Patrick Lei

All notes

DAHA and knot homology

Columbia, Fall 2021.

DAHA and Knot Homology Learning Seminar
Fall 2021
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

These notes were taken during the seminar using the vimtex package of the editor neovim. Any errors are mine and not the speakers’. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the lecturers. If you find any errors, please contact me at plei@math.columbia.edu.

I patched the notes from Álvaro’s lecture using his notes, which contain some material not covered during the lecture. The notes here only contain material covered during the lecture.

1 Sebastian (Sep 14): Khovanov Homology

Recall that the Kauffman bracket of a link diagram D has axioms ⟨∅⟩=1, ⟨0⊔D⟩=(q+q−1)⁢⟨D⟩, and for any crossing

If D has n+ positive crossings and n− negative crossings, then the Jones polynomial of the underlying link L is J⁡(L)=(−1)n⁢qn+−2⁢n−⁢⟨D⟩.

We may have heard that Khovanov homology is a categorification of the Jones polynomial. To motivate this, there are quantum invariants like the Jones polynomial, the HOMFLY polynomial, WRT, and others. On the other side, there are gauge-theoretic invariants like instanton Floer homology and the Casson invariant. The quantum invariants generally come from representations of quantum groups are are combinatorial, but it is unclear what geometric information we obtain. On the other hand, the gauge theoretic invariants are more powerful, but of course harder to compute. We want to consider relations between the two types of invariants.

Some of the gauge-theoretic invariants end up being Euler characteristics; for example, the Casson invariant is twice the Euler characteristic of instanton Floer homology.

For a link diagram L, we will introduce a bigraded ℤ-module Kh∗,∗⁢(L) such that

1.1 Category of Pictures

Given D, we will order its crossings. For each crossing we have a 0-resolution (horizontal) and a 1-resolution (vertical). If n=n++n−, then each element of {0,1}n gets a complete resolution, which is a collection of circles in the plane. For example, consider the Hopf link with resolution.

12H=n+=0,n−=200011011
Figure 1.1: Hopf link and resolution

Then we have

and thus

Changing a 0-resolution to a 1-resolution is given by a saddle, or a cobordism.

Let 𝖢𝗈𝖻3¯ be the category with objects collections of oriented circles in ℝ2 and morphisms oriented cobordisms in ℝ2×[0,1]. For each Σ∈Hom⁡(O1,O2), we define deg⁡Σ=χ⁡(Σ). Then we add objects O⁢{m} for each O∈Obj⁡(𝖢𝗈𝖻3¯) and m∈ℤ. This gives a graded category 𝖢𝗈𝖻3. Now let 𝖬𝖺𝗍⁡(𝖢𝗈𝖻3) denote the additive closure of 𝖢𝗈𝖻3 and 𝖪𝗈𝗆⁡(𝖬𝖺𝗍⁡(𝖢𝗈𝖻3)) be its category of chain complexes.

Now for each α∈{0,1}n, we assign the object O⁢{m}, where O is the resolution of L obtained from the αi-resolution at the i-th crossing and m=∑αi. For each edge, we will assign the saddle cobordism between objects at each edge.

Figure 1.2: Saddle cobordism

To obtain d2=0, we need that all square faces in the diagram anti-commute. The problem is that in the diagram, the two cobordisms are actually the same (we can reorder the saddles), so now we need to add signs to the edges such that each square face has an odd numbder of minus signs. In the example of the Hopf link, we just add a minus sign to the 10→11 edge.

Now we may construct the m-th chain group [[L]]m of [[L]] to be

The chain homotopy type of [[L]] is a link invariant after passing to a quotient 𝖢𝗈𝖻3/ℓ of 𝖢𝗈𝖻3 by the following relations:

T=2,S=0,1234−1234−1234+1234=04⁢Tu
Figure 1.3: Relations for 𝖢𝗈𝖻3//ℓ

Now we define the Bar-Natan category 𝖡𝖭≔𝖪𝗈𝗆⁡(𝖬𝖺𝗍⁡(𝖢𝗈𝖻3/ℓ)). Of course, we now need to prove invariance under the Reidemeister moves. For the first Reidemeister move, we have the following diagram:

0000F0=−=G0d=
Figure 1.4: Diagram for R1

Then here hare the easy relations.

d⁢F0=−=0,G0⁢F0==2−=id
Figure 1.5: Easy relations

Finally, here we see that we have a homotopy:

Clearly F1⁢G1⏟=0−id+d⁢h=0.F0⁢G0−id+h⁢d=−−+=F0⁢G0idh⁢d=0by 4⁢Tu applied to1234
Figure 1.6: The homotopy really is a homotopy

The proofs of invariance under the other Reidemeister moves are similar, but significantly more complicated. But now any functor from 𝖢𝗈𝖻3/ℓ to any abelian category will give us a knot invariant. Note that such a functor does not have to be a 2-dimensional TQFT.

1.2 Khovanov Homology

Returning to something more concrete, any cobordism in 𝖢𝗈𝖻3/ℓ can be decomposed into pairs of pants and caps and cups, so to specify a functor, we only need to specify what it does to four morphisms. Here, the cobordisms correspond to

where V=ℱ⁡(S1) for ℱ a TQFT. Then Khovanov homology is the functor where V=ℤ⁢{−1}⊕ℤ⁢{1}, where m is given by

Δ is given by

ι is given by 1↦v+, and ε is given by v−↦1,v+↦0. The grading on V is called the Jones grading because it records power of q in the Jones polynomial.

Now we are finally able to define the Khovanov homology

Theorem 1.2.1.

We have the formula

For the Hopf link, the chain complex is given by

and thus we can compute the Khovanov homology.

Theorem 1.2.2 (Jacobsson, Bar-Natan).

The map on [[L]] associated to a cobordism in ℝ3×[0,1] is independent of its decomposition into elementary cobordisms up to a sign and cobordisms in the Bar-Natan category.

2 Álvaro (Sep 21): HOMFLY-PT homology

If we already know about Hecke algebras, then we could discover the HOMFLY-PT polynomial with the unknot being 1 and the skein relation

Here, if we set t=q2⁢N and z=q+q−1, we recover the S⁢L2 link polynomial and if we set N=1, we recover the Jones polynomial.

The reason that we could have discovered the HOMFLY-PT polynomial is using the representation theory of the braid group, which is generated by s1,…,sn−1 with relations si⁢si+1⁢si=si+1⁢si⁢si+1 and si⁢sj=sj⁢si if |i−j|≥2.

First, note that every link is the closure of a braid. Then two braids give rise to isotopic links if they are related by Markov moves. Now we consider

Now we can define a trace map

with the condition that Tr⁡(x⁢sn⁢y)=z⁢Tr⁡(x⁢y) and Tr⁡(1)=1. In particular, these imply that Tr⁡(x⁢y)=Tr⁡(y⁢x), and this is called the Oceanu trace. If we normalize the si, then Tr⁡(x⁢si)=Tr⁡(x⁢si−1). Then under the correct normalization, we obtain the HOMFLY-PT polynomial.

2.1 Triply graded homology

This was discovered by Khovanov-Rozansky in 2004 using matrix factorizations and Khovanov in 2005 using Soergel bimodules. Then work of various subsets of Elias, Hogancamp, and Mellit computed HHH for torus (m,n)-links.

Consider the symmetric group Sm=⟨s1,…,sm−1⟩. Then consider R=ℚ⁡[x1,…,xm], where the xi have degree 2. Then we have a natural action of Sm on R, and RSi are the invariants. Then we will define Bsi=R⊗RsR⁡(1) as an R-bimodule, and for w¯=si1⁢⋯⁢sin, we will define the Bott-Samuelson bimodule B⁢S⁢(w¯)=Bsi1⊗⋯⊗Bsin. Next, if we consider the Bott-Samuelson bimodules and add in all tensor products and direct sums, we obtain the Hecke category 𝖲𝖡𝗂𝗆, which categorifies the Hecke algebra.

In our running example, we will take S2={1,s} and R=ℚ⁡[x1,x2]. Here, we will take Rs=ℚ⁡[x1+x2,(x1−x2)2], and we will write r=x1+x2,t=x1−x2. Therefore, we have

Theorem 2.1.1 (Soergel’s categorification theorem).

Note that H⁡(Sn) has a KL-basis {bw∣w∈Sn}. Then there is a bijection between indecomposables in 𝖲𝖡𝗂𝗆 and the KL-basis taking the tensor product to multiplication.

Now if we write bsi=δsi+q, we have bsi2=q⁢bsi+q−1⁢bsi. On the other side, we have

Now we will categorify the braid group and construct HHH. We have bs=δs+q and thus δs=bs−q, so in the categorified version we will have qi:Bsi→R. Then we consider the complex

and analogously

Theorem 2.1.2 (Rouquier).

The assignment w¯→F∙⁢(w¯)=Fs1±⊗⋯⊗Fsin± is a well-defined homomorphism Brn→Kb⁢(𝖲𝖡𝗂𝗆).

Now we apply the Hochschild homology H⁢Hn to F∙⁢(w), and we obtain the complex

and taking the homology, we obtain the HHH homology. The grading is given by the homological grading H⁢Hi⁢(Fj⁢(w¯)) that modifies j, the Hochschild grading, and the internal grading as R-graded bimodules.

2.2 Hochschild homology

Next we observe that Hochschild homology is easier for polynomial rings. Recall that if R is a k-algebra, then if we write Re⁢n⁢v=R⊗kRo⁢p, we have

Instead of the bar complex, we will consider the Koszul complex

which is a free resolution of R, where V=ℚm. Then this complex computes the Hochschild homology. For example, if R=ℚ⁡[x]=M where deg⁡x=2, we have

and R⁡(−2)→0R, and thus

for another example, when R=ℚ⁡[x,y] and M is a graded R-bimodule, and deg⁡x=deg⁡y⁢ 2, we have

where the maps are given by (x⊗1−1⊗x,1⊗y−y⊗1) and (y⊗1−1⊗y,x⊗1−1⊗x), and the Koszul resolution of M is similar.

2.3 Example of triply graded homology

Now the unknot is the closure of 1∈Br1, and so we want to compute its homology HHH∗∗∗. We know that

and therefore

The Euler characteristic is

and this recovers the unknot.

Now if we take the closure of s∈Br2, which is still the unknot, we have

where R=ℚ⁡[x1,x2]. But now we need to compute the Hochschild homology

in all degrees. For R⁡(1), the Koszul resolution is

and for Bs it is

where Rs=ℚ⁡[s,t2] and f⁡(m)=12⁢t⁢m−12⁢m⁢t. Now we have ker⁡(f)=⟨m⁢t+t⁢m⟩,Im(f)=m⁢t−t⁢m. Now we compute H⁢H0⁢(Bs)=Bs/Imf, H⁢H∗⁢(Bs)=R⁡(−1)⊕ker⁡(f)⁢(−2), and H⁢H2⁢(Bs)=ker⁡(f)⁢(−4). Then the three maps are the identity in degree 0, (id,i) in degree 1, and i in degree 2, where i is the counit. Finally we obtain the following table for the HHH:

Table 2.1: HHH homology
Hochschild degree HHH0∗∗(s1) HHH1∗∗(s1)
0 0 0
1 0 ℚ⁡[r]⁢(−1)
2 0 ℚ⁡[r]⁢(−3)

This is isomorphic to H⁢H⁢H⁢(1) in the previous example after a shift.

Finally consider the Hopf link, which is the closure of s2. Here F∙⁢(s2) is given by

This complex is homotopic to Bs⁢(−1)→Bs⁢(1)→R⁡(2).

2.4 Connections to geometry

Define the matrix Bi⁢(z) to have the matrix (011z) in the i-th position. Then we have the identity

For a positive braid β=si1⁢⋯⁢sin, we assign the matrix Bβ⁢(z1,…,zn)=Bi1⁢(z1)⁢⋯⁢Bin⁢(zn), and finally we consider the variety

For example, if β is the trefoil knot, then X⁡(β)={1+z1⁢z2=0}. There is a torus action on this variety, where if we have n strands there is an action of (ℂ×)n−1. In this example, we have (z1,z2,z3)↦(z1⋅t,z2⋅t−1,z3).

Theorem 2.4.1 (Trinh, 2021).

We have H∗,B⁢MT⁢(X⁡(β)) has a nontrivial weight filtration whose associated graded module is isomorphic to HHH0∗∗(β).

For the trefoil knot, we obtain a grading on H∗⁢(S1).

Finally, relating this to Hilbert schemes we have the Oblomkov-Shende conjecture, where if C is an integral plane curve and p is a singularity, then

Next, there is the Oblomkov-Rasmussen-Shende conjecture, which says that replacing χ with H∗, we obtain HHH. Then there is Oblomkov-Rozansky homology, and finally a conjecture of Gorsky-Negut-Rasmussen, which says that if X=F⁢Hilbnd⁢g⁢(𝔸2) is a dg version of the flag Hilbert scheme, then there is a commutative diagram of monoidal functors

where 𝖵𝖾𝖼𝗍ℚℤ⊕3 is the category of ℤ⊕3-graded ℚ-vector spaces.

3 Avi (Sep 28): Hilbert Schemes

3.1 Definitions

Let X be a projective variety over a field k=ℂ. For example, if X=ℙn, we in algebraic geometry are interested in projective varieties, which are closed subschemes of X. We are interested in classifying closed subschemes of X, and to do this we will construct a functor Hilb⁡(X) given by

We are interested in the representability of this functor, and unfortunately this functor is not representable in any reasonable way.

For example, consider X=ℙ1. If we consider subschemes with finite support, we obtain the space

This has infinitely many connected components and is infinite-dimensional in general. This tells us that we want to stratify our functor into pieces.

Now consider Z⊂X=ℙn. Then define graded coordinate ring R=⨁iRi of Z and the function i↦dimRRi. For i≫0, this function equals some polynomial in i. More generally, let ℱ be a quasicoherent sheaf on X. Then we can consider the function

and in particular, for i≫0, this function equals the Euler characteristic χ⁡(ℱ⁡(i)) because for i sufficiently large, all twists have vanishing higher cohomology. Finally, χ⁡(ℱ⁡(i)) is a polynomial, which we will call the Hilbert polynomial Pℱ⁢(i).

For a subscheme Z⊂X with ideal sheaf ℐZ⊆𝒪X, we will define PZ=PℐZ. Returning to X=ℙ1, then if Z is a scheme which is n points, then dimH0⁢(X,ℐZ)=n. For the subscheme X, we have h0⁢(X,𝒪X⁢(i))=i+1. We are now able to define the functor

In particular, we have a decomposition

In our case, we will focus on 0-dimensional subschemes, and we will denote both the functor and the scheme that represents it as X[n]=Hilbn⁡(X).

Example 3.1.1.

Suppose n=0. All subschemes of T×X with PZt=0 must have empty fibers over T and thus must be empty themselves.

If n=1, then all ideals of colength 1 are maximal ideals, which are the same as closed points of X, so X[1]=X.

When n=2, we are looking for ideals ℐ such that 𝒪X/ℐ has length 2. Thus either this subscheme is supported at two distinct points or a single point. If the support of the subscheme is two distinct points x1≠x2, then at each xi, the ring 𝒪X/ℐ must have length 1. Thus we have a locus (X×X∖Δ)/S2. In the second case, where 𝒪X/ℐ is supported at a single point x, then we have a sequence of inclusions

Thus we are looking for morphisms φ:𝔪x/𝔪x2→k up to a scalar, so above the point x we have ℙ⁢TX,x, and therefore we have

More generally, we have a morphism π:X[n]→Symn⁢X sending a subscheme Z to its support |Z|, called the Hilbert-Chow morphism. If λ=(λ1,…,λr) is a partition of n, we have a stratum Symλ⁡X with fixed multiplicities given by λ. Thus we have

which induces a stratification of X[n] into pieces Xλ[n].

For example, when λ=(n), we have the Hilbert-Chow morphism

and so the fibers are schemes Xp[n] parameterizing length n subschemes of X supported at p.

Example 3.1.2.

When X is a smooth curve, π is actually an isomorphism. This is because all local rings 𝒪X,p are discrete valuation rings, and in particular they have a unique idea of length n. Therefore all fibers of π must be points.

When X is a smooth surface, then X[n] is always smooth, but in fact Hilbert schemes of points are highly singular for n>2 in higher dimension.

3.2 Hilbert schemes of smooth surfaces

Now suppose that X is a smooth surface. Suppose that Z⊆X is supported at a point p with length n. In particular, we have a chain of inclusions

and thus this data is equivalent to I/𝔪xn⊂𝒪X,x/𝔪xn−1. Up to completion, all of this data is independent of both X and x, and in fact

Therefore, Xx[n] is independent of X,x, so we may consider X=𝔸2,x=0. We will see that (𝔸2)0[n] has dimension n−1. In general, we have

In particular, dimX[n]=2⁢n.

Now we will compute the tangent space of X[n]. If a subscheme Z is given by an ideal ℐ, we first show that the tangent space to X[n] at ℐ is given by Hom⁡(ℐ,𝒪X/ℐ), and then we can use Hirzebruch-Riemann-Roch. Alternatively, consider the action of Gm×Gm on 𝔸2. This lifts to an action on (𝔸2)[n], and the fixed points are given by monomial ideals. Each monomial ideal corresponds to a Young diagram (just consider all monomials not in the ideal), and of course these are in bijection with partitions of n. If we do some allegedly very elegant combinatorics, we obtain the desired result.

We will now connect Hilbert schemes of surfaces to representation theory. We have the formula of Göttsche, which states that

and another formula which states that

In particular, doing some generatingfunctionology gives us the Betti numbers.

These formulas are similar to those coming from representations of the Heisenberg algebra. In fact, the total cohomology of the Hilbert scheme is a representation of the Heisenberg algebra. We consider the incidence schemes

which classifies subschemes Z1⊂Z2 which differ at a single point x.

3.3 Plane curve singularities

There exists a similar formula to those of Göttsche for smooth curves due to Macdonald, which says that

We would like to say something about the LHS, which we will call W⁡(C), in general when C is not smooth. Write C=Csm∪{pi}, where the pi are the singularities of C. Then we have

by the cut-and-paste relation for the Euler characteristic. To make sense of what happens at the singular points, we will consider knot theory.

Near p⊂C⊂ℂ2, we will consider C∩S3, which has real dimension 1 and is therefore a link Lp. We would hope that some knot invariant of Lp will recover the data of the fomula above, and according to the Oblomkov-Shende conjecture, this is true. The conjecture says that

where P is the HOMFLY polynomial. To recover the entire Homfly polynomial, we have an upgraded version of the conjecture, which says that

where Cp,m[n] parameterizes colength n ideals ℐ generated by m elements. This conjecture is known for torus knots and in the special case when t=1, among others.

4 Patrick (Oct 5 and Oct 12): The HOMFLY polynomial and enumerative geometry

Note: these are the speaker’s notes. There are several mistakes, but I am too lazy to correct them.

4.1 Introduction

Let C⊂ℂ2 be a reduced curve and suppose p∈C is an (isolated) singularity. Then define a constructible function m:Cp[n]→ℕ given by m⁡([Z]) being the number of generators of the ideal IZ,p⊂𝒪C,p. Now consider the generating function

Additionally, taking a small S3 around p∈ℂ2 and intersecting with C gives us a link ℒC,p. Recall the HOMFLY polynomial 𝖯⁡(ℒ;v,s)∈ℤ⁡[v±,(s−s−1)±]. If μ is the Milnor number of the singularity (for example, the middle Betti number of the Milnor fiber), then the Oblomkov-Shende conjecture is

Theorem 4.1.1 (Maulik).

The proof of this result relates both sides of the equality to enumerative geometry, and in particular Pandharipande-Thomas, or stable pairs, curve counting. In this lecture, we will (attempt to) sketch a proof of the Oblomkov-Shende conjecture, but first we will introduce PT theory to familiarize ourselves with the objects involved in the proof.

Remark 4.1.2.

Maulik proves everything for a colored version of the HOMFLY polynomial, but here we will only work with uncolored data, which corresponds to all partitions being (1).

4.2 Introduction to PT Theory

In this part, we are following the papers Curve counting via stable pairs in the derived category and Stable pairs and BPS invariants by Pandharipande and Thomas.

Let X be a smooth threefold and β∈H2⁢(X,ℤ). The PT moduli space Pn⁢(X,β) parameterizes two-term complexes

where ℱ is a pure 1-dimensional sheaf supported on a Cohen-Macaulay subcurve of X, s has 0-dimensional cokernel, χ⁡(ℱ)=n, and [supp⁡F]=β. The space Pn⁢(X,β) has a virtual fundamental class coming from the deformation theory of complexes in the derived category. Here, note that the deformation theory of (ℱ,s) (really of the corresponding complex ℐ∙) is given by

and the virtual fundamental class lives in dimension

In particular, for a Calabi-Yau threefold, cβ=0. Given this data, we may define the PT invariant

and the P⁢T partition function

There is an alternative way to compute the PT invariants for X a projective Calabi-Yau threefold, due to Behrend (originally done for DT theory). In this case, the moduli space is actually a projective scheme, and if Pn⁢(X,β) is smooth everywhere, then

Of course, moduli spaces are almost never smooth for nontrivial moduli problems, so instead, we have

Proposition 4.2.1 (Behrend).

There exists an integer-valued constructible function χB such that

4.3 A computation in PT theory

This computation comes from The 3-fold vertex via stable pairs by Pandharipande and Thomas.

We will now compute the T-equivariant PT vertex of ℂ3 (this is the local model for toric varieties) to give us all a feel for this enumerative theory. First, we need to define some combinatorial data. Let μ=(μ1,μ2,μ3) be a triple of partitions. Then there exists a unique minimal T-fixed subscheme Cμ with outgoing partitions the μi (simply take the curves Cμi given by the three partitions and take the union), whose ideal we will denote ℐμ. Then we define

Every T-invariant pair (ℱ,s) on ℂ3 corresponds to a finitely-generated T-invariant submodule

and we will now give a combinatorial description of such submodules.

For each μi, we may consider the module Mi as an infinite cylinder Cyli⊂ℤ3 (extending in both directions). Then for every w∈ℤ3, consider the vectors 𝟏w,𝟐w,𝟑w representing w in each copy of ℤ3 (for each of the Mi). Clearly x1 shifts w by (1,0,0) and similarly for x2,x3. We will now consider the decomposition of the union of the Cyli into the following types:

  • •

    Type I+ are those which have only nonnegative coordinates and lie in exactly one cylinder;

  • •

    Type II (resp III) are those which lie in exactly 2 (resp 3) cylinders;

  • •

    Type I− are those with at least one negative coordinate.

Clearly M/⟨(1,1,1)⟩ is supported on types II,III,I−, and now we have three cases:

  • •

    If w∈I−, then clearly ℂ⋅𝐢w⊂M/𝒪Cμ;

  • •

    If w∈II, then ℂ⋅𝐢w⊕ℂ⋅𝐣wℂ⋅(𝐢w+𝐣w)≅ℂ;

  • •

    If w∈III, then ℂ⋅𝟏w⊕ℂ⋅𝟐w⊕ℂ⋅𝟑wℂ⋅(𝟏w+𝟐w+𝟑w)≅ℂ2.

In particular, we need to consider labelled box configurations, where type III boxes may be labeled by an element of ℙ1 (where ℂ2 is identified with the vector space above) or unlabeled (corresponding to the inclusion of the entire ℂ2 in Q). Now we will denote by 𝒬μ the set of components of the moduli space of T-invariant submodules of M/𝒪Cμ.

We are now able to define the equivariant vertex. Let ℓ⁡(𝒬) be the number of boxes in the labelled configuration associated to 𝒬∈𝒬μ. Then let |μ| denote the renormalized volume of the partition π corresponding to ℐCμ, which is defined as

which is independent of a sufficiently large N≫0.

We need to define a few characters of T, which we will need to define the vertex and compute our example. Let P be the Poincaré polynomial of a free resolution of the universal complex 𝕀 on 𝒬×ℂ3. Denote by 𝖥 the character of ℱ. In particular, we have

Then define a vertex character 𝖵 by

where 𝖦α⁢β is a certain character defined by edge data. Now let

be the contribution of 𝖵 on 𝒬. Then the equivariant vertex is defined to be

Example 4.3.1.

For μ=((1),∅,∅), we have

To see this, note that 𝒬μ=ℤ>0, where k corresponds to the length k box configuration in the negative x1-direction.

Now we simply compute that

This implies that

and therefore that

as desired.

4.4 Flop invariance of PT theory

We will now begin the proof of the Oblomkov-Shende conjecture. The first step is to understand what happens when we blow up C at p via flop invariance of PT partition functions.

Let Y be the total space of the bundle 𝒪⁡(−1)⊕𝒪⁡(−1) and Y− be the threefold obtained via a flop of the zero section. The flop is some birational map ϕ:Y⇢Y−. If π:Y→ℙ1 is the projection, choose an identification of ℂ2 with π−1⁢(0). Then the strict transform of π−1⁢(0) with respect to ϕ is Bl0⁡ℂ2 with exceptional fiber E− which is the zero section of Y− (isomorphic to Y). This implies that the strict transform of C with respect to ϕ is Bl0⁡C.

Definition 4.4.1.

A stable pair (ℱ,s) is C-framed if on Y∖E if after restricting to Y∖E, we have an isomorphism (ℱ,s)≃𝒪Y→𝒪C.

Given r,n∈ℤ, we define the moduli space P⁡(Y,C,r,n) of C-framed stable pairs such that supp⁡F has generic multiplicity r along E and for any projective compactification Y¯ of Y, we have χ⁡(ℱ¯)=n+χ⁡(𝒪C¯). This is a locally closed subscheme of the space of stable pairs on Y¯ and is independent of the choice of Y¯. Now we define the C-framed PT partition function

Proposition 4.4.2.

Let m1,…,mr be the multiplicities of the branches of C at p. We have the flop identity

where

is the normalized PT partition function and δ=(∑mi2).

4.5 Algebraic links

We will now study what happens to an algebraic link under blowup. Recall that if the singularity (C,0) is irreducible, then we can describe ℒC,0 as an iterated torus knot using the Puiseux series of C at 0. If the Puiseux series is

then ℒC,0 is simply the iterated torus knot with parameters (q0,p0),…,(qs,ps). To help us compute things, we will project all of our diagrams into the annulus and use skein theory. Before we do this, we need to review skein theory.

Definition 4.5.1.

Let F⊂ℝ2 be a surface with boundary and designated input and output points. The framed Homfly skein over Λ≔ℤ⁡[v±,s±,(sr−s−r)−1∣r≥1] is the Λ-module generated by oriented diagrams in F up to isotopy, R1, R2, and the skein relations1

If F is a rectangle with m inputs and outputs, then the skein ℋm has a product given by stacking diagrams, and is isomorphic to the Am Hecke algebra. If F is an annulus, the skein 𝒞 is a commutative algebra with product obtained by placing one annulus inside another. There is a Λ-module morphism ⨁ℋm→𝒞 given by sending a braid to its closure. The algebra 𝒞+ generated by the image is isomorphic to the ring of symmetric functions with coefficients in Λ, and we will denote by Qλ the diagram associaated to the Schur function sλ. Finally, if F=ℝ2, then the skein is simply the ring Λ. This gives us a trace map ⟨⟩:𝒞→Λ. Up to some monomial factor, the trace gives us the HOMFLY polynomial.

Now we will discuss a satellite construction. This is how we will turn algebraic links into diagrams in the annulus for computations.

Definition 4.5.2.

Let ℒ be a framed link with r components and Q1,…,Qr be diagrams in the annulus with counterclockwise orientation. The satellite link ℒ∗(Q1,…,Qr) is obtained by drawing Qi on the neighborhood of the i-th strand of L. If ℒ comes from a counterclockwise-oriented diagram in the annulus, then this construction only depends on the equivalence classes of the decorations.

Remark 4.5.3.

Using these terms, coloring a link ℒ just means giving each component of ℒ a partition and considering the link ℒ∗(Qλ1,…,Qλr).

Now the iterated torus knot ℒC,0 with parameters (qi,pi) can be embedded as a diagram LC in the annulus, where

and Tpq is the diagram of the (q,p)-torus knot.

For the general case of an algebraic link ℒC,0 where (C,0) is not irreducible, there is a more complicated satellite construction for constructing a diagram in the annulus. This produces satellite operators Spq∗(−,−).

Before we continue, we will construct several objects that we will need later. Recall that the diagram Tmn is the n-th power of the diagram below:

Refer to caption
Figure 4.1: The diagram β5

Next, we will consider the diagram σm below and denote its n-th power by Smn. These are required for the satellite construction for links, but we will not need it here.

Refer to caption
Figure 4.2: The diagram σ5

To conclude this section, we will explain how to actually construct the link diagram in the annulus for a non-reducible singularity. Let C1,…,Cr be the branches of C at 0 and consider truncated Puiseux series

Write αi=qipi and consider the pairs (αi,ai). It is always possible to find a finite truncation of the Puiseux series that does not affect the topological type of the link, so the inductive process we will define is actually finite. For each (α,a), let {i0,…,in} be the set of indices with associated pair (α,a). Assume we have an annulus diagram L(α,a) and set

We may assume that α1<⋯<αk, and the link ℒC can be represented by the annulus diagram

4.6 Some explicit calculations for the unknot and Hopf link

The following is already apparently known.

Proposition 4.6.1.

For any partition λ,

Let X∈𝒞+ be a counterclockwise-oriented diagram. Define the meridian operator 𝖬X on 𝒞+ by the construction below:

Refer to caption
Figure 4.3: Meridian operator

For a partition μ, the Schur function Qμ is an eigenvector for 𝖬X with eigenvalue tμ⁢(X). Then the HOMFLY polynomial for the colored Hopf link decorated by μ,λ is simply tμ⁢(Qλ)⁢(tμ). In the remainder of this section, we will describe the operator tμ, which is a ring homomorphism 𝒞+→Λ. Define the function

Then we have tμ⁢(Qλ)=sλ⁢(Eμ⁢(t)), where for a power series E⁡(t)=∑Ek⁢tk, we write sλ as a polynomial in the ek and then substitute ek←Ek.

4.7 Behavior of links with respect to blowup

Having discussed the behavior of enumerative invariants with respect to blowups, we need to discuss the behavior of the link ℒC,0 with respect to blowing up the origin. Let C1,…,Cr be the irreducible components of C through 0 and denote their strict transforms by Ci′. At each point p1,…,pe∈E (the exceptional divisor) let Dk denote the singularity of C′∪E at pk and Bk be the singularity of C′ at pk. Choose (truncated) Puiseux expansions

For each of the branches Ci. We may also assume that qipi≥1 for all i. If we blow up at the origin, consider the chart with coordinates (x,y=x⁢w). Substitution, we obtain the new Puiseux expansion

for Ci′ at pk. In particular, we obtain the relation

If we perform deeper analysis, we obtain the following result:

Proposition 4.7.1.

If any αi=qipi>1, then we have

Otherwise, we have

Now we will write down a blowup identity for links. The idea is to use the topological vertex (originally introduced by Aganagic-Klemm-Marino-Vafa) and its relationship with Chern-Sinons invariants of the unknot. For a partition μ, define

Here, κμ=2⁢∑□∈μc⁡(□). By the computation of the colored HOMFLY polynomial of the colored unknot (sorry), we obtain the identity

4.8 Relation between PT theory and HOMFLY

We will prove a relationship between the PT partition function for C-framed stable pairs in Y=𝒪ℙ1⁢(1)⊕𝒪ℙ1⁢(1) and the HOMFLY polynomial for ℒC,0. We will focus on the simplest case of a node with two branches. The general case reduces to this one by careful checking of what happens on both sides under blowup of C at 0.

Proposition 4.8.1.

We have the identity (possibly up to monomials)

where the superscript low means we take the lowest degree terms.

We only need to prove this for the unknot and the Hopf link.

Proof.

This apparently follows from the fact that the topological vertex calculates both the v=0 specialization of HOMFLY of the Hopf link and the stable pairs vertex. See the references to Maulik’s paper for a reference. □

5 Che (Oct 19): Braid varieties and Khovanov-Rozansky homology

5.1 Khovanov-Rozansky homology

Let G=G⁢Ln and 𝔥 be the Cartan. Write R=ℂ⁡[𝔥]=ℂ⁡[x1,…,xn], and this has an action of the Weyl group permuting the coordinates. We will specify deg⁡xi=2. We may now define Soergel bimodules

Now define Ti=[Bi⁢(−1)→R] by 1⊗1↦1 and Ti−1=[R→Bi⁢(−1)] by 1↦xi⊗1−1⊗xi+1′. We may now form the Rouquier complex for a braid β=σi1⁢⋯⁢σik as

For example, if β=σ2, then we have

We may now take the Hochschild homology HHi⁢(Tβ). We will mostly be concerned with HH0⁢(M)=Hom⁡(R,M). Continuing our example, we have

We are finally able to define the Khovanov-Rozansky homology

This has gradings A for the Hochschild grading, Q for the internal degree in R, and T for the homological grading. Continuing our example again, we have

Remark 5.1.1.

We will usually multiply our HHH by 1−Q2 as a useful normalization. This is a consequence of us considering G=G⁢Ln instead of G=S⁢Ln.

5.2 Braid varieties

Write Bi⁢(z) for the matrix with 1s on the diagonal except for a (011z) block on the diagonal in the i,i+1 positions. Given β=σi1⁢⋯⁢σik, we consider the matrix

Then we may define the braid variety

For example, if β=σ2 for 2 strands, we have

and so

Similarly, for β=σ3, we have the matrix

and thus we have X⁡(β)=(1+z1⁢z2=0)×ℂz3. Similarly, we may compute

These braid varieties come with a torus action. We would like to write

For example, we have

Doing this on Bβ⁢(z1,…,zk), we obtain an action of T=(ℂ×)n on X⁡(β). Continuing our example, we have

Here are some facts:

  1. 1.

    Under mild assumptions, X⁡(β) is smooth.

  2. 2.

    If β closes to a knot, then the action of T is free.

We are now ready to state the main result.

Theorem 5.2.1.

We have an equality

up to a global shift in grading, where w is the weight filtration.

5.3 Mixed Hodge Structures

To understand the weight filtration, we will discuss mixed Hodge structures. For a smooth projective variety X of dimension n, recall the Hodge decomposition

Of course, for general X which are not necessarily smooth or projective, this decomposition will fail. In this case, there is a weight filtration

which morally measures how much the Hodge decomposition fails.

Theorem 5.3.1.

Let X be a smooth algebraic variety inside a smooth and projective X¯ such that D=X¯∖X is a normal crossings divisor. Write D=⋃iDi as a union of its irreducible components. Then there exists a spectral sequence

which collapses at the E2 page. Here, D(p) is the locus of the p-fold intersection.

Note that our spectral sequence looks like

For example, when X=ℂ× and X¯=ℙ1, we have D=0+∞, and so we obtain

and for example, grw2⁢H1⁢(ℂ×)=ℂ.

5.4 Sketch of proof of main theorem

What we will do is to construct a Bott-Semelson variety BS⁡(β) from the Khovanov-Rozansky homology, then as a fiber over the flag variety we will consider the brick variety of β, which is a compactification of X⁡(β).

Definition 5.4.1.

Given β=σi1⁢⋯⁢σik, define the Bott-Samelson variety

For example, when β=σ1⁢σ2, we have

Then (F0,F1,F2)∈BS⁡(β). Now we define the open Bott-Samelson variety

We have the projection pr:BS⁡(β)→G/B sending a tuple of flags to Fk. Then we define

where δ⁡(β)∈W is the Demazure product. Finally, define brick0⁢(β)=brick⁡(β)∩OBS⁡(β).

Theorem 5.4.2.
  1. 1.

    X⁡(β;δ⁡(β))≅brick0⁢(β), where β is β with the transpositions reversed.

  2. 2.

    The complement brick⁡(β)∖brick0⁢(β) is a normal crossings divisor.

  3. 3.

    The brick variety brick⁡(β) has a stratification by brick⁡(β′) for subwords β′ of β such that δ⁡(β)=δ⁡(β′).

Remark 5.4.3.

We define X⁡(β,w) for some w∈W as

For example, we have X⁡(σ3,(12))={1+z1⁢z2≠0}. On the other hand, brick⁡(σ3)=ℙ1×ℙ1, brick⁡(σ2)=ℙ1, and brick⁡(σ)=pt.

Theorem 5.4.4.

As an R−R bimodule, we have

6 Sam (Oct 26 and Nov 09): Finite dimensional type A rational Cherednik representations

Let R be the An−1 root system, W be the Weyl group Sn−1, and 𝔥 be the n−1-dimensional representation. Sitting inside W there is a canonical set of generators.

Definition 6.0.1.

Given a constant c, define the rational Cherednik algebra to be the algebra generated by Sym⁡(𝔥∗)=ℂ⁡[𝔥], Sym⁡(𝔥)=ℂ⁡[𝔥∗], and ℂ⁢W with the relations

  • •

    [xi,xj]=0=[yi,yj];

  • •

    w⁢x⁢w−1=w⁡(x) and −w⁢y⁢w−1=w⁡(y);

  • •

    [y,x]=⟨y,x⟩−c⁢∑s∈S⟨y,αs⟩⁢⟨αs∨,x⟩⁢s,

where x1,…,xn−1∈Sym⁡(𝔥∗) and y1,…,yn−1∈Sym⁡(𝔥) are the canonical set of generators.

This has some induced representations. If τ is a representation of W, then there is an action of ℂ⁡[𝔥∗]⋊W on τ̂, which is the same vector space. This is given by P↦P⁡(0). Then we may define

Here are some basic properties:

  1. 1.

    There is a grading where xi have degree 1, yi have degree −1, and w has degree 0.

  2. 2.

    There is a filtration given by taking the total degree in x and y.

A stranger property is that there is some 𝔰⁢𝔩2-triple. Here, if

then E,𝐡,F is an 𝔰⁢𝔩2-triple. This integrates to an S⁢L⁢(2,ℂ) action given by xi↦yi,yi↦−xi, which is called the Fourier transform.

6.1 Relation to KZ-connection

Let ℂ⁡[𝔥]=H0⁢(𝔥)=M⁡(triv) and define operators

These are called the Dunkel operators.

Exercise 6.1.1.

If yi=Di, then this gives the relations in the rational DAHA for 𝔤⁢𝔩n−1.

Corollary 6.1.2.

Because [Di,Dj]=0, then a module M⁡(τ) gives us a bundle with meromorphic connection on 𝔥. Note that the poles along hyperplanes are of order 1, which means we have regular singularities and thus we can study representations of DAHA using the Riemann-Hilbert correspondence.

6.2 Background on ℋqc-representations

The Hecke algebra ℋq is a deformation of Sn=ℋ1. Given a parameter c, we obtain a parameter qc=e2⁢π⁢i⁢c. Now there is a theory of Specht-modules.

Theorem 6.2.1 (Dipper-James).
  1. 1.

    There is a flat (over 𝔸q1) family Sλ of ℋQ-modules nduced by any partition λ of n. This exhausts without repetition the representations for q not a root of unity and agrees with the usual Specht module Sλ for q=1.

  2. 2.

    When q is a root of unity, these are not always irreducible. If e is the order of q and λ is e-regular (the multiplicity of every λ(i) is less than e), then Sλ is irreducible.

  3. 3.

    There always exists a simple quotient Dλ of Sλ.

  4. 4.

    There is a bijection between blocks of representations of ℋq and e-cores. An e-core is a partition λ of n without e-hooks.

For e=n, either λ is an e-core or λ is one of (n),(n−1,1),…,(1,…,1). In this case, we obtain the representations ⋀0𝔥,⋀1𝔥,…,⋀n−1𝔥. Clearly ⋀0𝔥 is the trivial representation, ⋀1𝔥=𝔥, ad ⋀n−1𝔥 is the sign representation.

Proposition 6.2.2.

The composition series of Sλi is given by

whenever these are defined. Moreover, Dλ−a=0,Dλn−1=0, but the other Dλi≠0 for i<n−1.

Example 6.2.3.

For n=3, we have the composition series

If we take ℋζ3, the action on 𝔥 has a trivial 1-dimensional submodule.

6.3 The KZ functor

Consider the weight zero summand 𝒪⁡(ℋc) of category 𝒪. Then the monodromy gives us a local system on 𝔥reg=𝔥∖⋃Hα. Remember that the root hyperplanes were the singularities of our connection Di. Given M⁡(τ), we obtain the module

and in fact this is the monodromy. This gives us a W-equivariant local system on 𝔥reg, and these are the same thing as 𝖱𝖾𝗉⁡(π1⁢(𝔥reg/W))=𝖱𝖾𝗉⁡(Bn). We claim that this gives an equivalence

Proof.

We will give a “proof” of this result. The first thing we need to prove that (Ti−qc)⁢(Ti−1), where Ti is the monodromy around γi, where γi is a small loop around Hαi. In the n=2 case, our Dunkel operator is simply

For the trivial and nontrivial representation of ℤ/2⁢ℤ, we obtain either

But now we may compute the monodromy. In the case of ∂∂xi, we have trivial monodromy. In the harder case, if we complexify to z and take ξ=z (corresponding to going halfway around a hyperplane), then we have

This clearly has monodromy e2⁢π⁢i⁢c=qc because we have the fundamental solution ϕ=ec⁢log⁡ξ. Then there is something that reduces the general case to this calculation.

This functor factors through torsion because torsion is the kernel of ℂ[𝔥reg]⊗−. This was the first step, so this is clear. We now prove equivalence. This follows from the Riemann-Hilbert correspondence once we prove regularity at infinity in some compactification. This computation is omitted. □

We will use

to give a representation of L⁡(triv)=M⁡(triv)/J⁡(triv).

Theorem 6.3.1 (BGG (BEG)).

There is a resolution

This result follows from the fact that the block of representations of the Hecke algebra has n−2 simple representations, so once we finish at M⁡(triv), we must have a torsion module, which must be L⁡(triv). If c=mn>0 and (m,n)=1, then L⁡(triv) is finite-dimensional.

Example 6.3.2.

By definition, it is clear that M⁡(triv)=ℂ⁡[𝔥]. Thus the BGG resolution is the Koszul resolution of Spec⁡L⁡(triv). When n=2 and c=23, the image Im(M⁡(Λ1⁢𝔥)→M⁡(triv)) is the ideal

where u1,u2 are x1−x2,x2−x3. These are just two reducible pairs of lines, and so then if we take Spec⁡L2/3⁢(triv), we obtain the subscheme defined by the above ideal. Clearly L2/3⁢(triv) is the ring of functions on this scheme.

In the A3 case, we have Spec⁡Lr/4⁢(triv) is a local complete intersection. This is derived smooth, and so we have some kind of Poincaré duality.

6.4 Characters for Hc representations

Let 𝐡=12⁢∑xi⁢yi, where yi is a basis for 𝔥∗ and xi is the dual basis. Then V∈𝒪⁡(Hc) splits as

where 𝐡 acts by a. Because W commutes with 𝐡 we can keep track of the 𝕙-grading and W-characters.

Definition 6.4.1.

We define

Now we will define the character for Mc⁢(τ). We an 𝐡×W-module, we have

Therefore we have

Here, κ⁡(c,τ) is the lowest weight.

We will now compute κ for τ=Λ∙⁢𝔥. To begin, suppose that τ is the trivial representation. Then the lowest weight vector is 1 and the yi act by 0. We now have

Next, we use the Koszul-BGG resolution and see that

Finally, we compute the characters for Lc⁢(triv) and prove that they exhaust finite-dimensional representations. Suppose [L]∈𝒪⁡(Hc) is finite-dimensional. For c>0 we have Hc≃H−c. We know that L is in the block containing Mc⁢(Λi⁢𝔥). Therefore we have

This gives us

This is a degree n−1 polynomial Q⁡(qc⁢n) and also that the trace is finite as q→1, so in fact Q⁡(qc⁢n)∝(1−q)n−1. Therefore ai=(−1)i assuming that dimL>0 and L is simple. But now we see that L=L⁡(triv), and so

7 Cailan (Nov 09 and Nov 16): Torus knots and the rational DAHA

7.1 Representation theory

We will begin by changing Sam’s formula until it is in the form that Cailan wants:

Definition 7.1.1.

Let L be a representation of Sn. Define ch:𝖱𝖾𝗉⁡(Sn)→Λn by

where pi are the power sums and ki⁢(θ) is the number of cycles of length i in θ.

Remark 7.1.2.

We have ch⁡(Sλ)=sλ. Also, [Sλ] is a basis for K0⁢(𝖱𝖾𝗉⁡(Sn)), and in fact, there is an isomorphism of Hopf algebras

with the ring of symmetric polynomials in infinitely many variables.

Lemma 7.1.3.

The reflection (representation) of Sn is isomorphic to ℂn/ℂ⋅(x1+⋯⁢xn).

Lemma 7.1.4.

Let T:V→V be a linear operator and let V=V1⊕⋯⊕Vk be a sum of T-invaraint subspaces. Then χT⁢(q)=χT|V1⁢(q)⁢⋯⁢χT|Vk⁢(q).

Proof.

Clearly T−q⁢I is a block matrix. □

Proposition 7.1.5.

For each σ∈Sn acting in the reflection representation h, we have

Proposition 7.1.6.

For A:V→V with dimV=n, we have

Proof.

Note that ℂn=h⊕triv. Therefore by the previous lemma, we have

By conjugation invariance of the characteristic polynomial, it is invariant under permutation matrices, which corresponds to conjugation in Sn, and so the trace depends only on the cycle type. For each cycle c in σ of length i, there exists a σ-invariant subspace Vc of ℂn of dimension i. Clearly the trace only depends on the length i. Also, we know that

and each Vci is σ-invariant. Using the lemma again, we have

It is easy to check that χTi⁢(q)=(−1)i⁢(qi−1), and therefore we have

as desired. □

Theorem 7.1.7.

Let Fm/n⁢(q,pi)=gch⁡(Lm.n) be the graded Frobenius character. This is equal to ∑ich⁡((Lm/n)i)⁢qi. Fixing m, we claim that the function

satisfies the identity

where [m]q=qm/2−q−m/2q1/2−q−1/2.

Proof.

Let δm,n=(m−1)⁢(n−1)2. Using Sam’s formula, we have

Therefore

By an easy exercise, it is easy to see that q−δm,n⁢1−q1−qm=qn⁡(1−m)2[m]1. Therefore we have

Using the fact that ∑ii⁢ki⁢(λ)=n, we have

This tells us that

We now have

Remark 7.1.8.

The Frobenius character is really the same information as {χL⁢(g)}g∈Sn. For id=(1n), we see that ⟨p1n⟩⁢ch⁡L=dimLn!, and thus dimL=n!⁢⟨ch⁡L,p1n⟩. Also, ch⁡(L) is the generating function for

Theorem 7.1.9.

As a graded Sn-representation, we have

Proof.

It suffices to prove this at the level of the Frobenius character. We have the Cauchy identity

and here we will specialize to

This implies that

Proposition 7.1.10.

We have

Example 7.1.11.

We have L3/2=(q+q−1)⁢S(2)⊕S(1,1).

7.2 HOMFLY polynomial of torus knots

Definition 7.2.1.

We will define the HOMFLY polynomial using the same skein relations as before but with P⁡(unknot)=1.

Example 7.2.2.

For the trefoil T⁡(2,3) we have P⁡(T⁡(2,3))=a2⁢(q2−q−2−a2).

Definition 7.2.3.

Define Hq⁢(n) to be the quotient of ℤ⁡[q±]⁢[Bn] by the relation σi2=(q−1)⁢σi+q. Let gi=[σi]∈Hq⁢(n).

Here, q will always be generic, unlike in Sam’s lecture.

Definition 7.2.4.

The Jones-Ocneanu trace tr:⋃n≥1Hq⁢(n)→ℤ⁡[q±]⁢[z] is the unique linear map such that tr⁡(a⁢b)=tr⁡(b⁢a), tr⁡(1)=1, and tr⁡(x⁢gn)=z⁢tr⁡(x) for any x∈Hq⁢(n).

Theorem 7.2.5 (Jones).

Let B∈Bn. Let χB⁢(q,λ)=f⁡(q,λ)⁢tr⁡([B])|z=1−q1−λ⁢q. Then

Theorem 7.2.6 (Ocneanu).

Let x∈Hq⁢(n). Then

Proof.

We know that Hq⁢(n)=⨁λ⊢nEnd⁡(Sλ⁢(q)) for q generic and any f:Mk⁢(ℂ)→ℂ satisfying commutativity is a scalar multiple of the trace. □

Definition 7.2.7.

We define T⁡(m,n) to be the closure of the braid (σ1⁢⋯⁢σn−1)m.

Remark 7.2.8.

T⁡(m,n) is a knot if and only if (m,n)=1. In addition, T⁡(m,n)=T⁡(n,m).

Let πλ:Hq⁢(n)→End⁡(Sλ⁢(q)).

Lemma 7.2.9.

Fix λ⊢n. Define ei=1+πλ⁢(gi)1+q. Then ei2=ei. Moreover, let dimSλ⁢(q)=d and rkSλ⁢(q)⁡ei=r. Then

Proof.

We omit checking that ei is idempotent. Next, we have a very important fact that F⁢Tn=(σ1⁢⋯⁢σn−1)n is central in Bn. Therefore, πλ⁢((g1⁢⋯⁢gn−1)n)=c⁢idSλ for some c∈C. By definition, πλ⁢(gi)=q⁢ei−(1−ei), and so in some basis of Sλ⁢(q), we have

This implies that det⁡πλ⁢(gi)=±qr, and therefore det⁡(πλ⁢((g1⁢⋯⁢gn−1)n))=qr⁢n⁢(n−1), so cd=qr⁢n⁢(n−1), and thus c=w⁡(q)⁢qr⁢n⁢(n−1)/d, where w:ℂ→ζd is continuous. But now w is constant because ℂ is connected and ζd is discrete.

We can now thus evaluate at q=1. Note that g1⁢⋯⁢gn−1=(1⁢n⁢(n−1)⁢⋯⁢2) at q=1, and therefore because (1⁢n⁢⋯⁢2)n=1, we see that c=1. □

Lemma 7.2.10.

The matrix Aλ⁢(q)=q−r⁡(n−1)/d⁢πλ⁢(g1⁢⋯⁢gn−1) is conjugate to the matrix for (1⁢n⁢⋯⁢2) acting on Sλ.

Proof.

We know that Aλ⁢(q)n−I=0, and thus Aλ⁢(q) is diagonalizable. In particular, the conjugacy class is determined by the eigenvalues and their multiplicities. By the same continuity argument, we can evaluate at q=1. □

Corollary 7.2.11.

Suppose (m,n)=1. Then

Proof.

By the lemma, we have

by Galois theory, where λi are the eigenvalues of (n⁢⋯⁢1). Now everything follows from the Murnaghan-Nakayama rule. □

Theorem 7.2.12.

Suppose (m,n)=1. Then

Here, ⟨n⟩=qn−q−n.

Proof.

The only thing left is r=rkSλ⁢(q)⁡(ei). By the braid relations, all σi are conjugate to σ1 in Bn, and thus we only need to find the rank of e1. Note that e1∈Hq⁢(2). Then the rank can be computed in Hq⁢(2)⊂Hq⁢(n), and so we use the branching rule. There are two irreducibles, where g1 acts by q on the trivial representation and by −1 on the sign representation. Therefore π(2)⁢(e1)=1 and π(1,1)⁢(e1)=0.

This implies that r is the multiplicity of the trivial representation S(2)⁢(q), and this is given by (a+b−1a). Finally, d=dim is given by (a+ba) by the hook length formula (here we specialize λ=Ha,b). □

Theorem 7.2.13 (GORS).

The graded Frobenius character of Lm/n coincides with the HOMFLY polynomial of the (m,n) torus knot when (m,n)=1. Specifically, we have

Proof.

We simply compute

Because Λk⁢𝔥=Sk,n−1−k, we use some generating function magic to obtain the desired result. □

Corollary 7.2.14 (Rank-level duality).

For (m,n)=1, we have

8 Davis (Nov 23): Hilbert schemes, Coulomb branches, and DAHA

8.1 Affine grassmannians

We will begin with some abstract nonsense discussing affine Grassmannians.

Definition 8.1.1.

An A-family of vector bundles on the formal disc D (or the punctured formal disc D∘) is a projective finitely generated module over A⁡[[t]] (or A⁡((t))).

These give categories 𝖵𝖾𝖼𝗍A⁢(D),𝖵𝖾𝖼𝗍A⁢(D∘).

Definition 8.1.2.

The affine Grassmannian for a group G, denoted GrG is the functor

where P is an A-family of G-bundles (𝖱𝖾𝗉⁡(G)→𝖵𝖾𝖼𝗍A⁢(D)) over D and γ is a trivialization of P over D∘.

Example 8.1.3.

If G=G⁢Ln, then GrG⁢Ln⁢(A) the set of finitely generated projecive A⁡[[t]]-submodules of A⁢((t))⊕n such that for some m, we have

Remark 8.1.4.

We can define GrG⁢Ln≤m to be those M which satisfy the relation for some fixed m. Using this, we may view M as an A⁡[t]/t2⁢m-module satisfying some conditions. Therefore, GrG⁢Ln≤m is a closed subfunctor of Gr⁡(2⁢m⁢n) (which means things inside a 2⁢m⁢n-dimensional vector space).

Remark 8.1.5.

Unfortunately, the affine Grassmannian is not a scheme, but at least it is an inductive limit of schemes.

Now consider g∈G⁡((t))⁢(A). This induces an automorphism of the trivial A-family of G-bundles on D∘, and so there is a natural action

Remark 8.1.6.

We have a simply transitive action on the fiber by G⁡[[t]], and thus

Example 8.1.7.

Consider the case of G⁢L1=𝔾m. Then the only data that matters is the power of zn near 0, and so GrG⁢L1⁢(ℂ)≅ℤ.

Now we can give a lattice description. Remember that we are considering t-stabble subspaces of t−m⁢ℂ⁢[[t]]/tm⁢ℂ⁢[[t]]. This has a basis (t−m,t1−m,…,tm−1). But now GrG⁢L1≤m⁢(ℂ) is a union of 2⁢m points, and so in the increasing limit we obtain ℤ.

8.2 Coulomb branches

Definition 8.2.1.

Fix a G-representation N. We will define the space

where (P,γ) is as before and s is a section of Ptriv×GN over D∘.

This has a subset RNG⊂TNG, where we require s to extend to D and γ−1⁢(s) to extend to a section of P. We will call RNG the BFN space, where BFN is Braverman, Finkelberg, and Nakajima.

Definition 8.2.2.

We define the Coulomb branch algebra to be

Example 8.2.3.

Consider G=ℂ× and N=0. Then RNG=GrG, and then we have Heq⁢(ℤ)=ℂ⁡[w,x±].

Example 8.2.4.

Consider G=ℂ× and N the adjoint representation. If the trivialization is simply zn=γ, then we need γ⁢s to have no singularities, and therefore s∈zmax⁡(0,n)⁢ℂ⁢[z]. Therefore

and therefore we have

If we modify the second ℂ× to act with weight m, we instead obtain ℂ⁡[w,x,y]/(w|m|=x⁢y).

Remark 8.2.5.

For T⊆G, we have (ATℏ)W→A~Gℏ, and after localization, this is an isomorphism.

8.3 Hilbert schemes

Definition 8.3.1.

Fix v∈N⁡((t)), where N is a G-representation. Then the generalized affine Springer fiber is defined by

The main claim is the following: Let Ĉ be a germ of plane curve singularity. Then there is an Ad⊕V-generalized affine Springer fiber MvAd⊕V such that

By the Weierstrass preparation theorem, we assume that f⁡(x,t)=xn−an−1⁢xn−1−⋯−a0. Then as ℂ⁡[[t]]-modules, we can identify ℂ⁡[[x,t]]/f=⟨1,x,…,xn−1⟩⁢ℂ⁢[[t]]. Then if R=ℂ⁡[[x,t]]/f, we can identify the fraction field of R with ℂ⁢((t))⊕n,∗.

Unwinding definitions, points in Hilb0⁢(Ĉ) are the same as x-stable, finitely generated, projective ℂ⁡[[t]]-submodules of ℂ⁢((t))⊕n such that there exists m such that

Then we have

We now choose v=(γ,en), where en corresponds to xn−1. We also need ad⁡(g)⁢γ∈Ad⁡[[t]]. We will also choose g such that for some Λ⊂ℂ⁢[[t]]⊕n,∗, we have ℂ⁢[[t]]n,∗⁢g−1=Λ. This implies that

and so Λ⁢γ⊂Λ. The second condition gives us g−1⁢ei∈ℂ⁢[[t]]n. By definition of the affine Springer fiber, we only need g−1⁢en∈ℂ⁢[[t]]n. This implies that γ⁢ek+1=ek+ak⁢en and so we can induct downwards.

Example 8.3.2.

Let f=x−t. Then x-stability is the same as t-stability, and so we are looking for something t-stable in

which is the same as m points. Thus Hilb⁡ℂ⁡[[x,t]]/t≃⨆ℕpt.

Remark 8.3.3.

When f=xn−tk, there is another ℂ× scaling x and t (called the flavor symmetry). Then we have an action of G⁢Ln⁢[[t]]~⋉ℂ×, and so (RNG)=A~ℏ. If we denote the group by GF~, then we define LV to be the stabilizer of v, and this is isomorphic to ℂ×. By some general theory, there is an action of A~ℏ on HLV⁢(MNV). We claim that for N=Ad⊕V, A~ℏ,Ad⊕V is the spherical rational DAHA.