De Rham Cohomology Operators and indices K -theory and the index

Coarse Index Theory Lecture 1

John Roe
Penn State University

May 2006

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Outline

De Rham Cohomology

Operators and indices

K -theory and the index

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

The de Rham complex

Let M be a smooth manifold. Recall that a k -form on M is a section of the k th exterior power of the cotangent bundle. The exterior derivative is the differential operator d : k (M) k +1 (M) which is uniquely determined by the facts that if f 0 (M) is a smooth function, then df (X ) = X f for all vector elds f , and d is a derivation of the graded algebra (M).

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Lemma For any form , d(d) = 0. This lemma means that the differential forms form a complex 0 (M) 1 (M) 2 (M) called the de Rham complex, and we have Theorem The cohomology of the de Rham complex is isomorphic to the topologically dened (singular or Cech) cohomology of M.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

In particular, if M is a compact manifold, the de Rham cohomology spaces H i (M) are nite dimensional. This is a non-trivial statement about partial differential equations. From the de Rham complex of a (compact) manifold we can extract two important numerical invariants:
1 2

The Euler characteristic (M) =

(1)i dim H i (M), and

(in the oriented case) the signature, discussed below.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Poincar duality and the signature

An orientation on a compact n-manifold M denes an integration functional : H n (M) R. The Poincar duality theorem states Theorem The intersection form (, )
M

denes a nondegenerate pairing H k (M) H nk (M) R. In particular if n = 4k then the intersection form gives a symmetric bilinear form on H 2k (M). The signature of M is by denition the signature of this symmetric bilinear form.
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

The Hodge theorem

In order to nd a realization of de Rham cohomology in terms of analysis, we need to introduce a Riemannian metric. This gives rise to L2 inner products on all spaces of forms. Theorem Each de Rham cohomology class on a compact manifold contains a unique element of minimal L2 norm. Simple Hilbert space geometry shows that such a minimal element must be orthogonal to the range of d, and therefore it must be in the kernel of the Hilbert space adjoint operator d . This directs our attention to the elliptic operator D = d + d on the space of all differential forms.
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Consider a rst order linear differential operator D on sections of a vector bundle E over M. There is a unique endomorphism-valued homogeneous function D on T M with the property that D(fs) = fDs + D (df )s for all (smooth) functions f and sections s of E. Denition D is called the symbol of D. Example The symbol of the exterior derivative operator d is the 1-form that takes a covector to the operation exterior multiplication by .
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Denition The linear differential operator D is called elliptic if, for every nonzero , the endomorphism D () is invertible. For example, the de Rham operator d + d that we discussed earlier is elliptic. The basic analytical information about elliptic operators on compact manifolds is contained in the following theorem. Theorem Let D be a self-adjoint elliptic operator on a compact manifold. Then D has compact resolvent: for any f C0 (R), the operator f (D) (dened by the functional calculus) is compact.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Corollary The kernel of an elliptic operator is a nite-dimensional vector space. Together with the Hodge theorem this provides an analytic explanation for the nite dimensionality of de Rham cohomology. (In fact, the basic analytical facts about elliptic operators are needed in the proof of the Hodge theorem.)

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Gradings
Denition Let D be an elliptic operator on a bundle E. A grading for D is a self-adjoint involution of E that anticommutes with D (D + D = 0). In other words, splits E into two orthogonal subbundles E , and D interchanges them. Denition The index of a graded operator (D, ) is the integer dim ker D+ dim ker D where D denotes the restriction of D to an operator E E .
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Let D = d + d denote the de Rham operator on a compact Riemannian manifold. Example The involution which equals (1)k on the space of k -forms is a grading, and the index of D with respect to this grading is the Euler characteristic. Let M be oriented and of dimension 4k . The Hodge star operator is a (linear) identication p np coming from the Riemannian metric. Up to some powers of i = 1, the Hodge star operator is a grading for D. Example The index of D with respect to the Hodge star grading is the signature.
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

The Atiyah-Singer index theorem

It is not hard to prove that the index of an elliptic operator on a compact manifold M depends only on its symbol, and indeed only on the homotopy class of that symbol in the space of elliptic symbols. The Atiyah-Singer index theorem gives an explicit topological formula for the index in terms of characteristic classes associated to the homotopy class of the symbol. Thus it makes a connection between topology and analysis.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Example The index theorem applied to the signature operator produces the Hirzebruch Signature Formula Sign(M) =
M

L(M)

where L is a certain combination of Pontrjagin classes. Explicitly, we have for example Dimension of M L-class 4 p1 /3 2 8 (7p2 p1 )/45 3 12 (62p3 13p1 p2 + 2p1 )/945 in terms of the Pontrjagin classes of the tangent bundle.
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

K-theory
The version of algebraic topology that is most suitable for calculations with indices is K -theory. Let A be a C -algebra with unit. A projection in A is an element p such that p = p2 = p . Let Pn (A) be the collection of all projections in the matrix algebra Mn (A), and let P(A) be the direct limit P(A) = lim Pn (A).
n

This is a topological space. Moreover, its set of connected components, 0 (P(A)), has a natural addition operation deform two projections until they are disjoint, then take the direct sum.
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Denition The K -theory group K0 (A) is the Grothendieck (or universal) group associated to the abelian semigroup 0 (P(A)). Example The group K0 (C) is isomorphic to Z. The isomorphism associates to a projection in Mn (C) the dimension of its range. One can extend the denition of K0 to algebras without unit. If J is an algebra without unit, embed it as an ideal in a larger algebra A. Then K0 (J) is generated by formal differences [p] [q] of projections in Mn (A) such that p q Mn (J).

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Again let A be a unital algebra. A unitary in A is an element u such that uu = u u = 1. Let Un (A) denote the collection of unitaries in Mn (A) and let U(A) be the direct limit lim Un (A). Denition The K -theory group K1 (A) is the collection of connected components 0 (U(A)). Here, because U(A) is already a topological group, there is no need to use the Grothendieck construction to obtain a group.The group operation in K1 (A) can be obtained from multiplication of unitaries, or from direct sum. In particular, K1 (A) is abelian.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Example The group K1 (C) is zero, because all the unitary groups Un (C) are connected. Example Consider the algebra A = C(S 1 ) of continuous functions on the circle. Then K1 (A) Z. The isomorphism is given by sending a = loop of unitary matrices to the winding number of its determinant (a loop in C \ {0}) about the origin. One can also extend the denition of K1 to non-unital algebras.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

K -theory is a functor: for any -homomorphism : A B one can nd an induced homomorphism of abelian groups : Ki (A) Ki (B). Homotopic -homomorphisms induce identical maps on K -theory.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Six term exact sequence

The most important property of K -theory is the six term exact sequence. Let 0JAB0 be a short exact sequence of C -algebras. Then there is a commutative exact diagram K0 (J)
O / K0 (A) / K0 (B) 

K1 (B) o

K1 (A) o

K1 (J)

The connecting homomorphisms (vertical maps) can be described in explicit terms.

John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Consider the case A = B(H), J = K(H). The quotient B = A/J is then the Calkin algebra. Any unitary u B lifts to V A which is an essentially unitary Fredholm operator. On the other hand, u also denes a class [u] K1 (B). Theorem The class [u] K0 (J) = Z is the index of the Fredholm operator V . Notice that the statement the index of an invertible operator is zero is encoded in the exactness of the K -theory sequence.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Abstract index theory

Motivated by the above, we are going to reinterpret the index of an elliptic operator (on a compact manifold) in terms of operator K -theory. This will show us the correct generalization of the index to non-compact manifolds. Our abstract set-up will be H is a Hilbert space. J is a C -algebra of operators on H. D is an unbounded, self-adjoint operator on H such that f (D) J for all f C0 (R).

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Gradings

We distinguish the graded and ungraded cases of our basic set-up. In the ungraded case the data is exactly as I have said. In the graded case we assume in addition that H is a graded Hilbert space it is equipped with a self-adjoint involution . The grading operator is a multiplier of J, that is J = J = J. D is odd relative to the grading, that is, D + D = 0.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Now let A be the multiplier algebra of J (the largest C -algebra of operators on H within which J is an ideal). There is a short exact sequence 0 J A A/J 0.

Denition A normalizing function is a continuous increasing odd function : R (1, 1) such that (t) 1 as t . Observe
1

Any two normalizing functions differ by an element of C0 (R). The square of a normalizing function is equal to 1 modulo C0 (R).
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Thus the operator F = (D) is an element of M(J), and its class in the quotient algebra M(J)/J is a symmetry (a self-adjoint involution). In the ungraded case, a self-adjoint involution in M(J)/J denes a class in K0 (M(J)/J). In the graded case, an odd self-adjoint involution in M(J)/J denes a class in K1 (M(J)/J). Denition The index class of D is the image of [F ] K (M(J)/J) under the boundary map : K (M(J)/J) K1 (J).

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Theorem The index is an obstruction to invertibility. In other words, if 0 does not belong to the spectrum of D, then Index(D) = 0. Proof. If the spectrum of D does not contain 0 then one can nd a normalizing function that is equal to 1 everywhere on the spectrum of D. Then (D) itself is an involution so denes a class in Ki (M(J)). By exactness, the index in Ki1 (J) is equal to zero. In the ungraded case there is nothing special about 0 here if the spectrum of D has any gap at all, then Index(D) = 0.

John Roe

Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

Index formulas
To conclude we give formulae for the index. These depend on knowledge of the explicit form of the connecting homomorphisms. In the ungraded case the index is the homotopy class of the unitary u = ei(D) where is a normalizing function. (This is a unitary differing from the identity by an element of J.) It is sometimes convenient to select so that u becomes the Cayley transform (D + i)(D i)1 of D.
John Roe Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

In the graded case we rst dene a sort of relative index. Denition Let S, T be symmetries in M(J) that anticommute modulo J (that is, ST + TS J). Dene [S : T ] K0 (J) to be the K -theory class of the formal difference [STS] [T ]

of symmetries (or of the corresponding projections). The denition of the boundary map in K -theory gives that the index of the graded operator D is equal to [S : ] where S A is any symmetry equal to (D) modulo J. An example of such a symmetry is S = (D) +
John Roe

1 (D)2 .
Coarse Index Theory Lecture 1

De Rham Cohomology Operators and indices K -theory and the index

We can simplify this expression. Lemma If S + T is invertible then [S : T ] = [T : S]. If we choose S = (D) + 1 (S)2 then S + is invertible so the index is equal to [ : S]. Now take () = (1 + 2 )1/2 ; then S is the phase (D + )|D + |1 , and similarly S is the phase of D . Thus the graded index is the difference of the phases of D .

John Roe

Coarse Index Theory Lecture 1