)(X, Z). Proof. By Corollary 2.3 of Chapter V, we know that &’(X, ¥)Z is tangent to M; hence its normal component is 0. QED.26 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Example 4.3. If NV = R"** and if M is a hypersurface, then we have Dyé = 0, and hence (VxA)(¥, Z) = (Veh) (X, Z), or, equivalently, (FxA)(¥) (WyA)(X). This is the classical equation of Codazzi (when n = 2). Remark. At the end of §3, we showed how the second funda~ mental form « can be expressed in terms of the canonical form @ and the connection form y on the bundle O(N, M). We shall indicate how the equations of Gauss and Codazzi can be derived from that viewpoint. Using the same convention as in §3 that 1S4,B,...5n+p, while 1 Sij,...snandnt+1sr, “+p, we express the second structure equation of N as Y= dvi + Sut awh + Dad An where (Vf) is the curvature form of N. Setting A =iand B =j and restricting this second structure equation to O(N, M), we obtain ¥=U+ Dery, where (91) is the curvature form of M (lifted from 0(M) to O(N, M)). Since yf = —yf = —¥, Ajg* and yj = E, Aj’, we have Vim 4S Aids — Ande ag This is equivalent to Proposition 4.1 (Equation of Gauss). Similarly, setting A = s and B = j in the second structure equa- tion of N and restricting it to O(N, M), we obtain an equation equivalent to Proposition 4.3 (Equation of Codazzi). In the rest of the section, we shall prove some theorems involving sectional curvature. First we prove preparatory results, Propostrion 4.5, Let M be an n-dimensional submanifold immersed in an (n + p)-dimensional Riemannian manifold N. Let X and ¥ be a 355044 VIL SUBMANIFOLDS 7 pair of orthonormal vectors in T,(M), where x © M. For the plane X n shanned by X and Y, we have * ky(X An ¥) = kylX 4 ¥) + gla(X, ¥), a(X, ¥)) — a(x(X, X), @(¥, ¥)), where ky (resp. kyy) denotes the sectional curvature in N (resp. M). In particular, if N = R™, then ky (X 4 Y) = g(a(X, X), a(¥, ¥)) — g(a(X, ¥), aX, ¥)). Proof. This is immediate from the equation of Gauss in Proposition 4.1, since ky(X a Y) = R(X,Y, X,Y) and ky(X a Y) = R(X, ¥, XY). QED. Propostrion 4.6. Let M be an nedimensional compact manifold immersed in R*?, Then there is a point xy € M such that a(X,X)e0 for any Xe T,(M), X40. Proof. Let y(x) denote the position vector, for the point in R™? corresponding to xe M. Let g(x) = (y(x),.9(x))/2, where (, ) is the Euclidean inner product. The differentiable func- tion g on M takes a maximum, say, at xy. Namely, x» is a point such that the point y(%) = yp is of maximum distance from the origin in R'?, The following argument is valid if g has a local maximum at x9; hence we now assume that a neighborhood U7 of x is imbedded in R"*?, and we identify x € U with y(2), so that (x) = (x, x). For any vector field X on U, Xx (that is, X applied to the vector-valued function x) is the vector-valued function which expresses X. ‘This being said, we have Xp = (X, x) and this is 0 at xp. Thus (Xap 0) = 0. Since X is arbitrary, it shows that the vector x is normal to M at xy. We have further Bp = (Wey 8) + (XX) = (2(% 2), 4) + e(% A) at rey since VieX = VxX + a(X, X). Since g has a local maximum at %qy it follows that X%p = 0 at 2». Thus we obtain (a(Xey Xay)s to) + g(May Xe,) = 9,28 FOUNDATIONS OF DIFFERENTIAL GEOMETRY that is, (2(Xeyy Xe,)s eo) S —e(Ney X,,) <0, and a(X,,, X,,) #0 for X,, # 0. This proves that «(X, X) + 0 for every non-zero Xe T,(M). QED. ‘We now state the main result, ‘Turorem 4.7. Let M be an n-dimensional compact Riemannian ‘manifold isometrically immersed in R'*”. If, at every point x of M, the tangent space T,{M) contains an m-dimensional subspace T such that the sectional curcature for any plane in T; is non-positive, then we have Proof. By Proposition 4.6, there exists a point x of M such that a(X, X) # 0 for any non-zero vector X in T,(M). Consider the restriction of a to T; x T;. By assumption, the sectional curvature k(X 0 Y) is non-positive for any plane X A ¥in T;(M). By Proposition 4.5, we have g(a(X, X), a(¥, Y)) — g(alX, ¥), a(X, ¥)) 50 whenever X and Y are a pair of orthonormal vectors in 7:(_M). Actually, the preceding inequality holds for all X and Y in 7(M) by Proposition 1.3 of Chapter V (or as can be directly verified by orthonormalizing X and ¥). Our conclusion will therefore follow from the following lemma. Lena. Let a: R x Rm > RP be a symmetric bilinear mapping and let g be a positive definite inner product in RP. If (1) (a(x 4), 2(9,9)) — g(a( 9), al) = 0 Sor all x, ER” and if (2) a(x, x) #0 for all non-zero x €R™, then we have p = m. Proof. We may extend « to a symmetric complex bilinear mapping of G™ x * -» C?, Consider the equation a(z,z) =0. Since « is C*-valued, this equation is equivalent to a system of p 4 quadratic equations: ah(2, 2) = 0,444, a%(z, 2) = 0. ‘VIL. SUBMANIFOLDS 29 Suppose p A). Thus 1/n (trace 4,) is a linear function on T,(M)+. There is a unique element in’ 7,(M)+, say, n, such that 1 (ace A) = gl) — for every €€ T,(M)+. We call y the mean curvature normal at x. If &, «.., &,i8 an ortho- normal basis in T,(M)+, and if 4, = 4,,, then (Eon) lise trace A, so that y= 13 (uace A). In particular, for p = 1 we have the definition given in Example 5.3 in the case of N = R"1, Mis said to be minimal in N (for the immersion) if the mean curvature normal vanishes at each point. We may prove the following: There is no compact minimal submanifold in a Euclidean space. From the proof of Proposition 4.6, we know that there is a point x9 of the compact submanifold such that the posi- tion vector x» is normal and such that (a(X, X), x9) <0 for every tangent vector X # 0 at xq, Ifwelet € = xq, then from Proposition 3.3 (2), we have (A(X), X) = g(a(X, X), €) <0 for every tangent vector X + 0, which implies that 4, is negative- definite. Thus trace 4, cannot be 0. Remark. Lot Ay... , Ay be the principal curvatures of a hyper- surface Mat xe M, and let Xy..., X, be the corresponding orthonormal basis for 7,(M) so that AX, = 4,X;fori = 1,...,m VIL SUBMANIFOLDS 35, ‘The Riemannian curvature tensor R may be considered as a symmetric bilinear mapping A* T,(M) x At T,(M) Rat each pohnt x ¢M. Let & be the corresponding symmetric linear trans- formation At T,(M) -+ A? T.(M). By the equation of Gauss (cf. Example 4.1) we have RIX AY) = AX A AY for X, Ye T,(M). Hence SE eee ee eee Since {X, 4 X,; 1 0, then every point of M is umbilical and M is locally a Aypersphere. Proof. We first prove (2) and (3). If S = pg, then by Propo- sition 5.2 we have (trace A)A — 4? = pl, where J's the identity transformation. At a point x of M we take an orthonormal basis X,,..., X, in T,(M) such that AX, 14,X,, 1 & i 0, we show that all 4,’s are equal. Suppose that 4, # 4,. Then the other 4's are equal to A oF Ay. ‘Thus we assume that 4, appears p times and A, appears q times (so that n =p + q) among the eigenvalues of 4. Then we have from the quadratic equation Ay t+ Wa = 8 = phy + ghey that is, (p — 1)ar + (9 — Ia = 0 and dyh = p > 0. nam 5 I eee Vil. SUBMANIFOLDS 37 The second relation shows that 4, and Z have the same sign. ‘Thin the first relation implies that p = 1, q = I and hence n = 2, contrary to the assumption. This proves that all 4,’s are equal (alld not equal to 0, because p # 0). ‘We now prove (1). Consider the equation () Wp—sh+p=0, wherep <0. Ifall A's are equal, then s = nA, and hence p = (n — 1)28, which contradicts the assumption p <0. Thus the equation above has two distinct roots. Assume that the eigenvalues of 4 are given by aha =A and dy where Atm=s phe (n=p~uas so that (o-DA+ @—p— Dye and ay From these relations we obtain (2-0 +p = Dp If p =1, then (n —p — I)p =0, and hence n=p +1 which is not the case. Thus p # I and we obtain 2) B= —pln—p —Vi(p = 1). ‘The argument above applies to each point x of M. Sinces = trace A is differentiable and since the equation (1) has two distinct roots at each point, it follows that the two roots 4 and y are differentiable functions. From the equation (2) we conclude that 1 is a constant function (since M is connected), and so is x. It abo follows that » and n — 9 are also constant. We define two distributions A, and A, on M as follows: + A) = (Ke TUM); AX = ax} Ad(s) = (Ye T,(M); AX = pX}. We shall show that both A, and A, are differentiable and in- volutive, and that M is locally a direct product of the maximal integral manifolds M, and M, of A, and A, as a Riemannian38 FOUNDATIONS OF DIFFERENTIAL GEOMETRY manifold. This will finish the proof, because if Y and Y are tangent to M, and M,, respectively, wehave 0 = R(X 4 Y) = AX AAY = auX 4 Y, which implies 4p = 0, contrat Ap <0. First, to prove that A, and A, are differentiable, let X,, -. +, X,, Xpuy .++, X, be differentiable vector fields such that X,, 1Si Sp), and X, p +1555 n, form a basis of A(x) and Ag(%p), respectively, at a point 2». We define vector fields Yi, ..., Y, by Y=(4-WXy 1818p, and Y,=(4-4X, p+isjsn Since (4 — A), = (A — a)(A — p)X, = 0 for 1s i's p and (A = WY, = (4 — wld — DX, = 0 for p +1 sj sn, we see that Yi, 1's iS p, belong to A, and ¥,, p +1 0.) QED. Corollary 5.4 is valid for the case n = 2 as well, but the proof is more difficult (see Note 15). Both the second fundamental form and the Gaussian curvature are closely related to the convexity of a hypersurface. A hyper- surface M in RM is saith to be conver at a point x eM if the hyperplane H, of R'+4 tangent to M at x does not separate a neighborhood of x in M into two parts. Moreover, if is the only point of a neighborhood which lies on H,, then M is said to be strictly convex at x. If, for every x € M, H, does not separate M into two parts, then M is said to be convet. Moreover, if, for every x eM, x is the only point of M which lies on H,, then M is said to be strictly convex. A convex hypersurface is always orientable. Choosing at each point x of M the unit normal vector pointed outward (i.e., to the opposite side of M with respect to H,), we obtain a continuous field of unit normal vectors. ‘We say that the second fundamental form a of a hypersurface M is definite at x6 M if a(X,X) #0 for all non-zero vectors Xe T,(M). If we choose a unit normal vector £ at x and write @ =-Aé (cf. §3), then ais definite if and only if the classical second fundamental form Ais either positive or negative defin'te. Similarly, wwe say that a is non-degenerate at x if h is non-degenerate at x. Propostrion 5.5. A hypersurface M in R** is strictly convex at 4 point x if its second fundamental form a is definite atx. Proof. Fixing the point x, we choose a Euclidean coordinate system}, ...,y"+1 for RR" such that 4y"*? = O defines the tangent space T,(M). Then y**#, considered as a function on M, is critical atx. Its Hessian at xis nothing but the second fundamental form of M at x (cf. Examples 3.3 and 3.4). If « is definite at x, then y*#, as a function on M, attains an isolated local minimum or maximum. Hence a neighborhood of x in M lies strictly in one side of the tangent hyperplane at x. QED. VII. SUBMANIFOLDS 41 Taeorem 5.6. For a connected compact hypersurface M in Ret (n & 2), the following conditions are equivalent: Fy Fee con findamel form sof Ms tafe eeyutere ox Ms (2) M is orientable and the spherical map of Gauss M —» S” is a diffeomorphism; (3) The Gaussian curvature K,, of M never vanishes on M. Moverover, any one of the conditions above implies that M is strictly concer, Proof, (1) (2). At each point x €.M we choose a unit nore mal vector £, in such a way that the classical second fundamental form h defined by a(X, ¥) = A(X, ¥)&, for X, Ye T,(M) is negative definite. Then ¢ is continuous and hence M is orientable. Ata point where a is non-degenerate, the Jacobian of the spherical map M->S" is non-degenerate (cf Example 3.5). Since M is compact, the spherical map M > S* is a covering projection by Corollary 4.7 of Chapter IV. Since S* is simply connected, the spherical map is a diffeomorphism. (2) + (3). Since the Jacobian of the spherical map Ms is non-degenerate everywhere, so is the second fundamental form (cf. Example 3.5). Hence K, # 0 everywhere. (3) + (1). Since K, #0 everywhere, a is non-degenerate everywhere. Since M is compact, there is a point x) € M such that a@(X,X) 40 forall Xe T,,(M), X40 by Proposition 4.6. Since a is definite at x, and is non-degenerate everywhere, « is definite everywhere. ‘To prove the last assertion, let x be any point of M, and choos a Euclidean coordinate system y*, .. ., y"*! of R'* such that the tangent hyperplane H, is given by 7**1 = 0 and a neighborhood of x in M lies in the region y"*? <0, Let x* be a point of M where the function y"* attains its maximum on M. Then H,. is parallel to H,, and the outward unit normal vector at x* is parallel to the cone at x (in the same direction). Since the spherical map is one-to- one, we have x* = x. Hence M — {x} lies in the region y"*" <0. QED. ‘The last assertion of Theorem 5.6 is due to Hadamard [1]. Chern and Lashof [1] proved that a compact surface in R® with Ky 2 is convex and constructed for n Z 3 a non-convex compact42 FOUNDATIONS OF DIFFERENTIAL GEOMETRY hypersurface M in R* with K, = 0, For hypersurfaces with K, = 0orX, = 0 which are not necessarily compact, see Hartman and Nirenberg [1] and Nirenberg [3]. (See Note 15 for these and other results.) 6. Type number and rigidity Let M be a hypersurface immersed in R**, At each point x of M, the type number of M at x, denoted by t(x), is defined to be the rank of the linear transformation 4 of T.(M). Of course, it is determined independently of the choice of the field of unit normals E because A, and A, have the same rank. By Example 3.5 we know that ¢(x) is the rank of the Jacobian of the spherical map $: MS", which is at least locally defined. We shall prove Tuones 6.1. For a hypersurface M immersed in R™%, (1) (x) #50 or 1 ff and only if R = 0 at x. (2) Iftls) = 2, then t(x) =n — dim TS, where Tz = (Xe TAM); R(X, Y) = 0 for all ¥ © T{M)}. Proof. (1) If ¢(x) = 0, that is, A =0, then the equation of Gauss implies that R = 0 at x. Suppose (x) = 1. Then there is non-zero, vector We T,(M) such that for each X ¢ T,(M) we have AN = cW, where cis a scalar depending on X. The equation of Gauss again implies that R(X, ¥) = 0 for all X, ¥. Conversely, suppose that R = O atx, If¢(x) & 2, there exist a pair of non-zero veetors ¥ and Yin T,(M) such that AX = 4X, AY = pY, where 4, w# 0, and such that g(X, Y) = 0, We find that BRK, Y)Y, X) = tH, which is a contradiction to the assumption R = 0. (2) Let 73 be the null space of 4, and let T;, be the orthogonal complement of 7. Since A is symmetric, we may easily see that A maps T; onto itself in a one-to-one manner. If X ¢ 77, then the equation of Gauss implies that R(X, ¥) = 0 for all Ye T,(M), that is, Xe T?. Now assuming dim T: = ¢(x) 2 2, we prove that Ty = T?. Suppose there is an Y ¢ T# which is not in T7. Since X= X,"+ Xp where Xe 7; and ke 7] ¢ Tt, we sce that ‘VIL, SUBMANIFOLDS 43 Xe TE 9 T,, Since dim T; z 2, there exists a Ye Ti such that %, and Y are linearly independent (and so are AX, and AY). By the fquation of Gauss we have R(X, Y)AY = g(AY, AY)AX, — g(AX,, AY) AY. Since X, € T+, we have R(X,, ¥) = 0, that is, the right hand side is also equal to.0. Since g(AY, AY) + 0, this relation means that AX, and AY are linearly dependent, which is a contradiction. We have thus shown that 7; = 7. Hence ¢(x) = dim T! dim T; =n — dim Tt, completing the proof of Theorem 6. QED. Theorem 6.1 shows that, unless M is locally Euclidean, the rank of the second fundamental form is independent of an im- mersion, We shall prove that under a stronger assumption the immersion itself is unique (up to an isometry of RY), First we show that the second fundamental form is uniquely determined (up to a sign) at each point. Tueones 6.2. Let M be an redimensional Riemannian manifold and let f and f be isometric immersions of M into RM, If the type number t(x) of the immersion f at x is %3, then the second fundamental form h for f and the second fundamental form h for f coincide at x up to a sign. Proof. Let us first remark that, by virtue of Theorem 6.1, R is not 0 at x, and hence the type number /(x) for fis equal to f(x). ‘As we see from the proof of Theorem 6.1, the null space of A and the null space of 4 coincide with TS = {X; R(X, Y) = 0 for all Ye T,}. This means that in the decomposition 7,(M) = Ty + T? in that proof, 77 is the null space for A as well as for 4, and both 4 and Fmap 7 onto iuelf in one-to-one manner, This being said, consider the exterior product TA T; and introduce an inner product s such that S(XA¥, ZnW) = g(X, Z)g(¥, W) — g(X, W)g(¥, Z). ‘The equation of Gauss implies that B(R(X,Y)Z, W) = s(X a Y, AZ 0 AW) =s(X 0 ¥,4Z 9 AW).44 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Since s is positive definite on T; a T;, we have (1) AZNAW =AZNAW forall Z, We Ty We shall now prove that, for each Xe T;, AX = cAX for some ¢ (this constant ¢ may depend on X, although we shall eventually show that ¢ = 1 independently of X). Since AX # 0 for X #0. it is sufficient to show that AX and AX are linearly dependent, Suppose that they are linearly independent so that AX a AX # 0. Since dim T? = i{x) 2 3 by assumption, we may choose a Ye T! such that AX, 4X, AY are linearly independent (and hence AX NAY NAX #0). By property (I), we have AXAAY = AX AAY so that AX » AY 4 AX = AX » AY 2 AX = 0, which is a contradiction. This proves our assertion that AY = cAX for some c. Let X,,..., X, bea basis in 77, There exist c,,...,¢-such that AX, = cAX, for 1 Sir. At the same time, A(X, + X,) = A(X, + X,) for some c. We have then A(X, +X) = AX, + AX, (AX, + AX). For any i#j, 1 g(Z, X) — @(Z*, [% ¥]) = e(VxZ", ¥) — e(VyZ", X). Hence da? = 0 if and only if 8(V x2", Y) = aX V2")50 FOUNDATIONS OF DIFFERENTIAL GEOMETRY for all vector fields X, Y. since this last condition is indeed satisfied, a(VxZ?, ¥) = g(GAX, ¥) = eX, SAY) = a(X, VyZ") by virtue of (5) and of the symmetry of A. In Lemma 2 it is clear that, given any point, say, 94 in R™, we may choose an isometry f which maps x, upon yp. ‘Thus it follows that in order to prove the existence of an isometric imbedding in ‘Theorem 7.1 it is sufficient to prove that the system of equations (5) and (6) have a solution with any arbitrary initial conditions. The uniqueness follows from the uniqueness of the solution of (5) and (6) with a preassigned initial condition, which is almost obvious. We shall now prove Lenoma 3. Let xy be a point of M and let U be a neighborhood with normal coordinates , «2%, |x! 0 such that if ls = < 6, then x(s) € U,, and f, and its associated field of unit normals coincide with f, and its associated field of unit normals in a neighborhood of x(s); (3) fa coincides with fy in a neighborhood of x», and so do their associated fields of unit normals. If a continuation exists along a curve, it is unique. More precisely, iff, and f; are continuations of fy along x(t) with fields of unit normals € and ¢', respectively, then each point x(t) has a neighborhood on which f, = f’ and § = ¢”. This assertion follows from the uniqueness part of Theorem 7.1. ‘Now a continuation exists along any curve x(t). Let fy be the Vi. SUBMANIFOLDS 53 supremum of f, > 0 such that a continuation of for exists for 0 ¢ t <4, Let W be a convex neighborhood of x(f,) such that evbry point of W’ has a normal neighborhood containing HW” (cf. ‘Theorem 8.7 of Chapter IIT). Take f, < fq such that x(¢,) « W and let V be a normal neighborhood of x(j) which contains W. By the existence part of Theorem 7.1 we may extend the imbedding fy, to an allowable imbedding f' of V. For a 6 > 0 such that x(0) € W for St L(N) | M+ LN) M L(M) induces a connection in L(M) by Proposi- tion 6.1 of Chapter II. The connection form @ on L(M) so determined is related to the connection form y on L(N) by Be) = h(i* ©j*y), where js i: L(N, M) + L(N) is the natural imbedding, and A on the right hand side is the natural homo- morphism of gl(n + 2, n; R) onto gl(n, R). Propostrion 8.3, Every auto-parallel submanifold M of N is totally geodesic. Proof, Let + be a geodesic of NV starting from a point x of M and tangent to M at x. Let +’ be the geodesic of M (with respect to the induced affine connection of M) starting from x and tangent to + at x. Then +’ is a geodesic of N, Indeed, if +” = x, then the vector field &, along 7’ is parallel with respect to the connection of M by definition of a geodesic. It follows that this vector field is parallel with respect to the affine connection of . 29. ‘VIL, SUBMANIFOLDS 57 Thus +’ is a geodesic of N. Since any two geodesics tangent to each ottigr at» point must eoincse, we have r — 1 that i + i ia QED. ‘The following result is in E. Cartan [8; Chapter V]. Tueorem 8.4. Let N be a manifold with an affine connection with zero torsion. Then every totally geadesic submanifold M of N is auto- parallel. Proof. For a given point of M we choose V and x}, . x and define U as in the proof of (2) -> (3) in Proposition 8.2. For any xe U, we take a geodesic x; such that x = x and i Zp, a'(9/ax"), (that is, tangent to M at the initial point x), where a}, ..., a" are arbitrarily fixed. Since M is totally geodesic by assumption, the geodesic x, stays in M for small values of t. The equation (cf. Proposition 7.8 of Chapter IT) reduces, forn +15 1s.n +p, to z de! dst Eto In particular, we have for t = 0 E Pylxdaict = 0. Since the torsion is 0, we have fs, =z, Since a}, ..., a" are arbitrary, we may conclude [i,(s) = 0 for 1S j, kS.n and n+157sn-+p. This is valid at every point ¥ of U. Thus Vaun(8/2x!), 1 < i,j Sm, are tangent to U at every point of U. Ivfollows that if ¥ and ¥ are vector fields on U (or on M), then VY is tangent to U (or M) everywhere. By Proposition 8.2 we conclude that M is an auto-parallel submanifold of M. QED. Let M be a submanifold of N. For any covariant tensor field R on N we speak of its restriction to M which is defined in a natural way. For a tensor field X of type (I, s) on N, we shall say that K can be restricted to M if, by considering K as an s-linear mapping X(N) x -++ x X(N) “»3() in the manner of Proposition 3.1 of Chapter I, K(¥, ..., 2%) is tangent to M at every point of /Mf whenever the vector fields %, ..., Z, are tangent to M at every58 FOUNDATIONS OF DIFFERENTIAL GEOMETRY point of M. Under this condition it is obvious that K induces a tensor field of type (I, 5), called the restriction of K to M, in the natural fashion; at each point x of M, K, is an s-linear mapping of T,(M) x +++ x T,(M) into T,(M) which is the restriction of the s-linear mapping K of T,(N) x +++ x T,(N) into T.(N). Now we have Prorosrion 8.5. Let N be a manifold with an affine connection Y and let M be an auto-parallel submanifold of N with induced afine connection V. (1) PX and ¥ are vector fields on M, then GY is tangent to M at every point of M and Vx¥ = Vx¥ on M. (2) If Kis the restriction to M of a covariant tensor field K on N, then ‘UK is the restriction to M of VR. (3) Let R be a tensor field of type (1, 8) on N which can be restricted to M and let K be its restriction. Then VR can be restricted to M and its restriction coincides with UK. Proof. (1) is contained in Proposition 8.2. To prove (2) for K oftype (0, 5), let X, Xy.... ,X, be vector fields on M and extend them to vector fields X, X,, ..., X, on N; actually, we do this locally and that is sufficient for our purpose. Then (PRX, Xs X) (Ry eee BN) — ERB Vela »%) and its value at x € M is equal to the value of (PK)(Xy oy XX) = Vix (K(Xyy 0 AQ) — ER My es Vad ee) at x by virtue of (1). The proof of (3) is quite similar. QED. Proposrrion 8.6. Let N and M be as in Proposition 8.5. (1) The torsion tensor field T and the curvature tensor field R of N can be restricted to M and their restrictions are equal to the torsion tensor ‘field T and the curvature tensor field R of M, respectively. (2) The m-th covariant differentials VT and VR can be restricted to M and their restrictions are equal to V"T and VR, respectively, where m is an arbitrary positive integer. ‘VIL. SUBMANIFOLDS 59 Proof. We first prove (1). If ¥ and F are vector fields on N whith are tangent to Mat every point of M, then Ve¥, Vp and LZ, F] have the same property. Thus 7(X, f) is also tangent to ‘Mat every point of M and is equal to T(X, ¥) on M. The proof for Ris similar. (2) follows from (1) and (3) of Proposition 8.5. QED. Coronary 8.7. Let M be an auto-parallel submanifold of a manifold N with an affine connection. Each of the following properties for Nis inherited by M (with induced affine connection): (1) the torsion tensor is zero; (2) the curoature tensor is zero; (3) the torsion tensor field is parallel; (4) the curvature tensor field is parallel In the Riemannian case we have Proposrrion 8.8, Let N be a Riemannian manifold. For a sub- manifold M of N, the following conditions are equivalent. (1) M is auto-parallel; (2) Mis totally geodesic (3) the second fundamental form of M is identically zero. When M is auto-parallel, the Riemannian connection of M with respect to the induced Riemannian metric coincides with the induced affine connection. Proof. By definition of the second fundamental form a in §3: Vi¥ =Va¥ +(X, ¥), wwe see that a is identically 0 if and only if ¥xY is tangent to M whenever X and Y are tangent vector fields on M. Thus the muivalence of (1), (2), and (3) follows immediately from Propo- ion 8.2, Proposition 8.3, and Theorem 8.4. Also when a is identically 0, we know that the affine connection ¥ induces the affine connection V on M, which is the Riemannian connection for the induced Riemannian metric on M by Proposition 3.1. QED. Turorem 8.9. Let M be an auto-parallel submanifold of a Rie- mannian manifold N. Let X be an infnitesinal isometry of NV. At each point of M, decompose X into a vector tangent to M and a vector normal to M. Then the tangential component of X is an infinitesimal isometry of M. Proof. We first prove the following60 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Lemma. Every normal vector to M remains normal under the parallel displacement along any curve contained in M. Proof. Let 7 be a curve in M. Let X and Y be vector fields of N along = parallel with respect to the Riemannian connection of N. If g is the Riemannian metric tensor of N, then g(X, Y) is constant along r. If Y is tangent to M ata point, then Yis tangent to M at every point of + as M is auto-parallel. If X is furthermore normal to M at a point, then the constant function g(X, Y) vanishes everywhere along r. Hence, X is normal to M at every point of r. This proves our lemma, At each point of M, we decompose an infinitesimal isometry X of N as follows: Kax yx where X’ is tangent to M and X" is normal to M. Let g denote also the induced Riemannian metric tensor on M. By Proposition 2.5 and Proposition 3.2 of Chapter VI, it suffices to prove that (VpX', Y) = 0 for all vector fields ¥ of M. By the same proposi- tions, we have g(VyX, Y) = 0. Hence, we have 0 = a(VrX, Y) = (Vy X", ¥) + e(TrX", ¥). By the lemma above, VX" is normal to Mso that g(Vy-X", ¥) = 0. Hence, g(VyX", Y) = 0. QED. Conortary 8.10, Every closed auto-parallel submanifold of a homogeneous Riemannian manifold is homogeneous. Proof. Assume Nis homogencous and M is auto-parallel in N. Given a tangent vector of M, there exists an infinitesimal isometry X of NV which extends it as N is homogeneous. The tangential component X" of X in Theorem 8.9 is an infinitesimal isometry of M which extends the given vector of M. On the other hand, every Cauchy sequence of M is a Cauchy sequence of V and, being homogeneous, NV is complete (ef. Theorem 4.5 of Chapter IV). Hence, M is also complete. By Theorem 2.4 of Chapter VI, X’ generates a global 1-parameter group of isometries of M. This shows that the set of all X’ thus obtained generates a transitive Sroup of isometries of M. QED. Remark, Evidently, the above argument shows that the as- sumption that M is closed can be replaced by the one that M is complete. Vit, SUBMANIFOLDS ot We shall concludé this section by an important example of auto-parallel submanifold. Eljample 8.1. Let N be a manifold with an affine connection and let G be any set of affine transformations of . Let F be the set of points of IV which are left fixed by G. Then cach connected component M of F is a closed auto-parallel submanifold of N (provided that F is non-empty). Indeed, let x € Mf. Then every clement of G induces an endomorphism of T_(V). Let V be the subspace of T,() consisting of vectors which are left fixed by the endomorphisms of T,(1V) induced by G. Let U* be aneighborhood of the origin in T(N) such that the exponential mapping exp,: Ut — N is one-to-one. Let U = exp, (U*). It is easy to see that U 0 F = exp, (U* 0 V). Since U A Fis a neighborhood of xin F and exp, (U* 9 V) is a submanifold of N, M is also a submanifold of NV. Evidently, M is closed in N. If X is a’parallel vector field along a curve + in N, then every affine transformation fof N maps X into a parallel vector field along f(r). Assuming that r isin M and Xis tangent to M ata single point of 7, we shall show that X is tangent to M at every point of 7. Since + is in M, f(s) =r, and hence f(X) is a parallel vector field along 7 for every fe G. Since f leaves M fixed, X and f(X) coincide at the point where X is tangent to M. ‘The uniqueness of the parallel displacement implies that X and f(X) coincide at every point of 7. ‘Thus, X is invariant by every element f of G. This implies that X is tangent to M at every point of +, thus completing the proof of our assertion. Similarly, if F is the set of common zeros of any set of infinitesimal affine transformations of N, then each connected component M of F is a closed, auto-parallel submanifold of N. We have only to apply the same argument to the set of local -parameter group of local affine transformations of N’ generated by the given set of infinitesimal affine transformations,CHAPTER VIL Variations of the Length Integral 1. Jacobi fields Let M be an n-dimensional manifold with an affine connection and let T and R be the torsion and the curvature tensor fields on M. A vector field X along a geodesic + = x, of M is called a Jacobi feld if it satisfies the following second order ordinary linear differential equation, called the Jacobi equation: VEX + Va(T(X, 4) + RX 4)4 = 0, where %, is the vector tangent to 7 at the point x, We denote by J, the set of Jacobi fields along 7; obviously, it forms a vector space over R. Propostrion Il, A Jacobi field X along + =x, is uniquely determined by the values of X and V,,X at one point x, of r. In particular, taacal dim J, = 2n. Proof. This is a consequence of the fact that the Jacobi equa- tion isa second order ordinary differential equation. QED. We shall now give a geometric interpretation of a Jacobi field. By a variation of a geodesic r = x, 0S S 1, we shall mean a 1-parameter family of geodesics +4, —e <5 <¢, such that 7 = 7°. More precisely, it is a differentiable mapping of class C® of (0, 1] x (=e, ¢) into M, (ts) +t, such that (1) For each fixed s € (—e, ¢), x = xf is.a geodesics Q) 4#= ford stsl. An infinitesimal variation X of a geodesic + is a vector field along ‘induced by some variation +* = x{ of r in the following manner. 664 FOUNDATIONS OF DIFFERENTIAL GEOMETRY For each fixed ¢, we denote by 7, = xj, the curve described by x}, —e <5 0; x, is a conjugate point of x along x,, 0 < fs 3}, and let 6 = inf'S. Since exp, is singular at oX, itis singular at bX, that is, x, is a conjugate point of x. ‘Thus we may speak of the first conjugate point of x along 7. ‘Jacobi equations of non-symmetrical invariant connections on homogencous spaces have been considered by Chavel [1], [2] (see also Rauch [5]). . 2. Jacobi fields in a Riemannian manifold From now on we shall assume that M is a Riemannian mani- fold and that every geodesic + = x, is parametrized by its arc length unless otherwise stated. For any vector field X along r, the vector fields V,,X and V3.X along + will be denoted by X” and X”, respectively. Every geodesic r = x, admits two Jacobi fields in a natural way. One is given by #, and will be denoted by #. The other is given by 2, and will be denoted by #. It is a trivial matter to verify that # and # satisfy the Jacobi equation. Proposirion 2.1. Every Jacobi field X along a geodesic + = x, of @ Riemannian manifold M can be uniquely decomposed in the following form: Xaai+he+y, where a and b are real numbers and Y is a Jacobi field along + which is everywhere perpendicular to 7. Proof. Let g be the Riemannian metric on M and set 0 = gli X,,), 8 = gles Xi), ¥ = X — at — 0. VIL, VARIATIONS OF THE LENGTH INTEGRAL 69 Since X, , and # satisfy the Jacobi equation, so does ¥. Taking the infer product of # with the Jacobi equation satsied by Y, we obtain al?", #) + e(RUY, #4, 7) =0. Since R(Y, #) is a skew-symmetric linear transformation of the tangent space at each point (cf. Proposition 2.1 of Chapter V), the second term in the above equation vanishes identically. Hence, the first term also vanishes. From V;# = 0 and Vg =0, we obtain Sa) Thus, g(¥,#) = At +B, where A and B are constants, Since fz, = 0, we have Baa Viel, #) ss) = 8(Xay — Oayy #5) = 8( eqs fo) — 8(2% 0» 40) =a—a=0. As ris a geodesic, we have Y’ = X’ — b#, Hence, 4 — 8(b%y 4) = 6 — 5 = Fal Ao = Thus, g(¥, #) = 0. In order to prove the uniqueness of the decomposition, let Xaaepve+Z be another decomposition of X such that Z is perpendicular to 1. Forveach ¢, we have 4 = (a + b)%, + Ye, = (a! + 6%, + Z,. Since both ¥,, and Z,, are perpendicular to #,, we have and hence QED. Conortary 2.2. Ifa Jacobi field along a geodesic + is perpendicular to + at two points, it is perpendicular to x at every point of 1.70 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Proof, Suppose a Jacobi field X is perpendicular to + = x, atx, and x,. Let X = at + 67 + Yasin Proposition 2.1, Then we have O=(a+ Hk, 0 = (a +b5)& Since r #5, we havea = b = 0. QED. We shall establish a formula which will be needed in sub- sequent sections. . Proposrriox 2.3. Let X be a Jacobi field and Y a piecewise differentiable vector feld along a geodesic + = x,. Then, for any parameter values a and b of t, we have [eer 1) ~ 60a 4+ oe Ke Proof. We obtain the desired formula by integrating the elk Y) = alk) + al", Y) = (X,Y) — a(R 44, Y). QED. Conotnany 24. If g(R(Ysi)% ¥) = 0 for every vector field Y along zr, no two points of are conjugate along +. In particular, a Riemannian manifold with non-positive sectional curvature has no conjugate points. Proof. Let X be a Jacobi field aiong + which vanishes at x, and x,, In the formula of Proposition 2.3, set Y = X. Then the ‘unintegtated terms vanish, On the other hand, the integrand in the integrated term is non-negative and hence must be zero. This implies g(X’, X’) = 0. Since X vanishes at x, X' = 0 implies X=0. QED. ‘The following proposition is also for later use. Propostrion 2.5. If X and ¥ are Jacobi fields along a geodesic +, then (X,Y!) — g(X", ¥) = constant, Moreover, if X and ¥ vanish at some point, say xz, of 7, then (X,Y) — (4%, ¥) = 0. VII, VARIATIONS OF THE LENGTH INTEGRAL a Proof. d ! gS ¥) = 8X P) +a(KP) (X', ¥') — a(%, RY, 9) Similarly, 4 ye. ” yn : Belt rY) = a F) ~ (R(X 4, ¥)- a(R(X, #)#, ¥) (cf. Proposition 2.1 of Since g(X, R(Y, #)#) Chapter V), we have Si&Y) - aX.) QED. Example 2.1. Let + = x; be a geodesic in a Riemannian mani- fold M with positive constant curvature k. Let % Yu «+05 Yuan be an orthonormal basis for T,,(.M). By the parallel displacement along 1, we extend ¥;,..., ¥,. to parallel vector fields along + so that +, Y;,..-, Y,-, form an orthonormal basis for the tangent space at every point of x. Set (Uda, = Sin (VEN ay — (Vida, = 008 (VE) (Y, By Corollary 2.3 of Chapter V we can verify that Uy, ..-, Up-a Vy ++) Vara are Jacobi fields along +. It follows (cf. Proposition 111) that these 2n'— 2 Jacobi fields, + and # form a basis for the space J, of Jacobi fields along +. It follows also that the conjugate points of xy along + are given by the points with parameter values t = mniVk, m= +1, +2,.... 3. Conjugate points Asin §2, let M be a Riemannian manifold and + = x, geodesic. in M parametrized by its arc length. For each piecewise differenti- able vector field X along r, we set (ay =f Caco, 2) — e184 42 FOUNDATIONS OF DIFFERENTIAL GEOMETRY where X’ denotes the covariant derivative of X in the direction of +. In this section, we shall investigate properties of [%(X) and shall derive some consequences on the distance between two consecutive conjugate points. The integral /%(X) is also closely related with the index form of Morse, which will be defined and studied later in §6. . Proposimion 3.1. Let r=, a 2125, be a geodesic in M such that x, has no conjugate point along + = x, for aS tsb. Lat Y bea Tacobi field along + which vanishes at x, and is perpendicular to 1. Let X be a piecewise differentiable vector field along + which vanishes at x, and is perpenticular to x. If Xz, = Yay then woe have TX) = RY), and the equality holds only when X = Y. Proof. Let J,, be the space of all Jacobi fields along + which vanish at x, and is perpendicular to +. By Proposition 1.1, the space of all Jacobi fields along 7 which vanish at x, is of dimension 1. It follows (cf. Proposition 2.1) that dimJ,, =n —1. Let Yy, sey Fqoa be a basis for J,,. Since ¥ € J, we may write YsaYte where ay ..., @-1 are constants, Since there is no conjugate point of x, on r for aSts b, Yy,..+, Yq are linearly in- dependent at every point x, for a <1 5 6, There exist therefore functions f,(t), .. «+ f.-a(¢) stich that XaAYy tot Suae We have (2 standing for 275!) (XX) = (LAT LAY) + (ELA TAY + (2 Ate TAM» where each f; denotes df, dt. We also have —8(R(X, #)#, X) = —¥ fag(R(Yo #4, X) = Y falVi, X) = ALAM TI), VII, VARIATIONS OF THE LENGTH INTEGRAL B where the second equality is a consequence of the assumption that cach Y, is a Jacobi field. On the other hand, we have FaB Xo ELM) = AE LTo BIN) + (B f8 BALD HelE flo BLM) + Elo BL Combining these three equalities we obtain B(X', X’) — g(R(X, #)#, X) = AE AT BAN) + GelB flo Bf) + AE Ate Sf) eB 06 3.47) Since, by Proposition 2.5, we have ELT EIN) ~ aE Lo ELM) = Effie Yo Y) — e(¥ YD) = we obtain BX) = [EL EHV a + (Eo ZIM ne Similarly, we obtain RY) = [Sak VaH) dt + EAN BaP nw Since a; = dadt = 0, we have (P) = LAX LAP enw By our assumption that X,, = Y,y we have a, = f,(b) for i = 1, syn — 1. Hence, nay — Re) = fe fv, BAY) 20. Obviously, the equality holds if and only iff? = Ofori = 1,..., n—1.Sincea, = f(b), implies a, =f (() for all t and hence x=. QED.4 FOUNDATIONS OF DIFFERENTIAL GEOMETRY. Corottary 3.2. Let r= x, a StS b, be a geodesic in M such that x has no conjugate point along + = x, for a1. b. If Xisa piecewise differentiable vector field along + vaniskiig at x, and x, and ‘perpendicular to 7, then IX) 20, and the equality holds only when X = 0. . Proof. Set ¥ = 0 in Proposition 3.1. QED. As an application of this corollary we prove ‘Turoren 3.3. Let M be a Riemannian manifold with sectional curvature % ky > 0. Then, for every geodesic + of M, the distance between two consecutive conjugate points of + is at most m/V kp. Proof. Let + = xa S 1S ¢, be a geodesic such that x, is the first conjugate point of x, on 7. Let } be an arbitrary number such that a 0. Then, for every geodesic + of M, the distance between any two consecutive con jugate points of 7 is at most 7{/Vhq. VIL, VARIATIONS OF THE LENGTH INTEGRAL 75 Proof. Let + = % a, ¢ and b be exactly as in the proof of ‘Theprem 3.3. Let ¥,,..., Y,- be parallel unit vector fields along + sugh that 7, Y;,..-, ¥,-1 are orthonormal at each point of r. Let f(t) be a non-zero function with f(a) = f(b) = 0. By Corol- lary 3.2, we have J*(Y,) = Ofori = 1,...,n — 1. On the other hand, we have Sawn) = [arr 5%) —PEARM > ¥)] at s fe = Df? = haf) dt. ‘The rest of the proof is similar to that of Theorem 3.3. QED. Propostrion 3.5, Let r= xy a StS b, be a geodesic in a Riemannion manifold M. Then there is a conjugate point x=, @- 0 be such that x,y and x,,, are in U, Then since x44 is not a conjugate point of x,_, along the segment +’ of r from %,_5 to x,» the linear mapping of the set of Jacobi fields on +’ into T,,(M) + Ty,,,(M) given by Z—+ (Zz, Z,,,) 8 one-to-one and therefore onto, because both vector spaces have dimension 2n, Hence there is a Jacobi field on +’ with prescribed values at both ends. We now choose a Jacobi field Z on +’ such that Z,, , = ¥,,,and Z,,,, = 0. ‘We now define a vector field X along + as follows: X=Y from x, t0 x» Z from x4 10 tay =0 from x5 to ty76 FOUNDATIONS OF DIFFERENTIAL GEOMETRY 4(Y) + I,(Y) by Proposition 2.3, we F(X) — BY) TY) + F(Z) — IY) — TY) Tei(Z) — 1(). Let Y be a vector field along + from x,_, to x,,,"efined as follows: Y=¥ from x4 to xy from x, 0 Xp Applying Proposition 3.1 to vector fields P and Z, we have Te4(Z) < TP) = Tel), Hence, we have J9(X) <0. QED. 4. Comparison theorem ‘The main purpose of this section is to prove the following comi- parison theorem of Rauch [1]: Tueorem 4.1. Let M and N be Riemannian manifolds of dimension n with metric tensor g and h respectively. Let o = x, aS 42 b, bea geodesic in M and X a non-zero Sacobi field along o perpendicular to 0. Let r= yy aS tS b, be a geodesic in N and ¥ a non-zero Jacobi field along x perpendicular to +. Assume: (1) Both X and Y vanish at t = a; (2) X' and ¥" have the same length at t = a; (3) x has no conjugate point on ¢ = x, a St 0 (or more generally, with Ricci tensor S whose eigenvalues are all % (n + 1)ky > 0). Then (1) The diameter of M is at most [VF yi (2) M is compact ; (3) The fundamental group =,(M) is finite. Proof. Let x and y be arbitrary points of M and let r be a minimizing geodesic joining x toy. By Theorems 3.3, 3.4, and 5.7, the length of r is at most /V/,, which proves (1). Since M_ is bounded and complete, it is compact (cf Theorem 4.1 of Chapter IV). The universal covering space Mf of M with the naturally induced Riemannian metric satisfies the assumption of Theorem 5.8. Hence, M1 is also compact. This means that 7,(M) is finite. QED. Theorem 5.8 is due to Myers [1]. 6. Index theorem of Morse Let r =x, 4515 5, bea geodesic fromy = #, to = 4 in a Riemannian manifold M. Given a conjugate point x,,@ (X;, Xia) defines a linear isomorphism of the space of Jacobi fields along + | {2 4:41] into the direct sum of the tangent spaces at x,, and %,,,* Since it isa linear isomorphism of a vector space of dimension VIL, VARIATIONS OF THE LENGTH INTEGRAL 91 2n into a vector space of the same dimension (cf. Proposition 1.1), it mbst be surjective. This completes the proof of (1). (2) With the notations in §3, we have 12,8) = "Zrg(X), ok, oX) ='S teenie). By Proposition 3.1, we have Tein(X) & [ee(pX) and the equality holds if and only if X is a Jacobi field along 7 | au ausale (3) If U is a subspace of T on which I is negative semi- definite, then J is negative semi-definite on pU by (2). Morcover, p: U- pUisa linear isomorphism. In fact, if X ¢ U and p(X) = 0, then (2) implies 0 2 UX, X) = I(pX, pX) =0, and hence 1(X, X) = I(pX, pX). Again by (2), we have X = pX = 0, Thus we have a(I|T2) s a(| J). ‘The reverse inequality is obvious. The proof for the index i(/ | 7+) is similar. Finally, to prove n(I|T}) =n(I|J), let X be an element of J such that J(X, Y) = 0 for all YeJ. Since X is a Jacobi field along + | [4,, a,,,] for all i, the formula in Theorem 5.5 reduces to the following 4X, ¥) =2['S (x - 2, 1).,]- In the same, way as we proved Theorem 5.6, we conclude that X’- = X’+ at s,, for all iso that X is a Jacobi field along +. This means that n(I[7) = n(I | J). The reverse inequality is obvious, Lemma | and the next lemma imply (2) of Theorem 6.1. Leuata 2. If B is a symmetric bilinear form on a finite-dimensional vector space V, then a(B) =i(B) +0(B).92 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Proof. Let »,,..+, 0, be a basis for V with respect to which B is a diagonal matrix with diagonal entries dy, ..., dye Set V,, = the space spanned by {045 4, > 0}; V_ = the space spanned by {045 dy <0) Vo = the space spanned by {0,5 dew= 0}, + V+ Vo We shall show that n(B) = dim V_ and a(B) = dim (Vo + V-). Clearly, {XeV;B(X,Y)=0 forall Ye V} so that V dim V, iB % and hence dim Vy = n(B). Let U be any subspace of V on which B is negative semi-definite. We claim that the projection p: U—> Vy + V_ along V, is injective, In fact, if Xe U and p(X) thea X Y,, Since B is negative semi-definite on U and positive definite on V,, X must be zero. Thus, dim U = dim (Vy + V-) and hence a(B) = dim (V5 + V-). Similarly, if U’is any subspace of Von which B is negative-definite, then the projection p’: U’ > V_ is injective and hence i(B) = dim V_. This completes the proof of Lemma 2. Since dimJ = ( —1)(n—1) < oo, Lemma 1 implies that both a(|T+) and i(I |T}) are finite, The finiteness of conjugate points follows from the next lemma, Lemna 3. For any finite number of conjugate points uy +++ 5%, (a N, be the linear isomorphism induced by parallel displacement of N(a,) to N(ra,) along +. Let J, be the index form for the geodesic + = x, 051s rb, We then set BAX, ¥) = Irae * pe 2 a(X), aF** pe @(¥)) for X,Y © Say «+ 0)sof FOUNDATIONS OF DIFFERENTIAL GEOMETRY Lewma 4, (1) By(X, ¥) = 1(X, ¥) for X, Ye Say. + a5) (2) B, is positive definite for small values of r5 (3) The nullity n(B,) is equal tothe order ofthe point xq asa conjugate point of x along +. In particular, n(B,) = 0 if xy is not conjugate to xy along 7; (3) The family B, is cominuous inv Proof. (I) Since « =, 1, =, and py: > N, = Nis the identity mapping, we have B, = I. (2) Since a7* « p, « a is an isomorphism of J(ag, . . ., a) ontoJ,, it suffices to show that I, is positive-definite on J, for small values of r. But Corollary 3.2 implies that if r is small enough so that + | (0, 78] has no conjugate point of x, then J, is positive-definite on J, (3) Since a:*» p,° a: Say, -.., 4) J, is an isomorphism, we have n(B,) = n(I, | J,). Applying (3) of Lemma | and (3) and (4) of Theorem 6.1 to 1,, we see that (J, | J,) is equal to the order of x3, a8 a conjugate point of xy along 7, (8) Forfixed X, ¥ © J(ap, ..-, 4), we shall show that B,(X, ¥) is continuous in r for 0 (X(7), X(r)'~, X(r)"*) om, i8 continuous. Define vectors Xy,5 ++ +5 Xy BY (Bap eres Seay) =p 20(X), Xe Ang We fix an integer i, 0 < iS h — 1, and to simplify notations, we set 0 may dm tun Ve Kin We Kane ‘Then for each r, V, is a vector at the point x,., and W, is a vector at the point x,.. Evidently, both V, and W, depend differentiably on r, We recall that 7|[¢, d] is contained in a convex normal VII. VARIATIONS OF THE LENGTH INTEGRAL 95 coordinate neighborhood. Both exp uV, and exp uW, are also in theyame neighborhood for |u| (1) is trivial. (2) may be proved as follows. lims =r. dry) = lim dl, x) ‘These two properties of A imply that either A = (0, 20) or A = (0, 7] where r is some positive number. If A = (0, r], then the point 2, is called the cut point (or the minimum point) of x along r. If A = (0, «), we say that there is no cut point of x along 7. VIL, VARIATIONS OF THE LENGTH INTEORAL 97 Tweorem'7.1. Let x, be the cut point of x» along a geodesic 1% 0S 1 < 00, Then, at least one (possibly both) of the following stalfments holds: (1) 4, isthe first conjugate point of xy along 7; (2) There exist, at least, two minimizing geodesics from xp to x, Proof. Let a), a, ... be a monotone decreasing sequence of real numbers converging to r. For each natural number &, let exp (X,, 0S 1S dy be a minimizing geodesic from % tO yy where X, is a unit tangent vector at x, and 5, is the Qistance from. 4% to x,,. Let X be the unit tangent vector at x determined by 4% = exp £X for all ¢, Since x, is the cut point of x along + and a > 1, we have X#X > be Since 4, is the distance between x, and x,,, we have r= limb. Hence, the set of vectors 4%; is contained in some compact subset of T,,(M). By taking a subsequence if necessary, we may assume that the sequence b,X,, bsXp,.. . converges to some vector of length 1, say r¥, where ¥ is a unit tangent vector. Then, exp tY, 0-5 = 1, isa geodesic from x, to x, since exp r¥ = lim... exp bX, limy..s X= % It is of length r and hence minimizing. If X # ¥, then exp /X and exp (¥, 0 ¢ = r, are two distinct minimizing ‘geodesics joining x, and x, and (2) holds. Assume X = ¥. Assuming also that x, is not conjugate to x, along 7, we shall obtain a contradiction. Since the differential of exp: T,(M) —> M is non- singular at rX (cf. §1), exp is a diffeomorphism of a neighborhood U of rX in T,,(M) onto an open neighborhood of x, in M. Let be a large integer such that both aX and bX; are in U. Since exp a,X = x, = exp 5,X,, we can conclude that a4 = bXy thus contradicting the fact that X # X,. We have shown that if X =, then g, is conjugate to *y along +. On the other hand, there is no conjugate point of x, before x, along 7. Indeed, if x,, where 0 = 5 5 1, were conjugate to x, along 7, then 7 would not be minimizing beyond x, by Theorem 5.7, which contradicts the, definition of r. Hence, x, is the first conjugate point of xy along 7. QED.98 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Corortary 7.2. If x, is the cut point of xp along 7 = xy 0 St < 0, then xyisthe cut point of , along + (inthe reversed direction). Proof, Extending the geodesic + in the negative direction, we may assume that x, is defined for —co << 00. Let —a be any negative number. We claim that | [—a,] is not minimizing. Assume (1) of Theorem 7.1. Then x is a conjugate point of x, along + in the reversed direction and, by Theore™ 5.7, + | [—a, r] cannot be minimizing. Assume (2) of Theorem 7.1. Then the join of + | [—a, 0] and a minimizing geodesic from x, to x, other than + | (0,1) gives us a non-geodesic curve of length @ + r joining x_ and x,. Hence, + | [—a, F] whose length is also a +r, cannot be minimizing. This proves our claim, There exists therefore a non- negative number b lim... u(%)). Set % = H(%), a= Him (sy). s ce p(X) >a, exp aX cannot be conjugate to x along the geodesic exp ¢X. By Theorem 14, exp: T,,(M) + M maps a neighborhood, say U, of aX diffeomorphically onto a neighborhood of exp aX. We may assume, by omitting a finite number of a,X, ifnecessary, that all of a,X, are in U. Since exp maps U diffeomor- phically onto exp (U), exp a, cannot be conjugate to x5 along the geodesic exp 1X,. By Theorem 7.1, there is another minimizing geodesic from x, to exp 4,X,. In other words, there exists, for each , a unit vector Y, # X, at 4p such that exp a exp a,F. Since exp maps U one-to-one into M, each a,¥, does not lie in U. By taking a subsequence if necessary, we may assume that Yy, Yay . . converges to some unit vector, say Y. Then aY, which is the limit vector of a,¥,, does not lie in U. We have exp a¥ = exp (lim a,¥,) = lim (exp ae¥,) = lim (exp a-%,) = exp (lim a.Xi) = exp aX. Hence, both exp 1X and exp #¥, 0.5 t/a, are minimizing geodesics from x, to exp aX = exp a¥. This implies that if b is any number greater than a, then the geodesic exp tX,0 <1 SB, is no longer minimizing, contradicting the assumption 4(X) > a. We consider next the case w(X) < lim. u(%;). Let be a positive number such that lim... a(X,) > #(X) + 5. By omitting a finite number of X; if necessary, we may assume that, w(%) > W(X) +5 forall k. By the very definition of p(X), there exists a unit vector X” 4X at x» such that exp ¢X’, 0 <1 = u(X) +8, is a minimizing geodesic from xy to exp (u(X) + 5)X, where 6” < 5. (Note that 5° may be negative.) In particular, exp (u(X) + )X = exp (u(X) + 6)". We set % —.100 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Since the sequence of points exp (u(X) + 6)X, converges to exp (u(X) + 0)X, we may assume, by omitting a finite number of ¥, if necessary, that the distances between exp (u(X) + 6)X and exp (u(X) + 6)X, are all less than c, For each fixed , consider the curve from 4y to exp (u(X) + 5)X, defined as follows. It consists of the geodesic expiX’, 05 15 u(X) +6', from x to exp (u(X) + 6')X’ = exp (u(X) + 5)X and a miniMizing geodesic from exp (u(X) + 5)X to exp (u(X) + 6)X,. Then the length of this curve is less’ than (X) + b° +¢ =y(X) +5 —c. This means that the geodesic expt%,, 051 n(X) +5, is not minimizing, which contradicts the inequality u(,) > u(X) +B. QED. Let G(x) denote the set of all u(X)X, where X are unit vectors at xp such that y(X) are finite. Let C(xq) be the image of E(x.) under exp, Obviously, C(x) consists of all cut points of x, along all geodesics starting from %. We shall call C(x,) the cut locus of x, and C(x) the cut locus of x» in T.,(M). Turonem 74, Let E = {0X30 5 ¢ < p(X) and X unit vectors at xq}. Then (1) Bis an open call in T,,(M) ; (2) exp maps E difeomorphically onto an open subset of Mi (3) Mis a disjoint union of exp (E) and the cut lacus C(t) of %. Proof. (1) follows from Theorem 7.3. Obviously, exp maps E One-to-one into M. By Theorem 5.7, there cannot be any con- jugate point of xq in Z, Hence, the differential of exp: E> M is non-singular at every point of E, This implies (2). Since E and G(x) are disjoint, exp (E) and C(x) are also disjoint, To show that M is a union of exp (£) and C(%), lety be an arbitrary point of M, Let exp tX, 0 < f = a, be a minimizing geodesic from % to Jy where X is a unit tangent vector at xy and a is the distance from % toy, From the very definition of u(X), it follows that a < m(X): Hence, aX is either in E or in C(x»). The point y = exp aX is therefore either in exp (E) or in C(x). QED. Remark, The open subset exp (E) of M is the largest open subset of M in which a normal coordinate system around *) can be defined, scorer enamenenenanenencctnatte’ Seman pman ses mem: tstayneeeremmmtnt sat Teen nee VII, VARIATIONS OF THE LENGTH INTEGRAL 101 ‘Tueoren 7.5. Let M be a complete Riemannian manifold and xy a it of M. Then M is compact if and only if, for every unit tangent ed X at oy (X) is finite. ‘Proof, Suppose M is compact and let d be the diameter of M. If @>d, then exptX, 05134, cannot be a minimizing geodesic from x, to exp aX, Hence, u(X) = d. Conversely, assume that, for every unit vector X at x, u(X) is finite. Since wis a continuous function defined on the unit sphere in 7,,(M) (cf. Theorem 7.3), u is bounded by a positive number, say b. Let B be the set of tangent vectors at x» whose length are less than or equal to b. Then B is a compact set containing E and C(x.) of ‘Theorem 7.4. By Theorem 7.4, exp maps B onto M. Hence, M is compact. QED. Remark, Theorems 7.3 and 7.5 imply that M is compact if and only if the cut locus G(x) of x» in T,,(M) is homeomorphic with a sphere of dimension n — 1, where n = dim M. Example 7.1. Let M be an n-dimensional unit sphere and x its north pole. The cut locus of x in T,(M) is the sphere of radius 7 with center at the origin of T,(M). The cut locus of x in M reduces to the north pole. ‘Example 7... Let S* be the unit sphere in R™, Identifying each point of S* with its antipodal point, we obtain the n-dimen- sional real projective space, which will be denoted by M. The Riemannian metric of S* induces a Riemannian metric on M in a natural manner so that the projection of S* onto M is a local isometry, The cut locus of a point x € M in T,(M) is the sphere of radius 7/2 in T,(M) with center at the origin. If x corresponds to the north and south poles of S*, then the cut locus of x in M is the image of the equator of S* under the projection $* -» M. The cut locus C(x) is therefore a naturally imbedded (n — 1)-dimensional projective space. Example 7.3. In the Euclidean plane R* with coordinate sys- tem (x,y), consider the closed square given by 0 = x,y = 1. By identifying (x, 0) with (x, 1) for all 0 $x 1 and (0,») with (1,9) for all 0's y = 1, we obtain a torus, which we denote by M. ‘The natural Riemannian metric on R? induces a Riemannian metric on M. Let p be the point with coordinate (4, #). Then the cut locus ofp in T,(M) can be identified with the boundary of the102 FOUNDATIONS OF DIFFERENTIAL GEOMETRY square in R* under the natural identification of T,(M) with R¥. The cut locus ofp in M consists of two closed curves which form a basis for H4(M,Z) ~Z + Z. ‘A list of papers on conjugate and cut loci can be found in an introductory article by Kobayashi [27]. For more recent results, see Warner [3], Weinstein [1], and Wong [4]. » 8. Spaces of non-positive curvature We have already seen (cf. Corollary 2.4) that a Riemannian manifold with non-positive curvature has no conjugate point, In this section, we shall prove more. ‘Tueonem 8.1. Let M be a complete Riemannian manifold with non-positive curvature. Let x be an arbitrary point of M. Then the ex- ponential map exp: T,(M) —+ M at x is a covering map. In particular, if M is simply connected, then exp, isa difeomorphism of T.(M) onto M. Proof. In general, let f be a map froma Riemannian manifold Nwith metric gy into a Riemannian manifold M with metric 2. We say that f increases (resp. decreases) the distance if Stan = ay — (resp. fea = gy), IX = 1feXI (resp. 1X2 WfeX) for every tangent vector X of N. We first prove the following general lemma. that is, if Lemma 1. Let f be a map from a connected complete Riemannian manifold N onto another connected Riemannian manifold M of the same dimension. If f increases the distance, then f is a covering map and M is also complete. Proof. Let gsr and gy be the metrics for M and N respectively. Since f increases the distance, /*2, is also a Riemannian metric on N, For every curve r in N, its arc length with respect to gy is less than or equal to the one with respect to,f*g.y. It follows that if d and d’ denote the distance functions for gy and f*g y respec tively, then a'(9) 2 d(x9) for xy eN. It is then clear that every Cauchy sequence for d’ is a Cauchy ee eenaasihemmmemeemestomenl Vi, VARIATIONS OF THE LENGTH INTEGRAL 103 sequence for d and hence is convergent as NV is complete with resffect to gy. Since the topology of N is compatible with any iefiannian metric on N, N is complete with respect to f*ga- Replacing gy by fg we may assume that f: NM is an isometric immersion, ie., that gy =/*gy. Lemma 1 now follows from Theorem 4.6 of Chapter IV. Lena 2. Let M be an arbitrary Riemannian manifold and let x € M, Let p be aray in T,(M) emanating from the origin and + = exp, (p) the corresponding geodesic in M starting from x. Then every vector tangent to T,(M) and perpendicular to p is mapped by exp. into a vector tangent to M and perpendicular to 7. Proof. Let p = tX,0 == 1, where Xe T,(M). Let W* be a vector tangent to T,(M) at X and perpendicular to p. Let Xe T,(M), —€ 5 < ¢, bea l-parameter family of vectors such that (a) X° =X; (b) each X* has the same length as X; (c) W* is the tangent vector to the curve Xt, —e S 5 Se, at x. For each 5, let p! be the ray given by #X, 0. sts 1. The 1- parameter family of rays p', —e < s 5 ¢,inducesin a natural way a vector field Y* along p = p° which is perpendicular to p; for , each fixed ¢, ¥* at (X is the tangent vector to the curve described by ¢Xt, —c = 5 = ¢, at (X®. The vector field Y* vanishes at the origin of T,(M) and coincides with the vector W* at X, If we set = exp, (pt) and ¥ = (exp,)(¥*), then Y is the Jacobi field along * induced by the variation + of the geodesic +". We apply Theorem 5.1 to the variation 7*, Since the length L(r!) is independent of s, the left hand side of the formula in Theorem 5.1 vanishes. Since Y¥ vanishes at x and + is a geodesic, it follows that ¥ is perpendicular to 7° at the point exp, (X). This completes the proof of Lemma 2. Lewaca 3, Let M be a complete Riemannian manifold with non~ ‘positive curvature and x a point of M. Then exp,: T,(M) — M increases the distance. Proof. Let W* be a vector tangent to T,(M) at a point X of T,(M). We set W = (exp,)a(W*),los FOUNDATIONS OF DIFFERENTIAL GEOMETRY Then W is a tangent vector of M at exp, X. To show that the length [|W] of W’ is not less than the length |W*) of W*, we decompose W/* into a sum of two mutually perpendicular vectors at X: Wea we + Ws, where W; is in the direction of the ray p joining the origin of T,{M) to X and Wy is perpendicular to the ray p, We set W, = (exps)a( Wi") and We = (exp,)4(W2)- Since W+ is in the direction of p, it follows from the definition of exp, that Is = 1M . ‘As W, is perpendicular to W, by Lemma 2, we have iwi Wy + Wall = WI + | Wall Hence, (Wit = |e = yt + Wal? — We IW — WW ewer The problem is thus reduced to proving the inequality | Wz] = [W3I. In other words, it suffices to prove the inequality |W] = IW*| under the assumption that W’* is perpendicular to p. We define vector fields ¥* and Y along p and r respectively asin the proof of Lemma 2. Since Y* and ¥ are induced by l-parameter families of geodesics p* and +* ~ exp, (p%) respectively, they are ‘Jacobi fields along p and + respectively vanishing at ¢ = 0. We ‘hall now apply the comparison theorem of Rauch (Theorem 4.1) to Y* and Y, Assumptions (1) and (4) in Theorem 4.1 are obviously satisfied. By Corollary 2.4, assumption (3) is also satisfied. To complete the proof of Lemma 3, itis therefore sufficient to verify assumption (2) for the vector fields ¥* and Y. Take a Euclidean coordinate system in 7,(M) with origin at the origin of T,(M) and the corresponding normal coordinate system in a neighbor- hood of x. From the construction of Y, it follows that Y* and ¥ have the same components with respect to the coordinate systems chosen above. On the other hand, the Christoffel symbols vanish £229 VILL, VARIATIONS OF THE LENGTH INTEGRAL 105 at x (cf. Proposition 8.4 of Chapter IIT) and hence the covariant diffgrentiation and the ordinary differentiation coincide at x (cf. Capa 8.5 of Chapter IIT). These two facts imply assumption (2) br Theorem 4.1. ‘Combining Lemma 1 and Lemma 3, we obtain Theorem 8.1. QED. Remark 1. If M has vanishing curvature in Lemma 3, then ‘we may apply the comparison theorem of Rauch in both directions to obtain |W*l| = |W) and | 1W*) > |W. Hence we have: If M has vanishing curcature, then exp,: T,(M) —> M is an isometric immersion Combining this with Theorem 8.1 we obtain Conorzary 82. If M is a complete, simply connected Riemannian manifold with vanishing curvature and x € M, then exp,: T,(M) > M is an isometry. Remark 2. Kobayashi [16] strengthened Theorem 8.1 asfollows: If Mis a connected complete Riemannian manifold and a point x of M has no conjugate point, then exp,: T,(M) —+ M is a covering map. ‘The proof may be achieved by replacing Lemma 3 by the following: Ifa point x € M has no conjugate point, there is a complete Riemannian metric on T,{M) which makes exp, distance-increasing. For the proof we refer the reader to Kobayashi [16]; it is more elementary than that of Lemma 3. ‘Theorem 8.1 is due to Hadamard [1] and E. Cartan [8]. The proof presented here is based on Ambrose’s lectures at MIT in 1957-58. Theorem 8.1 and its generalization as stated in Remark 2 may be proved by means of the theory of Morse (cf. Klingenberg [8] and Milnor [3)). Tueorem 8.3. Let M be a connected homogeneous Riemannian manifold with non-positive sectional curvature and negative-definite Ricci tensor. Then M is simply connected and, for every x © M, exp,: T,(M) M is a diffeomorphism. Proof. An isometry of a Riemannian manifold is called a Gliford translation if the distance between a point and its image under the isometry is the same for every point. The following Jemma is due to Wolf [1].106 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Lemma 1, Let M be a homogeneous Riemannian manifold and Ifa covering manifold with the induced metric so thatthe covering projection fp: M > M is an isometric immersion, Then a difeomorphism f of i onto itself satisfying p of = p is a Cliford translation of M1. Proof of Lemma 1. Let G be a connected Lie group of isom- tries acting transitively on M, and g the Lie algebra of G. Considering every element X 4 as an infinitesimal isometry of M, let X* be the lift of X to AZ. Then the set of these vector fields X* generates a transitive Lie group G* of isometries of M whose Lie algebra is isomorphic to g. Since f induces the identity trans- formation of M, it leaves every X* invariant. Hence f commutes with every element of G*. For any two points y, y' €7, let y be an clement of G* such that 7’ = p(y). Then - A LID = dea). Fo v0) = dv), ve £0) = 4,0), where d denotes the distance between two points. This completes the proof of Lemma 1. Lemma 2. Let M, M, and f be as in Lemma 1, Let x) € M and let 1 = y,0 StS a, bea minimizing geodesic from yo t0 f (3) s0 that In =F). Set x = ply) for 0 St Za. Then =x, 05154, 4s a smoothly closed geodesic, that is, the outgoing direction of + at xy coincides with the incoming direction of 7 at x. Proof of Lemma 2, Let r be a small positive number such’ that the r-neighborhoods V(j9; r) and V(y45 1) of yp and y, are homeo- morphic with the r-neighborhood U(x; r) of x) = %, under the projection p. Assume that + is not smoothly closed at x = x. Then there is a small positive number 8 such that the points x, 5 and x, can be joined by a curve a in U(x9; 1) whose length is less than 26 (where 26 is equal to the length of + from x,_, through 4%, = fp t0 x,). Let o* be the curve in V(y45 1) such that p(a*) = 0. Tet y* be the end point of o*. Then y* = f(y,). Then A SID) = I09*) $ In Ied) + II") S (a — 26) + (length of o*) = (a — 26) + (length of ) <(a~ 28) +28 = a= d(Jn,f(d))- This contradicts Lemma 1, thus completing the proof of Lemma 2. Vi, VARIATIONS OF THE LENGTH INTEGRAL 107 We shall now complete the proof of Theorem 8.3. Assuming that M is not simply connected, let AT be the universal covering marffold of M. Let f be a covering transformation of 7 different frond the identity transformation. Let + = xy 0 St a, be a smoothly closed geodesic of M given in Lemma 2. Let X be any infinitesimal isometry of M. We define a non-negative function A(), —20 << +00, as follows: AM) = 2%, X), for stsa, and then extend 4 to a periodic function of period a. By Lemma 2, A(t) is differentiable at every point ‘, —co M is a diffeo- morphism follows from Theorem 8.1. QED. ‘Theorem 8.3 is due to Kobayashi [19]. Tueore 8.4, Let M be a connected homogeneous Riemannian manifold with non-positive sectional curcature and negative-definite Ricti tensor. If a Lie subgroup G of the group I(M) of isometries of M is transitive on M, then G has trivial center. Proof, Let € be the center of G. Let G be the closure of G in 1(M), Then every element of C commutes with every element of G and hence lies in the center of G. We may therefore assume that Gis a closed subgroup of 1(M). We first prove that Cis discrete. Suppose that X is an infinites- imal isometry of M which generates a l-parameter group belong- ing to the center C. Since X is invariant under the action of G,108 FOUNDATIONS OF DIFFERENTIAL GEOMETRY the length g(X, X)' is constant on M. Since the Ricci tensor of ‘M is negative-definite, Theorem 5.3 of Chapter VI implies that X vanishes identically. Let g be any element of C. Then it commutes with each element of G and hence is a Clifford translation in the sense defined in the proof of Theorem 8.3. In fact, for x, «” €.M we choose an element y of G such that x’ = y(x). Then we have d(x’, o(x')) = dlyla), pe v(x)) = diplx), ve o(x)) = d(x l2)). It follows that the action of € on M is free. We shall show that C is properly discontinuous on M. Let H be the isotropy subgroup of G at a point of M so that M = G/H. Then H is compact (cf. Corollary 4.8 of Chapter I). By Proposition 4.5 of Chapter I, is discontinuous on M. By Proposition 4.4 of Chapter I, © is properly discontinuous on M. Then the quotient space MIC is a manifold (cf. Proposition 4.3 of Chapter I) and M is a covering manifold of M/C with the natural projection p: M— M[C as a covering projection (ef. pp. 61-62 of Chapter I), Since C is the center of G, the action of G on M induces an action of G on M/C. It follows that with respect to the induced Riemannian metric M{C is also a homogeneous Riemannian manifold with non- positive sectional curvature and negative-definite Ricci tensor. By Theorem 8.3, M/C is simply connected. Hence C reduces to the identity element. QED. 9. Center of gravity and fixed points of isometries Let A be a compact topological space and (4) the algebra of real-valued continuous functions fon A. With the norm |ifil defined by Wf = sup (1, (A) is a Banach algebra. A Radon measure (or simply, measure) on Ais a continuous linear mapping 4: C(A) > R. For each f €C(4), u(f) is called the integral of f with respect to the measure js and will be denoted by | F(a) du(a) or simply by i) f du. A measure Mis positive if u(f) ='0 for all non-negative f ¢ C(4) and if p #0, VII, VARIATIONS OF THE LENGTH INTEGRAL 109 We need the following theorem to prove the existence of a fixed point of a compact group of isometries of a complete, simply connected Riemannian manifold with non-positive curvature. Turonen 9.1. Let 4 be a positive measure on a compact topological space A. Let f be a continuous mapping from A into a complete, simply ‘connected Riemannian manifold M with non-positive curvature. We set J) -f dls, f(a))* dula)— forxe M, where d(x, f(a)) is the distance between x and f(a). Then J attains its sminimurn at precisely one point. ‘The point where J takes its minimum will be called the center of gravity of f(A) with respect toy. Proof. By normalizing the measure if necessary, we may assume that the total measure of 4, i.e., (1), is 1. (1). The case where M = R°, Let x!,..., 2" be a Euclidean coordinate system in R" and let f: A > R* be given by v=f(d, aedin Then Eto F(a)" dn ante) —2E fF) dole) + Ef ray ena E(x)? — 2H + 0, where =| 1) dua. ‘ iF fila) dula) and Hence, we have Sea) = (Gt He — HN, showing that J takes its minimum at b = (88, ..., 6") and only at b.110 FOUNDATIONS OF DIFFERENTIAL GEOMETRY (2) Existence of the center of gravity in the general case. Since J'is continuous, it suffices to show the existence ofa compact subset K of M and a positive number r such that Je) 7 for some y5 in K and Js) > 1 for all x not in K. Choose an arbitrary point 7 of M and take a positive number r such that d(yf(@)) Sr forallae A. K = (re M; d(x f(A) 3 Since f(A) is compact, K is bounded and closed. Being a bounded, closed subset of a complete Riemannian manifold M, K is compact by Theorem 4.1 of Chapter IV. Evidently, yp € X. We have We set Jn) = [rae =r, If x is not in K, then d(x, f(a)) > r for all a € A. Hence, Sts) > [rene =r peseK. (3) The uniqueness of the center of gravity in the general case. We shall reduce the problem to the Euclidean case. Let o be a joint of M where J takes its minimum. Since exp,: T4(M) > M is a diffeomorphism by Theorem 8.1, there is a unique mapping F: A+ T,(M) such that f= exp, We set (a) =[e%royane), Xe TM), A where d" denotes the Euclidean distance in T,(M). Let 0 be the origin of T,(M). We shall prove the following relations which imply a that ois the only point where J takes its minimum: Slo) =S'(0) Mas follows: f(a) =a(x,) for ae. Let y be a left invariant measure on G, (It is known that every compact group admits a bi-invariant measure; see for example, Nachbin [1; p. 81].) By setting A = G, we apply Theorem 9.1 and claim that the center of gravity of f(G) with respect x is a fixed point of G. Evidently, it suffices to show that J(b(x)) = J(x) for x © M and for b €G. Since the distance function d is invariant by G and jis left invariant, we have Kota) = fete) a(59* daa) = fas Ota) * daa) = fo bax.) * du(o-ta) = Jia). QED.2 FOUNDATIONS OF DIFFERENTIAL GEOMETRY We owe the formulation of Theorem 9.1 to Iwahori [1]. ‘Theorem 9.2 was originally proved by E. Cartan [16]. (See also Borel [4]. ‘As an immediate consequence of Theorem 9.2 we have Conottary 9.3, ‘Let M be a connected, simply connected homo- ‘gentous Riemannian manifold of non-positive sectional curvature and let G be a closed subgroup of the group I(M) of isometries of M. Assume that G is transitive so that M = G/H, where His the isotropy subgroup of G at 4 point of M. Then (1) isa maximal compact subgroup of G and every maximal compact subgroup of G is conjugate to H; (2) If G is connected, so is H. Proof. (1) Let K be any maximal compact subgroup of G. By Theorem 9.2, X is contained in the isotropy subgroup of G at a point say x, of M. Since the isotropy subgroup at x is compact (cf. Corollary 4.8 of Chapter 1), it coincides with K. Since G is transitive, the isotropy subgroups are all conjugate to each other. Hence H is conjugate to. X and is a maximal compact subgroup of 6. (2) Since M is simply connected, a simple homotopy argument shows that if G is connected, so is Hf. QED. The following theorem will not be used except in §11 of Chapter XL Tuonen 9.4. Let M be a connected, simply connected homogencous Riemannian manifold with non-positive sectional curvature, Let G be a closed subgroup of the group of isometries of M which is transitive on M 0 that M = G[H, where H is the isotropy subgroup of G at a point, say 0, of M. Assure thatthe linear isotropy representation of Hf leaves no non-zero vector of T(M) invariant. Then we have (1) M is in one-to-one correspondence with the set of all maximal wompact subgroups of G under the correspondence which assigns to each point x6 M the isotropy subgroup Gy of G at x; (2) If «is an automorphism of G of prime period, then there is a maximal compact subgroup of G which is invariant by a. Proof. (1) By Corollary 9.3, G, is a maximal compact sub- group of G and, conversely, every maximal compact subgroup of G coincides with G, for some x € M. It remains to prove that, for VII. VARIATIONS OF THE LENGTH INTEGRAL 113, every x € M, xis the only fixed point of G.. Because of homogeneity, it suffices to show that o is the only fixed point of H = G,. Assume that there is another fixed point x of H. By Theorem 8.1 there is a ‘unique geodesic + from o to x. By uniqueness of +, H leaves + pointwise invariant. Hence the tangent vector to 7 at ois invariant by H, in contradiction to our assumption. (2) Let 7 be any automorphism of G. Then y permutes the maximal compact subgroups of G. Let j denote the corresponding transformation of M; if 7(G,) = G,, then 7(x) = by definition, Let I’ be the cyclic group generated by a and assume that the order of Tis prime. Then every element y # 1 of Tis a generator of P, and a fixed point of 7 for any y ¥ 1 is a fixed point of a. If a has no fixed point on M, then I’ acts freely on M. But this is impossible since M is homeomorphic to a Euclidean space by ‘Theorem 8.1 of Chapter VIII. If Pacts freely on M, then EP(D; Z) = H*(MP; Z) by a theorem of Eckmann, Eilenberg-MacLane and Hopf (see Cartan-Eilenberg [1; p. 356]). On the other hand, H(MjT, Z) = O for k > dim M. On the other hand, H*(T'; Z) = Z, for all even k (cf, Cartan-Eilenberg [1; p. 251]). This shows that T° cannot act, freely on M. In other words, & has a fixed point, say x ¢ M. Then « leaves G, invariant. QED.CHAPTER IX Complex Manifolds 1, Algebraic preliminaries ‘The linear-algebraic results on real and complex vector spaces obtained in this section will be applied to tangent spaces of manifolds in subsequent sections. ‘A complex structure on a real vector space V is a linear endomor- phism J of V such that 7 = —1, where 1 stands for the identity transformation of V. A real vector space V with a coriplex struc- ture J can be turned into a complex vector space by defining scalar multiplication by complex numbers as follows: (a+ i)X =aX + 6X forXeV and a, beR. Clearly, the real dimension m of V must be even and 4m is the complex dimension of V. Conversely, given a complex vector space V of complex dimen- sion n, let J be the linear endomorphism of V defined by JX =iX forked. If we consider V as a real vector space of real dimension 2x, then J isa complex structure of V. Propostrion 1.l, Let J be a complex structure on a n-dimensional real vector space V. Then there exist elements Xyy ..., Xq of V such that Oe creer Anes xe) a a base for We Proof. We turn V into an n-dimensional complex vector space as above. Let Xj, ..., X, be a basis for V as a complex vector space. It is easy to see that (X,...,X,, JX... -, JX} is a basis for Vas a real vector space. QED. ut TX, COMPLEX MANIFOLDS 15 Let Gr be the complex vector space of mtuples of complex numbers Z = (z4,..., 2"). Ifwe set P+, ow gteR, k=lj.yn, then O* can be identified with the real vector space R® of 2n-tuples of real numbers (x!,..., 2°, y!,...,)"). In the following, unless otherwise stated, the identification of G" with R® will always be done by means of the correspondence (21, ..., 2") -» (x,...,.", Jh +++"). The complex structure of R™ induced from that of "maps (s,.. 2494, 9") into... 7% —al,.., ox) and is called the canonical complex structure of B®, In terms of the natural basis for R®, it is given by the matrix a4, where J, denotes the identity matrix of degree n. 2 Proposrrton 1.2. Let J and J’ be complex structures on real vector spaces V and V’, respectively. If we consider V and V’ as complee vector spaces in a natural manner, then a real linear mapping f of V into V" is complex linear when and only when J’ f = f oJ. Proof. . This follows from the fact that J or J" is the multiplica- tion by i when V or V’ is considered as a complex vector space. QED. In particular, the complex general linear group GL(n; C) of degree n can be identified with the subgroup of GL(2n; R) con- sisting of matrices which commute with the matrix a-(i a} A simple calculation shows that this representation of GL(n; C) into GL(2n; R), called the real representation of GL(n; ©), is given by ae A+iBs +i (64 where both 4 and B are real n x n matrices, ) for 4 +iBeGL(n;),116 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Proposirion 1.3, There is a natural one-to-one correspondence between the set of complex structures on RY and the homogencous space GL(2n; R)/GL(n; C) ; the coset represented by an element S © GL(2n;R) corresponds to the complex. structure SJgS-*, where Jy is (the matrix representing) the canonical complex structure of R™. Proof. Every element 5 of GL(2n;R) sends every complex structure J of R® into a complex structure SJS~1 of R™; we con- sider GL(2n; R) as a transformation group acting on the set of complex structures of R'. It suffices to prove that this action is transitive and that the isotropy subgroup of GL(2n;R) at Jy is GL{n; ©). To prove the transitivity of the action, let J and J" be two complex structures of R**. By Proposition 1.1, there are bases Keay c++ Can Sty «=p Seg) and {045 «0 thy S'Cy «og J'04) for REN, Tf we define an element S of GL(2n; R) by Seat Sey then J’ = SJS-, thus proving the transitivity. On the other hand, an element S of GL(2n; R) is in GL(n; C) ifand only if it commutes with Jp, that is, Jy = SJqS- (cf. the argument following Proposi- tion 1.2). QED. Proposrrion 1.4. Let J be a complex structure on a real vector space V. Then a real vector subspace V" of V is invariant by J if and only if V isa complex subspace of V when V is considered as a complex vector space. Proof. As in the case of Proposition 1.2, this also follows from the fact that J is the multiplication by i when V is ¢onsidered as a complex vector space. QED. fork = 1y...5m Let V be a real vector space and V* its dual space. A complex structure J on V induces a complex structure on V*, denoted also by J, as follows: (IX, X*) = (X,JX*) for Xe V and X+ eV. Let V be a real m-dimensional vector space and V* the com- plexification of V, ic., V! = V @g G. Then Vis areal subspace of V* in a natural manner. More generally, the tensor space T:(V) of type (r,s) over V can be considered as a real subspace of the tensor space T7(V*) in a natural manner. The complex conjugation in V* is the real linear endomorphism defined by ZaX+i¥+Z=X-i¥ fork, Yev. TX, COMPLEX MANIFOLDS 7 ‘The complex conjugation of V* extends in a natural manner to that of THF). Assume now that V is a real 2n-dimensional vector space with a complex structure J. Then J can be uniquely extended to a com- plex linear endomorphism of V*, and the extended endomorphism, denoted also by J, satisfies the equation J? = —1, The eigen values of J are therefore i and —i. We set V8 = (ZEW; JZ Ve = (ZeV; JZ = —iZ}. ‘The following proposition is evident, {KL iX; Xe V} and VX = Propostrion 1.5. (1) Vi {X4+UX; XeV}; (2) Ve = V*® + V°"(complex vector space direct sum) ; (3) ‘The complex conjugation in V* defines a real linear isomorphism between V" and V%, Let V* be the dual space of V. Its complexification V* is the dual space of V*, With respect to the eigenvalues ++i of the complex structure Jon V*, we haye a direct sum decomposition as above: Ve = Vat Vane ‘The proof of the following proposition is also trivial. Proposrrion 1.6. Vig = {X* EV; (X,X*} = 0 forall Xe V}, Vox = (Xe V*; (X,X*) = 0 forall Xe V9}, ‘The tensor space T;(V*) may be decomposed into a direct sum of tensor products of vector spaces each of which is identical with cone of the spaces V**, V4, V,o, and Vox. We shall study the decomposition of the exterior algebra A'V* more closely. The exterior algebras A Vy and A Vax can be considered as sub- algebras of A V** in a natural manner. We denote by A’* V# the subspace of A V*" spanned by A B, where « € A?V,. and B € A°V,,. The following proposition is evident. Propostrion 1.7. The exterior algebra A V** may be decomposed as follows: AV =S NV" with AV! = > WN “yes,18 FOUNDATIONS OF DIFFERENTIAL GEOMETRY and the complex conjugation in V*, extended to A V** in a natural manner, gives a real linear isomorphism between AMY* and At?V*, If {¢, ..., et) is a basis for the complex vector space Vj», then @,..., &), where # = @, is a basis for V,, (cf. Proposition 1.5) and the set of elements As Ae ABA AR 1S fst) 0” for alt non-zero Ze Ve; (3) (Z,W)=0 forZeV® and Wer, Conversely, every complex symmetric bilinear form h on V* satisfying (1), (2), (8) is the natural extension of a Hermitian inner product of V. To each Hermitian inner product # on V with respect to a complex structure J, we associate an element @ of A*V* as follows: (X,Y) = A(X, SY) for X, Ye. We have to verify that g is skew-symmetric: (Y, X) = AY, IX) = AIX, Y) = AUX, PY) = AX, -JY) = -9(X, ¥).120 FOUNDATIONS OF DIFFERENTIAL GEOMETRY It can be easily seen that g is-also invariant by J. Since At V* can be considered as a subspace of A* V*, ¢ may be considered as an element of A*V*. In other words, y may be uniquely extended to a skew-symmetric bilinear form on Vé, denoted also by g. By Propositions 1.5, 1.6, and 1.10, we have Proposrrion 1.11. Let @ be the skew-symmetric bilinear form on V° associated to a Hermitian inner product h of V. Then g & A¥ V*, We prove Proposrrtox 1.12. Let hand g be as in Proposition 111, (Zip «5 Z,} a basis for V* over G and {2, ..., 2} the dual basis for Vi. We set hg = WZ, 2) forjsk=Vyeeyn Then () hg hy forj k= dee ms (2) p= 2B Ey Agel at. Proof. (1) follows from (1) of Proposition 1.10. As for (2), given Z, We V*, we may write Z= F(Z), +H 2/2), W= SWZ, + HLW)Z)) ‘A simple calculation then shows 9(Z, W) = = E al (ZIE) — #(W)E(Z)) 2 QED. Example 1.1, Let g be a Lie algebra over C. Considering g as ‘a real vector space we have a complex structure J defined by JX = iX, The complex structure J satisfies (JX, Y] = JLX, ¥] = [X%, JY] for all X, ¥ eg, thatis, J» ad (X) = ad (X) ° J for every X Eg, Conversely, suppose that g is a Lie algebra over R with a complex structure J satisfying J* ad (X) = ad (X) e J for every X €g. Then, defining (a + ib)X = aX + 6JX, where a,b eR, we get a complex Lie algebra; we may verify complex bilinearity of the bracket operation as follows: [(a + i8)X, ¥] = (aX, ¥] + (6X, ¥] = af X, ¥] + 1%, ¥] = aX, ¥] + 6J[X, ¥] = (a + i0)[, ¥]- == IX, COMPLEX MANIFOLDS 121 2. Almost complex manifolds and complex manifolds A definition of complex manifold was given in Chapter I. For the better understanding of complex manifolds, we shall define the notion of almost complex manifolds and apply the results of §1 to tangent spaces of almost complex manifolds. ‘An almost complex structure on a real differentiable manifold M isa tensor field J which is, at every point x of M, an endomorphism of the tangent space T.(M) such that J* = —1, where 1 denotes the identity transformation of T,(M). A manifold with a fixed almost complex structure is called an almost complex manifold. Prorostrion 2.1. Every almost complex manifold is of even dimen- sions and is orientable. Proof. An almost compléx structure J on M defines a complex structure in each tangent space T,(M). As we have shown at the beginning of §1, dim T,(M) is even. Let 2n = dim M. In each tangent space T,(M) we fix a basis Xj, .-+5 Xp SXy «+5 SX qe ‘The existence of such a basis was proved in Proposition 1.1 and it is easy to see that any two such bases differ from each other by a linear transformation with positive determinant. To give an orieitation to M, we consider the family of all local coordinate systems st, ..., x of Af such that, at each point x where the coordinate system x!, ..., x! is valid, the basis (2/@x1),, . (aa), of T.(M) differs from the above chosen basis X, ..., “y JX, «JX, by a linear transformation with positive deter- minant. It is a simple matter to, verify that the family of local coordinate systems thus obtained gives a complete atlas compatible with the pseudogroup of orientation-preserving transformations of B®, QED. ‘The orientation of an almost complex manifold M given in the proof:above is called the natural orientati To show that every complex manifold carries a natural almost complex structure, we consider the space C* of n-tuples of com- plex numbers (24, ..., 2°) with Jaa +i, j=l... m With respect to the coordinate system (x4, ... **, sn) we define an almost complex structure Jon G* by Hafas!) = aay, Jay") = -(2]2#), 5122 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Prorostrion 2.2, A mapping f of an open subset of O® into Cm preserves the almost complex structures of ©* and ©", ie., fye J =S of, af and only if f is holomorphic. Proof, Let (w!, ..., w®) with wf = at + it, k= 1, be the natural coordinate system in ©, If we express of these coordinate systems in C* and G”: whats aI], mest es aa J), then fis holomorphic when and only when the following Cauchy- Riemann equations hold: ae'/ae ~ a4/2)/ = 0, aut/ay! + Oet]Ox! = 0. On the other hand, we have always (whether f is holomorphic or not) Fa(9/dx’) = 5 (au yaey(ajau) + 5 (ant/axry(a/ae'), 5 (8/) = S (aut/ay(aaus) + ¥ (aet] ay (a/00%), forj=ly....m From these formulas and the definition of J in @* and ©" given above, we see that fy*J =J-f, if and only if f satisfies the Cauchy-Riemann equations. QED. To define an almost complex structure on a complex manifold ‘M, we transler the almost complex structure of €* to M by means of charts. Proposition 2.2 implies that an almost complex structure can be thus defined on M independently of the choice of charts. An almost complex structure Jon a manifold M is called a complex structure if M is an underlying differentiable manifold of a complex manifold which induces J in the way just described. Let M and M’ be almost complex manifolds with almost com- plex structures J and J’, respectively. A mapping f: M—> M’ is said to be almast complex if J’*f, =f, ° J. From Proposition 2.2 we obtain 1X, COMPLEX MANIFOLDS 123 Proposrrion 2.3, Let M and M’ be complex manifolds. A mapping fi M+ M’ is holomorphic if and only if f is almost complex with respect to the complex structures of M and M’. In particular, two complex manifolds with the same underlying differentiable manifold are identical if the corresponding almost complex structures coincide, Given an almost complex structure J on a manifold M, the tensor field —J'is also an almost complex structure which is said to be conjugate to J. If M is a complex manifold with atlas {(U;, 9,)} then the family of charts (U, ¢,), where g, is the complex con- jugate of ¢,, is an atlas of the topological space underlying which is compatible with the pseudogroup of holomorphic trans- formations of Cr. The atlas {(U;, §,)} defines a complex manifold ‘whose underlying differentiable manifold is the same as that of M; this new complex manifold is said to be conjugate to M and will be denoted by M. It is easy to verify that if Jis the complex structure of a complex manifold M, then —J'is the complex structure of Af. Propostrion 2.4, Let M be a n-dimensional orientable manifold and L(M) the bundle of linear frames over M. Then the set of almost complex structures on M are in one-to-one correspondence with the set of crossesections of the associated bundle B = L(M){GL(n; C) with fibre GL(2n; R)/GL(n; ©), where GL(n; C) is considered as a subgroup of GL(Qn; R) by its real representation, Proof. This follows from Proposition 5.6 of Chapter I (sce also the remark following it) and Proposition 1.3. QED. In general, given two tensor fields A and B of type (1, 1) on a manifold M, we can construct the torsion of A and B, which is a tensor field of type (1,2) (cf. Proposition 3.12 of Chapter I). Specializing to the case where both A and B are an almost complex structure J, we define the torsion of J to be the tensor field N of type (1, 2) given by N(X, ¥) = 2(UX, JY] — (X, ¥] — JEN, JY] — JX, YD} for X, ¥eX(M). Let x4, ..., x8" be a local coordinate system in M. By setting X = fax! and ¥ = @/@s* in the equation defining N, we see that the components Nj, of N with respect to *', ..., x may be124 FOUNDATIONS OF DIFFERENTIAL GEOMETRY expresed in terms of the components Ji of J and its partial derivatives as follows: Ni = 25 (9 agNf — JET — Sea + HB where 8 denotes the partial differentiation 0/@x'. An almost complex structure is said to be integrable if it has no torsion. Tuworem 2.5. An almost complex structure is a complex structure if and only if it has no torsion. Proof. We shall only prove here that a complex structure has no torsion. The converse will be proved in Appendix 8 only in the case where the manifold and its almost complex structure are real analytic. Let 2',..., 2", 2! =x) + i, be a complex local co- ordinate system in a complex manifold M. From the construction of the complex structure J given before Proposition 2.2, itis clear that the components of J with respect to the local coordinate system 4, ..., x", 1, ..., 9 are constant functions in the co- ordinate neighborhood and hence have vanishing partial deriva- tives. By the expression above for Nj, it is clear that the torsion N is zero. QED. The complex tangent space T:(M) of a manifold M at x is the complexification of the tangent space T.(M) and its elements are called complex tangent vectors at x. If we denote by D*(M) the space of rforms on M, then an element of the complexification ©°(M) of D'(M) is called a complex r-form on M. Every complex r-form @ may be written uniquely aso’ + io”, where ” and o" are (real) r-forms. If we denote by T3¢ the complexification of the dual space of T,(M), then a complex r-form w on M gives an element of A’ T3* at each point x of M; in other words, a skew- symmetric r-linear mappings Ts(M) x «++ x.Ts(M) + © at each point x of M. More generally, we can define the space of complex tensor fields on M as the complexification of the space of (real) tensor fields. Such operations as contractions, brackets, exterior differentiation, Lie differentiations, interior products, etc. (cf. §3 of Chapter I) can be extended by linearity to complex tensor fields or complex differential forms. If M is an almost complex manifold with almost complex structure J, then by Proposition 1.5 Lease i 1 1X, COMPLEX “MANIFOLDS 125 where Ti and Te? are the eigenspaces of J corresponding to the eigenvalties i and —i respectively. A complex tangent vector (field) is of type (1,0) (resp. (0, 1)) if it belongs to 7}° (resp. T?#), By Proposition 1.5 we have Proposition 2.6. A complex tangent vector Z of an almost complex manifold M is of type (1, 0) (resp. (0, 1)) if and only if Z = X —iIX (resp, Z = X 4 iFX) for some real tangent vector X. By Proposition 1.7, the space © = €(M) of complex differential forms on an almost complex manifold M of dimension 2n may be Dbigraded as follows: caso. ‘An element of 6 is called a (complex) form of degree (p, 9). By Proposition 1.6, a complex I-form o is of degree (1, 0) (resp. (0, 1)) if'and only if @(Z) = 0 for all complex vector fields Z of type (0, 1) (resp. (1, 0)). Ifo, ..., w” is a local basis for ©, then its complex conjugate at, ..., a is a local basis for © (cf. Proposition 1.5). It follows that the set of forms a! A - OF AER ADL Shy <0 M’, the differential f, extends to a complex linear mapping of T:(M) into T5(M’),7 = f(x), which will be dénoted by the same symbol fa» Similarly f* maps each complex differential form of M’ into a complex differential form of M. Proposition 2.9. For a mapping f of an almost complex manifold -M into another almost complex manifold M’, the following conditions are equivalent: (a) If Z is a complex vector of type (1,0) of M; so is the vector Sa(Z) of M's (b) Uf Zis a complex vector of type (0, 1) of M, 50 is the vector fy(Z) of M5 (c) If @ is a form of degree (p, g) on M’, then f*(w) is @ form of degree (p, 9) on M for all p and q; (d) f is almost complex. Proof. The equivalence of (a) and (b) follows from the fact that Zs of type (0, 1) ifand only if Zs of type (1, 0) (cf. Proposi- tion 1.5) and the fact that f, commutes with the complex conjuga- tion. Since (a) (resp. (b)) is equivalent to the condition that f* maps every form of degrce (0, 1) (resp. (1, 0)) into a form of the same degree (cf. Proposition 1.6) and since the algebra of complex forms is locally generated by functions, forms of degree (1, 0) and forms of degree (0, 1), we may conclude that (c) is equivalent to (a) and (b). To prove the equivalence of (a) and (d), let J and J’ denote the almost complex structures of M and M’, respectively. Assume (a). If X isa real tangent vector of M, then X — iJX is of type (1, 0) by Proposition 2.6. Since f,(X — iSX) = f,(X) — i{fy(JX)), the vector f,(X) — i(f,(FX)) is of type (1, 0) by (a). Hence, f, (JX) = J'(f,(X)), showing that f is almost complex. Assume (d). If Z is a complex tangent vector of type (1, 0) of M, then Z = X —iJX for some real tangent vector X by Proposition 2.6. From f, (JX) =J'(f,(X)), we obtain SfalX = 15K) =fo(X) ~ i) =fel0) - VO), showing that Z is mapped by f into a vector of type (1, 0) QED. An infinitesimal automorphism of an almost complex structure J128 FOUNDATIONS OF DIFFERENTIAL GEOMETRY on Misa vector field X such that LJ = 0, where Ly denotes the Lie differentiation with respect to X. By Corollary 3.7 of Chapter I, a vector field X is an infinitesimal automorphism of J if and only if it generates a local 1-parameter group of local almost complex transformations. Proposimion 2.10, A vector field X is an infinitesimal automorphism of an almost complex structure J on a manifold M if and only if [X, JY] = J(%, ¥]) for all vector fields ¥ on M. Proof. Let X and ¥ be any vector fields on M. By Proposition 3.2 of Chapter I, we have [X, JY] = Lx (JY) = (LxJ)¥ + J(Lx¥) = (LxJ)¥ + J(LX, ¥)). Hence, LJ = 0 if and only if [X, J¥] ([X, Y]) for all ¥. QED. If Xis an infinitesimal automorphism of J, JX need not be. In fact, if Xis an infinitesimal automorphism of J and Y is arbitrary, then by Proposition 2.10 we have MX, ¥) = (UX, JY] — JUX, ¥)), showing that JX is also an infinitesimal automorphism of J if and only if M(X, ¥) = 0 for all ¥. If Y=, then the Lie algebra a of infinitesimal automorphisms of J is stable under J and, by Proposition 2.10, [X, JY] = J(LX, ¥]) for X, ¥ € a. Consequently, a is a complex Lie algebra (possibly of infinite dimensions), the complex structure being defined by J (cf. Example 1.1). Let M be a complex manifold and 2!,..., 2", 2 = + ia complex local coordinate system of M. Then dz! = dx! + i dy! and de) = ds! — idys, We set ade! = 4(a]0x' — i(8/a)), AfaB! = Hajar + (9/2)). From the construction of the complex structure J given before Proposition 2.2, it follows that 2/224, ..., 8/82" (resp. 8/@2,..., aa2") form a basis for T!* (resp. 2+) at each point x of the coordinate neighborhood and dz!, ..:, d2* (resp. di, ..., dz") form a local basis for © (resp. ©), Consequently, the set of 4g Soe TX, COMPLEX MANIFOLDS 129 forms dz" A+++ Adz AGM As A ABLE fy Soi < jp St and 1 3 ky <-+* < hy Sm forms a local basis for ©", A form @ of degree (p,0) is said to be holomorphic if Jo = 0. If we express @ in terms of z!,..., 2*: = SZ _ Sys dh nee h dol, snes then So = 0 if and only if § . In general, if fis a complex-valued function, then 4/0 if and only if 4/182 =0 for j=1, -.., . On the other hand, the definition of /a2! given above shows that the equations @f}22/ = 0 are nothing but the Cauchy-Riemann equations. It follows that « is holomorphic if and only if the coefficient functions f,.., are all holomorphic. AA holomorphic vector field on. a complex manifold M is a complex vector field Z of type (1, 0) such that Zs holomorphic for every locally defined holomorphic function f. If we write z= prajae’ my in terms of z4,..., 2, then Z is holomorphic if and only if the components f? are all holomorphic functions. PRoposrrion 2.11. On a complex manifold M, the Lie algebra of infinitesimal automorphisms of the complex structure J is isomorphic with the Lie algebra of holomorphic vector fields, the isomorphism being given by X+Z = HX ~idX). Proof. Let X and ¥ be vector fields on M and, in terms of a local coordinate system 24, ... , 2", we write (cf. Proposition 2.6) HX UX) = Speen, HY — iY) = 3 pia] ‘A simple calculation shows that the equation [X, JY] = J([X, ¥]) is equivalent to the system of equations E7., £(2/1/a2) = 0 for j=1,..., 1. From Proposition 2.10 we may conclude that X is ‘an infinitesimal automorphism of J if and only if 3/62" = 0 for 5k =1,...,1, thatis, ifand only if }(X — iJ) is holomorphic. Let X and Y¥ be infinitesimal automorphisms of J. Since J has no torsion, both JX and JY are also infinitesimal automorphisms of J as we saw just after Proposition 2.10. By Proposition 2.10, we have J[X, Y] = [JX, ¥] = [X, JY] and [X, ¥] = -(VX, JY].130 FOUNDATIONS OF DIFFERENTIAL GEOMETRY Using these identities we may easily verify that the mapping 6: X— {X — iJX) satisfies 0([X, Y]) = [0(X), 0(¥)]- Clearly, 6 is one-to-one. Finally, given a holomorphic vector field Z we show that there is an infinitesimal automorphism X of J such that @(X) = Z. In any coordinate neighborhood U, we write Z = Ep. /0(a/a2!) and set f? = & + inf. Then we have Z = (a) E(H(2/ax) + (A/} — EX (—W1(00e) + (2/8)}. If we take Xy = ER, ( (9/2) + n'(9/ay")}, then Z = (9) x (Xy —iJXy). If Xp is a vector field on another coordinate neighborhood V such that Z = (4)(Xy — iJX,), then it is clear that Xy = X, on U 0 V. Thus there is a vector field X on M such that Z = 6(X). We have shown that @ is an isomorphism of the Lie algebra of infinitesimal automorphisms of J onto the Lie algebra of holomorphic vector fields on M. QED. We shall give a few examples of complex and almost complex manifolds. Example 2.1. A complex Lie group G is a group which is at the same time a complex manifold such that the group operations (a, 8) €@ x G+ab-‘eG is a holomorphic mapping of G x G into G, Here @ x G is provided with the product complex struc- ture. (More generally, if M and M’ have almost complex structures Sand J’, respectively, then M x M’ has a product almost com- plex structure J x J’ whose definition is rather obvious. If J and J’ are integrable (or complex structures), so is J x J’.) The group GL(n; ©) isa complex Lic group; its complex structure is obtained by considering it as an open subset of C™, Its Lie algebra gl(n; C) consists of all x x m complex matrices. IfG is a complex Lie group, its complex structure J is invariant by the mappings L,: x ax, Ry: x + xa, ad (a): x > axa™, and yx x4, IPX is a left invariant vector field, so is JX. Thus J induces a complex structure in the Lie algebra g of G. Since J is invariant by ad (a), a € G, it follows that J's ad (X) = ad (X)°J for every X € g. Thus gis a complex Lie algebra (cf. Example 1.1). Conversely, let G be a Lie group whose Lie algebra g is a complex 1X, COMPLEX MANIFOLDS 131 Lie algebra (with coniplex structure J). Assuming first that G is connected, we get from Jad (X) =ad(X)eJ, Xeg, that Joad (a) = ad (2) °J for every a €G. If we transfer J to each tangent space T,(G) by Z4, we get a left invariant almost complex structure J on G which is also right invariant. Since [JX, Y] = JLX, ¥] for all left invariant vector fields X and ¥, we see that the torsion N(X, ¥) is 0 for such X and ¥. Thus N = 0. Since J is real analytic (as a left invariant tensor field on a Lie group), we conclude that G is a complex manifold. Since L, and Ry 2 €G, preserve J, the Leibniz formula (Proposition 1.4 of Chapter I} shows that the mapping (x, 7) €G x G + xy €G is holomorphic. ‘The mapping : x -+ x is holomorphic at ¢ (since its differential is —1, which commutes with J) and hence everywhere, because yo Ly = Reve y implies that (pe), = (Ry-t)ne° (Pas ® (La)ee BPE serves J. Thus G is a complex Lie group. The case where Gis not connected is left to the reader. Example 2.2. Consider C* as a vector space and let ry, +++ 5 T1» be any basis of C* over the field of real numbers. Let D be the subgroup of C* generated by ty, -++5 Taq? D = (28% myrys my integers}. The action of D on C* is properly discontinuous and the quotient manifold €*/D is called an n-dimensional complex torus (cf. Proposition 4.3 of Chapter I). In contrast to the case of real tori, two complex tori of the same dimension need not be holomorphically isomorphic. A complex torus C*/D is called an ‘abelian variety if there is a real skew-symmetric bilinear form E on © such that (1) EX, ¥) = BUY, X) for X, Yer; (2) BUX, X) >0 "for every nonzero X ¢ (3) E(X, ¥) is an integer if X, Ye D, For n = 1, every complex torus is an abelian variety. As a quotient group of C*, C*/D is a connected compact com- plex Lie group. Conversely, every connected compact complex Lie group G is a complex torus, In fact, the adjoint representation of G is a holomorphic mapping of G into GL(n; €) < G", where n = dim G. Since a holomorphic function on a compact complex manifold ‘must be constant, the adjoint representation of G is trivial, that is, Gis abelian. Thus G is an even-dimensional torus (Pontrjagin [1]) and hence a complex torus.