You are on page 1of 177

On the Geometry of Grassmannians and the Symplectic Group: the Maslov Index and Its Applications.

Paolo Piccione Daniel Victor Tausk


Notes for a short course given at the XI Escola de Geometria Diferencial, Universidade Federal Fluminense, Niteroi, RJ, Brazil

Table of Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v 1 Symplectic Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 A Short Review of Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Complex Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3 Complexication and Real Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.3.1 Complex structures and complexications . . . . . . . . . . . . . . . . 14 1.4 Symplectic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.4.1 Isotropic and Lagrangian subspaces . . . . . . . . . . . . . . . . . . . . . . 22 1.4.2 Lagrangian decompositions of a symplectic space . . . . . . . . . 26 Exercises for Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 The Geometry of Grassmannians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.1 Differentiable Manifolds and Lie Groups . . . . . . . . . . . . . . . . . . . . . . . 32 2.1.1 Classical Lie Groups and Lie Algebras . . . . . . . . . . . . . . . . . . . 34 2.1.2 Actions of Lie Groups and Homogeneous Manifolds . . . . . . 38 2.1.3 Linearization of the Action of a Lie Group on a Manifold . . 42 2.2 Grassmannians and Their Differentiable Structure . . . . . . . . . . . . . . . 44 2.3 The Tangent Space to a Grassmannian . . . . . . . . . . . . . . . . . . . . . . . . . 47 2.4 The Grassmannian as a Homogeneous Space . . . . . . . . . . . . . . . . . . . 50 2.5 The Lagrangian Grassmannian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.5.1 The submanifolds k (L0 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Exercises for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Topics of Algebraic Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.1 The Fundamental Groupoid and Group . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.1.1 Stability of the homotopy class of a curve . . . . . . . . . . . . . . . . 70 3.2 The Homotopy Exact Sequence of a Fibration . . . . . . . . . . . . . . . . . . 73 3.2.1 Applications to the theory of classical Lie groups . . . . . . . . . 84 3.3 Singular Homology Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 3.3.1 The Hurewiczs homomorphism . . . . . . . . . . . . . . . . . . . . . . . . 102 Exercises for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
iii

iv

TABLE OF CONTENTS

The Maslov Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.1 Index of a Symmetric Bilinear Form . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.1.1 The evolution of the index of a one-parameter family of symmetric bilinear forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 4.2 Denition and Computation of the Maslov Index . . . . . . . . . . . . . . . 121 Exercises for Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 Some Applications to Differential Systems . . . . . . . . . . . . . . . . . . . . . . . . 134 5.1 Symplectic Differential Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .134 5.2 The Maslov Index of a Symplectic Differential System . . . . . . . . . 138 5.3 The Maslov Index of semi-Riemannian Geodesics and Hamiltonian Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 5.3.1 Geodesics in a semi-Riemannian manifold . . . . . . . . . . . . . . 142 5.3.2 Hamiltonian systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 5.3.3 Further developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 Exercises for Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 Answers and Hints to the Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 From Chapter 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .151 From Chapter 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .154 From Chapter 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .156 From Chapter 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .157 From Chapter 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .158 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

Introduction
It has become evident through many mathematical theories of our century that Geometry and Topology offer very powerful tools in the study of qualitative and also quantitative properties of differential equations. The main idea behind these theories is that some equations, or better, some classes of equations can be studied by means of their symmetries, where by symmetry we mean generically any algebraic or geometric structure which is preserved by their ow. Once such invariant structures are determined for a class of differential equations, many properties of the solutions of the class can be read off from the geometry of the curve obtained by the ow, taking values in the space (typically a Lie group) of all structurepreserving morphisms. A simple, but instructive, example is given by the Sturmian theory for second order ordinary differential equations in I R. The Sturm oscillation theorem deals with equations of the form (px ) + rx = x, where p and r are functions, p > 0, 1 [, ] the space and is a real parameter. The theorem states that, denoting by Co of C 1 -functions on [, ] vanishing at and , the index of the symmetric bilinear b 1 [a, b] is equal to the sum over t ]a, b[ form B (x, y ) = a [px y + rxy ] dt in Co t 1 [a, t]. of the dimension of the kernel of the bilinear form a [px y + rxy ] dt in Co The classical proof of the Sturm oscillation theorem (see for instance [4, Chapter 8]) is obtained by showing that the two quantities involved in the thesis can be obtained as the winding number of two homotopic closed curves in the real projective line. The class of differential equations that we are interested in consists in the so called symplectic differential systems; these are linear systems in I Rn I Rn whose ow preserve the canonical symplectic form, given by (v, ), (w, ) = (v ) (w). Recall that a symplectic form is a nondegenerate skew-symmetric bilinear form on a (necessarily even dimensional) vector space. These differential systems appear naturally in a great variety of elds of pure and applied mathematics, and many areas of mathematics and physics, like Calculus of Variations, Hamiltonian systems, (Pseudo-)Riemannian Geometry, Symplectic Geometry, Mechanics and Optimal Control Theory produce examples of symplectic systems as basic objects of investigation. For instance, MorseSturm systems are special cases of symplectic systems; such systems are obtained from the Jacobi equation along any pseudo-Riemannian geodesic by means of a parallel trivialization of the tangent bundle of the pseudo-Riemannian manifold along the geodesic. More in general, symplectic systems are obtained by considering the linearized Hamilton equations along any solution of a (possibly time-dependent) Hamiltonian problem,
v

vi

INTRODUCTION

using a symplectic trivialization along the solution of the tangent bundle of the underlying symplectic manifold. Another large class of examples where the theory leads naturally to the study of symplectic systems is provided by Lagrangian variational theories in manifolds, possibly time-dependent, even in the case of constrained variational problems. Indeed, under a suitable invertibility assumption called hyper-regularity, the solutions to such problems correspond, via the Legendre transform, to the solutions of an associated Hamiltonian problem in the cotangent bundle. The fundamental matrix of a symplectic system is a curve in the symplectic group, denoted by Sp(2n, I R), which is a closed subgroup of the general linear group GL(2n, I R), hence it has a Lie group structure. This structure is extremely rich, due to the fact that symplectic forms on a vector space are intimately related to its complex structures, and such relation produces other invariant geometric and algebraic structures, such as inner products and Hermitian products. Many interesting questions can be answered by studying solutions of symplectic systems whose initial data belong to a xed Lagrangian subspace of I Rn I Rn . Recall that a Lagrangian subspace of a symplectic space is a maximal subspace on which the symplectic form vanishes. Such initial conditions are obtained, for instance, in Riemannian or pseudo-Riemannian geometry when one considers Jacobi elds along a geodesic that are variations made of geodesics starting orthogonally at a given submanifold. Since symplectic maps preserve Lagrangian subspaces, the image of the initial Lagrangian by the ow of a symplectic system is a curve in the set of all Lagrangian subspaces of I Rn I Rn . The set is a smooth (indeed, real-analytic) submanifold of the Grassmannian Gn (I Rn I Rn ) of all n-dimensional subspaces of I Rn I Rn ; is called the Lagrangian Grassmannian n n of the symplectic space I R I R . The original interest of the authors was the study of conjugate points along geodesics in a semi-Riemannian manifold and their stability (see [29, 34]), with the aim of developing an innite dimensional Morse Theory (see [30, 12, 26]) for semi-Riemannian geodesics. A few decades ago a new integer valued homological invariant, called the Maslov index, was introduced by the Russian school (see for instance [1] and the references therein) for closed curves in a Lagrangian submanifold M of the space I R2n endowed with its canonical symplectic structure. The notion of Maslov index has been immediately recognized as an important tool in the study of conjugate points, and it has has been thoroughly investigated and extended in several directions by mathematical-physicists, geometers and analysts. There is nowadays a very extensive literature about the subject, and it is almost impossible to acknowledge the work of all the many authors who have given signicant contributions to the eld. Our list of references ([6, 9, 11, 14, 15, 16, 24, 28, 36, 42]) is far from being exhaustive. Periodic or non periodic solutions of Hamiltonian systems, like for instance geodesics in a semi-Riemannian manifold, dene a curve in the symplectic group, or in the Lagrangian Grassmannian, hence they dene a Maslov index. Roughly speaking, the Maslov index gives a sort of algebraic count of the conjugate points along a solution; here are some of the main properties of this invariant:

INTRODUCTION

vii

it is always nite (even when the number of conjugate points is innite); it is stable by small perturbations of the data; it coincides with the geometric index in the case of a causal (timelike or lightlike) Lorentzian geodesic; it is related to the analytic index (or, more in general, to the relative index) of the solution, which is the index of the second variation of an associated Lagrangian action functional; it is related to the spectral properties of the associated Hamiltonian second order differential operator.

Conjugate and focal points appear naturally in Optics, both classical and relativistic, and the Maslov index provides a new topological invariant. For instance, the optics of light rays in a general-relativistic medium, like vacuum, dust, or plasma, magnetized or not, etc., can be described using a Hamiltonian formalism. As the underlying spacetime model one assumes an arbitrary 4-dimensional Lorentzian manifold (M, g ), and the trajectories of light rays are projections onto M of solutions in the cotangent bundle T M of some Hamiltonian function H : TM I R. Typically, the explicit form of the function H involves the spacetime metric and a number of tensor elds by which the medium is characterized. For a comprehensive discussion of this subject we refer to Perlick [33]. The aim of this booklet is to provide a complete, self-contained study of the geometry of the Grassmannian manifolds, the symplectic group and the Lagrangian Grassmannian. This study will lead us naturally to the notion of Maslov index, that will be introduced in the context of symplectic differential systems. These notes are organized as follows. In Chapter 1 we describe the algebraic setup; we will study complex structures and symplectic structures on nite dimensional vector spaces. Special attention is given to the Lagrangian subspaces and to the Lagrangian decompositions of a symplectic space. Chapter 2 is entirely dedicated to Differential Geometry; we will study at the differentiable structure of the Grassmannian manifolds and of the Lagrangian Grassmannians. We will develop briey the theory of Lie groups and their actions on differentiable manifolds, so that we will be able to describe the Grassmannians as homogeneous spaces. In Chapter 3 we will develop the algebraic topological framework of the theory, including the basics of homotopy theory and of singular homology theory for topological spaces. Using the long exact sequences in homotopy and in homology, and using the Hurewicz homomorphism we will compute the rst homology and relative homology group of the Grassmannians. In Chapter 4 we will introduce the notion of Maslov index for curves in the Lagrangian Grassmannian, and we will present some methods to compute it in terms of the change of signature of curves of bilinear forms. In Chapter 5 we will show how symplectic differential systems are produced in several geometrical problems; more precisely, we will consider the case of the Jacobi equation along a semi-Riemannian geodesic, and the case of the linearized

viii

INTRODUCTION

Hamilton equation along the solution of a Hamiltonian problem in a symplectic manifold. At the end of each Chapter we have given a list of exercises whose solution is based on the material presented in the chapter. The reader should be able to solve the problems as he/she goes along; the solution or a hint for the solution of (almost) all the exercises is given in Appendix A.

INTRODUCTION

ix

Ognuno sta solo sul cuor della terra tratto da un raggio di sole: ed e ` subito sera.

CHAPTER 1

Symplectic Spaces
1.1. A Short Review of Linear Algebra In this section we will briey review some well known facts concerning the identication of bilinear forms and linear operators on vector spaces. These identications will be used repeatedly during the exposition of the material, and, to avoid confusion, the reader is encouraged to take a few minutes to go through the pain of reading this section. The results presented are valid for vector spaces over an arbitrary eld K , however we will mainly be interested in the case that K = I R or K = C. Moreover, we emphasize that even though a few results presented are also valid in the case of innite dimensional vector spaces, in this chapter we will always assume that the vector spaces involved are nite dimensional. Let V and W be vector spaces. We denote by Lin(V, W ) and by B(V, W ) respectively the vector spaces of all the linear operators T : V W and of bilinear operators, called also bilinear forms, B : V W K ; by V we mean the dual space Lin(V, K ) of V . Shortly, we set Lin(V ) = Lin(V, V ) and B(V ) = B(V, V ). There is a natural isomorphism: (1.1.1) Lin(V, W ) B(V, W ), which is obtained by associating to each linear operator T : V W the bilinear form BT B(V, W ) given by BT (v, w) = T (v ) w . 1.1.1. R EMARK . Given vector spaces V, W, V1 , W1 and a pair (L, M ) of linear operators, with L Lin(V1 , V ) and M Lin(W, W1 ), one denes another linear operator: (1.1.2) by: (1.1.3) Lin(L, M ) T = M T L. In this way, Lin(, ) becomes a functor, contravariant in the rst variable and covariant in the second, from the category of pairs of vector spaces to the category of vector spaces. Similarly, given linear operators L Lin(V1 , V ) and M Lin(W1 , W ), we can dene a linear operator B(L, M ) : B(V, W ) B(V1 , W1 ) by setting B(L, M ) B = B (L, M ). In this way, B(, ) turns into a functor, contravariant in both variables, from the category of pairs of vector spaces to the category of vector spaces. This abstract formalism will infact be useful later (see Section 2.3).
1

Lin(L, M ) : Lin(V, W ) Lin(V1 , W1 )

1. SYMPLECTIC SPACES

The naturality of the isomorphism (1.1.1) may be meant in the technical sense of natural isomorphism between the functors Lin(, ) and B(, ) (see Exercise 1.1). To avoid confusion, in this Section we will distinguish between the symbols of a bilinear form B and of the associated linear operator TB , or between a linear operator T and the associated bilinear form BT . However, in the rest of the book we will implicitly assume the isomorphism (1.1.1), and we will not continue with this distinction. Given another pair of vector spaces V1 and W1 and operators L1 Lin(V1 , V ), L2 Lin(W1 , W ), the bilinear forms BT (L1 , ) and BT (, L2 ) correspond via (1.1.1) to the linear operators T L1 and L 2 T respectively. Here, L2 : W W1 denotes the transpose linear operator of L2 given by: L 2 () = L2 , W .

We will identify every vector space V with its bidual V and every linear operator T with its bitranspose T . Given T Lin(V, W ) we will therefore look at T as an element in Lin(W, V ); if BT is the bilinear operator associated to T , then the bilinear operator BT associated to T is the transpose bilinear B(W, V ) dened by B (w, v ) = B (v, w ). operator BT T T Given B B(V ), we say that B is symmetric if B (v, w) = B (w, v ) for all v, w V ; we say that B is anti-symmetric if B (v, w) = B (w, v ) for all v, w V (see Exercise 1.2). The sets of symmetric bilinear forms and of antisymmetric bilinear forms are subspaces of B(V ), denoted respectively by Bsym (V ) and Basym (V ). The reader is warned that, unfortunately, the identication (1.1.1) does not behave well in terms of matrices, with the usual convention for the matrix representations of linear and bilinear operators. m n If (vi )n i=1 and (wi )i=1 are bases of V and W respectively, we denote by (vi )i=1 m and (wi )i=1 the corresponding dual bases of V and W . For T Lin(V, W ), the matrix representation (Tij ) of T satises:
m

T (vj ) =
i=1

Tij wi .

On the other hand, if B B(V, W ), the matrix representation (Bij ) of B is dened by: Bij = B (vi , wj ); hence, for all T Lin(V, W ) we have: Tij = T (vj )(wi ) = BT (vj , wi ) = [BT ]ji . Thus, the matrix of a linear operator is the transpose of the matrix of the corresponding bilinear operator; in some cases we will be considering symmetric operators, and there will be no risk of confusion. However, when we deal with symplectic forms (see Section 1.4) one must be careful not to make sign errors.

1.1. A SHORT REVIEW OF LINEAR ALGEBRA

1.1.2. D EFINITION . Given T Lin(V, W ), we dene the pull-back associated to T to be map: T # : B(W ) B(V ) given by T # (B ) = B (T , T ). When T is an isomorphism, we can also dene the push-forward associated to T , which is the map: T# : B(V ) B(W ) dened by T# (B ) = B (T 1 , T 1 ). 1.1.3. E XAMPLE . Using (1.1.1) to identify linear and bilinear operators, we have the following formulas for the pull-back and the push-forward: (1.1.4) T # (B ) = T TB T, T# (B ) = (T 1 ) TB T 1 .

The identities (1.1.4) can be interpreted as equalities involving the matrix representations, in which case one must use the matrices that represent B , T # (B ) and T# (B ) as linear operators. For B B(V ), the kernel of B is the subspace of V dened by: (1.1.5) Ker(B ) = v V : B (v, w) = 0, w V .

The kernel of B coincides with the kernel of the associated linear operator T : V V . The bilinear form B is said to be nondegenerate if Ker(B ) = {0}; this is equivalent to requiring that its associated linear operator T is injective, or equivalently, an isomorphism. 1.1.4. E XAMPLE . If B B(V ) is nondegenerate, then B denes an isomorphism TB between V and V and therefore we can dene a bilinear form [TB ]# (B ) in V by taking the push-forward of B by TB . By (1.1.4), such bilinear form is 1 ) ; if B is symmetric, then [TB ]# (B ) is the associated to the linear operator (TB 1 bilinear form associated to the linear map TB . 1.1.5. D EFINITION . Let B Bsym (V ) be a symmetric bilinear form in V . We say that a linear operator T : V V is B -symmetric (respectively, B -antisymmetric) if the bilinear form B (T , ) is symmetric (respectively, anti-symmetric). We say that T is B -orthogonal if T # [B ] = B , i.e., if B (T , T ) = B . 1.1.6. E XAMPLE . Given B Bsym and T Lin(V ), the B -symmetry of T is equivalent to: (1.1.6) TB T = (TB T ) ; clearly, the B -anti-symmetry is equivalent to TB T = (TB T ) . When B is nondegenerate, we can also dene the transpose of T relatively to Lin(V ) such that B (T v, w) = B (v, T w) for all B , which is the operator T v, w V . Explicitly, we have (1.1.7) = T 1 T T B . T B

1. SYMPLECTIC SPACES

= T (resp., iff T = T ), Then, T is B -symmetric (resp., B -anti-symmetric) iff T = T 1 . and it is B -orthogonal iff T . We also say that T is B -normal if T commutes with T Given a subspace S V and a bilinear form B B(V ), the orthogonal complement S of S with respect to B is dened by: (1.1.8) S = v V : B (v, w) = 0, w S .

In particular, Ker(B ) = V . The annihilator S o of S is the subspace of V dened as: S o = V : (w) = 0, w S .


1 Observe that S = TB (S o ).

1.1.7. E XAMPLE . Assume that B Bsym (V ) is nondegenerate and let T the B -transpose of T . If S V is an invariant subspace for Lin(V ); denote by T T , i.e., if T (S ) S , then the B -orthogonal complement S of S is invariant for . This follows from (1.1.7) and from the identity S = T 1 (S o ), observing that T B the annihilator S o of S is invariant for T . 1.1.8. P ROPOSITION . If B B(V ) is nondegenerate and S V is a subspace, then dim(V ) = dim(S ) + dim(S ). P ROOF. Simply note that dim(V ) = dim(S ) + dim(S o ) and that dim(S ) = 1 (S o ), with TB an isomorphism, because B is nondegendim(S o ), and S = TB erate. If B is either symmetric or anti-symmetric, then it is easy to see that S (S ) ; the equality does not hold in general, but only if B is nondegenerate. 1.1.9. C OROLLARY. Suppose that B B(V ) is either symmetric or antisymmetric; if B is nondegenerate, then S = (S ) . P ROOF. It is S (S ) ; by Proposition 1.1.8 dim(S ) = dim (S ) . If B B(V ) is nondegenerate and S V is a subspace, then the restriction of B to S S may be degenerate. We have the following: 1.1.10. P ROPOSITION . The restriction B |S S is nondegenerate if and only if V = S S. P ROOF. The kernel of the restriction B |S S is S S ; hence, if V = S S , it follows that B is nondegenerate on S . Conversely, if B is nondegenerate on S , then S S = {0}. It remains to show that V = S + S . For, observe that the map: (1.1.9) S x B (x, )|S S is an isomorphism. Hence, given v V , there exists x S such that B (x, ) and B (v, ) coincide in S , thus x v S . This concludes the proof.

1.2. COMPLEX STRUCTURES

1.1.11. C OROLLARY. Suppose that B B(V ) is either symmetric or antisymmetric; if B is nondegenerate, then the following are equivalent: B is nondegenerate on S ; B is nondegenerate on S . P ROOF. Assume that B is nondegenerate on S . By Proposition 1.1.10 it is V = S S ; by Corollary 1.1.9 we have V = S (S ) , from which it follows that B is nondegenerate on S by Proposition 1.1.10. The converse is analogous, since (S ) = S . 1.1.12. E XAMPLE . Proposition 1.1.10 actually does not hold if V is not nite dimensional. For instance, if V is the space of square summable sequences x = 2 (xi )iI N of real numbers, i.e., iI N xi < +, B is the standard Hilbert product in V given by B (x, y ) = iI N xi yi and S V is the subspace consisting of all almost null sequences, i.e., xi = 0 only for a nite number of indices i I N, then it is easy to see that S = {0}. What happens here is that the map (1.1.9) is injective, but not surjective. 1.1.13. R EMARK . Observe that Proposition 1.1.10 is indeed true if we assume only that S is nite dimensional; for, in the proof presented, only the niteness of dim(S ) was used to conclude that the map (1.1.9) is an isomorphism. As an application of Proposition 1.1.10 we can now prove that every symmetric bilinear form is diagonalizable. We say that a basis (vi )n i=1 of V diagonalizes the bilinear form B if B (vi , vj ) = 0 for all i = j , i.e., if B is represented by a diagonal matrix in the basis (vi )n i=1 . 1.1.14. T HEOREM . Suppose that K is a eld of characteristic different from 2. Given B Bsym (V ), there exists a basis (vi )n i=1 of V that diagonalizes B . P ROOF. We prove the result by induction on dim(V ). If dim(V ) = 1 the result is trivial; assume dim(V ) = n and that the result holds true for every vector space of dimension less than n. If B (v, v ) = 0 for all v V , then B = 0. For, 0 = B (v + w, v + w) = 2 B (v, w), and the led K has characteristic different from 2. Since the result in the case that B = 0 is trivial, we can assume the existence of v1 V such that B (v1 , v1 ) = 0. It follows that B is nondegenerate on the one-dimensional subspace K v1 generated by v1 ; by Proposition 1.1.10 we get: V = K v1 (K v1 ) .
By the induction hypothesis, there exists a basis (vi )n i=1 of (K v1 ) that diagonaln izes the restriction of B ; it is then easy to check that the basis (vi )i=2 diagonalizes B.

1.2. Complex Structures In this section we will study the procedure of changing the scalar eld of a real vector space, making it into a complex vector space. Of course, given a complex

1. SYMPLECTIC SPACES

vector space, one can always reduce the scalars to the real eld: such operation will be called reduction of the scalars. Passing from the real to the complex eld requires the introduction of an additional structure, that will be called a complex structure. Many of the proofs in this section are elementary, so they will be omitted and left as an exercise for the reader. For clarity, in this section we will refer to linear operators as I R-linear or Clinear, and similarly we will talk about I R-bases or C-bases, real or complex dimension, etc. Let V be a complex vector space; we will denote by VI R the real vector space obtained by restriction of the multiplication by scalars C V V to I RV V . Observe that the underlying set of vectors, as well as the operation of sum, coincides in V and VI R . We say that VI R is a realication of V , or that VI R is obtained by a reduction of scalars from V . The endomorphism v iv of V given by the multiplication by the imaginary unit i = 1 is C-linear, hence also I R-linear. The square of this endomorphism is given by minus the identity of V . This suggests the following denition: 1.2.1. D EFINITION . Let V be a real vector space. A complex structure in V is a linear operator J : V V such that J 2 = J J = Id. Clearly, a complex structure J is an isomorphism, since J 1 = J . Given a complex structure J on V it is easy to see that there exists a unique way of extending the multiplication by scalars I R V V of V to a multiplication by scalar C V V in such a way that J (v ) = iv . Explicitly, we dene: (1.2.1) (a + b i)v = av + b J (v ), a, b I R, v V. Conversely, as we had already observed, every complex extension of multiplication by scalars for V denes a complex structure on V by J (v ) = iv . We will henceforth identify every pair (V, J ), where V is a real vector space and J is a complex structure of V , with the complex vector space V obtained from (V, J ) by (1.2.1). Observe that V is the realication VI R of V . 1.2.2. E XAMPLE . For every n I N , the space I R2n has a canonical complex structure dened by J (x, y ) = (y, x), for x, y I Rn . We can identify (I R 2n , J ) with the complex vector space Cn by (x, y ) x + iy . In terms of matrix representations, we have: 0 I (1.2.2) J= , I 0 where 0 and I denote respectively the 0 and the identity n n matrices. We have the following simple Lemma: 1.2.3. L EMMA . Let (V1 , J1 ) and (V2 , J2 ) be real vector spaces endowed with complex structures. A I R-linear operator T : V1 V2 is C-linear if and only if T J1 = J2 T . In particular, the C-linear endomorphisms of a vector space with complex structure (V, J ) are the I R-linear endomorphisms of V that commute with J.

1.2. COMPLEX STRUCTURES

P ROOF. Left to the reader in Exercise 1.3. 1.2.4. R EMARK . Observe that if J is a complex structure on V , then also J is a complex structure, that will be called the conjugate complex structure. For C and v V , the product of and v in the complex space (V, J ) is given by the and v in the complex space (V, J ), where is the complex conjugate product of of . The set of complex bases of (V, J ) and (V, J ) coincide; observe however that the components of a vector in a xed basis are conjugated when replacing J by J . A C-linear operator T between complex spaces is still C-linear when replacing the complex structures by their conjugates in both the domain and the counterdomain. The representations of T with respect to xed bases in the complex structures and the same bases in the conjugate complex structures are given by conjugate matrices. 1.2.5. D EFINITION . A map T between complex vector spaces is said to be (v ) for all anti-linear, or conjugate linear, if it is additive and if T (v ) = T C and all v in the domain of T . An anti-linear map is always I R-linear when we see it as a map between the realications of the domain and the counterdomain. Moreover, a map is anti-linear if and only if it is C-linear when the complex structure of its domain (or of its counter domain) is replaced by the complex conjugate. In particular, the anti-linear endomorphisms of (V, J ) are the I R-linear endomorphisms of V that anti-commute with J . We have the following relation between the bases of (V, J ) and of V : 1.2.6. P ROPOSITION . Let V be a (possibly innite dimensional) real vector space and J a complex structure on V . If (bj )j J is a C-basis of (V, J ), then the union of (bj )j J and J (bj ) j J is an I R-basis of V . P ROOF. Left to the reader in Exercise 1.4. 1.2.7. C OROLLARY. The real dimension of V is twice the complex dimension of (V, J ); in particular, a (nite dimensional) vector space admits a complex structure if and only if its dimension is an even number. P ROOF. We only need to show that every real vector space of even dimension admits a complex structure. This is easily established by choosing an isomorphism with I R2n and using the canonical complex structure given in Example 1.2.2. 1.2.8. E XAMPLE . If (V, J ) is a real vector space with complex structure, then the dual complex space of (V, J ) is given by the set of I R-linear maps : V C such that: (1.2.3) J (v ) = i (v ), v V. It is easy to see that (1.2.8) determines the imaginary part of when it is known its real part; hence we have an I R-linear isomorphism: (1.2.4) (V, J ) V ,

1. SYMPLECTIC SPACES

where : C I R denotes the real part operator. The isomorphism (1.2.4) therefore induces a unique complex structure of V that makes (1.2.4) into a C-linear isomorphism. Such complex structure is called the dual complex structure, and it is easy to see that it is given simply by the transpose operator J . We conclude this section with a relation between the matrix representations of vectors and operators in real and complex bases. Let (V, J ) be a 2n-dimensional vector space with complex structure; a basis of V adapted to J , shortly a J -basis, is a basis of the form (1.2.5) (b1 , . . . , bn , J (b1 ), . . . , J (bn ));

in this case, (bj )n j =1 is a complex basis of (V, J ). For instance, the canonical basis 2 n of I R , endowed with the canonical complex structure, is a J -basis corresponding to the canonical basis of Cn . In other words, the J -bases of a vector space are precisely those with respect to which the matrix representations of J is that given by (1.2.2). The existence of J -bases is given by Proposition 1.2.6. Let a J -basis of V be xed, corresponding to a complex basis B = (bj )n j =1 of (V, J ). Given v V with coordinates (z1 , . . . , zn ) in the basis B , then its coordinates in the (real) J -basis of V are: v (x1 , . . . , xn , y1 , . . . , yn ), where zj = xj + i yj , xj , yj I R. If T is a C-linear operator represented in the complex basis by the matrix Z = A + i B (A and B real), then its representation in the corresponding J -basis is: (1.2.6) T A B B A .

1.2.9. R EMARK . Formula (1.2.6) tells us that the map Z =A+Bi A B B A

is an injective homomorphism of the algebra of complex n n matrices into the algebra of real 2n 2n matrices. 1.3. Complexication and Real Forms In this section we show that any real vector space can be extended in a canonical way to a complex vector space, by mimicking the relation between I Rn and Cn ; such an extension will be called a complexication of the space. We also show that, given a complex space, it can be seen as the complexication of several of its real subspaces, that will be called the real forms of the complex space. We will only consider the case of nite dimensional spaces, even though many of the results presented will hold in the case of innite dimensional spaces, up to minor modications. Some of the proofs are elementary, and they will be omitted and left to the reader as Exercises.

1.3. COMPLEXIFICATION AND REAL FORMS

1.3.1. D EFINITION . Let V be a complex vector space; a real form in V is a real subspace V0 of V (or, more precisely, a subspace of the realication VI R of V ) such that: VI R = V0 i V0 . In other words, a real form V0 in V is a real subspace such that every v V can be written uniquely in the form v = v1 + i v2 , with v1 , v2 V0 . To a real form V0 we associate maps: (1.3.1) : V V0 , : V V0 , c : V V , given by (v1 + i v2 ) = v1 , (v1 + i v2 ) = v2 and c(v1 + i v2 ) = v1 i v2 , for all v1 , v2 V . We call , and c respectively the real part, imaginary part, and conjugation operators associated to the real form V0 . All these operators are I R-linear; the operator c is also anti-linear. For v V , we also say that c(v ) is the conjugate of v relatively to the real form V0 , and we also write: c(v ) = v . 1.3.2. D EFINITION . Let V be a real vector space. A complexication of V is a pair (V C , ), where V C is a complex vector space and : V V C is an injective I R-linear map such that (V ) is a real form in V C . The result of the following Proposition is usually known as the universal property of the complexication: 1.3.3. P ROPOSITION . Let V be a real vector space, (V C , ) a complexication of V and W a complex vector space. Then, given an I R-linear map f : V WI R, C there exists a unique C-linear map f : V W such that the following diagram commutes: (1.3.2) VOC
f

! /W

P ROOF. Left to the reader in Exercise 1.5. As corollary, we obtain the uniqueness of a complexication, up to isomorphisms: 1.3.4. C OROLLARY. Suppose that (V1C , 1 ) and (V2C , 2 ) are complexications of V . Then, there exists a unique C-linear isomorphism : V1C V2C such that the following diagram commutes: V1C

`AA AA 1 AAA

/ VC > 2 ~~ ~ ~ ~~ 2 ~~

P ROOF. Left to the reader in Exercise 1.6.

10

1. SYMPLECTIC SPACES

If V is a real vector space, we can make the direct sum V V into a complex space by taking the complex structure J (v, w) = (w, v ). Setting (v ) = (v, 0), it is easy to see that (V V, ) is a complexication of V , that will be called the canonical complexication of the real vector space V . By Corollary 1.3.4, we do not need to distinguish between complexications of a vector space; so, from now on, we will denote by V C the canonical complexication of V , or, depending on the context, we may use the symbol V C to denote some other complexication of V , which is necessarily isomorphic to the canonical one. The original space V will then be identied with (V ), so that we will always think of an inclusion V V C ; since (V ) is a real form in V C , then V C is a direct sum of V and i V : V C = V i V. 1.3.5. E XAMPLE . The subspace I Rn Cn is a real form in Cn , hence Cn is a n complexication of I R . 1.3.6. E XAMPLE . The space Mn (I R) of real n n matrices is a real form in the space Mn (C) of complex n n matrices. A less trivial example of a real form in Mn (C) is given by u(n), which is the space of anti-Hermitian matrices, i.e., matrices A such that A = A, where A denotes the conjugate transpose matrix of A. In this example, i u(n) is the space of Hermitian matrices, i.e., the space of those matrices A such that A = A. It is easy to see that Mn (C) = u(n) i u(n), and so u(n) is a real form in Mn (C) and Mn (C) is a complexication of u(n). 1.3.7. E XAMPLE . If V is a complex vector space and if (bj )n j =1 is a complex basis of V , then the real subspace V0 of VI R given by:
n

V0 =
j =1

j bj : j I R, j

is a real form in V . Actually, every real form of V can be obtained in this way; for, if V0 V is a real form, then an I R-basis (bj )n j =1 of V0 is also a C-basis of V . It follows in particular that the real dimension of a real form V0 is equal to the complex dimension of V . Example 1.3.7 tells us that every complex space admits innitely many real forms; in the following proposition we give a characterization of the real forms in a complex space. We recall that a bijection of a set is said to be an involution if 2 = = Id. 1.3.8. P ROPOSITION . Let V be a complex space. Then there exists a bijection between the set of real forms in V and the set of the anti-linear involutive automorphisms of V . Such bijection is obtained by: associating to each real form V0 V its conjugation operator c (see (1.3.1));

1.3. COMPLEXIFICATION AND REAL FORMS

11

associating to each anti-linear involutive automorphism c of V the set of its xed points V0 = {v V : c(v ) = v }. The above result suggests an interesting comparison between the operation of realication of a complex space and the operation of complexication of a real space. In Section 1.2 we saw that, roughly speaking, the operations of realication and of addition of a complex structure are mutually inverse; the realication is a canonical procedure, while the addition of a complex structure employs an additional information, which is an automorphism J with J 2 = Id. In this section we have the opposite situation. The complexication is a canonical operation, while its inverse operation, which is the passage to a real form, involves an additional information, which is an anti-linear involutive automorphism. Let us look now at the complexication as a functorial construction. Let V1 and V2 be real spaces; from the universal property of the complexication (Proposition 1.3.3) it follows that each linear operator T : V1 V2 admits a unique C-linear extension T C : V1C V2C . We have the following commutative diagram: V1C V2C V1 V2
T TC

The operator is called the complexication of T ; more concretely, we have that C T is given by: T C (v + i w) = T (v ) + i T (w), It is immediate that: (1.3.3) (1.3.4)
C C (T1 T2 )C = T1 T2 ,

TC

v, w V1 .

IdC = Id,

and, when T is invertible (T C )1 = (T 1 )C . The identities (1.3.3) imply that the complexication V V C , T T C is a functor from the category of real vector spaces with morphisms the I R-linear operators to the category of complex vector spaces with morphisms the C-linear operators. Given a linear operator T : V1 V2 , it is easy to see that: (1.3.5) Ker(T C ) = (Ker(T ))C , Im(T C ) = Im(T )C ; in the context of categories, the identities (1.3.5) say that the complexication is an exact functor, i.e., it takes short exact sequences into short exact sequences. If U V is a subspace, it is easy to see that the complexication iC of the inclusion i : U V is injective, and it therefore gives an identication of U C with a subspace of V C . More concretely, the subspace U C of V C is the direct sum of the two real subspaces U and i U of V C ; equivalently, U C is the complex subspace of V C generated by the set U V C . However, not every subspace of V C is the complexication of some subspace of V . We have the following characterization:

12

1. SYMPLECTIC SPACES

1.3.9. L EMMA . Let V be a real vector space and let Z V C be a complex subspace of its complexication. Then, there exists a real subspace U V with Z = U C if and only if Z is invariant by conjugation, i.e., c(Z ) Z , where c is the conjugation relative to the real form V V C . If Z = U C , then such U is uniquely determined, and it is given explicitly by U = Z V . P ROOF. Left to the reader in Exercise 1.7. Given real vector spaces V1 and V2 , observe that the map: (1.3.6) Lin(V1 , V2 ) T T C Lin(V1C , V2C )

is I R-linear; we actually have the following: 1.3.10. L EMMA . The map (1.3.6) takes Lin(V1 , V2 ) isomorphically onto a real form in Lin(V1C , V2C ). P ROOF. Since (V2C )I R = V2 i V2 , it is easy to see that: (1.3.7) Lin V1 , V2C
I R

= Lin(V1 , V2 ) iLin(V1 , V2 ).

From the universal property of the complexication, it follows that the restriction operator (1.3.8) Lin V1C , V2C S S|V1 Lin V1 , V2C
=

I R

is an isomorphism. From (1.3.7) and (1.3.8) it follows: (1.3.9) Lin V1C , V2C = Lin(V1 , V2 ) Lin(V1 , V2 ),

where the two summands on the right of (1.3.9) are identied respectively with the image of (1.3.6) and with the same image multiplied by i. From Lemma 1.3.8 it follows in particular that the dual V = Lin(V, I R) can be identied with a real form of the dual of the complexication (V C ) = Lin(V C , C) (compare with Example 1.2.8). Along the same lines of Lemma 1.3.9, in the next lemma we characterize the image of (1.3.6): 1.3.11. L EMMA . Let V1 , V2 be real vector spaces. Given a C-linear map S : V1C V2C , the following statements are equivalent: there exists an I R-linear map T : V1 V2 such that S = T C ; S preserves real forms, i.e., S (V1 ) V2 ; S commutes with conjugation, i.e., c S = S c, where c denotes the conjugation operators in V1C andV2C with respect to the real forms V1 and V2 respectively. When one (hence all) of the above conditions is satised, there exists a unique T Lin(V1 , V2 ) such that S = T C , which is given by the restriction of S .

1.3. COMPLEXIFICATION AND REAL FORMS

13

1.3.12. E XAMPLE . Let V1 , V2 be real vector spaces; choosing bases for V1 and V2 , the same will be bases for the complexications V1C and V2C (see Example 1.3.7). With respect to these bases, the matrix representation of a linear operator T : V1 V2 is equal to the matrix representation of its complexication T C : V1C V2C (compare with the result of Section 1.2, and more in particular with formula (1.2.6)). In terms of matrix representations, the map (1.3.6) is simply the inclusion of the real matrices into the complex matrices. 1.3.13. E XAMPLE . The real form in Lin(V1C , V2C ) dened in the statement of Lemma 1.3.10 corresponds to a conjugation operator in Lin(V1C , V2C ); given S Lin(V1C , V2C ), we denote by S its conjugate operator. Explicitly, S is given by: S = c S c. For, using Proposition 1.3.8 and Lemma 1.3.11, it sufces to observe that S c S c denes an anti-linear involutive automorphism of Lin(V1C , V2C ) whose xed point set is the image of (1.3.6). Observe that we have the identity: S (v ) = S ( v ), v V1C .

In terms of bases, the matrix representation of S is the complex conjugate of the matrix representation of S . The theory presented in this section can be easily generalized to the case of multi-linear operators, anti-linear operators and operators with mixed multilinearity, like sesquilinear operators. The latter case has special importance: 1.3.14. D EFINITION . Given complex vector spaces V1 , V2 and V , we say that a map B : V1 V2 V is sesquilinear if for all v1 V1 the map B (v1 , ) is anti-linear and for all v2 V2 the map B (, v2 ) is C-linear. If V1 = V2 and if a real form is xed in V , we say that a sesquilinear map B is Hermitian (respectively, anti-Hermitian) if B (v1 , v2 ) = B (v2 , v1 ) (respectively, B (v1 , v2 ) = B (v2 , v1 )) for all v1 , v2 V1 . A Hermitian form in a complex space V is a sesquilinear Hermitian map B : V V C; if B is positive denite, i.e., B (v, v ) > 0 for all v = 0, we also say that B is a positive Hermitian product, or simply an Hermitian product, in V . In the same way that we dene the complexication T C for an I R-linear operator, we can dene the complexication B C of an I R-multilinear operator B : V1 Vp V as its unique extension to a C-multi-linear operator B C : V1C VpC V C . Similarly, we can associate to an I R-linear operator its C C C R-bilinear unique extension T : V1 V2 to an anti-linear operator, and to an I operator B : V1 V2 V its unique sesquilinear extension B Cs : V1C V2C V C . In Exercise 1.8 the reader is asked to generalize the results of this section, in particular Proposition 1.3.3, Lemma 1.3.10 and Lemma 1.3.11, to the case of multi-linear, conjugate linear or sesquilinear operators.

14

1. SYMPLECTIC SPACES

1.3.15. E XAMPLE . If V is a real vector space and B Bsym (V ) is a symmetric bilinear form on V , then the bilinear extension B C of B to V C is symmetric; on the other hand, the sesquilinear extension B Cs of B is a Hermitian form on V C . Similarly, the bilinear extension of an anti-symmetric bilinear form is anti-symmetric, while the sesquilinear extension of an anti-symmetric form is anti-Hermitian. The notions of kernel (see (1.1.5)), nondegeneracy and orthogonal complement (see (1.1.8)) for symmetric and anti-symmetric bilinear forms generalize in an obvious way to sesquilinear Hermitian and anti-Hermitian forms. If B is symmetric (or anti-symmetric), it is easy to see that the condition of nondegeneracy of B is equivalent to the nondegeneracy of either B C or B Cs . Moreover, if B Bsym (V ) is positive denite, i.e., B (v, v ) > 0 for all v = 0, then its sesquilinear extension B Cs is also positive denite. Observe that the C-bilinear extension B C will be nondegenerate, but it is not positive denite (see Exercise 1.9). For instance, the canonical inner product of I Rn is given by:
n

x, y =
j =1

xj yj .

Its sesquilinear extension denes the canonical Hermitian product in Cn , given by


n

(1.3.10)

z, w

Cs

=
j =1

zj w j ,

while its C-bilinear extension is given by:


n

z, w

=
j =1

zj wj .

1.3.16. R EMARK . In the spirit of Denition 1.1.5, given a complex space V and a Hermitian form B in V , we say that a C-linear operator T Lin(V ) is B Hermitian (respectively, B -anti-Hermitian) if B (T , ) is a Hermitian (respectively, anti-Hermitian) form. We also say that T is B -unitary if B (T , T ) = B . Given a real vector space V , B Bsym (V ) and if T Lin(V ) is a B symmetric (respectively, B -anti-symmetric) operator, then its complexication T C in Lin(V C ) is a B Cs -Hermitian (respectively, B Cs -anti-Hermitian) operator. If T is B -orthogonal, then T C is B Cs -unitary. 1.3.1. Complex structures and complexications. The aim of this subsection is to show that there exists a natural correspondence between the complex structures of a real space V and certain direct sum decompositions of its complexication V C . Let V be a real vector space and let J : V V be a complex structure in V ; we have that J C is a C-linear automorphism of the complexication V C that satises (J C )2 = Id. It is then easy to see that V C decomposes as the direct sum of the two eigenspaces of J C corresponding to the eigenvalues i and i respectively;

1.3. COMPLEXIFICATION AND REAL FORMS

15

more explicitly, we dene: V h = v V C : J C (v ) = iv , V a = v V C : J C (v ) = iv . Then, V h and V a are complex subspaces of V C , and V C = V h V a ; the projections onto the subspaces V h and V a are given by: v iJ C (v ) v + iJ C (v ) , a (v ) = , v V C. 2 2 We call the spaces V h and V a respectively the holomorphic and the anti-holomorphic subspaces of V C . Next proposition justies the names of these spaces (see also Example 1.3.18 below): (1.3.11) h (v ) = 1.3.17. P ROPOSITION . Let V be a real vector space and J a complex structure in V . Then, the projections h and a given in (1.3.11) restricted to V dene respectively a C-linear isomorphism of (V, J ) onto V h and a C-anti-linear isomorphism of (V, J ) onto V a . Proposition 1.3.17 tells us that, if we complexify a space V that already possesses a complex structure J , we obtain a complex space V C that contains a copy of the original space (V, J ) (the holomorphic subspace) and a copy of (V, J ) (the anti-holomorphic subspace). Observe also that the holomorphic and the antiholomorphic subspaces of V C are mutually conjugate: Va =c Vh , Vh =c Va ,

where c denotes the conjugation of V C relative to the real form V . In our next example we make a short digression to show how the theory of this subsection appears naturally in the context of calculus with functions of several complex variables. 1.3.18. E XAMPLE . The construction of the holomorphic and the anti-holomorphic subspaces appears naturally when one studies calculus of several complex variables, or, more generally, when one studies the geometry of complex manifolds. In this example we consider the space Cn , that will be thought as the real space 2 n I R endowed with the canonical complex structure. The real canonical basis of Cn (I R2n , J ) will be denoted by: ,..., n, 1,..., n ; x1 x y y this is a basis of I R2n adapted to J , and the corresponding complex basis of Cn is given by: ,..., n . 1 x x We now consider another complex space, given by the complexication (I R2n )C 2 n n C . We denote by J the multiplication by the scalar i in C , while in C2n such multiplication will be denoted in the usual way v i v . Let J C : C2n C2n be

16

1. SYMPLECTIC SPACES

the complexication of J , which denes the holomorphic and the anti-holomorphic subspaces of C2n . By Proposition 1.3.17, the projections h and a dened in (1.3.11) map the canonical complex basis of Cn respectively into a basis of the holomorphic subspace and a basis of the anti-holomorphic subspace of C2n . These bases are usually n n denoted by z ; using (1.3.11) we compute explicitly: j j =1 and z j j =1 1 = j z 2 i j j x y , 1 = j z 2 +i j j x y .

Observe that the vector is conjugate to the vector z j. z j The notation xj , yj for the canonical basis of I R2n is justied by the identication of vectors in I R2n with the partial derivative operators on differentiable functions f : I R 2n I R. The complexication of I R2n is therefore identied with the space of partial derivative operators acting on complex differentiable functions f :I R2n C; in this notation, the CauchyRiemann equations, that characterize the holomorphic functions, are given by setting equal to 0 the derivatives in the directions of the anti-holomorphic subspace: f = 0, j = 1, . . . , n. (1.3.12) z j Observe that f satises (1.3.12) if and only if its differential at each point is a C-linear operator from Cn (I R2n , J ) to C.

We now show that the decomposition into holomorphic and anti-holomorphic subspace determines the complex structure: 1.3.19. P ROPOSITION . Let V be a real vector space and consider a direct sum of the complexication V C = Z1 Z2 , where Z1 and Z2 are mutually conjugate subspaces of V C . Then, there exists a unique complex structure J on V such that Z1 = V h ; moreover, for such J , it is also Z2 = V a . P ROOF. The uniqueness follows from the fact that V h is the graph of J when we use the isomorphism V C V V . For the existence, consider the unique C C-linear operator in V that has Z1 and Z2 as eigenspaces corresponding to the eigenvalues i and i respectively. Clearly, such operator commutes with the conjugation and its square equals Id. From Lemma 1.3.11 it follows that it is of the form J C for some complex structure J : V V . Let now T be a C-linear endomorphism of (V, J ), i.e., an I R-linear endomorphism of V such that T J = J T ; let T C be its complexication. It is easy to see that the the holomorphic and the anti-holomorphic subspaces of V C are invariant by T C ; moreover, we have the following commutative diagrams: V V = =
TC T

(1.3.13)

h |V

h |V

a |V

V V = = a |
TC

Vh Vh

Va Va

1.3. COMPLEXIFICATION AND REAL FORMS

17

It follows from Proposition 1.3.17 that the vertical arrows in the diagram on the left are C-linear isomorphisms of (V, J ) with V h and the vertical arrows in the diagram on the right are C-linear isomorphisms of (V, J ) in V a . n Let now (bj )n j =1 be a complex basis of (V, J ) and let (bj , J (bj ))j =1 be the corresponding real basis of V adapted to J . The latter is also a complex basis for V C (see Example 1.3.7). By Proposition 1.3.17, the vectors uj , u j dened by: (1.3.14) uj = bj iJ (bj ) V h, 2 u j = bj + iJ (bj ) V a, 2 j = 1, . . . , n

form a complex basis of (V, J ). If T is represented by the matrix Z = A + B i, C with A, B real matrices, in the basis (bj )n j =1 of V (hence it is represented by the matrix (1.2.6) with respect to the real basis of V ), then it follows from (1.3.13) that C the matrix representation of T C with respect to the basis (uj , u j )n j =1 of V is given by: (1.3.15) TC Z 0 0 Z .

On the other hand, the matrix representation of T C with respect to the basis (bj , J (bj ))n j =1 is again (1.2.6) (see Example 1.3.12). This shows in particular that the matrices in (1.2.6) and in (1.3.15) are equivalent (or conjugate, i.e., representing the same operator in different bases). We summarize the above observations into the following: 1.3.20. P ROPOSITION . Let V be a real vector space and J a complex structure in V . If T is a C-linear endomorphism of (V, J ), then: the trace of T as an operator on V is twice the real part of the trace of T as an operator on (V, J ); the determinant of T as an operator on V is equal to the square of the absolute value of the determinant of T as an operator on (V, J ). More explicitly, if A, B and real n n matrices, Z = A + B i and C is the matrix given in (1.2.6), then we have the following identities: tr(C ) = 2 tr(Z ) , det(C ) = |det(Z )|2 ,

where tr(U ), det(U ) denote respectively the trace and the determinant of the matrix U , and (), || denote respectively the real part and the absolute value of the complex number . 1.3.21. R EMARK . Suppose that V is endowed with a positive denite inner product g and that J : V V is a complex structure which is g -anti-symmetric, Then, we have J # g = g , i.e., J is g -orthogonal. The operator J C on V C will then be anti-Hermitian (and unitary) with respect to the Hermitian product g Cs in V C (see Remark 1.3.16). It is easy to see that the holomorphic and the antiholomorphic subspaces of J are orthogonal with respect to g Cs : g Cs (v, w) = 0, v V h, w V a.

18

1. SYMPLECTIC SPACES

Using g and J , we can also dene a Hermitian product gs in V by setting: gs (v, w) = g (v, w) + ig (v, Jw), v, w V. Actually, this is the unique Hermitian form in (V, J ) that has g as its real part. We have the following relations: gs (v, w) gs (v, w) , g Cs a (v ), a (w) = , v, w V ; 2 2 they imply, in particular, that if (bj )n orthonormal complex basis of (V, J ) j =1 is an uj , j = 1, . . . , n, (see (1.3.14)) form with respect to gs , then the vectors 2 uj , 2 an orthonormal real basis of V C with respect to g Cs . Also the vectors bj and J (bj ), j = 1, . . . , n, form an orthonormal real basis of V with respect to g , and therefore they form a complex orthonormal basis of V C with respect to g Cs . We conclude then that if Z = A + B i (A, B real matrices), then the matrices in formulas (1.2.6) and (1.3.15) are unitarily equivalent, i.e., they represent the same complex operator in different orthonormal bases. g Cs h (v ), h (w) = 1.4. Symplectic Forms In this section we will study the symplectic vector spaces. We dene the notion of symplectomorphism, which is the equivalence in the category of symplectic vector spaces, and we show that symplectic vector spaces of the same dimension are equivalent. 1.4.1. D EFINITION . Let V be a real vector space; a symplectic form on V is an anti-symmetric nondegenerate bilinear form : V V I R. We say that (V, ) is a symplectic vector space. 1.4.2. R EMARK . If Basym (V ) is a possibly degenerate anti-symmetric bilinear form on V , then denes an anti-symmetric bilinear form on the quotient V /Ker( ); it is easy to see that is nondegenerate, hence (V /Ker( ), ) is a symplectic space. We start by giving a canonical form for the anti-symmetric bilinear forms; the proof is similar to the proof of Theorem 1.1.14. 1.4.3. T HEOREM . Let V be a p-dimensional vector space and Basym (V ) an anti-symmetric bilinear form on V . Then, there exists a basis of V with respect to which the matrix of (as a bilinear form) is given by: 0n In 0nr 0n 0nr , (1.4.1) In 0rn 0rn 0r where r = dim(Ker( )), p = 2n + r, and 0 , 0 and I denote respectively the zero matrix, the zero square matrix and the identity matrix. P ROOF. In rst place, it is clear that, if a basis as in the thesis is found, then the last r vectors of this basis will be a basis for Ker( ), from which we get r = dim(Ker( )) and p = 2n + r.

1.4. SYMPLECTIC FORMS

19

For the proof, we need to exhibit a basis (bi )p i=1 of V such that: (1.4.2) (bi , bn+i ) = (bn+i , bi ) = 1, i = 1, . . . , n, and (bi , bj ) = 0 otherwise. We use induction on p; if p 1 then necessarily = 0 and the result is trivial. Lets assume p > 1 and that the result is true for all vector spaces of dimension less than p. If = 0 the result is trivial; lets assume then that v, w V are chosen in such a way that (v, w) = 0, for instance (v, w) = 1. Then, it is easy to see that is nondegenerate when restricted to the two-dimensional plane generated by v and w; from Proposition 1.1.10 it follows that: V = (I Rv + I Rw) (I Rv + I Rw) . We now use the induction hypothesis to the restriction of to the (p 2)-dimensional vector space (I Rv + I Rw) , and we obtain a basis (b2 , . . . , bn , bn+2 , . . . , bp ) of (I Rv + I Rw) in which takes the canonical form. This means that equality (1.4.2) holds for i = 2, . . . , n, and (bi , bj ) = 0 otherwise. The desired basis for V is then obtained by setting b1 = v and bn+1 = w. 1.4.4. C OROLLARY. If (V, ) is a symplectic space, then V is even dimenn sional, and there exists a basis (bi )2 i=1 of V with respect to which the matrix of as a bilinear form is given by: (1.4.3) 0 I I 0 ,

where 0 and I denote respectively the zero and the identity n n matrices.
n 1.4.5. D EFINITION . We say that (bi )2 i=1 is a symplectic basis of (V, ) if the matrix of as a bilinear form in this basis is given by (1.4.3).

Observe that the matrix of the linear operator : V V is given by the transpose of (1.4.3), i.e., it coincides with the matrix given in (1.2.2). Corollary 1.4.4 tells us that every symplectic space admits a symplectic basis. We now dene sub-objects and morphisms in the category of symplectic spaces. 1.4.6. D EFINITION . Let (V, ) be a symplectic space; We say that S is a symplectic subspace if S V is a subspace and the restriction |S S is nondegenerate. Hence, (S, |S S ) is a symplectic space. Let (V1 , 1 ) and (V2 , 2 ) be symplectic spaces; a linear operator T : V1 V2 is a symplectic map if T # (2 ) = 1 , i.e., if 2 (T (v ), T (w)) = 1 (v, w), v, w V1 .

We say that T is a symplectomorphism if T is an isomorphism and a symplectic map. A symplectomorphism takes symplectic bases in symplectic bases; conversely, if T : V1 V2 is a linear map that takes some symplectic basis of V1 into some symplectic basis of V2 , then T is a symplectomorphism.

20

1. SYMPLECTIC SPACES

In terms of the linear operators 1 Lin(V1 , V1 ) and 2 Lin(V2 , V2 ), a map T Lin(V1 , V2 ) is symplectic if and only if: (1.4.4) T 2 T = 1 . 1.4.7. R EMARK . Observe that the right hand side of equality (1.4.4) is an isomorphism, from which it follows that every symplectomorphism T is an injective map. In particular, the image T (V1 ) is always a symplectic subspace of V2 . 1.4.8. E XAMPLE . We dene a symplectic form in I R2n by setting: (1.4.5) (v1 , w1 ), (v2 , w2 ) = v1 , w2 w1 , v2 , for v1 , v2 , w1 , w2 I Rn , where , denotes the canonical inner product of I Rn . 2 n We say that (1.4.5) is the canonical symplectic form of I R ; the canonical basis 2 n of I R is a symplectic basis for , hence the matrix of (as a bilinear map) with respect to the canonical basis of I R2n is (1.4.3). The existence of a symplectic basis for a symplectic space (Corollary 1.4.4) implies that every symplectic space admits a symplectomorphism with (I R2n , ), hence the proof of every theorem concerning symplectic spaces can be reduced to the case of (I R2n , ). We can also dene a canonical symplectic form in I Rn I Rn by setting: (v1 , 1 ), (v2 , 2 ) = 2 (v1 ) 1 (v2 ), where v1 , v2 and 1 , 2 I Rn . Again, the canonical basis of I Rn I Rn n n is a symplectic basis for the canonical symplectic form of I R I R . I Rn 1.4.9. R EMARK . Denoting by (dq1 , . . . , dqn , dp1 , . . . , dpn ) the canonical ba sis of I R2n (dual of the canonical basis of I R2n ), the canonical symplectic form of 2 n I R is given by:
n

=
i=1

dqi dpi .

It follows easily: n = . . . = (1) n


n(n1) 2

dq1 . . . dqn dp1 . . . dpn .

Hence, is a volume form in for all symplectomorphism T of (I R2n , ) we therefore have: T # ( n ) = n = det(T ) n , from which it follows det(T ) = 1. In general, not every linear map T with det(T ) = 1 is a symplectomorphism of (I R2n , ); when n = 1 the symplectic form is a volume form, hence T is a symplectomorphism if and only if det(T ) = 1. The symplectomorphisms of a symplectic space (V, ) form a group by composition. 1.4.10. D EFINITION . Let (V, ) be a symplectic space; the symplectic group of (V, ) is the group of all symplectomorphisms of (V, ), denoted by Sp(V, ). We denote by Sp(2n, I R) the symplectic group of I R2n endowed with the canonical symplectic form.

I R 2n ;

1.4. SYMPLECTIC FORMS

21

Using a symplectic basis of (V, ), a map T Lin(V ) is a symplectomorphism if and only if the matrix M that represents T in such basis satises: (1.4.6) M M = , A B C D where is the matrix given in (1.4.3). Writing (1.4.7) T ,

then (1.4.6) is equivalent to the following relations: (1.4.8) D A B C = I, A C and B D are symmetric, where A, B, C, D are n n matrices, I is the n n identity matrix, and means transpose (see Exercise 1.10). A matrix of the form (1.4.7) satisfying (1.4.8) will be called a symplectic matrix. We dene direct sum of symplectic spaces. 1.4.11. D EFINITION . Given symplectic spaces (V1 , 1 ) and (V2 , 2 ), we dene a symplectic form = 1 2 on V1 V2 by setting: ((v1 , v2 ), (w1 , w2 )) = 1 (v1 , w1 ) + 2 (v2 , w2 ), v1 , w1 V1 , v2 , w2 V2 . The space (V1 V2 , 1 2 ) is called the direct sum of the symplectic spaces (V1 , 1 ), (V2 , 2 ). If (V, ) is a symplectic space, two subspaces S1 , S2 V are said to be orthogonal if (v1 , v2 ) = 0 for all vi Si , i = 1, 2. If V = S1 S2 with S1 and S2 -orthogonal, then it is easy to see that both S1 and S2 are symplectic subspaces of (V, ); in this case we say that V is the symplectic direct sum of the subspaces S1 and S2 . Observe that the notion of direct sum for symplectic spaces is not meant as a sum in a categorical sense, i.e., it is not true that a symplectic map on a direct sum V1 V2 is determined by its restriction to V1 and V2 (see Exercise 1.15). 1.4.12. E XAMPLE . If Ti : Vi Vi , i = 1, 2, are symplectic maps, then the map T = T1 T2 : V1 V2 V1 V2 dened by: T (v1 , v2 ) = (T1 (v1 ), T2 (v2 )), vi Vi , i = 1, 2, is also symplectic. If both T1 and T2 are symplectomorphisms, then also T is a symplectomorphism. One needs to be careful with the notion of direct sum of symplectic spaces when working with symplectic bases; more explicitly, the concatenation of a sym2m n plectic basis (bi )2 i=1 of V1 and a symplectic basis (bj )j =1 of V2 is not a symplectic basis of V1 V2 . In order to obtain a symplectic basis of V1 V2 we need to rearrange the vectors as follows: (b1 , . . . , bn , b1 , . . . , bm , bn+1 , . . . , b2n , bm+1 , . . . , b2m ). Similar problems are encountered when dealing with symplectic matrices: the simple juxtaposition of along the diagonal of an element of Sp(2n, I R) and an element

22

1. SYMPLECTIC SPACES

of Sp(2m, I R) does not produce an element of Sp(2(n + m), I R); in order to obtain a symplectic matrix it is necessary to perform a suitable permutation of the rows and the columns of such juxtaposition. 1.4.1. Isotropic and Lagrangian subspaces. In this subsection we consider a xed symplectic space (V, ), with dim(V ) = 2n. 1.4.13. D EFINITION . A subspace S V is said to be isotropic if |S S = 0. Observe that S is isotropic if and only if it is contained in its orthogonal S with respect to ; from Proposition 1.1.8 we have: (1.4.9) dim(S ) + dim(S ) = 2n,

from which it follows that the dimension of an isotropic subspace is at most n. Observe that the notion of isotropic subspace is, roughly speaking, opposite to the notion of symplectic subspace; for, by Proposition 1.1.10, S is a symplectic subspace iff S S = {0}. We have the following: 1.4.14. L EMMA . Let L V be a subspace; the following statements are equivalent: L is maximal isotropic, i.e., L is isotropic and it is not properly contained in any other isotropic subspace of V ; L = L ; L is isotropic and dim(L) = n. P ROOF. If L is maximal isotropic, then L L and for v L the subspace L+I R v is isotropic and it contains L. It follows that L = L + I R v , hence v L and L = L . If L = L , then L is isotropic, and from (1.4.9) it follows that dim(L) = n. Finally, if L is isotropic and dim(L) = n, then L is maximal isotropic, because the dimension of an isotropic subspace is at most n. 1.4.15. D EFINITION . A subspace L V is said to be Lagrangian subspace if it satises one (hence all) of the statements in Lemma 1.4.14. 1.4.16. E XAMPLE . The subspaces {0} I Rn and I Rn {0} are Lagrangian 2 n subspaces of I R endowed with the canonical symplectic structure. Given a linear map T Lin(I Rn ), then its graph Graph(T ) = {v + T (v ) : v I Rn } is a 2 n Lagrangian subspace of I R endowed with the canonical symplectic structure if and only if T is symmetric with respect to the canonical inner product of I Rn . 1.4.17. E XAMPLE . If S V is an isotropic subspace, then the kernel of the restriction of to S is the subspace (S ) S = S (see Corollary 1.1.9). It follows that denes by passing to the quotient a symplectic form in S /S (Remark 1.4.2). In the following denition we relate symplectic forms and complex structures on V :

1.4. SYMPLECTIC FORMS

23

1.4.18. D EFINITION . A complex structure J : V V is said to be compatible with the symplectic form if (, J ) is an inner product in V , i.e., a symmetric, positive denite bilinear form on V . More explicitly, J is compatible with if: (Jv, w) = (v, Jw), and if (Jv, v ) > 0 for all v = 0. 1.4.19. E XAMPLE . The canonical complex structure of I R2n (Example 1.2.2) is compatible with the canonical symplectic structure of I R2n . The inner product 2 n (, J ) is simply the canonical inner product of I R . It follows that every symplectic space admits a complex structure compatible with the symplectic form: it is enough to dene J by the matrix (1.2.2) with respect to any xed symplectic basis. Such basis will then be an orthonormal basis with respect to the inner product (, J ). Lets assume that J is a given complex structure on V which is compatible with , and lets denote by g the inner product (, J ); J is a symplectomorphism of (V, ) (see Exercise 1.16) and the following identity holds: g (J , ) = . A compatible complex structure J can be used to construct a Lagrangian which is complementary to a given Lagrangian: 1.4.20. L EMMA . If L V is a Lagrangian subspace and J is a complex structure compatible with , then V = L J (L). P ROOF. It sufces to observe that L and J (L) are orthogonal subspaces with respect to the inner product g . 1.4.21. C OROLLARY. Every Lagrangian subspace admits a complementary Lagrangian subspace. P ROOF. It follows from Lemma 1.4.20, observing that J (L) is Lagrangian, since J is a symplectomorphism (Exercise 1.16). We can dene a complex valued sesquilinear form gs (see Denition 1.3.14) in the complex space (V, J ) by setting: (1.4.10) gs (v, w) = g (v, w) i (v, w). v, w V,

It is easy to see that gs is a positive Hermitian product in (V, J ). Recall from Remark 1.3.16 that a C-linear endomorphism is gs -unitary when gs (T , T ) = gs ; in this situation we also say that T preserves gs . We have the following: 1.4.22. P ROPOSITION . Let T Lin(V ) be an I R-linear map; the following statements are equivalent: T is C-linear in (V, J ) and gs -unitary; T is orthogonal with respect to g and T Sp(V, ).

24

1. SYMPLECTIC SPACES

P ROOF. If T is C-linear and gs -unitary, then T preserves gs , hence it preserves separately its real part, which is g , and its imaginary part, which is . Hence T is an orthogonal symplectomorphism. Conversely, if T is an orthogonal symplectomorphism, then the following identities hold: T g T = g, T T = , = g J, considering g and as linear operators in Lin(V, V ) (see Example 1.1.3)). It follows easily that J T = T J , i.e., T is C-linear. Since T preserves both the real and the imaginary part of gs , we conclude that T is gs -unitary. 1.4.23. E XAMPLE . The canonical complex structure J of I R2n (see Example 1.4.8) is compatible with its canonical symplectic structure (Example 1.2.2), and the inner product g corresponds to the canonical inner product of I R2n . If we 2 n n identify (I R , J ) with C (Example 1.2.2), the Hermitian product gs coincides with the canonical Hermitian product of Cn given in (1.3.10). 1.4.24. R EMARK . Observe that if (V, J ) is a complex space endowed with a Hermitian product gs , then the real part of gs is a positive inner product g on V and the imaginary part of gs is a symplectic form on V ; moreover, dening as minus the imaginary part of gs , it follows that J is compatible with and g = (, J ). 1.4.25. R EMARK . If V is a real vector space, g is a positive inner product on V and J is a complex structure which is g -anti-symmetric (or, equivalently, g -orthogonal), then we get a symplectic form on V by setting = g (J , ). The complex structure J will then be compatible with , and g = (, J ). Again, we also get a Hermitian product gs in (V, J ) dened by (1.4.10). We have the following relation between Lagrangian subspaces and the Hermitian product gs : 1.4.26. L EMMA . A subspace L V is Lagrangian if and only if it is a real form which is preserved by gs , i.e., V = L J (L) and gs (L L) I R. P ROOF. It follows from Lemma 1.4.20 and the observation that the imaginary part of gs equals . As a corollary, we now prove that the group of gs -unitary isomorphisms of (V, J ) act transitively on the set of Lagrangian subspaces of (V, ): 1.4.27. C OROLLARY. Given any pair of Lagrangian subspaces L1 , L2 of V , there exists a C-linear isomorphism T of (V, J ) which is gs -unitary and such that T (L1 ) = L2 . P ROOF. Let (bj )n j =1 be an orthonormal basis of L1 with respect to the inner product g ; since L1 is a real form of (V, J ), it follows that (bj )n j =1 is a complex basis of (V, J ) (see Example 1.3.7). Moreover, since gs is real on L1 , it follows that (bj )n j =1 is an orthonormal basis of (V, J ) with respect to gs . Similarly, we consider a basis (bj )n j =1 of L2 which is orthonormal with respect to g , and we n obtain that (bj )j =1 is a gs -orthonormal basis of (V, J ). It follows that the C-linear

1.4. SYMPLECTIC FORMS

25

isomorphism T dened by T (bj ) = bj , for all j = 1, . . . , n, is unitary and satises T (L1 ) = L2 . It follows that also the symplectic group acts transitively on the set of Lagrangian subspaces: 1.4.28. C OROLLARY. Given any pair L1 , L2 of Lagrangian subspaces of the symplectic space (V, ), there exists a symplectomorphism T Sp(V, ) such that T (L1 ) = L2 . P ROOF. It follows from Corollary 1.4.27, observing that every gs -unitary map is a symplectomorphism (Proposition 1.4.22). 1.4.29. R EMARK . For later use, we will mention a mild renement of the result of Corollary 1.4.27. Given Lagrangian subspaces L1 , L2 V and chosen orientations O1 and O2 respectively on the spaces L1 and L2 , it is possible to nd a C-linear and gs -unitary endomorphism T of (V, J ) such that T (L1 ) = L2 and such that T |L1 : L1 L2 is positively oriented. To see this, it sufces to choose n in the proof of Corollary 1.4.27 the g -orthonormal bases (bj )n j =1 and (bj )j =1 of L1 and L2 respectively in such a way that they are positively oriented. 1.4.30. R EMARK . Given a Lagrangian subspace L0 V , then it is always n possible to nd a basis (bj )2 j =1 of V which is at the same time symplectic, adapted n to J , and such that (bj )j =1 is a basis of L0 . For, if (bj )n j =1 is a g -orthonormal basis of L0 , then the basis dened in (1.2.5) satises the required properties; moreover, such basis is g -orthonormal and the complex basis (bj )n j =1 of (V, J ) is gs orthonormal. We therefore obtain a basis that puts simultaneously all the objects (V, , J, g, gs , L0 ) in their canonical forms. In the spirit of Remark 1.4.24 and Remark 1.4.25, one can ask himself whether given a real space V endowed with a symplectic form and a positive inner product g , it is possible to construct a complex structure J and a Hermitian product gs which are naturally associated to g and . If one requires the condition = g (J , ), then this is clearly impossible in general, because there exists a unique operator H Lin(V ) such that = g (H , ), and such H does not in general satisfy H 2 = Id. We conclude the subsection with a result in this direction: 1.4.31. P ROPOSITION . Let (V, ) be a symplectic space and g a positive inner product in V . Then there exists a unique complex structure J in V which is g -antisymmetric (or, equivalently, g -orthogonal) and compatible with . P ROOF. The uniqueness is the hard part of the thesis, which we now prove. Suppose that J is a given g -anti-symmetric complex structure in V which is compatible with , and let H Lin(V ) be the unique operator such that = g (H , ). Then, H is a g -anti-symmetric isomorphism of V . The compatibility of J with is equivalent to the condition that g (HJ , ) be a symmetric bilinear form on V which is negative denite. By the usual identication

26

1. SYMPLECTIC SPACES

of linear and bilinear maps, we see that the g -anti-symmetry property of H and J , together with the g -symmetry of HJ are expressed by the following relations: g J = J g, g H = H g, g H J = J H g,

from which it follows easily that H J = J H . We now consider the complexications J C , H C Lin(V C ) and the unique sesquilinear extension g Cs of g to V C ; clearly, g Cs is a positive Hermitian product in V C , with respect to which H C and J C are anti-Hermitian operators (see Example 1.3.15 and Remark 1.3.16); moreover, H C J C = J C H C and (J C )2 = Id. Since H C is g Cs -anti-Hermitian, then H C can be diagonalized in a g Cs -orthonormal basis of V C (see Exercise 1.18); its eigenvalues are pure imaginary (non zero, because H C is invertible), and since H C commutes with the conjugation, it follows that eigenspaces of H C corresponding to two conjugate eigenvalues are mutually conjugate (see Lemma 1.3.11). We can then write a g Cs -orthogonal decomposition:
r r

=
j =1

Zi j
j =1

Zi j ,

where j > 0 for all j , Zi the eigenspace of H C corresponding to the eigenvalue i; also, Zi is the conjugate of Zi . Since J C commutes with H C , it follows that the eigenspaces of H C are invariant by J C . The restriction of J C to each Zij is an anti-Hermitian operator whose square is Id, from which it follows that such restriction is diagonalizable, and its possible eigenvalues are i and i. The restriction of g Cs (J C H C , ) to Zij , that coincides with the restriction of ij g Cs (J C , )) must be Hermitian and negative denite, from which it follows that the unique eigenvalue of the restriction of J C to Zij must be equal to i. We conclude that the restriction of J C to Zij is the operator of multiplication by i, and the restriction of J C to Zij is the operator of multiplication by i; such conditions determine J C , which shows the uniqueness of J . For the existence, simply consider the unique complex structure J on V whose holomorphic space coincides with j Zij (see Proposition 1.3.19). 1.4.2. Lagrangian decompositions of a symplectic space. In this subsection we study the properties of Lagrangian decompositions of a symplectic space, that will be fundamental in the study of the Lagrangian Grassmannian in Section 2.5. Throughout this subsection we will x a symplectic space (V, ), with dim(V ) = 2n. We start with a denition: 1.4.32. D EFINITION . A Lagrangian decomposition of (V, ) is a pair (L0 , L1 ) of Lagrangian subspaces of V with V = L0 L1 . 1.4.33. E XAMPLE . The pair (I Rn {0}, {0} I Rn ) is a Lagrangian decom2 n position of I R endowed with the canonical symplectic structure. More generally, if L V is a Lagrangian subspace and J is a complex structure on V compatible

1.4. SYMPLECTIC FORMS

27

with , then (L, J (L)) is a Lagrangian decomposition of (V, ) (see Lemma 1.4.20 and the proof of Corollary 1.4.21). Given a Lagrangian decomposition (L0 , L1 ) of (V, ), we dene a a map: L0 ,L1 : L1 L 0 by setting (1.4.11) L0 ,L1 (v ) = (v, )|L0 for all v L1 ; it is easy to see that L0 ,L1 is an isomorphism (see Exercise 1.19). 1.4.34. R EMARK . The isomorphism L0 ,L1 gives us an identication of L1 with the dual space L 0 , but the reader should be careful when using this identication for the following reason. The isomorphism L0 ,L1 induces an isomorphism (L0 ,L1 ) : L L0 L 0 1 ; however, (L0 ,L1 ) does not coincide with L1 ,L0 , but with its opposite: (1.4.12) (L0 ,L1 ) = L1 ,L0 . L : V /L L , by setting L (v + L) = (v, )|L . Given a Lagrangian decomposition (L0 , L1 ) of (V, ), we have the following commutative diagram of isomorphisms: (1.4.13) L1 E E
q

If L V is a Lagrangian subspace, we also dene an isomorphism:

EEL0 ,L1 EE EE "

z< zz z zz zz L0

L 0

V /L0 where q is the restriction to L1 of the quotient map V V /L0 . An application of the isomorphism L0 ,L1 is given in the following: 1.4.35. L EMMA . If L0 V is a Lagrangian subspace, then every basis (bi )n i=1 n of L0 extends to a symplectic basis (bi )2 i=1 of V ; moreover, given any Lagrangian n L1 which is complementary to L0 , one can choose the basis (bi )2 i=1 in such a way 2 n that (bi )i=n+1 is a basis of L1 . P ROOF. Observe rst that the Lagrangian L0 admits a complementary Lagrangian L1 (see Corollary 1.4.21); given one such Lagrangian L1 , we dene:
1 bn+i = L0 ,L1 (bi ),

i = 1, . . . , n,

n n where (b i )i=1 is the basis of L0 which is dual to (bi )i=1 .

1.4.36. C OROLLARY. Given any Lagrangian decompositions (L0 , L1 ) and (L0 , L1 ) of V then every isomorphism from L0 to L0 extends to a symplectomorphism T : V V such that T (L1 ) = L1 .

28

1. SYMPLECTIC SPACES

n P ROOF. Let (bi )n i=1 be a basis of L0 and let (bi )i=1 be the basis of L0 which n corresponds to (bi )i=1 by the given isomorphism. Using Lemma 1.4.35 we can n 2n 2n nd symplectic bases (bi )2 i=1 and (bi )i=1 of V in such a way that (bi )i=n+1 and 2 n (bi )i=n+1 are bases of L1 and L1 respectively; to conclude the proof one simply chooses T such that T (bi ) = bi , i = 1, . . . , 2n.

1.4.37. C OROLLARY. If L0 V is a Lagrangian subspace, then every isomorphism of L0 extends to a symplectomorphism of V . P ROOF. Choose a Lagrangian L1 complementary to L0 (see Corollary 1.4.21) and apply Corollary 1.4.36. The technique of extending bases of Lagrangians to symplectic bases of the symplectic space may be used to give an alternative proof of Corollary 1.4.28. Roughly speaking, Corollary 1.4.28 tells us that Lagrangian subspaces are indistinguishable from the viewpoint of the symplectic structure; our next Proposition tells us that the only invariant of a pair (L0 , L1 ) of Lagrangian subspaces is the dimension of their intersection L0 L1 : 1.4.38. P ROPOSITION . Given three Lagrangian subspaces L0 , L, L V with dim(L0 L) = dim(L0 L ), there exists a symplectomorphism T of (V, ) such that T (L0 ) = L0 and T (L) = L . P ROOF. By Corollary 1.4.37, there exists a symplectomorphism of (V, ) that takes L0 into itself and L0 L onto L0 L ; we can therefore assume without loss of generality that L0 L = L0 L . Set S = L0 L = L0 L ; clearly S is isotropic and L0 , L, L S . We have a symplectic form in S /S obtained from by passing to the quotient (see Example 1.4.17). Denote by q : S S /S the quotient map; it is easy to see that q (L0 ), q (L) and q (L ) are Lagrangian subspaces of (S /S, ); moreover, q (L0 ), q (L) and q (L0 ), q (L ) are both Lagrangian decompositions of S /S and hence there exists a symplectomorphism T of (S /S, ) such that T q (L0 ) = q (L0 ) and T q (L) = q (L ) (see Corollary 1.4.28). The required symplectomorphism T Sp(V, ) is obtained from the following Lemma. 1.4.39. L EMMA . Let L0 V be a Lagrangian subspace and let S L0 be any subspace. Consider the quotient symplectic form on S /S ; then, given any symplectomorphism T of (S /S, ) with T (q (L0 )) = q (L0 ), there exists a symplectomorphism T of (V, ) such that T (S ) = S (hence also T (S ) = S ), T (L0 ) = L0 , and such that the following diagram commutes S
q T |S

/ S 
q

S /S

/ S /S

where q : S S /S denotes the quotient map.

EXERCISES FOR CHAPTER 1

29

o o o o P ROOF. Write L0 = S R; hence L 0 = S R , where S and R are the annihilators of S and R respectively. Let L1 be any complementary Lagrangian to L0 in V (Corollary 1.4.21). We have: 1 1 o o L1 = L0 ,L1 (S ) L0 ,L1 (R ).

We obtain a direct sum decomposition V = V1 V2 into -orthogonal subspaces given by: 1 1 o o V 1 = S V 2 = R L0 ,L1 (R ), L0 ,L1 (S ), from which it follows that V is direct sum of the symplectic spaces V1 and V2 . Observe that S = V2 S , hence the quotient map q restricts to a symplectomorphism of V2 into S /S ; therefore, we have a unique symplectomorphism T of V2 such that the diagram: V2
q |V2 T

/ V2 
q |V2

S /S

/ S /S

commutes. Since T preserves q (L0 ) it follows that T preserves R; we then dene T by setting T |V1 = Id and T |V2 = T (see Example 1.4.12). 1.4.40. R EMARK . We claim that one can actually choose the symplectomorphism T in the thesis of Proposition 1.4.38 in such a way that T restricts to a positively oriented isomorphism of L0 ; namely, if dim(L0 L) = dim(L0 L ) = 0 then this claim follows directly from Corollary 1.4.36. For the general case, we observe that in the last part of the proof of Lemma 1.4.39 one can dene T |V1 to be any symplectomorphism of V1 which preserves S (while T |V2 = T is kept unchanged); since S is Lagrangian in V1 , using Corollary 1.4.37, we get that T |S can be choosen to be any isomorphism A of S given a priori (and T |R does not depend on A). Since dim(S ) 1, this freedom in the choice of A can be used to adjust the orientation of T |L0 . Exercises for Chapter 1 E XERCISE 1.1. Show that the isomorphism between the spaces Lin(V, W ) and B(V, W ) given in (1.1.1) is natural in the sense that it gives a natural isomorphism of the functors Lin(, ) and B(, ) from the category of pairs of vector spaces to the category of vector spaces. E XERCISE 1.2. Prove that B(V ) = Bsym (V ) Basym (V ). E XERCISE 1.3. Prove Lemma 1.2.3. E XERCISE 1.4. Prove Proposition 1.2.6. E XERCISE 1.5. Prove Proposition 1.3.3. E XERCISE 1.6. Prove Corollary 1.3.4.

30

1. SYMPLECTIC SPACES

E XERCISE 1.7. Prove Lemma 1.3.9. E XERCISE 1.8. Generalize the results of Section 1.3, in particular Proposition 1.3.3, Lemma 1.3.10 and Lemma 1.3.11, to the case of anti-linear, multi-linear and sesquilinear operators. E XERCISE 1.9. Prove that if V is a non trivial complex vector space, then there exists no C-bilinear form on V which is positive denite. E XERCISE 1.10. Prove that T Lin(V ) is a symplectomorphism of (V, ) if and only if its matrix representation with respect to a symplectic basis of (V, ) satises the relations (1.4.8). E XERCISE 1.11. Consider the symplectic space I Rn I Rn endowed with its canonical symplectic structure. Prove that to each Lagrangian subspace L there corresponds a unique pair (P, S ), where P I Rn is a subspace and S : P P I R is a symmetric bilinear form on P , such that: L = (v, ) I Rn I Rn : v P, |P + S (v, ) = 0 . More generally, if (L0 , L1 ) is a Lagrangian decomposition of the symplectic space (V, ), there exists a bijection between the Lagrangian subspaces L V and the pairs (P, S ), where P L1 is any subspace and S Bsym (P ) is a symmetric bilinear form on P , so that (recall formula (1.4.11)): (1.4.14) L = v + w : v P, w L0 , L1 ,L0 (w)|P + S (v, ) = 0 . A B be an element in Sp(2n, I R) (recall forC D mula (1.4.8)) and let L0 = {0} I Rn . Prove that the following two statements are equivalent: (a) T (L0 ) is transverse to L0 ; (b) B is invertible. Prove also that, in this case, the n n matrices DB 1 , B 1 A and C DB 1 A B 1 are symmetric. E XERCISE 1.12. Let T = E XERCISE 1.13. Prove that the transpose of a symplectic matrix in Sp(2n, I R) is again symplectic. E XERCISE 1.14. Every invertible matrix M can be written in polar form: M = P O, P = (M M ) 2 ,
1

O = P 1 M,

where P is symmetric and positive denite and O is orthogonal. Such decomposition is unique and it depends continuously on M . Prove that M Sp(2n, I R) if and only if both P and O are in Sp(2n, I R). E XERCISE 1.15. Prove that the direct sum of symplectic spaces is not categorical, i.e., it is not true in general that if a linear map T : V1 V2 W is such that its restrictions T |V1 and T |V2 are symplectic, then T is symplectic. E XERCISE 1.16. Prove that a complex structure on a symplectic space which is compatible with the symplectic form is a symplectomorphism.

EXERCISES FOR CHAPTER 1

31

E XERCISE 1.17. Let V be a real vector space and g a positive inner product. Prove that a complex structure J in V is g -anti-symmetric iff it is g -orthogonal. E XERCISE 1.18. Let V be a complex space, gs a positive Hermitian product in V and T Lin(V ) a gs -normal operator. Show that T is diagonalizable in a gs -orthonormal basis of V . E XERCISE 1.19. Given a Lagrangian decomposition (L0 , L1 ) of a symplectic space (V, ), prove that the map L0 ,L1 : L1 L 0 dened in (1.4.11) page 27 is an isomorphism. E XERCISE 1.20. Let (V, ) be a symplectic space, S V an isotropic subspace, and consider the quotient symplectic space (S /S, ) dened in Example 1.4.17. Prove that if L V is a Lagrangian subspace of (V, ), then (L) is Lagrangian in (S /S, ), where : S S /S is the projection.

CHAPTER 2

The Geometry of Grassmannians


2.1. Differentiable Manifolds and Lie Groups In this section we give the basic denitions and we x some notations concerning calculus on manifolds. In this text, the term manifold will always mean a real, nite dimensional differentiable manifold whose topology satises the Hausdorff property and the second countability axiom, i.e., it admits a countable basis of open sets. The term differentiable will always mean of class C ; we will describe below the terminology used in the construction of a differentiable manifold structure. Let M be a set; a chart in M is a bijection: : U U, where U M is any subset and U is an open set in some Euclidean space I Rn ; in some situation, with a slight abuse of terminology, we will allow that U be an open subset of some arbitrary real nite dimensional vector space. We say that two charts : U U and : V V in M are compatible if U V = or if (U V ) and (U V ) are both open sets and the transition function: 1 : (U V ) (U V ) is a differentiable diffeomorphism. A differentiable atlas A in M is a set of charts in M that are pairwise compatible and whose domains form a covering of M . A chart is said to be compatible with a differentiable atlas if it is compatible with all the charts of the atlas; it is easy to see that two charts that are compatible with an atlas are compatible with each other. Hence, every differentiable atlas A is contained in a unique maximal differentiable atlas which is obtained as the collection of all the charts in M that are compatible with A. A differentiable atlas A induces on M a unique topology such that each chart of A is a homeomorphism dened in an open subset of (M, ); such topology is dened as the set of parts A M such that (A U ) is an open subset of U for every chart : U U in A. A (differentiable) manifold is then dened as a pair (M, A), where M is a set and A is a maximal differentiable atlas in M whose corresponding topology is Hausdorff and second countable; a chart, or a coordinate system, in a differentiable manifold (M, A) is a chart that belongs to A.
32

2.1. DIFFERENTIABLE MANIFOLDS AND LIE GROUPS

33

2.1.1. R EMARK . Observe that some authors replace the assumption of second countability for a differentiable manifold with the assumption of paracompactness. In Exercise 2.1 the reader is asked to show that such assumption is weaker, but indeed not so much weaker. Let M be a manifold and N M be a subset; we say that a chart : U U I Rn is a submanifold chart for N if (U N ) is equal to the intersection of U with a vector subspace S of I Rn . We then say that: |U N : U N U S is the chart in N induced by . The subset N is said to be an embedded submanifold of M if for all x N there exists a submanifold chart for N whose domain contains x. The inclusion i : N M will then be an embedding of N in M , i.e., a differentiable immersion which is a homeomorphism onto its image endowed with the relative topology. An immersed submanifold N in M is a manifold N such that N is a subset of M and such that the inclusion i : N M is a differentiable immersion. Observe that a subset N M may admit several differentiable structures that make it into an immersed submanifold; however, if we x a topology in N , then there exists at most one differentiable structure in N compatible with such topology and for which N is an immersed submanifold of M (see Exercise 2.3). In general, if N and M are any two manifolds, and if f : N M is an injective differentiable immersion, then there exists a unique differentiable structure on f (N ) that makes f into a differentiable diffeomorphism onto f (N ); hence, f (N ) is an immersed submanifold of M . If f is an embedding, then it follows from the local form of immersions that f (N ) is an embedded submanifold of M . From now on, unless otherwise stated, by submanifold we will always mean embedded submanifold. 2.1.2. R EMARK . If P and M are two manifolds, N M is an embedded submanifold and f : P M is a differentiable map such that f (P ) N , then there exists a unique map f0 : P N such that the following diagram commutes: (2.1.1)
>M O }} } } i }} }} /N P
f f0

where i denotes the inclusion. We say that f0 is obtained from f by change of counterdomain, and we will often use the same symbol f for f0 ; the map f0 is differentiable. The same results does not hold in general if N is only an immersed submanifold; it holds under the assumption of continuity for f0 (see Exercise 2.2). Immersed submanifolds N M for which the differentiability of f in (2.1.1) implies the differentiability of f0 are known as almost embedded submanifolds of M ; examples of such submanifolds are integral submanifolds of involutive distributions, or immersed submanifolds that are subgroups of Lie groups.

34

2. GRASSMANNIANS

2.1.3. R EMARK . If f : M N is a differentiable submersion, then it follows from the local form of the submersions that for all y Im(f ) and for all x f 1 (y ) M there exists a local differentiable section of f that takes y into x, i.e., there exists a differentiable map s : U M dened in an open neighborhood U of y in N such that s(y ) = x and such that f (s(z )) = z for all z U . The existence of local differentiable sections allows to prove that differentiable submersions that are surjective have the quotient property; this means that if f : M N is a surjective submersion and g : M P is a differentiable map, and if there exists a map g : N P such that the following diagram commutes: MA
f

AA g AA AA A  /P N g

then also g is differentiable. In particular, if M is a manifold and f : M N is a surjective map, then there exists at most one differentiable structure on N that makes f into a differentiable submersion; such structure is called a quotient differentiable structure induced by f. 2.1.1. Classical Lie Groups and Lie Algebras. In this subsection we give a short description and we introduce the notations for the classical Lie groups and Lie algebras that will be used in the text. A Lie group is a group G endowed with a differentiable structure such that the map G G (x, y ) xy 1 G is differentiable; the unit of G will be denoted by 1 G. A Lie group homomorphism will always means a group homomorphism which is also continuous; then, it will be automatically differentiable (see for instance [47, Theorem 2.11.2] and [48, Theorem 3.39]). For g G, we denote by lg and rg respectively the diffeomorphisms of G given by the left-translation lg (x) = gx and by the right-translation rg (x) = xg ; 1 we denote the inner automorphism of G associated to g . If g G by Ig = lg rg and v Tx G is a tangent vector to G, we write: gv = dlg (x) v, vg = drg (x) v ;

for all X T1 G we dene vector elds X L and X R in G by setting: (2.1.2) X L (g ) = gX, X R (g ) = Xg,

for all g G. We say that X L (respectively, X R ) is the left-invariant (respectively, the right-invariant ) vector eld in X associated to X T1 G. The Lie algebra corresponding to G, denoted by g, is dened as the tangent space at 1 of the manifold G: g = T1 G; the Lie bracket, or commutator, in g is obtained as the restriction of the Lie brackets of vector elds in G where we identify each X g with the left-invariant vector eld X L .

2.1. DIFFERENTIABLE MANIFOLDS AND LIE GROUPS

35

We denote by exp : g G the exponential map of G, dened in such a way that, for each X g, the map: (2.1.3) I R t exp(tX ) G

is a Lie group homomorphism whose derivative at t = 0 is equal to X . Then, the curve (2.1.3) is an integral curve of the vector elds X L and X R , that is: (2.1.4) d exp(tX ) = X L (exp(tX )) = X R (exp(tX )), dt

for all t I R (see [48, Theorem 3.31]). A Lie subgroup of G is an immersed submanifold which is also a subgroup of G; then, H is also a Lie group with the group and the differentiable structure inherited from those of G (see Remark 2.1.2). A Lie subgroup H G will be an embedded submanifold if and only if H is closed in G (see [47, Theorem 2.5.4] and [48, Theorem 3.21]); moreover, every closed subgroup of a Lie group is a Lie subgroup of G (see [47, Theorem 2.12.6] and [48, Theorem 3.42]). If H G is a Lie subgroup, then the differential of the inclusion map allows to identify the Lie algebra h of H with a Lie subalgebra of g (see [48, Proposition 3.33]); explicitly, we have: (2.1.5) h = X g : exp(tX ) H, t I R .

Observe that every discrete subgroup H G is an embedded (and closed) Lie subgroup of G with dim(H ) = 0; in this case h = {0}. If Go denotes the connected component of G containing the identity (which is also an arc-connected component), then it is easy to see that Go is a normal subgroup of G which is closed and open. Actually, every open subgroup of G is also closed, as its complementary is union of cosets of this subgroup, that are open. It follows that every open subgroup of G is the union of some connected components of G, and the Lie algebra of an open subgroup of G is identied with the Lie algebra of G. 2.1.4. R EMARK . If G is a Lie group and h is a subspace of g, then there exists a unique left-invariant distribution DL and a unique right-invariant distribution DR in G such that DL (1) = DR (1) = h. We have that DL , or DR , is involutive if and only if h is a Lie subalgebra of g. In this case, the maximal connected integral submanifold of DL , or of DR , passing through 1 G is a (connected) Lie subgroup of G whose Lie algebra is h; moreover, if H G is any Lie subgroup whose Lie algebra is h, then H o is the maximal connected integral submanifold of DL , or of DR passing through 1 G. The other maximal connected integral submanifolds of DL (respectively, of DR ) are the left cosets gH (respectively, the right cosets Hg ) of H . A proof of these facts can be found in [47, Theorem 2.5.2] and [48, Corollary (b), Theorem 3.19]; for the basic notions of involutive distributions, integral submanifolds and the Frobenius Theorem the reader may use, for instance, [47, Section 1.3] or [48, pages 4149].

36

2. GRASSMANNIANS

From the above observations we obtain that a curve t (t) G of class C 1 has image contained in some left coset of H if and only if (t)1 (t) h, for all t; similarly, it has image in some right coset of H if and only if: (t) (t)1 h, for all t. We will now present a short list of the classical Lie groups that will be encountered in this text, and we will describe their Lie algebras. All these groups and algebras are formed by real or complex matrices, or by linear operators on real or complex vector spaces. The group multiplication will always be the multiplication of matrices, or the operator composition, and the Lie bracket will always be given by: X, Y = XY Y X ; nally, the exponential map will always be:

exp(X ) =
n=0

Xn . n!

Typically, we will use capital letters to denote Lie groups and the corresponding small letters to denote their Lie algebras; all the vector spaces below will be meant to be nite dimensional. The general linear group. Let V be a real or a complex vector space; we denote by GL(V ) the group of all linear automorphisms of V ; its Lie algebra gl(V ) coincides with the space of all linear endomorphisms Lin(V ) of V . We call GL(V ) the general linear group of V . We write GL(I Rn ) = GL(n, I R), gl(I Rn ) = gl(n, I R), GL(Cn ) = GL(n, C) n and gl(C ) = gl(n, C); obviously, we can identify GL(n, I R) (respectively, GL(n, C)) with the group of invertible real (respectively, complex) n n matrices, and gl(n, I R) (resp., gl(n, C)) with the algebra of all real (resp., complex) n n matrices. Observe that if V is a real space and J is a complex structure on V , so that (V, J ) is identied with a complex space, then GL(V, J ) (resp., gl(V, J )) can be seen as the subgroup (resp., the subalgebra) of GL(V ) (resp., of gl(V )) consisting of those operators that commute with J (see Lemma 1.2.3). In this way we obtain an inclusion of GL(n, C) into GL(2n, I R) and of gl(n, C) into gl(2n, I R) (see Example 1.2.2 and Remark 1.2.9). The special linear group. If V is a real or complex vector space, we denote by SL(V ) the special linear group of V , given by the closed subgroup of GL(V ) consisting of those endomorphisms with determinant equal to 1. Its Lie algebra sl(V ) is given by the set of endomorphisms of V with null trace. We also write SL(I Rn ) = n n n SL(n, I R), SL(C ) = SL(n, C), sl(I R ) = sl(n, I R) and sl(C ) = sl(n, C).

2.1. DIFFERENTIABLE MANIFOLDS AND LIE GROUPS

37

We identify SL(n, I R) (resp., SL(n, C)) with the group of real (resp., complex) n n matrices with determinant equal to 1, and sl(n, I R) (resp., sl(n, C)) with the algebra of real (resp., complex) n n matrices with null trace. As in the case of the general linear group, we have inclusions: SL(n, C) SL(2n, I R) and sl(n, C) sl(2n, I R). The orthogonal and the special orthogonal groups. If V is a real vector space endowed with a positive inner product g , we denote by O(V, g ) the orthogonal group of (V, g ), which is the closed subgroup of GL(V ) consisting of the g -orthogonal operators. The special orthogonal group of (V, g ) is dened by: SO(V, g ) = O(V, g ) SL(V ). The Lie algebras of O(V, g ) and of SO(V, g ) coincide, and they are both denoted by so(V, g ); this is the subalgebra of gl(V ) consisting of g -antisymmetric operators. If V = I Rn and g is the canonical inner product, then we write O(I Rn , g ) = n n R , g ) = SO(n) and so(I R , g ) = so(n); O(n) is identied with O(n), SO(I the group of n n orthogonal matrices (a matrix is orthogonal if its transpose coincides with its inverse), SO(n) is the subgroup of O(n) consisting of those matrices with determinant equal to 1, and so(n) is the Lie algebra of real n n anti-symmetric matrices. The unitary and the special unitary groups. Let V be a complex vector space endowed with a positive Hermitian product gs . The unitary group of (V , gs ), denoted by U(V , gs ), is the closed subgroup of GL(V ) consisting of the gs unitary operators on V ; the special unitary group of (V , gs ) is dened by: SU(V , gs ) = U(V , gs ) SL(V ). The Lie algebra u(V , gs ) of U(V , gs) is the subalgebra of gl(V ) consisting of the gs -anti-Hermitian operators, and the Lie algebra su(V , gs ) of SU(V , gs ) is the subalgebra of u(V , gs ) consisting of operators with null trace. If V is a real space and J is a complex structure in V in such a way that (V, J ) is identied with a complex vector space V , then given a Hermitian product gs in (V, J ) we also write U(V , gs ) = U(V, J, gs ), SU(V , gs ) = SU(V, J, gs ), u(V , gs ) = u(V, J, gs ) and su(V , gs ) = su(V, J, gs ). If V = Cn and gs is the canonical Hermitian product in Cn , then we write U(Cn , gs ) = U(n), SU(Cn , gs ) = SU(n), u(Cn , gs ) = u(n) and su(Cn , gs ) = su(n); then U(n) is the group of complex n n unitary matrices (a matrix is unitary if its conjugate transpose is equal to its inverse), SU(n) is the subgroup of U(n) consisting of matrices with determinant equal to 1, u(n) is the Lie algebra of all complex n n anti-Hermitian matrices (a matrix is anti-Hermitian if its conjugate transpose equals its opposite), and su(n) is the subalgebra of u(n) consisting of matrices with null trace. The symplectic group.

38

2. GRASSMANNIANS

Let (V, ) be a symplectic space; in Denition 1.4.10 we have introduced the symplectic group Sp(V, ). We have that Sp(V, ) is a closed subgroup of GL(V ); its Lie algebra consists of those linear endomorphisms X of V such that (X , ) is a symmetric bilinear form, that is: (2.1.6) (X (v ), w) = (X (w), v ), v, w V. In terms of the linear operator : V V , formula (2.1.6) is equivalent to the identity: (2.1.7) X = X . If is the canonical symplectic form of I R2n , then we write Sp(I R 2n , ) = 2 n Sp(2n, I R) and sp(I R , ) = sp(2n, I R). The matrix representations of elements of Sp(V, ) with respect to a symplectic basis are described in formulas (1.4.7) and (1.4.8). Using (2.1.7) it is easy to see that the matrix representation of elements of sp(V, ) in a symplectic basis is of the form: A B C A , B, C symmetric,

where A denotes the transpose of A. 2.1.2. Actions of Lie Groups and Homogeneous Manifolds. In this subsection we state some results concerning actions of Lie groups on manifolds and we study the homogeneous manifolds, that are manifolds obtained as quotients of Lie groups. If G is a group and M is a set, a (left) action of G on M is a map: (2.1.8) GM (g, m) g m M

such that g1 (g2 m) = (g1 g2 ) m and 1 m = m for all g1 , g2 G and for all m M , where 1 is the unit of G. Given an action of G on M , we get a map (2.1.9) m : G M g : M M of M given by g (m) = g m; the map g g is a group homomorphism from G to the group of bijections of M . For all m M , we dene the orbit of m relative to the action of G by: G(m) = g m : g G ; the orbits of the action of G form a partition of M ; we also dene the isotropy group of the element m M by: Gm = g G : g m = m . It is easy to see that Gm is a subgroup of G. We say that the action of G on M is transitive if G(m) = M for some, hence for all, m M ; we say that the action is free, or without xed points, if Gm = {1}

given by m (g ) = g m, and for all g G we get a bijection:

2.1. DIFFERENTIABLE MANIFOLDS AND LIE GROUPS

39

for all m M . The action is effective if the homomorphism g g is injective, i.e., if mM Gm = {1}. If H is a subgroup of G, we will denote by G/H the set of left cosets of H in G: G/H = gH : g G , where gH = {gh : h H } is the left coset of g G. We have a natural action of G on G/H given by: (2.1.10) G G/H (g1 , g2 H ) (g1 g2 )H G/H ;

this action is called action by left translation of G in the left cosets of H . The action (2.1.10) is always transitive. If G acts on M and Gm is the isotropy group of the element m M , then the map m of (2.1.9) passes to the quotient and denes a bijection: (2.1.11) m : G/Gm G(m) m (gGm ) = g m. We therefore have the following commutative diagiven by gram:
FF m FF FF F#  = /M G/Gm
q m

G FF

where q : G G/Gm denotes the quotient map. 2.1.5. D EFINITION . Given actions of the group G on sets M and N , we say that a map : M N is G-equivariant if the following identity holds: (g m) = g (m), for all g G and all m M . If is an equivariant bijection, we say that is an equivariant isomorphism; in this case 1 is automatically equivariant. The bijection (2.1.11) is an equivariant isomorphism when we consider the action of G on G/Gm by left translation and the action of G on G(m) obtained by the restriction of the action of G on M . 2.1.6. R EMARK . It is possible to dene also a right action of a group G on a set M as a map: (2.1.12) M G (m, g ) m g M

that satises (m g1 ) g2 = m (g1 g2 ) and m 1 = m for all g1 , g2 G and all m M . A theory totally analogous to the theory of left actions can be developed for right actions; as a matter of facts, every right action (2.1.12) denes a left action by (g, m) m g 1 . Observe that in the theory of right actions, in order to dene m in formula (2.1.11), the symbol G/H has to be meant as properly the bijection the set of right cosets of H .

40

2. GRASSMANNIANS

Lets assume now that G is a Lie group and that M is a manifold; in this context we will always assume that the map (2.1.8) is differentiable, and we will say that G acts differentiably on M . If H is a closed subgroup of G, then there exists a unique differentiable structure in the set G/H such that the quotient map: q : G G/H is a differentiable submersion (see Remark 2.1.3). The kernel of the differential dq (1) is precisely the Lie algebra h of H , so that the tangent space to G/H at the point 1H may be identied with the quotient space g/h. Observe that, since q is open and surjective, it follows that G/H has the quotient topology induced by q from the topology of G. By continuity, for all m M , the isotropy group Gm is a closed subgroup of G, hence we get a differentiable structure on G/Gm ; it can be shown that the map gGm g m is a differentiable immersion, from which we obtain the following: 2.1.7. P ROPOSITION . If G is a Lie group that acts differentiably on the manifold M , then for all m M the orbit G(m) has a unique differentiable structure that makes (2.1.11) into a differentiable diffeomorphism; with such structure G(m) is an immersed submanifold of M , and the tangent space Tm G(m) coincides with the image of the map: dm (1) : g Tm M, where m is the map dened in (2.1.9). 2.1.8. R EMARK . If we choose a different point m G(m), so that G(m ) = G(m), then it is easy to see that the differentiable structure induced on G(m) by m coincides with that induced by m . We also have the following: 2.1.9. C OROLLARY. If G acts transitively on M , then for all m M the map (2.1.11) is a differentiable diffeomorphism of G/Gm onto M ; in particular, the map m of (2.1.9) is a surjective submersion. In the case of transitive actions, when we identify G/Gm with M by the diffeomorphism (2.1.11), we will say that m is the base point for such identication; we then say that M (or G/Gm ) is a homogeneous manifold. 2.1.10. C OROLLARY. Let M, N be manifolds and let G be a Lie group that acts differentiably on both M and N . If the action of G on M is transitive, then every equivariant map : M N is differentiable. P ROOF. Choose m M ; the equivariance property of gives us the following commutative diagram: GB
m

BB (m) BB BB B  /N M

and the conclusion follows from Corollary 2.1.9 and Remark 2.1.3.

2.1. DIFFERENTIABLE MANIFOLDS AND LIE GROUPS

41

In some situations we will need to know if a given orbit of the action of a Lie group is an embedded submanifold. Let us give the following denition: 2.1.11. D EFINITION . Let X be a topological space; a subset S X is said to be locally closed if S is given by the intersection of an open and a closed subset of X . Equivalently, S is locally closed when it is open in the relative topology of its closure S . Exercise 2.4 is dedicated to the notion of locally closed subsets. We have the following: 2.1.12. T HEOREM . Let G be a Lie group acting differentiably on the manifold M . Given m M , the orbit G(m) is an embedded submanifold of M if and only if G(m) is locally closed in M . P ROOF. See [47, Theorem 2.9.7]. We conclude the subsection with a result that relates the notions of bration and homogeneous manifold. 2.1.13. D EFINITION . Given manifolds F , E and B and a differentiable map p : E B , we say that p is a differentiable bration with typical ber F if for all b B there exists a diffeomorphism: : p1 (U ) U F such that 1 = p|p1 (U ) , where U B is an open neighborhood of b in B and 1 : U F U is the projection onto the rst factor. In this case, we say that is a local trivialization of p around b. 2.1.14. T HEOREM . Let G be a Lie group and H, K closed subgroups of G with K H ; then the map: p : G/K G/H dened by p(gK ) = gH is a differentiable bration with typical ber H/K . P ROOF. It follows from Remark 2.1.3 that p is differentiable. Given gH G/H , let s : U G be a local section of the submersion q : G G/H dened in an open neighborhood U G/H of gH ; it follows that q s is the inclusion of U in G/H . We dene a local trivialization of p: : p1 (U ) U H/K by setting (xK ) = (xH, s(xH )1 xK ). The conclusion follows. 2.1.15. C OROLLARY. Under the assumptions of Corollary 2.1.9, the map m given in (2.1.9) is a differentiable bration with typical ber Gm . 2.1.16. C OROLLARY. Let f : G G be a Lie group homomorphism and let H G, H G be closed subgroups such that f (H ) H ; consider the map: : G/H G /H f (gH ) = f (g )H for all g G. If induced from f by passage to the quotient, i.e., f f is surjective, then f is a differentiable bration with typical ber f 1 (H )/H .

42

2. GRASSMANNIANS

P ROOF. Consider the action of G on G /H given by G G /H (g, g H ) (f (g )g )H G /H .

, and its isotropy group The orbit of the element 1H G /H is the image of f 1 is surjective, it follows from Corollary 2.1.9 that the map is f (H ); since f 1 f : G/f (H ) G /H induced from f by passage to the quotient is a diffeomorphism. We have the following commutative diagram:
rr rrr r r r y rr
p

G/H

G/f 1 (H )

HH HH f HH HH H# / G /H

where p is induced from the identity of G by passage to the quotient; it follows from Theorem 2.1.14 that p is a differentiable bration with typical ber f 1 (H )/H . This concludes the proof. A differentiable covering is a differentiable bering whose ber is a discrete manifold (i.e., zero dimensional). We have the following: 2.1.17. C OROLLARY. Under the assumptions of Corollary 2.1.16, if H and is a differentiable covering. f 1 (H ) have the same dimension, then f 2.1.18. R EMARK . Given a differentiable bration p : E B with typical ber F , then every curve : [a, b] B of class C k , 0 k +, admits a lift : [a, b] E (i.e., p = ) which is of class C k :

=E
p

[a, b]

/B

The proof of this fact is left to the reader in Exercise 2.9. 2.1.3. Linearization of the Action of a Lie Group on a Manifold. In this subsection we will consider a Lie group G with a differentiable (left) action on the manifold M ; we show that such action denes a anti-homomorphism of the Lie algebra g of G to the Lie algebra of the differentiable vector elds on M . Given X g, we dene a differentiable vector eld X on M by setting: X (m) = dm (1) X, m M,

where m is the map dened in (2.1.9). Recall that if f : N1 N2 is a differentiable map, the vector elds Y1 and Y2 on N1 and N2 respectively are said to be f -related if: Y2 (f (n)) = dfn (Y1 (n)), n N1 .

2.1. DIFFERENTIABLE MANIFOLDS AND LIE GROUPS

43

2.1.19. R EMARK . If Y1 , Z1 are differentiable vector elds on the manifold N1 that are f -related respectively with the elds Y2 , Z2 on the manifold N2 , then the Lie bracket [Y1 , Z1 ] is f -related to the Lie bracket [Y2 , Z2 ]. Observe that, for all g G and all m M , we have gm = m rg , hence (2.1.13) dgm (1) = dm (g ) drg (1). If X R denotes the right invariant vector eld on G corresponding to the element X g, then, using (2.1.13), we have: (2.1.14) X (g m) = dm (g ) X R (g ), m M. The identity (2.1.14) tells us that, for all m M , the eld X in M is m -related with the eld X R in G. 2.1.20. R EMARK . Let us denote by X L the left invariant vector eld on G corresponding to X g; if G acts on the left on M , then in general it is not possible to construct a vector eld in M which is m -related to X L . Observe also that, in general, the eld X is not invariant by the action of G in M ; actually, it is not possible in general to construct a vector eld on M which is invariant by the action of G and whose value at a given point is given. As a corollary of (2.1.14) we get the following: 2.1.21. P ROPOSITION . Given X, Y g, then we have: [X, Y ] = [X , Y ], where the bracket on the left of the equality is the Lie product in g and the bracket on the right denotes the Lie bracket of vector elds in M . P ROOF. Choose m M ; since the vector elds X and Y are m -related respectively to the right invariant vector elds X R and Y R , it follows from Remark 2.1.19 that [X , Y ] is m -related to [X R , Y R ]. To conclude the proof, we will show that: (2.1.15) [X R , Y R ] = [X, Y ]R ; observe now that from (2.1.15) it will follow that both [X , Y ] and [X, Y ] are m -related to [X R , Y R ], hence they must coincide on Im(m ) = G(m). Since m is arbitrary, the proof of Proposition 2.1.21 will follow. In order to show (2.1.15), consider the inversion map inv : G G given by inv(g ) = g 1 ; we have that d(inv)(1) = Id. Then, it is easy to see that X R is inv-related to the left invariant eld X L , and, by Remark 2.1.19, [X R , Y R ] is inv-related to [X L , Y L ] = [X, Y ]L ; also, [X, Y ]R is inv-related to [X, Y ]L . The conclusion now follows from the fact that inv is surjective. The map X X is called the linearization of the action of G in M ; Proposition 2.1.21 tells us that this map is a anti-homomorphism of the Lie algebra g into the Lie algebra of differentiable vector elds on M .

44

2. GRASSMANNIANS

2.1.22. R EMARK . From (2.1.14) it follows easily that, for all m M , the map t exp(tX ) m is an integral curve of X . More generally, given any map I t X (t) g dened in an interval II R, we obtain a time-dependent right invariant vector eld in G given by: (2.1.16) I G (t, g ) X (t)R (g ) = X (t)g Tg G; (t, m) X (t) (m) Tm M.

we also have a time-dependent vector eld in M by setting: (2.1.17) I M From (2.1.14) it follows also that, for any m M , the map m takes integral curves of (2.1.16) into integral curves of (2.1.17); more explicitly, if t (t) G satises (t) = X (t) (t), for all t then: d ( (t) m) = X (t) ( (t) m). dt 2.2. Grassmannians and Their Differentiable Structure In this section we will study the geometry of the set of all k -dimensional subspaces of a Euclidean space. Let n, k be xed integers, with n 0 and 0 k n; we will denote by Gk (n) the set of all k -dimensional vector subspaces of I Rn ; Gk (n) is called the n Grassmannian of k -dimensional subspaces of I R . Our goal is to describe a differentiable atlas for Gk (n), and the main idea is to view the points of Gk (n) as graphs of linear maps dened on a xed k -dimensional subspace of I Rn and taking values in another xed (n k )-dimensional subspace n of I R , where these two xed subspaces are transversal. To this aim, we consider a direct sum decomposition I Rn = W0 W1 , where dim(W0 ) = k (and obviously dim(W1 ) = n k ). For every linear operator T : W0 W1 , the graph of T given by: Gr(T ) = { v + T (v ) : v W0 } is an element in Gk (n). Moreover, an element W Gk (n) is of the form Gr(T ) if and only if it is transversal to W1 , i.e., iff it belongs to the set: G0 k (n, W1 ) = W Gk (n) : W W1 = {0} Gk (n). In this situation, the operator T is uniquely determined by W . We can therefore dene a bijection: (2.2.1) W0 ,W1 : G0 k (n, W1 ) Lin(W0 , W1 ),

by setting W0 ,W1 (W ) = T when W = Gr(T ). More concretely, if 0 and 1 denote respectively the projections onto W0 and W1 in the decomposition I Rn = W0 W1 , then the operator T = W0 ,W1 (W ) is given by: T = (1 |W ) (0 |W )1 .

2.2. GRASSMANNIANS AND THEIR DIFFERENTIABLE STRUCTURE

45

Observe that the condition that W be transversal to W1 is equivalent to the condition that the restriction 0 |W be an isomorphism onto W0 . We will now show that the collection of the charts W0 ,W1 , when (W0 , W1 ) run over the set of all direct sum decomposition of I Rn with dim(W0 ) = k , is a differentiable atlas for Gk (n). To this aim, we need to study the transition functions between these charts. Let us give the following: 2.2.1. D EFINITION . Given subspaces W0 , W0 I Rn and given a common n n complementary subspace W1 I R of theirs, i.e., I R = W0 W1 = W0 W1 , then we have an isomorphism:
W1 = W : W0 W0 , 0 ,W
0

obtained by the restriction to W0 of the projection onto W0 relative to the decomW1 position I Rn = W0 W1 . We say that W is the isomorphism of W0 and W0 0 ,W0 determined by the common complementary subspace W1 .
W1 W1 The inverse of W is simply W ; we have the following commutative 0 ,W0 0 ,W0 diagram of isomorphisms:

q |W0

W0

w; ww w w ww ww

I Rn /W1

W1 W0 ,W0

cGG q| GG W0 GG GG G /W

where q : I Rn I Rn /W1 is the quotient map. Let us consider charts W0 ,W1 and W0 ,W1 in Gk (n), with k = dim(W0 ) = dim(W0 ); observe that they have the same domain. In this case it is easy to obtain the following formula for the transition function: (2.2.2)
W1 W0 ,W1 (W0 ,W1 )1 (T ) = (1 |W0 + T ) W , ,W0
0

where 1 denotes the projection onto W1 relative to the decomposition I Rn = W0 W1 . Let us now consider decompositions I Rn = W0 W1 = W0 W1 , with dim(W0 ) = k , and let us look at the transition function W0 ,W1 (W0 ,W1 )1 . In rst place, we observe that its domain consists of those operators T Lin(W0 , W1 ) such that Gr(T ) G0 k (n, W1 ); it is easy to see that this condition is equivalent to the invertibility of the map: Id + (0 |W1 ) T, where 0 denotes the projection onto W0 relative to the decomposition I Rn = W0 W1 and Id is the identity operator on W0 . We have the following formula for W0 ,W1 (W0 ,W1 )1 : (2.2.3)
W0 W0 ,W1 (W0 ,W1 )1 (T ) = W T Id + (0 |W1 ) T 1 ,W
1

We have therefore proven the following:

46

2. GRASSMANNIANS

2.2.2. P ROPOSITION . The set of all charts W0 ,W1 in Gk (n), where the pair (W0 , W1 ) run over the set of all direct sum decompositions of I Rn with dim(W0 ) = k , is a differentiable atlas for Gk (n). P ROOF. Since every subspace of I Rn admits one complementary subspace, it follows that the domains of the charts W0 ,W1 cover Gk (n). The transition functions (2.2.2) and (2.2.3) are differentiable maps dened in open subsets of the vector space Lin(W0 , W1 ). The general case of compatibility between charts W0 ,W1 and W0 ,W1 follows from transitivity. 2.2.3. R EMARK . As to the argument of transitivity mentioned in the proof of Proposition 2.2.2, we observe that in general the property of the compatibility of charts is not transitive. However, the following weaker transitivity property holds, and that applies to the case of Proposition 2.2.2: if 0 , 1 and 2 are charts on a set such that 0 is compatible with 1 , 1 is compatible with 2 and the domain of 0 coincides with the domain of 1 , then 0 is compatible with 2 . 2.2.4. R EMARK . Formulas (2.2.2) and (2.2.3) show indeed that the charts W0 ,W1 form a real analytic atlas for Gk (n). 2.2.5. R EMARK . Given a nite collection V1 , . . . , Vr of k -dimensional subspaces of I Rn , it is possible to nd a subspace W which is complementary to all of the Vi s. For, if k < n, we can choose a vector v1 I Rn \ r i=1 Vi . Let us now consider the subspaces Vi = Vi I R v1 of dimension k + 1; by repeating the construction to the Vi s, we determine inductively vectors v1 , . . . , vnk that form a basis for a common complementary to the Vi s. This argument shows that every nite subset of Gk (n) belongs to the domain of some chart W0 ,W1 . In Exercise 2.6 the reader is asked to show that the same holds for countable subsets of Gk (n). We nally prove that Gk (n) is a manifold: 2.2.6. T HEOREM . The differentiable atlas in Proposition 2.2.2 makes Gk (n) into a differentiable manifold of dimension k (n k ). P ROOF. If dim(W0 ) = k and dim(W1 ) = n k , then dim(Lin(W0 , W1 )) = k (n k ). It remains to prove that the topology dened by the atlas is Hausdorff and second countable. The Hausdorff property follows from the fact that every pair of points of Gk (n) belongs to the domain of a chart. The second countability property follows from the fact that, if we consider the nite set of chart W0 ,W1 , where both W0 and W1 are generated by elements of the canonical basis of I Rn , we obtain a nite differentiable atlas for Gk (n). 2.2.7. R EMARK . It follows immediately from the denition of topology induced by a differentiable atlas that the subsets G0 k (n, W1 ) Gk (n) are open; moreover, since the charts W0 ,W1 are surjective, it follows that G0 k (n, W1 ) is homeomorphic (and diffeomorphic) to the vector space Lin(W0 , W1 ). 2.2.8. E XAMPLE . The Grassmannian G1 (n) of all the lines through the origin in I Rn is also known as the real projective space I RP n1 . By taking W0 = n 1 n 1 {0} I R and W1 = I R {0}, the chart W0 ,W1 gives us what is usually

2.3. THE TANGENT SPACE TO A GRASSMANNIAN

47

known in projective geometry as the homogeneous coordinates. The space I RP n1 n 1 can also be described as the quotient of the sphere S obtained by identifying the antipodal points. The real projective line I RP 1 is diffeomorphic to the circle S 1 ; in fact, con1 sidering S C, the map z z 2 is a two-fold covering of S 1 over itself that identies antipodal points. 2.2.9. R EMARK . The theory of this section can be repeated verbatim to dene a manifold structure in the Grassmannian of all k -dimensional complex subspaces of Cn . Formulas (2.2.2) and (2.2.3) are holomorphic, which says that such Grassmannian is a complex manifold, whose complex dimension is k (n k ). 2.3. The tangent Space to a Grassmannian In this section we give a concrete description of the tangent space TW Gk (n) for W Gk (n), by showing that it can be naturally identied with the space Lin(W, I Rn /W ). This identication will allow to compute in a simple way the derivative of a curve in Gk (n). We start with an informal approach. Suppose that we are given a differentiable curve t W (t) in Gk (n), i.e., for all instants t we have k -dimensional subspace W (t) of I Rn . How can we think of the derivative W (t0 ) in an intuitive way? Consider a curve of vectors t w(t) I Rn , with v (t) W (t) for all t; in some sense, the derivative v (t0 ) must encode part of the information contained in the derivative W (t0 ). We now try to formalize these ideas. For all t, write W (t) = Ker A(t) , where A(t) Lin(I Rn , I Rnk ); differentiating the identity A(t)w(t) = 0 in t = t0 we get: A (t0 )w(t0 ) + A(t0 )w (t0 ) = 0. This identity shows that the value of w (t0 ) is totally determined by w(t0 ) up to elements of W (t0 ). More precisely, to all w0 W (t0 ), we can associate a class w0 + W (t0 ) I Rn /W (t0 ) by setting w0 = w (t0 ), where t w(t) is any differentiable curve in I Rn with w(t) W (t) for all t and w(0) = w0 . Using the above identity it is easy to see that such map is well dened, i.e., it does not depend on the choice of the curve w(t). The map w0 w0 + W (t0 ) is a linear operator from W (t0 ) to I Rn /W (t0 ), and we can look at it as the derivative of the curve of subspaces W (t) in t = t0 . We can now prove the existence of a canonical isomorphism of the tangent space TW Gk (n) with Lin(W, I Rn /W ); in the following proposition we will use the abstract formalism concerning the functor Lin(, ) introduced in Remark 1.1.1. 2.3.1. P ROPOSITION . Let W Gk (n) and W1 be a complementary subspace of W in I Rn . Denote by q1 : W1 I Rn /W the restriction of the quotient map onto I Rn /W . We have an isomorphism: (2.3.1) where (2.3.2) Lin(Id, q1 ) : Lin(W, W1 ) Lin(W, I Rn /W ) Lin(Id, q1 ) dW,W1 (W ) : TW Gk (n) Lin(W, I Rn /W ),

48

2. GRASSMANNIANS

is the operator of composition on the left T q1 T (recall formulas (1.1.2) and (1.1.3)). The isomorphism (2.3.1) does not depend on the choice of the complementary subspace W1 . P ROOF. Since q1 is an isomorphism and W,W1 is a chart around W , obviously (2.3.1) is an isomorphism. The only non trivial fact in the statement is the independence of (2.3.1) from the choice of the subspace W1 . To prove this fact, consider a different complementary subspace W1 of W in I Rn ; observe that W,W1 (W ) = W,W1 (W ) = 0. By differentiating the transition function (2.2.3) in T = 0 we see that the following diagram commutes:
p dW,W1 (W ) ppp p ppp p p w p

TW Gk (n)

Lin(W, W1 )

Lin Id, W

OOO d W,W1 (W ) OOO OOO OO' / Lin(W, W ). 1

W1 ,W1

The conclusion now follows easily from the observation that also the diagram
W

(2.3.3)

W1 G

W1 ,W1

GG GG G q1 GGG #

I Rn /W

/W 1 w w w ww ww {ww q1

is commutative, where q1 denotes the restriction to W1 of the quotient map onto Rn /W . 2.3.2. R EMARK . Observe that, from a functorial point of view, the conclusion of Proposition 2.3.1 follows by applying the functor Lin(W, ) to the diagram (2.3.3). Keeping in mind Proposition 2.3.1, we will henceforth identify the spaces TW Gk (n) and Lin(W, I Rn /W ). Our next proposition will provide a justication for the informal reasons of such identication given at the beginning of the section: 2.3.3. P ROPOSITION . Let W : I Gk (n) and w : I I Rn be curves dened in an interval I containing t0 , both differentiable at t = t0 . Suppose that w(t) W (t) for all t I . Then, the following identity holds: W (t0 ) w(t0 ) = w (t0 ) + W (t0 ) I Rn /W (t0 ), where we identify W (t0 ) with an element in Lin(W, I Rn /W (t0 )) using the isomorphism (2.3.1). P ROOF. Set W0 = W (t0 ) and choose a complementary subspace W1 of W0 in I Rn . Set T = W0 ,W1 W , so that, for all t I sufciently close to t0 , we have W (t) = Gr(T (t)). Denoting by 0 the projection onto W0 relative to the decomposition I Rn = W0 W1 , we set u = 0 w.

2.3. THE TANGENT SPACE TO A GRASSMANNIAN

49

Since w(t) W (t), we have: (2.3.4) w(t) = u(t) + T (t) u(t), t I.

Using the isomorphism (2.3.1) we see that W (t0 ) TW0 Gk (n) is identied with: Lin(Id, q1 ) dW0 ,W1 (W0 ) W (t0 ) = q1 T (t0 ) Lin(W0 , I Rn /W0 ), where q1 and Lin(Id, q1 ) are dened as in the statement of Proposition 2.3.1. Hence, it remains to show that: q1 T (t0 ) w(t0 ) = w (t0 ) + W0 I Rn /W0 . Differentiating (2.3.4) in t = t0 and observing that T (t0 ) = 0, u(t0 ) = w(t0 ), we obtain: w (t0 ) = u (t0 ) + T (t0 ) w(t0 ), where u (t0 ) W0 . The conclusion follows. 2.3.4. R EMARK . Given a curve W : I Gk (n), t0 I and a vector w0 W0 = W (t0 ), we can always nd a curve t w(t) I Rn dened in a neighborhood of t0 in I , with w(t) W (t) for all t, with w(t0 ) = w0 and such that w has the same regularity as W . Indeed, for t near t0 , we write W in the form W (t) = Gr(T (t)) using a local chart W0 ,W1 ; then we can dene w(t) = w0 + T (t) w0 . This implies that Proposition 2.3.3 can always be used to compute differentials of functions dened on, or taking values in, Grassmannian manifolds. Indeed, the computation of differentials may always be reduced to the computation of tangent vectors to curves, and to this aim we can always use Proposition 2.3.3 (see for instance the proofs of Lemma 2.3.5, Proposition 2.4.11 and Proposition 2.4.12). We now compute the differential of a chart W0 ,W1 at a point W of its domain using the identication TW Gk (n) Lin(W, I Rn /W ): 2.3.5. L EMMA . Consider a direct sum decomposition I Rn = W0 W1 , with 0 dim(W0 ) = k , and let W Gk (n, W1 ); then the differential of the chart W0 ,W1 at W is the operator:
W1 1 Lin W , q1 : Lin(W, I Rn /W ) Lin(W0 , W1 ), 0 ,W

that is:
W1 1 dW0 ,W1 (W ) Z = q1 Z W , 0 ,W

Z Lin(W, I Rn /W ) = TW Gk (n),

W1 where q1 denotes the restriction to W1 of the quotient map onto Rn /W and W 0 ,W is the isomorphism of W0 onto W determined by the common complementary W1 (cf. Denition 2.2.1).

P ROOF. It is a direct application of the technique described in Remark 2.3.4. Let t W(t) be a differentiable curve in Gk (n) with W(0) = W , W (0) = Z ; write T (t) = W0 ,W1 (W(t)), so that W(t) = Gr(T (t)) for all t; observe that T (0) = dW0 ,W1 (W ) Z .

50

2. GRASSMANNIANS

Let w W ; since W = Gr(T (0)), we can write w = w0 + T (0) w0 with w0 W0 . Then, t w(t) = w0 + T (t) w0 is a curve in I Rn with w(t) W(t) for all t and w(0) = w. By Proposition 2.3.3 we have: W (0) w = Z w = w (0) + W = T (0) w0 + W I Rn /W.
W1 (w), we conclude that Observing that w0 = W,W 0 W1 . Z = q1 T (0) W,W 0

The conclusion follows. 2.4. The Grassmannian as a Homogeneous Space In this section we will show that the natural action of the general linear group of I Rn on Gk (n) is differentiable. This action is transitive, even when restricted to the special orthogonal group; it will follow that the Grassmannian is a quotient of this group, and therefore it is a compact and connected manifold. Each linear isomorphism A GL(n, I R) denes a bijection of Gk (n) that associates to each W Gk (n) its image A(W ); with a slight abuse of notation, this bijection will be denoted by the same symbol A. We therefore have a (left) action of GL(n, I R) on Gk (n), that will be called the natural action of GL(n, I R) on Gk (n) . We start by proving the differentiability of this action: 2.4.1. P ROPOSITION . The natural action GL(n, I R) Gk (n) Gk (n) is differentiable. P ROOF. We simply compute the representation of this action in local charts. Let A GL(n, I R) and W0 Gk (n) be xed. Let W1 be a common complementary for W0 and A(W0 ); hence, W0 ,W1 is a chart whose domain contains both W0 and A(W0 ). We compute W0 ,W1 (B (W )) for B in a neighborhood of A and W in a neighborhood of W0 ; writing T = W0 ,W1 (W ) we have: (2.4.1) W0 ,W1 (B (W )) = (B10 + B11 T ) (B00 + B01 T )1 , where Bij denotes the component i (B |Wj ) of B and i , i = 0, 1, denotes the projection onto Wi relative to the decomposition I Rn = W0 W1 . Obviously, (2.4.1) is a differentiable function of the pair (B, T ). The action of GL(n, I R) on Gk (n) is transitive; actually, we have the following stronger result: 2.4.2. P ROPOSITION . The natural action of SO(n) in Gk (n), obtained by restriction of the natural action of GL(n, I R), is transitive. P ROOF. Let W, W Gk (n) be xed; we can nd orthonormal bases (bj )n j =1 n such that (b )k k and (bj )n of I R is a basis of W and ( b ) is a basis of W . j j =1 j =1 j j =1 By possibly replacing b1 with b1 , we can assume that the two bases dene the same orientation of I Rn . We can therefore nd A SO(n) such that A(bj ) = bj for all j = 1, . . . , n, hence in particular A(W ) = W .

2.4. THE GRASSMANNIAN AS A HOMOGENEOUS SPACE

51

2.4.3. C OROLLARY. The Grassmannian Gk (n) is diffeomorphic to the quotients: SO(n) O(n) and ( k ) ( n k ) S ( k ) O(n k ) O O O where S O(k ) O(n k ) denotes the intersection: SO(n) O(k ) O(n k ) . It follows in particular that Gk (n) is a compact and connected manifold. P ROOF. The isotropy of the point I Rk {0}nk by the action of O(n) is given by the group of orthogonal operators that leave the subspaces I Rk {0}nk and k n k {0} I R invariant; this group is clearly isomorphic to O(k ) O(n k ). A similar argument applies to the case of the action of SO(n). The conclusion follows from Corollary 2.1.9 and Proposition 2.4.2. 2.4.4. R EMARK . Obviously, we could have added to the statement of Corollary 2.4.3 a representation of Gk (n) as a quotient of GL(n, I R). Observe that in k n k this case the isotropy of I R {0} is not GL(k ) GL(n k ) (see Exercise 2.7). 2.4.5. R EMARK . As a matter of facts, formula (2.4.1) shows that the natural action of GL(n, I R) on Gk (n) is real analytic. In the case of a complex Grassmannian, the natural action of the linear group GL(n, C) on Cn is holomorphic. An obvious generalization of Proposition 2.4.2 shows that the action of the special unitary group SU(n) on the complex Grassmannian is transitive. Analogously to the result of Corollary 2.4.3, we conclude that the complex Grassmannian is compact, connected and isomorphic to the quotients U(n)/ U(k ) U(n k ) and SU(n)/S U(k ) U(n k )), where S U(k ) U(n k ) denotes the intersection SU(n) U(k ) U(n k ) . We have two more interesting corollaries of the representation of Gk (n) as the quotient of a Lie group. 2.4.6. P ROPOSITION . In an open neighborhood U of any point of Gk (n) we can dene a differentiable map A : U GL(n, I R) such that A(W )(I Rk {0}nk ) = W for all W U . P ROOF. It follows from Propositions 2.4.1, 2.4.2 and from Corollary 2.1.9 that the map: GL(n, I R) B B I Rk {0}nk Gk (n) is a submersion; the required map is simply a local differentiable section of this submersion (see Remark 2.1.3). 2.4.7. C OROLLARY. In an open neighborhood U of any point of Gk (n) there exist differentiable maps: Zker : U Lin(I Rn , I Rnk ) and Zim : U Lin(I Rk , I Rn ) such that W = Ker Zker (W ) = Im Zim (W ) for all W U .

52

2. GRASSMANNIANS

P ROOF. Dene A as in Proposition 2.4.6 and take Zker = A(W )1 and Zim = A(W ) i, where i : I Rk I Rn is the inclusion in the rst k -coordinates and : I Rn I Rnk is the projection onto the last n k coordinates. 2.4.8. C OROLLARY. Let S I Rn be any subspace and let r Z be a non negative integer; then, the set of subspaces W Gk (n) such that dim(W S ) r is open in Gk (n). P ROOF. Let W0 Gk (n) be xed and let Zker be a map as in the statement of Corollary 2.4.7 dened in an open neighborhood U of W0 in Gk (n). For all W U we have: W S = Ker Zker (W )|S , from which we get that dim(W S ) r if and only if the operator Zker (W )|S Lin(S, I Rnk ) has rank greater or equal to dim(S ) r; this condition denes an open subset of Lin(S, I Rnk ), and the conclusion follows. We now consider the action of the product of Lie groups GL(n, I R) GL(m, I R) on the vector space Lin(I Rn , I Rm ) given by: (2.4.2) (A, B, T ) B T A1 ,

for A GL(n, I R), B GL(m, I R) and T Lin(I Rn , I Rm ). An elementary linear algebra argument shows that the orbits of the action (2.3.4) are the sets: Linr (I Rn , I Rm ) = T Lin(I Rn , I Rm ) : T is a matrix of rank r , where r = 1, . . . , min{n, m}. It is also easy to see that the sets: Lini (I Rn , I Rm ) and
ir ir

Lini (I Rn , I Rm )

are respectively an open and a closed subset of Lin(I Rn , I Rm ); it follows that each Linr (I Rn , I Rm ) is locally closed in Lin(I Rn , I Rm ). Thus, we have the following: 2.4.9. L EMMA . For each r = 1, . . . , min{n, m}, the set Linr (I Rn , I Rm ) is an n m embedded submanifold of Lin(I R ,I R ). P ROOF. It follows from Theorem 2.1.12. We also obtain directly the following: 2.4.10. P ROPOSITION . Given non negative integers m, n and r, with r min{n, m}, then the maps: (2.4.3) (2.4.4) are differentiable. Linr (I Rn , I Rm ) Lin (I R ,I R )
r n m

T Im(T ) Gr (m) T Ker(T ) Gnr (n)

2.4. THE GRASSMANNIAN AS A HOMOGENEOUS SPACE

53

P ROOF. The product GL(n, I R) GL(m, I R) acts transitively on the orbit Linr (I Rn , I Rm ), and it also acts transitively on Gr (m), by considering the action for which GL(n, I R) acts trivially and GL(m, I R) acts on Gr (m) with its natural action. The map (2.4.3) is equivariant, hence its differentiability follows from Corollary 2.1.10 and Proposition 2.4.1. The differentiability of (2.4.4) follows similarly. In the next two propositions we compute the differential of the natural action of GL(n, I R) on Gk (n). 2.4.11. P ROPOSITION . For A GL(n, I R), let us consider the diffeomorphism of Gk (n), also denoted by A, given by W A(W ). For W Gk (n), the differential dA(W ) of A at the point W is the operator: : Lin(W, I Lin (A|W )1 , A Rn /W ) Lin(A(W ), I Rn /A(W )) Z (A|W )1 , where given by Z A : I A Rn /W I Rn /A(W ) is induced from A by passing to the quotient. P ROOF. It is a direct application of the technique described in Remark 2.3.4. Let t W (t) a differentiable curve in Gk (n) with W (0) = W and W (0) = Z ; let t w(t) be a differentiable curve in I Rn with w(t) W (t) for all t. It follows that t A(w(t)) is a differentiable curve in I Rn with A(w(t)) A(W (t)) for all t; by Proposition 2.3.3 we have: (2.4.5) (2.4.6) (A W ) (0) A(w(0)) = A(w (0)) + A(W ) I Rn /A(W ). W (0) w(0) = w (0) + W I Rn /W. Using again Proposition 2.3.3, we get: The conclusion follows from (2.4.5) and (2.4.6). 2.4.12. P ROPOSITION . For W Gk (n), the differential of the map: W : GL(n, I R) Gk (n) given by W (A) = A(W ) is: dW (A) X = q X A1 |A(W ) , for all A GL(n, I R), X Lin(I Rn ), where q : I Rn I Rn /A(W ) is the quotient map. P ROOF. We use again the technique described in Remark 2.3.4. Let t A(t) be a differentiable curve in GL(n, I R) with A(0) = A and A (0) = X ; x w0 W . It follows that t A(t)(w0 ) is a differentiable curve in I Rn with A(t)(w0 ) W (A(t)) for all t. Using Proposition 2.3.3 we get: (W A) (0) A(w0 ) = X (w0 ) + A(W ) I Rn /A(W ). The conclusion follows.

54

2. GRASSMANNIANS

2.5. The Lagrangian Grassmannian In this section we will show that the set of all Lagrangian subspaces of a 2n-dimensional symplectic space (V, ) is a submanifold of the Grassmannian of all n-dimensional subspaces of V . We will call the Lagrangian Grassmannian of (V, ). We will study in detail the charts of , its tangent space and the action of the symplectic group Sp(V, ) on ; we will show that, like the total Grassmannian, the Lagrangian Grassmannian is a homogeneous manifold. We will make systematic use of the results concerning the Grassmannian manifolds presented in Sections 2.2, 2.3 and 2.4, as well as the results concerning the symplectic spaces presented in Section 1.4, and especially in Subsection 1.4.2. We start with the observation that the theory of Grassmannians of subspaces of I Rn developed in Sections 2.2, 2.3 and 2.4 can be generalized in an obvious way if we replace I Rn with any other arbitrary nite dimensional real vector space V ; let us briey mention the changes in the notation that will be used in order to consider Grassmannians of subspaces of an arbitrary space V . We will denote by Gk (V ) the set of all k -dimensional subspaces of V , with 0 k dim(V ); this set has a differentiable structure of dimension k (dim(V ) k ), with charts described in Section 2.2. If W1 V is a subspace of codimension k , 0 we will denote by G0 k (V, W1 ) (or more simply by Gk (W1 ) when the space V will be clear from the context) the subset of Gk (V ) consisting of those subspaces that are transversal to W1 :
0 G0 k (V, W1 ) = Gk (W1 ) = W Gk (V ) : V = W W1 . 0 If W0 G0 k (W1 ), then Gk (W1 ) is the domain of the chart W0 ,W1 . For W Gk (V ), we will always consider the following identication of the tangent space TW Gk (V ):

TW Gk (V )

Lin(W, V /W ),

that is constructed precisely as in Section 2.3. In Section 2.4 we must replace the general linear group GL(n, I R) of I Rn by the general linear group GL(V ) of V ; in Proposition 2.4.2 and in Corollary 2.4.3 the orthogonal and the special orthogonal group O(n) and SO(n) of I Rn must be replaced by the corresponding group O(V, g ) and SO(V, g ) associated to an arbitrary choice of an inner product g in V . Let now be xed for the rest of this section a symplectic space (V, ) with dim(V ) = 2n. We denote by (V, ), or more simply by , the set of all Lagrangian subspaces of (V, ): (V, ) = = L Gn (V ) : L is Lagrangian . We say that is the Lagrangian Grassmannian of the symplectic space (V, ). We start with a description of submanifold charts for :

2.5. THE LAGRANGIAN GRASSMANNIAN

55

2.5.1. L EMMA . Let (L0 , L1 ) be a Lagrangian decomposition of V ; then a subspace L G0 n (L1 ) is Lagrangian if and only if the bilinear form: (2.5.1) is symmetric. P ROOF. Since dim(L) = n, then L is Lagrangian if and only if it is isotropic. Let T = L0 ,L1 (L), so that T Lin(L0 , L1 ) and L = Gr(T ); we have: v + T (v ), w + T (w) = (T (v ), w) (T (w), v ). The conclusion follows by observing that the bilinear form (2.5.1) coincides with (T , )|L0 L0 . If L1 V is a Lagrangian subspace, we denote by 0 (L1 ) the set of all Lagrangian subspaces of V that are transversal to L1 : (2.5.2) 0 (L1 ) = G0 n (L1 ). It follows from Lemma 2.5.1 that, associated to each Lagrangian decomposition (L0 , L1 ) of V we have a bijection: (2.5.3) L0 ,L1 : 0 (L1 ) Bsym (L0 ) given by L0 ,L1 (L) = L0 ,L1 L0 ,L1 (L). We therefore have the following: 2.5.2. C OROLLARY. The Grassmannian Lagrangian is an embedded sub1 manifold of Gn (V ) with dimension dim() = 2 n(n + 1); the charts L0 ,L1 dened in (2.5.3) form a differentiable atlas for as (L0 , L1 ) runs over the set of all Lagrangian decompositions of V . P ROOF. Given a Lagrangian decomposition (L0 , L1 ) of V , it follows from Lemma 2.5.1 that the chart: (2.5.4) G0 n (L1 ) W L0 ,L1 L0 ,L1 (W ) Lin(L0 , L 0) B(L0 ) of Gn (V ) is a submanifold chart for , that induces the chart (2.5.3) of . More1 n(n + 1). The conclusion follows from the fact that, over, dim(Bsym (L0 )) = 2 since every Lagrangian admits a complementary Lagrangian (Corollary 1.4.21), the domains of the charts (2.4.5) cover as (L0 , L1 ) runs over the set of all Lagrangian decompositions of V . 2.5.3. R EMARK . It follows from formula (2.5.2) and Remark 2.2.7 that the subset 0 (L1 ) is open in ; moreover, since the chart (2.5.3) is surjective, we have that 0 (L1 ) is homeomorphic (and diffeomorphic) to the Euclidean space Bsym (L0 ). It is sometimes useful to have an explicit formula for the transition functions between the charts (2.5.3) of the Lagrangian Grassmannian; we have the following: 2.5.4. L EMMA . Given Lagrangian decompositions (L0 , L1 ) and (L0 , L1 ) of V then:
L1 (2.5.5) L0 ,L1 (L0 ,L1 )1 (B ) = L0 ,L1 (L0 ) + L ,L0
0

L0 ,L1 L0 ,L1 (L) Lin(L0 , L 0)

B(L0 )

(B ) Bsym (L0 ),

56

2. GRASSMANNIANS

L1 denotes the isomorphism of L0 onto L0 for every B Bsym (L0 ), where L 0 ,L0 determined by the common complementary L1 (recall Denitions 1.1.2 and2.2.1); if (L0 , L1 ) is also a Lagrangian decomposition of V then the following identity holds:

(2.5.6)

1 L0 ,L1 (L0 ,L1 )1 (B ) = B Id + (0 |L1 ) L0 ,L1 B

for all B L0 ,L1 (0 (L1 )) Bsym (L0 ), where 0 denotes the projection onto L0 relative to the decomposition V = L0 L1 . Observe that the following identity holds: (2.5.7)
1 (0 |L1 ) L0 ,L1 = (L0 ,L1 )# L1 ,L0 (L1 ) .

P ROOF. Using (2.2.2) it is easy to see that: (2.5.8)


L1 1 L0 ,L1 (L0 ,L1 )1 (B ) = L0 ,L1 (1 |L0 + L0 ,L1 B ) L ,L0 ,
0

where 1 denotes the projection onto L1 relative to the decomposition V = L0 L1 ; it is also easy to prove that:
L1 1 L0 ,L1 L0 ,L1 = L ,L0
0

: L 0 L0

and substituting in (2.5.8) we obtain (see also (1.1.4)):


L1 L1 (2.5.9) L0 ,L1 (L0 ,L1 )1 (B ) = L0 ,L1 (1 |L0 ) L + L ,L0 ,L0
0 0

(B ).

Setting B = 0 in (2.5.9) we conclude that


L1 L0 ,L1 (L0 ) = L0 ,L1 (1 |L0 ) L , ,L0
0

which completes the proof of (2.5.5). Now, using (2.2.3) it is easy to see that: L0 ,L1 (L0 ,L1 )1 (B ) =
L0 1 1 L0 ,L1 L L0 ,L1 B Id + (0 |L1 ) L0 ,L1 B 1 ,L
1

and it is also easy to prove that:


L0 1 L0 ,L1 L L0 ,L1 = Id : L0 L0 , 1 ,L
1

and this concludes the proof. In our next Lemma we show an interesting formula that involves the charts (2.5.3). 2.5.5. L EMMA . Let L0 , L1 and L be Lagrangian subspaces of V that are pairwise complementary; the following identities hold: (2.5.10) (2.5.11) L0 ,L1 (L) = L0 ,L (L1 ), L0 ,L1 (L) = (L1 ,L0 )# L1 ,L0 (L)1 ;

2.5. THE LAGRANGIAN GRASSMANNIAN

57

P ROOF. Let T = L0 ,L1 (L); then T Lin(L0 , L1 ) and L = Gr(T ). Observe that Ker(T ) = L0 L = {0} and so T is invertible; hence: L1 = {v + (v T (v )) : v L0 } and therefore: L0 ,L (L1 ) : L0 v v T (v ) L. For all v, w L0 , we now compute, : L0 ,L (L1 ) (v, w) = (v T (v ), w) = (T (v ), w) = L0 ,L1 (L) (v, w), which completes the proof of (2.5.10). To show (2.5.11) observe that L1 ,L0 (L) = T 1 ; then: L1 ,L0 (L) = L1 ,L0 T 1 , from which we get: L0 ,L1 (L) = L0 ,L1 L1 ,L0 (L)1 L1 ,L0 . The conclusion follows from (1.4.12) and (1.1.4). We will now study the tangent space TL of the Lagrangian Grassmannian. 2.5.6. P ROPOSITION . Let L be xed; then the isomorphism: (2.5.12) Lin(Id, L ) : Lin(L, V /L) Lin(L, L ) B(L) given by Z L Z takes TL TL Gn (V ) Bsym (L) B(L). Lin(L, V /L) onto the subspace L0 ,L1 (L) = L0 ,L1 T,

P ROOF. Let L1 be a Lagrangian complementary to L. As in the proof of Corollary 2.5.2, the chart: (2.5.13) G0 n (L1 ) W L,L1 L,L1 (W ) B(L) of Gn (V ) is a submanifold chart for that induces the chart L,L1 of ; hence, the differential of (2.5.13) at the point L is an isomorphism that takes TL onto 1 Bsym (L). By Lemma 2.3.5, the differential of L,L1 at the point L is Lin(Id, q1 ), where q1 denotes the restriction to L1 of the quotient map onto V /L; it follows from the diagram (1.4.13) that the differential of (2.5.13) at L coincides with the isomorphism (2.5.12). Using the result of Proposition 2.5.6, we will henceforth identify the tangent space TL with Bsym (L). We will now prove versions of Lemma 2.3.5 and Propositions 2.4.11 and 2.4.12 for the Lagrangian Grassmannian; in these proofs we must keep in mind the isomorphism (2.5.12) that identies TL and Bsym (L). 2.5.7. L EMMA . Consider a Lagrangian decomposition (L0 , L1 ) of V and let L 0 (L1 ) be xed; then, the differential of the chart L0 ,L1 at the point L is the push-forward operator:
L1 L,L 0 #

: Bsym (L) Bsym (L0 ),

L1 where L,L denotes the isomorphism of L onto L0 determined by the common 0 complementary L1 (see Denition 2.2.1).

58

2. GRASSMANNIANS

P ROOF. By differentiating the equality: L0 ,L1 = Lin(Id, L0 ,L1 ) L0 ,L1 at the point L and keeping in mind the identication TL Bsym (L), we obtain:

L1 1 1 dL0 ,L1 (L) = Lin L , L0 ,L1 q1 L |Bsym (L) : Bsym (L) Bsym (L0 ), 0 ,L

where q1 denotes the restriction to L1 of the quotient map onto V /L. On the other hand, it is easy to see that:
L1 1 1 L0 ,L1 q1 L = L0 ,L .

This concludes the proof. Clearly, the natural action of GL(V ) on the Grassmannian Gn (V ) restricts to an action of the symplectic group Sp(V, ) on the Lagrangian Grassmannian ; we have the following: 2.5.8. P ROPOSITION . The natural action of Sp(V, ) on is differentiable. P ROOF. It follows directly from Proposition 2.4.1. Let us now compute the differential of the action of Sp(V, ) on : 2.5.9. P ROPOSITION . For A Sp(V, ), consider the diffeomorphism, also denoted by A, of given by L A(L). For L , the differential dA(L) is the push-forward operator: (A|L )# : Bsym (L) Bsym (A(L)). P ROOF. Using Proposition 2.4.11 and keeping in mind the identications of the tangent spaces TL Bsym (L) and TA(L) Bsym (A(L)), we see that the differential dA(L) is obtained by the restriction to Bsym (L) of the map dened by the following commutative diagram: B(L)
O
Lin(Id,L )

/ B(A(L)) O
Lin Id,A(L)

Lin(L, V /L)
Lin (A|L )1 ,A

/ Lin A(L), V /A(L)

: V /L V /A(L) is induced from A by passing to the quotient, hence: where A 1 . = Lin (A|L )1 , A(L) A L It is easy to see that: 1 = (A|L ) 1 . A(L) A L This concludes the proof.

2.5. THE LAGRANGIAN GRASSMANNIAN

59

2.5.10. P ROPOSITION . For L , the differential of the map: L : Sp(V, ) given by L (A) = A(L) is: dL (A) X = (X A1 , )|A(L)A(L) , for all A Sp(V, ), X TA Sp(V, ) = sp(V, ) A. P ROOF. It follows easily from Proposition 2.4.11, keeping in mind the identication TA(L) Bsym (A(L)) obtained by the restriction of the isomorphism Lin(Id, A(L) ). We will now show that the Lagrangian Grassmannian can be obtained as a quotient of the unitary group. Let J be a complex structure in V which is compatible with the symplectic form ; consider the corresponding inner product g = (, J ) on V and the Hermitian product gs in (V, J ) dened in (1.4.10). Using the notation introduced in Subsection 2.1.1, Proposition 1.4.22 tells us that U(V, J, gs ) = O(V, g ) Sp(V, ). Let us now x a Lagrangian L0 V ; by Lemma 1.4.26, L0 is a real form in (V, J ) where gs is real. It follows that gs is the unique sesquilinear extension of the inner product g |L0 L0 in L0 . Since L0 is a real form in (V, J ), we have that (V, J ) is a complexication of L0 , from which it follows that every I R-linear endomorphism T Lin(L0 ) extends uniquely to a C-linear endomorphism of (V, J ). From Remark 1.3.16 it follows that T Lin(L0 ) is g -orthogonal if and only if T C is gs -unitary; we therefore have an injective homomorphism of Lie groups: (2.5.14) O L0 , g |L0 L0 T T C U(V, J, gs ) whose image consists precisely of the elements in U(V, J, gs ) that leave L0 invariant (see Lemma 1.3.11). Corollary 1.4.27 tells us that the subgroup U(V, J, gs ) of Sp(V, ) acts transitively on ; from Corollary 2.1.9 we therefore obtain the following: 2.5.11. P ROPOSITION . Fix L0 and a complex structure J on V which is compatible with ; the map: U(V, J, gs ) induces a diffeomorphism U(V, J, gs )/O L0 , g |L0 L0 , where O(L0 , g |L0 L0 is identied with a closed subgroup of U(V, J, gs ) through (2.5.14). Obviously, the choice of a symplectic basis in V induces an isomorphism between the Lagrangian Grassmannian of (V, ) and the Lagrangian Grassmannian of I R2n endowed with the canonical symplectic structure. Hence we have the following: A A(L0 )

60

2. GRASSMANNIANS

2.5.12. C OROLLARY. The Lagrangian Grassmannian is isomorphic to the quotient U(n)/O(n); in particular, is a compact and connect manifold. 2.5.1. The submanifolds k (L0 ). In this subsection we will consider a xed symplectic space (V, ), with dim(V ) = 2n, and a Lagrangian subspace L0 V . For k = 0, . . . , n we dene the following subsets of : k (L0 ) = L : dim(L L0 ) = k . Observe that, for k = 0, the above denition is compatible with the denition of 0 (L0 ) given in (2.5.2). Our goal is to show that each k (L0 ) is a submanifold of and to compute its tangent space; we will also show that 1 (L0 ) has codimension 1 in , and that it admits a canonical transverse orientation in . Let us denote by Sp(V, , L0 ) the closed subgroup of Sp(V, ) consisting of those symplectomorphisms that preserve L0 : (2.5.15) Sp(V, , L0 ) = A Sp(V, ) : A(L0 ) = L0 . It is easy to see that the Lie algebra sp(V, , L0 ) of Sp(V, , L0 ) is given by (see formula (2.1.5)): sp(V, , L0 ) = X sp(V, ) : X (L0 ) L0 . In the next Lemma we compute more explicitly this algebra: 2.5.13. L EMMA . The Lie algebra sp(V, , L0 ) consists of those linear endomorphisms X Lin(V ) such that (X , ) is a symmetric bilinear form that vanishes on L0 . P ROOF. It follows from the characterization of the algebra sp(V, ) given in Subsection 2.1.1, observing that (X , )|L0 L0 = 0 if and only if X (L0 ) is contained in the -orthogonal complement L 0 of L0 . But L0 is Lagrangian, hence L0 = L0 . It is clear that the action of Sp(V, ) on leaves each subset k (L0 ) invariant; moreover, by Proposition 1.4.38, it follows that k (L0 ) is an orbit of the action of Sp(V, , L0 ). The strategy then is to use Theorem 2.1.12 to conclude that k (L0 ) is an embedded submanifold of ; to this aim, we need to show that k (L0 ) is locally closed in . For each k = 0, . . . , n we dene:
n k

(L0 ) =
i=k

(L0 ),

(L0 ) =
i=0

i (L0 ).

We have the following: 2.5.14. L EMMA . For all k = 0, . . . , n, the subset k (L0 ) is open and the subset k (L0 ) is closed in . P ROOF. It follows from Corollary 2.4.8 that the set of spaces W Gn (V ) such that dim(W L0 ) k is open in Gn (V ); since has the topology induced by that of Gn (V ), it follows that k (L0 ) is open in . Since k (L0 ) is the complementary of k1 (L0 ), the conclusion follows.

2.5. THE LAGRANGIAN GRASSMANNIAN

61

2.5.15. C OROLLARY. For all k = 0, . . . , n, the subset k (L0 ) is locally closed in . P ROOF. Simply observe that k (L0 ) = k (L0 ) k (L0 ). As a corollary, we obtain the main result of the subsection: 2.5.16. T HEOREM . For each k = 0, . . . , n, k (L0 ) is an embedded submani1 fold of with codimension 2 k (k + 1); its tangent space is given by: (2.5.16) TL k (L0 ) = B Bsym (L) : B |(L0 L)(L0 L) = 0 , for all L k (L0 ). P ROOF. It follows from Proposition 1.4.38 that k (L0 ) is an orbit of the action of Sp(V, , L0 ) on . From Theorem 2.1.12 and Corollary 2.5.15 it follows that k (L0 ) is an embedded submanifold of . It remains to prove the identity in (2.5.16), because then it will follow that (2.5.17) TL = Bsym (L) B B |(L L)(L L) Bsym (L0 L)
0 0

is a surjective linear operator whose kernel is which implies the claim on the codimension of k (L0 ). Using Propositions 2.1.7, 2.5.10 and Lemma 2.5.13, we have that: TL k (L0 ) = B |LL : B Bsym (V ), B |L0 L0 = 0 , for all L k (L0 ). It remains to prove that every symmetric bilinear form B Bsym (L) that vanishes on vectors in L L0 can be extended to a symmetric bilinear form on V that vanishes on L0 . This fact is left to the reader in Exercise 2.8. 2.5.17. R EMARK . One can actually prove that the manifolds k (L0 ) are connected; namely, Remark 1.4.40 implies that the group Sp+ (V, , L0 ) of symplectomorphisms of V which restrict to a positive isomorphism of L0 acts transitively on k (L0 ). The connectedness of k (L0 ) then follows from the conectedness of Sp+ (V, , L0 ) (see Example 3.2.36). 2.5.18. R EMARK . It follows from Theorem 2.5.16 that 0 (L0 ) is a dense open subset of ; indeed, its complement 1 (L0 ) is a nite union of positive codimension submanifolds, all of which have therefore null measure. It follows that given any sequence (Li )iI N of Lagrangian subspaces of V , then the set 0 (Li ) = L : L Li = {0}, i I N
iI N

TL k (L0 ),

is dense in , because its complement is a countable union of sets of null measure. The same conclusion can be obtained by using Baires Lemma instead of the null measure argument. We are now able to dene a transverse orientation for 1 (L0 ) in . Recall that if N is a submanifold of M , then a transverse orientation for N in M is an orientation for the normal bundle i (T M )/T N , where i : N M denotes the inclusion; more explicitly, a transverse orientation for N in M is a choice of an

62

2. GRASSMANNIANS

orientation for the quotient space Tn M/Tn N that depends continuously on n I N. The continuous dependence of the choice of an orientation has to be meant in the following sense: given any n0 N there exists an open neighborhood U N of n0 and there exist continuous functions Xi : U T M , i = 1, . . . , r, such that (Xi (n) + Tn N )r i=1 is a positively oriented basis of Tn M/Tn N for all n U . It follows that, if such continuous maps Xi exist, then we can replace them with differentiable maps Xi that satisfy the same condition. Observe that, for each L k (L0 ), the map (2.5.17) passes to the quotient and denes an isomorphism: (2.5.18) TL /TL k (L0 ) Bsym (L0 L).
=

2.5.19. D EFINITION . For each L 1 (L0 ) we dene an orientation in the quotient TL /TL 1 (L0 ) in the following way: we give an orientation to the unidimensional space Bsym (L0 L) by requiring that an element B Bsym (L0 L) is a positively oriented basis if B (v, v ) > 0 for some (hence for all) v L0 L with v = 0; we consider the unique orientation in TL /TL 1 (L0 ) that makes the isomorphism (2.5.18) positively oriented. 2.5.20. P ROPOSITION . The orientation chosen in Denition 2.5.19 for the space TL /TL 1 (L0 ) makes 1 (L0 ) into a transversally oriented submanifold of ; this transverse orientation is invariant by the action of Sp(V, , L0 ), i.e., for all A Sp(V, , L0 ) and for all L 1 (L0 ) the isomorphism: TL /TL 1 (L0 ) TA(L) /TA(L) 1 (L0 ) induced from dA(L) by passage to the quotient is positively oriented. P ROOF. It follows from Proposition 2.5.9 that the differential dA(L) coincides with the push-forward A# ; hence we have the following commutative diagram: (2.5.19) TL

dA(L)

/ TA(L)  / Bsym (A(L) L0 )

Bsym (L L0 )

(A|LL0 )#

where the vertical arrows are the operators of restriction of bilinear forms. Then, the orientation given in Denition 2.5.19 is Sp(V, , L0 )-invariant. The continuous dependence on L of such orientation now follows from the fact that the action of Sp(V, , L0 ) on 1 (L0 ) is transitive.1
dle over 1 (L0 ) whose ber at the point L 1 (L0 ) is the set consisting of the two possible orientations of TL /TL 1 (L0 ). Under this viewpoint, the Sp(V, , L0 )-invariance of this transverse orientation means that the map O is Sp(V, , L0 )-equivariant, and the differentiability of O follows then from Corollary 2.1.10.
1The required transverse orientation can be seen as a section O of the (Z -principal) ber bun2

EXERCISES FOR CHAPTER 2

63

2.5.21. R EMARK . If A Sp(V, ) is a symplectomorphism with A(L0 ) = L0 , then, as in the proof of Proposition 2.5.20, it follows that the isomorphism: TL /TL 1 (L0 ) TA(L) /TA(L) 1 (L0 ) induced by the differential dA(L) by passage to the quotient is positively oriented for all L 1 (L0 ). To see this, simply replace L0 by L0 in the right column of the diagram (2.5.19).

Exercises for Chapter 2 E XERCISE 2.1. Let X be a locally compact Hausdorff topological space. Show that if X is second countable then X is paracompact; conversely, show that if X is paracompact, connected and locally second countable then X is second countable. E XERCISE 2.2. Suppose that P, M are manifolds, N M is an immersed submanifold and f : P M is a differentiable map. Suppose that f (P ) N ; prove that if f0 : P N is continuous (f0 is dened by the diagram (2.1.1)) when N is endowed with the topology induced by its differentiable atlas, then f0 : P N is differentiable. E XERCISE 2.3. Let M be a manifold, N M a subset and a topology for N . Prove that there exists at most one differentiable structure on N that induces the topology and that makes N an immersed submanifold of M . E XERCISE 2.4. Prove that every locally compact subspace of a Hausdorff space is locally closed and, conversely, that in a locally compact Hausdorff space every locally closed subset is locally compact in the induced topology. E XERCISE 2.5. Let G be a Lie group acting differentiably on the manifold M ; let X g and let X be the vector eld given by (2.1.14). Prove that X is complete in M , i.e., its maximal integral lines are dened over the whole real line. E XERCISE 2.6. Show that, given any countable family {Vi } i=1 of k -dimensional subspaces of I Rn , with k < n, then there exist a (n k )-dimensional subspace W I Rn which is complementary to all the Vi s. E XERCISE 2.7. Determine the isotropy of the element I Rk {0}nk Gk (n) with respect to the natural action of GL(n, I R) on Gk (n). E XERCISE 2.8. Let (V, ) be a (nite dimensional) symplectic space and L, L0 be Lagrangian subspaces of V . Suppose that B Bsym (L) is a symmetric bilinear form on L that vanishes in L L0 . Prove that B extends to a symmetric bilinear form on V that vanishes in V0 . E XERCISE 2.9. Prove that if P : E B is a differentiable bration, then every curve of class C k , : [a, b] B , admits a lift : [a, b] B of class C k , 0 k + (see Remark 2.1.18).

64

2. GRASSMANNIANS

E XERCISE 2.10. Show that the map Lin(I Rn , I Rm ) T Gr(T ) Gn (n + m) is a diffeomorphism onto an open set and compute its differential. E XERCISE 2.11. Prove that a map D : [a, b] Gk (n) is of class C p if and only if there exist maps Y1 , . . . , Yk : [a, b] I Rn of class C p such that (Yi (t))k i=1 is a basis of D(t) for all t. E XERCISE 2.12. The Grassmannian of oriented k -dimensional subspaces of I Rn is the set G+ Rn is a k -dimensional k (n) of all pairs (W, O ) where W I subspace and O is an orientation in W . Dene an action of GL(n, I R) in G+ k (n) and show that its restriction to SO(n) is transitive if k < n. Conclude that, if k < n , G+ k (n) has a natural structure of homogeneous manifold which is compact and connected. E XERCISE 2.13. Given a Lagrangian L0 of a symplectic space (V, ), denote by FixL0 the subgroup of Sp(V, ) consisting of those symplectomorphisms T such that T |L0 = Id, i.e., such that T (v ) = v for all v L0 . Prove that FixL0 is a Lie subgroup of Sp(V, ), and that it acts freely and transitively on 0 (L0 ). Conclude that FixL0 is diffeomorphic to 0 (L0 ). E XERCISE 2.14. In the notations of Exercise 2.13, prove that FixL0 is isomorphic as a Lie group to the additive group of n n real symmetric matrices. E XERCISE 2.15. Given L0 , L with L L0 = {0} and B Bsym (L0 ) a nondegenerate symmetric bilinear form on L0 , prove that there exists L1 with L1 L0 = {0} and such that L0 ,L1 (L) = B .

CHAPTER 3

Topics of Algebraic Topology


3.1. The Fundamental Groupoid and Group In this section we will give a short summary of the denition and of the main properties of the fundamental groupoid and group of a topological space X . We will denote by I the unit closed interval [0, 1] and by C 0 (Y, Z ) the set of continuous maps f : Y Z between any two topological spaces Y and Z . Let us begin with a general denition: 3.1.1. D EFINITION . If Y and Z are topological spaces, we say that two maps f, g C 0 (Y, Z ) are homotopic when there exists a continuous function: H : I Y Z such that H (0, y ) = f (y ) and H (1, y ) = g (y ) for every y Y . We then say that H is a homotopy between f and g and we write H : f = g . For s I , we denote by Hs : Y Z the map Hs (y ) = H (s, y ). Intuitively, a homotopy H : f = g is a one-parameter family (Hs )sI in 0 C (Y, Z ) that deforms continuously H0 = f into H1 = g . In our context, the following notion of homotopy is more interesting: 3.1.2. D EFINITION . Let , : [a, b] X be continuous curves in a topological space X ; we say that is homotopic to with xed endpoints if there exists a homotopy H : = such that H (s, a) = (a) = (a) and H (s, b) = (b) = (b) for every s I . In this case, we say that H is a homotopy with xed endpoints between and . Clearly, two curves , : [a, b] X can only be homotopic with xed endpoints if they have the same endpoints, i.e., if (a) = (a) and (b) = (b); given a homotopy with xed endpoints H the stages Hs are curves with the same endpoints as and . It is easy to see that the homotopy and the homotopy with xed endpoints are equivalence relations in C 0 (Y, Z ) and in C 0 ([a, b], X ) respectively. For this section we will x a topological space X and we will denote by (X ) the set of all continuous curves : I X : (X ) = C 0 (I, X ). For (X ), we denote by [ ] the equivalence class of all curves homotopic to with xed endpoints; we also denote by (X ) the set of such classes: (X ) = [ ] : (X ) .
65

66

3. ALGEBRAIC TOPOLOGY

If , (X ) are such that (1) = (0), we dene the concatenation of and to be the curve in (X ) dened by: ( )(t) = (2t), t [0, 1 2 ], , 1] . (2t 1), t [ 1 2

In this way, the map (, ) denes a partial binary operation in the set (X ). For (X ), we dene 1 (X ) by setting: 1 (t) = (1 t), t I.

For each point x X we denote by ox (X ) the constant curve equal to x: ox (t) = x, [ ] = [1 1 ], t I.


1 1 = 1 .

It is not hard to prove that, if (1) = (0), [ ] = [1 ] and [] = [1 ], then: These identities show that the operations (, ) and 1 pass to the quotient and they dene operations in the set (X ); we then dene: [ ] [] = [ ], [ ]1 = 1 .

The homotopy class [ ] of a curve is invariant by reparameterizations: 3.1.3. L EMMA . Let (X ) be a continuous curve and consider a reparameterization of , where : I I is a continuous map. If (0) = 0 and (1) = 1, then [ ] = [ ]; if (0) = (1), then is homotopic with xed endpoints to a constant curve, i.e., [ ] = [o ((0)) ]. P ROOF. Dene H (s, t) = (1 s)t + s (t) to prove the rst statement and H (s, t) = (1 s) (t) + s (0) to prove the second statement. 3.1.4. R EMARK . In some cases we may need to consider homotopy classes of curves : [a, b] X dened on an arbitrary closed interval [a, b]; in this case we will denote by [ ] the homotopy class with xed endpoints of the curve: (3.1.1) I t (b a)t + a X ;

it follows from Lemma 3.1.3 that (3.1.1) is homotopic with xed endpoints to every reparameterization of , where : I [a, b] is a continuous map with (0) = a and (1) = b. Also the concatenation of curves dened on arbitrary closed intervals should be understood in the sense of the concatenation of their afne reparameterizations on the interval I . 3.1.5. C OROLLARY. Given , , (X ) with (1) = (0) and (1) = (0), then: (3.1.2) [ ] [] [] = [ ] [] [] .

Moreover, for (X ) we have: (3.1.3) [ ] [o (1) ] = [ ], [o (0) ] [ ] = [ ]

3.1. THE FUNDAMENTAL GROUPOID AND GROUP

67

and also: (3.1.4) [ ] [ ]1 = [o (0) ], [ ]1 [ ] = [o (1) ]. P ROOF. The identity (3.1.2) follows from the observation that ( ) is a reparameterization of ( ) by a continuous map : I I with (0) = 0 and (1) = 1. Similarly, the identities in (3.1.3) are obtained by observing that o (1) and o (0) are reparameterizations of by a map with (0) = 0 and (1) = 1. The rst identity in (3.1.4) follows from the fact that 1 = where : I I satises (0) = (1) = 0; the second identity in (3.1.4) is obtained similarly. The identity (3.1.2) tells us that the concatenation is associative in (X ) when all the products involved are dened; the identities in (3.1.3), roughly speaking, say that the classes [ox ], x X , act like neutral elements for the operation of concatenation, and the identities in (3.1.4) tell us that the class [ 1 ] acts like the inverse of the class [ ] with respect to the concatenation. If we x a point x0 X , we denote by x0 (X ) the set of loops in X with basepoint x0 : x0 (X ) = (X ) : (0) = (1) = x0 . We also consider the image of x0 (X ) in the quotient (X ), that will be denoted by: 1 (X, x0 ) = [ ] : x0 (X ) . The (partially dened) binary operation of concatenation in (X ) restricts to a (totally dened) binary operation in 1 (X, x0 ); from Corollary 3.1.5 we obtain the following: 3.1.6. T HEOREM . The set 1 (X, x0 ) endowed with the concatenation operation is a group. This is the main denition of the section: 3.1.7. D EFINITION . The set (X ) endowed with the (partially dened) operation of concatenation is called the fundamental groupoid of the topological space X . For all x0 X , the group 1 (X, x0 ) (with respect to the concatenation operation) is called the fundamental group of X with basepoint x0 . 3.1.8. R EMARK . A groupoid is normally dened as a small category, i.e., a category whose objects form a set, whose morphisms are all isomorphisms. In this context it will not be important to study this abstract notion of groupoid, nevertheless it is important to observe that Corollary 3.1.5 shows that the fundamental groupoid of a topological space is indeed a groupoid in this abstract sense. 3.1.9. R EMARK . If X0 X is the arc-connected component of x0 in X , then 1 (X, x0 ) = 1 (X0 , x0 ), since every loop in X with basepoint in x0 has image contained in X0 , as well as every homotopy between such loops has image in X0 . In the following lemma we describe the functoriality properties of the fundamental groupoid and group:

68

3. ALGEBRAIC TOPOLOGY

3.1.10. L EMMA . Let f : X Y be a continuous map; for (X ), the homotopy class [f ] depends only on the homotopy class [ ] of ; hence, we have a well dened map f : (X ) (Y ) given by f ([ ]) = [f ]. For , (X ) with (1) = (0) and for every x0 X the following identities hold: f ([ ] []) = f ([ ]) f ([]), f [ ]1 = f ([ ])1 , f ([ox0 ]) = [of (x0 ) ]. In particular, if f (x0 ) = y0 then f restricts to a map f : 1 (X, x0 ) 1 (Y, y0 ) which is a group homomorphism. Clearly, given f C 0 (X, Y ) and g C 0 (Y, Z ) then: (g f ) = g f , and that, if Id denotes the identity of X , then Id is the identity of (X ); it follows that, if f : X Y is a homeomorphism, then f is a bijection, and it induces an isomorphism of 1 (X, x0 ) onto 1 (Y, f (x0 )). The map f is said to be induced by f in the fundamental groupoid or in the fundamental group. The following proposition relates the fundamental groups relative to different basepoints: 3.1.11. P ROPOSITION . Given x0 , x1 X and a continuous curve : I X with (0) = x0 and (1) = x1 , we have an isomorphism: # : 1 (X, x0 ) 1 (X, x1 ) dened by # ([ ]) = []1 [ ] [], for every x0 (X ). 3.1.12. C OROLLARY. If x0 and x1 belong to the same arc-connected component of X , then the groups 1 (X, x0 ) and 1 (X, x1 ) are isomorphic. The following commutative diagram relates the homomorphisms f and # : 1 (X, x0 ) 1 (Y, y0 ) (f )
#

1 (X, x1 ) 1 (Y, y1 )
f

where f C 0 (X, Y ), x0 , x1 X , y0 = f (x0 ), y1 = f (x1 ) and (X ) is a curve from x0 to x1 . 3.1.13. R EMARK . In spite of the fact that 1 (X, x0 ) and 1 (X, x1 ) are isomorphic if x0 and x1 are in the same arc-connected component of X , such isomorphism is not canonical; more explicitly, if 0 , 1 (X ) are curves from x0 to x1 , then:
1 (1 ) # (0 )# = I[] , 1 where = 1 0 and I[] denotes the operator of conjugation by the element [] in 1 (X, x0 ). If 1 (X, x0 ) is abelian it follows that (0 )# = (1 )# , and therefore

3.1. THE FUNDAMENTAL GROUPOID AND GROUP

69

the fundamental groups with basepoints in the same arc-connected components can be canonically identied (compare with Remark 3.3.34). 3.1.14. D EFINITION . We say that a topological space X is simply connected if it is arc-connected and if 1 (X, x0 ) is the trivial group {ox0 } for some (hence for all) x0 X . Observe that, if X is simply connected, then [ ] = [] for all continuous curves , : I X such that (0) = (0) and (1) = (1); for, in this case, [ ][]1 = [ox0 ]. 3.1.15. E XAMPLE . A subset X I Rn is said to be star-shaped around the point x0 X if for every x X the segment: [x0 , x] = (1 t)x0 + tx : t I is contained in X ; we say that X is convex if it is star-shaped at each one of its points. If X is star-shaped at x0 , then X is simply connected; indeed, X is clearly arc-connected, and, given a loop x0 (X ), we can dene a homotopy: I I between and ox0 . 3.1.16. R EMARK . Two loops x0 (X ) and x1 (X ) are said to be freely homotopic if there exists a homotopy H : = such that, for every s I , the curve Hs is a loop in X , i.e., H (s, 0) = H (s, 1) for every s. In this situation, if we set (s) = H (s, 0), we have the following identity: (3.1.5) # ([ ]) = []. The identity (3.1.5) follows from the fact that, since the square I I is convex, the homotopy class in (I I ) of the loop that is obtained by considering the boundary of I I run counterclockwise is trivial, hence so is its image by H . Such image is precisely the difference of the terms on the two sides of the equality in (3.1.5). In Exercise 3.3 the reader is asked to show that, conversely, any loop is always freely homotopic to 1 , for any curve with (0) = (0). In particular, if , x0 (X ) are freely homotopic, then the classes [ ] and [] are conjugate in 1 (X, x0 ); it follows that x0 (X ) is such that [ ] = [ox0 ] if and only if is freely homotopic to a constant loop. With this argument we have shown that an arc-connected topological space X is simply connected if and only if every loop in X is freely homotopic to a constant loop. 3.1.17. E XAMPLE . A topological space X is said to be contractible if the identity map of X is homotopic to a constant map, i.e., if there exists a continuous map H : I X X and x0 X such that H (0, x) = x and H (1, x) = x0 for every x X . For instance, if X I Rn is star-shaped at x0 , then X is contractible: the required homotopy H is given by H (s, x) = (1 s)x + s x0 . It is easy to see that every contractible space is arc-connected (see Exercise 3.1). Moreover, if X is contractible then X is simply connected; indeed, if H : Id = x0 is a homotopy and (X ) is a loop, then the map (s, t) H (s, (t)) is a free homotopy between and the constant loop ox0 (see Remark 3.1.16). (s, t) (1 s) (t) + s x0 X

70

3. ALGEBRAIC TOPOLOGY

3.1.1. Stability of the homotopy class of a curve. In this subsection we show that, under reasonable assumptions on the topology of the space X , two continuous curves in X that are sufciently close belong to the same homotopy class. We begin with a denition of proximity for continuous maps: 3.1.18. D EFINITION . Let Y, Z be topological spaces; for K Y compact and U Z open, we dene: V (K ; U ) = f C 0 (Y, Z ) : f (K ) U . The compact-open topology in C 0 (Y, Z ) is the topology generated by the sets V (K ; U ) with K Y compact and U Z open; more explicitly, an open set in the compact-open topology is union of intersections of the form: V (K1 ; U1 ) . . . V (Kn ; Un ) with each Ki Y compact and each Ui Z open, i = 1, . . . , n. 3.1.19. R EMARK . When the topology of the counterdomain Z is metrizable, i.e., it is induced by a metric d, the compact-open topology in C 0 (Y, Z ) is also called the topology of the uniform convergence on compacta; in this case it is not too hard to prove that, for f C 0 (Y, Z ), a fundamental systems of open neighborhood of f is obtained by considering the sets: V (f ; K, ) = g C 0 (Y, Z ) : sup d(f (y ), g (y )) < ,
y K

where K Y is an arbitrary compact set and > 0. In this topology, a sequence (or a net) fn converges to f if and only if fn converges uniformly to f on each compact subset of Y . In the context of differential topology, if Y and Z are manifolds (possibly with boundary), the compact-open topology in C 0 (Y, Z ) is also known as the C 0 topology or as the C 0 -weak Whitney topology. 3.1.20. R EMARK . To each map f : X Y Z which is continuous in the second variable there corresponds a map: : X C 0 (Y, Z ). f An interesting property of the compact-open topology in C 0 (Y, Z ) is that, if Y is is equivalent to the continuity of f |X K for every Hausdorff, the continuity of f o 21, 8, Cap compact K Y (see [22, Proposic a tulo 9]). In particular, if Y are is Hausdorff and locally compact, the continuity of f and the continuity of f equivalent. We will now introduce suitable conditions on the topological space X that will allow to prove the stability of the homotopy class of curves. 3.1.21. D EFINITION . We say that the topological space X is locally arc-connected if every point of X has a fundamental system of open neighborhoods consisting of arc-connected subsets, i.e., if for every x X and every neighborhood V of x in X there exists an open arc-connected subset U X with x U V .

3.1. THE FUNDAMENTAL GROUPOID AND GROUP

71

We say that X is semi-locally simply connected if every x X has a neighborhood V such that every loop on V is contractible in X , i.e., given (X ) with (0) = (1) and Im( ) V , then is homotopic (in X ) with xed endpoints to a constant curve. 3.1.22. E XAMPLE . If every point of X has a simply connected neighborhood, then X is semi-locally simply connected; in particular, every differentiable (or even topological) manifold is locally arc-connected and semi-locally simply connected. This is the main result of the subsection: 3.1.23. T HEOREM . Let X be a locally arc-connected and semi-locally simply connected topological space; given a curve (X ), there exists a neighborhood U of in the space C 0 (I, X ) endowed with the compact-open topology such that for every U , if (0) = (0) and (1) = (1) then [] = [ ]. P ROOF. Write X = A U , where each U X is open and such that every lace in U is contractible in X . Then, the inverse images 1 (U ), A, form an open covering of the compact space I , which has a Lebesgue number > 0, i.e., every subset of I whose diameter is less than is contained in some 1 (U ). Let 0 = t0 < t1 < < tk = 1 be a partition of I with tr+1 tr < and let r A be such that ([tr , tr+1 ]) Ur for every r = 0, . . . , k 1. For each r, the point (tr ) Ur1 Ur has an open arc-connected neighborhood Vr contained in the intersection Ur1 Ur ; dene the neighborhood U of in C 0 (I, X ) by:
k 1 k 1

U=
r=0

V ([tr , tr+1 ]; Ur )
r=1

V ({tr }; Vr ).

Clearly, U . Let now U be such that (0) = (0) and (1) = (1); we need to show that [ ] = []. For each r = 1, . . . , k 1 choose a curve r (Vr ) with r (0) = (tr ) and r (1) = (tr ); set 0 = o (0) and k = o (1) . For r = 0, . . . , k 1, we have (see Remark 3.1.4): (3.1.6) [|[tr ,tr+1 ] ] = [r ]1 [ |[tr ,tr+1 ] ] [r+1 ], because the curve on the right hand side of (3.1.6) concatenated with the inverse of the curve on the left hand side of (3.1.6) is the homotopy class of a loop in Ur , hence trivial in (X ). Moreover, (3.1.7) [] = [|[t0 ,t1 ] ] [|[tk1 ,tk ] ], [ ] = [ |[t0 ,t1 ] ] [ |[tk1 ,tk ] ].

The conclusion now follows from (3.1.7) by concatenating the curves on both sides of the identities (3.1.6) for r = 0, . . . , k 1. 3.1.24. E XAMPLE . Let S n I Rn+1 be the unit n-dimensional sphere. From the proof of Theorem 3.1.23 it follows that every curve : I S n is homotopic with xed endpoints to a curve which is piecewise C 1 . If n 2, such curve cannot

72

3. ALGEBRAIC TOPOLOGY

be surjective onto the sphere, because its image must have null measure in S n . Hence, if n 2 and : I S n is a piecewise C 1 loop, there exists x S n such that Im( ) S n \ {x}. Using the stereographic projection, we see that S n \ {x} is homeomorphic to I Rn , therefore it is simply connected. From this argument it follows that the sphere S n is simply connected for n 2; the circle S 1 is not simply connected (see Example 3.2.24). We will need also a version of Theorem 3.1.23 for the case of homotopies with free endpoints in a given set. 3.1.25. D EFINITION . Let A X be a subset and let , : [a, b] X be given curves with (a), (a), (b), (b) A; we say that and are homotopic with endpoints free in A if there exists a homotopy H : = such that Hs (a), Hs (b) A for every s I ; in this case we say that H is a homotopy with free endpoints in A between and . The relation of homotopy with free endpoints in A is an equivalence relation in the set of curves C 0 ([a, b], X ) such that (a), (b) A; obviously, if two curves with endpoints in A are homotopic with xed endpoints then they will be homotopic with free endpoints in A. 3.1.26. R EMARK . If (X ) is a curve with endpoints in A and (A) is such that (1) = (0), then the concatenation is homotopic to with free endpoints in A. Indeed, for each s I , denote by s (A) the curve s (t) = (1 s)t . Then, Hs = s denes a homotopy with free endpoints in A between and o(0) ; the conclusion follows from the fact that and o(0) are homotopic with xed endpoints. Similarly, one shows that if (A) is such that (1) = (0), then is homotopic to with free endpoints in A. We have the following version of Theorem 3.1.23 for homotopies with free endpoints in a set: 3.1.27. T HEOREM . Let X be a locally arc-connected and semi-locally simply connected topological space; let A X be a locally arc-connected subspace of X . Given a curve : I X with endpoints in A, then there exists a neighborhood U of in C 0 (I, X ) endowed with the compact-open topology such that, for every U with endpoints in A, the curves and are homotopic with free endpoints in A. P ROOF. We will only show how to adapt the proof of Theorem 3.1.23 to this case. Once the open sets Ur and Vr are constructed, we also choose open neighborhood V0 and Vk of (t0 ) and (tk ) respectively in such a way that V0 A and Vk A are arc-connected and contained respectively in U0 and in Uk1 . Then, we dene U by setting:
k 1 k

U=
r=0

V ([tr , tr+1 ]; Ur )
r =0

V ({tr }; Vr ).

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

73

Let U be a curve with endpoints in A; we must show that and are homotopic with free endpoints in A. The curves 0 and k are now chosen in such a way that r (0) = (tr ), r (1) = (tr ) and Im(r ) Vr A for r = 0, k . The identity (3.1.6) still holds for r = 0, . . . , k 1. Using the same argument of that proof, we now obtain: [] = [0 ]1 [ ] [k ]; and the conclusion follows from Remark 3.1.26. 3.2. The Homotopy Exact Sequence of a Fibration In this section we will give a short exposition of the denition and the basic properties of the (absolute and relative) homotopy groups of a topological space; we will describe the exact sequence in homotopy of a pair (X, A), and as a corollary we will obtain the homotopy exact sequence of a bration p : E B . As in Section 3.1, we will denote by I the closed unit interval [0, 1] and by C 0 (Y, Z ) the set of continuous maps from Y to Z . We will denote by I n the unit n-dimensional cube, and by I n its boundary, that is: I n = t I n : ti {0, 1} for some i = 1, . . . , n . If n = 0, we dene I 0 = {0} and I 0 = . Let I R denote the space of all sequences (ti )i1 of real numbers; we identify n I with the subset of I R : In R = {(t1 , . . . , tn , 0, 0, . . .) : 0 ti 1, i = 1, . . . , n} I in such a way that, for n 1, the cube I n1 will be identied with the face of I n : I n1 = {t I n : tn = 0} I n ; we will call this face the initial face of I n . We denote by J n1 the union of the other faces of I n : J n1 = t I n : tn = 1 or ti {0, 1} for some i = 1, . . . , n 1 . We will henceforth x a topological space X ; for every x0 X we denote by n x0 (X ) the set: n x0 (X ) = C 0 (I n , X ) : (I n ) {x0 } . If n = 0, we identify a map : I 0 X with the point (0) X , so that 0 x0 (X ) is identied with the set X (observe that 0 ( X ) does not actually depend on x0 ). x0 1 The set x0 (X ) is the loop space with basepoint x0 introduced in Section 3.1. We say that (X, A) is a pair of topological spaces if X is a topological space and A X is a subspace. If (X, A) is a pair of topological spaces, x0 A and n 1 we denote by n x0 (X, A) the set: n x0 (X, A) = (3.2.1) C 0 (I n , X ) : (I n1 ) A, (J n1 ) {x0 } .
n n x0 (X ) = x0 (X, {x0 }), n Observe that, for n x0 (X, A), we have (I ) A; also:

n 1.

74

3. ALGEBRAIC TOPOLOGY

If n = 1, the cube I n is the interval I , the initial face I n1 is the point {0} and J n1 = {1}; the set 1 x0 (X, A) therefore is simply the set of continuous curves : I X with (0) A and (1) = x0 . 3.2.1. D EFINITION . If X is a topological space, x0 X and n 0, we n say that , n x0 (X ) are homotopic in x0 (X ) if there exists a homotopy n H : = such that Hs x0 (X ) for every s I ; the homotopy in n x0 (X ) is an equivalence relation, and for every n ( X ) we denote by [ ] its equivalence x0 class. The quotient set is denoted by: n (X, x0 ) = [] : n x0 (X ) . We say that [] is the homotopy class dened by in n (X, x0 ). Similarly, if (X, A) is a pair of topological spaces, x0 A and n 1, we say n that , n x0 (X, A) are homotopic in x0 (X, A) when there exists a homotopy n H : = such that Hs x0 (X, A) for every s I ; then we have an equivalence relation in n x0 (X, A) and we also denote the equivalence classes by []. The quotient set is denoted by: n (X, A, x0 ) = [] : n x0 (X, A) . We say that [] is the homotopy class dened by in n (X, A, x0 ). Observe that the set 0 (X, x0 ) does not depend on the point x0 , and it is identied with the set of arc-connected components of X ; for every x X , [x] will denote then the arc-connected component of X that contains x. From (3.2.1) it follows that: (3.2.2) n (X, {x0 }, x0 ) = n (X, x0 ), n 1.
n Given , n x0 (X ) with n 1, or given , x0 (X, A) with n 2, we dene the concatenation of with as the map : I n X given by:

(3.2.3)

( )(t) =

1 (2t1 , t2 , . . . , tn ), t1 [0, 2 ], 1 (2t1 1, t2 , . . . , tn ), t1 [ 2 , 1],

for every t = (t1 , . . . , tn ) I n . Observe that the denition (3.2.3) does not make 1 sense in general for , 0 x0 (X ) or for , x0 (X, A). n The concatenation is a binary operation in x0 (X ) for n 1 and in n x0 (X, A) for n 2; it is easy to see that this binary operation passes to the quotient and it denes operations in the sets n (X, x0 ) and n (X, A, x0 ) of the homotopy classes, given by: [] [ ] = [ ]. We generalize Theorem 3.1.6 as follows: 3.2.2. T HEOREM . For n 1, the set n (X, x0 ) is a group (with respect to the concatenation operation) and for n 2 also the set n (X, A, x0 ) is a group; in both cases, the neutral element is the class ox0 of the constant map ox0 : I n X : (3.2.4) ox0 (t) = x0 , t I n,

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

75

and the inverse of [] is the homotopy class [1 ] of the map 1 : I n X given by: 1 (t) = (1 t1 , t2 , . . . , tn ), t I n . 3.2.3. D EFINITION . A pointed set is a pair (C, c0 ) where C is an arbitrary set and c0 C is an element of C . We say that c0 is the distinguished element of (C, c0 ). A map of pointed sets f : (C, c0 ) (C , c0 ) is an arbitrary map f : C C such that f (c0 ) = c0 ; in this case we dene the kernel of f by: (3.2.5) Ker(f ) = f 1 (c0 ), If Ker(f ) = C we say that f is the null map of (C, c0 ) in (C , c0 ). A pointed set (C, c0 ) with C = {c0 } will be called the null pointed set. Both the null pointed set and the null map of pointed sets will be denoted by 0 when there is no danger of confusion. Given a group G, we will always think of G as the pointed set (G, 1), where 1 is the identity of G; with this convention, the group homomorphisms are maps of pointed sets, and the denition of kernel (3.2.5) coincides with the usual denition of kernel of a homomorphism. 3.2.4. D EFINITION . For n 1, the group n (X, x0 ) is called the n-th (absolute) homotopy group of the space X with basepoint x0 ; for n 2, the group n (X, A, x0 ) is called the n-th relative homotopy group of the pair (X, A) with basepoint x0 A. We call 0 (X, x0 ) and 1 (X, A, x0 ) respectively the zero-th set of homotopy of X with basepoint x0 X and the rst set of homotopy of the pair (X, A) with basepoint x0 A; all the sets and groups of homotopy (absolute or relative) will be seen as pointed sets, being the class [0x0 ] their distinguished element. 3.2.5. R EMARK . Arguing as in Example 3.1.9, one concludes that if X0 is the arc-connected component of X containing x0 , then n (X, x0 ) = n (X0 , x0 ) for every n 1; if x0 A X0 , then also n (X, A, x0 ) = n (X0 , A, x0 ) for every n 1. If x0 A X and if A0 denotes the arc-connected component of A containing x0 , then n (X, A, x0 ) = n (X0 , A0 , x0 ) for every n 2. 3.2.6. E XAMPLE . If X I Rd is star-shaped around the point x0 X , then n (X, x0 ) = 0 for every n 0; for, given n x0 (X ) we dene a homotopy H: = ox0 by setting: H (s, t) = (1 s)(t) + s x0 , s I, t I n . 3.2.7. E XAMPLE . For n 1, if n x0 (X, A) is such that Im() A, then [] = [ox0 ] in n (X, A, x0 ); for, a homotopy H : = ox0 in n x0 (X, A) can be dened by: H (s, t) = t1 , . . . , tn1 , 1 (1 s)(1 tn ) , In particular, we have n (X, X, x0 ) = 0. t I n , s I.

76

3. ALGEBRAIC TOPOLOGY

3.2.8. D EFINITION . Let X, Y be topological spaces and let x0 X , y0 Y be given. If f : X Y is a continuous map such that f (x0 ) = y0 , we say that f preserves basepoints, and we write f : (X, x0 ) (Y, y0 ). Then, for n 0, f induces a map of pointed sets: (3.2.6) f : n (X, x0 ) n (Y, y0 ) dened by f ([]) = [f ]. Given pairs (X, A) and (Y, B ) of topological spaces, then a map of pairs f : (X, A) (Y, B ) is a continuous map f : X Y such that f (A) B . If a choice of basepoints x0 A and y0 B is done, we say that f preserves basepoints if f (x0 ) = y0 , in which case we write: f : (X, A, x0 ) (Y, B, y0 ). For n 1, such a map induces a map f of pointed sets: (3.2.7) f : n (X, A, x0 ) n (Y, B, y0 ) dened by f ([]) = [f ]. It is easy to see that the maps f are well dened, i.e., they do not depend on the choice of representatives in the homotopy classes. Given maps: f : (X, A, x0 ) (Y, B, y0 ), g : (Y, B, y0 ) (Z, C, z0 ) then (g f ) = g f ; if Id denotes the identity of (X, A, x0 ), then Id is the identity of n (X, A, x0 ). It follows that if f : (X, A, x0 ) (Y, B, y0 ) is a homeomorphism of triples, i.e., f : X Y is a homeomorphism, f (A) = B and f (x0 ) = y0 , then f is a bijection. Similar observations can be made for the absolute homotopy groups n (X, x0 ). We also have the following: 3.2.9. P ROPOSITION . Given f : (X, x0 ) (Y, y0 ), then, for n 1, the map f given in (3.2.6) is a group homomorphism; moreover, if f : (X, A, x0 ) (Y, B, y0 ), then for n 2 the map f given in (3.2.7) is a group homomorphism. 3.2.10. E XAMPLE . If X = X1 X2 , and pr1 : X X1 , pr2 : X X2 denote the projections, then a continuous map : I n X is completely determined by its coordinates: pr1 = 1 : I n X 1 , pr2 = 2 : I n X 2 , from which it is easy to see that, given x = (x1 , x2 ) X and n 0, we have a bijection: n (X, x) n (X1 , x1 ) n (X2 , x2 )
=

(pr1 ) ,(pr2 )

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

77

which is also a group homomorphism if n 1. More generally, given A1 X1 , A2 X2 , x A = A1 A2 , then for n 1 we have a bijection: n (X, A, x) n (X1 , A1 , x1 ) n (X2 , A2 , x2 )
=

(pr1 ) ,(pr2 )

which is also a group homomorphism if n 2. Similar observations can be made for products of an arbitrary number (possibly innite) of topological spaces. Give a pair (X, A) and x0 A, we have the following maps: i : (A, x0 ) (X, x0 ), q : (X, {x0 }, x0 ) (X, A, x0 ),

induced respectively by the inclusion of A into X and by the identity of X . Keeping in mind (3.2.2) and Denition 3.2.8, we therefore obtain maps of pointed sets: (3.2.8) i : n (A, x0 ) n (X, x0 ), q : n (X, x0 ) n (X, A, x0 );

explicitly, we have i ([]) = [] and q ([]) = []. For n 1 we dene the connection operator relative to the triple (X, A, x0 ): (3.2.9) : n (X, A, x0 ) n1 (A, x0 )

by setting ([]) = [|I n1 ]; it is easy to see that is well dened, i.e., it does not depend on the choice of a representative of the homotopy class. Moreover, is always a map of pointed sets, and it is a group homomorphism if n 2. 3.2.11. D EFINITION . A sequence of pointed sets and maps of pointed sets of the form: (Ci1 , ci1 ) (Ci+1 , ci+1 ) (Ci , ci ) is said to be exact at (Ci , ci ) if Ker(fi ) = Im(fi+1 ); the sequence is said to be exact if it is exact at each (Ci , ci ) for every i. We can now prove one of the main results of this section: 3.2.12. T HEOREM . If (X, A) is a pair of topological spaces and x0 A, then the sequence: (3.2.10)
n (A, x0 ) n (X, x0 ) n (X, A, x0 ) n1 (A, x0 ) 1 (X, A, x0 ) 0 (A, x0 ) 0 (X, x0 )

fi+2

fi+1

fi

fi1

is exact, where for each n the pointed set maps i , q and are given in formulas (3.2.8) and (3.2.9) P ROOF. The proof is done by considering several cases in which the homotopies are explicitly exhibited.

78

3. ALGEBRAIC TOPOLOGY

Exactness at n (X, x0 ). The fact that Im(i ) Ker(q ) follows from Exam ple 3.2.7. Let n x0 (X ) be such that there exists a homotopy H : = ox0 n n in x0 (X, A). Dene K : I I X by setting: Ks (t) = K (s, t) = H2tn (t1 , . . . , tn1 , 0),
tn s Hs t1 , . . . , tn1 , 22 s

0 2tn s, , s 2tn 2;

It is easy to see that = K1 n x0 (A) and that K : = is a homotopy in n x0 (X ). It follows [] = i ([ ]). Exactness at n (X, A, x0 ). n1 (X, A) be The inclusion Im(q ) Ker( ) is trivial. Let x 0 n such that there exists a homotopy H : |I n1 = ox0 in x0 (A). Dene K : I I n X by the following formula: Ks (t) = K (s, t) = Hs2tn (t1 , . . . , tn1 ),
tn s t1 , . . . , tn1 , 22 s

0 2tn s, , s 2tn 2;

It is easy to see that = K1 n x0 (X ) and that K : = is a homotopy in n x0 (X, A). It follows that [] = q ([ ]). Exactness at n (A, x0 ). +1 We rst show that Im( ) Ker(i ). To this aim, let n x0 (X, A). n Dene H : I I X by setting: Hs (t) = H (s, t) = (t, s), s I, t I n ; It is easy to see that H : |I n = ox is a homotopy in n (X ), so that
0

x0

(i )([]) = [ox0 ]. Let now n x0 (A) be such that there exists a homotopy K : = ox0 n in x0 (X ). Then, dene: (t) = Ktn+1 (t1 , . . . , tn ), t I n+1 ;
+1 it follows that n x0 (X, A) and ([]) = [ ].

This concludes the proof. The exact sequence (3.2.10) is known as the the long exact homotopy sequence of the pair (X, A) relative to the basepoint x0 . The exactness property of (3.2.10) at 1 (X, A, x0 ) can be rened a bit as follows: 3.2.13. P ROPOSITION . The map (3.2.11) 1 (X, A, x0 ) 1 (X, x0 ) ([ ], []) [ ] 1 (X, A, x0 ) denes a right action of the group 1 (X, x0 ) on the set 1 (X, A, x0 ); the orbit of the distinguished element [ox0 ] 1 (X, A, x0 ) is the kernel of the connection operator : 1 (X, A, x0 ) 0 (A, x0 ),

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

79

and the isotropy group of [ox0 ] is the image of the homomorphism: i : 1 (A, x0 ) 1 (X, x0 ); in particular, the map (3.2.12) q : 1 (X, x0 ) 1 (X, A, x0 ) induces, by passage to the quotient, a bijection between the set of right cosets 1 (X, x0 )/Im(i ) and the set Ker( ). P ROOF. It is easy to see that (3.2.11) does indeed dene a right action (see Corollary 3.1.5). The other statements follow from the long exact sequence of the pair (X, A) and from the elementary theory of actions of groups on sets, by observing that the map of action on the element ox0 : [ox0 ] : 1 (X, x0 ) 1 (X, A, x0 ) given by [ox0 ] ([]) = [ox0 ] coincides with (3.2.12). We now proceed with the study of brations. 3.2.14. D EFINITION . Let F, E, B be topological spaces; a continuous map p : E B is said to be a locally trivial bration with typical ber F if for every b B there exists an open neighborhood U of b in B and a homeomorphism: (3.2.13) : p1 (U ) U F such that pr1 = p|p1 (U ) , where pr1 denotes the rst projection of the product U F ; we then say that is a local trivialization of p around b, and we also say that the bration p is trivial on the open set U B . We call E the total space and B the base of the bration p; for every b B the subset Eb = p1 (b) E will be called the ber over b. Clearly, any local trivialization of p around b induces a homeomorphism of the ber Eb onto the typical ber F . We have the following: 3.2.15. L EMMA . Let p : E B a locally trivial bration, with typical ber F ; then, given e0 E , b0 B with p(e0 ) = b0 , the map: (3.2.14) p : n (E, Eb0 , e0 ) n (B, {b0 }, b0 ) = n (B, b0 ) is a bijection for every n 1. The proof of Lemma 3.2.15 is based on the following technical Lemma: 3.2.16. L EMMA . Let p : E B be a locally trivial bration with typical ber F ; then, for n 1, given continuous maps : I n B and : J n1 E with p = |J n1 , there exists a continuous map : I n E such that |J n1 = and such that the following diagram commutes:
e

>E
p

In

/B

80

3. ALGEBRAIC TOPOLOGY

P ROOF. The proof is split into several steps. (1) There exists a retraction r : I n J n1 , i.e., r is continuous and r|J n1 = Id. = 1 , . . . , 1 , 1 I Fix t Rn ; for each t I n dene r(t) as the unique 2 2 and t. point of J n1 that belongs to the straight line through t (2) The Lemma holds if there exists a trivialization (3.2.13) of p with Im() U. Let 0 : J n1 F be such that ( (t)) = ((t), 0 (t)), then, we consider: (t) = 1 (t), 0 (r(t)) , t I n. t J n1 ;

(3) The Lemma holds if n = 1. Let 0 = u0 < u1 < . . . < uk = 1 be a partition of I such that, for i = 0, . . . , k 1, ([ui , ui+1 ]) is contained in an open subset of B over which the bration p is trivial (see the idea of the proof of Theorem 3.1.23); using step (2), dene on the interval [ui , ui+1 ] starting with i = k 1 and proceeding inductively up to i = 0. (4) The Lemma holds in general. We prove the general case by induction on n; the base of induction is step (3). Suppose then that the Lemma holds for cubes of dimensions less than n. Consider a partition: (3.2.15) 0 = u 0 < u1 < . . . < u k = 1 of the interval I ; let a = (a1 , . . . , an1 ) be such that for each i = 1, . . . , n 1, the set ai is equal to one of the intervals [uj , uj +1 ], j = 0, . . . , k 1 of the partition (3.2.15), or else ai is equal to one of the points {uj }, j = 1, . . . , k 1; dene: Ia = Ia1 Ian1 I n1 . If r {0, . . . , n 1} is the number of indices i such that ai is an interval (containing more than one point), we will say that Ia is a block of dimension r. The partition (3.2.15) could have been chosen in such a way that each Ia [uj , uj +1 ] is contained in an open subset of B over which the bration is trivial (see the idea of the proof of Theorem 3.1.23). Using the induction hypotheses (or step (3)) we dene the map on the subsets Ia I where Ia is a block of dimension one. We then proceed inductively until when is dened on each Ia I such that Ia is a block of dimension r n 2. Fix now a in such a way that Ia is a block of dimension n 1; using step (2) we dene on ia [uj , uj +1 ] starting with j = k 1 and continuing inductively until j = 0. This concludes the proof.

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

81

The map in the statement of Lemma 3.2.16 is called a lifting of relatively to p. P ROOF OF L EMMA 3.2.15. Given [] n (B, b0 ), by Lemma 3.2.16 there exists a lifting : I n E of relatively to p, such that is constant equal to e0 on J n1 ; then n (E, Eb0 , e0 ) and p = []. This shows that p is surjective; we now show that p is injective. Let [1 ], [2 ] n (E, Eb0 , e0 ) be such that p ([1 ]) = p ([2 ]); then, there exists a homotopy H : I I n = I n+1 B such that H0 = p 1 , H1 = p 2 and Hs n b0 (B ) for every s I . Observe that: J n = I J n1 ({0} I n ) ({1} I n ); we can therefore dene a continuous map : J n E by setting (0, t) = 1 (t), (1, t) = 2 (t) for t I n , and (s, t) = e0 for s I , t J n1 . It follows from Lemma 3.2.16 that there exists a continuous map: : I I n = I n+1 E H = H eH |J n = ; it is then easy to see that H : 1 such that p H = 2 is n a homotopy in e0 (E, Eb0 ) and therefore [1 ] = [2 ] n (E, Eb0 , e0 ). This concludes the proof. The idea now is to replace n (E, Eb0 , e0 ) by n (B, b0 ) in the long exact homotopy sequence of the pair (E, Eb0 ), obtaining a new exact sequence. Towards this goal, we consider a locally trivial bration p : E B with typical ber F ; choose b0 B , f0 F , a homeomorphism h : Eb0 F and let e0 Eb0 be such that h(e0 ) = f0 . We then dene maps e in such a way that the following diagrams commute: (3.2.16)
pp ppp p p p = p x pp
h

n Eb0 , e0

n (F, f0 ) (3.2.17) n E, Eb0 , e0


p

NNN NNN NN i NNN& / n (E, e0 )

/ n1 Eb , e0 0
=

n (B, b0 )

/ n1 (F, f0 )

where i is induced by inclusion, and is the connection operator corresponding to the triple (E, Eb0 , e0 ). We then obtain the following:

82

3. ALGEBRAIC TOPOLOGY

3.2.17. C OROLLARY. Let p : E B be a locally trivial bration with typical ber F ; choosing b0 B , f0 F , a homeomorphism h : Eb0 F and taking e0 Eb0 such that h(e0 ) = f0 we obtain an exact sequence (3.2.18)
n (F, f0 ) n (E, e0 ) n (B, b0 ) n1 (F, f0 ) 1 (B, b0 ) 0 (F, f0 ) 0 (E, e0 ) 0 (B, b0 )

where e are dened respectively by the commutative diagrams (3.2.16) and (3.2.17). P ROOF. Everything except for the exactness at 0 (E, e0 ) follows directly from the long exact sequence of the pair (E, Eb0 ) and from the denitions of and . The exactness at 0 (E, e0 ) is obtained easily from Lemma 3.2.16 with n = 1. The exact sequence (3.2.18) is known as the long exact homotopy sequence of the bration p. 3.2.18. D EFINITION . A map p : E B is said to be a covering if p is a locally trivial bration with typical ber F that is a discrete space. We have the following: 3.2.19. C OROLLARY. If p : E B is a covering, then, given e0 E and b0 B with p(e0 ) = b0 , the map: p : n (E, e0 ) n (B, b0 ) is an isomorphism for every n 2. P ROOF. It follows directly from the long exact homotopy sequence of the bration p, observing that, since F is discrete, it is n (F, f0 ) = 0 for every n 1. 3.2.20. R EMARK . Let p : E B be a locally trivial bration with typical ber F ; choose b0 B and a homeomorphism h : Eb0 F . Let us take a closer look at the operator dened by diagram (3.2.17), in the case n = 1. f the operator dened by diagram (3.2.17) For each f F , we denote by taking n = 1 and replacing f0 by f and e0 by h1 (f ) in this diagram. We have the following explicit formula: (3.2.19)
f ([ ]) = h (0)

0 (F, f ),

1 b0 (B ),

where : I E is any lifting of (i.e., p = ) with (1) = h1 (f ). The existence of the lifting follows from Lemma 3.2.16 with n = 1. f Using (3.2.19), it is easy to see that only depends on the arc-connected component [f ] of F containing f ; for, if f1 , f2 F and : I F is a continuous curve with (0) = f1 and (1) = f2 , then, given a lifting of with (1) = h1 (f1 ), it follows that = (h1 ) is a lifting of = ob0 with (1) = h1 (f2 ), and so
f1 ([ ]) = h (0)

= h (0)

f2 f2 = ([]) = ([ ]).

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

83

Denoting by 0 (F ) the set of arc-connected components of F (disregarding the distinguished point) we obtain a map (3.2.20) 1 (B, b0 ) 0 (F ) 0 (F )
f given by ([ ], [f ]) ([ ]). It follows easily from (3.2.19) that (3.2.20) denes a left action of the group 1 (B, b0 ) on the set 0 (F ). Let us now x f0 F and let us set e0 = h1 (f0 ); using the long exact sequence of the bration p it follows that the sequence
1 (E, e0 ) 1 (B, b0 ) 0 (F, f0 ) 0 (E, e0 )

f0

is exact. This means that the orbit of the point [f0 ] 0 (F ) relatively to the action (3.2.20) is equal to the kernel of and that the isotropy group of [f0 ] is equal to the image of p ; hence the operator induces by passing to the quotient a bijection between the set of left cosets 1 (B, b0 )/Im(p ) and the set Ker( ). 3.2.21. E XAMPLE . Let p : E B be a locally trivial bration with discrete typical ber F , i.e., p is a covering. Choose b0 B , e0 Eb0 and a homeomorphism h : Eb0 F (actually, in the case of discrete ber, every bijection h will be a homeomorphism); set f0 = h(e0 ). Since 1 (F, f0 ) = 0, it follows from the long exact sequence of the bration that the map p : 1 (E, e0 ) 1 (B, b0 ) is injective; we can therefore identify 1 (E, e0 ) with the image of p . Observe that the set 0 (F, f0 ) may be identied with F . Under the assumption that E is arc-connected, we have 0 (E, e0 ) = 0, and it follows from Remark 3.2.20 that the map induces a bijection: (3.2.21) 1 (B, b0 )/1 (E, e0 ) F.
=

Unfortunately, since F has no group structure, the bijection (3.2.21) does not give any information about the group structures of 1 (E, e0 ) and 1 (B, b0 ). Let us now assume that the ber F has a group structure and that there exists a continuous right action: (3.2.22) EF (e, f ) e f E

of F on E (since F is discrete, continuity of (3.2.22) in this context means continuity in the second variable); let us also assume that the action (3.2.22) is free, i.e., without xed points, and that its orbits are the bers of p. If f0 = 1 is the unit of F and the homeomorphism h : Eb0 F is the inverse of the bijection: e0 : F we will show that the map (3.2.23) : 1 (B, b0 ) 0 (F, f0 ) =F f e0 f Eb0 ,

is a group homomorphism; it will then follow that Im(p ) 1 (E, e0 ) is a normal subgroup of 1 (B, b0 ) and that the bijection (3.2.21) is an isomorphism of groups.

84

3. ALGEBRAIC TOPOLOGY

Let us show that (3.2.23) is a homomorphism. To this aim, let , 1 b0 (B ) and let , : I E be lifts of and respectively, with (1) = (1) = e0 ; using (3.2.19) and identifying 0 (F, f0 ) with F we obtain: ([ ]) = h (0) , Dene : I E by setting: (t) = (t) h (0) , t I; then = is a lifting of = with (1) = e0 and, using again (3.2.19) we obtain: ([ ] []) = h (0) = h (0) = h( (0)) h (0) = ([ ]) ([]), which concludes the argument. 3.2.22. R EMARK . The groups 1 (X, x0 ) and 2 (X, A, x0 ) may not be abelian, in general; however, it can be shown that n (X, x0 ) is always abelian for 2 and n (X, A, x0 ) is always abelian for n 3 (see for instance [18, Proposition 2.1, Proposition 3.1, Chapter 4]). 3.2.23. R EMARK . Generalizing the result of Proposition 3.1.11, given n 1 it is possible to associate to each curve : I X with (0) = x0 and (1) = x1 an isomorphism: # : n (X, x0 ) n (X, x1 ); in particular, if x0 e x1 belong to the same arc-connected component of X then n (X, x0 ) is isomorphic to n (X, x1 ). The isomorphism # is dened by setting: # ([]) = [ ], where is constructed using a homotopy H : = such that Hs (t) = (t) for every t I n and all s I (for the details, see [18, Theorem 14.1, Chapter 4]). Then, as in Example 3.1.17, it is possible to show that if X is contractible, then n (X, x0 ) = 0 for every n 0. If Im() A X then, given n 1, we can also dene a bijection of pointed sets: # : n (X, A, x0 ) n (X, A, x1 ), which is a group isomorphism for n 2 (see [18, Exercises of Chapter 4]). 3.2.1. Applications to the theory of classical Lie groups. In this subsection we will use the long exact homotopy sequence of a bration to compute the fundamental group and the connected components of the classical Lie groups introduced in Subsection 2.1.1. All the spaces considered in this section are differentiable manifolds, hence the notions of connectedness and of arc-connectedness will always be equivalent (see Exercise 3.2). We will assume familiarity with the concepts and the notions introduced in Subsections 2.1.1 and 2.1.2; in particular, without explicit mention, we will make systematic use of the results of Theorem 2.1.14 and of Corollaries 2.1.9, 2.1.15, 2.1.16 and 2.1.17. ([]) = h (0) .

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

85

The relative homotopy groups will not be used in this Section; from Section 3.2 the reader is required to keep in mind the Examples 3.2.6, 3.2.10 and 3.2.21, and, obviously, Corollary 3.2.17. In order to simplify the notation, we will henceforth omit the specication of the basepoint x0 when we refer to a homotopy group, or set, n (X, x0 ), provided that the choice of such basepoint is not relevant in the context (see Corollary 3.1.12 and Remark 3.2.23); therefore, we will write n (X ). We start with an easy example: 3.2.24. E XAMPLE . Denote by S 1 C the unit circle; then, the map p : I R given by p(t) = e2it is a surjective group homomorphism whose kernel is Ker(p) = Z. It follows that p is a covering map. Moreover, the action of Z on I R by translation is free, and its orbits are the bers of p; it follows from Example 3.2.21 that we have an isomorphism: S1 : 1 (S 1 , 1) Z given by ([ ]) = (0), where :II R is a lifting of such that (1) = 0. In 1 particular, the homotopy class of the loop : I S given by: (3.2.24) is a generator of 1 (S 1 , 1) (t) = e2it , Z. t I,

3.2.25. E XAMPLE . Let us show that the special unitary group SU(n) is (connected and) simply connected. First, observe that the canonical action of the group SU(n + 1) on Cn+1 restricts to an action of SU(n + 1) on the unit sphere S 2n+1 ; it is easy to see that this action is transitive, and that the isotropy group of the point en+1 = (0, . . . , 0, 1) Cn+1 is identied with SU(n). It follows that the quotient SU(n + 1)/SU(n) is diffeomorphic to the sphere S 2n+1 ; we therefore have a bration: p : SU(n + 1) S 2n+1 , with typical ber SU(n). Since the sphere S 2n+1 is simply connected (see Example 3.1.24), the long exact homotopy sequence of the bration p gives us: (3.2.25) (3.2.26) 0 (SU(n)) 0 (SU(n + 1)) 0 1 (SU(n)) 1 (SU(n + 1)) 0.

Since SU(1) = {1} is clearly simply connected, from the exactness of (3.2.25) it follows by induction on n that SU(n) is connected. Moreover, from the exactness of (3.2.26) it follows by induction on n that SU(n) is simply connected. 3.2.26. E XAMPLE . Let us show now that the unitary group U(n) is connected, and that 1 (U(n)) Z for every n 1. Consider the determinant map det : U(n) S 1 ; we have that det is a surjective homomorphism of Lie groups, and therefore it is a bration with typical ber Ker(det) = SU(n). Keeping in mind that SU(n)

86

3. ALGEBRAIC TOPOLOGY

is simply connected (see Example 3.2.25), from the bration det we obtain the following exact sequence: (3.2.27) (3.2.28) 0 0 (U(n)) 0
0 1 (U(n), 1) 1 (S 1 , 1) 0

det

From (3.2.27) we conclude that U(n) is connected, and from (3.2.28) we obtain that the map (3.2.29) is an isomorphism. 3.2.27. E XAMPLE . We will now show that the special orthogonal group SO(n) is connected for n 1. The canonical action of SO(n + 1) on I Rn+1 restricts to n an action of SO(n + 1) on the unit sphere S ; it is easy to see that this action is transitive, and that the isotropy group of the point en+1 = (0, . . . , 0, 1) I Rn+1 is identied with SO(n). It follows that the quotient SO(n + 1)/SO(n) is diffeomorphic to the sphere S n , and we obtain a bration: (3.2.30) p : SO(n + 1) S n , 0 (SO(n)) 0 (SO(n + 1)) 0 from which it follows, by induction on n, that SO(n) is connected for every n (clearly, SO(1) = {1} is connected). The determinant map induces an isomorphism between the quotient O(n)/SO(n) and the group {1, 1} Z2 , from which it follows that O(n) has precisely two connected components: SO(n) and its complementary. 3.2.28. E XAMPLE . We now show that the group GL+ (n, I R) is connected. If n , it is easy to see that there exists a unique we choose any basis (bi )n of I R i=1 orthonormal basis (ui )n Rn such that, for every k = 1, . . . , n, the vectors i=1 of I k k (bi )i=1 and (ui )i=1 are a basis of the same k -dimensional subspace of I Rn and n dene the same orientation of this subspace. The vectors (ui )i=1 can be written explicitly in terms of the (bi )n i=1 ; such formula is known as the GramSchmidt orthogonalization process. Given any invertible matrix A GL(n, I R), we denote by r(A) the matrix obtained from A by an application of the Gram-Schmidt orthogonalization process on its columns; the map r from GL(n, I R) onto O(n) is differentiable (but it is not a homomorphism). Observe that if A O(n), then r(A) = A; for this we say that r is a retraction. Denote by T+ the subgroup of GL(n, I R) consisting of upper triangular matrices with positive entries on the diagonal, i.e., T+ = {(aij )nn GL(n, I R) : aij = 0 if i > j, aii > 0, i, j = 1, . . . , n}. Then, it is easy to see that we obtain a diffeomorphism: (3.2.31) GL(n, I R) A r(A), r(A)1 A O(n) T+ . with typical ber SO(n); then, we have an exact sequence:
= det : 1 (U(n), 1) 1 (S 1 , 1) =Z

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

87

We have that (3.2.31) restricts to a diffeomorphism of GL+ (n, I R) onto SO(n) T+ . It follows from Example 3.2.27 that GL+ (n, I R) is connected, and that the general linear group GL(n, I R) has two connected components: GL+ (n, I R) and its complementary. 3.2.29. R EMARK . Actually, it is possible to show that GL+ (n, I R) is connected by an elementary argument, using the fact that every invertible matrix can be written as the product of matrices corresponding to elementary row operations. Then, the map r : GL(n, I R) O(n) dened as in Example 3.2.28 gives us an alternative proof of the connectedness of SO(n). 3.2.30. E XAMPLE . We will now show that the group GL(n, C) is connected and that: 1 (GL(n, C)) = Z. We use the same idea as in Example 3.2.28; observe that it is possible to dene a Gram-Schmidt orthonormalization process also for bases of Cn . Then, we obtain a diffeomorphism: GL(n, C) A r(A), r(A)1 A U(n) T+ (C),

where T+ (C) denotes the subgroup of GL(n, C) consisting of those upper triangular matrices having positive real entries on the diagonal: T+ (C) = {(aij )nn GL(n, C) : aij = 0 if i > j, aii I R and aii > 0, i, j = 1, . . . , n}. It follows from Example 3.2.26 that GL(n, C) is connected and that 1 (GL(n, C)) is isomorphic to Z for n 1; more explicitly, we have that the inclusion i : U(n) GL(n, C) induces an isomorphism: i : 1 (U(n), 1) 1 (GL(n, C), 1). 3.2.31. R EMARK . Also the connectedness of GL(n, C) can be proven by a simpler method, using elementary row reduction of matrices. Then, the GramSchmidt orthonormalization process gives us an alternative proof of the connectedness of U(n) (see Remark 3.2.29). 3.2.32. E XAMPLE . We will now consider the groups SL(n, I R) and SL(n, C). We have a Lie group isomorphism: SL(n, I R) I R+ (T, c) c T GL+ (n, I R),
=

where I R+ = ]0, +[ is seen as a multiplicative group; it follows from Example 3.2.28 that SL(n, I R) is connected, and that the inclusion i : SL(n, I R) GL+ (n, I R) induces an isomorphism: i : 1 (SL(n, I R), 1) 1 (GL+ (n, I R), 1). The group 1 (GL+ (n, I R)) will be computed in Example 3.2.35 ahead.
=

88

3. ALGEBRAIC TOPOLOGY

Let us look now at the complex case: for z C \ {0} we dene the diagonal matrix: z 1 0 (z ) = GL(n, C); 0 ... 1 we then obtain a diffeomorphism (which is not an isomorphism): SL(n, C) I R+ S 1 (T, c, z ) (cz )T GL(n, C).

Then, it follows from Example 3.2.30 that SL(n, C) is connected and that: 1 (GL(n, C)) =Z = Z 1 (SL(n, C)), from which we get that SL(n, C) is simply connected. In order to compute the fundamental group of the special orthogonal group SO(n) we need the following result: Sn, 3.2.33. L EMMA . If S n I Rn+1 denotes the unit sphere, then, for every x0 n we have k (S , x0 ) = 0 for 0 k < n.

n n P ROOF. Let k x0 (S ). If is not surjective, then there exists x S with n n n Im() S \ {x}; but S \ {x} is homeomorphic to I R by the stereographic pron jection, hence [] = [ox0 ]. It remains to show that any k x0 (S ) is homotopic n in k x0 (S ) to a map which is not surjective. Let > 0 be xed; it is known that there exists a differentiable1 map : I k I Rn+1 such that (t) (t) < for every t I k (see [23, Teorema 10, 5, Cap tulo 7]). Let : I R [0, 1] be a differentiable map such that (s) = 0 for |s| and (s) = 1 for |s| 2. Dene : I Rn+1 I Rn+1 by setting

(x) = ( x x0 )(x x0 ) + x0 ,

xI Rn+1 ;

then, is differentiable in I Rn+1 , (x) = x0 for x x0 and (x) x 2 n +1 for every x I R . It follows that : I k I Rn+1 is a differentiable map k ( )(I ) {x0 } and ( )(t) (t) 3 for every t I k . Choosing n > 0 with 3 < 1, then we can dene a homotopy H : = in k x0 (S ) by setting: (1 s)(t) + s( )(t) Hs (t) = , t I k , s I, (1 s)(t) + s( )(t) where (t) = ( )(t)/ ( )(t) , t I k , is a differentiable map; since k < n, it follows that cannot be surjective, because its image has null measure in S n (see [23, 2, Cap tulo 6]). This concludes the proof.
1The differentiability of a map dened in a non necessarily open subset of I Rk means that the

map admits a differentiable extension to some open subset containing its domain.

3.2. THE HOMOTOPY EXACT SEQUENCE OF A FIBRATION

89

3.2.34. E XAMPLE . The group SO(1) is trivial, therefore it is simply connected; the group SO(2) is isomorphic to the unit circle S 1 (see Example 3.2.27, hence: 1 (SO(2)) = Z. n For n 3, Lemma 3.2.33 tells us that 2 (S ) = 0, and so the long exact homotopy sequence of the bration (3.2.30) becomes:
0 1 (SO(n), 1) 1 (SO(n + 1), 1) 0 =

where i is induced by the inclusion i : SO(n) SO(n + 1); it follows that 1 (SO(n)) is isomorphic to 1 (SO(n + 1)). We will show next that 1 (SO(3)) = Z2 , from which it will then follow that 1 (SO(n)) = Z2 , n 3. Consider the inner product g in the Lie algebra su(2) dened by g (X, Y ) = tr(XY ),

X, Y su(2),

where Y denotes here the conjugate transpose of the matrix Y and tr(U ) denotes the trace of the matrix U ; consider the adjoint representation of SU(2): (3.2.32) Ad : SU(2) SO(su(2), g ) given by Ad(A) X = AXA1 for A SU(2), X su(2); it is easy to see that the linear endomorphism Ad(A) of su(2) is actually g -orthogonal for every A SU(2) and that (3.2.32) is a Lie group homomorphism. Clearly, SO(su(2), g ) is isomorphic to SO(3). An explicit calculation shows that Ker(Ad) = {Id, Id}, and since the domain and the counterdomain of (3.2.32) have the same dimension, it follows that the image of (3.2.32) is an open subgroup of SO(su(2), g ); since SO(su(2), g ) is connected (Example 3.2.27), we conclude that (3.2.32) is surjective, and so it is a covering map. Since SU(2) is simply connected (Example 3.2.25), it follows from Example 3.2.21 that 1 (SO(3)) = Z2 , keeping in mind the action of Z2 = {Id, Id} on SU(2) by translation. The non trivial element of 1 (SO(3), 1) coincides with the homotopy class of any loop of the form Ad , where : I SU(2) is a curve joining Id and Id. 3.2.35. E XAMPLE . The diffeomorphism (3.2.31) shows that the inclusion i of SO(n) into GL+ (n, I R) induces an isomorphism: (3.2.33) i : 1 (SO(n), 1) 1 (GL+ (n, I R), 1).
=

It follows from Example 3.2.34 that 1 (GL+ (n, I R)) is trivial for n = 1, it is isomorphic to Z for n = 2, and it is isomorphic to Z2 for n 3. 3.2.36. E XAMPLE . We will now look at the symplectic group Sp(2n, I R) and we will show that it is connected for every n 1. Let be the canonical symplectic form of I R2n and let + be the Grassmannian of oriented Lagrangians of the symplectic space (I R2n , ), that is: + = (L, O) : L I R2n is Lagrangian, and O is an orientation of L .

90

3. ALGEBRAIC TOPOLOGY

We have an action of the symplectic group Sp(2n, I R) on the set + given by T (L, O) = (T (L), O ), where O is the unique orientation of T (L) that makes T |L : L T (L) positively oriented. By Remark 1.4.29, we have that the restriction of this action to the unitary group U(n) is transitive. Consider the Lagrangian L0 = I Rn {0} and let O be the orientation of L0 corresponding to the canonical basis of I Rn ; then, the isotropy group of (L0 , O) relative to the action of U(n) is SO(n). The isotropy group of (L0 , O) relative to the action of Sp(2n, I R) will be denoted by Sp+ (2n, I R, L0 ). In formulas (1.4.7) and (1.4.8) we have given an explicit description of the matrix representations of the elements of Sp(2n, I R); using these formulas it is easy to see that Sp+ (2n, I R, L0 ) consists of matrices of the form: (3.2.34) T = A AS 0 A 1 , A GL+ (n, I R), S n n symmetric matrix,

where A denotes the transpose of A. It follows that we have a diffeomorphism: (3.2.35) Sp+ (2n, I R, L0 ) T (A, S ) GL+ (n, I R) Bsym (I Rn )

where A and S are dened by (3.2.34). We have the following commutative diagrams of bijections: n) U(n)/SO( K
i

KK KK KK 1 KKK %

/ Sp(2n, I R)/Sp+ (2n, I R, L0 ) l l l lll lll l l 2 l l v ll

where the maps 1 and 2 are induced respectively by the actions of U(n) and of Sp(2n, I R) on + and i is induced by the inclusion i : U(n) Sp(2n, I R) by passage to the quotient; we have that i is a diffeomorphism. Hence, we have a bration: (3.2.36) p : Sp(2n, I R) Sp(2n, I R)/Sp+ (2n, I R, L0 ) = U(n)/SO(n)

GL+ (n, I whose typical ber is Sp+ (2n, I R, L0 ) = R) Bsym (I Rn ). By Example 3.2.28 this typical ber is connected, and by Example 3.2.26 the base manifold U(n)/SO(n) is connected. It follows now easily from the long exact homotopy sequence of the bration (3.2.36) that the symplectic group Sp(2n, I R) is connected. 3.2.37. E XAMPLE . We will now show that the fundamental group of the symplectic group Sp(2n, I R) is isomorphic to Z. Using the exact sequence of the bration (3.2.36) and the diffeomorphism (3.2.35), we obtain an exact sequence:
(3.2.37) 1 (GL+ (n, I R)) 1 (Sp(2n, I R)) 1 U(n)/SO(n) 0

where is induced by the map : GL+ (n, I R) Sp(2n, I R) given by: (A) = A 0 0 A 1 , A GL+ (n, I R).

3.3. SINGULAR HOMOLOGY GROUPS

91

We will show rst that the map is the null map; we have the following commutative diagram (see (3.2.29) and (3.2.33)):
0

(3.2.38)

1 (SO(n))
=

/ 1 (U(n))

det
=

* / 1 (S 1 )

1 (GL+ (n, I R))

 / 1 (Sp(2n, I R))

where the unlabeled arrows are induced by inclusion.2 A simple analysis of the diagram (3.2.38) shows that = 0. Now, the exactness of the sequence (3.2.37) implies that p is an isomorphism of 1 (Sp(2n, I R)) onto the group 1 U(n)/SO(n) ; let us compute this group. Consider the quotient map: q : U(n) U(n)/SO(n); we have that q is a bration. We obtain a commutative diagram: (3.2.39) 1 (SO(n))
/ 1 (U(n)) MMM MMM MM = det 0 MMM& 
q

/ 1 U(n)/SO(n)

/0

1 (S 1 )

The upper horizontal line in (3.2.39) is a portion of the homotopy exact sequence of the bration q ; it follows that q is an isomorphism. Finally, denoting by i the inclusion of U(n) in Sp(2n, I R) we obtain a commutative diagram:
oo7 ooo o o oo ooo
i

1 (Sp(2n, I R))
=

1 (U(n))

RRR RRR p RRR RRR = R( / 1 U(n)/SO(n)

from which it follows that i is an isomorphism:


Z 1 (Sp(2n, I R), 1). = 1 (U(n), 1) =

3.3. Singular Homology Groups In this section we will give a brief exposition of the denition and the basic properties of the group of (relative and absolute) singular homology of a topological space X ; we will describe the homology exact sequence of a pair of topological spaces. For all p 0, we will denote by (ei )p Rp and by e0 the i=1 the canonical basis of I p 0 zero vector of I R ; by I R we will mean the trivial space {0}. Observe that, with this notations, we will have a small ambiguity due to the fact that, if q p i,
R) depends on the identication of n n complex matrices U(n) into Sp(2n, I with 2n 2n real matrices; see Remark 1.2.9.
2The inclusion of

92

3. ALGEBRAIC TOPOLOGY

the symbol ei will denote at the same time a vector of I Rp and also a vector of I Rq ; however, this ambiguity will be of a harmless sort and, if necessary, the reader may consider identications I R0 I R1 I R2 I R . Given p 0, the p-th standard simplex is dened as the convex hull p of the set {ei }p Rp ; more explicitly: i=0 in I p =
p i=0

ti ei :

p i=0

ti = 1, ti 0, i = 0, . . . , p .

Observe that 0 is simply the point {0} and 1 is the unit interval I = [0, 1]. Let us x some terminology concerning the concepts related to free abelian groups: 3.3.1. D EFINITION . If G is an abelian group, then a basis3 of G is a family (b )A such that every g G is written uniquely in the form g = A n b , where each n is in Z and n = 0 except for a nite number of indices A. If G is another abelian group, then a homomorphism f : G G is uniquely determined when we specify its values on the elements of some basis of G. An abelian group that admits a basis is said to be free. If A is any set, the free abelian group GA generated by A is the group of all almost zero maps N : A Z, i.e., N () = 0 except for a nite number of indices A; the sum in GA is dened in the obvious way: (N1 + N2 )() = N1 () + N2 (). We then identify each A with the function N GA dened by N () = 1 and N ( ) = 0 for every = . Then, G is indeed a free abelian group, and A GA is a basis of GA . 3.3.2. D EFINITION . For p 0, a singular p-simplex is an arbitrary continuous map: T : p X. We denote by Sp (X ) the free abelian group generated by the set of all singular p-simplexes in X ; the elements in Sp (X ) are called singular p-chains. If p = 0, we identify the singular p-simplexes in X with the points of X , and S0 (X ) is the free abelian group generated by X . If p < 0, our convention will be that Sp (X ) = {0}. Each singular p-chain can be written as: c=
T singular p-simplex

nT T,

where nT Z and nT = 0 except for a nite number of singular p-simplexes; the coefcients nT are uniquely determined by c. Given a nite dimensional vector space V and given v0 , . . . , vp V , we will denote by (v0 , . . . , vp ) the singular p-simplex in V dened by:
p p

(3.3.1)

(v0 , . . . , vp )
i=0

ti ei =
i=0

ti vi ,

3An abelian group is a Z-module, and our denition of basis for an abelian group coincides with

the usual denition of basis for a module over a ring.

3.3. SINGULAR HOMOLOGY GROUPS

93

where each ti 0 and p i=0 ti = 1; observe that (v0 , . . . , vp ) is the unique afne function that takes ei into vi for every i = 0, . . . , p. For each p Z, we will now dene a homomorphism: p : Sp (X ) Sp1 (X ). If p 0 we set p = 0. For p > 0, since Sp (X ) is free, it sufces to dene p on a basis of Sp (X ); we then dene p (T ) when T is a singular p-simplex in X by setting:
p

p (T ) =
i=0

(1)i T (e0 , . . . , ei , . . . , ep ),

where ei means that the term ei is omitted in the sequence. For each i = 0, . . . , p, the image of the singular (p 1)-simplex in I Rp (e0 , . . . , ei , . . . , ep ) can be visualized as the face of the standard simplex p which is opposite to the vertex ei . If c Sp (X ) is a singular p-chain, we say that p (c) is its boundary; observe that if T : [0, 1] X is a singular 1-simplex, then 1 (T ) = T (1) T (0). We have thus obtained a sequence of abelian groups and homomorphisms (3.3.2) Sp (X ) Sp1 (X )
p+1 p p1

The sequence (3.3.2) has the property that the composition of two consecutive arrows vanishes: 3.3.3. L EMMA . For all p Z, we have p1 p = 0. P ROOF. If p 1 the result is trivial; for the case p 2 it sufces to show that p1 (p (T )) = 0 for every singular p-simplex T . Observing that (v0 , . . . , vq ) (e0 , . . . , ei , . . . , eq ) = (v0 , . . . , vi , . . . , vq ) we compute as follows: p1 (p (T )) =
j<i

(1)j +i T (e0 , . . . , ej , . . . , ei , . . . , ep ) (1)j +i1 T (e0 , . . . , ei , . . . , ej , . . . , ep ) = 0.


j>i

Let us give the following general denition: 3.3.4. D EFINITION . A chain complex is a family C = (Cp , p )pZ where each Cp is an abelian group, and each p : Cp Cp1 is a homomorphism such that p1 p = 0 for every p Z. For each p Z we dene: Zp (C) = Ker(p ), Bp (C) = Im(p+1 ), and we say that Zp (C), Bp (C) are respectively the group of p-cycles and the group of p-boundaries of the complex C. Clearly, Bp (C) Zp (C), and we can therefore dene: Hp (C) = Zp (C)/Bp (C); we say that Hp (C) is the p-th homology group of the complex C.

94

3. ALGEBRAIC TOPOLOGY

If c Zp (C) is a p-cycle, we denote by c + Bp (C) its equivalence class in Hp (C); we say that c + Bp (C) is the homology class determined by c. If c1 , c2 Zp (C) determine the same homology class (that is, if c1 c2 Bp (C)) we say that c1 and c2 are homologous cycles. Lemma 3.3.3 tells us that S(X ) = (Sp (X ), p )pZ is a chain complex; we say that S(X ) is the singular complex of the topological space X . We write: Zp (S(X )) = Zp (X ), Bp (S(X )) = Bp (X ), Hp (S(X )) = Hp (X ); and we call Zp (X ), Bp (X ) and Hp (X ) respectively the group of singular p-cycles, the group of singular p-boundaries and the p-th singular homology group of the topological space X . Clearly, Hp (X ) = 0 for p < 0 and H0 (X ) = S0 (X )/B0 (X ). We dene a homomorphism (3.3.3) : S0 (X ) Z by setting (x) = 1 for every singular 0-simplex x X . It is easy to see that 1 = 0; for, it sufces to see that (1 (T )) = 0 for every singular 1-simplex T in X . We therefore obtain a chain complex: (3.3.4) Sp (X ) Sp1 (X )
1 S0 (X ) Z 0

p+1

p1

3.3.5. D EFINITION . The homomorphism (3.3.3) is called the augmentation map of the singular complex S(X ); the chain complex in (3.3.4), denoted by (S(X ), ) is called the augmented singular complex of the space X . The groups of p-cycles, of p-boundaries and the p-th homology group of (S(X ), ) are denoted p (X ), B p (X ) and H p (X ) respectively; we say that H p (X ) is the p-th reduced by Z singular homology group of X . Clearly, for p 1 we have: p (X ) = Zp (X ), B p (X ) = Bp (X ), Z p (X ) = Hp (X ). H

From now on we will no longer specify the index p in the operator p , and we will write more concisely: p = , p Z. 3.3.6. E XAMPLE . If X = is the empty set, then obviously Sp (X ) = 0 for p (X ) = 0 for every p = 1; on every p Z, hence Hp (X ) = 0 for every p, and H 1 (X ) = Z. the other hand, we have H If X is non empty, then any singular 0-simplex x0 X is such that (x0 ) = 1 (X ) = 0. Concerning the relation 1, and so is surjective; it follows that H 0 (X ), it is easy to see that we can identify H 0 (X ) with a between H0 (X ) and H subgroup of H0 (X ), and that H 0 (X ) Z (x0 + B0 (X )) = 0 (X ) Z, H0 (X ) = H where Z (x0 + B0 (X )) is the subgroup (innite cyclic) generated by the homology class of x0 in H0 (X ).

3.3. SINGULAR HOMOLOGY GROUPS

95

3.3.7. E XAMPLE . If X is arc-connected and not empty , then any two singular 0-simplexes x0 , x1 X are homologous; indeed, if T : [0, 1] X is a continuous curve from x0 to x1 , then T is a singular 1-simplex and T = x1 x0 B0 (X ). It follows that the homology class of any x0 X generates H0 (X ), and since (x0 ) = 1, it follows that no non zero multiple of x0 is a boundary; therefore: 0 (X ) = 0. H0 (X ) = Z, H 3.3.8. E XAMPLE . If X is not arc-connected, we can write X = A X , where each X is an arc-connected component of X . Then, every singular simplex in X has image contained in some X and therefore: Sp (X ) =
A

Sp (X ),

from which it follows that: Hp (X ) =


A

Hp (X ).

In particular, it follows from Example 3.3.7 that: H0 (X ) =


A

Z.

The reader should compare this fact with Remark 3.2.5. 3.3.9. E XAMPLE . Suppose that X I Rn is a star-shaped subset around the point w X . For each singular p-simplex T in X we dene a singular (p + 1)simplex [T, w] in X in such a way that the following diagram commutes: I p

p+1 where and are dened by: (s, t) = (1 s)t + s ep+1 ,

OOO OOO OOO OOO O' /X


[T,w]

(s, t) = (1 s)T (t) + sw,

t p , s I ;

geometrically, the singular (p + 1)-simplex [T, w] coincides with T on the face p p+1 , it takes the vertex ep+1 on w and it is afne on the segment that joins t with ep+1 for every t p . The map T [T, w] extends to a homomorphism: Sp (X ) c [c, w] Sp+1 (X ). [c, w] + (1)p+1 c, p 1 (c)w c, p = 0; It is easy to see that for each singular p-chain c Sp (X ) we have: (3.3.5) [c, w] =

for, it sufces to consider the case that c = T is a singular p-simplex, in which case (3.3.5) follows from an elementary analysis of the denition of [T, w] and of the

96

3. ALGEBRAIC TOPOLOGY

denition of the boundary operator. In particular, we have [c, w] = (1)p+1 c for p (X ) and therefore c B p (X ); we conclude that, if X is star shaped, every c Z then p (X ) = 0, p Z. H 3.3.10. D EFINITION . Let C = (Cp , p ), C = (Cp , p ) be chain complexes; a chain map : C C is a sequence of homomorphisms p : Cp Cp , p Z, such that for every p the diagram Cp1 Cp
p

p1

Cp Cp1
p

commutes; in general, we will write rather than p . It is easy to see that if is a chain map, then (Zp (C)) Zp (C ) and (Bp (C)) Bp (C ), so that induces by passage to quotients a homomorphism : Hp (C) Hp (C ); we say that is the map induced in homology by the chain map . Clearly, if : C C and : C C are chain maps, then also their composition is a chain map; moreover, ( ) = , and if Id is the identity of the complex C, i.e., Idp is the identity of Cp for every p, then Id is the identity of Hp (C) for every p. It follows that if is a chain isomorphism, i.e., p is an isomorphism for every p, then is an isomorphism between the homology groups, and (1 ) = ( )1 . If X, Y are topological spaces and f : X Y is a continuous map, then for each p we dene a homomorphism: f# : Sp (X ) Sp (Y ) by setting f# (T ) = f T for every singular p-simplex T in X . It is easy to see that f# is a chain map; we say that f# is the chain map induced by f . It is clear that, given continuous maps f : X Y , g : Y Z then (g f )# = g# f# , and that if Id is the identity map of X , then Id# is the identity of S(X ); in particular, if f is a homeomorphism, then f# is a chain isomorphism, and (f 1 )# = (f# )1 . We have that the chain map f# induces a homomorphism f : Hp (X ) Hp (Y ) between the groups of singular homology of X and Y , that will be denoted simply by f . 3.3.11. R EMARK . If A is a subspace of X , then we can identify the set of singular p-simplexes in A with a subset of the set of singular p-simplexes in X ; then Sp (A) is identied with a subgroup of Sp (X ). If i : A X denotes the inclusion, then i# is simply the inclusion of Sp (A) into Sp (X ). However, observe that the induced map in homology i is in general not injective, and there exists no identication of Hp (A) with a subgroup of Hp (X ).

3.3. SINGULAR HOMOLOGY GROUPS

97

Recall that (X, A) is called a pair of topological spaces when X is a topological space and A X is a subspace. We dene the singular complex S(X, A) of the pair (X, A) by setting: Sp (X, A) = Sp (X )/Sp (A); the boundary operator of S(X, A) is dened using the boundary operator of S(X ) by passage to the quotient. Clearly, S(X, A) is a chain complex; we write Hp (S(X, A)) = Hp (X, A). We call Hp (X, A) the p-th group of relative homology of the pair (X, A). If f : (X, A) (Y, B ) is a map of pairs (Denition 3.2.8) then the chain map f# passes to the quotient and it denes a chain map f# : S(X, A) S(Y, B ) that will also be denoted by f# ; then f# induces a homomorphism between the groups of relative homology, that will be denoted by f . Clearly, if f : (X, A) (Y, B ) and g : (Y, B ) (Z, C ) are maps of pairs, then (g f )# = g# f# and that if Id is the identity map of X , then Id# is the identity of S(X, A); also, if f : (X, A) (Y, B ) is a homeomorphism of pairs, i.e., f is a homeomorphism of X onto Y with f (A) = B , then f# is a chain isomorphism. 3.3.12. R EMARK . An intuitive way of thinking of the groups of relative ho p (X/A) mology Hp (X, A) is to consider them as the reduced homology groups H of the space X/A which is obtained from X by collapsing all the points of A to a single point. This idea is indeed a theorem that holds in the case that A X is closed and it is a deformation retract of some open subset of X . The proof of this theorem requires further development of the theory, and it will be omitted in these notes (see [31, Exercise 2, 39, Chapter 4]). 3.3.13. E XAMPLE . If A is the empty set, then S(X, A) = S(X ), and therefore Hp (X, A) = Hp (X ) for every p Z; for this reason, we will not distinguish between the space X and the pair (X, ). 3.3.14. E XAMPLE . The identity map of X induces a map of pairs: (3.3.6) q : (X, ) (X, A);
1 Bp (X, A) = q # Bp (S(X, A)) ;

then q# : S(X ) S(X, A) is simply the quotient map. We dene


1 Zp (X, A) = q # Zp (S(X, A)) ,

we call Zp (X, A) and Bp (X, A) respectively the group of relative p-cycles and the group of relative p-boundaries of the pair (X, A). More explicitly, we have Zp (X, A) = c Sp (X ) : c Sp1 (A) = 1 (Sp1 (A)), Bp (X, A) = c + d : c Sp+1 (X ), d Sp (A) = Bp (X ) + Sp (A); Observe that Zp (S(X, A)) = Zp (X, A)/Sp (A), Bp (S(X, A)) = Bp (X, A)/Sp (A);

98

3. ALGEBRAIC TOPOLOGY

it follows from elementary theory of quotient of groups that: (3.3.7) Hp (X, A) = Hp (S(X, A)) = Zp (X, A)/Bp (X, A). Given c Zp (X, A), the equivalence class c + Bp (X, A) Hp (X, A) is called the homology class determined by c in Hp (X, A); if c1 , c2 Zp (X, A) determine the same homology class in Hp (X, A), i.e., if c1 c2 Bp (X, A), we say that c1 and c2 are homologous in S(X, A). 3.3.15. E XAMPLE . If X is arc-connected and A = , then arguing as in Example 3.3.7 we conclude that any two 0-simplexes in X are homologous in S(X, A); however, in this case every point of A is a singular 0-simplex which is homologous to 0 in S(X, A), hence: H0 (X, A) = 0. If X is not arc-connected, then we write X = A X , where each X is an arc-connected component of X ; writing A = A X , as in Example 3.3.8 we obtain: Sp (X, A) = Sp (X , A );
A

and it follows directly that: Hp (X, A) =


A

Hp (X , A ).

In the case p = 0, we obtain in particular that: H0 (X, A) =


A

Z,

where A is the subset of indices A such that A = . Our goal now is to build an exact sequence that relates the homology groups Hp (X ) and Hp (A) with the relative homology groups Hp (X, A). 3.3.16. D EFINITION . Given chain complexes C, D, E , we say that (3.3.8) 0 C D E 0

is a short exact sequence of chain complexes if and are chain maps and if for every p Z the sequence of abelian groups and homomorphisms 0 Cp Dp Ep 0 is exact. We have the following result of Homological Algebra: 3.3.17. L EMMA (The Zig-Zag Lemma). Given a short exact sequence of chain complexes (3.3.8), there exists an exact sequence of abelian groups and homomorphisms: (3.3.9)
Hp (C) Hp (D) Hp (E ) Hp1 (C)

3.3. SINGULAR HOMOLOGY GROUPS

99

where and are induced by and respectively, and the homomorphism is dened by: (3.3.10) e + Bp (E ) = c + Bp1 (C), e Zp (E ),

where c Cp1 is chosen in such a way that (c) = d and d Dp is chosen in such a way that (d) = e; the denition (3.3.10) does not depend on the arbitrary choices involved. P ROOF. The proof, based on an exhaustive analysis of all the cases, is elementary and it will be omitted. The details can be found in [31, 24, Chapter 3]. The exact sequence (3.3.9) is known as the long exact homology sequence corresponding to the short exact sequence of chain complexes (3.3.8) Coming back to the topological considerations, if (X, A) is a pair of topological spaces, we have a short exact sequence of chain complexes: (3.3.11) 0 S(A) S(X ) S(X, A) 0
i# q#

where i# is induced by the inclusion i : A X and q# is induced by (3.3.6). Then, it follows directly from the Zig-Zag Lemma the following: 3.3.18. P ROPOSITION . Given a pair of topological spaces (X, A) then there exists an exact sequence (3.3.12) q i i Hp (X, A) Hp1 (A) Hp (A) Hp (X ) where i is induced by the inclusion i : A X , q is induced by (3.3.6) and the homomorphism is dened by: c + Bp (X, A) = c + Bp1 (A), c Zp (X, A);

such denition does not depend on the choices involved. If A = we also have an exact sequence (3.3.13) q i i p1 (A) p (A) p (X ) H H Hp (X, A) H whose arrows are obtained by restriction of the corresponding arrows in the sequence (3.3.12). P ROOF. The sequence (3.3.12) is obtained by applying the Zig-Zag Lemma to the short exact sequence (3.3.11). If A = , we replace S(A) and S(X ) by the corresponding augmented complexes; we then apply the Zig-Zag Lemma and we obtain the sequence (3.3.13). The exact sequence (3.3.12) is known as the long exact homology sequence of the pair (X, A); the sequence (3.3.13) is called the long exact reduced homology sequence of the pair (X, A).

100

3. ALGEBRAIC TOPOLOGY

3.3.19. E XAMPLE . If A = is homeomorphic to a star-shaped subset of I Rn , p (A) = 0 for every p Z (see Example 3.3.9); hence, the long exact then H reduced homology sequence of the pair (X, A) implies that the map: p (X ) Hp (X, A) q : H is an isomorphism for every p Z. Now, we want to show the homotopical invariance of the singular homology; more precisely, we want to show that two homotopic continuous maps induce the same homomorphisms of the homology groups. We begin with an algebraic denition. 3.3.20. D EFINITION . Let C = (Cp , p ) and C = (Cp , p ) be chain complexes. Given a chain map , : C C then a chain homotopy between and is a sequence (Dp )pZ of homomorphisms Dp : Cp Cp+1 such that p p = p+1 Dp + Dp1 p , for every p Z; in this case we write D : = and we say that and are chain-homotopic. (3.3.14) The following Lemma is a trivial consequence of formula (3.3.14) 3.3.21. L EMMA . If two chain maps and are chain-homotopic, then and induce the same homomorphisms in homology, i.e., = . Our next goal is to prove that if f and g are two homotopic continuous maps, then the chain maps f# and g# are chain-homotopic. To this aim, we consider the maps: (3.3.15) iX : X I X, jX : X I X dened by iX (x) = (0, x) and jX (x) = (1, x) for every x X , where I = [0, 1]. We will show rst that the chain maps (iX )# and (jX )# are chain-homotopic: 3.3.22. L EMMA . For all topological space X there exists a chain homotopy DX : (iX )# = (jX )# where iX and jX are given in (3.3.15); moreover, the association X DX may be chosen in a natural way, i.e., in such a way that, given a continuous map f : X Y , then the diagram Sp (X ) Sp+1 (I X ) ((Idf ) (f )
# p

(DX )p

(3.3.16)

# )p

Sp (Y ) Sp+1 (I Y )
(DY )p

commutes for every p Z, where Id f is given by (t, x) (t, f (x)). P ROOF. For each topological space X and each p Z we must dene a homomorphism (DX )p : Sp (X ) Sp+1 (I X );

3.3. SINGULAR HOMOLOGY GROUPS

101

for p < 0 we obviously set (DX )p = 0. For p 0 we denote by Idp the identity map of the space p ; then Idp is a singular p-simplex in p , and therefore Idp Sp (p ). The construction of DX must be such that the diagram (3.3.16) commutes, and this suggests the following denition: (3.3.17) (DX )p (T ) = ((Id T )# )p (Dp )p (Idp ), for every singular p-simplex T : p X (observe that T# (Idp ) = T ); hence, we need to nd the correct denition of (3.3.18) (Dp )p (Idp ) = ap Sp+1 (I p ), for each p 0. Keeping in mind the denition of chain homotopy (see (3.3.14)), our denition of ap will have to be given in such a way that the identity (3.3.19) ap = (ip )# (Idp ) (jp )# (Idp ) (Dp )p1 (Idp ) be satised for every p 0 (we will omit some index to simplify the notation); observe that (3.3.19) is equivalent to: (3.3.20) ap = ip jp (Dp )p1 (Idp ). Let us begin by nding a0 S1 (I 0 ) that satises (3.3.20), that is, a0 must satisfy a0 = i0 j0 ; we compute as follows: (i0 j0 ) = 0. 0 (I 0 ) = 0 (see Example 3.3.9) we see that it is indeed possible to Since H determine a0 with the required property. We now argue by induction; x r 1. Suppose that ap Sp+1 (I p ) has been found for p = 0, . . . , r 1 in such a way that condition (3.3.20) be satised, where (DX )p is dened in (3.3.17) for every topological space X ; it is then easy to see that the diagram (3.3.16) commutes. An easy computation that uses (3.3.16), (3.3.18) and (3.3.20) shows that: (3.3.21) (iX )#
p

(jX )#

= (DX )p + (DX )p1 ,

for p = 0, . . . , r 1. Now, we need to determine ar that satises (3.3.20) (with p = r). It follows from (3.3.21), where we set X = r and p = r 1, that: (3.3.22) (Dr )r1 (Idr ) = (ir )# (Idr ) (jr )# (Idr ) (Dr )r2 (Idr ) = (ir jr ); using (3.3.22) we see directly that (3.3.23) ir jr (Dr )r1 (Idr ) Zr (I r ). Since Hr (I r ) = 0 (see Example 3.3.9) it follows that (3.3.23) is an rboundary; hence it is possible to choose ar satisfying (3.3.20) (with p = r). This concludes the proof. It is now easy to prove the homotopical invariance of the singular homology.

102

3. ALGEBRAIC TOPOLOGY

3.3.23. P ROPOSITION . If two continuous maps f, g : X Y are homotopic, then the chain maps f# and g# are homotopic. P ROOF. Let H : f = g be a homotopy between f e g ; by Lemma 3.3.22 there exists a chain homotopy DX : iX = jX . Then, it is easy to see that we obtain a chain homotopy between f# and g# by considering, for each p Z, the homomorphism (H# )p+1 (DX )p : Sp (X ) Sp+1 (Y ). 3.3.24. C OROLLARY. If f, g : X Y are homotopic, then f = g . P ROOF. It follows from Proposition 3.3.23 and from Lemma 3.3.21. 3.3.1. The Hurewiczs homomorphism. In this subsection we will show that the rst singular homology group H1 (X ) of a topological space X can be computed from its fundamental group; more precisely, if X is arc-connected, we will show that H1 (X ) is the abelianized group of 1 (X ). In the entire subsection we will assume familiarity with the notations and the concepts introduced in Section 3.1; we will consider a xed topological space X . Observing that the unit interval I = [0, 1] coincides with the rst standard simplex 1 , we see that every curve (X ) is a singular 1-simplex in X ; then, S1 (X ). We will say that two singular 1-chains c, d S1 (X ) are homologous when c d B1 (X ); this terminology will be used also in the case that c and d are not necessarily cycles.4 We begin with some Lemmas: 3.3.25. L EMMA . Let (X ) and let : I I be a continuous map. If (0) = 0 and (1) = 1, then is homologous to ; if (0) = 1 and (1) = 0, then is homologous to . P ROOF. We suppose rst that (0) = 0 and (1) = 1. Consider the singular 1-simplexes and (0, 1) in I (recall the denition of in (3.3.1)). Clearly, (0, 1) = 0, i.e., (0, 1) Z1 (I ); since H1 (I ) = 0 (see Example 3.3.9) it follows that (0, 1) B1 (I ). Consider the chain map # : S(I ) S(X ); we have that # ( (0, 1)) B1 (X ). But # (0, 1) = B1 (X ), from which it follows that is homologous to . The case (0) = 1, (1) = 0 is proven analogously, observing that + (0, 1) Z1 (I ). 3.3.26. R EMARK . In some situations we will consider singular 1-chains given by curves : [a, b] X that are dened on an arbitrary closed interval [a, b] (rather than on the unit interval I ); in this case, with a slight abuse, we will denote by S1 (X ) the singular 1-simplex (a, b) : I X ; it follows from Lemma 3.3.25 that (a, b) is homologous to any reparameterization of , where : I [a, b] is a continuous map such that (0) = a and (1) = b (see also Remark 3.1.4).
4Observe that a singular 1-chain c denes a homology class in H (X ) only if c Z (X ). 1 1

3.3. SINGULAR HOMOLOGY GROUPS

103

3.3.27. L EMMA . If , (X ) are such that (1) = (0), then is homologous to + ; moreover, for every (X ) we have that 1 is homologous to and for every x0 X , ox0 is homologous to zero. P ROOF. We will basically use the same idea that was used in the proof of 1 1 Lemma 3.3.25. We have that 0, 2 + 2 , 1 (0, 1) Z1 (I ) = B1 (I ); considering the chain map ( )# we obtain:
1 ( )# (0, 1 2 ) + ( 2 , 1) (0, 1) = + B1 (X ),

from which it follows that is homologous to + . The fact that 1 is homologous to follows from Lemma 3.3.25; nally, if T : 2 X denotes the constant map with value x0 , we obtain T = ox0 B1 (X ). 3.3.28. L EMMA . Let K : I I X be a continuous map; considering the curves: 1 = K 3 = K (0, 0), (1, 0) , (1, 1), (0, 1) , 2 = K 4 = K (1, 0), (1, 1) , (0, 1), (0, 0) ,

we have that the singular 1-chain 1 + 2 + 3 + 4 is homologous to zero. P ROOF. We have that H1 (I I ) = 0 (see Example 3.3.9); moreover (3.3.24) (0, 0), (1, 0) + + (1, 0), (1, 1) + (1, 1), (0, 1) (0, 1), (0, 0) Z1 (I I ) = B1 (I I ).

The conclusion follows by applying K# to (3.3.24). We now relate the homotopy class and the homology class of a curve (X ). 3.3.29. C OROLLARY. If , (X ) are homotopic with xed endpoints, then is homologous to . P ROOF. It sufces to apply Lemma 3.3.28 to a homotopy with xed endpoints K: = , keeping in mind Lemma 3.3.27. 3.3.30. R EMARK . Let A X be a subset; if : I X is a continuous curve with endpoints in A, i.e., (0), (1) A, then S0 (A), and therefore Z1 (X, A) denes a homology class + B1 (X, A) in H1 (X, A). It follows from Lemma 3.3.28 (keeping in mind also Lemma 3.3.27) that if and are homotopic with free endpoints in A (recall Denition 3.1.25) then and dene the same homology class in H1 (X, A). 3.3.31. R EMARK . If , are freely homotopic loops in X (see Remark 3.1.16) then it follows easily from Lemma 3.3.28 (keeping in mind also Lemma 3.3.27) that is homologous to . We dene a map: (3.3.25) : (X ) S1 (X )/B1 (X )

104

3. ALGEBRAIC TOPOLOGY

by setting ([ ]) = + B1 (X ) for every (X ); it follows from Corollary 3.3.29 that is well dened, i.e., it does not depend on the choice of the representative of the homotopy class [ ] (X ). Then, Lemma 3.3.27 tells us that: (3.3.26) ([ ] []) = ([ ]) + ([]), [ ]1 = ([ ]), ([ox0 ]) = 0, for every , (X ) with (1) = (0) and for every x0 X . If (X ) is a loop, then Z1 (X ); if we x x0 X , we see that restricts to a map (also denoted by ): (3.3.27) : 1 (X, x0 ) H1 (X ). It follows from (3.3.26) that (3.3.27) is a group homomorphism; this homomorphism is known as the Hurewiczs homomorphism. The Hurewiczs homomorphism is natural in the sense that, given a continuous map f : X Y with f (x0 ) = y0 , the following diagram commutes: 1 (X, x0 ) H1 (X ) f f

1 (Y, y0 ) H1 (Y )

If : I X is a continuous curve joining x0 and x1 , then the Hurewiczs homomorphism ts well together with the isomorphism # between the fundamental groups 1 (X, x0 ) and 1 (X, x1 ) (see Proposition 3.1.11); more precisely, it follows from (3.3.26) that we have a commutative diagram: (3.3.28) 1 (X, x0 )
LLL LLL LLL L& 8 rrr r r rr rrr 

H1 (X )

1 (X, x1 ). We are now ready to prove the main result of this subsection. We will rst recall some denitions in group theory. 3.3.32. D EFINITION . If G is a group, the commutator subgroup of G, denoted by G , is the subgroup of G generated by all the elements of the form ghg 1 h1 , with g, h G. The commutator subgroup G is always a normal subgroup5 of G, and therefore the quotient G/G is always a group. We say that G/G is the abelianized group of G. The group G/G is always abelian; as a matter of facts, if H is a normal subgroup of G, then the quotient group G/H is abelian if and only if H G .
5actually, the commutator subgroup G of G is invariant by every automorphism of G.

3.3. SINGULAR HOMOLOGY GROUPS

105

3.3.33. T HEOREM . Let X be an arc-connected topological space. Then, for every x0 X , the Hurewiczs homomorphism (3.3.27) is surjective, and its kernel is the commutator subgroup of 1 (X, x0 ); in particular, the rst singular homology group H1 (X ) is isomorphic to the abelianized group of 1 (X, x0 ). P ROOF. Since the quotient 1 (X, x0 )/Ker() = Im() is abelian, it follows that Ker() contains the commutator subgroup 1 (X, x0 ) , and therefore denes a homomorphism by passage to the quotient: : 1 (X, x0 )/1 (X, x0 ) H1 (X ); our strategy will be to show that is an isomorphism. For each x X , choose a curve x (X ) such that x (0) = x0 and x (1) = x; we are now going to dene a homomorphism : S1 (X ) 1 (X, x0 )/1 (X, x0 ) ; since 1 (X, x0 )/1 (X, x0 ) is abelian and the singular 1-simplexes of X form a basis of S1 (X ) as a free abelian group, is well dened if we set ( ) = q [ (0) ] [ ] [ (1) ]1 , where q denotes the quotient map q : 1 (X, x0 ) 1 (X, x0 )/1 (X, x0 ) . We are now going to show that B1 (X ) is contained in the kernel of ; to this aim, it sufces to show that (T ) is the neutral element of 1 (X, x0 )/1 (X, x0 ) for every singular 2-simplex T in X . We write: (3.3.29) T = 0 1 + 2 , where 0 = T (e1 , e2 ), 1 = T (e0 , e2 ) and 2 = T (e0 , e1 ). Applying to both sides of (3.3.29) we obtain: (3.3.30) (T ) = (0 )(1 )1 (2 ) = q [T (e1 ) ] [0 ] [1 ]1 [2 ] [T (e1 ) ]1 . (X ),

Writing [] = [ (e1 , e2 )] [ (e2 , e0 )] [ (e0 , e1 )] (2 ) then (3.3.30) implies that: (T ) = q [T (e1 ) ] T ([]) [T (e1 ) ]1 ; since [] 1 (2 , e1 ), we have that [] = [oe1 ] (see Example 3.1.15), from which it follows (T ) = q ([ox0 ]). Then, we conclude that B1 (X ) Ker(), from which we deduce that passes to the quotient and denes a homomorphism : S1 (X )/B1 (X ) 1 (X, x0 )/1 (X, x0 ) . The strategy will now be to show that the restriction |H1 (X ) is an inverse for . Let us compute ; for (X ) we have: (3.3.31) ( )( ) = ([ (0) ]) + ([ ]) ([ (1) ]) = (0) + (1) + B1 (X ).

106

3. ALGEBRAIC TOPOLOGY

Dene a homomorphism : S0 (X ) S1 (X ) by setting (x) = x for every singular 0-simplex x X ; then (3.3.31) implies that: (3.3.32) = p (Id ),

where p : S1 (X ) S1 (X )/B1 (X ) denotes the quotient map and Id denotes the identity map of S1 (X ). If we restricts both sides of (3.3.32) to Z1 (X ) and passing to the quotient we obtain: |H1 (X ) = Id. Let us now compute ; for every loop x0 (X ) we have: ( ) q ([ ]) = ( ) = q ([x0 ])q ([ ])q ([x0 ])1 = q ([ ]), observing that 1 (X, x0 )/1 (X, x0 ) is abelian. It follows that: |H1 (X ) = Id, which concludes the proof. 3.3.34. R EMARK . If X is arc-connected and 1 (X ) is abelian, it follows from Theorem 3.3.33 that the Hurewiczs homomorphism is an isomorphism of 1 (X, x0 ) onto H1 (X ); this fact explains why the fundamental groups with different basepoints 1 (X, x0 ) and 1 (X, x1 ) can be canonically identied when the fundamental group of the space is abelian. The reader should compare this observation with Remark 3.1.13 and with the diagram (3.3.28). Exercises for Chapter 3 E XERCISE 3.1. Prove that every contractible space is arc-connected. E XERCISE 3.2. Prove that a topological space X which is connected and locally arc-connected is arc-connected. Deduce that a connected (topological) manifold is arc-connected. E XERCISE 3.3. Let (X ) be a loop and let (X ) be such that (0) = (0); show that the loops and 1 are freely homotopic. E XERCISE 3.4. Let f, g : X Y be homotopic maps and let H : f = g be a homotopy from f to g ; x x0 X and set (s) = Hs (x0 ), s I . Show that the following diagram commutes: 1 (Y, f (x0 )) o7 f oooo
=

ooo ooo

1 (X, x0 )

OOO OOO O g OOOO '

1 (Y, g (x0 ))

EXERCISES FOR CHAPTER 3

107

E XERCISE 3.5. A continuous map f : X Y is said to be a homotopy equivalence if there exists a continuous map g : Y X such that g f is homotopic to the identity map of X and f g is homotopic to the identity map of Y ; in this case we say that g is a homotopy inverse for f . Show that if f is a homotopy equivalence then f : 1 (X, x0 ) 1 (Y, f (x0 )) is an isomorphism for every x0 X . E XERCISE 3.6. Show that X is contractible if and only if the map f : X {x0 } is a homotopy equivalence. E XERCISE 3.7. Prove that a homotopy equivalence induces an isomorphism in singular homology. Conclude that, if X is contractible, then H0 (X ) = Z and Hp (X ) = 0 for every p 1. E XERCISE 3.8. If Y X , a continuous map r : X Y is said to be a retraction if r restricts to the identity map of Y ; in this case we say that Y is a retract of X . Show that if r is a retraction then r : 1 (X, x0 ) 1 (Y, x0 ) is surjective for every x0 Y . Show also that if Y is a retract of X then the inclusion map i : Y X induces an injective homomorphism i : 1 (Y, x0 ) 1 (X, x0 ) for every x0 Y . E XERCISE 3.9. Let G1 , G2 be groups and f : G1 G2 a homomorphism. Prove that the sequence 0 G1 G2 0 is exact if and only if f is an isomorphism. E XERCISE 3.10. Let p : E B be a locally injective continuous map with E Hausdorff and let f : X B be a continuous map dened in a connected topological space X . Given x0 X , e0 E , show that there exists at most one : X E with p f = f and f (x0 ) = e0 . Show that if p is a covering map map f then the hypothesis that E is Hausdorff can be dropped. E XERCISE 3.11. Let X I R2 be dened by: X = (x, sin(1/x)) : x > 0 {0} [1, 1] . Show that X is connected but not arc-connected; compute the singular homology groups of X . E XERCISE 3.12. Prove the Zig-Zag Lemma (Lemma 3.3.17). E XERCISE 3.13. Let G be a group and let G act on a topological space X by homeomorphisms. We say that such action is properly discontinuous if for every x X there exists a neighborhood U of x such that gU U = for every g = 1, where gU = {g y : y U }. Let be given a properly discontinuous action of G in X and denote by X/G the set of orbits of G endowed with the quotient topology. Show that the quotient map p : X X/G is a covering map with typical ber G. Show that, if X is arc-connected, there exists an exact sequence of groups and group homomorphisms: 0 1 (X ) 1 (X/G) G 0.
p f

108

3. ALGEBRAIC TOPOLOGY

If X is simply connected conclude that 1 (X/G) is isomorphic to G. E XERCISE 3.14. Let X = I R2 be the Euclidean plane; for each m, n Z let gm,n be the homeomorphism of X given by: gm,n (x, y ) = ((1)n x + m, y + n). Set G = {gm,n : m, n Z}. Show that: G is a subgroup of the group of all homeomorphisms of X ; show that X/G is homeomorphic to the Klein bottle; show that the natural action of G in X is properly discontinuous; conclude that the fundamental group of the Klein bottle is isomorphic to G; show that G is the semi-direct product6 of two copies of Z; compute the commutator subgroup of G and conclude that the rst singular homology group of the Klein bottle is isomorphic to Z (Z/2Z). E XERCISE 3.15. Prove that if X and Y are arc-connected, then H1 (X Y ) = H1 (X ) H1 (Y ). E XERCISE 3.16. Compute the relative homology group H2 (D, D), where D is the unit disk {(x, y ) I R2 : x2 + y 2 1} and D is its boundary.

6Recall that a group G is the (inner) semi-direct product of two subgroups H and K if G = HK

with H K = {1} and K normal in G.

CHAPTER 4

The Maslov Index


4.1. Index of a Symmetric Bilinear Form In this section we will dene the index and the co-index of a symmetric bilinear form; in nite dimension, these numbers are respectively the number of negative and of positive entries of a symmetric matrix when it is diagonalized as in the Sylvester Inertia Theorem (Theorem 4.1.10). We will show some properties of these numbers. In this Section, V will always denote a real vector space, not necessarily nite dimensional. Recall that Bsym (V ) denotes the space of symmetric bilinear forms B :V V I R. We start with a denition: 4.1.1. D EFINITION . Let B Bsym (V ); we say that B is: positive denite if B (v, v ) > 0 for all v V , v = 0; positive semi-denite if B (v, v ) 0 for all v V ; negative denite if B (v, v ) < 0 for all v V , v = 0; negative semi-denite if B (v, v ) 0 for all v V . We say that a subspace W V is positive with respect to B , or B -positive, if B |W W is positive denite; similarly, we say that W is negative with respect to B , or B -negative, if B |W W is negative denite. The index of B , denoted by n (B ), is dened by: (4.1.1) n (B ) = sup dim(W ) : W is a B -negative subspace of V . The index of B can be a non negative integer, or +. The co-index of B , denoted by n+ (B ), is dened as the index of B : n+ (B ) = n (B ). Obviously, the co-index of B can be dened as the supremum of the dimensions of all B -positive subspaces of V . When at least one of the numbers n (B ) and n+ (B ) is nite, we dene the signature of B by: sgn(B ) = n+ (B ) n (B ). If B Bsym (V ) and W V is a subspace, then clearly: (4.1.2) n (B |W W ) n (B ), n+ (B |W W ) n+ (B ). The reader should now recall the denitions of kernel of a symmetric bilinear form B , denoted by Ker(B ), and of orthogonal complement of a subspace S V with respect to B , denoted by S . Recall also that B is said to be nondegenerate if Ker(B ) = {0}.
109

110

4. THE MASLOV INDEX

Observe that in Section 1.1 we have considered only nite dimensional vector spaces, but obviously the denitions of kernel, orthogonal complement and nondegeneracy make sense for symmetric bilinear forms dened on an arbitrary vector space V . However, many results proven in Section 1.1 make an essential use of the niteness of the vector space (see Example 1.1.12). For instance, observe that a bilinear form is nondegenerate if and only if its associated linear operator (4.1.3) V v B (v, ) V is injective; if dim(V ) = +, this does not imply that (4.1.3) is an isomorphism. 4.1.2. D EFINITION . Given B Bsym (V ), the degeneracy of B , denoted by dgn(B ) is the possibly innite dimension of Ker(B ). We say that a subspace W V is nondegenerate with respect to B , or also that W is B -nondegenerate, if B |W W is nondegenerate. 4.1.3. E XAMPLE . Unlike the case of the index and the co-index (see (4.1.2)), the degeneracy of a symmetric bilinear form B is not monotonic with respect to the inclusion of subspaces. For instance, if V = I R2 and B is the symmetric bilinear form: (4.1.4) B (x1 , y1 ), (x2 , y2 ) = x1 x2 y1 y2 then dgn(B ) = 0; however, if W is the subspace generated by the vector (1, 1), we have: dgn(B |W W ) = 1 > 0 = dgn(B ). On the other hand, if B is dened by B (x1 , y1 ), (x2 , y2 ) = x1 x2 and if W is the subspaces generated by (1, 0), then dgn(B |W W ) = 0 < 1 = dgn(B ). 4.1.4. E XAMPLE . If T : V1 V2 is an isomorphism and if B Bsym (V1 ), then we can consider the push-forward of B , T# (B ) Bsym (V2 ). Clearly, T maps B -positive subspaces of V1 into T# (B )-positive subspaces of V2 , and B -negative subspaces of V1 into T# (B )-negative subspaces of V2 ; moreover, Ker(T# (B )) = T (Ker(B )). Hence we have: n+ T# (B ) = n+ (B ), n T# (B ) = n (B ), dgn T# (B ) = dgn(B ).

4.1.5. R EMARK . It follows from Proposition 1.1.10 and from remark 1.1.13 that if W V is a nite dimensional B -nondegenerate subspace, then V = W W , even in the case that dim(V ) = +. Recall that if W V is a subspace, then the codimension of W in V is dened by: codimV (W ) = dim(V /W ); this number may be nite even when dim(W ) = dim(V ) = +. The codimension of W in V coincides with the dimension of any complementary subspace of W in V .

4.1. INDEX OF A SYMMETRIC BILINEAR FORM

111

The following Lemma and its Corollary are the basic tool for the computation of indices of bilinear forms: 4.1.6. L EMMA . Let B Bsym (V ); if Z V is a subspace of V on which B is positive semi-denite, then: n (B ) codimV (Z ). P ROOF. If B is negative denite on a subspace W , then W Z = {0}, and so the quotient map q : V V /Z takes W isomorphically onto a subspace of V /Z . Hence, dim(W ) codimV (Z ). 4.1.7. C OROLLARY. Suppose that V = Z W with B positive semi-denite on Z and negative denite on W ; then n (B ) = dim(W ). P ROOF. Clearly, n (B ) dim(W ). From Lemma 4.1.6 it follows that n (B ) codimV (Z ) = dim(W ). 4.1.8. R EMARK . Note that every result concerning the index of symmetric bilinear forms, like for instance Lemma 4.1.6 and Corollary 4.1.7, admits a corresponding version for the co-index of forms. For shortness, we will only state these results in the version for the index, and we will understand the version for the co-index. Similarly, results concerning negative (semi-)denite symmetric bilinear forms B can be translated into results for positive (semi-)denite symmetric forms by replacing B with B . 4.1.9. P ROPOSITION . If B Bsym (V ) and V = Z W with B positive denite in Z and negative denite in W , then B is nondegenerate. P ROOF. Let v Ker(B ); write v = v+ + v with v+ Z and v W . Then: (4.1.5) (4.1.6) B (v, v+ ) = B (v+ , v+ ) + B (v , v+ ) = 0, B (v, v ) = B (v+ , v ) + B (v , v ) = 0;

from (4.1.5) we get that B (v+ , v ) 0, and from (4.1.6) we get B (v+ , v ) 0, from which it follows B (v+ , v ) = 0. Then, (4.1.5) implies v+ = 0 and (4.1.6) implies v = 0. 4.1.10. T HEOREM (Sylvesters Inertia Theorem). Suppose dim(V ) = n < + and let B Bsym (V ); then, there exists a basis of V with respect to which the matrix form of B is given by: Ip 0pq 0pr 0qr , (4.1.7) B 0qp Iq 0r p 0r q 0r where 0 , 0 and I denote respectively the zero matrix, the zero matrix and the identity matrix. The numbers p, q and r are uniquely determined by the bilinear form B ; we have: (4.1.8) n+ (B ) = p, n (B ) = q, dgn(B ) = r.

112

4. THE MASLOV INDEX

P ROOF. The existence of a basis (bi )n i=1 with respect to which B has the canonical form (4.1.7) follows from Theorem 1.1.14, after suitable rescaling of the vectors of the basis. To prove that p, q and r are uniquely determined by B , i.e., that they do not depend on the choice of the basis, it is actually enough to prove (4.1.8). n To this aim, let Z be the subspace generated by the vectors {bi }p i=1 {bi }i=p+q +1 +q and W the subspace generated by {bi }p i=p+1 ; then V = Z W , B is positive semi-denite in Z and negative denite in W . It follows from Corollary 4.1.7 that n (B ) = dim(W ) = q . Similarly, we get n+ (B ) = p. It is easy to see that Ker(B ) is generated by the vectors {bi }n i=p+q +1 and we conclude that dgn(B ) = r. 4.1.11. C OROLLARY. Let B Bsym (V ), with dim(V ) < +. If g is an inner product in V and if T Lin(V ) is such that B = g (T , ), then the index (resp., the co-index) of B is equal to the sum of the multiplicities of the negative (resp., the positive) eigenvalues of T ; the degeneracy of B is equal to the multiplicity of the zero eigenvalue of T . P ROOF. Since T is g -symmetric, there exists a g -orthonormal basis that diagonalizes T , and this diagonal matrix has in its diagonal entries the eigenvalues of T repeated with multiplicity. In such basis, the bilinear form B is represented by the same matrix. After suitable rescaling of the vectors of the basis, B will be given in the canonical form (4.1.7); this operation does not change the signs of the elements in the diagonal of the matrix that represents B . The conclusion now follows from Theorem 4.1.10 4.1.12. E XAMPLE . The conclusion of Corollary 4.1.11 holds in the more general case of a matrix T that represents B in any basis; indeed, observe that any basis is orthonormal with respect to some inner product of V . Recall that the determinant and the trace of a matrix are equal respectively to the product and the sum of its eigenvalues (repeated with multiplicity); in the case dim(V ) = 2 it follows that the determinant and the trace of a matrix that represents B in any basis determine uniquely the numbers n (B ), n+ (B ) and dgn(B ). 4.1.13. L EMMA . Suppose that B Bsym (V ) is positive semi-denite; then Ker(B ) = {v V : B (v, v ) = 0}. P ROOF. Let v V with B (v, v ) = 0 and let w V be arbitrary; we need to show that B (v, w) = 0. If v and w are linearly dependent, the conclusion is trivial; otherwise, v and w form the basis of a two-dimensional subspace of V in which the restriction of B is represented by the matrix: (4.1.9) B (v, v ) B (v, w) B (v, w) B (w, w) .

It follows from Corollary 4.1.11 (see Example 4.1.12) that the determinant of (4.1.9) is non negative, that is: B (v, w)2 B (v, v )B (w, w) = 0, which concludes the proof.

4.1. INDEX OF A SYMMETRIC BILINEAR FORM

113

4.1.14. C OROLLARY. If B Bsym (V ) is positive semi-denite and nondegenerate, then B is positive denite. We now prove a generalized version of the CauchySchwarz inequality for symmetric bilinear forms: 4.1.15. P ROPOSITION . Let B Bsym (V ) and vectors v, w V be given. We have: if v , w are linearly dependent or if v , w generate a B -degenerate twodimensional subspace, then B (v, w)2 = B (v, v )B (w, w); if v , w generate a B -positive or B -negative two-dimensional subspace, then B (v, w)2 < B (v, v )B (w, w); if v , w generate a two-dimensional subspace where B has index equal to 1, then B (v, w)2 > B (v, v )B (w, w); the above possibilities are exhaustive and mutually exclusive. P ROOF. The case that v and w are linearly dependent is trivial; all the others follow directly from Corollary 4.1.11 (see also Example 4.1.12), keeping in mind that the matrix that represents the restriction of B to the subspace generated by v and w is given by (4.1.9). 4.1.16. D EFINITION . Given B Bsym (V ), we say that two subspaces V1 and V2 of V are orthogonal with respect to B , or B -orthogonal, if B (v1 , v2 ) = 0 for all v1 V1 and all v2 V2 ; a direct sum V = V1 V2 with V1 and V2 B -orthogonal will be called a B -orthogonal decomposition of V . 4.1.17. L EMMA . Let B Bsym (V ); if V = V1 V2 is a B -orthogonal decomposition of V and if B is negative denite (resp., negative semi-denite) in V1 and in V2 , then B is negative denite (resp., negative semi-denite) in V . P ROOF. It is obtained from the following simple computation: B (v1 + v2 , v1 + v2 ) = B (v1 , v1 ) + B (v2 , v2 ), v1 V 1 , v2 V 2 . 4.1.18. D EFINITION . Given B Bsym (V ), we say that a subspace W V is maximal negative with respect to B if W is B -negative and if it is not properly contained in any other B -negative subspace of V . Similarly, we say that W V is maximal positive with respect to B if W is B -positive and if it is not properly contained in any other B -positive subspace of V . 4.1.19. C OROLLARY. Let B Bsym (V ) and W V be a maximal negative subspace with respect to B . Then, if Z V is a subspace which is B -orthogonal to W , it follows that B is positive semi-denite in Z . P ROOF. By Lemma 4.1.17, the sum of any non zero B -negative subspace of Z with W would be a B -negative subspace of V that contains properly W . The conclusion follows.

114

4. THE MASLOV INDEX

Observe that Corollary 4.1.19 can be applied when n (B ) < + and W is a B -negative subspace with dim(W ) = n (B ). 4.1.20. C OROLLARY. Given B Bsym (V ), then dim(V ) = n+ (B ) + n (B ) + dgn(B ). P ROOF. If either one of the numbers n+ (B ) or n (B ) is innite, the result is trivial. Suppose then that both numbers are nite; let W V be a B -negative subspace with dim(W ) = n (B ) and let Z V be a B -positive subspace with dim(Z ) = n+ (B ). By Proposition 4.1.9 we have that B is nondegenerate in Z W , and it follows from Remark 4.1.5 that V = Z W (Z W ) . By Corollary 4.1.19, we have that B is positive semi-denite and also negative semi-denite in (Z W ) , hence B vanishes in (Z W ) . It follows now that Ker(B ) = (Z W ) , which concludes the proof. 4.1.21. C OROLLARY. If W V is a maximal negative subspace with respect to B Bsym (V ), then n (B ) = dim(W ). P ROOF. If dim(W ) = + the result is trivial; for the general case, it follows from Remark 4.1.5 that V = W W . By Corollary 4.1.19, B is positive semidenite in W , and then the conclusion follows from Corollary 4.1.7. 4.1.22. R EMARK . We can now conclude that the supremum that appears in the denition of index in (4.1.1) is in facts a maximum, i.e., there always exists a B -negative subspace W V with n (B ) = dim(W ). If n (B ) is nite, this statement is trivial. If n (B ) = +, it follows from Corollary 4.1.21 that no nite-dimensional subspace of V is maximal B -negative. If there were no innitedimensional B -negative subspace of V , we could construct a strictly increasing sequence W1 W2 of B -negative subspaces; then W = n1 Wn would be an innite-dimensional B -negative subspace, in contradiction with the hypothesis. As a matter of facts, it follows from Zorns Lemma that every symmetric bilinear form admits a maximal negative subspace (see Exercise 4.1). 4.1.23. P ROPOSITION . Let B Bsym (V ); if V = V1 V2 is a B -orthogonal decomposition, then: (4.1.10) (4.1.11) (4.1.12) n+ (B ) = n+ B |V1 V1 + n+ B |V2 V2 , n (B ) = n B |V1 V1 + n B |V2 V2 , dgn(B ) = dgn B |V1 V1 + dgn B |V2 V2 . Ker(B ) = Ker B |V1 V1 Ker B |V2 V2 . Let us prove (4.1.11). If B has innite index in V1 or in V2 the result is trivial; suppose then that these indices are nite. Let Wi Vi be a B -negative subspace

P ROOF. The identity (4.1.12) follows from

4.1. INDEX OF A SYMMETRIC BILINEAR FORM

115

with n B |Vi Vi = dim(Wi ), i = 1, 2. By Remark 4.1.5 we can nd a B orthogonal decomposition Vi = Zi Wi ; it follows from Corollary 4.1.19 that B must be positive semi-denite in Zi . Then: V = (W1 W2 ) (Z1 Z2 ), where, by Lemma 4.1.17, B is negative denite in W1 W2 and positive semidenite in Z1 Z2 . The identity (4.1.11) now follows from Corollary 4.1.7; the identity (4.1.10) follows by replacing B with B . 4.1.24. C OROLLARY. Let B Bsym (V ) and let N Ker(B ); if W V is any complementary subspace to N then the following identities hold: (4.1.13) n+ (B ) = n+ B |W W , n (B ) = n B |W W , dgn(B ) = dgn B |W W + dim(N );

if N = Ker(B ) then B is nondegenerate in W . P ROOF. The identities (4.1.13) follow immediately from Proposition 4.1.23, because V = W N is a B -orthogonal decomposition. If N = Ker(B ), the nondegeneracy of B in W is obvious. 4.1.25. R EMARK . If N is a subspace of Ker(B ) then we can dene by passing to the quotient a symmetric bilinear form B Bsym (V /N ): B (v1 + N, v2 + N ) = B (v1 , v2 ), v1 , v2 V. If W V is any subspace complementary to N , we have an isomorphism q : W V /N obtained by restriction of the quotient map; moreover, B is the pushforward of B |W W by q . It follows from Corollary 4.1.24 (see also Example 4.1.4) that n+ (B ) = n+ B , n (B ) = n B , dgn(B ) = dgn B + dim(N ); if N = Ker(B ) then it follows also that B is nondegenerate. 4.1.26. E XAMPLE . Lemma 4.1.17 does not hold if the subspaces V1 and V2 are not B -orthogonal. For instance, if V = I R2 and if we consider the symmetric bilinear form B given in (4.1.4), then n (B ) = n+ (B ) = 1, but we can write I R2 as the direct sum of the subspaces generated respectively by v1 = (0, 1) and v2 = (1, 2), that are both B -negative. In the next proposition we generalize the result of Lemma 4.1.17 by showing that if V = V1 V2 , where V1 and V2 are B -negative subspaces such that the product of elements of V1 with elements of V2 is relatively small with respect to their lengths, then V is B -negative. 4.1.27. P ROPOSITION . Let B Bsym (V ) and assume that V is written as the direct sum of B -negative subspaces V = V1 V2 ; if for all v1 V1 and v2 V2 , with v1 , v2 = 0, it is (4.1.14) B (v1 , v2 )2 < B (v1 , v1 )B (v2 , v2 ) then B is negative denite in V .

116

4. THE MASLOV INDEX

P ROOF. Let v V be non zero and write v = v1 + v2 , with v1 V1 and v2 V2 . We need to show that B (v, v ) < 0, and clearly it sufces to consider the case that both v1 and v2 are non zero. In this case, the hypothesis (4.1.14) together with Proposition 4.1.15 imply that the two-dimensional subspace generated by v1 and v2 is B -negative, which concludes the proof. 4.1.28. R EMARK . It can also be shown a version of Proposition 4.1.27 assuming only that B is negative semi-denite in V1 and in V2 , and that (4.1.15) B (v1 , v2 )2 B (v1 , v1 )B (v2 , v2 ), for all v1 V1 , v2 V2 . In this case, the conclusion is that B is negative semidenite in V (see Exercise 4.2). 4.1.1. The evolution of the index of a one-parameter family of symmetric bilinear forms. In this subsection we will study the evolution of the function n (B (t)), where t B (t) is a one parameter family of symmetric bilinear forms on a space V . We make the convention that in this subsection V will always denote a nite dimensional real vector space: dim(V ) < +. We choose an arbitrary norm in V denoted by bilinear form B B(V ) by setting:
v 1 w 1

; we then dene the norm of a

B = sup |B (v, w)|. Observe that, since V and B(V ) are nite dimensional, then any norm in these spaces induces the same topology. We will rst show that the condition n (B ) k (for some xed k ) is an open condition. 4.1.29. L EMMA . Let k 0 be xed; the set of symmetric bilinear forms B Bsym (V ) such that n (B ) k is open in Bsym (V ). P ROOF. Let B Bsym (V ) with n (B ) k ; then, there exists a k -dimensional B -negative subspace W V . Since the unit sphere of W is compact, we have: sup B (v, v ) = c < 0;
v W v =1

it now follows directly that if A Bsym (V ) and A B negative denite in W , and therefore n (A) k .

< |c|/2 then A is

4.1.30. C OROLLARY. Let k 0 be xed; the set of nondegenerate symmetric bilinear forms B Bsym (V ) such that n (B ) = k is open in Bsym (V ). P ROOF. If B Bsym (V ) is nondegenerate and n (B ) = k , then n+ (B ) = dim(V ) k (see Corollary 4.1.20); by Lemma 4.1.29, for A in a neighborhood of B in Bsym (V ) we have n (A) k and n+ (A) dim(V ) k , from which we get n (A) = k and dgn(A) = 0.

4.1. INDEX OF A SYMMETRIC BILINEAR FORM

117

4.1.31. C OROLLARY. Let t B (t) be a continuous curve in Bsym (V ) dened in some interval I I R; if B (t) is nondegenerate for all t I , then n (B (t)) and n+ (B (t)) are constant in I . P ROOF. By Corollary 4.1.30, the set of instants t I such that n (B (t)) = k is open in I for each k = 0, . . . , dim(V ) xed. The conclusion follows from the connectedness of I . Corollary 4.1.31 tells us that the index n (B (t)) and the co-index n+ (B (t)) can only change when B (t) becomes degenerate; in the next Theorem we show how to compute this change when t B (t) is of class C 1 : 4.1.32. T HEOREM . Let B : [t0 , t1 [ Bsym (V ) be a curve of class C 1 ; write N = Ker B (t0 ) . Suppose that the bilinear form B (t0 )|N N is nondegenerate; then there exists > 0 such that for t ]t0 , t0 + [ the bilinear form B (t) is nondegenerate, and the following identities hold: n+ (B (t)) = n+ (B (t0 )) + n+ B (t0 )|N N , n (B (t)) = n (B (t0 )) + n B (t0 )|N N . The proof of Theorem 4.1.32 will follow easily from the following: 4.1.33. L EMMA . Let B : [t0 , t1 [ Bsym (V ) be a curve of class C 1 ; write N = Ker(B (t0 )). If B (t0 ) is positive semi-denite and B (t0 )|N N is positive denite, then there exists > 0 such that B (t) is positive denite for t ]t0 , t0 + [. P ROOF. Let W V be a subspace complementary to N ; it follows from Corollary 4.1.24 that B (t0 ) is nondegenerate in W , and from Corollary 4.1.14 that B (t0 ) is positive denite in W . Choose any norm in V ; since the unit sphere of W is compact, we have: (4.1.16)
w W w =1

inf B (t0 )(w, w) = c0 > 0;

similarly, since B (t0 ) is positive denite in N we have: (4.1.17)


nN n =1

inf B (t0 )(n, n) = c1 > 0.

Since B is continuous, there exists > 0 such that c0 B (t) B (t0 ) , t [t0 , t0 + [ , 2 and it follows from (4.1.16) that: c0 (4.1.18) inf B (t)(w, w) > 0, t [t0 , t0 + [ . w W 2
w =1

Since B is differentiable at t0 we can write: (4.1.19) B (t) = B (t0 ) + (t t0 )B (t0 ) + r(t), with lim
tt0

r(t) = 0, t t0

118

4. THE MASLOV INDEX

and then, by possibly choosing a smaller > 0, we get: c1 (4.1.20) r(t) (t t0 ), t [t0 , t0 + [ ; 2 from (4.1.17), (4.1.19) and (4.1.20) it follows: c1 (4.1.21) inf B (t)(n, n) (t t0 ), t ]t0 , t0 + [ . nN 2
n =1

From (4.1.18) and (4.1.21) it follows that B (t) is positive denite in W and in N 1 for t ]t0 , t0 + [; taking c3 = B (t0 ) + c2 we obtain from (4.1.19) and (4.1.20) that: (4.1.22) B (t)(w, n) (t t0 )c3 , t [t0 , t0 + [ , provided that w W , n N and w = n = 1. By possibly taking a smaller > 0, putting together (4.1.18), (4.1.21) and (4.1.22) we obtain: c0 c1 (t t0 ) B (t)(w, n)2 (t t0 )2 c2 3 < 4 (4.1.23) B (t)(w, w) B (t)(v, v ), t ]t0 , t0 + [ , for all w W , n N with w = n = 1; but (4.1.23) implies: B (t)(w, n)2 < B (t)(w, w) B (t)(n, n), t ]t0 , t0 + [ , for all w W , n N non zero. The conclusion follows now from Proposition 4.1.27. P ROOF OF T HEOREM 4.1.32. By Theorem 4.1.10 there exists a decomposition V = V+ V N where V+ and V are respectively a B (t0 )-positive and a B (t0 )-negative subspace; similarly, we can write N = N+ N where N+ is a B (t0 )-positive and N is a B (t0 )-negative subspace. Obviously: n+ (B (t0 )) = dim(V+ ), n (B (t0 )) = dim(V ), n B (t0 )|N N = dim(N ); n+ B (t0 )|N N = dim(N+ ),

applying Lemma 4.1.33 to the restriction of B to V+ N+ and to the restriction of B to V N we conclude that there exists > 0 such that B (t) is positive denite in V+ N+ and negative denite in V N for t ]t0 , t0 + [;the conclusion now follows from Corollary 4.1.7 and from Proposition 4.1.9. 4.1.34. C OROLLARY. If t B (t) Bsym (V ) is a curve of class C 1 dened in a neighborhood of the instant t0 I R and if B (t0 )|N N is nondegenerate, where N = Ker B (t0 ) , then for > 0 sufciently small we have: n+ (B (t0 + )) n+ (B (t0 )) = sgn B (t0 )|N N . P ROOF. It follows from Theorem 4.1.32 that for > 0 sufciently small we have: (4.1.24) (4.1.25) n+ (B (t0 + )) = n+ (B (t0 )) + n+ B (t0 )|N N ; n+ (B (t0 )) = n+ (B (t0 )) + n B (t0 )|N N . applying Theorem 4.1.32 to the curve t B (t) we obtain:

4.1. INDEX OF A SYMMETRIC BILINEAR FORM

119

The conclusion follows by taking the difference of (4.1.24) and (4.1.25). We will need a uniform version of Theorem 4.1.32 for technical reasons: 4.1.35. P ROPOSITION . Let X be a topological space and let be given a continuous map X [t0 , t1 [ (, t) B (t) = B (, t) Bsym (V )

differentiable in t, such that B t is also continuous in X [t0 , t1 [. Write N = Ker B (t0 ) ; assume that dim(N ) does not depend on X and that B0 (t0 ) = B t (0 , t0 ) is nondegenerate in N0 for some 0 X . Then, there exists > 0 and a neighborhood U of 0 in X such that B (t0 ) is nondegenerate on N and such that B (t) is nondegenerate on V for every U and for every t ]t0 , t0 + [. P ROOF. We will show rst that the general case can be reduced to the case that N does not depend on X . To this aim, let k = dim(N ), that by hypothesis does not depend on . Since the kernel of a bilinear form coincides with the kernel of its associated linear operator, it follows from Proposition 2.4.10 that the map N Gk (V ) is continuous in X ; now, using Proposition 2.4.6 we nd a continuous map A : U GL(V ) dened in a neighborhood U of 0 in X such that for all U, the isomorphism A() takes N0 onto N . Dene: B (t) = A()# B (t) = B (t) A(), A() , for all U and all t [t0 , t1 [. Then, Ker B (t0 ) = N0 for all U; moreover, the map B dened in U [t0 , t1 [ satises the hypotheses of the Proposition, and the validity of the thesis for B will imply the validity of the thesis also for B . The above argument shows that there is no loss of generality in assuming that: Ker(B (t0 )) = N, for all X . We split the remaining of the proof into two steps. (1) Suppose that B0 (t0 ) is positive semi-denite and that B0 (t0 ) is positive denite in N . Let W be a subspace complementary to N in V ; then B0 (t0 ) is positive denite in W . It follows that B (t0 ) is positive denite in W and that B (t0 ) is positive denite in N for all in a neighborhood U of 0 in X . Observe that, by hypothesis, Ker(B (t0 )) = N for all U. Then, for all U, Lemma 4.1.33 gives us the existence of a positive number () such that B (t) is positive denite for all t ]t0 , t0 + ()[; we only need to look more closely at the estimates done in the proof of Lemma 4.1.33 to see that it is possible to choose > 0 independently of , when runs in a sufciently small neighborhood of 0 in X . The only estimate that is delicate appears in (4.1.20). Formula (4.1.19) denes now a function r (t); for each U, we apply the mean value inequality

120

4. THE MASLOV INDEX

to the function t (t) = B (t) tB (t0 ) and we obtain: (t) (t0 ) = r (t) (t t0 ) sup
s[t0 ,t]

(s) B (s) B (t0 ) .

= (t t0 ) sup
s[t0 ,t]

With the above estimate it is now easy to get the desired conclusion. (2) Let us prove the general case. Keeping in mind that Ker(B (t0 )) = N does not depend on X , we repeat the proof of Theorem 4.1.32 replacing B (t0 ) by B (t0 ), B (t0 ) by B0 (t0 ) and B (t) by B (t); we use step (1) above instead of Lemma 4.1.33 and the proof is completed.

4.1.36. E XAMPLE . Theorem 4.1.32 and its Corollary 4.1.34 do not hold without the hypothesis that B (t0 ) be nondegenerate in N = Ker(B (t0 )); counterexamples are easy to produce by considering diagonal matrices B (t) Bsym (I Rn ). A naive analysis of the case in which the bilinear forms B (t) are simultaneously diagonalizable would suggest the conjecture that when B (t0 ) is degenerate in Ker(B (t0 )) then it would be possible to determine the variation of the co-index of B (t) when t passes through t0 by using higher order terms on the Taylor expansion of B (t)|N N around t = t0 . The following example show that this is not possible. Consider the curves B1 , B2 : I R Bsym (I R2 ) given by: B1 (t) = 1 t t t3 , B2 (t) = 1 t2 t2 t 3 ;

we have B1 (0) = B2 (0) and N = Ker B1 (0) = Ker B2 (0) = {0} I R. Observe that B1 (t)|N N = B2 (t)|N N for all t I R, so that the Taylor expansion of B1 coincides with that of B2 in N ; on the other hand, for > 0 sufciently small, we have: n+ (B1 ()) n+ (B1 ()) = 1 1 = 0, n+ (B2 ()) n+ (B2 ()) = 2 1 = 1. Our next goal is to prove that the basis provided by Sylvesters Inertia Theorem can be written as a differentiable function of the parameter t when B (t) depends differentiably on t. Towards this goal, we consider the action of the general linear group GL(V ) in the space Bsym (V ) given by: (4.1.26) GL(V ) Bsym (V ) (T, B ) T# (B ) = B (T 1 , T 1 ) Bsym (V ); it follows from Sylvesters Inertia Theorem (Theorem 4.1.10) that the orbits of this action are the sets: Bp,q sym (V ) = {B Bsym (V ) : n+ (B ) = p, n (B ) = q },

4.2. DEFINITION AND COMPUTATION OF THE MASLOV INDEX

121

with p + q = 0, 1, . . . , dim(V ). Moreover, for p and q xed, the sets {B Bsym (V ) : n+ (B ) p, n (B ) q } {B Bsym (V ) : n+ (B ) p, n (B ) q } are respective an open and a closed subset of Bsym (V ), by Lemma 4.1.29. It follows that the set Bp,q sym (V ) is locally closed in Bsym (V ). From these observation we deduce the following 4.1.37. L EMMA . The set Bp,q sym (V ) is a connected embedded submanifold of Bsym (V ) for any integers p, q 0 with p + q = 0, 1, . . . , dim(V ). P ROOF. The fact that Bp,q sym (V ) is an embedded submanifold of Bsym (V ) follows from Theorem 2.1.12. The connectedness of Bp,q sym (V ) follows from the fact that the restriction of the action (4.1.26) to GL+ (V ) is still transitive in Bp,q sym (V ); this last statement follows from the fact that, once an orientation has been xed in V , the basis (bi )n i=1 given by the Sylvesters Inertia Theorem can be chosen positively oriented (possibly replacing b1 with b1 ). 4.1.38. C OROLLARY. The set of nondegenerate symmetric bilinear forms in V is an open subset of Bsym (V ) whose (arc-)connected components are the sets k,nk Bsym (V ), k = 0, 1, . . . , n, where n = dim(V ). P ROOF. It follows from Corollary 4.1.30 and Lemma 4.1.37. We nally obtain the desired extension of Sylvesters Inertia Theorem: 4.1.39. P ROPOSITION . Given a curve B : [a, b] Bsym (V ) of class C k (0 k +) such that the integers n (B (t)) and n+ (B (t)) do not depend on t [a, b], then there exist maps bi : [a, b] V of class C k , i = 1, . . . , n, such that for each t [a, b] the vectors (bi (t))n i=1 form a basis of V in which B (t) assumes the canonical form (4.1.7). P ROOF. Let p and q be such that n+ (B (t)) = p, n (B (t)) = q for all t [a, b]; keeping in mind the transitive action (4.1.26) of GL(V ) on Bp,q sym (V ), it follows from Corollary 2.1.15 that, for B0 Bp,q ( V ) xed, the map sym GL(V ) T T# (B0 ) = B0 (T 1 , T 1 ) Bp,q sym (V ) is a differentiable bration. It follows from Remark 2.1.18 that there exists a map T : [a, b] GL(V ) of class C k such that T (t)# (B0 ) = B (t) for all t [a, b]. Choosing a basis (bi )n i=1 of V with respect to which B0 has the canonical form (4.1.7), we dene bi (t) = T (t) bi for i = 1, . . . , n and t [a, b]. This concludes the proof. 4.2. Denition and Computation of the Maslov Index In this section we will introduce the Maslov index (relative to a xed Lagrangian subspace L0 ) of a curve in the Lagrangian Grassmannian of a symplectic space (V, ); this index is an integer number that corresponds to a sort of algebraic count of the intersections of this curve with the subset 1 (L0 ). and

122

4. THE MASLOV INDEX

The denition of Maslov index will be given in terms of relative homology, and we will therefore assume familiarity with the machinery introduced in Section 3.3. We will use several properties of the Lagrangian Grassmannian that were discussed in Section 2.5 (especially from Subsection 2.5.1). It will be needed to compute the fundamental group of , and to this aim we will use the homotopy long exact sequence of a bration, studied in Section 3.2. This computation follows the same line of the examples that appear in Subsection 3.2.1; following the notations of that subsection, we will omit for simplicity the specication of the basepoint of the fundamental groups studied. As a matter of facts, all the fundamental groups that will appear are abelian, so that the fundamental groups corresponding to different choices of basepoint can be canonically identied (see Corollary 3.1.12 and Remarks 3.1.13 and 3.3.34). Finally, in order to relate the fundamental group of with its rst singular homology group we will use the Hurewiczs homomorphism, presented in Subsection 3.3.1. Throughout this section we will consider a xed symplectic space (V, ), with dim(V ) = 2n; we will denote by the Lagrangian Grassmannian of this symplectic space. All the curves considered will be tacitly meant to be continuous curves; moreover, we will often use the fact that any two Lagrangian subspaces admit a common complementary Lagrangian subspace (see Remark 2.5.18). We know that the Lagrangian Grassmannian is diffeomorphic to the quotient U(n)/O(n) (see Corollary 2.5.12). Consider the homomorphism: d = det2 : U(n) S 1 , where S 1 C denotes the unit circle; if A O(n) then clearly det(A) = 1, hence O(n) Ker(d). It follows that d induces, by passing to the quotient, a map: (4.2.1) : U(n)/O(n) S 1 , d (A O(n)) = det2 (A). We have the following: given by d 4.2.1. P ROPOSITION . The fundamental group of the Lagrangian Grassmannian = U(n)/O(n) is innite cyclic; more explicitly, the map (4.2.1) induces an isomorphism: = : 1 U(n)/O(n) d 1 (S 1 ) = Z. is a bration with typical ber P ROOF. It follows from Corollary 2.1.16 that d Ker(d)/O(n). It is easy to see that the action of SU(n) on Ker(d)/O(n) by left translation is transitive, and that the isotropy group of the class 1 O(n) of the neutral element is SU(n) O(n) = SO(n); it follows from Corollary 2.1.9 that we have a diffeomorphism SU(n)/SO(n) = Ker(d)/O(n) induced by the inclusion of SU(n) in Ker(d). Since SU(n) is simply connected and SO(n) is connected, it follows easily from the homotopy long exact sequence of the bration SU(n) SU(n)/SO(n) that SU(n)/SO(n) is simply connected. Then, Ker(d)/O(n) is also simply connected, and the homotopy exact sequence

4.2. DEFINITION AND COMPUTATION OF THE MASLOV INDEX

123

becomes: of the bration d


0 1 U(n)/O(n) 1 (S 1 ) 0 =

This concludes the proof. 4.2.2. C OROLLARY. The rst singular homology group H1 () of is innite cyclic. P ROOF. Since is arc-connected and 1 () is abelian, it follows from Theorem 3.3.33 that the Hurewiczs homomorphism is an isomorphism: (4.2.2) : 1 () H1 () q : (, ) (, 0 (L0 )) induces an isomorphism: (4.2.3) H1 (, 0 (L0 )); q : H1 ()
= =

4.2.3. C OROLLARY. For a xed Lagrangian L0 , the inclusion

in particular, H1 (, 0 (L0 )) is innite cyclic. P ROOF. It follows from Remark 2.5.3 and from Example 3.3.19. 0 (L0 ); : [a, b] be a curve with endpoints in 0 (L0 ), i.e., (a), (b) then, denes a relative homology class in H1 (, 0 (L0 )) (see Remarks 3.3.30 and 3.3.26). Our goal is now to show that the transverse orientation of 1 (L0 ) given in Denition 2.5.19 induces a canonical choice of a generator of the innite cyclic group H1 (, 0 (L0 )). Once this choice is made, we will be able to associate an integer number to each curve in with endpoints in 0 (L0 ). Let 4.2.4. E XAMPLE . If we analyze the steps that lead us to the conclusion that H1 (, 0 (L0 )) is isomorphic to Z we can compute explicitly a generator for this group. In rst place, the curve it e i 0 3 [ , ] t A ( t ) = U(n) . 2 2 . 0 . i projects onto a closed curve A(t) = A(t) O(n) in U(n)/O(n); moreover, (4.2.4)
3 [ 2, 2 ]

t det2 A(t) = (1)n1 e2it

is a generator of the fundamental group of the unit circle S 1 . It follows from Proposition 4.2.1 that A denes a generator of the fundamental group of U(n)/O(n). Denoting by (I R2n ) the Lagrangian Grassmannian of the symplectic space I R 2n endowed with the canonical symplectic form, it follows from Proposition 2.5.11 that a diffeomorphism U(n)/O(n) R2n ) is given explicitly by: = (I U(n)/O(n) A O(n) A I Rn {0}n (I R2n ).

124

4. THE MASLOV INDEX

The Lagrangian A(t) I Rn {0}n is generated by the vectors1 {e1 cos(t) + en+1 sin(t), en+2 , . . . , e2n },
n where (ej )2 R 2n . j =1 denotes the canonical basis of I 2 n The choice of a symplectic basis (bj )j =1 of V induces a diffeomorphism of onto (I R2n ) in an obvious way. Consider the Lagrangian (t) given by: 2n

(4.2.5) then, the curve (4.2.6)

(t) = I R b1 cos(t) + bn+1 sin(t) +


j =n+2

I Rbj ;

3 [ 2, 2 ]

t (t)

is a generator of 1 (). By the denition of the Hurewiczs homomorphism (see (3.3.25)) we have that the same curve (4.2.6) denes a generator of H1 (); since the isomorphism (4.2.3) is induced by inclusion, we have that the curve (4.2.6) is also a generator of H1 (, 0 (L0 )). 4.2.5. L EMMA . Let A Sp(V, ) be a symplectomorphism of (V, ) and consider the diffeomorphism (also denoted by A) of induced by the action of A; then the induced homomorphism in homology: A : Hp () Hp () is the identity map for all p Z. P ROOF. Since Sp(V, ) is arc-connected, there exists a curve [0, 1] [0, 1] s A(s) Sp(V, ) such that A(0) = A and A(1) = Id. Dene (s, L) Hs (L) = A(s) L ; then H : A = Id is a homotopy. The conclusion follows from Corollary 3.3.24. 4.2.6. C OROLLARY. Let L0 be a Lagrangian subspace of (V, ) and let A Sp(V, , L0 ) (recall (2.5.15)); then the homomorphism A : H1 (, 0 (L0 )) H1 (, 0 (L0 )) is the identity map. P ROOF. It follows from Lemma 4.2.5 and from the following commutative diagram: H1 ()
q A =Id

/ H1 ()
=

H1 (, 0 (L0 )) where q is given in (4.2.3).

/ H1 (, 0 (L0 ))

1The complex matrix A(t) must be seen as a linear endomorphism of I R2n ; therefore, we need

the identication of n n complex matrices with 2n 2n real matrices (see Remark 1.2.9).

4.2. DEFINITION AND COMPUTATION OF THE MASLOV INDEX

125

4.2.7. E XAMPLE . Consider a Lagrangian decomposition (L0 , L1 ) of V and let L be an element in the domain of the chart L0 ,L1 , i.e., L 0 (L1 ). It follows directly from the denition of L0 ,L1 (see (2.5.3)) that the kernel of the symmetric bilinear form L0 ,L1 (L) Bsym (L0 ) is L0 L, that is: (4.2.7) Ker L0 ,L1 (L) = L0 L. Then, we obtain that for each k = 0, . . . , n the Lagrangian L belongs to k (L0 ) if and only if the kernel of L0 ,L1 (L) has dimension k , that is: L0 ,L1 0 (L1 ) k (L0 ) = {B Bsym (L0 ) : dgn(B ) = k }. In particular, we have L 0 (L0 ) if and only if L0 ,L1 (L) is nondegenerate. 4.2.8. E XAMPLE . Let t (t) be a curve in differentiable at t = t0 and let (L0 , L1 ) be a Lagrangian decomposition of V with (t0 ) 0 (L1 ). Then, for t in a neighborhood of t0 we also have (t) 0 (L1 ) and we can therefore dene (t) = L0 ,L1 ( (t)) Bsym (L0 ). Let us determine the relation between (t0 ) and (t0 ); by Lemma 2.5.7 we have:
1 (t0 ) = dL0 ,L1 ( (t0 )) (t0 ) = L (t0 ),L0

(t0 ).

1 Since L (t0 ),L0 xes the points of L0 (t0 ), we obtain in particular that the symmetric bilinear forms (t0 ) Bsym (L0 ) and (t0 ) Bsym ( (t0 )) coincide on L0 (t0 ).

4.2.9. L EMMA . Let L0 be a xed Lagrangian; assume given two curves


1, 2 :

[a, b]

Suppose that there exists a Lagrangian subspace L1 complementary to L0 such that 0 (L1 ) contains the images of both curves 1 , 2 ; if we have (4.2.8) n+ L0 ,L1 ( 1 (t)) = n+ L0 ,L1 ( 2 (t)) ,
1, 2

with endpoints in 0 (L0 ).

for t = a and t = b, then the curves

are homologous in H1 (, 0 (L0 )).

P ROOF. It follows from (4.2.8) and from Corollary 4.1.38 that there exist curves: 1 , 2 : [0, 1] Bsym (L0 ) such that 1 (t) and 2 (t) are nondegenerate for all t [0, 1] and also: 1 (0) = L0 ,L1 ( 1 (a)), 2 (0) = L0 ,L1 ( 1 (b)), 1 (1) = L0 ,L1 ( 2 (a)), 2 (1) = L0 ,L1 ( 2 (b)).

1 Dene mi = L0 ,L1 i , i = 1, 2; it follows from Example 4.2.7 that m1 and m2 have image in the set 0 (L0 ) and therefore they are homologous to zero in 1 H1 (, 0 (L0 )). Consider the concatenation = m 1 1 m2 ; it follows from Lemma 3.3.27 that 1 and are homologous in H1 (, 0 (L0 )). We have that and 2 are curves in 0 (L1 ) with the same endpoints, and since 0 (L1 ) is homeomorphic to the vector space Bsym (L0 ) it follows that and 2 are homotopic with

126

4. THE MASLOV INDEX 2

xed endpoints. By Corollary 3.3.29 we have that and concludes the proof.

are homologous, which

4.2.10. D EFINITION . Let : [a, b] be a curve of class C 1 . We say that intercepts transversally the set 1 (L0 ) at the instant t = t0 if (t0 ) 1 (L0 ) and (t0 ) T (t0 ) 1 (L0 ); we say that such transverse intersection is positive (resp., negative) if the class of (t0 ) in the quotient T (t0 ) /T (t0 ) 1 (L0 ) denes a positively oriented (resp., a negatively oriented) basis (recall Denition 2.5.19). From Theorem 2.5.16 it follows that intercepts 1 (L0 ) transversally at the instant t = t0 if and only if (t0 ) 1 (L0 ) and the symmetric bilinear form (t0 ) is non zero in the space L0 (t0 ); such intersection will be positive (resp., negative) if (t0 ) is positive denite (resp., negative denite) in L0 (t0 ). 4.2.11. L EMMA . Let L0 be a Lagrangian subspace and let
1, 2 :

[a, b]

be curves of class with endpoints in 0 (L0 ) that intercept 1 (L0 ) only once; suppose that such intersection is transverse and positive. Then, we have that 1 and 2 are homologous in H1 (, 0 (L0 )), and either one of these curves denes a generator of H1 (, 0 (L0 )) = Z. C1 P ROOF. Thanks to Lemma 3.3.25, we can assume that 1 , 2 intercept 1 (L0 ) at the same instant t0 ]a, b[. By Proposition 1.4.38 there exists a symplectomorphism A Sp(V, , L0 ) such that A( 1 (t0 )) = 2 (t0 ). It follows from Corollary 4.2.6 that A 1 and 1 are homologous in H1 (, 0 (L0 )); note that also A 1 intercepts 1 (L0 ) only at the instant t0 and that such intersection is transverse and positive (see Proposition 2.5.20). The above argument shows that there is no loss of generality in assuming ( t 1 0 ) = 2 (t0 ). By Lemma 3.3.27, it is enough to show that the restriction 1 |[t0 ,t0 +] is homologous to 2 |[t0 ,t0 +] for some > 0. Let L1 be a common complementary Lagrangian to 1 (t0 ) and L0 ; for t in a neighborhood of t0 we can write i (t) = L0 ,L1 i (t), i = 1, 2. By Example 4.2.8 we have that i (t0 ) and i (t0 ) coincide in L0 i (t0 ) = Ker i (t0 ) (see (4.2.7)); since by hypothesis i (t0 ) is positive denite in the unidimensional space L0 i (t0 ), it follows from Theorem 4.1.32 (see also (4.1.25)) that for > 0 sufciently small we have (4.2.9) n+ i (t0 + ) = n+ i (t0 ) + 1, n+ i (t0 ) = n+ i (t0 ) . Since 1 (t0 ) = 2 (t0 ), it follows from (4.2.9) that n+ 1 (t0 + ) = n+ 2 (t0 + ) , n+ 1 (t0 ) = n+ 2 (t0 ) , for > 0 sufciently small. Now, it follows from Lemma 4.2.9 that the curve 0 1 |[t0 ,t0 +] is homologous to the curve 2 |[t0 ,t0 +] in H1 (, (L0 )). This concludes the proof of the rst statement of the thesis. To prove the second statement it sufces to exhibit a curve that has a unique intersection with 1 (L0 ), being such intersection transverse and positive, so that n denes a generator of H1 (, 0 (L0 )). Let (bj )2 j =1 be a symplectic basis of V

4.2. DEFINITION AND COMPUTATION OF THE MASLOV INDEX

127

such that (bj )n j =1 is a basis of L0 (see Lemma 1.4.35); consider the generator of 0 H1 (, (L0 )) described in (4.2.5) and (4.2.6). It is easy to see that intercepts 1 (L0 ) only at the instant t = and L0 ( ) is the unidimensional space generated by b1 ; moreover, an easy calculation shows that: (4.2.10)
1

( )(b1 , b1 ) = (bn+1 , b1 ) = 1;

it follows that has a unique intersection with 1 (L0 ) and that this intersection is transverse and positive. By Lemma 3.3.27, the curve 1 is also a generator of H1 (, 0 (L0 )), which concludes the proof. 4.2.12. D EFINITION . Let L0 be a xed Lagrangian; we dene an isomorphism (4.2.11) L0 : H1 (, 0 (L0 )) Z
=

as follows: choose a curve of class C 1 in with endpoints in 0 (L0 ) such that has a unique intersection with 1 (L0 ) and such that this intersection is transverse and positive. Dene L0 by requiring that the homology class of be taken into the element 1 Z; by Lemma 4.2.11 the isomorphism (4.2.11) is well dened, i.e., independent of the choice of the curve . Suppose now that : [a, b] is an arbitrary curve with endpoints in 0 (L0 ), then we denote by L0 ( ) Z the integer number that corresponds to the homology class of by the isomorphism (4.2.11); the number L0 ( ) is called the Maslov index of the curve relative to the Lagrangian L0 . In the following Lemma we list some of the properties of the Maslov index: 4.2.13. L EMMA . Let : [a, b] be a curve with endpoints in 0 (L0 ); then we have: (1) if : [a , b ] [a, b] is a continuous map with (a ) = a, (b ) = b then L0 ( ) = L0 ( ); (2) if m : [a , b ] is a curve with endpoints in 0 (L0 ) such that (b) = m(a ), then L0 ( m) = L0 ( ) + L0 (m); (3) L0 1 = L0 ( ); (4) if Im( ) 0 (L0 ) then L0 ( ) = 0; (5) if m : [a, b] is homotopic to with free endpoints in 0 (L0 ) (see Denition 3.1.25) then L0 ( ) = L0 (m); (6) there exists a neighborhood U of in C 0 ([a, b], ) endowed with the compact-open topology such that, if m U has endpoints in 0 (L0 ), then L0 ( ) = L0 (m). P ROOF. Property (1) follows from Lemma 3.3.25; Properties (2) and (3) follow from Lemma 3.3.27. Property (4) follows immediately from the denition of the group H1 (, 0 (L0 )) (see (3.3.7)). Property (5) follows from Remark 3.3.30 and Property (6) follows from Theorem 3.1.27 and from Property (5). 4.2.14. E XAMPLE . The Maslov index L0 ( ) can be seen as the intersection number of the curve with the subset 1 (L0 ) ; indeed, it follows from Lemma 4.2.13 (more specically, from Properties (2), (3) and (4)) that if

128

4. THE MASLOV INDEX

: [a, b] is a curve of class C 1 with endpoints in 0 (L0 ) that has only transverse intersections with 1 (L0 ) then the Maslov index L0 ( ) is the number of positive intersections of with 1 (L0 ) minus the number of negative intersections of with 1 (L0 ). As a matter of facts, these numbers are nite (see Example 4.2.17 below). In Corollary 4.2.18 we will give a generalization of this result. We will now establish an explicit formula for the Maslov index L0 in terms of a chart L0 ,L1 of : 4.2.15. T HEOREM . Let L0 be a Lagrangian subspace and let : [a, b] be a given curve with endpoints in 0 (L0 ). If there exists a Lagrangian L1 complementary to L0 such that the image of is contained in 0 (L1 ), then the Maslov index L0 ( ) of is given by: L0 ( ) = n+ L0 ,L1 ( (b)) n+ L0 ,L1 ( (a)) . P ROOF. By Lemma 4.2.9, it sufces to determine for each i, j = 0, 1, . . . , n a curve i,j : [0, 1] Bsym (L0 ) such that: (4.2.12) (4.2.13) n+ i,j (0) = i, n+ i,j (1) = j, dgn i,j (0) = 0, dgn i,j (1) = 0

1 and such that the curve i,j = L0 ,L1 i,j satises L0 ( i,j ) = j i. If i = j , we simply take i,i to be any constant curve such that i,i (0) is nondegenerate and such that n+ i,i (0) = i. Property (3) in the statement of Lemma 4.2.13 implies that there is no loss of generality in assuming i < j . Let us start with the case j = i + 1; choose any basis of L0 and dene i,i+1 (t) as the bilinear form whose matrix representation in this basis is given by:

i,i+1 (t) diag(1, 1, . . . , 1, t 1 2 , 1, 1, . . . , 1),


i times ni1 times

t [0, 1],

where diag(1 , . . . , n ) denotes the diagonal matrix with entries 1 , . . . , n . Then, we have: n+ i,i+1 (0) = i, n+ i,i+1 (1) = i + 1, dgn i,i+1 (0) = 0, dgn i,i+1 (1) = 0;

1 moreover i,i+1 (t) is degenerate only at t = 1 2 and the derivative i,i+1 ( 2 ) is positive denite in the unidimensional space Ker i,i+1 ( 1 2 ) . It follows from Ex1 amples 4.2.7 and 4.2.8 that i,i+1 intercepts 1 (L0 ) only at t = 2 , and that such intersection is transverse and positive. By denition of Maslov index, we have:

L0 (

i,i+1 )

= 1;

and this completes the construction of the curve i,j in the case j = i + 1. Let us look now at the case j > i + 1. For each i = 0, . . . , n, let Bi Bsym (L0 ) be a nondegenerate symmetric bilinear form with n+ (Bi ) = i; choose i,i+1 : [0, 1] Bsym (L0 ) with i,i+1 (0) = Bi and i,i+1 (1) = Bi+1 any curve

4.2. DEFINITION AND COMPUTATION OF THE MASLOV INDEX

129

for i = 0, . . . , n 1. It follows from Lemma 4.2.9 and from the rst part of the 1 proof that i,i+1 = L0 ,L1 i,i+1 satises L0 i,i+1 = 1; for j > i + 1 dene: i,i+1 i+1,i+2 j 1,j . i,j = Then, the curve i,j satises (4.2.12), (4.2.13) and from Property (2) in the statement of Lemma 4.2.13 it follows that L0 ( i,j ) = j i. This concludes the proof. 4.2.16. D EFINITION . Given a curve t (t) of class C 1 we say that has a nondegenerate intersection with 1 (L0 ) at the instant t = t0 if (t0 ) 1 (L0 ) and (t0 ) is nondegenerate in L0 (t0 ). 4.2.17. E XAMPLE . If a curve in has a nondegenerate intersection with 1 (L0 ) at the instant t = t0 , then this intersection is isolated, i.e., (t) 0 (L0 ) for t = t0 sufciently close to t0 . To see this, choose a common complementary Lagrangian L1 to L0 and (t0 ) and apply Theorem 4.1.32 to the curve = L0 ,L1 , keeping in mind Examples 4.2.7 and 4.2.8. Since 1 (L0 ) is closed in , it follows that if a curve : [a, b] has only nondegenerate intersections with 1 (L0 ), then (t) 1 (L0 ) only at a nite number of instants t [a, b]. We have the following corollary to Theorem 4.2.15: 4.2.18. C OROLLARY. Let L0 be a Lagrangian subspace and let be given a curve : [a, b] of class C 1 with endpoints in 0 (L0 ) that has only nondegenerate intersections with 1 (L0 ). Then, (t) 1 (L0 ) only at a nite number of instants t [a, b] and the following identity holds: L0 ( ) =
t[a,b]

sgn

(t)|(L0

(t))(L0 (t))

P ROOF. It follows from Example 4.2.17 that (t) 1 (L0 ) only at a nite number of instants t [a, b]. Let t0 ]a, b[ be such that (t0 ) 1 (L0 ); keeping in mind Property (2) and (4) in the statement of Lemma 4.2.13, it sufces to prove that: L0 |[t0 ,t0 +] = sgn (t0 )|(L0 (t0 ))(L0 (t0 )) , for > 0 sufciently small. Choose a common complementary L1 of L0 and (t0 ); for t in a neighborhood of t0 we can write (t) = L0 ,L1 ( (t)). The conclusion now follows from Theorem 4.2.15 and from Corollary 4.1.34, keeping in mind Examples 4.2.7 and 4.2.8. In Example 4.2.17 we have seen that a nondegenerate intersection of a curve of class C 1 with 1 (L0 ) at an instant t0 is isolated, i.e., there exists > 0 such that (t) 1 (L0 ) for t [t0 , t0 [ ]t0 , t0 + ]. For technical reasons, we will need (in Proposition 5.2.8) a slightly stronger result and we will prove next that the choice of such > 0 can be made uniformly with respect to a parameter.

130

4. THE MASLOV INDEX

4.2.19. L EMMA . Let X be a topological space and suppose that it is given a continuous map: X [t0 , t1 [ (, t)
(t)

= (, t)

which is differentiable in the variable t and such that t : X [t0 , t1 [ T is also continuous. Fix a Lagrangian L0 ; suppose that dim( (, t0 ) L0 ) is independent of X and that the curve 0 = (0 , ) has a nondegenerate intersection with 1 (L0 ) at t = t0 for some 0 X . Then, there exists > 0 and a neighborhood U of 0 in X such that, for all U, has a nondegenerate intersection with 1 (L0 ) at t0 and such that (, t) 0 (L0 ) for all U and all t ]t0 , t0 + ]. P ROOF. Choose a common complementary Lagrangian L1 of L0 and (0 , t0 ) and dene (, t) = L0 ,L1 ( (, t)) for t in a neighborhood of t0 and in a neighborhood of 0 in X . Then, is continuous, it is differentiable in t, and the derivative t is continuous. The conclusion follows now applying Proposition 4.1.35 to the map , keeping in mind Examples 4.2.7 and 4.2.8. 4.2.20. R EMARK . A more careful analysis of the denition of the transverse orientation of 1 (L0 ) in (Denition 2.5.19) shows that the choice of the sign made for the isomorphism L0 is actually determined by the choice of a sign in the symplectic form . More explicitly, if we replace by , which does not affect the denition of the set , then we obtain a change of sign for the isomorphisms L0 ,L1 and L (dened in formulas (1.4.11) and (1.4.13)). Consequently, this change of sign induces a change of sign in the charts L0 ,L1 (dened in formula (2.5.3)) and in the isomorphism (2.5.12) that identies TL with Bsym (L). The conclusion is that changing the sign of causes an inversion of the transverse orientation of 1 (L0 ) in , which inverts the sign of the isomorphism L0 . 4.2.21. R EMARK . The choice of a Lagrangian subspace L0 denes an isomorphism: (4.2.14) L0 q : H1 () Z,
=

where q is given in (4.2.3). We claim that this isomorphism does not indeed depend on the choice of L0 ; for, let L0 be another Lagrangian subspace. By Corollary 1.4.28, there exists a symplectomorphism A Sp(V, ) such that A(L0 ) = L0 ; we have the following commutative diagram (see Lemma 4.2.5): H1 ()
q A =Id

/ H1 () 
q

H1 (, 0 (L0 ))

LLL LLL L0 LLL LL&

/ H1 (, 0 (L )) 0 rr r r r rrr L0 x rr r

4.2. DEFINITION AND COMPUTATION OF THE MASLOV INDEX

131

where the commutativity of the lower triangle follows from Remark 2.5.21. This proves the claim. Observe that if : [a, b] is a loop, i.e., (a) = (b), then, since denes a homology class in H1 (),we obtain the equality: L0 ( ) = L0 ( ), for any pair of Lagrangian subspaces L0 , L0 . 4.2.22. R EMARK . Let J be a complex structure in V compatible with ; consider the inner product g = (, J ) and the Hermitian product gs in (V, J ) dened in (1.4.10). Let 0 be a Lagrangian subspace; Proposition 2.5.11 tells us that the map (4.2.15) U(V, J, gs )/O
0, g| 0
0

AO

0, g| 0

A( 0 )

is a diffeomorphism. As in (4.2.1), we can dene a map : U(V, J, gs )/O 0 , g | S 1 d


0 0

obtained from d = det2 : U(V, J, gs ) S 1 induces an isomorphism d of the funby passage to the quotient; then the map d damental groups. Indeed, by Remark 1.4.30 we can nd a basis of V that puts all the objects (V, , J, g, gs , 0 ) simultaneously in their canonical forms, and then together with the everything works as in Proposition 4.2.1. The isomorphism d diffeomorphism (4.2.15) and the choice of (3.2.24) (or, equivalently, of (4.2.4)) as a generator of 1 (S 1 ) = H1 (S 1 ) produce an isomorphism (see also (4.2.2)): u = uJ, 0 : H1 () Z; this isomorphism does not indeed depend on the choice of J and of 0 . To see this, choose another complex structure J in V compatible with and another Lagrangian subspace 0 ; we then obtain an isomorphism u = uJ , 0 . From Remark 1.4.30 it follows that there exists a symplectomorphism A Sp(V, ) that takes 0 onto 0 and that is C-linear from (V, J ) into (V, J ); then, it is easy to see that the following diagram commutes: H1 ()
EE EE u EE EE E" y< Z yy y y yy yy u
=

H1 () By Lemma 4.2.5 we have that A = Id and the conclusion follows. As a matter of facts, formula (4.2.10) shows that the isomorphism u has the opposite sign of the isomorphism (4.2.14) obtained by using the transverse orientation of 1 (L0 ) in .

132

4. THE MASLOV INDEX

Exercises for Chapter 4 E XERCISE 4.1. Prove that every symmetric bilinear form B Bsym (V ) admits a maximal negative subspace. E XERCISE 4.2. Suppose that B Bsym (V ), with V = V1 V2 and B is negative semi-denite in V1 and in V2 . If the inequality (4.1.15) holds for all v1 V1 and v2 V2 , then B is negative semi-denite in V . E XERCISE 4.3. Let V be an n-dimensional real vector space, B Bsym (V ) a symmetric bilinear form and assume that the matrix representation of B in some basis {v1 , . . . , vn } of V is given by: X Z , Z Y where X is a k k symmetric matrix and Y is a (n k ) (n k ) symmetric matrix. Prove that, if X is invertible, then: n (B ) = n (X ) + n (Y Z X 1 Z ), and dgn(B ) = dgn(Y Z X 1 Z ), n+ (B ) = n+ (X ) + n+ (Y Z X 1 Z ).

E XERCISE 4.4. Let V be a nite dimensional real vector space and let U, Z Bsym (V ) be nondegenerate symmetric bilinear forms on V such that U Z is also nondegenerate. Prove that U 1 Z 1 is nondegenerate and that: n (Z ) n (U ) = n (Z 1 U 1 ) n (U Z ). E XERCISE 4.5. Consider the space I R2n endowed with its canonical symplectic form ; dene an isomorphism O : I R 2n I R2n by O(x, y ) = (x, y ), for n # all x, y I R . Show that O ( ) = and conclude that O induces a diffeomorphism of the Lagrangian Grassmannian to itself. Show that the homomorphism: O : H1 () H1 () is equal to minus the identity map (compare with Remark 4.2.20). E XERCISE 4.6. Let L0 be a Lagrangian in the symplectic space (V, ) and let A : [a, b] Sp(V, , L0 ), : [a, b] be continuous curves such that (a), (b) 0 (L0 ). Prove that the curve = A : [a, b] is homologous to in H1 (, 0 (L0 )). E XERCISE 4.7. Let L0 be a Lagrangian subspace of (V, ) and let L1 , [a, b] be curves such that: L1 (t) is transverse to L0 and to (t) for all t [a, b]; (a) and (b) are transverse to L0 . Show that the Maslov index L0 ( ) of the curve is equal to: L0 ( ) = n+ L0 ,L1 (b) ( (b)) n+ L0 ,L1 (a) ( (a)) . :

EXERCISES FOR CHAPTER 4

133

E XERCISE 4.8. Let L0 , L1 , L2 , L3 be four Lagrangian subspaces of the symplectic space (V, ), with L0 L1 = L0 L2 = L0 L3 = L2 L3 = {0}. Recall the denition of the map L0 ,L1 : L1 L 0 given in (1.4.11), the denition of pull-back of a bilinear form given in Denition 1.1.2 and the denition of the chart L0 ,L1 of given in (2.5.3). Prove that the following identity holds: L0 ,L1 (L3 ) L1 ,L0 (L2 ) = (L0 ,L1 )# L0 ,L3 (L2 )1 . E XERCISE 4.9. As in Exercise 4.8, prove that the following identity holds: n+ L0 ,L3 (L2 ) = n+ L1 ,L0 (L3 ) L1 ,L0 (L2 ) . E XERCISE 4.10. Let (L0 , L1 ) be a Lagrangian decomposition of the symplectic space (V, ) and let : [a, b] be a continuous curve with endpoints in 0 (L0 ). Suppose that there exists a Lagrangian L such that Im( ) 0 (L ). Prove that the Maslov index L0 ( ) is given by the following formula: L0 ( ) = n L1 ,L0 ( (b)) L1 ,L0 (L ) n L1 ,L0 ( (a)) L1 ,L0 (L ) . E XERCISE 4.11. Dene the following symplectic form in I R 4n : (v1 , w1 ), (v2 , w2 ) = (v1 , v2 ) (w1 , w2 ), v1 , w 1 , v2 , w 2 I R 2n . where is the canonical symplectic form of I R2n . Prove that A Lin(I R 2n , I R 2n ) 2 n is a symplectomorphism of (I R , ) if and only if its graph Gr(A) is a Lagrangian R) A Gr(A) (I R 4n , ) subspace of (I R4n , ). Show that the map Sp(2n, I is a diffeomorphism onto an open subset. E XERCISE 4.12. Prove that the set T Sp(2n, I R) : T (L0 ) L0 = {0} is an open dense subset of Sp(2n, I R) with two connected components. E XERCISE 4.13. Dene: + = T Sp(2n, I R) : det(T Id) > 0 ; = T Sp(2n, I R) : det(T Id) < 0 . Prove that + and are open and connected subsets of Sp(2n, I R) (see Exercise 4.15 for more properties of the sets + and ). E XERCISE 4.14. Consider the set: E = T Sp(2n, I R) : det(T Id) = 0, T (L0 ) L0 = {0} . Prove that E is a dense open subset of Sp(2n, I R) having 2(n + 1) connected comA B ponents. Prove that each connected component contains an element T = C D with A = 0 and B in diagonal form. E XERCISE 4.15. Recall from Exercise 4.13 the denition of the sets + , Sp(2n, I R). Prove that A + if and only if Gr(A) is a Lagrangian in 0 () (I R4n , ), where is the diagonal of I R 4n = I R 2n I R2n . Conclude that any loop in + is homotopic to a constant in Sp(2n, I R).

CHAPTER 5

Some Applications to Differential Systems


5.1. Symplectic Differential Systems In this section we will always consider the symplectic space I Rn I Rn endowed with its canonical symplectic form . Recall from Subsection 2.1.1 that the Lie algebra sp(I Rn I Rn , ) can be identied with the set of 2n 2n real matrices X of the form: A B (5.1.1) X= , B, C symmetric, C A where A, B, C are n n matrices and A is the transpose of A. The matrices A, B, C and A can be identied with linear operators A Lin(I Rn ), B n n n n n R ); we can also identify B R ), C Lin(I R ,I R ) and A Lin(I Lin(I R ,I n with a symmetric bilinear form in I R and C with a symmetric bilinear form in I Rn . We will be interested in homogeneous systems of linear differential equations of the form: d v (t) v (t) = X (t) , t [a, b], (5.1.2) (t) dt (t) where X : [a, b] sp(I Rn I Rn , ), v : [a, b] I Rn e : [a, b] I Rn . For all t [a, b], the linear operator X (t) determines operators A(t), B (t) and C (t) as in (5.1.1); we can then rewrite (5.1.2) more concisely in the form: (5.1.3) v = Av + B, = Cv A ,

where the variable t is omitted for simplicity. 5.1.1. D EFINITION . A homogeneous linear system of differential equations of the form (5.1.3), where A : [a, b] Lin(I Rn ), B : [a, b] Bsym (I Rn ) and n C : [a, b] Bsym (I R ) are smooth functions and B (t) is nondegenerate for all t [a, b] is called a symplectic differential system. If X (t) denotes the matrix dened by A(t), B (t) and C (t) as in (5.1.1), we say that X is the coefcient matrix of the symplectic system (5.1.3), and the maps A, B and C will be called the components of X . In general, we will identify the symplectic differential system (5.1.3) with its coefcient matrix X ; for instance, we will say that (v, ) is a solution of X meaning that (v, ) is a solution of (5.1.3), or, equivalently, of (5.1.2).
134

5.1. SYMPLECTIC DIFFERENTIAL SYSTEMS

135

In the rest of this section we will consider a xed symplectic differential system X : [a, b] sp(I Rn I Rn , ) with components A, B and C . 5.1.2. R EMARK . Since B (t) is nondegenerate for all t, then its index does not depend on t, and so we can write: n (B (t)) = n (B (t)1 ) = k, B (t)1 t [a, b]. Recall that the linear operator can be identied with a symmetric bilinear form in I Rn , which is the push-forward of the bilinear form B (t) by the linear operator B (t). Given a smooth map v : [a, b] I Rn , there exists at most one map : [a, b] such that (v, ) is a solution of X ; for, the invertibility of B (t) allows to solve the rst equation of (5.1.3) for in terms of v . We can therefore dene for any smooth map v : [a, b] I Rn a smooth map v : [a, b] I Rn by the following formula: I Rn (5.1.4) v (t) = B (t)1 v (t) A(t)v (t) . 5.1.3. D EFINITION . Given a smooth map v : [a, b] I Rn , we say that v is a solution of the symplectic differential system X if (v, v ) is a solution of X . From the elementary theory of ordinary linear differential equations we know that, given v0 I Rn and 0 I Rn , there exists a unique solution (v, ) of X in [a, b] satisfying v (a) = v0 and (a) = 0 (see for instance [4, Theorem 5.1, Chapter 1]); therefore we have a well dened linear isomorphism: (t) : I Rn I Rn I Rn I Rn such that (t) (v (a), (a) = v (t), (t) , for any solution (v, ) of X . We then obtain a smooth curve [a, b] GL(I Rn I Rn ) that satises: (5.1.5) (t) = X (t) (t), for all t [a, b] and t (t)

(a) = Id.

5.1.4. D EFINITION . The map determined by (5.1.5) is called the fundamental matrix of the symplectic differential system X . Since X takes values in the Lie algebra of the symplectic group, it follows from (5.1.5) that takes values in the symplectic group (see Remark 2.1.4), that is, the fundamental matrix of the symplectic differential system X is a differentiable map: : [a, b] Sp(I Rn I Rn , ). The fact that (t) is a symplectomorphism is expressed by the following identity (5.1.6) (v (t), v (t)), (w(t), w (t)) = w (t) v (t) v (t) w(t) = constant, for any solutions v and w of X . Let 0 I Rn I Rn be Lagrangian subspace; let us consider the following initial condition for the system (5.1.3): (5.1.7) (v (a), (a))
0.

136

5. APPLICATIONS

Recalling Exercise 1.11, there exists a bijection between the set of Lagrangian subspaces 0 I Rn I Rn and the set of pairs (P, S ), where P I Rn is a subspace and S Bsym (P ) is a symmetric bilinear form; such bijection is determined by the identity: (5.1.8)
0

= (v, ) I Rn I Rn : v P, |P + S (v ) = 0 ,

where S is identied with a linear operator S : P P . In terms of the pair (P, S ), the initial condition (5.1.7) can be rewritten in the form: (5.1.9) v (a) P, (a)|P + S v (a) = 0.

5.1.5. D EFINITION . We call (5.1.7) (respectively, (5.1.9)) the Lagrangian initial condition determined by the Lagrangian 0 (respectively, by the pair (P, S )); if (v, ) is a solution of X that satises (5.1.7), or, equivalently, (5.1.9), we say that (v, ) is a solution of the pair (X, 0 ), or also that (v, ) is a (X, 0 )-solution. We will denote by V = V(X, 0 ) the set of solutions of (X, 0 ), that is: (5.1.10) V(X,
0)

= V = v : v is a solution of (X,

0)

Clearly, V is a subspace of the space of all maps v : [a, b] I Rn ; moreover: dim(V) = dim( 0 ) = n. We will x for the rest of the section a Lagrangian 0 I Rn I Rn and we will denote by (P, S ) the pair corresponding to 0 as in (5.1.8). For each t [a, b] we dene the following subspace of I Rn : (5.1.11) It is easy to see that: (5.1.12) V[a] = P.
0 ),

V[t] = v (t) : v V I Rn ;

It follows directly from (5.1.6) that, given solutions v and w of (X, v (t) w(t) = w (t) v (t),

then:

for each t [a, b]; then, if v is a solution of (X, 0 ) with v (t) = 0, the functional v (t) annihilates the space V[t]. Conversely, if 0 I Rn is a functional that annihilates V[t], it follows from (5.1.6) that if v is the unique solution of X such that v (t) = 0 and v (t) = 0 , then 0 = (v (t), v (t)), (w(t), w (t)) = (v (a), v (a)), (w(a), w (a)) , for all (X, 0 )-solution w; hence, (v (a), v (a)) is -orthogonal to 0 and therefore v is a solution of (X, 0 ). These observations show that the annihilator of V[t] is given by: (5.1.13) V[t]o = v (t) : v V and v (t) = 0 , for all t [a, b]; keeping in mind (5.1.4), it follows directly from (5.1.13) that the orthogonal complement of V[t] with respect to B (t)1 is given by: (5.1.14) V[t] = B (t) V[t]o = v (t) : v V and v (t) = 0 .

5.1. SYMPLECTIC DIFFERENTIAL SYSTEMS

137

5.1.6. D EFINITION . We say that t ]a, b] is a focal instant for the pair (X, 0 ) (or that t is a (X, 0 )-focal instant) if there exists a non zero solution v V of (X, 0 ) such that v (t) = 0; the dimension of the space of solutions v V such that v (t) = 0 is called the multiplicity of the focal instant t, and it is denoted by mul(t). The signature of the focal instant t, denoted by sgn(t), is dened as the signature of the restriction of the symmetric bilinear form B (t)1 to the space V[t] , that is: sgn(t) = sgn B (t)1 |V[t] V[t] , where the orthogonal complement V[t] is taken relatively to the bilinear form B (t)1 . We say that the focal instant t is nondegenerate if B (t)1 is nondegenerate on V[t] . If t ]a, b] is not a (X, 0 )-focal instant, we dene mul(t) = 0 and sgn(t) = 0; if the pair (X, 0 ) has only a nite number of focal instants, we dene the focal index of (X, 0 ) as the integer number: ifoc (X,
0)

= ifoc =
t]a,b]

sgn(t).

5.1.7. R EMARK . For all t ]a, b] we have: (5.1.15) mul(t) = dim V[t]o = codimI Rn V[t],
0 )-focal if and only if V[t]

and in particular t is (X, (5.1.13) that the map

=I Rn ; indeed, it follows from

v V : v (t) = 0

v v (t) V[t]o I Rn

is an isomorphism. Keeping in mind Remark 5.1.2, we see that the symmetric bilinear form B (t)1 |V[t] V[t] is the push-forward of B (t)|V[t]o V[t]o by the isomorphism B (t)|V[t]o : V[t]o V[t] ; Then, we conclude that the signature of a (X,
0 )-focal

instant t coincides with:

sgn(t) = sgn B (t)|V[t]o V[t]o ; moreover, a focal instant t is nondegenerate if and only if V[t]o is a nondegenerate subspace for B (t). Observe also that Corollary 1.1.11 implies that a focal instant t is nondegenerate if and only if B (t)1 is nondegenerate in V[t]. 5.1.8. D EFINITION . We say that the Lagrangian initial condition determined by the Lagrangian subspace 0 (or, equivalently, by the pair (P, S )) is nondegenerate if the bilinear form B (a)1 is nondegenerate on P . In this case, we also say that the pair (X, 0 ) has a nondegenerate initial condition. 5.1.9. R EMARK . In Denition 5.1.6 we have explicitly excluded the possibility that t = a be a (X, 0 )-focal instant. Nevertheless, if we admit for the moment the terminology of Denition 5.1.6 also for t = a, we see that the nondegeneracy for the initial condition determined by 0 is indeed equivalent to the nondegeneracy of t = a as a focal instant. Arguing as in Remark 5.1.7 we see that the nondegeneracy of the initial condition determined by 0 is equivalent to the nondegeneracy of B (a) in the annihilator of P in I Rn , and also to the nondegeneracy of B (a)1 in

138

5. APPLICATIONS

P , where the orthogonal complement is taken with respect to B (a)1 . Observe that if P = I Rn , which is equivalent to 0 being transversal to {0}n I Rn , or if n n P = {0}, which is equivalent to 0 = {0} I R , then automatically (X, 0 ) has nondegenerate initial condition. 5.1.10. E XAMPLE . If g : [a, b] Bsym (I Rn ) and R : [a, b] Lin(I Rn ) are differentiable maps with g (t) nondegenerate and R(t) a g (t)-symmetric operator for all t, then the homogeneous linear differential equation: (5.1.16) g (t)1 g (t) v (t) = R(t) v (t), t [a, b], is called a Morse-Sturm equation. If g is constant, then (5.1.16) can be written in a simplied form: (5.1.17) v (t) = R(t) v (t), t [a, b]. Dening (t) = g (t) v (t), we can rewrite (5.1.16) as a system of differential equations: d v (t) v (t) 0 g (t)1 = (5.1.18) , t [a, b]. (t) g (t) R(t) 0 dt (t) The system (5.1.18) is a symplectic differential system, with A(t) = 0, B (t) = g (t)1 and C (t) = g (t) R(t). In general, any symplectic differential system X with A = 0 will be identied with a MorseSturm equation with g (t) = B (t)1 and R(t) = B (t) C (t). 5.2. The Maslov Index of a Symplectic Differential System In this section we show that if X is a symplectic differential system and 0 is a Lagrangian subspace of I Rn I Rn , then we can associate in a natural way a curve in the Lagrangian Grassmannian to the pair (X, 0 ), and under suitable hypotheses we can associate a Maslov index to the pair (X, 0 ). We will always denote by L0 the Lagrangian subspace: L0 = {0}n I Rn I Rn I Rn and by the Lagrangian Grassmannian of the symplectic space I Rn I Rn endowed with its canonical symplectic structure. As usual, we will denote by A, B and C the components of X and by its fundamental matrix. We dene a differentiable curve : [a, b] by setting: (5.2.1) (5.2.2) Observe that (a) = (5.2.3)
0;

(t) = (t)( 0 ), (t) = {(v (t), v (t)) : v V}. keeping in mind (5.2.2) and (5.1.13) we see that: L0 (t) = {0}n V[t]o ,

for all t [a, b]; more explicitly, we have:

for all t [a, b]. We have the following: 5.2.1. L EMMA . An instant t ]a, b] is (X, 0 )-focal if and only if (t) 1 (L0 ); moreover, (t) k (L0 ) if and only if mul(t) = k .

5.2. THE MASLOV INDEX OF A SYMPLECTIC DIFFERENTIAL SYSTEM

139

P ROOF. It follows easily from (5.2.3) and (5.1.15). Lemma 5.2.1 is a rst indication that the properties of the focal instants of the pair (X, 0 ) may be investigated by looking at the intersections of the curve with 1 (L0 ). In order to make more explicit the relations between the focal instants of (X, 0 ) and such intersections we now compute the derivative of . The linearization of the natural action of the symplectic group Sp(I Rn I Rn , ) in the Lagrangian Grassmannian gives us an anti-homomorphism of the Lie algebra sp(I Rn I Rn , ) into the Lie algebra of differentiable vector elds in . These concepts were dened in Subsection 2.1.3, and we will use here the notations of that subsection. The identity (5.1.5) tells us that is an integral curve of the time-dependent vector eld (t, g ) X (t)R (g ) in the Lie group Sp(I Rn I Rn , ); from Remark 2.1.22 and from (5.2.1) it then follows that is an integral curve of the timedependent vector eld (t, m) X (t) (m) in , i.e., (5.2.4) (5.2.5) (t) = X (t) ( (t)), X (t) (L) = (X (t), )|LL , (t) (0, ), (0, ) = X (t)(0, ), (0, ) = (B (t), A(t) ), (0, ) = B (t)(, ), for all t [a, b]. Proposition 2.5.9 gives us: for all L . Then, putting together (5.2.4) and (5.2.5) we obtain: (5.2.6)

for any (0, ), (0, ) L0 (t). We have therefore shown the following: 5.2.2. L EMMA . For all t [a, b], the restriction of the symmetric bilinear form B (t) Bsym (I Rn ) to V[t]o coincides with the push-forward of the restriction of (t) Bsym ( (t)) to L0 (t) by the isomorphism: L0 (t) (0, ) V[t]o P ROOF. It follows from (5.2.3) and (5.2.6). 5.2.3. C OROLLARY. An (X, 0 )-focal instant t ]a, b] is nondegenerate if and only if has a nondegenerate intersection with 1 (L0 ) at the instant t; moreover, sgn(t) = sgn (t)|L0
(t)

Also, the pair (X, 0 ) has nondegenerate initial condition if and only if either has a nondegenerate intersection with 1 (L0 ) at the instant t = a or (a) 1 (L0 ). P ROOF. It follows from Remark 5.1.7 and Remark 5.1.9. 5.2.4. C OROLLARY. If t0 ]a, b] is a nondegenerate (X, 0 )-focal instant, then it is isolated, i.e., no instant t = t0 sufciently close to t0 is (X, 0 )-focal. Moreover, if the initial condition of (X, 0 ) is nondegenerate, then there are no (X, 0 )-focal instants in a neighborhood of t = a. If (X, 0 ) has nondegenerate initial condition and if it has only nondegenerate focal instants, then (X, 0 ) has only a nite number of focal instants.

140

5. APPLICATIONS

P ROOF. It follows from Corollary 5.2.3, Lemma 5.2.2 and Example 4.2.17. We now want to dene the Maslov index of a pair (X, 0 ); essentially, it will be dened as the Maslov index L0 ( ) of the curve . The problem is that L0 ( ) only makes sense if has endpoints in 0 (L0 ); assuming that t = b is not a (X, 0 )focal instant, we have that (b) 0 (L0 ) (see Lemma 5.2.1). However, in general (a) = 0 may be in 1 (L0 ); to overcome this problem the idea is to erase a short initial portion of the curve . More precisely, let us give the following: 5.2.5. D EFINITION . If the pair (X, 0 ) has nondegenerate initial condition and if the nal instant t = b is not (X, 0 )-focal, then we dene the Maslov index imaslov (X, 0 ) of the pair (X, 0 ) by setting: imaslov (X,
0)

= imaslov = L0 |[a+,b] ,
0 )-focal

where > 0 is chosen in such a way that there are no (X, interval [a, a + ].

instants in the

From Corollary 5.2.4 it follows that, indeed, there exists > 0 such that (X, 0 ) does not have focal instants in the interval [a, a + ]. Moreover, the denition of imaslov does not depend on the choice of (see Exercise 5.4). Generically, the Maslov index imaslov (X, 0 ) can be thought as a sort of algebraic count of the focal instants of (X, 0 ): 5.2.6. P ROPOSITION . Suppose that (X, 0 ) has nondegenerate initial condition and that t = b is not (X, 0 )-focal. If (X, 0 ) has only nondegenerate focal instants, then the focal index coincides with the Maslov index: ifoc (X,
0)

= imaslov (X,

0 ).

P ROOF. It follows directly from Corollary 4.2.18 and Corollary 5.2.3. 5.2.7. E XAMPLE . If B (t) is positive denite for some (hence for all) t [a, b] then (X, 0 ) automatically has nondegenerate initial condition; moreover, every (X, 0 )-focal instant t ]a, b] is nondegenerate and sgn(t) = mul(t). Hence, if t = b is not (X, 0 )-focal, it follows from Proposition 5.2.6 that imaslov (X,
0)

=
t]a,b[

mul(t) < +.

One of the fundamental properties of the Maslov index of a pair is its stability; we have the following: 5.2.8. P ROPOSITION . Let X be a topological space and assume that for all X it is given a symplectic differential system X such that the map: X [a, b] (, t) X (t) sp(I Rn I Rn , ) is continuous; let 0 : X be a continuous curve in the Lagrangian Grassmannian such that dim(L0 0 ()) does not depend on X . If for some 0 X the pair (X0 , 0 (0 )) has nondegenerate initial condition and t = b is not (X0 , 0 (0 ))-focal, then there exists a neighborhood U of 0 in X such that for all U we have:

5.3. SEMI-RIEMANNIAN GEODESICS AND HAMILTONIAN SYSTEMS

141

(X , 0 ()) has nondegenerate initial condition; the instant t = b is not (X , 0 ())-focal; imaslov (X , 0 ()) = imaslov (X0 , 0 (0 )). P ROOF. Denote by the fundamental matrix of the symplectic differential system X . It follows from standard theory on continuous dependence with respect to a parameter of solutions of differential equations that the map (, t) (t) is continuous in X [a, b]. Dene (t) = (t)( 0 ()); then clearly (, t) (t) is a continuous map in X [a, b] and it follows from (5.2.4) that also (, t) 0 (t) is continuous in X [a, b]. In particular,since (L0 ) is open, we have 0 that (b) (L0 ) for in a neighborhood of 0 in X , and therefore for such values of the instant t = b is not (X , 0 ())-focal (see Lemma 5.2.1). Keeping in mind Corollary 5.2.3, it follows directly from Lemma 4.2.19 that there exists > 0 and a neighborhood U of in X such that (X , 0 ()) has nondegenerate initial condition, and such that there are no (X , 0 ())-focal instants in the interval [a, a + ] for all U. Hence, imaslov (X ,
0 ())

= L0

|[a+,b]

for all in a neighborhood of 0 in X . It follows from Remark 3.1.20 that the map C 0 ([a + , b], ) is continuous when we consider C 0 ([a + , b], ) endowed with the compact-open topology. By Property (6) in the statement of Lemma 4.2.13 we conclude that imaslov (X , 0 ()) is constant when runs in a neighborhood of 0 in X ; this concludes the proof. 5.2.9. C OROLLARY. Suppose that it is given a sequence (Xn )n1 of differentiable maps Xn : [a, b] sp(I Rn I Rn , ) that converges uniformly to a differentiable map X ; assume that X and Xn are symplectic differential systems for all n. Let also be given a sequence ( n 0 ) of Lagrangian subspaces that converges to some 0 , where dim(L0 n ) = dim(L0 0 ) for all n. Then, if (X, 0 ) has 0 nondegenerate initial condition and if t = b is not a (X, 0 )-focal instant, then for all n sufciently large also (Xn , n 0 ) has nondegenerate initial condition and t = b is not (Xn , n ) -focal, and: 0 imaslov (Xn ,
n 0)

= imaslov (X0 ,

0 ).

P ROOF. Consider the topological space X = I N {+}, where U X is open if and only if U I N or X \ U is a nite subset of I N . Setting X+ = X , then it is easy to see that the hypotheses of Proposition 5.2.8 are satised for 0 = +. The conclusion follows. 5.3. The Maslov Index of semi-Riemannian Geodesics and Hamiltonian Systems In this section we show how the theory of symplectic differential systems appears in several elds of geometry. In Subsection 5.3.1 we show how every geodesic in a semi-Riemannian manifold determines in a natural way a MorseSturm equation; such system is essentially obtained from the Jacobi equation along

142

5. APPLICATIONS

the geodesic through a parallel trivialization of the tangent bundle of the semiRiemannian manifold along the geodesic. In Subsection 5.3.2 we show that symplectic differential systems appear also as linearizations of Hamiltonian systems; we develop the theory in a very abstract and general formalism, using arbitrary symplectic manifolds. Finally, in Subsection 5.3.3 we give references for some further developments of the theory. 5.3.1. Geodesics in a semi-Riemannian manifold. Let M be a differentiable manifold; a semi-Riemannian metric in M is a differentiable (2, 0)-tensor eld g such that for every m M , gm is a nondegenerate symmetric bilinear form on Tm M . The pair (M, g) is called a semi-Riemannian manifold; when gm is positive denite for every m M we say that g is a Riemannian metric and that (M, g) is a Riemannian manifold. It is well known that there exists a unique connection on the tangent bundle T M of M which is torsion-free and such that g is parallel; such connection is called the Levi-Civita connection. The curvature tensor of is dened by: R(X, Y )Z = X Y Z Y X Z [X,Y ] Z, for every differentiable vector elds X, Y, Z in M . Given a differentiable curve : I M dened in some interval I I R and given a differentiable vector eld v : I T M along , i.e., v(t) T (t) M for every t I , then we denote by Dv dt (or just by v ) the covariant derivative of v along ; a geodesic is dened as a differentiable curve : I M whose derivative is parallel, i.e., = 0. A differentiable vector eld v along a geodesic : [a, b] M is called a Jacobi vector eld if it satises the differential equation: (5.3.1) D2 v(t) = R (t), v(t) (t), dt2 t [a, b];

equation (5.3.1) is known as the Jacobi equation. Set dim(M ) = n and chose parallel vector elds Zi : [a, b] T M along , i = 1, . . . , n such that (Zi (t))n i=1 form a basis of T (t) M for some (and hence for all) t [a, b]; we say that (Zi )n i=1 is a parallel trivialization of the tangent bundle T M along . The parallel trivialization (Zi )n i=1 induces a bijection between the set of differentiable vector elds v : [a, b] T M along and the set of differentiable maps v : [a, b] I Rn given by:
n

(5.3.2)

v(t) =
i=1

vi (t)Zi (t),

t [a, b],

where v (t) = (v1 (t), . . . , vn (t)) I Rn ; taking the covariant derivative along on both sides of (5.3.2) we get that if v corresponds to v then the covariant derivative v of v corresponds to the (standard) derivative v of v by means of the bijection induced by the parallel trivialization. For each t [a, b] we dene a nondegenerate symmetric bilinear form g (t) Bsym (I Rn ) and a linear operator R(t) Lin(I Rn )

5.3. SEMI-RIEMANNIAN GEODESICS AND HAMILTONIAN SYSTEMS

143

whose matrices with respect to the canonical basis of I Rn satisfy the identities:
n

gij (t) = g Zi (t), Zj (t) ,

R (t), Zj (t) (t) =


i=1

Rij (t)Zi (t),

for every i, j = 1, . . . , n. Observe that, since both g and the vector elds Zi are parallel, the symmetric bilinear form g (t) does not depend on t; moreover, standard symmetry properties of the curvature tensor imply that R(t) is g -symmetric for every t [a, b]. We can therefore consider the Morse-Sturm equation: (5.3.3) v (t) = R(t)v (t), t [a, b];

moreover, it is easy to see that a vector eld v along is a Jacobi vector eld iff the map v : [a, b] I Rn dened by (5.3.2) is a solution of (5.3.3). In Example 5.1.10 we have mentioned that every Morse-Sturm equation can be identied with a symplectic differential system X with components A(t) = 0, B (t) = g (t)1 and C (t) = g R(t); observe that v (t) = g (v (t)) I Rn for every t. Consider now a submanifold P M ; the second fundamental form of P at a point p P in a normal direction n Tp P (where the orthogonal complement is taken with respect to g) is the symmetric bilinear form Sn on Tp P dened by: Sn (v, w) = g(v W, n), v, w Tp P, where W is any differential vector eld which is tangent to P and such that W (p) = w. Suppose now that we have a geodesic : [a, b] M with (a) P and (a) T (a) P ; a vector eld v : [a, b] T M along is called a P -Jacobi eld if v is Jacobi and satisfy the condition: v (a) P and g v (a), |T (a) P + S (a) v (a), = 0 T (a) P . The basis (Zi (a))n Rn i=1 of T (a) M induces an isomorphism from T (a) M to I n which takes T (a) P onto some subspace P I R ; moreover, there exists a unique symmetric bilinear form S Bsym (P ) which is the push-forward of S (a) by (the restriction of) such isomorphism. The pair (P, S ) therefore denes a Lagrangian subspace 0 I Rn I Rn as in (5.1.8); it is easily seen that a vector eld v along is P -Jacobi if and only if the corresponding map v : [a, b] I Rn dened by (5.3.2) is a solution of (X, 0 ). In semi-Riemannian geometry one usually denes that a point (t), t ]a, b] is P -focal along when there exists a non zero P -Jacobi vector eld v along such that v(t) = 0; the dimension of the space of P -Jacobi elds v along such that v(t) = 0 is called the multiplicity of the P -focal point (t). Moreover, for each t ]a, b] one considers the space: J[t] = v(t) : v is P -Jacobi along T (t) M. The signature of the P -focal point (t) is dened as the signature of the restriction of the metric g to the orthogonal complement J[t] of J[t]; the P -focal point (t) is called nondegenerate if the space J[t] (or equivalently, J[t]) is nondegenerate for g. When there are only a nite number of P -focal points along we dene the focal index of the geodesic with respect to P as the sum of the signatures of the P -focal points along . The following facts are obvious:

144

5. APPLICATIONS

an instant t ]a, b] is (X, 0 )-focal iff (t) is a P -focal point; the multiplicity and signature of t as a (X, 0 )-focal instant or of (t) as a P -focal point coincide; the focal index of the pair (X, 0 ) coincide with the focal index of the geodesic with respect to P ; a (X, 0 )-focal instant t ]a, b] is nondegenerate if and only if the P focal point (t) is non-degenerate; the initial condition of (X, 0 ) is nondegenerate if and only if T (a) P is a nondegenerate subspace for g. If g is nondegenerate on T (a) P and if (b) is not a P -focal point we dene the Maslov index of the geodesic with respect to P as the Maslov index of the pair (X, 0 ). From Proposition 5.2.6 it follows immediately the following: 5.3.1. P ROPOSITION . Let : [a, b] M be a geodesic starting orthogonally to a submanifold P M ; suppose that the metric g is nondegenerate on T (a) P and that there are only nondegenerate P -focal points along . Then, if (b) is not P -focal, the Maslov index of coincides with the focal index of with respect to P. Observe that if (M, g) is Riemannian then the focal index of a geodesic is simply the sum of the multiplicities of the P -focal points along ; this (nonnegative) integer is sometimes called the geometric index of the geodesic . The geometric index of a geodesic is one of the numbers which enters into the statement of the celebrated Morse Index Theorem. A semi-Riemannian manifold (M, g) is called Lorentzian if the metric g has index 1 at every point; four-dimensional Lorentzian manifolds are mathematical models for general relativistic spacetimes. A vector v T M is said to be timelike, lightlike or spacelike respectively when g(v, v ) is negative, zero or positive. Similarly, we say that a geodesic is timelike, lightlike or spacelike when (t) is respectively timelike, lightlike or spacelike for all t. We have the following: 5.3.2. L EMMA . Let : [a, b] M be a timelike or a lightlike (non constant) geodesic in a Lorentzian manifold (M, g) starting orthogonally to a submanifold P M ; assume that g is nondegenerate on T (a) P (which is always the case if is timelike). Then, for every t ]a, b] the space J[t] is g-positive. P ROOF. Set v(t) = (ta) (t); then v is a P -Jacobi eld and therefore (t) J[t] for t ]a, b]]. It follows that J[t] is contained in the orthogonal complement of I R (t); if is timelike, this implies that g is positive denite on J[t]. If is lightlike we still have to show that (t) is not in J[t]; observe rst that, since T (a) P is g-nondegenerate it cannot be (a) T (a) P and therefore it cannot be T (a) P I R (a) ; hence we can nd a Jacobi eld v along with v(a) = 0 and v (a) orthogonal to T (a) P but not orthogonal to (a). Then v is a P -Jacobi eld and it cannot be that v(t) is orthogonal to (t); for, s g(v(s), (s)) is an afne map which vanishes at s = a and hence it cannot have another zero at s = t, since this would imply g(v, ) 0 and g(v (a), (a)) = 0. This proves that (t) J[t] and completes the proof.

5.3. SEMI-RIEMANNIAN GEODESICS AND HAMILTONIAN SYSTEMS

145

5.3.3. C OROLLARY. Under the hypotheses of Lemma 5.3.2, the geometric index and the focal index of with respect to P coincide. 5.3.2. Hamiltonian systems. In this subsection we will consider the following setup. Let (M, ) be a symplectic manifold, i.e., M is a smooth manifold and is a smooth closed skew-symmetric nondegenerate two-form on M, so that m is a = symplectic form on Tm M for each m M . We set dim(M) = 2n. Let H :U I R be a smooth function dened in an open set U I R M; we will call such function a Hamiltonian in (M, ). For each t I R, we denote by Ht the map m H (t, m) dened in the open set Ut M consisting of those m M such that (t, m) U . We denote by H the smooth time-dependent vector eld in M dened by dHt (m) = (H (t, m), ) for all (t, m) U ; let F denote the maximal ow of the vector eld H dened on an open set of I RI R M taking values in M, i.e., for each m M and t0 I R, the curve t F (t, t0 , m) is a maximal integral curve of H and F (t0 , t0 , m) = m. This means that F (, t0 , m) is a maximal solution of the equation: d F (t, t0 , m) = H (t, F (t, t0 , m)), F (t0 , t0 , m) = m. dt Recall that F is a smooth map; we also write Ft,t0 for the map m F (t, t0 , m); observe that Ft,t0 is a diffeomorphism between open subsets of M. We recall that a symplectic chart in M is a local chart (q, p) taking values in I Rn I Rn whose differential at each point is a symplectomorphism from the tangent space of M to I Rn I Rn endowed with the canonical symplectic struc , }, ture. We write q = (q1 , . . . , qn ) and p = (p1 , . . . , pn ); we denote by { q i pj i, j = 1, . . . , n the corresponding local referential of T M, and by {dqi , dpj } the local referential of T M . By Darbouxs Theorem, there always exists an atlas of symplectic charts. In a given symplectic chart (q, p), we have:
n n

=
i=1

dqi dpi ,

H=
i=1

H H pi qi qi pi

Let P be a Lagrangian submanifold of M, i.e., Tm P is a Lagrangian subspace of Tm M for every m P . We x an integral curve : [a, b] M of H , so that (t) = F (t, a, (a)) for all t [a, b]. We also say that is a solution of the Hamilton equation, i.e., in a symplectic chart (t) = (q (t), p(t)): H dq = , dt p (5.3.4) dp H = . dt q We assume that starts at P , that is (a) P . Finally, we will consider a xed smooth distribution L in M such that Lm is a Lagrangian subspace of Tm M for all m M.

146

5. APPLICATIONS

The basic example to keep in mind for the above setup is the case where M is the cotangent bundle T M of some smooth manifold M endowed with the canonical symplectic structure, P is the annihilator T P o of some smooth submanifold P of M , and L is the distribution consisting of the vertical subspaces, i.e., the subspaces tangent to the bers of T M . The Hamiltonian ow Ft,t0 is a symplectomorphism: 5.3.4. P ROPOSITION . The symplectic form is invariant by the Hamiltonian = for all (t, t ). ow F , i.e., Ft,t 0 0 In the case of a time-independent Hamiltonian, the result of Proposition 5.3.4 follows easily from the formula for the Lie derivative of forms: LH = d iH + iH d. For the general case, the proof is based on the following elementary Lemma which says how to compute the derivative of the pull-back of forms by a one-parameter family of functions: 5.3.5. L EMMA . Let G be a smooth map on an open subset of I R M taking values in M and let be a smooth r-form on M. For each t I R, denote by Gt t the (r 1)-form on an open subset of M given the map m G(t, m) and by G by: t )m = dGt (m) iv G(t,m) , (G
d where v = d t G(t, m) and iv is the interior product (or contraction in the rst variable) of a form with the vector v . Then, for all m M we have: d t )m + G t (d )m . (5.3.5) (G )m = d(G dt t P ROOF. The two sides of equality (5.3.5) are I R-linear maps of which have the same behavior with respect to exterior derivative and exterior products. Moreover, they agree on 0-forms. The conclusion follows from the fact that, locally, every r-form is a linear combination of products of derivatives of 0-forms.

P ROOF OF P ROPOSITION 5.3.4. We x an instant t0 I R; we consider the map G(t, m) = F (t, t0 , m) and we apply Lemma 5.3.5 to = . Observe that, t is equal to G (dHt ). From (5.3.5), we get: for each t, the 1-form G t d (F )m = 0, dt t,t0 is independent of t. The conclusion follows from the fact that F and so Ft,t t0 ,t0 0 is the identity map. A sextuplet (M, , H, L, , P ) where (M, ) is a symplectic manifold, H is a (time-dependent) Hamiltonian function dened on an open subset of I R M, L is a smooth distribution of Lagrangians in M, : [a, b] M is an integral curve of H and P is a Lagrangian submanifold of M with (a) P , will be called a set of data for the Hamiltonian problem. We give some more basic denitions.

5.3. SEMI-RIEMANNIAN GEODESICS AND HAMILTONIAN SYSTEMS

147

5.3.6. D EFINITION . A vector eld along in M is said to be a solution for the linearized Hamilton (LinH) equations if it satises: (5.3.6) (t) = dFt,a ((a)) (a).

We also say that is a P -solution for the (LinH) equations if in addition it satises (a) T(a) P . 5.3.7. D EFINITION . A point (t), t ]a, b] is said to be a P -focal point along if there exists a non zero P -solution for the (LinH) equations such that (t) L(t) . The multiplicity of a P -focal point (t) is the dimension of the vector space of such s. 5.3.8. D EFINITION . A symplectic L-trivialization of T M along is a smooth family of symplectomorphisms (t) : I Rn I Rn T(t) M such that, for all t [a, b], (t)(L0 ) = L(t) , where L0 = {0} I Rn . The existence of symplectic L-trivializations along is easily established with elementary arguments, using the fact that T M restricts to a trivial vector bundle along . We will be interested also in the quotient bundle T M/L and its dual bundle. We have an obvious canonical identication of the dual (T M/L) with the annihilator Lo T M ; moreover, using the symplectic form, we will identify Lo with L by the isomorphism: (5.3.7) Tm M (, ) Tm M , m M.

A symplectic L-trivialization induces a trivialization of the quotient bundle T M/L along , namely, for each t [a, b] we dene an isomorphism Zt : I Rn T(t) M/L(t) : (5.3.8) Zt (x) = (t)(x, 0) + L(t) , xI Rn . Given a symplectic L-trivialization of T M along , we dene a smooth curve : [a, b] Sp(2n, I R) by: (5.3.9) (t) = (t)1 dFt,a ((a)) (a).

The fact that (t) is a symplectomorphism follows from Proposition 5.3.4. We now dene a smooth curve X : [a, b] sp(2n, I R) by setting: (5.3.10) X (t) = (t)(t)1 ;

As customary, the components of the matrix X will be denoted by A, B and C . Finally, we dene a Lagrangian subspace 0 of I Rn I Rn by: (5.3.11)
0

= (a)1 (T(a) P ).

5.3.9. D EFINITION . The canonical bilinear form of (M, , H, L, , P ) is a family of symmetric bilinear forms HL (t) on (T(t) M/L(t) ) Lo L(t) (t) given by: (5.3.12)
HL (t) = Zt B (t) Zt ,

148

5. APPLICATIONS

where Z is the trivialization of T M/L relative to some symplectic L-trivialization of T M and B is the upper-right n n block of the map X in (5.3.10). In Exercise 5.5 the reader is asked to prove that the right hand side of (5.3.12) does not depend on the choice of the symplectic L-trivialization of T M. We say that the set of data (M, , H, L, , P ) is nondegenerate if HL (t) is nondegenerate for all t [a, b]. In this case, we can also dene the symmetric bilinear form HL (t)1 on T(t) M/L(t) . Given a nondegenerate set of data (M, , H, L, , P ), let us consider the pair (X, 0 ) dened by (5.3.10) and (5.3.11). It is easily seen that the submanifold P and the space P dened by 0 as in (5.1.8) are related by the following: Za (P ) = (T(a) P ), where : T(a) M T(a) M/L(a) is the quotient map. We set: (5.3.13) P0 = (T(a) P ).

To dene the signature of a P -focal point along , we need to introduce the following space: (5.3.14) V[t] = (t) : is a P -solution of the (LinH) equation L(t) , t [a, b]. Using the isomorphism (5.3.7), it is easy to see that V[a] is identied with the annihilator (T(a) P + La )o . It is easily seen that a point (t) is P -focal if and only if V[t] is not zero and that the dimension of V[t] is precisely the multiplicity of (t). 5.3.10. D EFINITION . Let (t) be a P -focal point along the solution . The signature sgn((t)) is the signature of the restriction of HL (t) to V[t] Lt Lo t. (t) is said to be a nondegenerate P -focal point if such restriction is nondegenerate. If has only a nite number of P -focal points, we dene the focal index ifoc () as: (5.3.15) ifoc () =
t ]a,b]

sgn((t)).

5.3.11. D EFINITION . Given a set of data (M, , H, L, , P ) such that: HL (a)1 is nondegenerate on P0 = T(a) P , where : T(a) M T(a) M/L(a) is the quotient map; (b) is not a P -focal point. We dene the Maslov index imaslov () as the Maslov index of any pair (X, associated to it by a symplectic L-trivialization of T M along .
0)

5.3.3. Further developments. In this short subsection we indicate some recent results by the authors of this book that contain further development of the theory of the Maslov index and its applications to semi-Riemannian geometry and Hamiltonian system.

EXERCISES FOR CHAPTER 5

149

In [29] prove the stability of the geometric index for timelike and lightlike Lorentzian geodesics (that follows from Corollary 5.3.3), it is given a counterexample to the equality of the Maslov index and of the focal index of a geodesic in the case that there are degenerate focal points. It is also studied the problem of characterizing those curves of Lagrangians that arise from the Jacobi equation along a semi-Riemannian geodesic. In [12] it is proven a Lorentzian extension of the Morse Index Theorem for geodesics of all causal character in a stationary Lorentzian manifold M, or, more generally, for geodesics that admit a timelike Jacobi eld. It is considered the case of a geodesic with initial endpoint variable in a submanifold P of M. Moreover, under suitable compactness assumptions it is developed an innite dimensional Morse theory for geodesics with xed endpoints in a stationary Lorentzian manifold. A version of the index theorem for periodic geodesics in stationary Lorentzian manifolds is proven in [25]. The Morse Index Theorem for timelike or lightlike geodesics in any Lorentzian manifold is proven in [2, 8]; for the case of both endpoints variable see [34]. In [36] the authors develop the Morse Index Theorem for the general case of a non periodic solution of a possibly time-dependent Hamiltonian system; it is used a suitable assumption that generalize the assumption of stationarity for the metric used in [12]. A general version of the semi-Riemannian Morse Index Theorem is proven in [45]; the Maslov index is proven to be equal to the difference of the index and of the coindex of suitable restrictions of the index form. Exercises for Chapter 5 E XERCISE 5.1. Consider the symplectic differential system given in formula (5.1.3) and initial condition (5.1.7), with X : [a, b] sp(I Rn I Rn , ) real1 analytic. Prove that either every instant t ]a, b] is (X, 0 )-focal, or else there are only a nite number of (X, 0 )-focal instants. Prove that if the initial condition is nondegenerate, then there are only a nite number of (X, 0 )-focal instants.
1A map f : U I Rn dened in an open subset U I Rm is said to be real-analytic if for all

x U , in a neighborhood of x0 we can write f as the sum of a power series centered at x0 , i.e., f (x) =
1 m , a (x1 x0 (xm x0 1) m)

for x near x0 , where = (1 , . . . , m ) runs over the set of all m-tuples of non negative integer numbers. A power series centered at x0 and convergent in a neighborhood of x0 , converges absolutely and uniformly in a (possibly smaller) neighborhood of x0 . It follows that the series above can be differentiated termwise; in particular, every real-analytic function is C , and the coefcient a is given by the Taylors formula: a = 1 || f 0 (x0 1 , . . . , xm ), m 1 1 ! m ! x 1 . . . xm

where || = 1 + + m . It follows easily from the two formulas above that the set of points of U where f and all its partial derivatives are zero is open and closed in U . In particular, a real-analytic function on a connected domain which is zero in a non empty open set is identically zero.

150

5. APPLICATIONS

E XERCISE 5.2. Consider the isomorphism O : I Rn I Rn I Rn I Rn dened by O(v, ) = (v, ); in terms of matrices: O= I 0 , 0 I

where I denotes the n n identity matrix. Dene: X op = O X O; prove that X op is a symplectic differential system and compute its components Aop , B op and C op , and its fundamental matrix op . The system X op is called the opposite symplectic differential system of X . Characterize the solutions of X op in terms of those of X . E XERCISE 5.3. In the notations of Exercise 5.2, prove that O : I Rn I Rn I Rn I Rn is not a symplectomorphism (with respect to the canonical symplectic structure), but it takes Lagrangian subspaces into Lagrangian subspaces. Given Rn I Rn , ), denote by op 0 (I 0 the Lagrangian O ( 0 ), which is called the opposite Lagrangian subspace of 0 . Denote by (P, S ) and (P op , S op ) respectively the pair associated to 0 and to the opposite Lagrangian subspace op 0 . Determine the relation between P and P op and between S and S op ; prove that (X, 0 ) has a nondegenerate initial condition if and only if (X op , op 0 ) does, and, in this case, nd the relation between the focal index (whenever dened) and the Maslov index of (X, 0 ) and of (X op , op 0 ). E XERCISE 5.4. Prove that, in Denition 5.2.5, the Maslov index of the curve |[a+,b] does not depend on the choice of > 0 provided that there are no (X, 0 )focal instants in the interval ]a, a + ]. E XERCISE 5.5. Show that the symmetric bilinear form HL introduced in 5.3.9 does not indeed depend on the choice of the symplectic trivialization.

APPENDIX A

Answers and Hints to the exercises


A.1. From Chapter 1 Exercise 1.1. The naturality of the isomorsm (1.1.1) means that: Lin(V, W ) B(V, W ) B(L,M ) Lin(L,M )
) Lin(V1 , W1 B(V1 , W1 )
= =

where L Lin(V1 , V ), M Lin(W1 , W ) and the horizontal arrows in the diagram are suitable versions of the isomorphism (1.1.1). Exercise 1.2. Write B = Bs + Ba , with Bs (v, w) = and Ba (v, w) = 1 2 B (v, w ) B (w, v ) . Exercise 1.3. Use formula (1.2.1). Exercise 1.4. Every v V can be written uniquely as v = j J zj bj , where zj = xj + i yj , xj , yj I R; therefore v can be written uniquely as a linear combination of the bj s and of the J (bj )s as v = j J xj bj + yj J (bj ). Exercise 1.5. The uniqueness follows from the fact that (V ) generates V C as (v ) = f 1 (v ) + i f a complex vector space. For the existence dene f 1 (v ), where and are the real part and the imaginary part operator relative to the real form (V ) of V C . Exercise 1.6. Use Proposition 1.3.3 to get maps : V1C V2C and : V2C V1C such that 1 = 2 and 2 = 1 ; the uniqueness of Proposition 1.3.3 gives the uniqueness of the . Using twice again the uniqueness in Proposition 1.3.3, one concludes that = Id and = Id. Exercise 1.7. If Z = U C , then obviously c(Z ) Z . Conversely, if c(Z ) Z then (Z ) and (Z ) are contained in U = Z V . It follows easily that Z = U C . Exercise 1.8. In the case of multi-linear operators, diagram (1.3.2) becomes: V1C VpC
O
1 r f 1 2

B (v, w) + B (w, v )

V1 Vp
151

% /W

152

A. ANSWERS AND HINTS TO THE EXERCISES

The identities (1.3.5) still hold when T C is replaced by T C ; observe that the same conclusion does not hold for the identities (1.3.3) and (1.3.4). Lemma 1.3.10 generalizes to the case of multi-linear operators; observe that such generalization gives us as corollary natural isomorphisms between the complexication of the tensor, exterior and symmetric powers of V and the corresponding powers of V C . Lemma 1.3.11 can be directly generalized to the case that S is an anti-linear, multi-linear or sesquilinear operator; in the anti-linear (respectively, sesquilinear) T C must be replaced by T C (respectively, by T Cs ). For a C-multilinear operator S : V1C VpC V C (or if p = 2 and S is sesquilinear) the condition that S preserves real forms becomes: S (V1 Vp ) V, while the condition of commuting with conjugation becomes: S (c, . . . , c) = c S . Exercise 1.9. If B B(V ), then B (v, v ) = B(iv, iv ). Exercise 1.10. Use (1.4.4). Exercise 1.11. Set 2n = dim(V ) and let P L1 be a subspace and S Bsym (P ) be given. To see that the second term in (1.4.14) denes an n-dimensional subspace of V choose any complementary subspace W of P in L1 and observe that the map: L v + w v, L1 ,L0 (w)|Q P Q , v L1 , w L0 , is an isomorphism. To show that L is isotropic, hence Lagrangian, one uses the symmetry of S : (A.1.1) (v1 +w1 , v2 +w2 ) = L1 ,L0 (w1 )v2 L1 ,L0 (w2 )v1 = S (v1 , v2 )S (v2 , v1 ) = 0, for all v1 , v2 L1 , w1 , w2 L0 with v1 + w1 , v2 + w2 L. Conversely, let L be any Lagrangian; set P = 1 (L), where 1 : V L1 is the projection relative to the direct sum decomposition V = L0 L1 . If v P and w1 , w2 L0 are such that v + w1 , v + w2 L, then w1 w2 L L0 ; since P L + L0 , it follows that the functionals L1 ,L0 (w1 ) and L1 ,L0 (w2 ) coincide in P . Conclude that if one chooses w L0 such that v + w L, then the functional S (v ) = L1 ,L0 (w)|P P does not depend on the choice of w. One obtains a linear map S : P P ; using the fact that L is isotropic the computation (A.1.1) shows that S is symmetric. The uniqueness of the pair (P, S ) is trivial. Exercise 1.12. The equality T (0, ) = (0, ) holds iff B = 0 and A = . If B is invertible, then clearly the only solution is = 0; conversely, if B is not invertible, then there exists a non zero solution of the equations. Since B D is symmetric, then so is B 1 (B D)B 1 = DB 1 . Moreover, since DB 1 is symmetric, then so is A DB 1 A; substituting A D = (Id + B C ) , we get that the matrix (Id+ B C ) B 1 A = (Id+ C B )B 1 A = B 1 A +

A.1. FROM CHAPTER 1

153

C A is symmetric. Since C A is symmetric, then B 1 A is symmetric. Finally, substituting C = B 1 (D A Id) and using the fact that DB 1 = B 1 D , we get C DB 1 A B 1 = B 1 D A B 1 DB 1 A B 1 = B 1 B 1 , which is clearly symmetric. Exercise 1.13. T is symplectic iff, in the matrix representations with respect to a symplectic basis, it is T T = ; this is easily established using the equalities T T = and 2 = Id. Exercise 1.14. Clearly, if P, O Sp(2n, I R) then M = P O Sp(2n, I R). Conversely, recall from (1.4.6) that M is symplectic if and only if M = 1 M J ; applying this formula to M = P O we get: P O = 1 P O = 1 P 1 O . Since is an orthogonal matrix, then 1 P is again symmetric and positive denite, while 1 O is orthogonal. By the uniqueness of the polar decomposition, we get P = 1 P and O = 1 O which, by (1.4.6), implies that both P and O are symplectic. Exercise 1.15. Use Remark 1.4.7: a symplectic map T : V1 V2 V must be injective. Use a dimension argument to nd a counterexample to the construction of a symplectic map on a direct sum whose values on each summand is prescribed. Exercise 1.16 (Jv, Jw) = (J 2 w, v ) = (w, v ) = (v, w). Exercise 1.17. If J is g -anti-symmetric g (Jv, Jw) = g (v, J 2 w) = g (v, w). Exercise 1.18. Use induction on dim(V ) and observe that the gs -orthogonal complement of an eigenspace of V is invariant by T . Note that if T is Hermitian, the its eigenvalues are real; if T is anti-Hermitian, then its eigenvalues are pure imaginary. Exercise 1.19. The linearity of L0 ,L1 is obvious. Since dim(L1 ) = dim(L 0 ), it sufces to show that L0 ,L1 is surjective. To this aim, choose L 0 and extend to the unique V such that (w) = 0 for all w L1 . Since is nondegenerate on V , there exists v V such that = (v, ). Since L1 is maximal isotropic it must be v L1 , and L0 ,L1 is surjective. Exercise 1.20. Clearly, (L) is isotropic in (S /S, ). Now, to compute the dimension of (L) observe that: dim (L S ) = dim(L S ) dim(L S ), (L S ) = L + S = L + S , dim(L S ) + dim (L S ) = dim(V ), 1 1 dim(S 1 /S ) = dim(V ) dim(S ) = dim(L) dim(S ). 2 2 The conclusion follows easily.

154

A. ANSWERS AND HINTS TO THE EXERCISES

A.2. From Chapter 2 Exercise 2.1. Suppose that X is locally compact, Hausdorff and second countable. Then, one can write X as a countable union of compact sets Kn , n I N, such that Kn is contained in the interior int(Kn+1 ) of Kn+1 for all n. Set C1 = K1 and Cn = Kn \ int(Kn1 ) for n 2. Let X = U be an open cover of X ; for each n, cover Cn with a nite number of open sets V such that each V is contained in some U ; each V is contained in Cn1 Cn Cn+1 . It is easily seen that X = V is a locally nite open renement of {U } . Now, assume that X is locally compact, Hausdorff, paracompact, connected and locally second countable. We can nd a locally nite open cover X = U such that each U has compact closure. Construct inductively a sequence of compact sets Kn , n 1, in the following way: K1 is any non empty compact set, Kn+1 is the union (automatically nite) of all U such that U Kn is non empty. Since Kn int(Kn+1 ), it follows that n Kn is open; since n Kn is the union of a locally nite family of closed sets, then n Kn is closed. Since X is connected, X = n Kn . Each Kn can be covered by a nite number of second countable open sets, hence X is second countable. Exercise 2.2. Let p P be xed; by the local form of immersions there exist open sets U M and V N , with f0 (p) V U , and a differentiable map r : U V such that r|V = Id. Since f0 is continuous, there exists a neighborhood W of p in P with f0 (W ) V . Then, f0 |W = r f |W . Exercise 2.3. Let A1 and A2 be differentiable atlases for N which induce the topology and such that the inclusions i1 : (N, A1 ) M and i2 : (N, A2 ) M are differentiable immersions. Apply the result of Exercise 2.2 with f = i1 and with f = i2 ; conclude that Id : (N, A1 ) (N, A2 ) is a diffeomorphism. Exercise 2.4. The proof follows from the following characterization of local closedness: S is locally closed in the topological space X if and only if every point p S has a neighborhood V in X such that V S is closed in V . Exercise 2.5. From (2.1.14) it follows easily that the curve: t exp(tX ) m is an integral line of X . Exercise 2.6. Repeat the argument in Remark 2.2.5, by observing that the union of a countable family of proper subspaces of I Rn is a proper subset of I Rn . To see this use the Baires Lemma. Exercise 2.7. It is the subgroup of GL(n, I R) consisting of matrices whose lower left (n k ) k block is zero. Exercise 2.8. Set k = dim(L L0 ). Consider a (not necessarily symplectic) 2nk n n basis (bi )2 i=1 of V such that (bi )i=1 is a basis of L0 and (bi )i=nk+1 is a basis of L. Dene an extension of B by setting B (bi , bj ) = 0 if either i or j does not belong to {n k + 1, . . . , 2n k }.

A.2. FROM CHAPTER 2

155

Exercise 2.9. The proof can be done in three steps: choose a partition a = t0 < t1 < < tk = b of the interval [a, b] such that for all i = 1, . . . , k 1 the portion |[ti1 ,ti+1 ] of has image contained in an open set Ui B on which the bration is trivial (an argument used in the proof of Theorem 3.1.23); observe that a trivialization i of the bration over the open set Ui induces a bijection between the lifts of |[ti1 ,ti+1 ] and the maps f : [ti1 , ti+1 ] F; construct : [a, b] E inductively: assuming that a lift i of |[a,ti ] is given, dene a lift i+1 of |[a,ti+1 ] in such a way that i+1 coincides with i on the interval [a, ti1 + ] for some > 0 (use the local trivialization i and a local chart in F ). Exercise 2.10. The map is differentiable because it is the inverse of a chart. Using the technique in Remark 2.3.4, one computes the differential of the map T Gr(T ) as: Lin(I Rn , I Rm ) Z q Z 1 |Gr(T ) TGr(T ) Gn (n + m), where 1 is the rst projection of the decomposition I Rn I Rm and q : I Rm n + m I R /Gr(T ) is given by: q (x) = (0, x) + Gr(T ). Exercise 2.11. Use the result of Exercise 2.9 and the fact that GL(n, I R) is the total space of a bration over Gk (n). Exercise 2.12. An isomorphism A GL(n, I R) acts on the element (W, O) G+ ( n ) and produces the element ( A ( W ) , O ) where O is the unique orientation k on A(W ) which makes A|W : (W, O) (A(W ), O ) a positively oriented isomorphism. The transitivity is proven using an argument similar to the one used in the proof of Proposition 2.4.2. Exercise 2.13. FixL0 is a closed subgroup of Sp(V, ), hence it is a Lie subgroup. Let L1 , L1 0 (L0 ) be given. Fix a basis B of L0 ; this basis extends in a unique way to a symplectic basis B1 in such a way that the last n vectors of such basis are in L1 (see the proof of Lemma 1.4.35). Similarly, B extends in a unique way to a symplectic basis B1 whose last n vectors are in L1 . The unique symplectomorphism T of (V, ) which xes L0 and maps L1 onto L1 is determined by the condition that T maps B1 to B1 . Exercise 2.14. Use (1.4.7) and (1.4.8) on page 21. 1 Exercise 2.15. Use formulas (2.5.6) and (2.5.7) on page 56: choose L with L1 L0 = {0} and set B = L0 ,L 1 (L). Now, solve for L1 the equation:
1 1 ( )# ( B = L0 ,L1 (B ) = B L0 ,L1 L1 ,L0 (L1 ) L ,L
0 1

156

A. ANSWERS AND HINTS TO THE EXERCISES

A.3. From Chapter 3 Exercise 3.1. A homotopy H between the identity of X and a constant map f x0 drags any given point of X to x0 . Exercise 3.2. For each x0 X , the set y X : a continuous curve : [0, 1] X with (0) = x0 , (1) = y is open and closed, since X is locally arc-connected.
1 Exercise 3.3. Dene s (t) = ((1 s)t) and Hs = ( s ) s . Observe that H1 is a reparameterization of .

Exercise 3.4. If [ ] 1 (X, x0 ) then H induces a free homotopy between the loops f and g in such a way that the base point travels through the curve ; use Exercise 3.3. Exercise 3.5. Using the result of Exercise 3.4, it is easily seen that g f and f g are isomorphisms. Exercise 3.6. The inclusion of {x0 } in X is a homotopy inverse for f iff X is contractible. Exercise 3.7. If g is a homotopy inverse for f , then it follows from Corollary 3.3.24 that g f = Id and f g = Id. Exercise 3.8. It follows from r i = Id. Exercise 3.9. Do you really need a hint for this Exercise? Exercise 3.10. If f 1 and f2 are such that p f1 = p f2 = f then the set {x : f 1 (x) = f2 (x)} is open (because p is locally injective) and closed (because E is Hausdorff). Exercise 3.11. X is connected because it is the closure of the graph of f (x) = sin(1/x), x > 0, which is connected. The two arc-connected components of X are the graph of f and the segment {0} [1, 1]. Both connected components are contractible, hence H0 (X ) = Z Z, and Hp (X ) = 0 for all p 1. Exercise 3.12. See [31, 24, Chapter 3]. Exercise 3.13. First, if U X is open then p(U ) is open in X/G since = gG gU ; moreover, if U is such that gU U = for every g = 1 then p is a trivial bration over the open set p(U ). This proves that p is a covering map. The other statements follow from the long exact homotopy sequence of p and more specically from Example 3.2.21. p1 (p(U )) Exercise 3.14. The restriction of the quotient map p : X X/G to the unit square I 2 is still a quotient map since I 2 is compact and X/G is Hausdorff; this gives the more familiar construction of the Klein bottle. To see that the action of G on X is properly discontinuous take for every x X = I R2 the open set U (see 1 Exercise 3.13) as an open ball of radius 2 . Exercise 3.15. Use Example 3.2.10 and Theorem 3.3.33.

A.4. FROM CHAPTER 4

157

Exercise 3.16. Use the exact sequence 0 = H2 (D) H2 (D, D) H1 (D) H1 (D) = 0. A.4. From Chapter 4 Exercise 4.1. It follows from Zorns Lemma, observing that the union of any increasing net of B -negative subspaces is B -negative. Exercise 4.2. The proof is analogous to that of Proposition 4.1.27, observing that if v1 , v2 V are linearly independent vectors such that B (vi , vi ) 0, i = 1, 2, and such that (4.1.5) holds, then B is negative semi-denite in the two-dimensional subspace generated by v1 and v2 (see Example 4.1.12). Exercise 4.3. Let V1 be the k -dimensional subspace of V generated by the vectors {v1 , . . . , vk } and V2 be the (n k )-dimensional subspace generated by {vk+1 , . . . , vn }; since X is invertible, then B |V1 V1 is nondegenerate, hence, by Propositions 1.1.10 and 4.1.23, n (B ) = n (B |V1 V1 ) + n (B |V V ). One 1 1 computes: V1 = (X 1 Zw2 , w2 ) : w2 V2 , and B |V V is represented by the matrix Y Z X 1 Z .
1 1

Exercise 4.4. Set W = V V and dene the nondegenerate symmetric bilinear form B Bsym (W ) by B ((a1 , b1 ), (a2 , b2 )) = Z (a1 , a2 ) U (b1 , b2 ). Let W denote the diagonal = {(v, v ) : v V }; identifying V with by v (v, v ), one computes easily B | = Z U , which is nondegenerate. Moreover, identing V with by V V (v, U 1 Zv ) , it is easily seen that B | = Z (Z 1 U 1 )Z . The conclusion follows. Exercise 4.5. Compute O on a generator of H1 (). Exercise 4.6. The map [0, 1] [a, b] (s, t) A((1 s)t + sa) (t) is a homotopy with free endpoints between and the curve A(a) . Using Remark 3.3.30 one gets that and A(a) are homologous in H1 (, 0 (L0 )); the conclusion follows from Corollary 4.2.6. Exercise 4.7. Using the result of Exercise 2.13 we nd a curve A : [a, b] Sp(V, ) such that A(t)(L2 ) = L1 (t) for all t and for some xed L2 0 (L0 ); it is easily seen that L0 ,L1 (t) ( (t)) = L0 ,L2 A(t)1 ( (t)) . The conclusion follows from Theorem 4.2.15 and Exercise 4.6. Exercise 4.8. Using formula (2.5.11) one obtains: (A.4.1) L3 ,L0 (L2 ) = (L0 ,L3 )# L0 ,L3 (L2 )1 ;
L0 L1 ,L0 (L3 ,L0 )1 (B ) = L1 ,L0 (L3 ) + L 1 ,L3 #

from (2.5.5) it follows that: (A.4.2) (B ) B(L1 ),

for any symmetric bilinear form B B(L3 ). It is easy to see that: (A.4.3)
L0 L0 ,L3 L = L0 ,L1 ; 1 ,L3

158

A. ANSWERS AND HINTS TO THE EXERCISES

and the conclusion follows by setting B = L3 ,L0 (L2 ) in (A.4.2) and then using (A.4.1) and (A.4.3). Exercise 4.9. See Examples 4.1.4 and 1.1.4. Exercise 4.10. By Theorem 4.2.15, it is L0 ( ) = n+ L0 ,L ( (b)) n+ L0 ,L ( (a)) ; Conclude using the result of Exercise 4.9 where L3 = L , setting L2 = (a) and then L2 = (b). Exercise 4.11. Use Exercise 2.10. Exercise 4.12. Observe that the map p : Sp(I Rn I Rn , ) given by n p(T ) = T ({0} I R ) is a bration; the set in question is the inverse image by p of the dense subset 0 (L0 ) of (see Remark 2.5.18). The reader can prove a general result that the inverse image by the projection of a dense subset of the basis of a bration is dense in the total space. For the connectedness matter see the suggested solution of Exercise 4.13 below. Exercise 4.13 and Exercise 4.14. These are the hardest problems on the book. The basic idea is the following; write every symplectic matrix T = A B C D I U 0 , I

with B invertible as a product of the form: T = 0 B B 1 D

with D = S B and S, U symmetric n n matrices. Observe that the set of symplectic matrices T with B invertible is diffeomorphic to the set of triples (S, U, B ) in Bsym (I Rn ) Bsym (I Rn ) GL(n, I R). Using also some density arguments (like the result of Exercise 4.12) the reader should be able to complete the details. Exercise 4.15. The map : Sp(2n, I R) (I R4n ) given by (t) = Gr(T ) induces a map : 1 Sp(2n, I R) R 4n ) = Z 1 ((I =Z which is injective (it is the multiplication by 2, up to a sign). This is easily checked by computing on a generator of 1 Sp(2n, I R) (see Remarks 4.2.21 and 4.2.22). It follows that if a loop in Sp(2n, I R) has image by which is contractible in (I R4n ) then the original loop is contractible in Sp(2n, I R). Now, use that 0 () is diffeomorphic to a Euclidean space. A.5. From Chapter 5 Exercise 5.1. If X is real-analytic, then also the fundamental matrix t (t) is real-analytic in [a, b]. If (bi )n i=1 is a basis of the Lagrangian 0 , then the (X, 0 )focal instants are the zeroes in ]a, b] of the real-analytic function: [a, b] t det (1 (t)) b1 , . . . , (1 (t)) bn ,

A.5. FROM CHAPTER 5

159

where 1 : I Rn I Rn I Rn denotes the projection onto the rst coordinate. If the initial condition is nondegenerate, then by Corollary 5.2.4 there are no (X, 0 )focal instants in a neighborhood of t = a. Exercise 5.2. Aop = A, B op = B , C op = C and (A.5.1) op = O O. Moreover, (v, ) is a solution of X op iff O (v, ) = (v, ) is a solution of X op . Exercise 5.3. It is O# ( ) = , which proves that O is not a symplectomorphism, but it takes Lagrangian subspaces into Lagrangian subspaces. It is op op P op = P , S op = S , ifoc (X op , op 0 ) = ifoc (X, 0 ) and imaslov (X , 0 ) = imaslov (X, 0 ). For the focal indexes, it sufces to observe that (X, 0 ) and (X op , op 0 ) have the same focal instants, with the same multiplicity but opposite signature. As to the Maslov indexes, one rst observe that, using (A.5.1), the curve op is given by O ; O gives a diffeomorphism of the Lagrangian Grassmannian , O(L0 ) = L0 and so O leaves 0 (L0 ) invariant. One gets an isomorphism O : H1 (, 0 (L0 )) H1 (, 0 (L0 )); it follows from the result of Exercise 4.5 that O = Id. Exercise 5.4. If 0 < < and if there are no (X, 0 )-focal instants in ]a, a + ], then |[a+,a+ ] is a curve in 0 (L0 ), and therefore L0 |[a+,a+ ] = 0. Exercise 5.5. See [36], in the remarks after Denition 3.1.6.

Bibliography
[1] V. I. Arnold, Characteristic Class Entering in Quantization Conditions, Funct. Anal. Appl. 1 (1967), 113. [2] J. K. Beem, P. E. Ehrlich, K. L. Easley, Global Lorentzian Geometry, Marcel Dekker, Inc., New York and Basel, 1996. [3] R. Bott, On the Iteration of Closed Geodesics and the Sturm Intersection Theory, Commun. Pure Appl. Math. 9 (1956), 171206. [4] E. A. Coddington, N. Levinson, Theory of Ordinary Differential Equations, McGrawHill Book Company, New York, Toronto, London, 1955. [5] M. do Carmo, Riemannian Geometry, Birkh auser, Boston, 1992. [6] J. J. Duistermaat, On the Morse Index in Variational Calculus, Adv. in Math. 21 (1976), 173 195. [7] H. M. Edwards, A Generalized Sturm Theorem, Ann. of Math. 80 (1964), 2257. [8] P. E. Ehrlich, S. Kim, A Focal Index Theorem for Null Geodesics, J. Geom. Phys. 6, n. 4 (1989), 657670. [9] I. Ekeland, An Index Theory for Periodic Solutions of Convex Hamiltonian Systems, Proc. of Symposia in Pure Math. vol. 45 (1986), Part I, 395423. o Matem [10] D. G. Figueiredo, A. F. Neves, Equac e a atica s Diferenciais Aplicadas, IMPA, Colec Universit aria, 1997. [11] D. B. Fuks, MaslovArnold Characteristic Classes, Soviet Math. Dokl. 9, n. 1 (1968), 9699. [12] F. Giannoni, A. Masiello, P. Piccione, D. Tausk, A Generalized Index Theorem for Morse Sturm Systems and Applications to semi-Riemannian Geometry, preprint 1999. (LANL math.DG/9908056) [13] F. Giannoni, P. Piccione, R. Sampalmieri, On the Geodesical Connectedness for a Class of Semi-Riemannian Manifolds, to appear in Journal of Mathematical Analysis and Applications. [14] V. Guillemin, S. Sternberg, Geometric Asymptotics, Mathematical Surveys and Monographs n. 14, AMS, Providence RI, 1990. [15] A. D. Helfer, Conjugate Points on Spacelike Geodesics or Pseudo-Self-Adjoint Morse-SturmLiouville Systems, Pacic J. Math. 164, n. 2 (1994), 321340. [16] A. D. Helfer, Conjugate Points and Higher ArnoldMaslov Classes, Contemporary Mathematics vol. 170 (1994), 135147. [17] L. H ormander, An Introduction to Complex Analysis in Several Variables, North-Holland Mathematical Library, 2nd ed., 1979. [18] S. T. Hu, Homotopy Theory, Academic Press, 1959. [19] D. Kalish, The Morse Index Theorem where the Ends are Submanifolds, Trans. Am. Math. Soc. 308, n. 1 (1988), 341348. [20] S. Lang, Differential Manifolds, Springer-Verlag, Berlin, 1985. [21] E. L. Lima, Espac os M etricos, Instituto de Matem atica Pura e Aplicada, (projeto Euclides), 1977. [22] E. L. Lima, Elementos de Topologia Geral, vol. 2, Instituto de Matem atica Pura e Aplicada, 1969. [23] E. L. Lima, Curso de An alise, Vol. 2, Instituto de Matem atica Pura e Aplicada (projeto Euclides), 1981.
160

BIBLIOGRAPHY

161

[24] Y. Long, A Maslov-type Index Theory for Symplectic Paths, Top. Meth. Nonlin. Anal. 10 (1997), 4778. [25] R. C. N. Marques, D. V. Tausk, The Morse Index Theorem for Periodic Geodesics in Stationary Lorentzian Manifolds, preprint 2000. [26] A. Masiello, Variational Methods in Lorentzian Geometry, Pitman Research Notes in Mathematics 309, Longman, London 1994. [27] D. Mc Duff, D. Salamon, Introduction to symplectic topology, Second edition. Oxford Mathematical Monographs. The Clarendon Press, Oxford University Press, New York, 1998. [28] J. Mawhin, M. Willem, Critical Point Theory and Hamiltonian Systems, SpringerVerlag, Berlin, 1989. [29] F. Mercuri, P. Piccione, D. Tausk, Stability of the Focal and the Geometric Index in semiRiemannian Geometry via the Maslov Index, Technical Report RT-MAT 99-08, Mathematics Department, University of S ao Paulo, Brazil, 1999. (LANL math.DG/9905096) [30] J. Milnor, Morse Theory, Princeton Univ. Press, Princeton, 1969. [31] J. R. Munkres, Elements of Algebraic Topology, The Benjamin/Cummings Publishing Company, Inc., 1984. [32] B. ONeill, SemiRiemannian Geometry with Applications to Relativity, Academic Press, New York, 1983. [33] V. Perlick, Ray Optics, Fermats Principle and Applications to General Relativity, Spinger Lecture Notes in Physics, Series m. [34] P. Piccione, D. Tausk, A Note on the Morse Index Theorem for Geodesics between Submanifolds in semi-Riemannian Geometry, J. Math. Phys. 40, vol. 12 (1999), 66826688. [35] P. Piccione, D. V. Tausk, On the Banach Differentiable Structure for Sets of Maps on Non Compact Domains, to appear in Nonlinear Analysis: Series A, Theory and Methods. [36] P. Piccione, D. V. Tausk, An Index Theorem for Non Periodic Solutions of Hamiltonian Systems, to appear in the Proceedings of the London Mathematical Society (LANL math.DG/9908056). [37] P. Piccione, D. V. Tausk, The Maslov Index and a Generalized Morse Index Theorem for Non Positive Denite Metrics, Comptes Rendus de lAcad emie de Sciences de Paris, vol. 331, 5 (2000), 385389. [38] P. Piccione, D. V. Tausk, The Morse Index Theorem in semi-Riemannian Geometry, preprint 2000 (LANL math.DG/..........). [39] P. Piccione, D. V. Tausk, On the Distribution of Conjugate Points along semi-Riemannian Geodesics, preprint 2000 (LANL math.DG/0011038). [40] P. H. Rabinowitz, Periodic Solutions of Hamiltonian Systems, Commun. Pure Appl. Math. 31 (1978), 157184. [41] J. Robbin, D. Salamon, The Maslov Index for Paths, Topology 32, No. 4 (1993), 827844. [42] D. Salamon, E. Zehnder, Morse Theory for Periodic Solutions of Hamiltonian Systems and the Maslov Index, Commun. Pure Appl. Math. 45 (1992), 13031360. [43] S. Smale, On the Morse Index Theorem, J. Math. Mech. 14 (1965), 10491056. [44] M. Spivak, A Comprehensive Introduction to Differential Geometry, vol. I, Publish or Perish, Inc., Boston, 1970. [45] D. V. Tausk, O Teorema do Indice de Morse para M etricas Indenidas e para Sistemas Hamiltonianos, Ph. D. Thesis in Mathematics, Universidade de S ao Paulo, Brazil, 2000. [46] F. Treves, Introduction to Pseudodifferential Operators, Plenum, New York, 1982. [47] V. S. Varadarajan, Lie Groups, Lie Algebras and Their Representations, Prentice-Hall series in Modern Analysis, 1974, New Jersey. [48] F. W. Warner, Foundations of Differentiable Manifolds and Lie Groups, Glenview, Scott, 1971. [49] J. W. Vick, Homology Theory, an Introduction to Algebraic Topology, 2nd edition, Graduate Texts in Mathematics 145, Springer-Verlag.

162

BIBLIOGRAPHY

Departamento de Matem atica Instituto de Matem atica e Estat stica Universidade de S ao Paulo The rst author is partially sponsored by CNPq (Brazil), Processo n. 301410/95, the second author is sponsored by FAPESP, Processo n. 98/12530-2

Index

abelianized group, 104 action by left translation, 39 effective, 39 free, 38 natural of GL(n, I R) on Gk (n), 50 properly discontinuous, 107 right, 39 transitive, 38 without xed points, 38 action of a group, 38 adjoint representation of SU(2), 89 annihilator of a subspace, 4 anti-Hermitian matrix, 10 anti-holomorphic subspace of V C , 15 anti-linear map, 7 augmentation map of the singular complex, 94 augmented singular complex of a topological space, 94 basis of an abelian group, 92 bidual, 2 bilinear form, 1 nondegenerate, 3 anti-symmetric, 2, 29 canonical form of, 18 kernel of, 3 norm of, 116 sesquilinear extension, 13 symmetric, 2, 29 CauchySchwarz inequality for, 113 coindex of, 109 degeneracy of, 110 diagonalization of, 5 index of, 109 maximal negative subsp. w. r. to, 113 maximal positive subsp. w. r. to, 113 negative denite, 109 negative semi-denite, 109
163

orthogonal subspaces w. resp. to, 113 positive denite, 14, 109 positive semi-denite, 109 signature of, 109 subspace negative with respect to, 109 subspace nondegenerate with respect to, 110 subspace positive with respect to, 109 transpose, 2 bitranspose, 2 B -orthogonal decomposition, 113 B -orthogonal subspaces, 113 boundary of a chain complex, 93 CauchyRiemann equations, 16 CauchySchwarz inequality, 113 chain complex, 93 isomorphism of, 96 map of, 96 map induced in homology by, 96 chain homotopy, 100 chain map, 96 chain homotopy of, 100 chain-homotopic maps, 100 change of counterdomain, 33 chart, 32 compatible, 32 submanifold chart, 33 co-index of a symmetric bilinear form, 109 codimension of a subspace, 110 commutator, 34 commutator subgroup of a group, 104 compact-open topology, 70 complementary Lagrangian subspace, 23 complex structure, 6 anti-holomorphic subspace of, 15 basis adapted to, 8 canonical of I R 2n , 6 complex, 7 dual, 8

164

INDEX

holomorphic subspace of, 15 complexication, 9 as a functor, 11 canonical of a real vector space, 10 of multilinear maps, 13 universal property of, 9 concatenation of curves, 66 concatenation of maps , : I n X , 74 conjugate linear map, 7 conjugate operator, 13 conjugation operator subspace invariant by, 12 connection on a semi-Riemannian manifold, 142 connection operator, 77 contractible topological space, 69 convex subset of I Rn , 69 coordinate system, 32 C 0 -topology, 70 covariant derivative along a curve, 142 covering map, 82 C 0 -weak Whitney topology, 70 curvature tensor, 142 cycle of a chain complex, 93 Darbouxs Theorem, 145 degeneracy of a symmetric bilinear form, 110 determinant map in U(n), 85 determinant of an endomorpshim, 17 differentiable atlas, 32 maximal, 32 differentiable covering, 42 lift of a curve of class C k , 42 differentiable manifold, 32 subito sera (S. Quasimodo), ix Ed e elementary row operation with matrices, 87 embedding, 33 equation MorseSturm, 138 equivariant isomorphism, 39 equivariant map, 39 exact sequence of pointed sets, 77 bration, 41, 79 base of, 79 ber over an element, 79 lifting of a map, 81 local trivialization of, 41, 79 total space of, 79 typical ber of, 79

rst set of homotopy of a pair of topological spaces, 75 focal index, 137 of a geodesic, 143 focal instant, 137 multiplicity of, 137 nondegenerate, 137 signature of, 137 free abelian group, 92 generated by a set, 92 freely homotopic loops, 69 f -related vector elds, 42 Frobenius Theorem, 35 functor, 11 exact, 11 fundamental group of a topological space w. basepoint, 67 fundamental groupoid of a topological space, 67 general linear group, 36 geodesic, 142 focal index of, 143 geometric index of, 144 lightlike, 144 spacelike, 144 timelike, 144 geometric index of a geodesic, 144 G-equivariant map, 39 Gk (n), 44 GramSchmidt orthogonalization process, 86 Grassmannian, 3263 of oriented subspaces, 64 Grassmannian of k-dimensional subspaces of I Rn , 44 Grassmannian of oriented Lagrangians, 89 group of p-boundaries of a chain complex, 93 group of p-cycles of a chain complex, 93 group of homology of a chain complex, 93 group of relative p-boundaries of a pair, 97 group of relative p-cycles of a pair, 97 group of singular p-boundaries of a topological space, 94 group of singular p-cycles of a topological space, 94 groups of relative homology of a pair, 97 groups of singular homology of a topological space, 94 Hamilton equation, 145 Hamiltonian function, 145

INDEX

165

Hermitian form, 13 Hermitian matrix, 10 Hermitian product, 13 canonical of Cn , 14 holomorphic function, 16 holomorphic subspace of V C , 15 homogeneous coordinates, 47 homogeneous manifold, 40 homogeneous manifolds, 3842 homologous cycles, 94 homologous relative cycles, 98 homology class determined by a cycle, 94 homology class determined by a relative cycle, 98 homotopic maps, 65 homotopy, 65 with xed endpoints, 65 homotopy equivalence, 107 homotopy groups of a topological space, 75 homotopy inverse, 107 homotopy of curves with free endpoint in a set, 72 Hurewiczs homomorphism, 104 index of a symmetric bilinear form, 109 inner product, 23 canonical of I Rn , 14 invariant subspace, 4 involution of a set, 10 isomorphism determined by a common complementary, 45 isotropy group, 38 Jacobi equation, 142 vector eld, 142 Klein bottle, 108 Lagrangian Grassmannian, 54 Lagrangian initial condition, 136 nondegenerate, 137 Lagrangian submanifold, 145 Lagrangian subspace opposite, 150 Lebesgue number of an open covering of a compact metric space, 71 left-invariant vector eld on a Lie group, 34 Levi-Civita connection, 142 Lie algebra, 34 Lie bracket, 34 Lie derivative, 146 Lie group, 34 action of, 40

linearization, 43 exponential map, 35 homomorphism of, 34 left invariant vector eld on, 34 left-invariant distribution in a, 35 left-translation in a, 34 right invariant vector eld on, 34 right-invariant distribution in a, 35 right-translation in a, 34 time-dependent right invariant vector eld on, 44 Lie subgroup, 35 lift of a curve to a ber bundle, 42 lightlike geodesic, 144 vector, 144 linear operator B-Hermitian, 14 B-anti-Hermitian, 14 B-unitary operator, 14 anti-linear extension, 13 anti-symmetric, 3 complexication of, 11 graph of, 22, 44 normal w. resp. to a bilinear form, 4 diagonalization of, 31 orthogonal, 3 symmetric, 3 transpose, 2 transpose w. resp. to a bilinear form, 3 linearized Hamilton (LinH) equations, 147 local section, 34 locally arc-connected topological space, 70 locally closed subset, 41 locally trivial bration, 79 local trivialization of, 79 typical ber of, 79 long exact homotopy sequence of a pair, 78 long exact homology sequence, 99 long exact homology sequence of a pair, 99 long exact homotopy sequence of a bration, 82 long exact reduced homology sequence of a pair, 99 Lorentzian manifold, 144 manifold, 32 map of pointed sets, 75 kernel of, 75 Maslov index, 121131 of a curve in with endpoints in 0 (L0 ), relative to a Lagrangian L0 , 127 of a geodesic, 144

166

INDEX

Maslov index of a pair (X, 0 ), 140 matrix representation of bilinear operators, 2 of linear operators, 2 maximal negative subspace w. resp. to a symmetric bilinear form, 113 maximal positive subspace w. resp. to a symmetric bilinear form, 113 Morse Index Theorem, 144 MorseSturm equation, 138 natural action of GL(n, I R) on Gk (n), 50 nondegenerate P -focal point, 143 nondegenerate intersection of a curve in with 1 (L0 ), 129 normal bundle, 61 n-th absolute homotopy group, 75 n-th relative homotopy group, 75 null measure, 61 opposite symplectic differential system, 150 orbit, 38 oriented Lagrangian, 25, 89 orthogonal complement of a subspace, 4 orthogonal goup, 37 orthogonal matrix, 37 P -focal point, 143 multiplicity of, 143 signature of, 143 P -Jacobi eld, 143 pair of topological spaces, 73 homeomorphism of, 97 map between pairs, 76 parallel trivialization, 142 partial binary operation, 66 pointed set, 75 distinguished element of, 75 null, 75 null map of, 75 polar form of an invertible matrix, 30 positive Hermitian product, 13 power series, 149 pull-back, 3 push-forward, 3 quotient differentiable structure, 34 quotient property for differentiable maps, 34 real form in a complex space, 9 conjugation operator associated to, 9 imaginary part operator associated to, 9 real part operator associated to, 9

real projective line, 47 real projective space, 46 real-analytic function, 149 realication of a complex vector space, 6 reduced singular homology group of a topological space, 94 reduction of scalars, 6 relative boundary of a pair, 97 relative cycle of a pair, 97 relative homotopy groups of a pair of topological spaces, 75 reparameterization of a curve, 66 retract, 107 retraction, 107 retraction of GL(n, I R) onto O(n), 86 Riemannian manifold, 142 metric, 142 right-invariant vector eld on a Lie group, 34 second countability axiom, 32 second fundamental form, 143 semi-direct product, 108 semi-locally simply connected space, 71 semi-Riemannian manifold, 142 connection of, 142 metric, 142 sesquilinear form, 13 anti-Hermitian, 13 Hermitian, 13 short exact sequence of chain complexes, 98 signature of a symmetric bilinear form, 109 simply connected topological space, 69 singular chain boundary of, 93 singular chain in a topological space, 92 singular complex of a pair, 97 singular complex of a topological space, 94 singular homology groups, 94 singular simplex in a topological space, 92 spacelike geodesic, 144 vector, 144 special linear group, 36 special orthogonal group, 37 special unitary group, 37 square summable sequence, 5 standard simplex, 92 star-shaped subset of I Rn , 69 submanifold almost embedded, 33 embedded, 33

INDEX

167

immersed, 33 sympectic group, 37 symplectic L-trivialization, 147 symplectic chart, 145 symplectic differential system, 134 coefcient matrix of, 134 focal instant of, 137 multiplicity of, 137 nondegenerate, 137 signature of, 137 fundamental matrix of, 135 Lagrangian initial condition for, 136 nondegenerate, 137 Maslov index of, 140 opposite, 150 symplectic form canonical of I Rn I Rn , 20 2n canonical of I R , 20 complex structure compatible with, 23 symplectic forms, 1829 symplectic manifold, 145 symplectic chart of, 145 symplectic map, 19 symplectic matrix, 21 symplectic space, 18 -orthogonal subspace of, 21 isotropic subspace of, 22 Lagrangian decomposition of, 26 Lagrangian subspace of, 22 maximal isotropic subspace of, 22 symplectic group of, 20 symplectic spaces, 129 direct sum of, 21 symplectic subspace, 19 symplectic vector space symplectic basis of, 19 symplectomorphism, 19 Taylors formula, 149 time-dependent right invariant vector eld, 44 timelike geodesic, 144 vector, 144 topology of uniform convergence on compact sets, 70 trace of an endomorphism, 17 transition function, 32 transverse interception of a curve in with 1 (L0 ), 126 transverse orientation, 61 unit n-dimensional cube (I n ), 73

initial face of, 73 unitary group, 37 Vector eld along a curve, 142 volume form, 20 zero-th set of homotopy of a topol. space, 75 Zig-Zag Lemma, 98

You might also like