Clifford Algebras and Bott PeriodicityBasic properties of Fredholm operators

2.1 Basic properties of Fredholm operators

Definition

Let HH be a Hilbert space. A bounded linear operator T:HHT: H \to H is Fredholm if kerT\ker T and cokerT\operatorname{coker}T are both finite-dimensional.

Exercise

TT induces an isomorphism between kerT\ker T^\perp and imT\operatorname{im}T. In particular, imT\operatorname{im}T is closed and (if kerT0\ker T \not= 0) 00 is an isolated point in the spectrum of TT.

Definition

The index of a Fredholm operator T:HHT: H \to H' is

idxT=dimkerTdimcokerT. \operatorname{idx}T = \dim \ker T - \dim \operatorname{coker}T.

Example

If H=HH = H' is finite-dimensional, then all operators are Fredholm with index 00, by the rank-nullity theorem.

Example

Let shift:22\mathrm{shift}: \ell ^2 \to \ell ^2 be the map

shift(x1,x2,x3,)=(x2,x3,x4,). \mathrm{shift}(x_1, x_2, x_3, \dots ) = (x_2, x_3, x_4, \dots ).

Then this has index 11. The adjoint of shift\mathrm{shift}, written shift1\mathrm{shift}^{-1}, is given by

shift1(x1,x2,x3,)=(0,x1,x2,). \mathrm{shift}^{-1}(x_1, x_2, x_3, \dots ) = (0, x_1, x_2, \dots ).

which has index 1-1.

One can similar see that idx(shiftk)=k\operatorname{idx}(\mathrm{shift}^k) = k for all kZk \in \mathbb {Z}. So in particular, every integer is the index of some operator.

The fact that the index of shiftk\mathrm{shift}^k is the negative of the index of its adjoint is not a coincidence.

Exercise

Let T:HHT: H \to H be Fredholm. Then so is TT^*, and idxT=idxT\operatorname{idx}T^* = - \operatorname{idx}T.

We shall prove some basic properties of Fredholm operators. They all follow from the following result:

Lemma

Consider the commutative diagram

\begin{useimager} 
    \[
      \begin{tikzcd}
        0 \ar[r] & H \ar[r] \ar[d, "T"] & H' \ar[r] \ar[d, "S"]& H'' \ar[r] \ar[d, "R"] & 0\\
        0 \ar[r] & H \ar[r] & H' \ar[r] & H'' \ar[r] & 0
      \end{tikzcd}
    \]
  \end{useimager}

If the rows are short exact, and any two of TT, SS and RR are Fredholm, then so is the third, and

idxS=idxT+idxR. \operatorname{idx}S = \operatorname{idx}T + \operatorname{idx}R.

Proof
Apply the snake lemma to obtain a long exact sequence

\begin{useimager} 
    \[
      \begin{tikzcd}[column sep=small]
        0 \ar[r] & \ker T \ar[r] & \ker S \ar[r] & \ker R \ar[r] & \coker T \ar[r] & \coker S \ar[r] & \coker R \ar[r] & 0
      \end{tikzcd}.\qedhere
    \]
  \end{useimager}

Proof

Corollary

If T:HHT: H \to H and S:HHS: H' \to H' are Fredholm, then so is TS:HHHHT \oplus S: H \oplus H' \to H \oplus H', and we have

idx(TS)=idxT+idxS. \operatorname{idx}(T \oplus S) = \operatorname{idx}T + \operatorname{idx}S.
Of course, we could have proved this directly.

Corollary

Let T:HHT: H \to H' and S:HHS : H' \to H” be Fredholm. Then STST is Fredholm, and

idx(ST)=idxT+idxS. \operatorname{idx}(ST) = \operatorname{idx}T + \operatorname{idx}S.

Proof

\begin{useimager} 
    \[
      \begin{tikzcd}
        0 \ar[r] & H \ar[r, "{(\id, T)}"] \ar[d, "T"] & H \oplus H' \ar[r, "T - \id"] \ar[d, "ST \oplus \id"]& H' \ar[r] \ar[d, "S"] & 0\\
        0 \ar[r] & H' \ar[r, "{(S, \id)}"] & H'' \oplus H' \ar[r, "\id - S"] & H'' \ar[r] & 0
      \end{tikzcd}\qedshift
    \]
  \end{useimager}

Proof

Corollary

Let I:HHI: H \to H be the identity and K:HHK : H \to H be a compact operator. Then I+KI+ K is Fredholm and has index 00.

Proof
First consider the case where KK has finite-dimensional image. Then consider the short exact sequences

\begin{useimager} 
    \[
      \begin{tikzcd}
        0 \ar[r] & \im K \ar[r] \ar[d] & H \ar[r] \ar[d]& \coker K \ar[r] \ar[d] & 0\\
        0 \ar[r] & \im K \ar[r] & H \ar[r] & \coker K \ar[r] & 0
      \end{tikzcd}
    \]
  \end{useimager}

where the vertical maps are all restrictions of I+KI+ K. One sees that the maps are well-defined.

Now note that the left-hand map is one between finite-dimensional vector spaces, so is Fredholm with index 00. On the other hand, the right-hand map is in fact the identity, since we quotiented out by the image of KK, and is in particular Fredholm with index 00. So we are done.

For general compact KK, we can approximate KK arbitrarily well by finite-rank operators 1 . So pick KK' such that KK' has finite rank and KK<1\| K - K'\| < 1. Thus I+(KK)I+ (K - K') is invertible, and we can write

I+K=(I+(KK))(I+(I+KK)1K). I+ K = (I+ (K - K')) (I+ (I+ K - K')^{-1} K').

The first term is invertible, hence Fredholm with index 00. The second is II plus something of finite rank, so is Fredholm with index 00. So the general result follows.

Proof

Theorem

Let TT be a bounded linear operator. Then TT is Fredholm iff there exists S:HHS: H \to H such that TSITS - I and STIST - I are compact.

Proof

Proof

Corollary

If TT is Fredholm and KK is compact, then T+KT + K is also Fredholm.

Proof
We can use the same SS, since the algebra of compact operators is a two-sided ideal in End(H)\operatorname{End}(H).
Proof