Skip to the Main Content

Note:These pages make extensive use of the latest XHTML and CSS Standards. They ought to look great in any standards-compliant modern browser. Unfortunately, they will probably look horrible in older browsers, like Netscape 4.x and IE 4.x. Moreover, many posts use MathML, which is, currently only supported in Mozilla. My best suggestion (and you will thank me when surfing an ever-increasing number of sites on the web which have been crafted to use the new standards) is to upgrade to the latest version of your browser. If that's not possible, consider moving to the Standards-compliant and open-source Mozilla browser.

March 6, 2006

nEFT at Schloss Mickeln, Part II

Posted by Urs Schreiber

Last time I tried to indicate, roughly, how a time-dependent euclidean (Wick-rotated) system of supersymmetric quantum mechanics defines a virtual vector bundle (with fibers the kernel minus the cokernel of the supercharges/Dirac operators) over the space which parameterizes the family of operators at different “times”, hence an element of the K-theory of that space. If this space consists of a single point (the “time-independent” case) then the dimension of the corresponding virtual vector space coincides with the partition function of the SQM system.

I ended by mentioning that in one dimension higher, for superconformal 2D euclidean field theories, the integer number called the partition function gets replaced by a modular form. Here I’ll try to review the mathematician’s description for what this means and how it comes about.

The following is essentially a transcript of Johannes Ebert’s talk at our workshop, which again roughly followed section 3.3 of WIAEO?.

We mainly want to understand and proof the following theorem:

Theorem: The partition function of a supersymmetric C n-linear conformal field theory E, with n even, is a weak integral modular form of weight n/2 .

The plan of J. Ebert’s talk was as follows:

1) modular forms

2) recall elements of CFTs

3) the partition function on the Teichmüller space of tori, the Pfaffian line bundle

4) the semigroup of E

5) supersymmetry and the proof of the theorem

1) modular forms

Definition: A modular function of weight k is a holomorphic function f: on the upper half plane , such that for all (a b c d)SL(2 ,) we have

(1)f(aτ+bcτ+d)=(cτ+d) kf(τ).

It follows that for every modular function f

(2)f(τ+1 )=f(τ)

which again implies that f only depends on

(3)q=e 2 πiτ

and hence has a Laurent expansion (“q-expansion”)

(4)f(q)= na nq n.

Definition: A modular function f is called a weak modular form if for all but a finite number of n<0 we have that a n=0 in the above expansion.

Furthermore, f is called a modular form if all a n for n<0 vanish.

Finally, f is called integral if all of the coefficients a n are integral.

Example: The crucial modular forms to know about are


(5)Δ=q n=1 (1 q n) 24 ,

which has weight 12 and is known as the discriminant,

ii) the Eisenstein series

(6)c 4 = 1 +240 k>0 σ 3 (k)q k c 6 = 1 504 k>0 σ 5 (k)q k,

where σ r(k)= dkd r. These have weight 4 and 6 , respectively.

Obviously, modular forms can be added and multiplied and yield further modular forms. This gives us a (graded) ring of modular forms. It turns out to be equivalent to the ring of polynomials over three indeterminates given by the above three modular forms, [c 4 ,c 6 ,Δ] , divided by the ideal generated by (c 4 3 c 6 2 1728 Δ). For the ring MF * of weak modular forms we have instead

(7)MF *[c 4 ,c 6 ,Δ,Δ 1 ]/(c 4 3 c 6 2 1728 Δ).

2) review of CFT

We need some concepts from Stolz-Teichner’s functorial description of CFT. I will not give the detailed definition of the following items. The main point is that we want to consider a bicategory whose 2-morphisms are (super-)conformal spin cobordisms, decorated with certain generalized Fock spaces on which certain Clifford algebras act. The crucial part of this definition for our purposes ist that for closed surfaces this generalized “Fock space” is not a Fock space but just the top exterior power of the kernel of some Dirac operator on the surface - the Pfaffian line.

Slightly more precisely we have the following.

Denote by 𝒮ℬ n 2 the category whose

- objects are 1-dimensiona spin manifolds Y

- morphisms are conformal spin bordisms from Y 1 to Y 2 labelled by a vector ψF alg(Σ) 1 from some sort of fermionic Fock space on which Clifford algebras cooked up from Σ,Y 1 and Y 2 are represented. (The reader should either look this up in Stolz/Teichner or trust that for a rough understanding of the main point to be discussed here the details can can be safely ignored.)

A CFT of degree n (similar remarks apply to the precise meaning of this degree) is (defined to be) a continuous functor

(8)E:𝒮ℬ n 2 Hilb

compatible with certain extra structure floating around, such as monoidal structure and certain involutions.

As mentioned above, by construction we have for closed surfaces that

(9)F alg(Σ)=Λ top(kerD +) *=:Pf(Σ)

is the Pfafiian line of some canonical Dirac operator D + defined on Σ.

It follows that the CFT E defines for any Riemannian spin surface Σ an element

(10)Z E(Σ)Pf(Σ) n


(11)Z E(Σ)(ψ):=E(Σ,ψ)Hilb(E(),E()),

where we use

(12)Pf(Σ) 1 =Pf(Σ) n*.

The map Z E is of course the partition function.

Before continuing, it proves convenient to pass from the language of spin structures and Dirac operators on conformal surfaces to the analogous concepts in complex analysis. Johannes Ebert noted the following small dictionary:

(13)2 Dspingeometry complexanalysis spinstructureonΣ squarerootSofcanonicallinebundleK Σ D + ¯ S:Γ(S)Γ(K Σ *S)=Γ(S 1 ).

Consider in particular the torus. Here we have four spin structures and K is the trivial holomorphic line bundle. Three of its square roots are nontrivial line bundles. For these we have ker(¯)=0 .

The fourth spin structure, the periodic one, has the property that dimker(¯)=1 , since this is just the trivial holomorphic line bundle itself.

The important point here is that on any torus there is a canonical spin structure with dimker(¯)=1 .

3) the partition function on the Teichmüller space of tori, the Pfaffian line bundle

For every point in the upper half plane there is a complex torus

(14) Τ τ Σ τ:=/τ

and every complex torus is of this form. Two such tori Σ τ, Σ τ are equivalent if and only if there is a gSL(2 ,) such that


In such a case the biholomorphic map relating both tori is given by

(16)Σ τ ϕ g Σ τ 1 z (cτ+d) 1 z.

For each such torus Σ we have its Pfaffian line Pf [Σ]:=ker(¯ Σ,perΣ) * discussed above. These fit together into a holomorphic line bundle

(17)PfΤ 1

on the torus Teichmüller space Τ 1 .

The bundle Pf 2 is always trivial. Its fibers are Pf Σ τ 2 =ker¯ K, which are nothing but the spaces of holomorphic 1-forms in one complex dimension. Hence dz τ with t τ the canonical coordinate on Σ τ is a canonical basis for Pf Σ τ 2 . Hence an explicit trivialization is given by mapping

(18)× (τ,x)xdz τ Pf 2 Τ 1 .

We have an action of SL(2 ,) not only on Τ 1 but also on Pf 2 over it. In terms of coordinates used in the above trivialization the action on the fibers is

(19)(ϕ g) *dz τ=(cτ+d) 1 dz.

Therefore for any point (τ,x) in the Pfaffian bundle over Teichmüller space we have the SL(2 ,) action

(20)(a b c d)(τ,x):=(aτ+bcτ+d,(cτ+d)x).

The reason for going through all of this is the following crucial consequence:

An equivariant section of Pf 2 k has the transformation behaviour of a modular form of weight k.

The partition function Z E mentioned above obviously defines a section of this bundle:

(21)Z E:Τ 1 Pf n .

This section is equivariant by construction, since we defined it from a functor E that is by definition insensitive to the transformation (Σ,ψ)(gΣ,g *ψ) induced by an element gSL(2 ,).

In order to show that the partition function Z E has a chance of being a modular form we need to show that it is a holomorphic function of τ. In order to get there we first need to re-express the partition function in terms of the transport over the annulus. This is the content of the next section.


We can build the partition function on the torus from a supertrace over the propagator of the annulus. More precisely, let

(22)A τ={0 zimτ}/

be the annulus defined by τ. A τ is a spin bordism from the circle S per with periodic spin structure to itself. We can evaluate our CFT functor E on this morphism to obtain

(23)E(A τ,Ω n),

where Ω is a certain canonical vacuum state in the above mentioned (but not explained) algebraic Fock spaces with which our morphisms are supposed to be labelled.

Lemma: The partition function evaluated on the torus is the supertrace of the transport over the annulus

(24)Z E(τ)=str(E(A τ,Ω n)).

This can be proven by cutting the torus in two equal halfs, each an annulus andf then using various properties satisfied by E.

The map

(25)ρ : Hilb(E(S per),E(S per)) τ E(A τ,Ω n)

defined by propagation over the annulus, has, by the properties of our CFT functor E, the following properties:

ρ is a semigroup homomorphism

ρ(τ) is a C n-linear even operator (but I haven’t explained this yet)

ρ(τ) is a Hilbert-Schmidt operator,

ρ is weakly differentiable, meaning that the function f v:τ{v ifτ ρ(τ)v otherwise is differentiable for all v.

Using this, we can prove the following

Lemma: Any semigroup homomorphisms τ from the upper half plane to self-adjoint C n-linear Hilbert-Schmidt operators on some Hilbert space H is of the form

(26)ρ(τ)=exp(2 πiτL)exp(2 πiτ¯L¯),

where L and L¯ are commuting, even C n-linear unbounded operators on H.

Proof: Define

(27)L:=1 2 πiτ τ=0 ρ(τ)


(28)L¯:=1 2 πiτ¯ τ=0 ρ(τ).

(Notice that the notation ρ(τ) is not supposed to imply that ρ is holomorphic, which it is not. You might prefer to write ρ(τ,τ¯).)

The fact that is a commutative semigroup and that ρ is a semigroup homomorphism implies that

(29)[L,L¯]=0 .

Some further important properties of ρ are

i) ρ(τ+1 )=ρ(τ) (because A τ+1 and A τ define the same annulus)

ii) ρ(τ¯)=ρ(τ) * (\because A¯ τ=A τ¯), this implies that both L and L¯ are self-adjoint

It also follows that

(30)exp(2 πi(LL¯))=exp(2 πiL)exp(2 πiL)=Id

and hence that the eigenvalues of LL¯ are all integral.

5) supersymmetry and the proof of the theorem

With all this information available, we can now quite easily show that the partition function is a modular form – if we introduce supersymmetry.

In stolz/Teichner’s work there is a little gap left which is filled by a hypothesis. We really want to define a susy CFT functor to be one which is defined not on the conformal spin bordisms mentioned above, but on superconformal bordisms. While it seems to be obvious how to do this, people are cautious about this point because the rigorous fortmulation has possibly not be written down yet. Anyway, we expect that when we introduce (1,0) superconformal symmetry (called 1 /2 supersymmetry by Stolz/Teichner and called the heterotic case by physicists) our CFT functor will be defined by L and L¯ such that there is an odd operator G¯ with

(31)L¯=D¯ 2 .

The “hypothesis” is that this is the case. So instead of honestly studying CFT functor E on superconformal worldsheets, we apply the above analysis and then restrict attention to the case where the above condition holds.

Now, by the content of section 4) we have that the partition function is given by

(32)Z E(τ)=str C n(q Lq¯ L¯).

Since L¯ has, by assumption of supersymmetry, a symmetric spectrum, it follows by the usual argument that only the kernel of D¯ can contribute nontrivially to the above graded trace

(33)str C n(q Lq¯ L¯)=str C n(q Lq¯ L¯ kerL¯).

Let H k be the kth Eigenspace of L. Then this can be rewritten as

(34)Z E(τ)= kq ksdim C n(H kKerL¯)

That the sum here is only over integers follows from the “level mathcing condition”, derived above, which says that spec(LL¯).

Finally, note that there cannot be an infinite number of negative k contributing to this sum, since if there were the propagator were not Hilbert-Schmidt, contrary to our definition of the CFT functor. Hence

Z E is a weak integral modular form.

Note that for each fixed k, we can interpret the above sum expression for the partition function as the partition function of a SQM system, whose Hamiltonian is given by the restriction of L¯ to H k.

Hence we should really think of the CFT partition function as a power series with coefficients in the K-theory of a point!

Judgig from the situation sketched in the previous entryI bet there is a neat way to express this in a way that manifestly realizes the CFT partition function as the dimension of a virtual 2-vector space which is the 2-kernel minus the 2-cokernel of some 2-operator. But that’s speculation.

Posted at March 6, 2006 10:32 AM UTC

TrackBack URL for this Entry:

1 Comment & 4 Trackbacks

Read the post Hilbert Uniformization of Moduli Space
Weblog: The String Coffee Table
Excerpt: On a result by Boedigheimer on Hilbert uniformization.
Tracked: March 7, 2006 5:13 PM

Re: nEFT at Schloss Mickeln, Part II

I wrote:

But that’s speculation.

Here is a rough idea on how this speculation might eventually be substantiated.

The elliptic genus is a power series in q whose coefficients take values in the K-theory of a point. But we know that this K-theory is really the decategorification of a virtual vector bundle. This suggests that the decategorification of the elliptic genus which I am after should involve a collection of (virtual) vector spaces.

But collections of vector spaces are precisely what the Kapranov-Voevodsky 2-vector spaces are made of.

Consider the following. Let

(1)H=diag(H k)

be the × matrix of vector spaces whose nondiagonal entries are all the 0-dimensional vector space and whose kth entry on the diagonal is the kth eigenspace for the left-moving Virasoro charge L.

Think of this as a 1-morphism in KV 2-vector spaces associated to the ingoing boundary of an annulus. The same 1-morphism is then assigned to the outgoing boundary. The 2-morphism in between should be the diagonal matrix

(2)P=diag(q τexp(τ¯L¯) H k)

whose kth diagonal entry is the propagator restricted to H k.

The point is that here all the virtual vector spaces which we expect to see in the categorification do appear as matrix entries. And using Kapranov’s notion of categorical trace we do have a nifty way do decategorify this and obtain the desired result:

Kapranov tells us that the right notion of trace of the above 2-morphism is to first take the categorical trace and then the ordinary trace over the resulting ordinary linear map. But the categorical trace sends the above matrix of vector spaces simply to the direct sum of all the K k, hence to the full Hilbert space of the string. Similarly it sends oll the restricted propagators to the full propagator acting on that Hilbert space. Taking the ordinary supertrace of the result is indeed the modular form which is the partition function.

This looks nice, but of course a full story would require an understanding of why precisely our would-be transport 2-functor assigns the above matrix H to the boundary of an annulus in the first place.

One natural idea would be to identitfy this matrix with a bimodule over the OPE algebra of the string states, obtained from some bimodule morphism given by transport along the annulus boundary. This almost seems to work, but not quite. I am not there yet. If anyone has any comments, I’d be grateful.

Posted by: urs on March 7, 2006 6:21 PM | Permalink | Reply to this
Read the post Seminar on 2-Vector Bundles and Elliptic Cohomology, V
Weblog: The String Coffee Table
Excerpt: Part V of a seminar on elliptic cohomology and 2-vector bundles. Review of relations between elliptic cohomology and strings.
Tracked: June 30, 2006 11:03 AM
Read the post Spectral Triples and Graph Field Theory
Weblog: The n-Category Café
Excerpt: Yan Soibelman is thinking about spectral stringy geometry.
Tracked: June 12, 2007 8:58 PM
Read the post QFT of Charged n-Particle: Towards 2-Functorial CFT
Weblog: The n-Category Café
Excerpt: Towards a 2-functorial description of 2-dimensional conformal field theory. A project description.
Tracked: August 6, 2007 11:45 PM

Post a New Comment