3
$\begingroup$

Let $P$ be a polarization of a Hilbert space $\mathcal{H}$, i.e. a bounded idempotent: consider a group $G=GL_{res}(\mathcal{H}):=\{g \in GL(\mathcal{H}): [g,P] \in HS\}$ (where $HS$ is the set of all Hilbert Schmidt operators). For $g \in G$ let $P_g:=gPg^{-1}$ and define $F_{g,h}=P_gP_h-P_h+I$. One can check that $F_{g,h}$ are Fredholm: therefore one can define the so called determinant line $det(F_{g,h})$ to be $\Lambda^{top}(ker(F_{g,h})) \otimes \Lambda^{top}(coker(F_{g,h}))^*$. Using this data one can construct a (small) category where (the set of) objects is $G$ and for $g,h \in G$ we put $Hom(g,h)=det(F_{g,h})$. However it is not immediate that this constitutes a category: in order to do so one have to prove some identifications:

How to prove that $det(F_{g,h} \cdot F_{h,k}) \cong det(F_{g,h}) \otimes det(F_{h,k})$?

Of course both spaces are in fact one dimensional therefore there is an abstract isomorphism but I suspect that this isomorphism should be somehow canonical and natural. I would guess that in order to construct such an isomorphism maybe one has to use the fact that $Index(F_{g,h}F_{h,k})=Index(F_{g,h})+Index(F_{h,k})$.

Once we have this identification one can show that in fact $F_{g,h}F_{h,k}-F_{g,k}$ is a trace-class operator.

How to prove that if $S,T$ are two Fredholm operators with $S-T$ being trace-class then $detS \cong detT$ are canonically isomorphic?

I would be very grateful for any help.

$\endgroup$
2
  • $\begingroup$ It is not exactly this, but there is a discussion about something similar in abbondandolo and majer’s paper infinite dimensional grassmannians $\endgroup$
    – Thomas Rot
    Commented May 27, 2023 at 6:30
  • $\begingroup$ Thank you for your comment! I think that one can deduce the first part from this paper-still I'm also interested in the second part (if a difference $S-T$ is trace class) $\endgroup$
    – truebaran
    Commented May 30, 2023 at 6:48

1 Answer 1

1
+100
$\begingroup$

As mentioned in the comments the first part of the question is in Abbonandolo and Majers "Infinite dimensional Grassmannians". My Hilbert spaces are real and separable and infinite dimensional.

I want to state that I basically know next to nothing about trace class operators, so take the rest with a grain of salt.

To every Fredholm operator one can associate its determinant line. An important fact is that this defines an actual line bundle over the space of Fredholm operators $\mathrm{det}\rightarrow \Phi(\mathbb{H})$. This is the unique (up to isomorphism) line bundle over $\Phi(\mathbb{H})$ that is non-trivial over each component of $\Phi(\mathbb{H})$. (This uses the fact that all components are homotopy equivalent and that $\pi_1(\Phi(\mathbb H))\cong\mathbb{Z}/2)$.

A continuous family of Fredholm operators is a continuous map $f:X\rightarrow \Phi(\mathbb H)$. One can use this map to pull back the determinant line bundle over $X$. Let's assume that $X$ is connected. Then the bundle $f^*\det$ is a trivial bundle over $X$ if and only if $f_*:\pi_1(X)\rightarrow \pi_1(\Phi(\mathbb H))$ is the trivial homomorphism.

Let $S$ be a Fredholm operator. Let $X$ be the space of Fredholm operators T such that $S-T$ is trace class, and $f:X\rightarrow \Phi(\mathbb H)$ the inclusion. I think that $X$ is an affine space, hence contractible, from which it follows that $f_*$ is trivial. This means that the determinant line bundle is trivial over $X$. You can use the triviality of the determinant line bundle to give a canonical isomorphism between two different operators of trace class.

$\endgroup$
6
  • 1
    $\begingroup$ Thank you, it sounds reasonable! I assume that you have meant $X$ to be convex instead of affine space, right? $\endgroup$
    – truebaran
    Commented May 31, 2023 at 16:01
  • $\begingroup$ Convex is fine as well, but Isnt X affine? (But maybe we have different definitions of affine in mind) $\endgroup$
    – Thomas Rot
    Commented May 31, 2023 at 16:22
  • $\begingroup$ I should say that this only gives an iso up to homotopy if you dont have a metric on the determinant line bundle. But i believe quillen also constructed such a metric. But i dont know the details $\endgroup$
    – Thomas Rot
    Commented May 31, 2023 at 17:12
  • $\begingroup$ Why do you need a metric? I though that if $\pi_1(X)$ is itrivial then so is $H^1(X,\mathbb{Z}/2)$ the latter being isomorphic to rank one real vector bundles. Also, I'm wondering about the following: as far as I know, one really needs the difference to be a trace class (not merely Hilbert-Schmidt for example)-but your proof seems to work if we replace trace class operators by Hilbert-Schmidt so I;m a little bit worried whether everything is all right $\endgroup$
    – truebaran
    Commented May 31, 2023 at 18:38
  • $\begingroup$ Dear @truebaran: It is true that the triviality of the vector bundle does not depend on the metric. However, to do parallel transport from one fiber to another we need to choose a metric I think. $\endgroup$
    – Thomas Rot
    Commented Jun 1, 2023 at 6:54

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .