ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES

AARON D. LAUDA AND HEATHER M. RUSSELL

arXiv:1203.0797v1 [math.RT] 5 Mar 2012

A BSTRACT. We identify the ring of odd symmetric functions introduced by Ellis and Khovanov as the
space of skew polynomials fixed by a natural action of the Hecke algebra at q = −1. This allows us
to define graded modules over the Hecke algebra at q = −1 that are ‘odd’ analogs of the cohomology
of type A Springer varieties. The graded module associated to the full flag variety corresponds to the
quotient of the skew polynomial ring by the left ideal of nonconstant odd symmetric functions. The top
degree component of the odd cohomology of Springer varieties is identified with the corresponding Specht
module of the Hecke algebra at q = −1.

1. I NTRODUCTION
Cn ,
The full flag variety X consists of the set of all flags in
X = {0 = U0 ⊂ U1 ⊂ · · · ⊂ Un = Cn | dimC Ui = i} .
The integral cohomology ring H(X) := H ∗ (X, Z) can be explicitly described as a quotient of the
graded polynomial ring Poln := Z[x1 , . . . , xn ] with deg(xj ) = 2. Write εr = εr (x1 , . . . , xn ) for the rth
elementary symmetric function in n variables. Denote by Λn := Z[x1 , . . . , xn ]S = Z[ε1 , ε2 , . . . , εn ] the
n

ring of symmetric functions, and let Λ+

n be the two-sided ideal of Poln generated by the elementary
symmetric functions with no constant term. The cohomology ring of X is the quotient
(1.1) H(X) = Z
∼ [x1 , . . . , xn ]/Λ+ .
n
Let λ = λ1 ≥ · · · ≥ λn ≥ 0 be a partition of n, and let λ′ = λ′1 ≥ · · · ≥ λ′n ≥ 0 be its transpose
partition. Consider a nilpotent matrix xλ of Jordan type λ. The Springer variety X λ is the closed
subvariety of X consisting of those flags preserved by xλ

X λ = (U0 , U1 , . . . , Un ) ∈ X | xλ Ui ⊆ Ui−1 .
Work of De Concini and Procesi [9] and Tanisaki [39] generalizes the above characterization of H(X)
Z
showing that the integral cohomology ring of the Springer variety H(X λ ) := H ∗ (X λ , ) is isomorphic
to the quotient of the polynomial ring Poln by the two-sided ideal Iλ generated by the following
elements:
 
S k ≥ r > k − δk (λ),
(1.2) Cλ := εr ,
S ⊆ {x1 , . . . , xn }, |S| = k
where εSr denotes the rth elementary symmetric function in the variables of S and
δk (λ) = λ′n + · · · + λ′n−k+1 .
For more details on the cohomology of Springer varieties see [5, 16, 18, 36], and for a generalization of
this ideal see [31].
The cohomology rings of Springer varieties carry additional structure summarized below.
(1) There is an action of the symmetric group Sn on H(X λ ) that preserves the cohomological
grading.

Date: March 5, 2012.

1
2 AARON D. LAUDA AND HEATHER M. RUSSELL

(2) The pull-back homomorphism i∗ : H(X) → H(X λ ) arising from the inclusion X λ → X is a
Z
surjective Sn -module map.
Q
(3) The top nonzero degree of rational cohomology H top (X λ , ) is isomorphic to the irreducible
Q Sn -module corresponding to the partition λ.
Q Q
(4) The full rational cohomology H ∗ (X λ , ) is isomorphic as a Sn -module to the permutation
module M λ .
From the polynomial presentation of the cohomology ring of the full flag variety in (1.1) the action
of the symmetric group is quite transparent (Springer’s original action is more sophisticated [36]). In
particular, w ∈ Sn acts on f ∈ Poln by permuting indices so that
w · f (x1 , x2 , . . . , xn ) = f (xw(1) , xw(2) , . . . , xw(n) ).
Since this action of Sn preserves the ideal of symmetric functions Λ+n , it descends to an action on the
quotient H(X). The ideal Iλ is preserved by the action of Sn , so there is also a natural Sn -action on
the quotient H(X λ ) = Poln /Iλ . The rings H(X λ ) inherit the grading from Poln . Since the ideal Iλ is
homogeneous, the action of Sn restricts to an action on each graded piece of H(X λ ).

Recent work of Ellis and Khovanov introduces a new “odd” analog of the Hopf algebra of sym-
metric functions [13]. The ring of odd symmetric functions, denoted by OΛn , is noncommutative
despite exhibiting behaviour surprisingly similar to the ordinary ring of symmetric functions; it has
odd analogs of elementary, complete, and Schur function bases. The ring OΛn is a certain quotient
of a related version of noncommutative symmetric functions defined by Thibon and Ung [41]. The
precise form of this quotient is essential to the results in this article.
The odd symmetric function and ordinary symmetric function rings have the same graded rank
and become isomorphic when coefficients are reduced modulo two. These quintessential rank and
reduction features characterize the “oddification” of an algebraic object. Recent work of the first
author with Ellis and Khovanov gives an explicit realization of odd symmetric functions in the ring
of skew polynomials
(1.3) Z
OPoln = hx1 , . . . , xn i/hxi xj + xj xi = 0 for i 6= ji,
and uses this to connect odd symmetric functions to a new odd analog of the nilHecke algebra [14]
inspired by constructions in Heegaard-Floer homology [28]. Kang, Kashiwara, and Tsuchioka [20]
independently studied the odd nilHecke algebra in their work on categorification of Hecke-Clifford
super algebras, and it is closely related to earlier work of Wang [40]. This algebra also appears in
recent work of Hill and Wang on categorifications of quantum Kac-Moody superalgebras [17].
The parallels between the odd and even nilHecke algebras extend further. Just as cyclotomic quo-
Z
tients of the nilHecke algebra are Morita equivalent to the integral cohomology rings H ∗ (Gr(k, N ), )
C
of Grassmannians of k-planes in N [27, Section 5], the corresponding cyclotomic quotients of the
odd nilHecke algebra are Morita equivalent to noncommutative rings OH(Gr(k, N )) [14]. The graded
ranks of these rings are the same, they become isomorphic when coefficients are reduced modulo
two, and OH(Gr(k, N )) has a basis consisting of certain odd Schur polynomials [14]. In other words,
OH(Gr(k, N )) is an oddification of the cohomology of the Grassmannian.
The nilHecke algebra, cohomology rings of Grassmannians, and the ring of symmetric functions
are vital components in the theory of categorified quantum sl2 [2, 15, 8, 26, 25, 23], which under-
lies Khovanov homology. All of the odd structures discussed above aim to establish a similar alge-
braic framework underlying the odd Khovanov homology theory of Ozsváth, Rasmussen, Szabó [33].
These invariants, which both categorify the Jones polynomial, are distinct (see [35, Section 3.1]) but
agree modulo two.
Another algebraic structure closely connected to Khovanov homology is the Springer variety X (n,n) .
In particular, the cohomology of X (n,n) is isomorphic to the center of Khovanov’s arc algebra Hn used
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 3

in one construction of Khovanov homology for tangles [22]. There is a generalization of Khovanov’s
arc algebra [37, 7] that Stroppel and Webster showed is isomorphic to the convolution algebra formed
on cohomologies of components of the Springer variety [38]. These Springer varietes also arise in
Cautis-Kamnitzer’s geometric construction of Khovanov homology using derived categories coher-
ent sheaves [6]. These connections lead one to ask whether the cohomology of Springer varieties also
have natural “odd” analogs. In this article we introduce graded quotient modules over the skew
polynomial ring which are oddifications of the cohomology of Springer varieties X λ for any partition.
In the odd setting there is an action of the symmetric group Sn on OPoln . However, odd symmetric
functions are not preserved by this action. This raises the question of what object plays the role of the
symmetric group Sn in the odd theory of Springer varieties. Here we show that the Hecke algebra
Hq (n) at q = −1 is the natural analog.
We begin by defining an action of the Hecke algebra Hq (n) at q = −1 on the space of skew polyno-
mials OPoln . A related action of Hq (n) on Poln was defined in [1] where Ti acts by the q-commutator
(1.4) [∂i , xi ]q = ∂i xi − qxi ∂i ,
and ∂i is the divided difference operator (see section 2.1.3). When q = 1 the q-commutator action is
Z
just the usual permutation action of H1 (n) = [Sn ] on Poln . In [1, Claim 3.1] it is shown that the ring
of symmetric functions Λn can also be understood as the space of invariants in Poln under this action
of Hq (n). The odd nilHecke algebra acts on skew polynomials by so-called odd divided difference
operators. Using these operators we define an action of Hq (n) at q = −1 by a similar q-commutator
formula and prove the following theorem.
Theorem. Odd symmetric functions OΛn ⊂ OPoln are precisely the skew-polynomials in OPoln that
are invariant under the action of H−1 (n). In other words, Ti (f ) = f for all generators Ti in H−1 (n) if
and only if f is odd symmetric.
Just as commutativity is lost in the passage from cohomology rings of Grassmanians to odd coho-
mology of Grassmanians, the odd cohomology of Springer varieties no longer possess a commutative
ring structure or even a ring structure at all. This occurs because the left ideal OI L := OPoln · OΛ+
generated by the elements in OΛ+ differs from the right ideal OI R := OΛ+ · OPoln . This fact was
first observed by Alexander Ellis. Thus, the definition of the odd cohomology of the full flag variety
requires a choice of the left or right ideal. Quotienting by the two-sided ideal leads to torsion in the
quotient ring, and the resulting ring will not have the same dimension as the ordinary cohomology
ring H(X) when coefficients are reduced modulo two.
The action of H−1 (n) on OPoln is homogeneous with the respect to the grading. Since the left ideal
OI L is preserved by the action of H−1 (n) and is generated by homogenous elements, the quotient
module OPoln /OI L is a left graded H−1 (n)-module. We denote this quotient by OH(X) since it will
play the role of the cohomology of the full flag variety in the odd theory.
Taking the analogy between the even and odd theories one step further, we construct odd analogs
OIλ of the ideals Iλ from (1.2) and show that these are also invariant under the Hecke algebra H−1 (n)
action. This allows us to define graded left H−1 (n)-modules
OH(X λ ) := OPoln /OIλ .
The ring OPoln is graded with deg(xi ) = 2, and the ideal OIλ is homogeneous with respect to
this grading, so that the action of H−1 (n) restricts to an action on each homogeneous component
of OH(X λ ). In this article we prove the following results.
Theorem.
(1) There is an isomorphism of graded vector spaces:
OH(X) ⊗Z 2 = Z ∼ H(X) ⊗Z 2 , Z and OH(X λ ) ⊗Z Z2 ∼= H(X λ) ⊗Z Z2 .
4 AARON D. LAUDA AND HEATHER M. RUSSELL

Z
(2) The underlying abelian group of OH(X λ ) is a free graded -module that admits an integral
basis identical in form to the Garsia-Procesi basis [16] of H(X λ ). Using this basis we show
that the graded ranks agree
rkq OH(X λ ) = rkq H(X λ ).
(3) There is a surjective homomorphism OH(X) → OH(X λ ) of graded H−1 (n)-modules.
F
(4) Working over a field , the top graded component OH top (X λ ) of OH(X λ ) is isomorphic as
λ
an H−1 (n)-module to the Specht module S−1 .
Though we utilize many of the combinatorial techniques introduced by Tanisaki [39] and by Garsia
and Procesi [16], we emphasize that the identification of OH top (X λ ) with the Specht module requires
new methods. This phenomenon is genuinely “odd” in that arguments make extensive use of cal-
culations in the odd nilHecke algebra. The original identification of the Specht module with the top
degree cohomology of the Springer variety X λ uses results from geometry. Since the precise connec-
tion between the odd theory and geometry has yet to be developed, we must prove directly that the
top cohomology of odd Springer varieties is isomorphic to the Specht module of the Hecke algebra
H−1 (n). This requires an explicit verification of row relations and Garnir relations (Propositions 4.3
and 4.6).

Acknowledgments: This project arose out of a general program started by the first author with
Alexander Ellis and Mikhail Khovanov to identify odd analogs of representation theoretic objects.
The idea of looking for odd analogs of cohomology of Springer fibers arose from these conversations.
Both authors are grateful to Monica Vazirani and Alexander Kleshchev for helpful discussions on the
representation theory of Hecke algebras, and to Sabin Cautis and Anthony Licata for helpful discus-
sions on Springer varieties. The first author was partially supported by the NSF grant DMS-0855713,
and the Alfred P. Sloan foundation. We would also like to acknowledge the use of the NC-Algebra
Mathematica package for calculations related to this project.

2. O DD SYMMETRIC FUNCTIONS AS INVARIANT FUNCTIONS

2.1. Reminders from the even theory.

2.1.1. The nilHecke ring (Even case). The nilHecke ring N Hn is the graded unital associative ring gener-
ated by elements x1 , . . . , xn of degree 2 and elements ∂1 , . . . , ∂n−1 of degree −2, subject to the relations
∂i2 = 0, ∂i ∂i+1 ∂i = ∂i+1 ∂i ∂i+1
xi ∂i + ∂i xi+1 = 1, ∂i xi + xi+1 ∂i = 1
(2.1)
xi xj + xj xi = 0 (i 6= j), ∂i ∂j + ∂j ∂i = 0 (|i − j| > 1),
xi ∂j + ∂j xi = 0 (|i − j| > 1).
Z
The nilHecke algebra N Hn acts as endomorphisms of the polynomial ring Poln := [x1 , x2 , . . . , xn ]
with xi acting by multiplication and ∂i acting by divided difference operators. That is for f ∈ Poln
f − si (f )
(2.2) ∂i (f ) := ,
xi − xi+1
where si (f ) denotes the standard action of the symmetric group Sn on the polynomial ring Poln by
exchanging the variables xi and xi+1 . Observe that the action of ∂i on any polynomial f can be
deduced from the rules

1
 if j = i
(2.3) ∂i (1) = 0, ∂i (xj ) = −1 if j 6= i + 1


0 otherwise,
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 5

and the Leibniz rule

(2.4) ∂i (f g) = ∂i (f )g + si (f )∂i (g) for all f, g ∈ Poln .
For any w ∈ Sn choose a reduced word decomposition w = si1 . . . sim into a product of elementary
transpositions. Define ∂w = ∂i−1 . . . ∂im . It is clear from the relations (2.1) that ∂w is independent of
the choice of reduced word decomposition for w.
2.1.2. Symmetric functions (Even case). Throughout the paper we refer to symmetric functions and
symmetric polynomials interchangeably. The ring of symmetric polynomials can be understood in
several ways:
Z
(1) Λn is the space of invariants for the standard action of Sn on Poln , i.e. Λn = [x1 , . . . , xn ]Sn .
(2) Λn is the intersection of the kernels or images of divided difference operators:
n−1
\ n−1
\
Λn = ker(∂i ) = im (∂i ).
i=1 i=1

The ring of symmetric functions Λn can be given a concrete presentation using the elementary
symmetric functions in Poln . Below we label these functions ‘even’ to contrast with the odd theory
being studied in this paper. Recall that the ring of symmetric polynomials is isomorphic to a graded
polynomial ring,
(2.5) Λn ∼ Z
= [εeven , εeven , . . . , εeven ],
1 2 n

where deg(εeven
k ) = 2k and
X
(2.6) εeven
k = εeven
k (x1 , . . . , xn ) = xi1 · · · xin .
1≤i1 <···<ik ≤n

The polynomial ring Poln is a free Λn module of rank n! with basis over Λn consisting of Schubert
polynomials seven
w ∈ Poln . These polynomials are defined for w ∈ Sn as
(2.7) sw (x1 , . . . , xn )even = ∂w−1 w0 (xδn ),
where w0 is the longest element of Sn and xδn := x1n−1 x2n−2 . . . x0n . The degree of sw is 2ℓ(w).

2.1.3. Hecke algebra action on Poln . The ring of symmetric functions also arises from a more general
action of the Hecke algebra Hq (n) of type A. The Hecke algebra Hq (n) is defined over and commuta-
tive domain R with 1 where q is any element of R. Here we work integrally whenever possible taking
Z
R = . In section 4 we take R = F for a field F.
The 0-Hecke algebra H0 (n) is the unital associative R-algebra with generators ∂ 1 , . . . , ∂ n−1 satisfy-
ing the relations
2
∂ i = ∂ i,
(2.8) ∂i∂j = ∂j ∂i (i 6= j),
∂ i+1 ∂ i ∂ i+1 = ∂ i+1 ∂ i ∂ i+1 .
The 0-Hecke algebra acts on Poln by the so-called isobaric divided difference operators ∂ i (f ) =
∂i xi (f ). The 0-Hecke algebra is the q = 0 specialization of the Hecke algebra associated to Sn . The
Hecke algebra Hq (n) is the R-algebra with generators Ti for 1 ≤ i ≤ n − 1 and relations
Ti2 = (1 − q)Ti + q,
(2.9) Ti Tj = Tj Ti (|i − j| > 1),
Ti+1 Ti Ti+1 = Ti+1 Ti Ti+1 .
6 AARON D. LAUDA AND HEATHER M. RUSSELL

Here we follow [1] and use a nonstandard presentation of the Hecke algebra so that our specialization
at q = 0 agrees with the form of the 0-Hecke algebra defined above. The usual conventions replace Ti
with −Ti .
In [1, Claim 3.1] it is shown that the Hecke algebra of type A at generic q acts on the polynomial ring
Poln with Ti acting by the q-commutator [∂i , xi ]q = ∂i xi − qxi ∂i . When q = 1 the q-commutator is just
the standard action of the symmetric group by permuting variables. When q = 0 the q-commutator
is the isobaric divided difference operator ∂ i of the 0-Hecke algebra. It is also shown in [1, Claim
3.2] that the ideal generated by the ring of symmetric functions Λn is invariant under this action.
This observation will be the starting point for our study of odd analogs of the cohomology rings of
Springer varieties.
2.2. The odd Theory. In this section we review the theory of the odd nilHecke ring and odd symmet-
ric functions from [13, 14], see also [20]
2.2.1. Odd NilHecke. Define the odd nilHecke ring ON Hn to be the graded unital associative ring gen-
erated by elements x1 , . . . , xn of degree 2 and elements ∂1 , . . . , ∂n−1 of degree −2, subject to the rela-
tions
∂i2 = 0, ∂i ∂i+1 ∂i = ∂i+1 ∂i ∂i+1 ,
xi ∂i + ∂i xi+1 = 1, ∂i xi + xi+1 ∂i = 1,
(2.10)
xi xj + xj xi = 0 (i 6= j), ∂i ∂j + ∂j ∂i = 0 (|i − j| > 1),
xi ∂j + ∂j xi = 0 (|i − j| > 1).
Define the ring of skew polynomials to be the free unital associative algebra on skew-commuting
variables x1 , . . . , xn ,
(2.11) Z
OPoln = hx1 , . . . , xn i/hxi xj + xj xi = 0 for i 6= ji.
The symmetric group Sn acts on the degree k part of OPoln as the tensor product of the permutation
representation and the k-th tensor power of the sign representation. That is, for 1 ≤ i ≤ n, the
transposition si ∈ Sn acts as the ring endomorphism


−xi+1 if j = i
(2.12) si (xj ) = −xi if j = i + 1


−xj otherwise.
The odd divided difference operators are the linear operators ∂i (1 ≤ i ≤ n − 1) on Zhx1 , . . . , xn i
defined by
(
1 if j = i, i + 1
(2.13) ∂i (1) = 0, ∂i (xj ) =
0 otherwise,
and the Leibniz rule
(2.14) ∂i (f g) = ∂i (f )g + si (f )∂i (g) Z
for all f, g ∈ hx1 , . . . , xn i.
Note that there is no analog of Equation 2.2 in the odd setting. It is easy to check from the definition
of ∂i that for all i ∂i (xj xk + xk xj ) = 0 for j 6= k, so ∂i descends to an operator on OPoln . Considering
∂i and (multiplication by) xj as operators on OPoln defines an action of ON Hn on OPoln , see [14].
For w = si1 . . . sim in Sn define ∂w = ∂i1 . . . ∂im . Unlike the even case, in the odd nilHecke algebra
the element ∂w depends on the choice of reduced expression for w up to a sign. For the longest element
w0 of Sn we fix a particular choice of reduced expression,
∂w0 = ∂1 (∂2 ∂1 ) · · · (∂n−1 · · · ∂1 ).
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 7

For each w ∈ Sn define the corresponding odd Schubert polynomial sw ∈ OPoln by

(2.15) sw (x1 , . . . , xn ) = ∂w−1 w0 (xδn ).
Again, the degree of sw is 2ℓ(w). For w, w′ ∈ Sn , the formula
(
±∂ww′ if ℓ(ww′ ) = ℓ(w) + ℓ(w′ )
(2.16) ∂w ∂w′ =
0 otherwise,
implies the following action of odd divided difference operators on odd Schubert polynomials
(
±swu−1 if ℓ(wu−1 ) = ℓ(w) − ℓ(u)
(2.17) ∂u sw =
0 otherwise.
Note that the exact signs in these relations can be determined for specific choices of reduced words
for the symmetric group elements involved, but we will not need these coefficients in what follows.

2.2.2. Odd symmetric polynomials. Define the ring of odd symmetric polynomials to be the subring
n−1
\ n−1
\
(2.18) OΛn = ker(∂i ) = im (∂i )
i=1 i=1

of OPoln . The kth odd elementary symmetric polynomial is defined as

X
(2.19) εk (x1 , . . . , xn ) = ei1 · · · x
x eik , ei = (−1)i−1 xi ,
where x
1≤i1 <···<ik ≤n

It was shown in [14] that the following relations hold in the ring OΛn :
εi ε2m−i = ε2m−i εi (1 ≤ i, 2m − i ≤ n)
(2.20) εi ε2m+1−i + (−1) ε2m+1−i εi = (−1)i εi+1 ε2m−i + ε2m−i εi+1
i
(1 ≤ i, 2m − i ≤ n − 1)
ε1 ε2m + ε2m ε1 = 2ε2m+1 (1 < 2m ≤ n − 1).
Note that the third is the i = 0 case of the second. In particular, the ring OΛn of odd symmetric func-
tions is noncommutative. These relations lead to an isomorphism of graded rings OΛn ∼ Z
= [ε1 , . . . , εn ].
See [41] and the references therein for a related version of noncommutative symmetric functions.
Odd symmetric functions also have natural bases corresponding to complete, monomial, and for-
gotten symmetric functions, as well as a Schur polynomials basis [13, 14]. There is also an odd analog
of the Littlewood-Richardson rule for Schur polynomials developed by Ellis [12]. Recall that the odd
Schubert polynomials defined above are a homogeneous basis for OPoln as a free left and right OΛn -
module of rank n! [14, Proposition 2.13].
Remark 2.1. It is not difficult to construct examples showing that odd symmetric functions are not the
Sn invariant functions for the action of Sn defined above. For example, s1 (e
x1 +e
x2 +e
x3 ) = (e
x1 +e
x2 −e
x3 ).
2.3. Modular reduction. For a graded abelian group V of finite rank in each degree define the graded
rank of V as
X
(2.21) rkq (V ) = rk (Vi )q i .
i∈ Z
F
Likewise, for a graded vector space W over a field of finite dimension in each degree, define the
graded dimension of W as
X
(2.22) dimq,F (W ) = dimF (Wi )q i .
i∈ Z
8 AARON D. LAUDA AND HEATHER M. RUSSELL

In what follows we make use of results from [14, Section 2] about various reductions mod 2. Con-
Z
sider the reduction map Λn → Λn ⊗Z /2. From the definition of elementary symmetric functions it is
Z
clear that their images under this map are nonzero in Λn ⊗Z /2. In particular, rkq (Λn ) = dimq,Z/2 (Λn ).
Similarly rkq (Poln ) = dimq,Z/2 (Poln ). The images of Schubert polynomials seven under the reduction
Z Z Z
w
map Poln → Poln ⊗Z /2 are nonzero and give a basis for Poln ⊗Z /2 as a free Λn ⊗Z /2-module.
A similar story holds in the odd setting where
rkq (OΛn ) = dimq,Z/2 (OΛn ) and rkq (OPoln ) = dimq,Z/2 (OPoln ).
Z
Products of odd elementary symmetric functions provide a /2-basis for OΛn ⊗Z /2, and odd Schu-Z
Z Z
bert polynomials provide a /2-basis for OPoln ⊗Z /2 as a free left and right OΛn ⊗Z /2-module. Z
Since the definitions of odd divided difference operators, odd elementary symmetric functions, and
odd polynomials agree with their even counterparts, when reduced modulo 2, we have isomorphisms
Z
OPoln ⊗Z ( /2) ∼ Z
= Poln ⊗Z ( /2) and OΛn ⊗Z ( /2) ∼ Z
= Λn ⊗Z ( /2). Z
2.4. Hecke invariants of OPoln . The 0-Hecke algebra acts on OPoln by odd isobaric divided differ-
ence operators ∂ i = ∂i xi , see [14, Section 3.2]. Furthermore, just as the Hecke algebra Hq (n) acts on
Z
the space of polynomials [x1 , x2 , . . . , xn ] we have the corresponding odd result.
Proposition 2.2. The Hecke algebra Hq (n) acts on OPoln with Ti acting by the q-commutator
(2.23) Ai := [∂i , xi ]q := ∂i xi − qxi ∂i
if and only if q = −1 or q = 0.
Proof. It is straightforward to check that the operators satisfy the first two relations of (2.9) for any q.
Computing the last relation gives
Ai+1 Ai Ai+1 − Ai Ai+1 Ai = (2q + 2q 2 )x1 x2 x3 ∂1 ∂2 ∂1 ,
which is only zero when q = −1 or q = 0. 
Note that, using the relations of the nilHecke algebra, Ai can also be written as
(2.24) Ai = 1 − (qxi + xi+1 )∂i .
In what follows we will abuse notation and write Ti for the action of the operators Ai at q = −1.
Observe that f ∈ OΛn is fixed by the action of the 0-Hecke algebra H0 (n) on OPoln by odd isobaric
divided difference operators since ∂ i (f ) = ∂i xi (f ) = f + xi ∂i (f ) = f .
Proposition 2.3. An odd polynomial f ∈ OPoln is fixed under the action of H−1 (n) defined in Propo-
sition 2.2 if an only if f is an odd symmetric function.
Proof. For any odd symmetric function f ∈ OΛn we have
(2.25) Ti (f ) := ∂ i (f ) + xi ∂i (f ) = ∂ i (f ) = f,
for all 1 ≤ i ≤ n − 1. To prove the converse take an arbitrary g ∈ OPoln and express it in the basis of
odd Schubert polynomials for the free left OΛn -module OPoln ,
X
(2.26) g= f w sw ,
w∈Sn

where fw ∈ OΛn . Then

X X (2.14) X
Ti (g) = ∂i xi (fw sw ) + xi ∂i (fw sw ) g + (xi − xi+1 ) si (fw )∂i (sw ),
w∈Sn w∈Sn w∈Sn

where we used that ∂i annihilates odd symmetric functions.

 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 9

We have shown that g is invariant under the action of H−1 (n) if and only if
X
(2.27) si (fw )∂i (sw ) = 0
w∈Sn

for all 1 ≤ i ≤ n − 1. Recall from (2.17) that ∂i (sw ) is either zero or the Schubert polynomial swsi if
ℓ(wsi ) = ℓ(w) − 1. Therefore g is invariant if and only if
X
(2.28) si (fw )swsi = 0,
w∈Sn
ℓ(wsi )=ℓ(w)−1

for all 1 ≤ i ≤ n − 1. As we vary over all values of i the only w ∈ Sn not appearing in (2.28) for some
i is the identity element e.
Since the action of Sn does not preserve the ring of odd symmetric functions, we cannot assume
that si (fw ) ∈ OΛn . However, we can still apply the standard reduction mod 2 argument from [14]
to deduce that the fw = 0 for w 6= e. Specifically, it is not hard to check that si (fw ) ∈ OΛn ⊗Z /2. Z
Therefore, dividing (2.28) by an appropriate power of 2 gives a relation between the reductions mod
two of odd Schubert polynomials contradicting that these polynomials form a basis for OPoln ⊗Z /2 Z
Z
as a free left OΛn ⊗Z /2-module. Therefore, g is invariant under the action of H−1 (n) if and only if
fw = 0 for all w 6= e. Thus, g = fe · se = fe and is therefore odd symmetric. 
The above result identifies OΛn with the ‘coinvariant algebra’ for H−1 (n) under the action on
OPoln . This result can also be interpreted as saying that H−1 (n) acts trivially on odd symmetric
functions. For the usual presentation of the Hecke algebra the trivial action of Ti in Hq (n) is an action
by q. Recall that we are using the nonstandard version of the Hecke algebra so that the trivial action
corresponds to Ti acting by −q, which for H−1 (n) amounts to Ti acting by 1.
Remark 2.4. The proof of Proposition 2.3 shows that for arbitrary q the odd symmetric functions are
also the invariants under the operators Ai .

3. O DD COHOMOLOGY OF THE S PRINGER VARIETY

3.1. The odd Tanisaki ideal. Let S be an ordered subset of {x1 , x2 , . . . , xn } and write |S| = k for the
size of S. If xi is in the ath position of S write S(i) = a. Let

S (−1)S(i)−1 xi for i ∈ S
(3.1) xi :=
0 for i ∈
/S
Note that when S = {x1 , x2 , . . . , xn } then xSi = x
ei = (−1)i−1 xi .
S
Define the odd partial symmetric function εr corresponding to the ordered subset S as follows:
X
(3.2) εSr = xSi1 . . . xSir .
1≤i1 <···<ir ≤n

Note that εSr = 0 when r > |S|. Also observe that given S ⊂ {x1 , . . . , xn−1 } with |S| = k
(3.3) εSr = εS∪{x
r
n}
+ (−1)k−1 εSr−1 xn .
Since OIλ is a left ideal, it convenient to write this as
(3.4) εSr = εS∪{x
r
n}
+ (−1)k+r xn εSr−1 .
Recall from (2.24) that H−1 (n) acts on OPoln with
Ti = ∂i xi + xi ∂i = 1 + (xi − xi+1 )∂i .
Below we compute the action of the operators Bi := (xi − xi+1 )∂i for 1 ≤ i ≤ n on odd partial
symmetric functions.
10 AARON D. LAUDA AND HEATHER M. RUSSELL

Lemma 3.1. The operators Bi := (xi − xi+1 )∂i for 1 ≤ i ≤ n act on odd partial symmetric functions
by the formulas

 0 if i, i + 1 ∈
/ S, or i ∈ S and i + 1 ∈ S
′
(3.5) Bi (εSr ) = εSr − εSr if i ∈ S and i + 1 ∈ /S
 S ′′ S
εr − εr if i ∈ / S and i + 1 ∈ S
where S ′ is the ordered subset obtained from S by replacing i with i + 1 and S ′′ is the subset obtained
from S by replacing i + 1 with i.
Proof. It is clear that ∂i (εSr ) is zero unless i ∈ S or i + 1 ∈ S. Suppose that both i, i + 1 ∈ S. Then
εSr is a sum of monomials of the form xSi1 . . . xSir . The odd divided difference operator annihilates any
monomial in which both i and i + 1 occur since ∂i (xi xi+1 ) = 0. Likewise, ∂i annihilates monomials
containing neither i nor i + 1. Hence, it is enough to consider monomials with exactly one element of
{i, i + 1}. Consider monomials that are identical except for i and i + 1
 
∂i xSi1 . . . xSis xSi xSis+2 . . . xSir + xSi1 . . . xSis xSi+1 xSis+2 . . . xSir = (−1)s xSi1 . . . xSis ∂i (xSi + xSi+1 )xSis+2 . . . xSir

it follows that Bi (εSr ) = 0 since xSi + xSi+1 = (−1)S(i)−1 (xi − xi+1 ).

Consider the case when i ∈ S and i + 1 ∈ / S. Then
(3.6)  
X X X
∂i  xSi1 . . . xSir  = ∂i (xSi1 . . . xSir ) = (−1)s xSi1 . . . xSis ∂i (xSi )xSis+2 . . . xSir
1≤i1 <···<ir ≤n 1≤i1 <···<ir ≤n

where the last summation is over all 1 ≤ i1 < · · · < ir ≤ n with some is+1 = i for 0 ≤ s ≤ r − 1.
Hence, the action of Bi on eSr is given by
X X
(xi − xi+1 ) (−1)s xSi1 . . . xSis (−1)S(i)−1 xSis+2 . . . xSir = xSi1 . . . xSis (−1)S(i)−1 (xi − xi+1 )xis+2 . . . xSir .

′
The claim then follows since xSj = xSj for j 6= i, i + 1 and i + 1 is in the same position in S ′ as i is in
′ ′
S, so that (−1)S(i)−1 xi+1 = (−1)S (i+1)−1 xi+1 = xSi+1 . A similar argument proves the case when i ∈ /S
and i + 1 ∈ S. 
Definition 3.2. Let λ be a partition of n. Consider the set of elements
 
k ≥ r > k − δk (λ),
(3.7) OCλ := εSr .
S ⊆ {x1 , . . . , xn }, |S| = k
Let OIλ := OPoln · OCλ denote the homogeneous (left) odd Tanisaki ideal of OPoln generated by the
elements of OCλ .
We use the left ideal generated by OCλ rather than the right ideal to be compatible with the left
action of H−1 (n) on OPoln .
Proposition 3.3. The odd Tanisaki ideal OIλ is preserved by the action of H−1 (n), and this action
preserves each homogeneous component of OIλ .
Proof. The second statement follows immediately from the definition of Ti in terms of homogeneous
elements of the odd nilHecke algebra. Lemma 3.1 and the Leibniz formula prove that OIλ is preserved
under the action of H−1 (n). 
We define the oddification of the cohomology of the Springer variety OH(X λ ) as the left OPoln graded
quotient module
(3.8) OH(X λ ) := OPoln /OIλ .
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 11

If p, q ∈ OPoln are equivalent modulo OIλ write p ∼

=λ q.
Corollary 3.4. The left action of H−1 (n) on OPoln restricts to a left action of H−1 (n) on OH(X λ ).

3.2. The size of OH(X λ ). Next we give a basis OB(λ) for OH(X λ ) as a free abelian group. The
monomials in this basis are the same as those for H(X λ ) first given by De Concini-Procesi [9] and
further studied by Garsia-Procesi [16] and others [5, 30, 3]. Our description most closely resembles
that of Mbirika [30, Section 2].
Given a partition λ = λ1 ≥ . . . ≥ λn ≥ 0 the height h(λ) of λ is the number of nonzero elements of
the partition. If λ is visualized as a top and left justified Young diagram, the height h(λ) is the number
of rows. For 1 ≤ a ≤ h(λ) and 1 ≤ b ≤ λa a node (a, b) of a Young diagram is the box in row a and
column b.
For 1 ≤ i ≤ h(λ), define λ(i) to be the partition of n − 1 obtained from λ by doing the following.
• Remove the rightmost box in the ith row of λ.
• If the resulting shape is top and left justified do nothing.
• If removing a box creates a broken column, shift all boxes below the gap directly upwards.
The resulting shape is now top and left justified.
Using the subpartitions λ(i) define the set OB(λ) for OH(X λ ) recursively by
h(λ)
G
OB(1) = {1} and OB(λ) = {bxi−1
n | b ∈ OB(λ(i) )}.
i=1

Example 3.5. Consider the partition λ = (2, 1, 1, 0). Garsia-Procesi make use of a tree to visualize the
full recursion used to build a basis. The tree corresponding to the basis OB(2, 1, 1) is constructed in
Figure 1 below.

PP
♠♠♠ ✵✵ PPPPP
♠ ♠♠♠♠ PPP x24
1 ♠♠ ✵✵ x4
♠ PPP
♠♠♠♠♠ ✵✵ PPP
♠ ♠♠ ✵ PPP
♠♠♠ PP
▼▼
☞ ✷✷ ▼▼▼▼ 2 ❄❄ ❄❄
1 ☞☞ ✷✷ x3 ▼▼▼x3 ❄❄x3 ❄❄x3
☞ ✷✷ ▼▼▼ 1 ❄❄ 1 ❄❄
☞ ✷ ▼▼▼
☞☞ ▼▼ ❄❄ ❄❄

✮ ✮ ✮ ✮✮ ✮
1 ✕✕
✕✕ ✮✮✮ x2 1 ✕✕
✕✕ ✮✮✮ x2 1 ✕✕
✕✕ ✮✮✮ x2 1 ✕✕
✕✕ ✮✮ x
1 ✕✕
✕✕ ✮✮✮ x2
✕ ✮✮ ✕ ✮✮ ✕ ✮✮ ✕ ✮✮ 2 1
✕ ✮✮ 1
✕✕ ✮ ✕✕ ✮ ✕✕ ✮ ✕✕ ✮ ✕✕ ✮

F IGURE 1. OB(2, 1, 1) = {1, x2 , x3 , x2 x3 , x23 , x2 x23 , x4 , x2 x4 , x3 x4 , x24 , x2 x24 , x3 x24 }

h(λ) h(λ) ∼
Lemma 3.6. The monomial xn is an element of the odd Tanisaki ideal OIλ , and thus xn =λ 0.
Proof. Observe that the conjugate partition λ′ = λ′1 ≥ · · · λ′n ≥ 0 for λ has the property that λ′1 = h(λ).
Thus, δn−1 (λ) = λ′n + · · · + λ′2 = δn (λ) − λ′1 = n − h(λ). It follows that h(λ) > (n − 1) − δn−1 (λ). Let
S = {x1 , . . . , xn−1 }. Then according to the definition of the odd Tanisaki ideal, εSh(λ) ∈ OIλ . Iteratively
12 AARON D. LAUDA AND HEATHER M. RUSSELL

applying Equation 3.4 gives

h(λ)−1
X i(i−1) S∪{x } h(λ)(h(λ)+1)
εSh(λ) = (−1)i(n−1+h(λ))+ 2 xin εh(λ)−i
n
+ (−1)h(λ)(n−1)+ 2 xnh(λ) .
i=0
S∪{xn } h(λ) ∼
Since εr ∈ OIλ for all 1 ≤ r ≤ n, it follows that xn =λ 0. 
Since OH(X λ ) is the quotient of OPoln by a homogenous left ideal, it has the structure of a graded
left OPoln -module. Let OH(X λ )i denote the subgroup of OH(X λ ) consisting of elements of the form
pxin for some p ∈ OPoln−1 . Lemma 3.6 implies that OH(X λ )h(λ) is represented by the zero coset in
OH(X λ ). Hence, there is an isomorphism of abelian groups
h(λ)
M
(3.9) OH(X λ ) ∼
= OH(X λ )i−1 /OH(X λ )i .
i=1

The following proof closely follows that of Garsia-Procesi [16, Proposition 2.1] (see also [39]) but
details are included for the reader’s convenience.
Proposition 3.7. The set OB(λ) spans OH(X λ ).
Proof. We prove this result via induction on the size of λ. For the base case, observe that OH(X 1 ) ∼
= Z.
Therefore OB(1) = {1} clearly spans OH(X 1 ).
Assume the result holds for all partitions of n − 1, and consider λ a partition of n. Given an element
of OH(X λ )i−1 /OH(X λ )i with representative of the form pxi−1 n where p ∈ OPoln−1 , we show that
 
X
pxi−1 ∼
=λ  αb bxi−1  + f xin
n n
b∈OB(λ(i) )
P
for some αb ∈ Z and f ∈ OPoln . By induction, assume that p ∼
=λ(i) b∈OB(λ(i) ) αb b and thus
X
pxi−1 ∼ αb bxi−1
n =λ(i) n .
b∈OB(λ(i) )

Therefore, it only remains to show that for any g ∈ OIλ(i) we have

gxi−1 ∼
=λ f xi n n

for some f ∈ OPoln . It is enough to show this for all of the odd partial elementary symmetric functions
εSr generating the odd Tanisaki ideal OIλ(i) .

Recall that for S ⊂ {x1 , . . . , xn−1 } with k = |S| we have r > k − δk λ(i) whenever εSr ∈ OIλ(i) .



Because λ(i) is subordinate to λ, it follows that δk λ(i) = δk+1 (λ) or δk λ(i) = δk+1 (λ) − 1 for all
1 ≤ k ≤ n −
1.
If δk λ(i) = δk+1 (λ) − 1, then
 
k − δk (λ(i) ) = k + 1 − δk (λ(i) ) + 1 = k + 1 − δk+1 (λ).
S∪{xn }
We conclude that whenever εSr ∈ OIλ(i) it follows that εr ∈ OIλ . By Equation 3.3,
εSr xi−1 = (εS∪{x n}
+ (−1)k−1 εSr−1 xn )xi−1 ∼
=λ εSr−1 xin .
n r n


Next assume that δk λ(i) = δk+1 (λ) and r > k + 1 − δk+1 (λ). Then
k + 1 − δk+1 (λ) = k + 1 − δk (λ(i) ) > k − δk (λ(i) ).
S∪{xn }
Since this implies εSr ∈ OIλ(i) and εr ∈ OIλ , we can argue as in the previous case.
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 13

S∪{xn }
Finally assume that δk λ(i) = δk+1 (λ) and r = k+1−δk+1 (λ). Then εSr ∈ OIλ(i) but er ∈
/ OIλ .


Observe that δk λ(i) = δk+1 (λ) = δk (λ) + λ′n−k−1 > δk (λ) + i. This is true since column n − k − 1 of
λ is to the right of the column that is altered in order to obtain λ(i) . Since λ(i) comes from removing
the rightmost box from the ith row, every column to the right of that location has at most i − 1 boxes.
Since r = k+1−δk+1 (λ), we have that r+i−1 = k+i−δk+1 (λ) > k−δk (λ) and εSr+i−1 ∈ OIλ . Notice
S∪{x } S∪{x }
also that εr+j n ∈ OIλ for all j ≥ 1. Rewriting Equation 3.4, we have xn εSr = (−1)k+r εr+1 n +
(−1)k+r+1 εSr+1 . Iterating the equation in this form, we get
εSr xi−1
n = (−1)r(i−1) εSr xi−1
n
 
i−1
X (j−1)(j−2) S∪{x } (i−2)(i−3)
=  (−1)j(r+k)+ 2 xi−j−1
n εr+j n  − (−1)(i−1)(r+k)+ 2 εSr+i−1
j=1
∼
=λ 0.
We have succeeded in showing that whenever εSr ∈ OIλ(i) it follows that
(3.10) εSr xi−1 ∼
=λ f xin
n

for some f ∈ OPoln , and so the proof is complete. 

Theorem 3.8. The spanning set OB(λ) is a homogeneous basis for OH(X λ ) as a free abelian group.
Z
Proof. Let R denote the -linear span of the set OB(λ). Since both of the sets B(λ) and OB(λ) consist
Z
of identical monomials with unit coefficients, it follows that they give a basis for the 2 -vector spaces
Z Z
H(X λ ) ⊗Z 2 and R ⊗Z 2 , respectively. Furthermore, this implies
(3.11) Z

rkq H(X λ ) = dimq,Z/2 H(X λ ) ⊗Z /2 = dimq,Z/2 (R ⊗Z /2) , Z
so there can be no relations among the elements of OB(λ). Any relation will, upon dividing by an
appropriate power of 2, give rise to a relation in H(X λ ) ⊗Z /2. Z 
Corollary 3.9. There is an equality of ranks
n!
(3.12) rkZ OH(X λ ) = rkZ H(X λ ) = .
λ1 ! · · · λn !
This equality extends to an equality of graded ranks, where the right hand side above is replaced by
an appropriate graded analog of the binomial coefficient, see [16, Equation 1.11].
Remark 3.10. While the bases B(λ) and OB(λ) consist of identical monomials, the rewriting rules for
non basis elements in OH(X λ ) differ from the corresponding rules in H(X λ ).
3.3. Maximal promotion. We now introduce a technique for determining if a monomial multiplied
by an elementary symmetric function is in the ideal OIλ . It follows from (3.4) that for S ⊂ {x1 , . . . , xn−1 }
and r ≥ 0
 
S∪{x }
(3.13) xn εSr = (−1)|S|+r εr+1 n − εSr+1 .

This equation promotes the degree of elementary symmetric functions from r to r + 1. Observe that,
starting with x2n εSr , we can promote a second time by applying (3.13) to the term xn εSr+1 .
ap
More generally, define the maximal promotion of xan1 xan+1
2
. . . xn+p+1 εSr to be the result of iteratively
applying (3.13). Start by skew-commuting xn to the right of the other variables and promoting wher-
ever possible until all powers of xn have been exhausted. The unused powers of xn are then skew
commuted back to the front. Repeat this process with the variable xn+1 , skew-commuting the unused
14 AARON D. LAUDA AND HEATHER M. RUSSELL

variables back to their original location when finished maximally promoting. Continue until all vari-
ables appearing before εSr have been maximally promoted. As an example, the maximal promotion of
x2n xn+1 εSr is
   
S∪{x } S∪{x ,x } S∪{x ,x } S∪{x }
x2n xn+1 εSr = xn εr+2 n − εr+2 n n+1 + (−1)|S|+r+3 εr+3 n n+1 − εr+3 n
 
S∪{x }
+ (−1)|S|+r+3 εr+3 n+1 − εSr+3 .

It is not too difficult to derive an explicit formula for maximal promotion. However, we will not
need such formulas in what follows. It suffices to know that the maximal promotion results in a sum
ap −κp S∪Rκ
of terms of the form xna1 −κ1 xan+1 2 −κ2
. . . xn+p+1 εr+|κ| where 1 ≤ κi ≤ ai , |κ| = κ1 + · · · + κp , and Rκ is a
subset of {xn , xn+1 , . . . , xn+p+1 } satisfying the condition that a variable xj ∈ / Rκ only if κj = aj .
For an analysis of whether or not f εSr is in the ideal OIλ for some monomial f , it suffices to check
S∪Rκ
εr+|κ| ∈ OIλ only for the minimal possible value of r + |κ| corresponding to each variable set S ∪ Rκ
arising in the maximal promotion.
Below we illustrate how maximal promotion can be used to identify elements in OIλ . The two
lemmas appearing below are somewhat technical, but they are essential for the proof of Theorem 4.7.
For any j ≥ 1 write Xj := {x1 , . . . , xj }.
Lemma 3.11. Given a partition λ with |λ| = n and h(λ) = m let d = n − λm−1 − 1, and T =
{xd+1 , xd+2 , . . . , xn }, see Figure 2. Then
(3.14) xm−2 m−2
ελm +α ∼
m−2 T
d+1 xd+2 . . . xn =λ 0
for any α > 0.
Proof. Write εTλm +α as a sum of monomials and reorder the variables in (3.14) in increasing order
resulting in a linear combination of terms of the form
xad+1
1
xad+2
2
. . . xnan−d
where λm + α of the exponents are m − 1 and the rest are m − 2. Consider the maximal promotion of
xad+1
1
xad+2
2
. . . xnan−d εX
0 .
d

The result is a linear combination of terms of the form

xad+1
1 −κ1 a2 −κ2 Xd ∪Rκ
xd+2 . . . xnan−d −κn−d ε|κ|

with 1 ≤ κj ≤ aj and Rκ ⊂ {xd+1 , . . . , xn } with xj ∈

/ Rκ only when κj = aj .
For a fixed set of variables Xd ∪ Rκ a careful analysis of the smallest r such that εrXd ∪Rκ occurs in
the maximal promotion reveals a minimal value of
r = ((h (λ) − 2) (λm−1 + 1) + λm + α) − ((h (λ) − 3) |Rκ | + ω)
where ω = min (λm + α, |Rκ |). The first set of parenthesis is the contribution if none of the variables
appear in Rκ and the second set of parenthesis is the contribution if κj = 1 for each variable xj
appearing in Rκ .
Write
ρ := |Xd ∪ Rκ | = n − (λm−1 + 1) + |Rκ |.
To show that εrXd ∪Rκ ∈ OIλ we must show r > ρ − δρ . After simplification this amounts to the
inequality
(h(λ) − 1)(λm−1 + 1) + (λm + α − ω) > (n − δρ (λ)) + (h(λ) − 2)|Rκ |
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 15

schematically illustrated below.

|Rκ |
z }| {

11111
00000
00000
11111
00000
11111
λm−1
+ (α − ω) >

λm−1
00000
11111
λm λm

Whenever |Rκ | ≤ λm it follows that ω = |Rκ |. In this case, one can check that the inequality fails
unless α > 0. 

Lemma 3.12. Given a partition λ with |λ| = n and h(λ) = m let c = n − λm−1 − λm , d = n − λm−1 − 1,
and T = {xd+1 , xd+2 , . . . , xn }, see Figure 2. Further suppose that S is a subset of {xc+1 , xc+2 . . . , xd }
and that K ≥ (λm + 1)(h(λ) − 2) − |S|(h(λ) − 3). Then
xm−2 m−2
d+1 xd+2 . . . xn εK+α ελm −α ∼
m−2 Xc ∪S T
=λ 0
for any α > 0.
Proof. The proof is similar to the previous lemma. Write εTλm −α as a sum of monomials, and for each
summand slide these monomials to the left of εX c ∪S
K+α acquiring a linear combination of terms of the
form
(3.15) xad+1
1
xad+2
2
. . . xnan−d εXc ∪S
K+α

where λm − α of the exponents are h(λ) − 1 and the rest are h(λ) − 2. The sign resulting from sliding
the monomials will not be relevant as we will show that each individual term in the sum is congruent
to zero in OIλ .
The maximal promotion of the expression in (3.15) results in a linear combination of terms of the
form
xad+1
1 −τ1 a2 −τ2
xd+2 . . . xann−d −τn−d εXc ∪S∪Rτ
K+α+|τ |

where 1 ≤ τj ≤ aj and xj ∈ / Rτ only when κj = aj . For a fixed set of variables Xc ∪ S ∪ Rτ one can
show that the smallest value of r such that εrXc ∪S∪Rτ occurs in the maximal expansion is
r = ((h(λ) − 2)(λm−1 + 1) + λm + α) − ((h(λ) − 3)|Rτ | − ω)
where ω = min (|Rτ |, λm − α). Set
ρ = |Xc ∪ S ∪ Rτ | = n − λm−1 − λm + |S| + |Rτ |.
Then εrXc ∪S∪Rτ ∈ OIλ if r > ρ − δρ , which after simplification amounts to the inequality
(h(λ) − 1)(λm + λm−1 ) + (λm − ω) > (n − δρ ) + (|S| + |Rτ |)(h(λ) − 2).
When |Rτ | = λm it follows that ω = λm − α and this inequality fails unless α > 0. 

4. I DENTIFICATION OF TOP DEGREE COMPONENTS OF OH(X λ )

In this section we work over a field F for convenience.
16 AARON D. LAUDA AND HEATHER M. RUSSELL

4.1. Specht modules for H−1 (n). For w ∈ Sn let w(t) be the action of Sn on a standard tableau t given
by permuting the entries. Given a standard tableau t we say that i precedes j in t if reading the entries
of t across rows, starting from the top left, i appears before j. Denote by tλ the row-filled tableau of
shape λ. For example, if λ = (5, 2, 1, 0, 0, 0, 0, 0) then
1 2 3 4 5
tλ = 6 7
8
Definition 4.1. The Specht module S−1 λ
F
of H−1 (n) is defined to be the -linear span of vectors vt
indexed by standard Young tableau t. The action of H−1 (n) is given by

 vt if i and i + 1 are in the same row of t, otherwise
(4.1) Ti (vt ) = −vsi (t) if i precedes i + 1 in t

−vsi (t) + 2vt if i + 1 precedes i in t.
(Recall our presentation of the Hecke algebra from Section 2.1.3.) To rewrite a nonstandard tableau in
terms of standards, one makes use of the following rewriting relations.
(1) Row relations: If t and t′ are identical except that two adjacent entries of a row have been
interchanged then
vt = vt′ .
(2) Garnir relations: A Garnir node of λ is a node (a, b) such that (a + 1, b) is also a node of λ. The
Garnir belt of λ associated to (a, b) consists of those nodes in the set A = {(a, f ) | b ≤ f ≤ λa }
and B = {(a + 1, g) | 1 ≤ g ≤ b}, see Figure 2.
Assume that t is row standard but not standard. Then there is a column containing adjacent
entries i above j for which i > j. Consider the Garnir node containing i. Let SA , SB , and
SA∪B denote the subgroups of Sn permuting only the entries of a tableau t in A, B, and A ∪ B,
respectively. For q = −1 the Garnir relation is the sum over minimal coset representatives of
SA × SB ⊂ SA∪B
X
(4.2) (−1)ℓ(w) vw(t) = 0.
w∈SA ×SB /SA∪B

As w varies over all coset representatives the tableaux w(t) give all row standard tableaux
obtained by permuting the entries A ∪ B in t. For general q the above formulas are modified
by powers of (−q).
Our description above follows [32]. See [10, 11, 29] for more details on Specht modules and [19] for
the specific case of q = −1. For a generalization to the affine Hecke algebra see [34].
In [21] an alternative presentation of the Specht module is given by generators and relations. For
λ
H−1 (n) the Specht module S−1 is freely generated by the vector vtλ subject to the relations below.
(1) Row relations: For all non identity permutations w in the row stabilizer W λ of tλ the relation
below.
Tw (vtλ ) = vtλ
holds.
(2) Garnir relations: For each Garnir node (a, b) of λ the relation
X
(−1)ℓ(w) Tw (vtλ ) = 0
w∈SA ×SB /SA∪B

holds.
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 17

Hence, one need only verify the row and Garnir relations on the generating basis vector vtλ . For a
second root of unity one can find a slightly smaller set of Garnir relations, but we will not need this
simplification here. For a completely different set of generators and relations of the Specht module
see [4, 24].
4.2. A bijection between monomials and standard tableaux. Given a monomial in n variables where
each variable appears with maximum degree n−1, consider the composition µ = µ1 , . . . , µn of n where
µi is the number of variables in the monomial of degree i − 1. We will say that the monomial has shape
µ. As an example, consider m = x1 x22 = x1 x22 x03 x04 ∈ OPol4 . Then m has shape (2, 1, 1, 0).
Given a λ-tableau t let j be the number occupying node (a, b), and set h(j) = a. Consider the map
h(1)−1 h(2)−1
(4.3) t → mt := x1 x2 . . . xnh(n)−1
sending a λ-tableau to a monomial of shape λ. The exponent of the variable xj is just the height
minus one of the corresponding box in the tableau t. It is not difficult to check that this map is a
bijection between the set of standard tableaux of shape λ and elements of maximal degree in OB(λ).
This bijection is extended to a bijection between all row standard tableau and monomials of shape λ
in [30].
Denote by mλ the basis element of OH(X λ ) corresponding to the tableau tλ . Given a part λi of a
partition λ = λ1 ≥ λ2 ≥ · · · ≥ λh(λ) , set p = λ1 + · · · + λi−1 and write
x(λi ) := xp+1 xp+2 · · · xp+λi .
Abusing notation, we denote by xi−1
(λi ) the monomial xi−1 i−1 i−1
p+1 xp+2 · · · xp+λi so that
h(λ)−1 h(λ)−1 h(λ)−1
(4.4) mλ = x0(λ1 ) x1(λ2 ) . . . x(λm ) = x01 . . . x0λ1 x1λ1 +1 . . . x1λ1 +λ2 . . . x|λ|−λm . . . x|λ| .
Remark 4.2. Our main theorem proves that incorporating certain signs in the bijection above yields
λ
an H−1 (n)-module isomorphism from the Specht module S−1 to OH(X λ ). The sign for an arbitrary
λ
standard tableau t can be deduced from the sign of m by acting by an appropriate Tw .
4.3. Proving Specht module relations in OH(X λ ).
4.3.1. Row relations.
Lemma 4.3 (Row relations). For all non-identity permutations w in the row stabilizer W λ of tλ the
relation
Tw (mλ ) = mλ
λ
holds in OH(X ).
Proof. For each si ∈ W λ the action of H−1 (n) is given by
(4.5) Ti (mλ ) = mλ + Bi (mλ ) = mλ ,
since Bi annihilates polynomials that are odd symmetric in variables i and i + 1. 
4.3.2. Garnir relations. For any Garnir node (a, b) define the Garnir element associated to (a, b) as
X
G(a,b) := Tw (mλ )
w∈SA ×SB /SA∪B
λ
in OH(X ).
We first focus on the Garnir relation associated to the bottom corner of a Young diagram. For a
partition λ with h(λ) = m, this Garnir element has the form G(m−1,λm ) . The Garnir belt associated to
G(m−1,λm ) consists of A = {(m − 1, f ) | λm ≤ f ≤ λm−1 } and B = {(m, g) | 1 ≤ g ≤ λm }. See Figure 2
for an example of a bottom corner Garnir belt.
18 AARON D. LAUDA AND HEATHER M. RUSSELL

λ1

c
λm−1 d ×××
λm ×××××

F IGURE 2. This figure features the Young diagram of the partition (9, 8, 8, 7, 5) with
the Garnir belt associated with the node (m − 1, λm ) marked with ×. In the tableau
tλ the number in the box labeled c is n − λm−1 and the box labeled d is n − λm−1 − 1.

Proposition 4.4. The Garnir element corresponding to (m − 1, λm ) can be written in the form
A∪B ∼
(4.6) G(m−1,λm ) = ±x0(λ1 ) . . . xm−3 m−2 m−2
(λm−2 ) x(λm−1 ) x(λm ) · ελm =λ 0,

where A and B are the set of terms in the Garnir belt associated to (m − 1, λm ).
Proof. Let Y(m−1,λm ) be the set of all row standard tableaux obtained by permuting elements of A ∪ B
within the Garnir belt of tλ and identical to tλ off of the Garnir belt. As observed earlier, varying over
all coset representatives w ∈ SA × SB /SA∪B , the collection of w(tλ ) is exactly Y(m−1,λm ) .
Consider t ∈ Y(m−1,λm ) . Let i1 < · · · < ik be the entries in the top row of the Garnir belt in t where
k = λm−1 − λm + 1. Let d = n − λm−1 − 1. Then A ∪ B = {d + 1, d + 2, . . . , n}, see Figure 2. Let w be
a minimal coset representative taking tλ to t, i.e. w(tλ ) = t. Since t is row standard d + j ≤ ij for all j.
For q < r define σq,r := sr−1 · · · sq+1 sq , and if q = r define σq,r = 1. Then we may assume w has the
form w := σd+1,i1 σd+2,i2 · · · σd+k,ik . A straightforward calculation shows that
k+λ
Ym h(j)−1
Tw (mλ ) = (−1)mℓ(w) x0(λ1 ) x1(λ2 ) · · · xm−3 m−2 m−2
(λm−2 ) · xn−λm−1 −λm +1 · · · xd · xd+j
j=1

λ
where h(j) is the height at which the entry d + j appears in the tableau w(t ).
Consider t ∈ Y(m−1,λm ) and its associated minimal coset representative w ∈ SA × SB /SA∪B . Let
ik+1 < · · · < in be the entries in the bottom row of t, and let xw := xik+1 · · · xin . Then we can write
 
k+λ
Ym h(j)−1 k+λ
Ym
xd+j = (−1)mℓ(w)  xm−2
d+j
 xw .
j=1 j=1

Putting these together we get the following expression for the Garnir element
X
G(m−1,λm ) = (−1)ℓ(w) Tw (mλ )
w∈SA ×SB /SA∪B
X
= x0(λ1 ) x1(λ2 ) · · · xm−3 m−2 m−2
(λm−2 ) x(λm−1 ) x(λm ) (−1)ℓ(w) xw .
w∈SA ×SB /SA∪B

It only remains to verify that the signed sum of terms xw is the appropriate odd partial elementary
symmetric function. Given two coset representatives w = si · w′ their monomials xw and xw′ are
identical except that one contains xi while the other contains xi+1 . Furthermore ℓ(w) = ℓ(w′ )+1, so xw
and xw′ are given opposite sign in the Garnir element G(m−1,λm ) . This is exactly the sign convention
P
in εA∪B
λm , so w∈SA ×SB /SA∪B (−1)
ℓ(w)
xw = ±εA∪B
λm . 

Proposition 4.5. Given a partition λ with |λ| = n and h(λ) = m, then G(m−1,λm ) ∼
=λ 0.
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 19

Proof. Let c = n − λm−1 − λm and d = n − λm−1 − 1, as in Figure 2. Recall that Xc := {x1 , x2 , . . . , xc }.

Consider the monomial xm−2 m−2 m−2
c+1 xc+2 . . . xd corresponding to those boxes in row (m − 1) not involved
m−2 Xc
in the Garnir relation. Maximally promoting xm−2 m−2
c+1 xc+2 . . . xd ε0 results in a linear combination of
terms of the form
m−2−κλm −1 Xc ∪Rκ
xm−2−κ
c+1
1
. . . xd ε|κ|
with 1 ≤ κj ≤ m − 2, as was explained in Section 3.3. Then by Proposition 4.4 (after rearranging
terms) we have
X m−2−κλm −1 m−2
m−2−κ1
G(m−1,λm ) = ακ x0(λ1 ) . . . xm−3 m−2
(λm−2 ) x(λm−1 ) xc+1 . . . xd xd+1 . . . xm−2
n εX c ∪Rκ A∪B
|κ| ελm
where the sum is over all terms in the maximal promotion. The precise form of the value of the
coefficients ακ will not be relevant here.
m−2 Xc ∪Rκ A∪B ∼
The result follows by proving that xm−2
d+1 . . . xn ε|κ| ελm =λ 0 for each term appearing in the
maximal promotion. To see this note that for any partition of a subset S ⊂ {x1 , . . . , xn } into ordered
subsets U and V it is evident that
Xr
(4.7) εSr = (−1)|U|·j eU V
r−j ej .
j=0

Then we have
 X 
εX c ∪Rκ A∪B
|κ| ε λm = (−1)|Xc ∪Rκ |·λm
ε Xc ∪Rκ ∪A∪B
|κ|+λm − (−1)|Xc ∪Rκ |·j Xc ∪Rκ
ε ε
|κ|+λm −j j
A∪B

where the sum is over all 0 ≤ j ≤ |κ| + λm with j 6= λm . The reader can easily verify that
Xc ∪Rκ ∪A∪B
ε|κ|+λm
∈ OIλ .
Using Lemma 3.11 when j > λm and Lemma 3.12 when j < λm , it follows that each term in the sum
is also congruent to zero in OIλ . 
Proposition 4.6. Given a Garnir node (a, b) of a partition λ then
(4.8) G(a,b) ∼
=λ 0
in OH(X λ ).
Proof. If (a, b) is not a bottom corner Garnir node, let λ′ be the subpartition of n′ < n obtained by
removing all boxes below and to the right of the Garnir belt in λ so that the node (a, b) = (h(λ′ ) −
1, λ′h(λ′ ) ) is a bottom corner in λ′ . It follows from Proposition 4.5 that G(a,b) ∼
=λ′ 0. Let µ be the
′
partition subordinate to λ obtained by adding a box to λ of minimal height. Then arguing as in (3.10)
h(µ)−1 h(µ)
if follows that G(a,b) xn′ +1 ∼ =µ f · xn′ +1 for some f ∈ OH(X µ ). Continuing in this way the Garnir
relation for the node (a, b) in λ follows from the Garnir relation for this node in λ′ . 
4.3.3. Main theorem. Propositions 4.3 and 4.6 prove our main result.
Theorem 4.7. There is an isomorphism S−1 λ
= OH top (X λ ) ⊗Z
∼ F
of H−1 (n)-modules sending the
λ
generator vtλ corresponding to the row filled tableau in the Specht module S−1 to the monomial mλ .

R EFERENCES
[1] R.M. Adin, A. Postnikov, and Y. Roichman. Hecke algebra actions on the coinvariant algebra. J. Algebra, 233(2):594–613,
2000.
[2] J. Bernstein, I. B. Frenkel, and M. Khovanov. A categorification of the Temperley-Lieb algebra and Schur quotients of
U(sl(2)) via projective and Zuckerman functors. Selecta Math. (N.S.), 5(2):199–241, 1999.
[3] R. Biagioli, S. Faridi, and M. Rosas. The defining ideals of conjugacy classes of nilpotent matrices and a conjecture of
Weyman. Int. Math. Res. Not. IMRN, 2008. arXiv:0803.0658.
20 AARON D. LAUDA AND HEATHER M. RUSSELL

[4] J. Brundan, A. Kleshchev, and W. Wang. Graded Specht modules. J. Reine Angew. Math., 655:61–87, 2011.
arXiv:math.RT/0901.0218.
[5] J. Brundan and V. Ostrik. Cohomology of Spaltenstein varieties. Transform. Groups, 16:619–648, 2011. arXiv:1012.3426.
[6] S. Cautis and J. Kamnitzer. Knot homology via derived categories of coherent sheaves. I. The sl(2)-case. Duke Math. J.,
142(3):511–588, 2008. arXiv:math/0701194.
[7] Y. Chen and M. Khovanov. An invariant of tangle cobordisms via subquotients of arc rings, 2006. arXiv:math/0610054.
[8] J. Chuang and R. Rouquier. Derived equivalences for symmetric groups and sl 2-categorification. Ann. of Math., 167:245–
298, 2008. arXiv:math.RT/0407205.
[9] C. De Concini and C. Procesi. Symmetric functions, conjugacy classes and the flag variety. Invent. Math., 64(2):203–219,
1981.
[10] R. Dipper and G. James. Representations of Hecke algebras of general linear groups. Proc. London Math. Soc. (3), 52(1):20–
52, 1986.
[11] R. Dipper and G. James. Blocks and idempotents of Hecke algebras of general linear groups. Proc. London Math. Soc. (3),
54(1):57–82, 1987.
[12] A. P. Ellis. The odd Littlewood-Richardson rule. 2012. arXiv:1111.3932.
[13] A. P. Ellis and M. Khovanov. The Hopf algebra of odd symmetric functions. 2011. arXiv:math.QA/1107.5610.
[14] A. P. Ellis, M. Khovanov, and A. Lauda. The odd nilhecke algebra and its diagrammatics. 2012. arXiv:1111.1320.
[15] I. B. Frenkel, M. Khovanov, and C. Stroppel. A categorification of finite-dimensional irreducible representations of quan-
tum sl(2) and their tensor products. Selecta Math. (N.S.), 12(3-4):379–431, 2006. math/0511467.
[16] A. M. Garsia and C. Procesi. On certain graded Sn -modules and the q-Kostka polynomials. Adv. Math., 94(1):82–138, 1992.
[17] D. Hill and W. Wang . Categorification of quantum Kac-Moody superalgebras. 2012. arXiv:1202.2769.
[18] R. Hotta and T. A. Springer. A specialization theorem for certain Weyl group representations and an application to the
Green polynomials of unitary groups. Invent. Math., 41(2):113–127, 1977.
[19] G. James and A. Mathas. Hecke algebras of type A with q = −1. J. Algebra, 184(1):102–158, 1996.
[20] S.-J. Kang, M. Kashiwara, and S. Tsuchioka. Quiver Hecke superalgebras. 2011. arXiv:math.QA/1107.1039v1.
[21] S.J. Kang, I.-S. Lee, K.H. Lee, and H. Oh. Hecke algebras, Specht modules and Gröbner-Shirshov bases. J. Algebra,
252(2):258–292, 2002.
[22] M. Khovanov. Crossingless matchings and the cohomology of (n, n) Springer varieties. Commun. Contemp. Math., 6(4):561–
577, 2004. arXiv:math/0202110.
[23] M. Khovanov, A. Lauda, M. Mackaay, and M. Stošić. Extended graphical calculus for categorified quantum sl(2). 2010.
arXiv:math.QA/1006.2866.
[24] A. Kleshchev, A. Mathas, and A. Ram. Universal Specht modules for cyclotomic Hecke algebras, 2011. arXiv:1102.3519.
[25] A. D. Lauda. Categorified quantum sl(2) and equivariant cohomology of iterated flag varieties. Algebr. Represent. Theory,
14(2):253–282, 2011. math.QA/0803.3848.
[26] A.D. Lauda. A categorification of quantum sl(2). Adv. Math., 225(6):3327–3424, 2010. arXiv:math.QA/0803.3652.
[27] A.D. Lauda. An introduction to diagrammatic algebra and categorified quantum sl(2). Bulletin of the Institute of Mathematics
Academia Sinica, 7(2):65–270, 2012. arXiv:1106.2128.
[28] R. Lipshitz, P. Ozsváth, and D. Thurston. Bordered Heegaard Floer homology: Invariance and pairing. 2008.
arXiv:math.GT/0810.0687.
[29] A. Mathas. Iwahori-Hecke algebras and Schur algebras of the symmetric group, volume 15 of University Lecture Series. AMS,
Providence, RI, 1999.
[30] A. Mbirika. A Hessenberg generalization of the Garsia-Procesi basis for the cohomology ring of Springer varieties. Electron.
J. Combin., 17(1):Research Paper 153, 29, 2010. arXiv:0911.4962.
[31] A. Mbirika and J. Tymoczko. Generalizing Tanisaki’s ideal via ideals of truncated symmetric functions. 2010.
arXiv:1012.1630.
[32] G. E. Murphy. The representations of Hecke algebras of type An . J. Algebra, 173(1):97–121, 1995.
[33] P. Ozsváth, J. Rasmussen, and Z. Szabó. Odd Khovanov homology. 2007. arXiv:0710.4300.
[34] A. Ram. Skew shape representations are irreducible. In Combinatorial and geometric representation theory (Seoul, 2001), vol-
ume 325 of Contemp. Math., pages 161–189. Amer. Math. Soc., Providence, RI, 2003.
[35] A. Shumakovitch. Patterns in odd Khovanov homology. J. Knot Theory Ramifications, 20(1):203–222, 2011. arXiv:1101.5607.
[36] T. A. Springer. A construction of representations of Weyl groups. Invent. Math., 44(3):279–293, 1978.
[37] C. Stroppel. Parabolic category O, perverse sheaves on Grassmannians, Springer fibres and Khovanov homology. Compos.
Math., 145(4):954–992, 2009. arXiv:math/0608234.
[38] C. Stroppel and B. Webster. 2-block Springer fibers: convolution algebras and coherent sheaves, 2008. arXiv:0802.1943.
[39] T. Tanisaki. Defining ideals of the closures of the conjugacy classes and representations of the Weyl groups. Tôhoku Math.
J. (2), 34(4):575–585, 1982.
[40] W. Wang. Double affine Hecke algebras for the spin symmetric group. Math. Res. Lett., 16:1071–1085, 2009.
arXiv:math.RT/0608074.
 ODDIFICATION OF THE COHOMOLOGY OF TYPE A SPRINGER VARIETIES 21

[41] J. Y.Thibon and B.C.V. Ung. Quantum quasi-symmetric functions and Hecke algebras. J. Phys. A, 29(22):7337–7348, 1996.

D EPARTMENT OF M ATHEMATICS , U NIVERSITY OF S OUTHERN C ALIFORNIA , L OS A NGELES , CA 90089, USA

E-mail address: [email protected]

D EPARTMENT OF M ATHEMATICS , U NIVERSITY OF S OUTHERN C ALIFORNIA , L OS A NGELES , CA 90089, USA

E-mail address: [email protected]