Вы находитесь на странице: 1из 48

THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS

ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA


arXiv:1111.1320v1 [math.QA] 5 Nov 2011

A BSTRACT. We introduce an odd version of the nilHecke algebra and develop an odd analogue of the
thick diagrammatic calculus for nilHecke algebras. We graphically describe idempotents which give a
Morita equivalence between odd nilHecke algebras and the rings of odd symmetric functions in finitely
many variables. Cyclotomic quotients of odd nilHecke algebras are Morita equivalent to rings which are
odd analogues of the cohomology rings of Grassmannians. Like their even counterparts, odd nilHecke
algebras categorify the positive half of quantum sl(2).

C ONTENTS
1. Introduction 2
1.1. Background 2
1.2. Outline 2
2. Odd symmetric polynomials and the odd nilHecke algebra 3
2.1. Odd symmetric polynomials 3
2.2. The odd nilHecke algebra 10
2.3. Odd Schur polynomials 18
3. Graphical calculus 20
3.1. Graphical calculus for the odd nilHecke algebra 20
3.2. Box notation for odd thin calculus 22
3.3. Automorphisms of the odd nilHecke algebra 30
4. Extended graphical calculus for the odd nilHecke algebra 30
4.1. Thick calculus 30
4.2. Explosions 33
4.3. Orthogonal idempotents 34
4.4. The odd nilHecke algebra as a matrix algebra 38
5. Cyclotomic quotients of the odd nilHecke algebra 43
6. Categorification 46
References 47

Date: November 5, 2011.


1
2 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

1. I NTRODUCTION
1.1. Background. The nilHecke algebra plays a central role in the theory of categorified quantum
groups, giving rise to an integral categorification of the negative half of Uq (sl2 ) [10, 7, 15]. One of
the original motivations for categorifying quantum groups was to provide a unified representation
theoretic explanation of the link homology theories that categorify various quantum link invariants.
Various steps in this direction have since been completed, most notably by Webster [18, 19].
Khovanov homology is one of the simplest of these link homology theories, categorifying a certain
normalization of the Jones polynomial. Just as the quantum group sl2 plays a fundamental role in
explaining the Jones polynomial, the categorification of quantum sl2 should play an equally important
role in Khovanov homology.
Surprisingly, the categorification of the Jones polynomial is not unique. Ozsvath, Rasmussen, and
Szabo found an odd analogue of Khovanov homology [14]. This odd homology theory for links agrees
with the original Khovanov homology modulo 2. Both of these theories categorify the Jones polyno-
mial, and results of Shumakovitch [16, Section 3.1] show that these categorified link invariants are not
equivalentboth can distinguish knots that are indistinguishable in the other theory.
This odd analogue of Khovanov homology hints that there should be a corresponding odd ana-
logue of categorified quantum groups. We expect that these odd categorified quantum groups will
not be equivalent in any appropriate sense to their even counterparts, though they will still categorify
the same quantum group.
In this paper we begin the program of odd categorification of quantum groups by studying an
odd analogue of the nilHecke algebra that categorifies the negative half of Uq (sl2 ). This algebra was
introduced by Kang, Kashiwara, and Tsuchioka [2] several months ago. We arrived at the odd nil-
Hecke algebra independently via the categorification of positive half of quantum gl(1|1) by way of
Lipshitz-Ozsvath-Thurston rings [6]. LOT rings provide the odd counterpart of the nilCoxeter alge-
bra, the subalgebra of the nilHecke algebra generated by the divided difference operators. Enlarging
LOT rings via suitable generators (represented diagrammatically by dots) and relations results in the
odd counterpart of the nilHecke algebra. A different motivation and approach to this odd nilHecke
algebra can be found in [2] together with its extension to odd quiver Hecke algebras for certain root
systems. Several years ago Weiqiang Wang [17] (see also the work of Khongsap and Wang [3, 4, 5])
introduced a so-called spin Hecke algebra and, as we learned via private communication, he was
aware of a nil version of his construction which he would call the spin nilHecke algebra and which
is identical to the odd nilHecke algebra.
Just as the nilHecke algebra has a close relationship with the combinatorics of symmetric functions,
the odd nilHecke algebra gives rise to an odd analogue of the ring of symmetric functions. This ring is
noncommutative, yet shares many of the same combinatorics of symmetric functions. Odd symmetric
functions were defined by the first two authors in [1]. Here we connect the theory of odd symmetric
functions with the odd nilHecke algebra and construct odd analogues of Schubert polynomials, Schur
polynomials and further develop their combinatorics, including the odd Pieri rule.
The nilHecke algebra is also pervasive throughout the theory of Grassmannians and partial flag
varieties. In particular, the nilHecke algebra admits so-called cyclotomic quotients that are Morita
equivalent to the cohomology rings of Grassmannians. Here we study the corresponding quotients of
the odd nilHecke algebra and show that they are Morita equivalent to certain noncommutative rings
that seem to be odd analogues of cohomology rings of complex Grassmannians. We hope that these
rings will show a path to a rather special case of quantum (noncommutative) geometry.

1.2. Outline. Section 2 introduces the main characters: odd analogues of divided difference opera-
tors, the nilHecke algebra, and symmetric polynomials. As in the even case, the odd nilHecke algebra
ON Ha is a matrix algebra over the ring of odd symmetric polynomials Oa . Basic results relating
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 3

these to their even counterparts are established, and odd analogues of some of the usual bases of
symmetric functions are given. An important technical result is the Omission Word Lemma (Lemma
2.18), which implies that the longest divided difference operator Da is left linear over the ring of
odd symmetric polynomials, up to an involution (it is automatically right linear). This investigation
culminates in the definition of odd Schur polynomials and an odd Pieri rule describing products s k .
In the following Section 3, we give a graphical description of the odd nilHecke algebra in the spirit
of [7]. The familiar planar isotopy relations only hold up to sign in the odd case; distant dots and
crossings anticommute. Much of Subsection 3.2 is devoted to keeping track of the resulting signs. A
useful tool here is a family of elements (3.16) which obey the (usual, not odd) 0-Hecke relations.
These 0-Hecke generators can be used to define idempotents ea which, when expressed diagram-
matically, we refer to as thick lines. Section 4 studies the resulting thick calculus, in analogy with
[8]. The twisted left linearity of Da is a key ingredient here, as it allows us to make sense of labelling
thick lines by elements of Oa . In Subsection 4.3, using some technical lemmas, we are able to de-
compose the unit element of ON Ha and ea eb ON Ha ON Hb ON Ha+b into indecomposable
orthogonal idempotents, see Theorems 4.15 and 4.16. In Section 6, we describe how these decompo-
sitions categorify the relations between divided powers in the positive part of Lusztigs integral form
of quantum sl2 .
In the even case, categorified quantum sl2 acts on a bimodule bicategory coming from the cohomol-
ogy rings of complex Grassmannians [10, 9]. Section 5 gives odd analogues of these rings; as expected,
they are quotients of the odd symmetric polynomials, and they have bases consisting of certain odd
Schur functions.

Acknowledgments:
The first author was supported by the NSF Graduate Research Fellowship Program. The second
author was partially supported by NSF grant DMS-0706924. The third author was partially supported
by NSF grant DMS-0855713 and by the Alfred P. Sloan foundation. All the authors would like to
acknowledge partial support from Columbia Universitys RTG grant DMS-0739392. A conversation
between the first author and Pedro Vaz led to a substantial simplification in the proof of the OWL
(Lemma 2.18).

2. O DD SYMMETRIC POLYNOMIALS AND THE ODD NIL H ECKE ALGEBRA

2.1. Odd symmetric polynomials.

2.1.1. Defining the odd symmetric polynomials. We define the ring of odd polynomials to be the free unital
associative algebra on skew-commuting variables x1 , . . . , xa ,

(2.1) Z
OPola = hx1 , . . . , xa i/hxi xj + xj xi = 0 for i 6= ji.

We let the symmetric group Sa act on the degree k part of OPola as the tensor product of the permu-
tation representation and the k-th tensor power of the sign representation. That is, for 1 j a 1,
the transposition sj Sa acts as the ring endomorphism


xi+1 if j = i,
(2.2) si (xj ) = xi if j = i + 1,


xj otherwise.
4 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Z
The odd divided difference operators are the linear operators i (1 i a 1) on hx1 , . . . , xa i defined
by
i (1) = 0,
(
(2.3) 1 if j = i, i + 1,
i (xj ) =
0 otherwise,

and the Leibniz rule


(2.4) Z
i (f g) = i (f )g + si (f )i (g) for all f, g hx1 , . . . , xa i.
It is easy to check from the definition of i that for all i,
(2.5) i (xj xk + xk xj ) = 0 for j 6= k,
so i descends to an operator on OPola .
Note that OPola is left and right Noetherian and has no zero divisors, but does not satisfy the
unique factorization property if a 2:
x21 + x22 = (x1 x2 )2 = (x1 + x2 )2 .
The following basic formulae in OPola can be derived from the above.
i (xi xi+1 ) = 0, i (xi xi+1 ) = 0,
i (x2m
i + x2m
i+1 ) = 0, i (xm m
i xi+1 ) = 0,
m1
X
(2.6) i (xm
i ) = (1)j xji+1 xm1j
i ,
j=0
m1
X
i (xm
i+1 ) = (1)j xji xm1j
i+1 .
j=0

Proposition 2.1. Considering i and (multiplication by) xj as operators on OPola , the following rela-
tions hold in End(OPola ):
(2.7) i2 = 0, i i+1 i = i+1 i i+1 ,
(2.8) xi i + i xi+1 = 1, i xi + xi+1 i = 1,
(2.9) xi xj + xj xi = 0 (i 6= j), i j + j i = 0 (|i j| > 1),
(2.10) 6 j, j + 1).
xi j + j xi = 0 (i =
Proof. We first prove that i2 (f ) = 0 for any f OPola . We can reduce to the case i = 1 and f being a
monomial in x1 , x2 , and then proceed by induction on the degree of f . When the degree is zero, f = 1
and 12 (1) = 0. Assume 12 (f ) = 0 for any monomial of degree m. Then
12 (x1 f ) = 1 (f x2 1 (f )) = 1 (f ) 1 (x2 )1 (f ) + x1 12 (f ) = 1 (f ) 1 (f ) = 0,
12 (x2 f ) = 1 (f x1 1 (f )) = 1 (f ) 1 (x1 )1 (f ) + x2 12 (f ) = 0,
which takes care of the inductive step. Next, we verify relations of the second type from (2.7). It
suffices to assume that i = 1 and the action is on a monomial in x1 , x2 , x3 . We proceed by induction
on the degree of the monomial. When f = 1, both sides act by 0. Assuming that
1 2 1 (f ) = 2 1 2 (f )
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 5

we compute
1 2 1 (x1 f ) = 1 2 (f x2 (1 (f ))) = 1 (2 (f ) 1 (f ) + x3 (2 1 (f )))
= 1 2 (f ) 12 (f ) x3 (1 2 1 (f )) = 1 2 (f ) x3 (1 2 1 (f )),
2 1 2 (x1 f ) = 2 1 (x2 2 (f )) = 2 (2 (f ) + x2 1 2 (f ))
= 1 2 (f ) x3 (2 1 2 (f )),

implying 1 2 1 (x1 f ) = 2 1 2 (x1 f ). Likewise,


1 2 1 (x2 f ) = 1 2 (f x1 (1 (f ))) = 1 (2 (f ) + x1 (2 1 (f )))
= 1 2 (f ) + 2 1 (f ) x2 (1 2 1 (f )),
2 1 2 (x2 f ) = 2 1 (f x3 2 (f )) = 2 (1 (f ) + x3 1 2 (f ))
= 1 2 (f ) + 2 1 (f ) x2 2 1 2 (f ),

proving the inductive step in this case. Checking that the two actions are the same on x3 f is equally
simple.
Relations (2.8) follow from the Leibniz rule (2.4):

i (xi+1 f ) = (xi+1 )f xi i (f ) = f xi i (f ),
i (xi f ) = f xi+1 i (f ).

The first type of relation in (2.9) consists of defining relations in OPola , and the second type in (2.9)
can be checked by applying i j + j i to monomials in x1 , . . . , xa to get 0. The relation (2.10) is again
a special case of (2.4). 

We can consider OPola as a graded ring by declaring each xi to be homogeneous of degree 2; the
operators i are all homogeneous of degree 2. Since i2 = 0, we can consider OPola as a chain
complex (taking the homological grading to be one-half of the grading deg(xi ) = 2, so (OPola )2k sits
in homological degree k). The multiplication operators
k : (OPola )k (OPola )k+1
(
(2.11) xi k is even
k =
xi+1 k is odd

give a chain homotopy between the identity and the zero maps, so the complex ((OPola )k , i ) is con-
tractible for each i, a. In particular, ker(i ) = im (i ) for each i. We define the ring of odd symmetric
polynomials to be the subring
a1
\ a1
\
(2.12) Oa = ker(i ) = im (i )
i=1 i=1

of OPola . Observe that


Z
OPola Z ( /2)
= Pola Z ( /2) Z
(2.13)
Z
Oa Z ( /2) = a Z ( /2), Z
the usual (commutative) rings of polynomials and symmetric polynomials in a variables over /2. Z
In particular, both OPola and Oa are free abelian groups whose ranks in each degree are bounded
below by those of Pola and a , since rkq (Pola ) = (dimq )Z/2 (Pola ) and rkq (a ) = (dimq )Z/2 (a ). Here,
6 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

for a graded abelian group V of finite rank in each degree and a graded vector space W over a field F
of finite dimension in each degree,
X
rkq (V ) = rk (Vi )q i ,
i Z
(2.14) X
(dimq )F (W ) = dimF (Wi )q i .
i Z
1
Both are power series in q . It is clear that Pola and OPola have the same graded rank.
2.1.2. Odd elementary symmetric polynomials and the size of OPola . Let
q n q n
[n] = ,
(2.15) q q 1
[n]! = [n][n 1] [2][1]
N
be the balanced q-numbers and q-factorial; both are elements of [q, q 1 ] and both reduce to their
non-quantum analogues n, n! as q 1. They are especially useful for keeping track of graded ranks.
For instance, the ring of symmetric polynomials is isomorphic to a graded polynomial algebra,
(2.16) a Z
= [even , even , . . . , even ],
1 2 a

where even
k is the usual elementary symmetric polynomial
X
(2.17) even
k (x1 , . . . , xa ) = xi1 xia
1i1 <<ik a

of degree 2k (since we are declaring deg(xi ) = 2). Hence the graded rank of a is
a
Y 1
rkq (a ) =
i=1
1 q 2i
Ya  1

(2.18) 1 i+1 q q
= q
(1 q 2 )a i=1 q a q a
1 1
= q a(a1)/2 ,
[a]! (1 q 2 )a
an element of Z0Jq2 K. This equals
rkq (Pola )
(2.19) rkq (a ) = P ()
,
Sa q

where denotes the Coxeter length function on Sa ; this makes one think of taking a quotient of the
ring of all polynomials by the symmetric group.

Proposition 2.2. The rings of symmetric and odd symmetric polynomials have the same graded ranks:
1 1
(2.20) rkq (Oa ) = rkq (a ) = q a(a1)/2 .
[a]! (1 q 2 )a
To prove this proposition, we will have to develop odd analogues of the elementary symmetric
polynomials.
By analogy with the even case, we introduce the odd elementary symmetric polynomials
X
(2.21) k (x1 , . . . , xa ) = ei1 x
x eik , ei = (1)i1 xi ,
where x
1i1 <<ik a
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 7

for 1 k a. If k = 0 define k = 1, and if k < 0 or k > a define k = 0. If we want to emphasize


(a)
the number of variables in shorthand, we will write k for k (x1 , . . . , xa ). Modulo 2, the elementary
symmetric and odd elementary symmetric polynomials are the same. But with signs, the relations
among the odd k are more complicated than mere commutativity. The following lemma will give us
enough relations to write down a presentation of Oa .
Lemma 2.3. The polynomial k (x1 , . . . , xa ) is odd symmetric for 1 k a. Furthermore, the follow-
ing relations hold in the ring Oa :
i 2mi = 2mi i (1 i, 2m i a),
(2.22) i 2m+1i + (1) 2m+1i i = (1)i i+1 2mi + 2mi i+1
i
(1 i, 2m i a 1),
1 2m + 2m 1 = 22m+1 (1 < 2m a 1).
Note that the third is the i = 0 case of the second.
By the first relation, odd subscripts commute with odd subscripts and even subscripts commute
Z
with even subscripts. These relations are enough to reduce any word i1 ir to a -linear combina-
tion of terms of the form j1 js with j1 js ; hence the rank of Oa in each degree is bounded
above by that of a in the same rank (cf. the proof of Proposition 2.2 below).
Proof. The relations are all true in degrees 0, 1. The third relation is the i = 0 case of the second. In
a = 1 variable, all the relations are clear. We now prove the first and second relations simultaneously
by induction on the number of variables a. The equation
(a1)
(2.23) (a) (a1)
m = m + m1 x
ea
(a1)
allows us to reduce calculations to fewer variables. Now we compute (writing just j for j ):
(2.24)
(a) (a) (a) (a)
i 2mi 2mi i = (i + i1 x ea ) (2mi + 2mi1 x
ea ) (2mi + 2mi1 x ea ) (i + i1 x
ea )
= (i 2mi 2mi i ) + (1)i1 (i1 2mi1 2mi1 i1 ) x
e2a

+ (1)i i1 2mi 2mi i1 + i 2mi1 (1)i 2mi1 i xea
= 0.
The second equality was grouping terms into powers of x ea ; the third equality used the induction hy-
e2a (first relation) and on the coefficient of x
pothesis on the coefficients of 1 and of x ea (second relation).
The second relation is proved similarly. 
Remark 2.4. The relations (2.22) allow one to sort a product i1 ik into non-increasing order of
subscripts (up to sign), modulo a set of terms which are lexicographically higher (that is, the word
Z
formed by their subscripts is greater in the lexicographic ordering) and in 2 Oa. This follows from
a consequence of the odd degree relation: suppose k < and k + is odd. Then
k
X i
k = k + 2 (1)(2) +i ki if k is even,
i=1
(2.25)
k
X i1
k = k + 2 (1)( 2 ) +i ki if k is odd.
i=1

Proof of Proposition 2.2. Let Oelem a be the subring of Oa generated by the odd elementary symmetric
polynomials k (x1 , . . . , xa ). For a partition = (1 , . . . , m ) of a written in non-increasing order, let
(2.26) = 1 m .
8 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Lemma 2.3 and the remark which follows it imply that these products span Oelem
a . Hence the rank
of each graded piece of Oelem
a is bounded above by that of the corresponding graded piece of a .
Conversely, we have an isomorphism

(2.27) a Z ( /2) Z
= Oa Z ( /2)
= Oelem
a Z
Z ( /2). Z
The identification of the first and third follows because the generators and relations of Oelem a and
a coincide mod 2; the identification of the first and second follows because mod 2, the action of the
divided difference operators on Pola and the action of the odd divided difference operators on OPola
Z
coincide. The graded rank of a and the graded dimension of a Z ( /2) are equal. This bounds
the graded rank of Oelema below by that of a : indeed, dividing any linear relation between the in
Oelem
a by a high enough power of 2, we would obtain a linear relation between their reductions mod
2, a contradiction. Thus rkq (a ) = rkq (Oelem
a ).
To conclude that Oa = Oelem , we prove that both have free abelian complements in OPola . For
Z
a
Oa , this is because if there were no free complement, some free direct summand (as a -submodule)
would be wholly divisible by an integer d > 1. But then we could divide generators of this summand
by d. The result would still be in the kernel of all the operators i , a contradiction. As for Oelem
a ,
one checks that with respect to a lexicographic order on monomials, the highest order term of is
x = x m
1 xr
1
with coefficient 1. Now since Oelem Oa and both have free complements, the
Z
a
graded dimensions over /2 of their reductions mod 2 coincide if and only if they are equal. As both
rings have mod 2 reduction isomorphic to a , these graded dimensions do coincide. So Oelem a = Oa ,
and we have established the formula
1 1
(2.28) rkq (Oa ) = q a(a1)/2 .
[a]! (1 q 2 )a


The following lemma is useful for passing between the rings Oa and Oa1 .
Lemma 2.5. For any k 0,
(a1) (a) (a) (a)
k = k k1 x e2a . . . + (1)k x
ea + k2 x eka
k
X
(2.29) (a)
= (1)j kj x
eja .
j=0

(a)
Proof. The polynomial k consists of terms without x ea and terms with xea . The former add up pre-
(a1) (a) (a)
cisely to k . The latter all appear as terms in k1 x
ea ; when we subtract off k1 x
ea , the extra terms
(a)
subtracted are precisely those which already had a factor of x
ea in k1 . The extra terms are now those
(a)
e2a , and they appear in k2 x
with a factor of x e2a . Continuing in this fashion (inclusion-exclusion), the
formula follows. 

By a slight abuse of notation, if R S is a subring and s S, we write R[s] for the subring of S
generated by R and s.
Corollary 2.6. Inside OPola , Oa [xa ] = Oa1 [xa ].
Proof. : Let f Oa . Using skew-commutativity, we can move all factors of xa in any term of f
all the way to the right. Collecting powers of xa , we see each coefficient of a given power of xa is an
element of Oa1 , so Oa Oa1 [xa ]. The converse follows from the previous lemma. 
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 9

2.1.3. Odd complete symmetric polynomials.

Definition 2.7. For k 1, the k-th odd complete symmetric polynomial is defined to be
X
(2.30) hk (x1 , . . . , xa ) = ei1 x
x eik .
1i1 ik a

Also define h0 = 1 and hk = 0 for k < 0.

Lemma 2.8. The polynomials k and hk in OPola satisfy


m
X 1
(2.31) (1) 2 k(k+1) k hmk = 0
k=0

for all m 1.

Proof. We proceed by induction on the number of variables a, the case a = 1 being clear. Let k , hk
denote the odd elementary and odd complete polynomials in a variables, so that
k
X
(a+1) (a+1)
k = k + k1 x
ea+1 , hk = eja+1 .
hkj x
j=0

Plugging these expressions into the left-hand side of equation (2.31) in a + 1 variables,

Xn Xn nk
X
1 (a+1) (a+1) 1
(1) 2 k(k+1) k hnk = ea+1 )
(1) 2 k(k+1) (k + k1 x eja+1
hnkj x
k=0 k=0 j=0
n nk
X X 1  j
= (1) 2 k(k+1) k hnkj + (1)nkj k1 hnkj x
ea+1 x
ea+1
k=0 j=0
n nj
X X 1
= eja+1
(1) 2 k(k+1) k hnkj x
j=0 k=0

X nj+1
n+1 X 1
+ eja+1 .
(1) 2 k(k+1)+nk+j1 k1 hnkj+1 x
j=1 k=0

In the last equality, the order of summation was reversed and the second term was re-indexed, j 7
j 1. Consider the last line: the inner sum of the first term is zero unless j = n, k = 0 (induction
hypothesis). Also, the second term is 0 when k = 0. Removing these vanishing terms, removing
boundary terms from the summations, re-combining the two summation terms, and re-indexing k 7
k + 1, this equals
n+1 nj
!
X X 1
ena+1 +
... = x (1)n+j1 (1) 2 k(k+1) k hnkj xeja+1 .
j=1 k=0

The inner sum here is zero unless j = n, k = 0 (induction hypothesis), in which case it cancels the
ena+1 . So the entire expression equals zero, and we are done.
x 

Equation (2.31) can be used inductively to solve for each hm as a polynomial in the various k , so
each hm is indeed odd symmetric.
10 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Lemma 2.9. The polynomials hk satisfy the same relations as the k in the ring Oa :
hi h2mi = h2mi hi (1 i, 2m i a),
(2.32) hi h2m+1i + (1)i h2m+1i hi = (1)i hi+1 h2mi + h2mi hi+1 (1 i, 2m i a 1),
h1 h2m + h2m h1 = 2h2m+1 (1 < 2m a 1).
Furthermore, we have the following mixed relations:
i h2mi = h2mi i (1 i, 2m i a),
(2.33)
i h2m+1i + (1) h2m+1i i = (1)i i+1 h2mi + h2mi i+1
i
(1 i, 2m i a 1).
Proof. The proofs of all these relations are similar to the proof of Lemma 2.3, using equation (2.23) and
its complete polynomial analogue
m
X (a1)
(2.34) h(a)
m = eja .
hkj x
j=0

As in the case of elementary functions, define h = h1 hm for a partition = (1 , . . . , m ).


In the even case, perhaps the most manifestly symmetric functions are the so-called monomial
symmetric functions
X
(2.35) m = x (sum over all distinct permutations of )

associated to a partition . However, for certain , no such analogous sum yields an odd symmetric
polynomial. An equivalent definition of these functions in the even setting is that they form the basis
dual to the basis {h } with respect to a standard bilinear form. For an approach to odd symmetric
functions along these lines, see [1].

2.2. The odd nilHecke algebra. By analogy with the even case [13], we define the odd nilHecke ring
ON Ha to be the graded unital associative ring generated by elements x1 , . . . , xa of degree 2 and el-
ements 1 , . . . , a1 of degree 2, subject to the relations (2.7), (2.8), (2.9), (2.10), which we repeat
here:
i2 = 0, i i+1 i = i+1 i i+1 ,
xi i + i xi+1 = 1, i xi + xi+1 i = 1,
xi xj + xj xi = 0 (i 6= j), i j + j i = 0 (|i j| > 1),
xi j + j xi = 0 (i 6= j, j + 1).
We define the odd nilCoxeter ring ON Ca to be the graded subring generated by the i s (this is the
LOT ring of [12]). As a consequence of these relations, ON Ha and ON Ca have natural representations
Z Z Z
on OPola . The -grading on ON Ha induces a /2-grading given by dividing the -grading by 2 and
then reducing mod 2. For f ON Ha we write degs (f ) for the super degree of f .
For each w Sa , choose a reduced expression w = si1 si in terms of simple transpositions
si = (i i + 1). Define
(2.36) w = i1 i .
For w = w0 we make a particular choice of word and re-name the operator,
Da = w0 = 1 (2 1 ) (a1 1 ).
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 11

Since the i satisfy a signed version of the relations of the simple transpositions si , the definition of w
is almost independent of choice of reduced expression for wthe only ambiguity is an overall sign.
For w, w Sa , the formula
(
ww if (ww ) = (w) + (w ),
(2.37) w w =
0 otherwise
is proven just as in the even case [13].
When no confusion will result, if A = (r1 , . . . , ra ) is an a-tuple of integers we define
(2.38) xA = xr11 xraa .
Now it is clear that the sets
(2.39) {xA w }wSa ,ANn and {w xA }wSa ,ANn
N
generate ON Ha (for us, 0 ). In fact, they are linearly independent as well. To prove this, we will
introduce odd Schubert polynomials.
Let a = (a 1, a 2, . . . , 1, 0). In what follows we will make repeated use of the monomial
(2.40) xa = x1a1 x2a2 x1a1 x0a .
For w Sa , define the corresponding odd Schubert polynomial sw OPola by
(2.41) sw (x1 , . . . , xa ) = w1 w0 (xa ),
where w0 is the longest element of Sa . As in the definition of the w , this is independent of choice of
reduced expression for w, up to sign. The degree of sw is 2(w). Equation (2.37) implies
(
swu1 if (wu1 ) = (w) (u),
(2.42) u sw =
0 otherwise.
a
Lemma 2.10. Let e Sa be the identity. Then se = (1)(3) .
Proof. This is a simple calculation, by induction on a. Alternatively, it follows from the graphical
arguments in Proposition 3.6 (whose proof does not depend on the present claim). 
It follows that
for (w) < (u), (xA u )(sw ) = 0,
(
(2.43) xA if w = u1
for (w) = (u), (xA u )(sw ) =
0 otherwise.

Proposition 2.11. There are no linear relations among the images of the {xA w }wSa ,ANn in End(OPola ),
nor among the images of the {w xA }wSa ,ANn . Thus the natural representations of the odd nilCox-
eter and odd nilHecke rings on OPola are faithful, and these rings have graded ranks
rkq (ON Ca ) = q a(a1)/2 [a]!,
(2.44) 1
rkq (ON Ha ) = q a(a1)/2 [a]! .
(1 q 2 )a
Proof. Since ON Ha is finite dimensional in each degree, the relations defining ON Ha imply that it
suffices to prove the Proposition for either one of the two spanning sets; we do so for the set {xA w }.
By equation (2.43),
( a
A (1)(3) xA if u = e,
(2.45) (x u )(se ) =
0 otherwise.
12 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Thus no element xA e is a linear combination of any other spanning set elements. Proceeding by
induction on the Coxeter length of w Sa , suppose that for all v Sa with (v) < (w), no element
xA v is a linear combination of any other spanning set elements. Then by equation (2.43), if xA u is
a linear combination of other terms xA v , all the v must be of shorter Coxeter length than u; but by
induction, there is no such relation. 
Our next goal is to express OPola as a free left and right module over Oa . The following lemma
describes the basis we will use. Our proofs of Lemma 2.12 and Proposition 2.13 closely follow the
proofs of Propositions 2.5.3 and 2.5.5 of [13], respectively.
Lemma 2.12. Let
Ha = spanZ {xA OPola : A a termwise}
(2.46)
= spanZ {xa1 1 xaaa : ai a i for 1 i a}.
Then the odd Schubert polynomials {sw (x)}wSa are an integral basis for Ha .
Proof. It is immediate from the definition of the Schubert polynomials that they are all contained in Ha ,
as the operators i only decrease exponents from xa . Since the set of Schubert polynomials and the
defining basis of Ha both have a! elements, it suffices to show linear independence and unimodularity.
Suppose
X
(2.47) cw sw (x) = 0
wSa

Z
with cw . Then by applying various operators u (as in the proof of Proposition 2.11), we see that
Q
all the cw = 0, proving linear independence over . So any f PHa is a rational linear combination of
Schubert polynomials; but applying u s to an expression f = w cw sw , we see that each cw .  Z
Proposition 2.13. OPola is a free left and right Oa -module of graded rank q a(a1)/2 [a]!, with a ho-
mogeneous basis given by the odd Schubert polynomials {sw }wSa .
Proof. We will show that multiplication
(2.48) Oa Ha OPola
is an isomorphism of abelian groups, realizing OPola as a free left Oa -module; a similar proof shows
it is a free right module with the same basis.
Our first claim is that any f OPola can be expressed in the form
a1
XX
(2.49) f= k,j hk,j xka hk,j Ha1 , k,j Oa .
k=1 j

This being clear for a = 1, we proceed by induction. Expand a given f OPola in powers of xa ,
X X a2
XX
(2.50) f= xka fk = xka i,j,k hi,j,k xia1 ,
k k i=1 j

where hi,j,k Ha2 , i,j,k Oa1 , and fk OPola1 for all i, j, k. Since hi,j,k xia1 Ha1 , we can
re-label and re-index to obtain an expression
X
(2.51) f= xka k,j hk,j hk,j Ha1 , k,j Oa1 .
j,k

By Corollary 2.6, each xka k,j
can be re-written as a sum over terms of the form xka , where Oa

and 1 k a 1. Re-indexing and collecting terms again, this proves the claim.
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 13

The above claim implies surjectivity of the multiplication map. Injectivity follows from a graded
rank count. We have shown that OPola is a free (left and right) Oa -module of graded rank
(2.52) (rkq )Oa (OPola ) = q a(a1)/2 [a]!,
P P
since the right-hand side is equal to w q 2(w) = w q deg(sw (x)) . 
We briefly recall the grading on matrix algebras over graded rings. Let f (q) be a Laurent series in q
and let h(q) be a Laurent polynomial in q. If A is a graded ring with rkq (A) = f (q) and M is a graded
A-module with (rkq )A (M ) = h(q), then
(2.53) rkq (EndA (M )) = f (q)h(q)h(q 1 ).
Corollary 2.14. The natural action of ON Ha on OPola by multiplication and odd divided difference
operators is an isomorphism

(2.54) : ON Ha
= // EndOa (OPola )
= Matqa(a1)/2 [a]! (Oa ).

Proof. Since the ON Ha action is by linearly independent operators (see proof of Proposition 2.11),
Q
Z (and hence ) is injective. Both ON Ha and EndOa (OPola ) have the same graded rank, so
is surjective as well. 
Proposition 2.15. For a 2, the center of ON Ha is the ring of symmetric polynomials in the squared
variables x21 , . . . , x2a . This coincides with the center of Oa .
Proof. Since ON Ha is a matrix ring over Oa , their centers are the same when we view Oa ON Ha
as scalar multiples of the identity. For the duration of this proof, then, we will just refer to cen-
tral (skew) polynomials. First claim: if f is central, then f is a polynomial in the squared variables
x21 , x22 , . . . , x2a . To see this, fix some 1 j a and expand the skew polynomial f as a polynomial
in xj ,
X
f= fk xkj .
k0
Then for each k, 0, let fk, be the degree part of fk , so
X
f= fk, xkj .
k,0

Multiplying this equation by xj on the left and on the right and comparing the results, it follows that
the degree of each fk, must be even. Doing this for each j separately, the first claim follows.
Second claim: if f is a polynomial in the squared variables, then i (f ) = 0 if and only if si (f ) = f .
First, suppose si (f ) = f . Then f is a linear combination of terms which are a product of a factor not
involving xi , xi+1 and a factor of the form x2k 2 2 2k
i xi+1 + xi xi+1 . The operator i annihilates both sorts
of factor, so i (f ) = 0. Conversely, suppose i (f ) = 0; without loss of generality suppose f is of
homogeneous degree. Expand
X
f= fk, x2k 2
i xi+1 ,
k,0
where fk, is a polynomial in the variables xj for j 6= i, i + 1. So
X
0 = i (f ) = fk, i (x2k 2
i xi+1 )
k,0
X  
2(k) 2(k)
= (x2i x2i+1 ) fk, i (xi ) + f,k i (xi+1 ) .
k,0

By decreasing induction on |k |, this implies that fk, = f,k for all k, , that is, si (f ) = f .
14 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

We now use the second claim to show that a skew polynomial f is central if and only if it is a sym-
metric polynomial in the squared variables x21 , . . . , x2a . By the second claim, f a symmetric polynomial
in the squared variables if and only if it is a polynomial in the squared variables and i (f ) = 0 for all
i. For such an f ,
(i f f i )(g) = i (f )g + si (f )i (g) f i (g) = 0,
so f commutes with all divided difference operators. So f Z(ON Ha ) if f is a symmetric polynomial
in the squared variables. Conversely, using the above observations, it is easy to see that all symmetric
polynomials in the squared variables are in Z(Oa ). 
2.2.1. Odd symmetrization.
Lemma 2.16. The longest element of Sa acts on odd elementary polynomials as
k a1
(2.55) (k )w0 = (1)(2)+k( 2 ) k .
In particular, (Oa )w0 = Oa .
Proof. The set of monomials appearing in k is unchanged by w0 , so we need only consider the in-
curred sign. The action on monomials is
a
xi1 xik 7 (1)k( 2) xa+1i1 xa+1ik .
The sign appears because elementary transpositions act on variables xi with a minus sign. Next, a sign
k
of (1)(2) is incurred in sorting the right hand side into ascending order. Finally, remember that the
monomials in k appear with tildes, x ei = (1)i1 xi . The sign difference between removing the tildes
on the left monomial and adding them in on the right monomial is (1)k(a1) . Putting all these signs
together, the sign is as described in the statement of the lemma. Since products of odd elementary
polynomials are a basis of Oa , it follows that (Oa )w0 = Oa . 
Remark 2.17. For a > 2 it is easy to see that given sj Sa , the action by sj does not preserve the ring
of odd symmetric functions. For example
s1 (1 (e
x1 , x
e2 , x
e3 )) = x e1 x
e2 + x e3 ,
which is not odd symmetric.
Recall our fixed choice of Coxeter word for the longest Weyl group element,
(2.56) Da = w0 = 1 (2 1 ) (a1 a2 1 ).
Z
A useful -linear map is
S : OPola Oa
(2.57) a w0
f 7 (1)(3) Da (f xa ) ,
which we call odd symmetrization. The name comes from the fact that, as we will prove in this section,
S is the projection operator from OPola onto its lowest-degree indecomposable summand, the subring
Oa . In order to prove this, we first establish a few lemmas.
A word w = si1 sir in the symmetric group Sa can act on a skew polynomial f in two ways:
act by i1 ir , as an element of the odd nilCoxeter ring,
act by si1 sir , as an element of Sa (equation (2.2)).
One way to hybridize and keep track of these two actions is to equip w with a function : {1, 2, . . . , r}
{0, 1} and to say (
sij if (j) = 0
the simple transposition sij acts as
ij if (j) = 1.
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 15

We will refer to the resulting operator as the generalized action of the pair (w, ), and denote its action

on a polynomial f by w f . Give such a pair, define its omission word wom to be the sub-word of
w consisting of just those sij such that (j) = 0; that is, the sub-word of w corresponding to those
transpositions which act via Sa rather than the odd nilCoxeter ring. In this language, the Leibniz rule
(2.4) can be generalized,
X
(2.58) w (f g) = i1 ir (f g) = (w f )wom
(g).


(w)
The sum is over all 2 possible choices of . Note that the action of w is a generalized action, while

the action of wom is an action by odd divided difference operators.
Lemma 2.18 (Omission Word Lemma (OWL)). Suppose w is a reduced expression for w0 . For any
as above, either
(1) w f = 0 for all f Oa ,
(2) (j) = 0 for all j = 1, . . . , r, or

(3) the omission word wom is non-reduced.
In order to prove the OWL, we first introduce odd divided difference operators for non-adjacent
transpositions. Notation: let sk, be the transposition of k and in the symmetric group Sa , even if
|k | > 1. For 1 i, j a and i 6= j, define the corresponding odd divided difference operator i,j
by
(
1 if h = i, j,
i,j (xh ) =
(2.59) 0 if h 6= i, j,
i,j (f g) = i,j (f )g + si,j (f )i,j (g).
Note that i,i+1 = i and that i,j is homogeneous of degree 2. It is not true in general that the kernel
of i,j contains Oa (unless, of course, j = i + 1).
Lemma 2.19. (1) For any i 6= j and k 6= , we have
(2.60) i,j sk, + sk, sk, (i,j) = 0
as operators on OPola . By sk, (i, j), we mean the pair obtained by applying the transposition
sk, to i and j. In particular, i,j and sk, anticommute if {i, j} {k, } = .
(2) If {i, j} {k, } = , then
(2.61) i,j k, + k, i,j = 0
as operators on OPola .
Proof. For both statements: first check on xh and then induct using the Leibniz rule. 
Lemma 2.20. Suppose 1 i < j a and f Oa . Then
(2.62) i+1 i+2 j1 i,j (f ) = 0.
Proof. The proof is by a slightly complicated induction. For now, consider all i, j simultaneously.

Step 1: We first prove the lemma in the case f = k , k 0. Let Di,j = i+1 i+2 j1 i,j . A
divided difference operator p,q applied to a monomial xJ = xj1 xj vanishes if and only if xp and
xq occur the same number of times in J. It follows that if the tuple J has no repeated entries, then

Di,j (xJ ) 6= 0 if and only if J contains either: all of i + 1, i + 2, . . . , j but not i, or all of i, i + 1, . . . , j 1

but not j. The terms of k on which Di,j does not vanish match into pairs
x ei+1 x
eI x ej x
eJ , x ei x
eI x ej1 x
eJ ,
16 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

where I and J are tuples whose degrees sum to k (j i). We compute



Di,j (e ei+1 x
xI x ej x
eJ + x ei x
eI x ej1 x
eJ )
= i+1 j1 i,j (e ei+1 x
xI x ej x
eJ + x ei x
eI x ej1 x
eJ )

= i+1 j1 i,j xI xi+1 xj xJ + (1) xI xi xj1 xJ
 
= i+1 j1 (1)|I|+1 xI xi+1 xj1 xJ + (1) (1)|I| xI xi+1 xj1 xJ
= 0.

It follows that Di,j (k ) = 0.
Step 2: We now induct on the degree of f . In each degree, we may assume f is a product k1 kr .
Step 1 covered the base cases r = 1 and degree 1. In suffices, then, to take f = gh, where both g and h
are odd symmetric and have positive degree. We will prove equation (2.62) simultaneously with the
following claim: for 1 j i,
(2.63) i+1 j (sj+1 sj1 si,j (g) j+1 j1 i,j (h)) = 0.
The = j i case of (2.63),
sji+1 sj1 si,j (g) i+1 j1 i,j (h) = 0,
follows from (2.62) in lower degree. Before proceeding, we fix i and induct on j, the base case j = i + 1
being obvious.
Step 3: To prove (2.63) by decreasing induction on (keeping i, j both fixed), we compute
i+1 j (sj+1 sj1 si,j (g) j+1 j1 i,j (h))
= i+1 j1 (sj+1 sj1 si,j i,j (g) j+1 j1 i,j (h)
+ sj sj1 si,j (g) j j1 i,j (h)),
by the Leibniz rule and part 1 of Lemma 2.19. Any one of the operators i+1 , . . . j1 annihilates the
second factor of the first term on the right-hand side by part 2 of Lemma 2.19, so this equals
. . . = sj+1 sj1 si,j i+1 j1 i,j (g) j+1 j1 i,j (h)
+ i+1 j1 (sj sj1 si,j (g) j j1 i,j (h)) .
This vanishes by induction: the first factor of the first term by (2.62) for lower j and the second term
by (2.63) for higher . This proves (2.63) in this degree and for this , hence for all j, .
Step 4: It remains to prove (2.62). We have
i+1 j1 i,j (gh) = i+1 j1 (i,j (g) h) + i+1 j1 (si,j (g) i,j (h)) .
The second term on the right-hand side is zero by (2.63) at = 1. Applying each of i+1 , . . . j1 to the
first term on the right-hand side and using the Leibniz rule, we always get zero for the term in which
the divided difference operator hits h. Therefore this term equals i+1 j1 i,j (g) h, which is zero
by (2.62) in lower degree. This completes the proof. 

Lemma 2.21. For any 1 i a 1, there is a reduced expression for w0 in Sa with si acting first.
(a) (a1)
Proof. This is clear for a = 2, 3. Let w0 denote w0 in Sa , w0 denote the longest word among
(a1)
strands 1, . . . , a 1 in Sa , and w0 denote the longest word among strands 2, . . . , a in Sa . Since
(a) (a1) (a1)
w0 = s1 sa1 w0 = sa1 s1 w0 ,
the lemma follows by induction. 
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 17

Proof of the OWL, Lemma 2.18. We are free to reorder Da up to sign; we choose the ordering Da such
a
that D2 = 1 and Da = a1 1 Da1

. (In fact, Da = (1)(3) Da .) Explicitly,
Da = (a1 1 ) (a1 a2 )a1 .
Fix some and consider the generalized action w f . Let sw0 ,b denote the longest element of the
symmetric group among strands a b + 1, . . . , a. If the rightmost a1 in Da acts with = 1, then any

f is annihilated and we are in Case 1. By induction on b 1, then, suppose the bottom 2b crossings
all act with = 0. The next b crossings are between strand pairs (a b, a b + 1), (a b + 1, a
b + 2), . . . , (a 1, a). Let i be the number of the left strand of any of these crossings which acts with

= 1. If i < a 1 and the (i + 1, i + 2) crossing acts with = 0, then the omission word wom is
non-reduced by Lemma 2.21 (pull a crossing si+1 to the top of sw0 ,b ), so we are in Case 3 and are done.
We are therefore reduced to the case in which once one of these crossings acts with = 1, then so do
all others to the left. That is,
w f = (. . .)a1 ab+ sab+1 sab sw0 ,b (f ).
By part 1 of Lemma 2.19 this equals
(. . .)sab+1 sab sw0 ,b ab+1 ab+1 ab,ab+ (f ),
which vanishes by Lemma 2.20. 

An immediate corollary of the OWL is that for any f Oa and g OPola ,


(2.64) Da (f g) = f w0 Da (g),
since v = 0 for any nonreduced word v.
Corollary 2.22. The odd symmetrization operator S is left Oa -linear and is a projection operator
onto the subring Oa OPola .
Proof. Left Oa -linearity of S follows immediately from equation (2.64). The image of S is inside Oa
due to the presence of the Da in its definition and by Lemma 2.16. This image is all of Oa and S 2 = S
a
because, by the OWL and the calculation Da (xa ) = (1)(3) (see (3.54)),
a a w0
(2.65) S(f xa ) = (1)(3) Da (f xa )w0 = (1)(3) f w0 Da (xa ) =f
for any f Oa . 

We conclude this section with another useful corollary of the OWL.


Corollary 2.23. For any f OPola ,
a
(2.66) Da (f )w0 = (1)(2 ) Da (f w0 ).
Proof. Expand f in the odd Schubert polynomial basis,
X
f= f w sw ,
wSa

where each fw Oa . Since the action of w0 preserves Oa OPola , equation (2.64) implies
X
Da (f )w0 = fw Da (sw )w0 = fw0 Da (xa )w0 ,
w
X
w0
Da (f )= Da ((fw )w0 (sw )w0 ) = fw0 Da ((xa )w0 ).
w
18 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

The last equality on each line is by degree reasons: the only Schubert polynomial of high enough
degree not to be annihilated by Da is sw0 . The corollary then follows from the relations
a a+1
Da (xa ) = (1)(3) , Da ((xa )w0 ) = (1)( 3 ) ,
proved in Section 3.3. 

2.3. Odd Schur polynomials.

2.3.1. Partitions. By a partition = (1 , 2 , . . .) we mean aP nonincreasing sequence of nonnegative


integers 1 2 . . . 0 with finite sum. We define || = ai=1 i when i = 0 for i > a. For short,
we represent repeated entries with exponents: for example, (. . . , 23 , . . .) means (. . . , 2, 2, 2, . . .).
By P (a, b) we denote the set of all partitions with at most a parts (that is, with a+1 = 0) such
that 1 b. That is, P (a, b) consists of partitions whose corresponding Young diagram fits into a box
of size a b. Moreover, the set of all partitions with at most a parts (that is, the set P (a, )) we denote
simply by P (a). P (0) = {} is the set of all partitions with at most 0 parts, so that P (0) contains only
the empty partition.

The cardinality of P (a, b) is a+b
a . We call
X  
a+b
(2.67) |P (a, b)|q := q 2||ab =
a
P (a,b)

the q-cardinality of P (a, b). The last term in the above equations is the balanced q-binomial.
The dual (or conjugate) partition of , denoted = (1 , 2 , . . .) with j = #{i|i j}, has Young
diagram given by reflecting the Young diagram of along the diagonal. For a partition P (a, b)
we define the complementary partition c = (b a , . . . , b 2 , b 1 ). To be more explicit we let
K = (ba ) P (a, b) and write K in place of c to emphasize that P (a, b). Finally we define
b := c . Note that and
b belong to P (b, a) and c to P (a, b).

2.3.2. Odd Schur polynomials. In the even case, one definition of the Schur polynomial corresponding
to a partition of length at most a is
(2.68) s (x1 , . . . , xa ) = w0 (xa + ) = w0 (xa1+
1
1 a2+2
x2 . . . x
a ).
a

In the odd case, we must be careful about the ordering of the terms in the above expression.
Definition 2.24. The odd Schur polynomial corresponding to a partition of length at most a is the
element of OPola given by odd-symmetrizing the monomial x ,
a
(2.69) s (x1 , . . . , xa ) = S(x ) = (1)(3) Da (x xa )w0 ,
see also equation (2.57). As usual, x = x a
1 xa .
1

Since S is the projection onto the subring of odd symmetric polynomials, s is odd symmetric.

2.3.3. The odd Pieri rule. In what follows, we will refer to the normal ordering of a monomial xi1 xi2 xir ;
this is the monomial obtained by sorting the subscripts into non-decreasing order (and not introduc-
ing a sign).
By analogy with the even case, we have the following.
Lemma 2.25. For 1 k a, we have
1
(2.70) s(1k ) (x1 , . . . , xa ) = (1) 2 k(k1) k (x1 , . . . , xa ).
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 19

Proof. Let i1 < < ik be a subset of {1, . . . , a}. Unless i1 = 1, . . . , ik = k, the normal ordering of the
monomial x eik xa will have a factor of the form xj xj+1 somewhere, so that Da (e
ei1 x eik xa ) =
xi1 x
eik xa ) = 0. Hence, using odd-symmetrization and the definitions of k and of s(1k ) ,
xi1 x
j (e
1
eik ) = (1) 2 k(k1) s(1k ) .
xi1 x
k = S(k ) = S(e


For a partition , let m denote the partition whose Young diagram is obtained from that of by
removing the m-th and all lower rows; let m
be likewise, removing the first through m-th rows.

Proposition 2.26 (Odd Pieri rule). Let be a partition and let 1 k a. Then in Oa ,

X i1 i
(1)| |
+...+ k
(2.71) s s(1k ) = s ,

where the sum is over all partitions with Young diagram obtained from that of by adding one box
each to the rows i1 < < ik .
Note that mod 2, this is the usual Pieri rule. The diagrams described in the statement of the
Proposition are precisely the diagrams obtained from by what is called adding a vertical strip of
length k; but we also need to know which rows we are adding to in order to get the correct sign.
k
Proof. Identifying s(1k ) with (1)(2) k by Lemma 2.25, we compute:
a k
s s(1k ) (2.69)
(1)(3) Da (x xa )w0 (1)(2) k
a a1
(1)(3 )+k( 2 ) Da (x xa k )w0
(2.55),(2.4)

a a1 X
(2.21)
(1)(3)+k( 2 ) Da (x xa x eik )w0 .
ei1 x
i1 <<ik

a1
Commuting the factor x eik past x a and
ei1 x
normal ordering it with x cancels the factor (1)k( 2 )
i1 i
and introduces a factor of (1)| |
+...+ k
to each term. So we appear to have the desired equation
(2.71), except that there are certain terms where is not non-decreasing, that is, does not correspond
to a Young diagram (the term involving xi1 xik corresponds to adding one box to in each of rows
i1 , . . . , ik ). This occurs when is a composition but not a partition. A term will fail to correspond to a
Young diagram if and only if the resulting monomial x xa xi1 xik , when normal ordered, has two
adjacent exponents equal:

x xa xi1 xik = . . . xm m
j xj+1 . . . for some j, m if and
only if is a composition but not a partition.

Such a monomial is sent to zero by j . Re-ordering Da = w0 so as to have j act first, we see the term
does not contribute to equation (2.71). 

The above proves the usual (even) Pieri rule as well; just drop all the signs.
Remark 2.27. Assume for a moment that the Schur functions of this paper agree with the Schur func-
tions of [1] (see Conjecture 5.3 below). Combining Proposition 2.26 with the automorphism 1 2 and
the anti-involution 3 of O which were introduced in Section 2.3 of [1], we have three more forms of
20 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

the Pieri rule:



P i1 i
(1)| |
+...+ k
(2.72) (1)(w ) s s(k) = (1)
(w )
s ,

P i i
s(1k ) s = (1)| |
1 +...+ k

(2.73) s ,

P i1 i
(1)(w ) s(k) s = (1)(w ) (1)| |
+...+ k
(2.74) s .

For definitions of the signs (1)(w ) and , see [1]. The second sum (respectively first and third
sums) is over all obtained from by adding a vertical (respectively horizontal) strip; the integers
i1 , . . . , ik are the rows (respectively columns) to which a box was added.
Remark 2.28. Again assuming Conjecture 5.3, no nave analogue of the Jacobi-Trudi relation can hold.
The usual (even) Jacobi-Trudi relation is
(2.75) s = det(i +ji ) = det(hi +ji ).
A minimal example of the problem is = (2, 2) (we order the terms of our determinants top-to-
bottom, though this isnt essential):
det(i +ji ) = 22 31
(2.76) det(hi +ji ) = h22 h31 = (22 2211 + 1111 ) (31 111 )
s22 = 22 + 31 24 .
Since 23 = 12 + 21 , it seems some clever rearrangement of terms and of signs could make the two
determinants equal each other. But neither determinant along with any possible re-arrangement of
signs can lead to a term involving 4 . That is, 4 is not in the subring generated by h1 , h2 , h3 , 1 , 2 , 3 .

3. G RAPHICAL CALCULUS

3.1. Graphical calculus for the odd nilHecke algebra. We find it convenient to use a graphical calcu-
lus to represent elements in ON Ha . The diagrammatic representation of elements in ON Ha is given
by braid-like dotted diagrams D equipped with the height Morse function h : D , such that R
h(g1 ) 6= h(g2 ) for any generators g1 , g2 that appear in the diagram.
We write

(3.1) ... ... := 1 ON Ha

with a total of a strands. The polynomial generators can be written as

(3.2) ... ... := xr

with the dot positioned on the r-th strand counting from the left, and

(3.3) ... ... := r ,

with the crossing interchanging the rth and (r + 1)-st strands.


THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 21

In the diagrammatic notation multiplication is given by stacking diagrams on top of each other,
left-to-right becoming top-to-bottom. Relations in the odd nilHecke ring acquire a graphical inter-
pretation. For example, the equalities r xr + xr+1 r = 1 = xr r + r xr+1 become diagrammatic
identities:


(3.4) + = = +

and the relation r r = 0 becomes

(3.5) = 0.

The relation r r+1 r = r+1 r r+1 is depicted as

(3.6) = .

The remaining relations in the nilHecke ring can be encoded by the requirement that the diagrams
super commute under braid-like isotopies:


(3.7) ... ... ... + ... ... ... = 0,


(3.8) ... ... ... + ... ... ... = 0,


(3.9) ... ... ... + ... ... ... = 0,

(3.10) ... ... ... + ... ... ... = 0.


22 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

3.2. Box notation for odd thin calculus. To simplify diagrams, write

:= = 1 ON Ha ,
a
{z } |
a
where we will omit the label a if it appears in a coupon as below. The operator Da is represented as

(3.11) Da = Da = Da := Da1 = Da1 (a1 . . . 2 1 ).

Next, let

a1
a2
a
x = a := 2 = x1a1 x2a2 . . . x2a2 xa1 .

Z
Recall that the super degree of an element of ON Ha is the -degree divided by 2. In particular, we
have
 
a a
(3.12) degs (Da ) degs (x ) (mod 2).
2
Throughout this section we make frequent use of the trivial binomial identities
  X a1
a a(a 1)
(3.13) = j= ,
2 j=1
2
  Xa1   j
a2 X
X
a j a(a 1)(a 2)
(3.14) = = j= ,
3 j=1
2 j=1
6
=1
  Xa Xk Xj
a+3 a(a + 1)(a + 2)(a + 3)
(3.15) = = .
4 j=1
24
k=1 =1

3.2.1. 0-Hecke crossings. We will use a diagrammatic shorthand


(3.16) r := := = xr r .

It is easy to check that the elements r for 1 r a 1 generate a copy of the 0-Hecke algebra:

2
(3.17) r = = = r,
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 23

(3.18) r r+1 r = = = r+1 r r+1 ,

and, since the element r has degree zero, we have


r s = sr if |r s| > 1.
Let w0 denote a reduced word presentation of the longest element in Sa and define

ea = ea = ea = w0 .

By the 0-Hecke relations above it is clear that w0 does not depend on the choice of reduced word
presentation w0 .
3.2.2. Relations for Da .
Lemma 3.1 (Crossing slide lemma).
(3.19) (a2 . . . 2 1 )(a1 . . . 2 1 ) = (a1 a2 . . . 2 1 )(a1 . . . 2 )

Proof. The proof is by induction on the number of strands. The base case of a = 3 strands follows
from the odd nilHecke relation (3.6). Assume the result holds for a 1 strands. Then

(3.6)
(3.20) = (1)a3 (1)a3

so that using the induction hypothesis the left side of (3.19) can be written as

(3.21) = (1)a3 = .



24 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Lemma 3.2 (Alternative definition of Da ). The element Da defined in (3.11) can also be inductively
defined by the equation

(3.22) Da = Da1

Proof. The proof is by induction with the base case following from (3.6). The left hand side can be
re-written using the definition of Da

Da2 Da2
(3.11)
(3.23) Da1 = = Da1 ,

where the second to last equality follows by repeatedly applying (3.6), and the last equality follows
from the induction hypothesis. 
Lemma 3.3 (Da slide). For a 1 the equation

a
(3.24) Da = (1)(3) Da

holds in ON Ha .
Proof. The proof is by induction on the number of strands. The base case is trivial, so we assume the
result holds for up to a 1 strands.

Da1 Da1
(3.11) (3.19)
(3.25) Da

a1
Da1 a1 a1
(3.26) = (1)( 2 ) = (1)( 2 )+( 3 ) ,
Da1

where we used the induction hypothesis in the last step. The result follows using the definition (3.11)
of Da and the identity
     
a1 a1 a
(3.27) + = .
2 3 3
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 25


3.2.3. Relations involving the 0-Hecke generators. It is a simple calculation to see that

(3.28) =

so that the relation

(3.29) = ea
ea

holds.
Remark 3.4. Note that an analogous relation

(3.30) 6= ea
ea

does not hold for sliding a projector ea from the bottom left to the top right through a strand. This is
because the reflection of equation (3.28) through a horizontal axis is false.
It follows from (3.4) that

(3.31) = .

It will be convenient to express elements ea in terms of one of the bases (2.39).


Proposition 3.5 (Projector in standard basis).

a
a
(3.32) ea = (1)(3)
Da

Proof. We prove the result by induction. The base case is trivial. Assume the result follows for ea1 .
From the definition of ea we have

(3.33) ea = ea = ea1
26 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Sliding the dots from the 0-Hecke generators up to the top of the diagram we can write


Pa2
j

(3.34) = (1) j=1 =

so that



(3.29)
(3.35) ea1 = ea1
ea1

Using the induction hypothesis we write ea1 in standard form



a1
(3.36) ... = (1)( 3 )
a1

Da1

then slide the dots into the correct position using the identity


Pa2 Pa2 a1
(3.37) = (1) j=1 =j
= (1)( 2 )
a1 a a

Pa2
which follows by sliding the lower right most dot down past each of the j=1 j dots, then the second
dot from the right, and so on. The Proposition follows from the inductive definition (3.11) of Da and
the binomial identity (3.27). 

The following proposition shows that the elements ea are idempotents in ON Ha .


Proposition 3.6.
The diagrammatic identities

Da
1) = Da ,


THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 27

Da
2) = Da ,
ea

ea
3) = ea
ea

hold in ON Ha .

Proof. Part 1) follows from the computation



Da1 Da1
Da (3.11) (3.31)

(3.24) a1 Da1 a1
(1)( 3 ) = (1)( 3 ) Da1 ,

where the last equality follows by the induction hypothesis. This proves part 1) using equation (3.24)
to slide Da1 back through the line.
For the second claim observe that

(3.38)

Da1 Da1
Da1
Da
Da (3.29) ea1 (3.22)
= = ,
ea ea1


28 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

where the third equality follows by induction. The second claim follows from part 1). Part 3) follows
from 2) using Proposition 3.5 since
a a
(3.39) ea ea = (1)(3) xa Da ea = (1)(3) xa Da = ea .

One can also prove the following equalities from the 0-Hecke relations.

a c
eb ea+b+c
(3.40) ea+b+c = =
ea+b+c eb
a c

3.2.4. Crossings and projectors. Just as a vertical lines can be represented by a single line labelled a, we
denote a specific choice of crossings of a strands with b strands as follows:
a b
z }| { z }| {

a b
(3.41) =

The first of the b strands crosses all of the a strands, then the second of the b strands crosses all of the
a strands, and so on. One can easily verify that
a b c a b c
a b+c a+b c
c
(3.42) = = (1)ab( )
2

The respective heights of ea and eb in the diagram

eb
ea eb
(3.43) ea =

b a
b a
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 29

are not relevant as these elements have even super-degrees and commute with each other. The equa-
tion

ea eb
ea eb
(3.44) =

ea+b
b a

holds.
Proposition 3.7.

Da Db
a b
(3.45) = Da+b = (1)(2)(2)
Db Da

b a b a

Proof. The second equality follows immediately from the first. To prove the first, observe that

Da Da Da
Da
Db1 Db1 Db1
b1 b1
= = (1)a( 2 ) = (1)a(2 )
Db

b a

where the sign in the second equality arises from sliding each of the b 1 crossings from Db down.
The third equality follows by repeated application of the Crossing Slide Lemma 3.1. Sliding the Db1
down past the a crossings and using the inductive definition of Da+1 shows that

Da Da+1

= ,
Db Db1

so that the result holds by induction. 


30 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

3.3. Automorphisms of the odd nilHecke algebra. Let : ON Ha ON Ha denote the automor-
phism given by reflecting diagrams across the vertical axis. Let : ON Ha ON Haop denote the
anti-automorphism given by reflecting diagrams across the horizontal axis. These automorphisms
commute with each other.
Below we collect the effect of applying these automorphisms to the elements xa , Da , and ea . One
can check that
(3.46) xa = x1a1 x2a2 . . . x1a1 x0a
a2
(3.47) (xa ) = xaa1 xa1 . . . x12 x01
(3.48) (xa ) = x0a x1a1 . . . x2a2 x1a1
a1 a1
(3.49) (xa ) = x01 x12 . . . xa2 xa .
These monomials are related by the equations
a a
(3.50) xa = (1)(4 ) (xa ), (xa ) = (1)(4) (xa ).
Similarly,
Da := 1 (2 1 ) . . . (a2 . . . 1 )(a1 . . . 1 )
(Da ) = a1 (a2 a1 )(2 . . . a1 )(1 . . . a1 )
(Da ) = (1 . . . a1 )(1 . . . a2 ) . . . (1 2 )1
(Da ) = (a1 . . . 1 )(a1 . . . 2 ) . . . (a1 a2 )a1 ,
and it follows from Lemma 3.2 that
a1
(3.51) Da = (Da ) = (1)( 4 ) (Da ) = (1)(a1
4 ) (D ).
a

We have shown in Proposition 3.5 that


a+1
(3.52) ea = (1)( 4 )+(a4 ) xa D = (1)(a3) xa D
a a

It is also worth while to write this equation in another form,


a+1
(3.53) ea = (1)( 4 ) (xa )Da .
It follows from Proposition 3.6 that
a a+1
(3.54) Da (xa ) = (1)(3) and Da ((xa )) = (1)( 4 ) .

4. E XTENDED GRAPHICAL CALCULUS FOR THE ODD NIL H ECKE ALGEBRA

4.1. Thick calculus. In this section we introduce an extension of the graphical calculus from the pre-
vious section.

4.1.1. Splitters. Define

:= ea
a

The notation is consistent, since ea is an idempotent, so that cutting a thick line into two pieces and
converting each of them into an ea box results in the same element of the odd nilHecke ring.
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 31

The following diagrams will be referred to as splitters or splitter diagrams.

a+b a+b
a b ea eb
(4.1) := := ea+b
a b
a b
a+b b a
Define a thick crossing by

b a
eb ea
eb ea
(3.44)
(4.2) := =

a b ea+b
a b a b

Proposition 4.1 (Associativity of splitters).

a b c c
a b c
(4.3) = (1)ab(2)

a+b+c a+b+c

a+b+c a+b+c

(4.4) =
a b c a b c

Proof. The proof of the first claim is a direct calculation making use of Proposition 3.5, equation 3.42,
and the identity

Da Da
Db Db
=
Dc Dc

The second claim follows from equation (3.40). 


The apparent asymmetry between a, c in the splitter associativity relation is a consequence of the
asymmetry in our choice of crossing between multiple strands, equation (3.41).
32 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Proposition 4.2. For a, b, c 0

a b+c
a
b
(4.5) =

c a+b
c a+b

b+c a
a
b a
(4.6) = (1)(2)bc

a+b c
a+b c

Proof. The proof uses equation 3.42 and equation (3.40). 

4.1.2. Dotted thick strands.

Definition 4.3. For any f Oa write

(4.7) := ea f ea .
f

For any polynomial f OPola observe that Da f Da = Da (f )Da in ON Ha . Now suppose f is odd
symmetric. It follows that
a
(4.8) ea f ea = xa Da f xa Da = xa Da (f xa )Da = (1)(3) xa f w0 Da .

Then by the corollary to the OWL (2.64),


a
(4.9) ea f ea = (1)(3) xa f w0 Da = ea f.

It follows that

(4.10) ea gf ea = (ea gea )(ea f ea ),


THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 33

so we have the diagrammatic identity

g
(4.11) = gf
f

a a

whenever f, g Oa .

4.2. Explosions. Sometimes it is convenient to split thick edges into thin edges. Define exploders

a
(4.12) = Da = ea
a

The associativity rules for exploded splitters follow immediately from the associativity rules for
general splitters derived above. Equation (3.45) gives the following diagrammatic identities.
(4.13)

a b
= = (1)(2)(2)
a b a b
a+b a+b a+b
Using equation (3.44) the equation below follows.

(4.14) = =

a+b a+b a+b


It is straightforward to show using the properties of the projectors ea that the following relations
hold,
a+b a+b a+b
a b a b
(4.15) = =


34 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

a+b a+b a+b

(4.16) = =

4.2.1. Exploding Schur polynomials. Recall that given a partition = (1 , . . . , a ) P (a) we have
defined the odd Schur polynomial s (x1 , . . . , xa ) as
a w0
(4.17) s (x1 , . . . , xa ) := (1)(3 ) Da (x xa ) .
It will be convenient to normal order the variables using the notation

(4.18) xa + = xa1+
1
1 a2+2
x2 . . . x
a .
a

One can show


a w0
(4.19) s (x1 , . . . , xa ) := (1) Da (xa + ) ,
where for P (a) we write
    X a  
a a aj +1
(4.20) a := + || + j .
3 2 j=1
2
a
That is, (1)(3) a is the sign needed to normal order the product x xa . For example, if = (1r ) then
j = 1 for 1 j r and
   
a a a r(3a2 3ar + r2 1)
(1r ) = +r + .
3 2 6
Observe that
(4.9) a
(1)(3 )+ xa Da (x+a )Da
a a
(4.21) ea s ea = (1) ea Da (x+a )w0 ea
a (3.52)
= (1)(3)+ xa Da x+a Da
a a
(4.22) (1) ea x+a Da .
This implies the diagrammatic identity

1
a 2
(4.23) = (1)
s a1
a
a

where j = j + a j.
Notice that writing such an equation would not be possible if we did not include the action of the
longest symmetric group element w0 in the definition of s .

4.3. Orthogonal idempotents.


THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 35

4.3.1. Some helpful lemmas.


Lemma 4.4 (Shuffle Lemma).

(1)m i (xim+k xm
i+1 ) if k = 1,
m m+k
i (xi xi+1 ) = i (xm+k
i xm
i+1 ) if k is even,

(1)m i (xm+k
i xm
i+1 ) (x
i i
m+k1 m+1
x m m+1 m+k1
i+1 ) + (1) i (xi xi+1 ) if k is odd and k 3.
Proof. Using the Leibniz rule for odd divided difference operators we have
m+k
(4.24) i (xm m m k m m si k m m k
i xi+1 ) = i (xi xi+1 )xi+1 + (xi xi+1 ) i (xi+1 ) = (xi+1 xi )i (xi+1 ).

eki + x
The sum x eki+1 is annihilated by the divided difference operator i when k = 1 and when k is
even. When k is odd and k 3, the sum
(4.25) eki + x
x eik1 x
eki+1 + x ei+1 + x
ei xk1
ei+1 ,
is annihilated by i . Using these facts the result follows. 

Note that for k odd and k 3 the relations above imply

(1)m i (xm+k
i xm m+k1 m+1
i+1 ) i (xi xm+1
xi+1 ) + (1)m i (ei
m+k1
xi+1 )
k
2  
X
(4.26) = (1)m i (xm+k
i xm
i+1 ) 2 (1)m(j+1) i xim+kj xm+j
i+1
j=1

so that
k
2  
X
(4.27) i (xm m+k
i xi+1 ) = (1) m
i (xm+k
i xm
i+1 ) 2 (1)m(j+1) i xim+kj xm+j
i+1 .
j=1

We call the relation above the big odd shuffle.


Proposition 4.5 (Vanishing of matched exponent in a partial staircase).
Let a m 2 and a (m 1) p a 1. Then
a(m2) a(m1) p
(4.28) Dm (x1a1 x2a2 . . . xm2 xm1 xm ) = 0.
Proof. We prove the proposition by induction on m. If m = 2 we are done since then p = a 1 and
1 (x1a1 x2a1 ) = 0. Assume by induction that
a(m2) p
Dm1 (x1a1 x2a2 . . . xm2 xm1 ) =0
for a (m 2) p a 1.
Suppose that p = a j for some 1 j m 1 in the expression
a(m1) p
(4.29) Dm (x1a1 x2a2 . . . xm1 xm ).
a(m1) a(m1)
If p = a(m1) then we are done since Dm = Dm1 (1 . . . m1 ) and m1 (x1a1 . . . xm1 xm ) =
0. Assume then that a (m 2) p a 1 so that
a(m1) p a(m1) p
Dm (x1a1 . . . xm1 xm ) = Dm1 (1 . . . m1 )(x1a1 . . . xm1 xm )
a(m2) a(m1) p
= Dm1 (1 . . . m2 )(x1a1 . . . xm2 m1 (xm1 xm )).
36 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Let q = p (a (m 1)). Lemma 4.4 together with the induction hypothesis show that the expression
vanishes whenever q = 1 or q is even. When q is odd and q 3 we write
a(m1) p a(m2) p
Dm (x1a1 . . . xm1 xm ) = (1)a(m1) Dm (x1a1 . . . xm2 a(m1)
xm1 xm )
a(m2) p1 a(m2) a(m2) a(m2) p1
Dm (x1a1 . . . xm2 xm1 xm ) + (1)a(m1) Dm (x1a1 . . . xm2 xm1 xm ).
The induction hypothesis shows that each of these three terms is zero, unless p = a (m 2) in
which case the induction hypothesis does not apply to the second term. But in this case, the term still
vanishes since the exponent of xm is also a (m 2). 
Proposition 4.6 (Adding a step to a full staircase). The equation
a1
(4.30) Da (x1a2 x2a3 . . . x0a1 xaa1 ) = (1)( 3 ) D (xa1 xa2 . . . x1 x0 ) = (1)(a1
2 )
a 1 2 a1 a
holds in ON Ha .
Proof. The proof follows by shuffling the exponent of xa left using Lemma 4.4 and the big odd shuffle.
Additional terms that arise from the big odd shuffle vanish by Proposition 4.5. In this calculation, the
a2
sign is (1) 2 when a is even, and (a 1) when a is odd. 
Proposition 4.7 (Reordering a reverse staircase).
a
(4.31) Da (x01 x12 . . . xa1 xa ) = (1)(4) Da (x1a1 x2a2 . . . x1a1 x0a ).
a2 a1

Proof. The proposition follows by induction from Proposition 4.6 together with some simplification
of the exponent of (1). 
Lemma 4.8. Let P (a, b) and P (b, a) be two partitions. Then
 
1+ a b 1+b1 b2+2 b1+1
(4.32) Da+b xa1+
1
1 a2+2
x2 . . . xa1 a1 x
a xa+1 xa+2 . . . xa+b1 xa+b =0
unless || + || = ab.

Proof. If the total degree deg(a )+deg(b )+||+|| of the dots is less than the degree a+b
2 of a+b , then
the total degree of the left-hand side of equation 4.33
 is negative, and the expression must therefore
be zero. If the degree is strictly greater than a+b2 , then using the Shuffle Lemma we can reorder
the exponents so that they are decreasing, acquiring additional terms each time the big odd slide is
performed. The requirements on and imply b 1 a and a 1 a 1. Since
the total degree of the dots is greater than a+b2 it follows that there must be a repeated exponent in
each term that arises from the shuffling process. Since the action of Da+b annihilates adjacent repeated
exponents, all of these terms must vanish. 
Lemma 4.9. Let P (a, b) and P (b, a). Then
  a+b
= ,b(1)()+( 3 ) ,
1+ a b 1+b1 b2+2 b1+1
(4.33) Da+b xa1+
1
1 a2+2
x2 . . . xa1 a1 x
a xa+1 xa+2 . . . xa+b1 xa+b
where
b1 
X 
bj + j
(4.34) () := .
j=0
3

Proof. By the previous lemma it suffices to consider the case when || + || = ab. The condition that
= b implies that the only time (4.33) is nonzero is when each of the exponents is unique, so that all
elements of the set {0, 1, . . . , a + b 1} occur as exponents. Note that the partition requirements for
and imply that there are chains of strict inequalities of exponents a 1 + 1 > a 2 + 2 > > a
with the first a variables and b < b1 + 1 < < 1 + b 1 with the last b variables.
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 37

We show that if an exponent is ever repeated in the left hand side of (4.33) then the expression
vanishes. We will prove this in the case a < b (the case a > b is similar; and if a = b , then the
expression contains repeated adjacent exponents and is therefore zero). In this case, the exponents of
xab 1 through xa must form a staircase
xa
b 1
xb 2 . . . x1a1 x0a ,
b 1 ab +1

or else the expression vanishes. To see this suppose that xaj for 0 j b 1 is the first j where the
exponent f of xfaj is not part of a reverse staircase, that is, f > j.
Since b > f all of the exponents of the last b strands must be larger than f , and all the variables
before xaj must also be larger than f . Hence the exponent j does not occur in the expression
 
j+1+ (j1)
Da+b xa1+
1
1
. . . xaj1 a(j+1) xfaj xa(j1) . . . x1a1 x0a xa+1
b b1+1
. . . xa+b .
Using the Shuffle Lemma and the big odd slide we reorder the exponents above so that they are
decreasing. Each of the resulting terms will have one element of the set {0, 1, . . . , j 1} missing
since the shuffling procedure takes pairs of exponents (t1 , t1 + k) with k > 0 and either shuffles them
(t1 + k, t1 ) or else creates terms (t1 + k , t1 + ) for 1 k2 . Observe that if a missing exponent
appears from a big odd slide (t1 +k , t1 +), then the lower exponent t1 has been removed. If t1 then
appears from a subsequent big odd slide, then again a lower exponent will have had to be removed.
Hence, at least one element in the set {0, 1, . . . , j 1} is missing in each term arising in the shuffling
procedure. For degree reasons, if one exponent in the set {0, 1, . . . , a + b 1} is missing, then at least
one exponent must be repeated, hence the expression contains adjacent repeated exponents and must
therefore vanish.
Thus, if our expression is nonvanishing, it must be of the form
 b +(ab ) b 1

Da+b xa1+
1
1
. . . xab xa(b 1) xa(
b 2
b 2) . . . x1
a1 x0 b
a xa+1 . . . xb1+1
a+b .
Now use Proposition 4.6 to slide the exponent b of xa+1 left to add a step to the staircase giving the
expression
 b +(ab ) b

1+
(1)(b ) Da+b xa1+
1
1
. . . xab xa(b 1) xa(
b 1
b 2)
b1+1
. . . x2a1 x1a x0a+1 xa+2(b1) . . . xa+b .
If b1 = b , then slide the exponent b1 + 1 = b + 1 of xa+2 left adding a step to the staircase
with sign (1)(b1 +1) . If this exponent b + 1 is repeated then it must be repeated as the exponent
b + (ab ) of xab , then the expression is zero. Otherwise we can assume that b1 + 1 = b + 1 is
not a repeated exponent and that b + (ab ) > b1 + 1.
If b1 > b so that b1 = b + g + 1 for some g 0, then the staircase must continue
b +g
xa( b +g)
. . . xa(
b +2
xb +1 xb xb 1
b +1) ab a(b 1) a(b 2)
. . . x1a x0a+1
or else the expression vanishes. If the staircase did not continue, then one of the exponents in the set
{b + 1, b + 2 . . . , b + g} would not occur in the expression, and arguing as above one can show
that all terms resulting from shuffling the expression to decreasing order must be missing at least one
exponent in the set {0, 1, . . . , b + g} and must therefore vanish.
Continuing in this way, it follows that if any exponent in the expression is repeated then the expres-
sion is zero. Otherwise, all the exponents of the last b variables can be slid through staircases so that
a+b
the resulting expression has strictly decreasing exponents. Since Da+b (xa+b ) = (1)( 3 ) the result
follows. 
Definition 4.10. Define dual Schur functions as
a a 1+a1
sb (x1 , . . . , xa ) := (1) Da ((xa + ))w0 = (1) Da (x
1 x2
a
. . . xa1+
a
1 w0
) .
38 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

a a 1+a1
It easy to see that ea sb ea = (1) ea (xa + )Da = (1) ea x
1 x2
a
. . . xa1+
a
1
Da .
Proposition 4.11. Let P (a, b) and P (b, a) be two partitions. Then






a
|+(2b))+(b
)+(a+b
(1) +b +(2)(|b 3 )
a b
b if =
b,
s
(4.35) =

a+b
sb

a



0 otherwise,
a+b
where () is as in Lemma 4.9 and
b is as in Subsection 2.3.
Proof. From the definitions we have

s1
s2
sa
b
s a b
(4.36) = (1) +

sb rb
rb1
a r1

a+b

a+b
where sj = aj+j and rj = bj+j . After sliding the splitters past dot terms and using associativity
for exploded splitters in equations (4.13) and (4.15), the left-hand side of the above equation becomes

s1
s2

a b sa
a
(1)(2)(||+(2))
b
= (1) + rb
.
rb1
r1

a+b
The result follows by Lemma 4.9. 
Remark 4.12. The ordering of the xs in the above equation is critical for the above to work. Notice
that this is a different ordering from what appears in [8]. To see why it is necessary consider the
example with a = 2, b = 2, = (2, 0), and = (2, 0). In this case, D4 (x31 x33 ) 6= 0, while D4 (x31 x34 ) = 0.
4.4. The odd nilHecke algebra as a matrix algebra. In this section we find it convenient to work with
odd elementary symmetric functions k rather than their Schur analogues s(1k ) . Recall from (2.70) that
k
the two are related by s k = (1)(2) .
(1 ) k
Let Sq(a) denote the set of sequences
(4.37) Sq(a) := { = 1 . . . a1 | 0 , = 1, 2, . . . , a 1}.
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 39

This set has size |Sq(a)| = a!. For Sq(a) let

2
a
a
(4.38) = a2 = (1)(3 )
ba1
b
a1 b a2

b1
r

Lemma 4.13.

(4.39) = , = , ea
a

for , Sq(a).

Proof. Consider the composite

a2a2
a1a1

a 22
(4.40) = (1)(3) 1
11

a2

a1

a
40 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Repeatedly apply the equality


,1
( )+X(1
(4.41) = , (1) 2 )
= , (1)(2)

+1 +1 +1

for 1 a 1, where the first equality follows from equation (2.70) and Proposition 4.11. The
Pa1  a

lemma follows since =1 2 = 3 . 
Using the definitions and equation (2.64) one can show that for 0 k a the diagrammatic identity
c c c
k

(4.42) = k
+ (1)c1 k1

c1 c1 c1
holds, where the first diagram on the right-hand side is zero by convention when k = c.
Lemma 4.14.
a 1

as
a
X
a
(4.43) = (1)( )
2

s=0
s
a 1 a 1

Proof. A direct computation gives


(4.44)



a1 a1
(4.41) a1 (3.4) a1
(1)( 2 ) (1)( 2 )

+

a1



a 1 a 1 a 1 a 1

By (4.42) the odd elementary symmetric function can be slid to the top of the diagram.


a a1s
a1
X
a1
(4.45) = (1)( 2 ) (1)(a1) + (1)as s

s=0


a 1 a 1
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 41

Sliding the s dots down on the second term using the odd nilHecke relation and the fact that

(4.46) = 0, if x < a 1
x

gives



a a1s
a1
a1 X
(4.47) (4.44) = (1)( 2 ) (1)(a1) + (1)as+s .

s=0 s + 1


a 1 a 1

In the last diagram on the right we can slide the s + 1 dots down past the degree a 1 splitter giving

a 1

a1s a1s a1s

= (1)(a1)(s+1) = (1)(a1)(s+1)
s + 1
s + 1 s + 1
a 1 a 1 a 1

where we used Proposition 4.2 in the last equality. Shifting the summation
 shows that both terms in
the parenthesis come with sign (a 1). Since (a 1) + a1
2 = a
2 the result follows. 

Theorem 4.15. Let e = . The set {e }Sq(a) consists of mutually orthogonal idempotents that
add up to 1 ON Ha :

X
(4.48) e e = , e , e = 1.
Sq(a)
42 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Proof. Lemma 4.13 shows that e are orthogonal idempotents. To see that they decompose the identity
we proceed by induction, the base case being trivial.
(4.49)
1

a2
a X
ON Ha+1 1 = = (1)(3)
Sq(a) a1

a
ba1
b ba2
b1 r

then apply Lemma 4.14 in the form

a 1

a
a
X
a
(4.50) = (1)(2)
a =0
aa
a 1 a 1

a
 a
 
a+1
and use that 3 + 2 3 mod 2, proving the inductive step. 

Elements , give an explicit realization of ON Ha as the algebra of a! a! matrices over the


ring of odd symmetric functions. Suppose that rows and columns of a! a! matrices are enumerated
by elements of Sq(a). The isomorphism takes the matrix with x Oa in the (, ) entry and zeros
elsewhere to x .

4.4.1. Decomposition of E (a) E (b) . Given P (a, b) let


     
a b a+b
(4.51) Xa,b := || |b
| + a + bb + |b
| + + (b
) + ,
2 2 3

where is defined by equation (4.34).


When P (a, 1), so that = (1as ) for some 0 s a, then || = a s, |b
| = s, and Xa,1 from
(4.51) simplifies to
 
a,1 ar+1
(4.52) X(1r) = a(a r) + .
2
a,1
In particular, X(1a ) = 1.
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 43

For every partition P (a, b) define


a+b
a b
s a,b
(4.53) := , := (1)X , e = .
sbb
a b
a+b
We view , , and e as elements of ON Ha+b with deg( ) = deg( ) = 2||2ab, and deg(e ) =
0. Proposition 4.11 says that
(4.54) = , ea+b , , P (a, b).
This implies
(4.55) e e = , e .
Theorem 4.16.
X
(4.56) ea,b = e
P (a,b)

or diagrammatically

a b
s
X a,b
X
(4.57) = (1)
P (a,b)

sbb
a b a b


Proof. Proposition 4.11 implies that the e form a collection of a+b
a mutually orthogonal idempotents
in ON Ha . The result follows from the isomorphism ON Ha = Mat(a)! 2 (Oa ) in Corollary 2.14. 
q

5. C YCLOTOMIC QUOTIENTS OF THE ODD NIL H ECKE ALGEBRA

Let a1 , a2 , . . . be a sequence of an algebra. We say this family satisfies the odd symmetric defining
relations if
ai aj = aj aj if i + j is even,
(5.1) i i
ai aj + (1) aj ai = (1) ai+1 aj1 + aj1 ai+1 if i + j is odd.
Usually, ai = 0 for large enough i.
In [1], the algebra O of odd symmetric functions (there denoted or 1 ) is constructed as the
quotient of the graded free algebra khh1 , h2 , . . .i, deg(hi ) = i, by the odd symmetric defining relations
among the hi s. The degree n part of O has a basis indexed by partitions of n,
On = k{h = h1 h2 hr : = (1 , . . . , r ) is a partition of n}.
The elements h are called odd complete symmetric functions.
However, we need to modify these gradings in order for the definitions of [1] to be compatible
with those of this paper. In [1], the super-degree of a homogeneous element is the mod 2 residue of its
44 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

Z Z
-degree. Instead, for our present purposes, double the -gradings but leave unchanged the super-
Z
gradings of [1]. So hi has -degree 2i and super-degree i. The odd defining relations are unchanged.
Inductively define elements n O by the relation
n
X k+1
(5.2) (1)( 2 ) k hnk = 0.
k=0
By convention, h0 = 0 = 1 and hi = i = 0 for i < 0. The products = 1 r form a basis
of Oa just like the h s; we call such products odd elementary symmetric functions. It follows from
Lemmas 2.3, 2.8, and 2.9 that O is isomorphic to an inverse limit (in the category of graded rings) of
the rings Oa ,
O
= lim O
a
with respect to the maps
Oa+1 Oa ,
(
j j 6= a + 1,
j 7
0 j = a + 1,
in such a way that the odd complete and odd elementary symmetric polynomials in Oa pull back to
the odd complete and odd elementary functions of [1].
In terms of generators and relations, then,
(5.3) Oa = O/hm : m > ai.
For each N 0 we define the odd Grassmannian ring to be the quotient of Oa by the ideal generated
by all hm with m > N a,
OHa,N = Oa /hhm : m > N ai.
Z
For a = 0, N this ring is just , and the it vanishes unless 0 a N . The even analogue of OHa,N
C
is the cohomology ring H (Gr(a, N )) of the Grassmannian of complex a-planes in N . The quotient
ON HaN = ON Ha /hxN 1 i is called the N -th cyclotomic quotient of ON Ha .
Let Ha = {e x : } be as in Lemma 2.12 and let : ON Ha EndOa (OPola ) be the iso-
morphism of Corollary 2.14; under , a polynomial f acts as multiplication by f . A set of defining
relations for the cyclotomic quotient ON HaN is given by the matrix entries of (e x1 )N with respect to
the basis Ha . We should therefore consider the operator (e x1 ) in some detail. Our analysis closely
follows [11, Section 5].
For each multi-index obtained by replacing 1 by zero in some appearing in Ha , consider the
Oa -submodule of OPola with basis
x1a1 x
B = {e e1a2 x
e , x e , . . . , x
e }.
The operator (e x1 ) sends the span of each B to itself; let (e
x1 ) be the resulting restricted map (or
the corresponding matrix with respect to the basis B ). So the defining relations given by the entries
x1 )N are all realized in the a a matrices (e
of (e x1 )N
.

Lemma 5.1. The restriction of (ex1 ) to the span of B has the matrix expression

1 1 0 0 0
0 1 0 0
2
3 0 0 1 0

(5.4) (ex1 ) = .
.. .. .. .. ..
. . . .
a2
(1)( 2 ) a1 0 0 0 1
a1
(1)( 2 ) a 0 0 0 0
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 45

with respect to the basis B .


ej1 x
e1 x
Proof. Since x e = x ej+1
1 x e , all entries other than the first column are clear. To get the first column,
a
we need to express x e1 x
e in the basis B . That is, we want to find a linear relation
ea1 x
x e1a1 x
e = f1 x e + . . . + fa1 x e ,
e1 x
with each fj Oa . Recall that 0 = 1. Now
x1ak x
k (x1 , . . . , xa )e e = x x1ak x
e1 k1 (x2 , . . . , xa )e x1ak x
e + k (x2 , . . . , xa )e e
X X
= x ei1 x
e1 x e1ak x
eik1 x e + ei1 x
x e1ak x
eik x e .
2i1 <<ik1 a 2i1 <<ik a
a(k+1)
Comparing this with the analogous expansion of k+1 x e1 e , the first sum of the latter and the
x
k
second sum of the former differ by the sign (1) . So telescoping cancellations occur if each fj = j
and fj+1 = (1)j j+1 . The claim follows; that is, we have found
n
X j1
ea1 x
x e = (1)( 2 ) j x
e1aj x
e ,
j=1

and that equation (5.4) is true. 


1
Proposition 5.2. The cyclotomic quotient ON HaN is isomorphic to a matrix algebra of size q 2 a(a1) [a]!
over the odd Grassmannian ring OHa,N .
The proof will use an alternate presentation of OHa,N . Let the algebra Oa [t] be obtained from
Oa by adjoining an element t of super-degree 1 which is super-central; that is, thk = (1)k hk t for all
Z
k (and likewise with hk replaced by k ). The -degree of t is immaterial, so set it to 2 for consistency
with the xi s. Then the relation
(5.5) (1 + 1 t + 2 t2 + . . .)(1 + z1 t + z2 t2 + . . .) = 1
k+1
holds if and only if zk = (1)( 2 ) hk for each k. So we can define OHa,N by taking the quotient of
Oa by the ideal generated by the coefficients of powers tk in
(5.6) (1 + 1 t + 2 t2 + . . . + a ta )(1 + z1 t + z2 t2 + . . . + zN a tN a ) = 1,
and furthermore we know that
k+1
(5.7) zk = (1)( 2 ) hk
for all k.
Proof. Let M be the matrix (e x1 ) of equation (5.4). By Lemma 5.1, the entries of M N generate the
ideal defining the cyclotomic quotient ON HaN . It is not hard to show that these relations are already
generated by the entries in the first column of M N a+1 (the proof is exactly as in the even case). These
relations are homogeneous of degrees 2(N a + 1), 2(N a + 2), . . . , 2N . Let v = (1, 0, . . . , 0)T be the
column vector with first entry 1 and all other entries 0, so that we are seeking to compute the entries
of M N a+1 v. We proceed by induction on N . We may as well assume N a a; the case N a < a
is similar. Let fj,N a be the relation of degree 2(N a + j) in equation (5.6), for each j = 1, . . . , a. We
claim that N a+j+1
(M N a+1 v)j = (1)( 2 ) fj,N a .
For N = a, a + 1, this is clear. Now replacing N by N + 1, matrix multiplication shows
( j1 N a N a+j+2

N a+2 (1)( 2 )+( 2 ) j f1,N a + (1)( 2 ) fj+1,N a for 1 j a 1,


(5.8) (M v)j = a1 N a+2
(1)( 2 )+( 2 ) f a 1,N afor j = a.
46 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

And by equation (5.6) with N + 1 in place of N ,


(
fj+1,N a (1)j(N a+1) j f1,N a for 1 j a 1,
(5.9) fj,N a+1 = a1 N a+2
(1)( 2 )+( 2 ) f a 1,N a for j = a.
The expressions (5.8),(5.9) differ only by a sign, proving the proposition. 
The algebra O has an important basis, the odd Schur functions. They are indexed by partitions as
the complete and elementary functions are, and they are defined by the change of basis relation
X
(5.10) h = K sH
,
n

where || = n and K is a signed count of semistandard Young tableaux of shape and content ; for
the signs, see [1]. The superscript H is to distinguish these from the Schur functions of Subsection
Z
2.3 The Schur functions form an integral basis over , and they are signed-orthonormal:
(5.11) (sH H
, s ) = (1)
hi+||
.
P j +1
Here we use the notation hi = j 2 . In the even setting, the images of those Schur functions s
Z
with 1 N a and () a form an integral basis of H (Gr(a, N ); ); the class s is Poincare dual
to the corresponding Schubert cycle. The odd Schur functions corresponding to the same partitions
give a basis in the odd case as well, although no geometric description is currently known.
Conjecture 5.3. The image of sH
under the map O Oa equals the odd Schur polynomial s .

Z
Proposition 5.4. Over k = , the images of the Schur functions sH
for having at most a rows and at
most N a columns form a homogeneous basis for OHa,N . All other Schur functions are 0 in OHa,N .
Proof. It is shown in [1] that
X
sH
= h + a sH
,
>
X
sH
= + b sH
,
>

for certain integers a , b ; the ordering on partitions here is the lexicographic one. As a result, if has
more than a rows, sH H
is a linear combination of s with 1 > a for all ; hence s = 0. Applying the
involution 1 2 of [1], analogous reasoning with the odd complete polynomials implies that sH = 0 if
Z
has more than N a columns. So OHa,N is spanned over by the Schur functions sH for having
at most a rows and at most N a columns. Reducing modulo 2, odd symmetric polynomials coincide
Z
with the usual (even) symmetric polynomials over /2. In particular, they have the same graded rank
Z Z
over /2. Since OHa,N is a free -module, this implies that these Schur functions are in fact a basis
Z
over . 

6. C ATEGORIFICATION
Consider the category ON Ha pmod of finitely generated
L graded left projective ON Ha -modules
and its Grothendieck group K0 (ON Ha ). Let ON H := a0 ON Ha and define
M
K0 (ON H) := K0 (ON Ha ).
a0

There is a natural inclusion of algebras ON Ha ON Hb ON Ha+b given on diagrams by placing


a diagram in ON Ha next to a diagram in ON Hb with the diagram from ON Ha appearing to the left
THE ODD NILHECKE ALGEBRA AND ITS DIAGRAMMATICS 47

of the one from ON Hb . These inclusions give rise to induction and restriction functors that equip
K0 (ON H) with the structure of a q-bialgebra (for the notion of q-bialgebra, see [1]).
Denote the regular representation of ON Ha by E a . By Corollary 2.14 this module decomposes into
the direct sum of a! copies of the unique indecomposable projective module of ON Ha . If we denote
by E (a) the projective module corresponding to the minimal idempotent ea with the grading shifted
down by a(a1)
2 , then we get a direct sum decomposition of graded modules
M a(a 1) M
E a = ON Ha
= (ON Ha ea ) { }= E (a) .
2
[a]! [a]!
a a
Here [a]! = [a][a 1] . . . [1] is the quantum factorial, [a] = qqqq
1 , and M
f
or f M , for a graded
P j 1
Z
module M and a Laurent polynomial f = fj q [q, q ], denotes the direct sum over j , of fj Z
copies of M {j}.
Diagrammatically, the projective module E a corresponds to the idempotent 1 ON Ha given by
a vertical lines. The idempotent ea corresponding to the indecomposable projective module E (a) is
represented in the graphical calculus by a thick edge of thickness a. Theorem 4.15 can be interpreted
as giving an explicit isomorphism
X M
(6.1) : E a [a]! E (a) = E (a) {a 1 2||},
Sq(a) Sq(a)

while Theorem 4.16 gives a canonical isomorphism


X M M
(6.2) : E (a) E (b) 1n E (a+b) = E (a+b) 1n {2|| ab}.

P (a,b) P (a,b)

a+b
a
In particular, we have an isomorphism
U+
q (sl2 )A K0 (ON H)

(6.3) (a) 7 E (a)


where U+ +
q (sl2 )A is the integral version of the algebra Uq (sl2 ). The same result was announced by
Kang, Kashiwara, and Tsuchioka [2].
LN
Denote by ON H N := N
a=0 ON Ha the direct sum of cyclotomic quotients of ON Ha . The odd
Grassmannian ring OHa,N is graded local, so Proposition 5.2 implies that K0 (ON H N ) has the same
size as the integral form of the irreducible representation of Uq (sl2 ) of highest weight N (the cyclo-
tomic quotient ON HaN is zero unless 0 a N ). The tensor products by bimodules which are
odd analogues of the cohomology of two-step flag varieties descend to the action of E and F on
K0 (ON H N ), as in the even case.

R EFERENCES
[1] A. P. Ellis and M. Khovanov. The Hopf algebra of odd symmetric functions. 2011. arXiv:math.QA/1107.5610.
[2] S.-J. Kang, M. Kashiwara, and S. Tsuchioka. Quiver Hecke superalgebras. 2011. arXiv:math.QA/1107.1039v1.
[3] T. Khongsap and W. Wang. Hecke-Clifford algebras and spin Hecke algebras I: The classical affine type. Transf. Groups,
13:389412, 2008. arXiv:math.RT/0704.0201.
[4] T. Khongsap and W. Wang. Hecke-Clifford algebras and spin Hecke algebras II: The rational double affine type. Pacific J.
Math., 238:73103, 2008. arXiv:math.RT/0710.5877.
[5] T. Khongsap and W. Wang. Hecke-Clifford algebras and spin Hecke algebras IV: Odd double affine type. SIGMA, 5, 2009.
arXiv:math.RT/0810.2068.
[6] M. Khovanov. How to category one-half of quantum gl(1|2). 2010. arXiv:math.QA/1007.3517.
48 ALEXANDER P. ELLIS, MIKHAIL KHOVANOV, AND AARON D. LAUDA

[7] M. Khovanov and A. Lauda. A diagrammatic approach to categorification of quantum groups I. Represent. Theory, 13:309
347, 2009. arXiv:math.QA/0803.4121.
[8] M. Khovanov, A. Lauda, M. Mackaay, and M. Stosic. Extended graphical calculus for categorified quantum sl(2). 2010.
arXiv:math.QA/1006.2866.
[9] A. Lauda. Categorified quantum sl(2) and equivariant cohomology of iterated flag varieties. arXiv:math.QA/0803.3848,
2008. To appear in Algebras and Representation Theory.
[10] A. Lauda. A categorification of quantum sl(2). Adv. Math., 225(6):33273424, 2010. arXiv:math.QA/0803.3652.
[11] A. D. Lauda. An introduction to diagrammatic algebra and categorified quantum sl(2). 2011. arXiv:math.QA/1106.2128.
[12] R. Lipshitz, P. Ozsvath, and D. Thurston. Bordered Heegaard Floer homology: Invariance and pairing. 2008.
arXiv:math.GT/0810.0687.
[13] L. Manivel. Symmetric functions, Schubert polynomials and degeneracy loci, volume 6 of SMF/AMS Texts and Monographs. AMS,
Providence, RI, 2001.
[14] P. Ozsvath, J. Rasmussen, and Z. Szabo.
Odd Khovanov homology. 2007. arXiv:math.QA/0710.4300.
[15] R. Rouquier. 2-Kac-Moody algebras, 2008. arXiv:math.RT/0812.5023.
[16] A. Shumakovitch. Patterns in odd Khovanov homology. J. Knot Theory Ramifications, 20(1):203222, 2011.
arXiv:math.GT/1101.5607.
[17] W. Wang. Double affine Heke algebras for the spin symmetric group. Math. Res. Lett., 16:10711085, 2009.
arXiv:math.RT/0608074.
[18] B. Webster. Knot invariants and higher representation theory I: Diagrammatic and geometric categorification of tensor
products, 2010. arXiv:math.QA/1001.2020.
[19] B. Webster. Knot invariants and higher representation theory II: The categorification of quantum knot invariants, 2010.
arXiv:math.QA/1005.4559.

D EPARTMENT OF M ATHEMATICS , C OLUMBIA U NIVERSITY, N EW 1 Y ORK , NY 10027, USA


E-mail address: ellis@math.columbia.edu

D EPARTMENT OF M ATHEMATICS , C OLUMBIA U NIVERSITY, N EW Y ORK , NY 10027, USA


E-mail address: khovanov@math.columbia.edu

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


E-mail address: lauda@usc.edu