New Perspectives in Geometric Combinatorics

MSRI Publications
Volume 38, 1999

Macdonald Polynomials and Geometry

MARK HAIMAN

Abstract. We explain some remarkable connections between the two-

parameter symmetric polynomials discovered in 1988 by Macdonald, and
the geometry of certain algebraic varieties, notably the Hilbert scheme
Hilbn (C 2 ) of points in the plane, and the variety Cn of pairs of commuting
n × n matrices.

Contents
1. Introduction 207
2. Symmetric Functions and Macdonald Polynomials 210
3. The n! Conjecture 218
4. The Hilbert Scheme and Xn 221
5. Frobenius Series 229
6. The Ideals J and J m 234
7. Diagonal Harmonics 244
8. The Commuting Variety 250
References 252

1. Introduction
This article is an explication of some remarkable connections between the
two-parameter symmetric polynomials discovered by Macdonald [1988] and the
geometry of certain algebraic varieties, notably the Hilbert scheme Hilbn (C 2 ) of
points in the plane and the variety Cn of pairs of commuting n × n matrices
(“commuting variety”, for short). The conjectures on diagonal harmonics intro-
duced in [Haiman 1994; Garsia and Haiman 1996a] also relate to this geometric
setting.

1991 Mathematics Subject Classification. Primary 05-02; Secondary 14-02, 05E05, 13H10,
14M05.
Key words and phrases. Macdonald polynomials, Hilbert scheme, commuting variety, sheaf
cohomology, Cohen–Macaulay, Gorenstein.
Supported in part by NSF Mathematical Sciences grant DMS-9400934.

207
208 MARK HAIMAN

I have sought to give a reasonably self-contained treatment of these topics,

by providing an introduction to the theory of Macdonald polynomials, to the
“plethystic substitution” notation for symmetric functions (which is invaluable
in dealing with them), and to the conjectures and related phenomena that we
aim to explain geometrically. The geometric discussion is less self-contained,
it being unavoidable to use scheme-theoretic language, constructions such as
blowups, and some sheaf cohomological arguments. I do however give geometric
descriptions in elementary terms of the various algebraic varieties encountered,
and review whatever of their special features we might use, so as to orient the
reader not previously familiar with them.
The linchpin of the geometric connections we consider is the so-called “n! con-
jecture” [Garsia and Haiman 1993; 1996b], which remains unproved at present.
The n! conjecture proposes a combinatorial interpretation of the famous Kostka–
Macdonald coefficients Kλµ (q, t), which relate the Macdonald polynomials to
Schur functions and which were conjectured by Macdonald to be polynomials
with nonnegative integer coefficients in the parameters q, t. (Part of the conjec-
ture is that the Kλµ (q, t) are polynomials at all, which is not obvious from their
definition and was only proved recently, in five independent papers [Garsia and
Remmel 1998; Garsia and Tesler 1996; Kirillov and Noumi 1998; Knop 1997;
Sahi 1996].)
The n! conjecture is really two conjectures: first, that certain simply defined
spaces, quotient rings of the polynomial ring

C[x, y] = C[x1 , y1 , . . . , xn , yn ],

have dimension n!; and second, that these spaces, viewed as doubly graded
representations of the symmetric group Sn , have Hilbert polynomials that are
essentially the Kostka–Macdonald coefficients. It turns out, as we shall show,
that the first (apparently weaker) part of the conjecture is equivalent to the
Cohen–Macaulay property of a certain “isospectral” variety Xn over Hilbn (C 2 ).
Using this fact we can prove that the first part of the conjecture actually im-
plies the second part, and with it the Macdonald positivity conjecture for the
Kλµ (q, t).
As we shall see, the variety Xn is the blowup of (C 2 )n at the ideal J generated
by those elements of its coordinate ring C[x, y] that alternate in sign under the
action of the symmetric group Sn . An obvious conjecture is that J is the ideal
of the locus in (C 2 )n where two or more of the n points (xi , yi ) coincide, that is,
\
J= (xi − xj , yi − yj ). (1.1)
i6=j

It is easy to see that J defines the coincidence locus set-theoretically, which is

to say, the radical of J is the intersection on the right hand side above, but it is
not obvious, and indeed it remains an open question, that J is a radical ideal.
 MACDONALD POLYNOMIALS AND GEOMETRY 209

More generally, the geometry of the blowup Xn depends on module-theoretic

properties of the powers J m . We are led to extend (1.1) and conjecture that
\
Jm = (xi − xj , yi − yj )m , (1.2)
i6=j

that is, the powers of J are the symbolic powers of the ideal of the coinci-
dence locus. In fact, we conjecture that the variables x1 , . . . , xn form a regular
sequence for the C[x, y]-module J m , for all m. As we show, this conjecture im-
plies (1.2). It further implies that the xi’s form a regular sequence on Xn (that
is, on its structure sheaf O, viewed as a sheaf of C[x, y]-algebras). Assuming this
regular-sequence conjecture, we are able to give an inductive sheaf-cohomological
argument to show that Xn is Cohen–Macaulay, and thus the n! and Macdonald
positivity conjectures follow.
An important point to remark on here is that the n! conjecture and many
of the related geometric conjectures have evident analogous statements in more
than two sets of variables X, Y, Z, . . . . For the most part, these analogs fail to
hold.1 However, the above conjectures on the ideals J m are an exception, as
we expect them to hold in any number of sets of variables (the last conjecture
then being that any one of the sets of variables forms a regular sequence). Of
course the reasoning leading from there to the n! conjecture makes essential use
of having only two sets.
The isospectral Hilbert scheme Xn also provides the geometric setting for the
study of diagonal harmonics, the subject of a series of conjectures [Garsia and
Haiman 1996a; Haiman 1994]. The space of diagonal harmonics may be iden-
tified with the quotient ring Rn of C[x, y] by the ideal I generated by all Sn
invariant polynomials with zero constant term. It is conjectured, among other
things, that the dimension of Rn as a vector space is (n + 1)n−1 . Further conjec-
tures in [Haiman 1994] describe aspects of its structure as a graded Sn module in
combinatorial terms. In [Garsia and Haiman 1996a] we conjectured a complete
formula for the doubly graded character of Rn , in terms of Macdonald polyno-
mials, and proved that this master formula implies all the earlier combinatorial
conjectures.
In geometric terms Rn is the coordinate ring of the scheme-theoretic fiber
over the origin under the natural map (C 2 )n → S n C 2 from ordered n-tuples of
points in the plane to unordered n-tuples. Now, there is a fiber square
Xn −−−−→ (C 2 )n
 

σy

y
τ
Hilbn (C 2 ) −−−−→ S n C 2 ,

1 The n! conjecture has acquired a minor history of exciting but unsuccessful ideas for simple

proofs, by the author and others. A good reality check on a contemplated proof is to ask where
the argument breaks down — as it must — in three sets of variables.
210 MARK HAIMAN

giving rise to a natural homomorphism from Rn to the global sections of the

structure sheaf on the fiber (τ σ)−1 (0) ⊆ Xn . If Xn is Cohen–Macaulay, that is,
if the n! conjecture is true, these sections may be identified with global sections
of a vector bundle on the zero-fiber τ −1 (0) in Hilbn (C 2 ). Under a suitable
cohomology vanishing hypothesis, the homomorphism from Rn to this space of
global sections will be an isomorphism. Moreover, we can give its character
explicitly, using a variant of the Atiyah–Bott Lefschetz formula. This yields the
master formula for diagonal harmonics, on the assumption that the n! conjecture
and vanishing hypotheses hold. The agreement of this master formula with
computational results for n ≤ 7 is in my view striking evidence for the probable
validity of the geometric conjectures that give rise to it.
Finally, the commuting variety Cn enters the picture because it contains a
natural nonsingular open set Cn0 with a smooth map to Hilbn (C 2 ). The analog in
this context of the isospectral Hilbert scheme Xn is the “isospectral commuting
variety” ICn of pairs (X, Y ) of commuting matrices, together with 2n-tuples
(a1 , b1 . . . , an , bn) for which the (ai , bi) are the joint eigenvalues of the matrices
X, Y , that is, they satisfy the equations
Y
n
det(I + rX + sY ) = (1 + rai + sbi ), (1.3)
i=1

where r, s are indeterminates. The open set ICn0 of ICn lying over Cn0 is the
fiber product of Xn with Cn0 over Hilbn (C 2 ):
ICn0 −−−−→ Xn
 
 
y y (1.4)

Cn0 −−−−→ Hilbn (C 2 ).

Thus ICn0 is smooth over Xn , and hence Xn is Cohen–Macaulay if and only if
ICn0 is. In fact, as we shall see, if Xn is Cohen–Macaulay it is even Gorenstein,
and thus the same is true of ICn0 . We are led to conjecture that the whole
isospectral commuting variety ICn is Gorenstein, and not just the open subset
ICn0 . We have been able to verify this for small values of n. This conjecture
not only implies the n! conjecture, it also implies that the ordinary commuting
variety Cn is Cohen–Macaulay, which is an open problem of long standing.

2. Symmetric Functions and Macdonald Polynomials

The general reference for material in this section is [Macdonald 1995], whose
notation and terminology we follow, except as to the plethystic substitution, and
as to the transformed Macdonald polynomials H̃µ defined below. We give some
definitions and derive some properties that are also in the same reference, both
for completeness and to illustrate the utility of the plethystic notation. We also
derive some additional facts that will be needed later.
 MACDONALD POLYNOMIALS AND GEOMETRY 211

We work throughout with symmetric functions in infinitely many indetermi-

nates x1 , x2 , . . . , with coefficients in the field Q(q, t) of rational functions of two
variables q and t. The various classical bases of the ring of symmetric func-
tions are indexed by integer partitions µ, and denoted as follows: the monomial
symmetric functions by mµ , the power-sums by pµ , the elementary symmetric
functions by eµ , the complete homogeneous symmetric functions by hµ , and the
Schur functions by sµ . In each basis, as µ ranges over partitions of a given integer
d, we obtain a basis for the symmetric polynomials homogeneous of degree d.
The standard partial ordering on partitions of d is the dominance order, de-
fined by λ ≤ µ if λ1 + · · · + λk ≤ µ1 + · · · + µk for all k. The triangularity of
transition matrices between certain bases of symmetric functions with respect
to dominance plays a crucial role in the definition and development of Mac-
donald polynomials, as well as in the reasoning we will use later to deduce the
Macdonald positivity conjecture from the n! conjecture.
We now turn to the important device of plethystic substitution. The fact that
the power-sums pµ form a basis means, equivalently, that the ring of symmet-
ric functions can be identified with the ring of polynomials in the power-sums
p1 , p2 , . . . . In particular, the pk ’s may be specialized arbitrarily to elements of
any algebra over the coefficient field, and the specialization extends uniquely to
an algebra homomorphism on all symmetric functions.
Now let A be a formal Laurent series with rational coefficients in indetermi-
nates a1 , a2 , . . . , which may include our parameters q and t. We define pk [A]
to be the result of replacing each indeterminate ai in A by aki . Extending the
specialization pk 7→ pk [A] to arbitrary symmetric functions f, we obtain the
plethystic substitution of A into f, denoted f[A].
If A is merely a sum of indeterminates, A = a1 + · · · + an , then we see that
pk [A] = pk (a1 , a2 , . . . , an ), and hence for every f we have f[A] = f(a1 , a2 , . . . , an ).
This is why we view the operation as a kind of substitution. Similarly, if A has a
series expansion as a sum of monomials, f[A] is f evaluated on these monomials;
for example,
f[1/(1−t)] = f(1, t, t2 , . . .).

Our convention will be that in a plethystic expression X stands for the sum of
the original indeterminates x1 + x2 + · · · , so that f[X] is the same as f(X),

f[X/(1−t)] = f(x1 , x2 , . . . , tx1, tx2 , . . . , t2 x1 , t2 x2 , . . .),

and so forth. Among the virtues of this notation is that the substitution of
X/(1 − t) for X as above has an explicit inverse, namely the plethystic substi-
tution of X(1−t) for X.
The one caution that must be observed with plethystic notation is that inde-
terminates must always be treated as formal symbols, never as variable numeric
quantities. For instance, if f is homogeneous of degree d then it is true (and
212 MARK HAIMAN

easy to see) that

f[tX] = td f[X],
but it is false that f[−X] = (−1)d f[X]; that is, we cannot set t = −1 in
the equation above. In fact f[−X] is a very interesting quantity: it equals
(−1)d ωf(X), where ω is the classical involution on symmetric functions defined
by ωpk = (−1)k+1 pk , which interchanges the elementary and complete symmet-
ric functions eλ and hλ, and more generally exchanges the Schur function sλ
with sλ0 , where λ0 is the conjugate partition.
It is convenient when using plethystic notation to define
 ∞ 
X
Ω(X) = exp pk (X)/k . (2.1)
k=1

Then since pk [A + B] = pk [A] + pk [B] and pk [−A] = −pk [A] we have

Ω[A + B] = Ω[A]Ω[B], Ω[−A] = 1/Ω[A]. (2.2)
P k


From this and the single-variable evaluation Ω[x] = exp k≥1 x /k = 1/(1−x)
we obtain
Y 1 ∞
X
Ω[X] = = hn (X) (2.3)
1−xi n=0
i
and
Y ∞
X
Ω[−X] = (1−xi) = (−1)n en (X). (2.4)
i n=0
Recall that the standard Hall inner product h · , · i is defined so that the Schur
functions sµ are an orthonormal basis, and the complete symmetric functions hµ
are dual to the monomials mµ . The Cauchy identity is that for Hall-dual bases
{uµ}, {vµ } we have
Y 1 X
= uµ (X)uν (Y ),
1−xiyj µ
i,j
or, plethystically, X
Ω[XY ] = uµ [X]vµ [Y ]. (2.5)
µ
This may be written in a basis-free way as
Ω[AX], f(X) = f[A], (2.6)
which follows from (2.5) by taking f = vµ , and extending to arbitrary f by
linearity. In particular,
Ω[B(AX)], Ω[CX] = Ω[BX], Ω[C(AX)] = Ω[ABC].
But since B and C are arbitrary, we may set f(X) = Ω[BX], g(X) = Ω[CX],
to obtain the identity
f[AX], g(X) = f(X), g[AX] , (2.7)
 MACDONALD POLYNOMIALS AND GEOMETRY 213

valid for all f, g. In other words, the plethystic substitution of AX for X is

self-adjoint.
Macdonald defines his polynomials by first introducing a q, t-analog of the
Hall inner product h · , · i, which in plethystic notation is simply
 h 1−q i
hf, giq,t = f(X), g X .
1−t

In view of (2.7), this definition is symmetric in f and g. If {uµ } and {vµ }

  
are h · , · iq,t -dual bases, then {uµ} and vµ 1−q
1−t
are Hall dual, so the Cauchy
identity gives
X h i h i X
1−q 1−t
Ω[XY ] = uµ [X]vµ X , or Ω XY = uµ [X]vµ [Y ]. (2.8)
µ
1−t 1−q µ
 1− t

Note that the nonplethystic expression for Ω XY 1−q , as in [Macdonald 1995],
is the rather mysterious product
Y (txi yj ; q)∞ ∞
Y
, where (a; q)∞ = (1 − aq k ).
i,j
(xi yj ; q)∞
k=0

As a particular case of (2.8), we see that the h · , · iq,t -dual basis to the monomials
 1−t 
mµ is the basis of transformed complete symmetric functions hµ X 1−q (denoted
gµ in [Macdonald 1995]).
The Macdonald polynomials Pµ (X; q, t) may be defined by requiring that they
are orthogonal with respect to h · , · iq,t , and lower unitriangular with respect to
the monomials, that is,
X
Pµ (X; q, t) = mµ (X) + cλµ (q, t)mλ , (2.9)
λ<µ

for some coefficients cλµ . Here, however, we shall define them directly as eigen-
functions of the plethystic operator

∆0 f(X) = f[X − (1−q)/z]Ω[zX(1−t−1 )] z0

, (2.10)

where the vertical bar indicates we are to take the constant term with respect
to z.
Before further examining the operator ∆0 we define an important quantity
Bµ that appears in the eigenvalues of the operator, and will turn out to have
geometric significance later on. (The theory of Macdonald polynomials is rife
with numerology. Quantities such as Bµ and various q, t-hook products crop up
again and again; see [Garsia and Haiman 1998; 1996a] for many examples. In our
geometric context, these quantities will turn out to have natural interpretations.)
First recall that the diagram of a partition µ is the array of lattice points

D(µ) = {(i, j) ∈ N × N : j < µi+1 }. (2.11)

214 MARK HAIMAN

As is customary, we regard i as indexing rows and j as indexing columns, so that

the rows of D(µ) have lengths equal to the parts of µ. We now set
X
Bµ (q, t) = ti q j , (2.12)
(i,j)∈D(µ)

a kind of generating function describing D(µ), with a term for each cell in the
diagram, as illustrated here.
• t2 +
µ : (4, 2, 1) Dµ : • • Bµ : t + qt + (2.13)
• • • • 1 + q + q2 + q3
We now return to the study of the operator ∆0 .
Proposition 2.1. A symmetric function f(X; q, t) is an eigenfunction of ∆0
with eigenvalue α(q, t−1 ) if and only if f[X/(1−t−1 ); q, t−1] is an eigenfunction
of the operator
 
∆f = f X + (1−q)(1−t)/z Ω[−zX] z0 ,
with eigenvalue α(q, t).
Proof. We verify directly from the definitions that

 
∆ f[X/(1−t−1 ); q, t−1] = (∆0 f) X/(1−t−1 ); q, t−1 ,
which implies the result. 
Proposition 2.2. The operator ∆0 is lower-triangular with respect to the basis
of monomial symmetric functions. More precisely,

X
∆0 mµ = 1 − (1−q)(1−t−1 )Bµ (q, t−1 ) mµ + bλµmλ ,
λ<µ

for some coefficients bλµ.

Proof. Since the Schur functions are lower unitriangular with respect to the
basis of monomials, it will do equally well to prove
X
∆0 mµ = aλµ sλ ,
λ≤µ

for some coefficients aλµ , with aµµ = 1−(1−q)(1−t−1 )Bµ (q, t−1 ). It is also suffi-
cient to restrict to a finite set of variables X = x1 +· · ·+xn . Then Ω[zX(1−t−1 )]
has the partial fraction expansion
Qn
Yn
1−t−1 zxi X n
1 (1−xj /txi)
−1 −n
Ω[zX(1−t )] = =t + Qj=1
i=1
1 − zx i i=1
1 − zx i j6=i (1−xj /xi )
X 1 v(X)(xi 7→txi )
= t−n + t1−n (1−t−1 ) , (2.14)
i
1 − zxi v(X)
Q
where v(X) = i<j (xi − xj ) is the Vandermonde determinant.
 MACDONALD POLYNOMIALS AND GEOMETRY 215

From (2.14), using the identity f(1/z)/(1 − zx) z0

= f(x), we see that for any
function f,
X v(X)(xi 7→txi)
f(1/z)Ω[zX(1− t−1 )] z0
= t−n f(1/z) z0
+ t1−n (1−t−1 ) f(xi ) ,
v(X)
i

and therefore, since mµ [X − xi(1−q)] = mµ (X)(xi 7→qxi ) ,

X v(X)(xi 7→txi )
∆0 mµ (X) = t−n mµ (X) + t1−n (1−t−1 ) mµ (X)(xi 7→qxi ) .
i
v(X)

Note that the substitution of xi for z −1 inside the plethysm is permissible, since
we are substituting one indeterminate for another.
Recall the Jacobi formula
 λ +n−j n
sλ (X)v(X) = det xi j i,j=1
.

From this we see that the coefficient of sλ in the Schur function expansion of
any symmetric function f(X) is the coefficient of xλ+δ in f(X)v(X), where
δ = (n−1, n−2, . . . , 1, 0). In particular the coefficient aλµ of sλ in ∆0 mµ is
given by
X
t−n kλµ + t1−n (1−t−1 ) mµ (X)(xi 7→qxi ) v(X)(xi 7→txi) , (2.15)
xλ+δ
i

where kλµ is the coefficient of sλ in mµ , so kµµ = 1. Now in each summand

above, the leading term in dominance order is clearly xµ+δ , establishing the
triangularity. This term arises from the term xµ in mµ , multiplied by the term
xδ in v(X). In the i-th summand the indicated substitutions multiply it by
q µi tn−i. Thus we find
X
n
aµµ = t−n + t1− n (1−t−1 ) q µi tn−i.
i=1

With the understanding that µi is zero for i exceeding the number of parts l(µ)
of µ, we readily verify that the above expression is independent of n for n > l(µ),
as it must be, and reduces to
∞
X
(1−t−1 ) q µi t1−i = 1 − (1−q)(1−t−1 )Bµ (q, t−1 ). 
i=1

Corollary 2.3. The operator ∆0 has distinct eigenvalues 1 − (1−q)(1−t−1 ) ×

Bµ (q, t−1 ) and its corresponding eigenfunction is a linear combination of the
monomial symmetric functions mλ : λ ≤ µ, with nonzero coefficient of mµ .
Definition. The Macdonald polynomial Pµ (X; q, t) is the eigenfunction of the
operator ∆0 ,


∆0 Pµ = 1 − (1−q)(1−t−1 )Bµ (q, t−1 ) Pµ ,
216 MARK HAIMAN

normalized so that X
Pµ = mµ + cλµ mλ .
λ<µ

The coefficients cλµ (q, t) are rational functions with nontrivial denominators.
Macdonald proposed an alternate normalization, called the integral form
Y
Jµ = (1−q a(s) t1+l(s) )Pµ , (2.16)
s∈D(µ)

and conjectured that its coefficients are polynomials in q and t, i.e., the above
product clears the denominators. Here a(s) and l(s) are the arm and leg of the
cell s in the diagram of µ, defined to be the number of cells strictly east and
north of s, respectively.
This integrality conjecture has recently been proved [Garsia and Remmel 1998;
Garsia and Tesler 1996; Kirillov and Noumi 1998; Knop 1997; Sahi 1996]. Mac-
donald made a further remarkable conjecture, that if we write the integral forms
as X
Jµ (X; q, t) = Kλµ (q, t)sλ [X(1−t)] (2.17)
λ
then the coefficients Kλµ (q, t) are not only polynomials in q and t, but they
have nonnegative integer coefficients. The search for an algebraic-combinatorial
proof of this Macdonald positivity conjecture, which remains open, has been the
moving force behind the work described in this article. It is pertinent to mention
here that Jµ (X; 0, t) specializes to the Hall–Littlewood function Qµ (X; t), and
therefore the coefficients Kλµ (0, t) specialize to the famous t-Kostka coefficients
Kλµ (t) that have been central to much beautiful work in combinatorics, geometry
and representation theory. This is one of many reasons for the great interest in
Macdonald polynomials in the decade since their discovery.
For our purposes it is convenient to work with the following variant. In all
that follows we fix |µ| = n (not to be confused with n(µ)).
Definition. The transformed Macdonald polynomials are
h X i
H̃µ (X; q, t) = tn(µ) Jµ ; q, t−1 , (2.18)
1−t−1
P P
where n(µ) = i (i − 1)µi = s∈D(µ) l(s).
From Proposition 2.1 and equation (2.17) we immediately obtain this result:
Proposition 2.4. The transformed polynomial H̃µ is an eigenfunction of the
operator ∆,


∆H̃µ = 1 − (1−q)(1−t)Bµ H̃µ ,
and its Schur function expansion is
X
H̃µ = K̃λµ (q, t)sλ ,
λ
 MACDONALD POLYNOMIALS AND GEOMETRY 217

where K̃λµ (q, t) = tn(µ) Kλµ (q, t−1 ). In particular (since it is known that K(n),µ =
tn(µ) [Macdonald 1995]), H̃µ is normalized so that its coefficient of s(n) is 1.
Now the operator ∆ is symmetric in q and t, while Bµ (t, q) = Bµ0 (q, t). Hence
we obtain
Proposition 2.5. For all µ we have H̃µ0 (X; q, t) = H̃µ(X; t, q) and, conse-
quently, K̃λµ0 (q, t) = K̃λµ (t, q).
As mentioned earlier, the functions H̃µ can be characterized by certain triangu-
larity relations.
Proposition 2.6. The transformed Macdonald polynomials H̃µ satisfy, and are
uniquely characterized by, these conditions:
(1) H̃µ[X(1−q); q, t] ∈ Q(q, t){sλ : λ ≥ µ};
(2) H̃µ[X(1−t); q, t] ∈ Q(q, t){sλ : λ ≥ µ0 };
(3) hH̃µ, s(n) i = 1.
Proof. Note that H̃µ[X(1−t); q, t] = t|µ| H̃µ [−X(1−t−1 ); q, t] is a scalar multi-
ple of Pµ [−X; q, t−1 ] and thus of ωPµ (X; q, t−1 ). Since Pµ belongs to the space
Q(q, t){sλ : λ ≤ µ}, and conjugation reverses the dominance order, ωPµ be-
longs to Q(q, t){sλ : λ ≥ µ0 }, which is (2) above. From the symmetry given by
Proposition 2.5 we then obtain (1). The normalization from Proposition 2.4 is
(3).
For uniqueness, suppose Hµ0 (X) is another solution of (1) and (2). Then
(1) implies that Hµ0 [X(1−q)] ∈ Q(q, t){H̃λ[X(1−q)] : λ ≥ µ} and hence that
Hµ0 ∈ Q(q, t){H̃λ : λ ≥ µ}. Similarly, (2) implies that Hµ0 ∈ Q(q, t){H̃λ : λ ≤ µ}.
Together these mean that Hµ0 is a scalar multiple of H̃µ, and (3) fixes the scalar
factor as 1. 
0
Corollary 2.7. For all µ we have ωH̃µ (X; q, t) = tn(µ) q n(µ ) H̃µ(X; q −1 , t−1 )
0
and, consequently, K̃λ0 µ (q, t) = tn(µ)q n(µ ) K̃λµ (q −1 , t−1 )
0
Proof. One verifies easily that ωtn(µ) q n(µ ) H̃µ (X; q −1 , t−1 ) satisfies (1) and (2)
of Proposition 2.6, and hence is a scalar multiple of H̃µ . To fix the scalar as 1
0
requires that K̃(1n ),µ = tn(µ) q n(µ ) . But this is known [Macdonald 1995], as it is
0
equivalent to K(1n ),µ = q n(µ ) . 
To conclude, we will recover the orthogonality of the Pµ ’s with respect to h · , · iq,t ,
as in Macdonald’s original definition. Replacing t by t−1 , we are to show that
  
−1 1−q
Pµ (X; q, t ), Pν X ; q, t−1 =0
1−t−1
for µ 6= ν, or equivalently that
  
−1 1−q
Pµ (X; q, t ), Pν −X ; q, t−1 = 0.
1−t
218 MARK HAIMAN

Since Pµ (X; q, t−1 ) is a scalar multiple of H̃µ [X(1−t)], we are to show that
H̃µ [X(1−t)], H̃ν [−X(1−q)] = 0.
Now from Proposition 2.6 and the orthogonality of Schur functions it is clear
that this last inner product vanishes unless ν ≤ µ. But then by symmetry it also
vanishes unless µ ≤ ν, that is, unless µ = ν.

3. The n! Conjecture
Let D = {(p1 , q1 ), . . . , (pn , qn )} be an n-element subset of N × N. We define
a polynomial in 2n variables x1 , y1 , . . . , xn , yn as follows:
 p q n
∆D (x, y) = det xi j yi j i,j=1 . (3.1)
Note that ∆D is well defined, up to a change of sign, independent of the ordering
chosen for the elements of D, and that it alternates in sign under the action of
the symmetric group Sn permuting the x and the y variables simultaneously.
That is,
w∆µ = ε(w)∆µ for all w ∈ Sn ,
where ε(w) is the sign of the permutation w. For a partition diagram, we set
∆µ = ∆D(µ) .
Then ∆µ is doubly homogeneous, of degree n(µ) in the x variables and n(µ0 ) in
the y variables. In the cases µ = (1n ) and µ = (n), ∆µ is the usual Vandermonde
determinant in the x and y variables, respectively.
Our conjectures concern the space of all derivatives of ∆µ,

Dµ = p(∂x1 , ∂y1 , . . . , ∂xn , ∂yn )∆µ : p ∈ Q[x, y] .
Since ∆µ is doubly homogeneous, this space is doubly graded:
M
Dµ = (Dµ )r,s ,
r,s

where (Dµ )r,s consists of those elements of Dµ that are doubly homogeneous of
degree r in the x variables and s in the y variables. Since ∆µ is alternating, the
space Dµ is stable under the action of Sn . Thus it affords a doubly graded repre-
sentation of the symmetric group Sn . We denote by mult(χλ , χ) the multiplicity
of the irreducible Sn -character χλ in a given character χ.
Conjecture 3.1 (the n! Conjecture). The space Dµ has dimension n!.
Conjecture 3.2. The bivariate character multiplicity Hilbert series
X
tr q s mult(χλ , ch(Dµ )r,s ) (3.2)
r,s

is equal to K̃λµ (q, t). In particular , the latter is a polynomial with nonnegative
integer coefficients.
 MACDONALD POLYNOMIALS AND GEOMETRY 219

It is known [Macdonald 1995] that K̃λµ (1, 1) = χλ(1), the degree of the character
χλ or the number of standard Young tableaux of shape λ. Hence, according to
Conjecture 3.2, we must have mult(χλ , ch(Dµ )) = χλ(1), so that when we ignore
the grading, Dµ affords the regular representation of Sn , and hence has dimension
n!. Thus Conjecture 3.2 implies Conjecture 3.1. One of the chief things we will
achieve in the geometric setting of Sections 4 and 5 is to prove the converse
implication.
It will be helpful to reformulate the conjectures in two ways. The first is to
introduce a more convenient notation for (3.2). Recall that the Frobenius map
from Sn characters to symmetric functions homogeneous of degree n is defined by
1 X
Φ(χ) = χ(w)pτ(w) (X), (3.3)
n!
w∈Sn

where τ (w) is the partition whose parts are the lengths of the cycles of the
permutation w, and where the indeterminates X are not to be confused with the
coordinates x. For the irreducible characters we have the symmetric function
identity Φ(χλ ) = sλ (X), from which follows, for any character,
X
Φ(χ) = mult(χλ , χ)sλ.
λ

Now, by analogy to the Hilbert series, we define the Frobenius series of a doubly
graded Sn representation D to be
X
FD (X; q, t) = tr q s Φ ch(D)r,s .
r,s

This given, Conjecture 3.2 takes the simple form

FDµ = H̃µ . (3.4)

In Section 5 we extend the notion of Frobenius series to Sn actions on modules

over a geometric regular local ring with an equivariant two-dimensional torus
action, providing the basic tool to link Conjecture 3.2 with the geometry.
The second reformulation we need is of the definition of Dµ itself — the def-
inition in terms of derivatives is simple, but geometrically misleading, and we
need a derivative-free version. This is given by the next propositions. We will
need the following definition here and later.
Definition. The alternation operator over the symmetric group is
X
Alt f = ε(w)w(f).
w∈Sn

Since the polynomials ∆D form a basis of all the Sn -alternating polynomials in

Q[x, y], it makes sense to speak of the coefficient of ∆D in Alt f. Indeed, it is
merely the coefficient of the monomial xp11 y1q1 · · · xpnn ynqn .
220 MARK HAIMAN

Proposition 3.3. The ideal Jµ of polynomials p(x, y) ∈ Q[x, y] for which the
differential operator p(∂x, ∂y) annihilates ∆µ can be characterized as follows:
p ∈ Jµ if and only if for all g ∈ Q[x, y], the coefficient of ∆µ in Alt gp is zero.
Proof. Observe that the constant term of g(∂x, ∂y)p(∂x, ∂y)∆µ is, apart from
a constant factor, the coefficient of ∆µ in Alt gp. Hence if p(∂x, ∂y)∆µ = 0, the
characterization certainly holds. Conversely, if the characterization holds, then
p(∂x, ∂y)∆µ has the property that it and all its partial derivatives of all orders
have zero constant term. By Taylor’s theorem this implies that p(∂x, ∂y)∆µ = 0.

Proposition 3.4. The quotient ring Q[x, y]/Jµ is isomorphic as a doubly
graded Sn representation to Dµ .
Proof. Define an inner product ( · , · ) on Q[x, y] by
(f, g) = f(∂x, ∂y)g(x, y) x,y=0
. (3.5)
One sees immediately that monomials are mutually orthogonal under ( · , · ).
Hence the form is symmetric and nondegenerate, separately on each doubly
homogeneous subspace (Q[x, y])r,s . It follows that Q[x, y]/Jµ is isomorphic as
a doubly graded Sn representation to the orthogonal complement Jµ⊥ .
Now I claim that Jµ⊥ = Dµ . Taking g = ∆µ and f ∈ Jµ in (3.5), we see that
∆µ ∈ Jµ⊥ . From the definition we have (p(x, y)f, g) = (f, p(∂x, ∂y)g) for all
p(x, y), i.e., multiplication is adjoint to differentiation. Since Jµ is an ideal it
follows that Jµ⊥ is closed under differentiation and hence contains Dµ .
Now both Jµ⊥ and Dµ are finite-dimensional, and the map p 7→ p(∂x, ∂y)∆µ
is an isomorphism of vector spaces from Q[x, y]/Jµ onto Dµ . Therefore Jµ⊥ and
Dµ have the same dimension and hence are equal. 
For geometric purposes we will replace Q by C, extending Jµ to Jµ ⊗Q C, which
we again denote Jµ , and for which the characterization in Proposition 3.3 still
holds.
Definition. The ring Rµ is C[x, y]/Jµ.
Of course Rµ has the same Frobenius series as Q[x, y]/Jµ, and also, by Propo-
sition 3.4, as Dµ .
The evidence for Conjecture 3.2 includes the fact that various known sym-
metries and specializations of H̃µ can also be established for FDµ . We conclude
this section by demonstrating a few of these.
Proposition 3.5. We have the identity FDµ0 (X; q, t) = FDµ (X; t, q) (compare
Proposition 2.5).
Proof. Obvious, since ∆µ0 (x, y) = ∆µ (y, x). 
Proposition 3.6. We have the identity
0
ωFDµ (X; q, t) = tn(µ) q n(µ ) FDµ (X; q −1 , t−1 )
 MACDONALD POLYNOMIALS AND GEOMETRY 221

(compare Corollary 2.7).

Proof. In the proof of Proposition 3.4 there are two isomorphisms of Q[x, y]/Jµ
with Dµ . The first is the inclusion of Dµ in Q[x, y], followed by projection mod
Jµ , which is an isomorphism of graded Sn representations. The second is the
map p 7→ p(∂x, ∂y)∆µ . As ∆µ is Sn -alternating and homogeneous of degrees
(n(µ), n(µ0 )), this map reverses degrees and tensors Sn characters by the sign
character. Recalling that the Frobenius map satisfies Φ(ε ⊗ χ) = ωΦ(χ), we
see that reversing degrees and tensoring with the sign character on Dµ yields
0
a space with Frobenius series ωtn(µ) q n(µ ) FDµ (X; q −1 , t−1 ). Combining the two
isomorphisms shows this is equal to FDµ . 
Remark: The algebraic correlate of this symmetry of Dµ is that the ring Rµ is
Gorenstein. More generally, for any homogeneous ideal J ⊆ Q[x], the quotient
Q[x]/J is finite-dimensional (as a vector space) and Gorenstein if and only if J
is the ideal of differential operators annihilating some homogeneous polynomial.
Proposition 3.7. We have the identity FDµ (X; 0, t) = H̃µ (X; 0, t).
Proof. The left-hand side is the Frobenius series of the subspace Q[x] ∩ Dµ .
The Garnir polynomial gµ (x) is defined as the product of the Vandermonde
determinants in the first µ01 variables, the next µ02 variables, and so on. It is not
hard to see that the space Q[x] ∩ Dµ is spanned by all derivatives of gµ and its
images under permutation of the variables.
This space has been well studied [Bergeron and Garsia 1992; De Concini
and Procesi 1981; Garsia and Procesi 1992; Hotta and Springer 1977; Kraft
1981; Springer 1978], and its Frobenius series (in one variable t) is known to
be the transformed Hall–Littlewood polynomial tn(µ)Qµ [X/(1−t−1 ); t−1 ]. Since
Jµ (X; 0, t) = Qµ (X; t), the result follows. 
It is also possible to give an elementary proof (see [Garsia and Haiman 1996b])
that the n! conjecture implies FDµ (X;1, t) = H̃µ (X;1, t). Since FDµ (X; 1, 1) de-
termines the character and thus the dimension of Dµ , one must of course assume
the n! conjecture for this. But as we are going to prove that the n! conjecture
implies FDµ = H̃µ , the proof for the q = 1 specialization would be redundant
here.

4. The Hilbert Scheme and Xn

Let C 2 = Spec C[x, y] be the affine plane over C. By definition, closed sub-
schemes S ⊆ C 2 correspond to ideals I ⊆ C[x, y]. The subscheme S is 0-
dimensional, of length n, if C[x, y]/I has dimension n as a vector space. The
generic example of such a subscheme S is a set of n points in C 2 , with the re-
duced subscheme structure. In this case I is a radical ideal and C[x, y]/I can be
identified with the ring of complex-valued functions on the finite set S, which is
clearly n-dimensional.
222 MARK HAIMAN

Other such subschemes S have fewer than n points, with nonreduced sub-
scheme structures at these points. If we define the multiplicity of each point
P ∈ S as the length of the local ring OS,P then the total length, n, is the sum of
the multiplicities. Associated to each partition µ of n is a nonreduced subscheme
Sµ whose ideal Iµ is spanned by the monomials {xr ys : (r, s) 6∈ D(µ)}. Note
that this is indeed an ideal. Since the remaining monomials

Bµ = {xr ys : (r, s) ∈ D(µ)} (4.1)

form a basis of C[x, y]/Iµ, Sµ has length n, and the Hilbert series of C[x, y]/Iµ
as a doubly graded algebra is the quantity Bµ (q, t) defined in (2.12). We readily
see that in fact, the Sµ ’s are the only 0-dimensional length-n subschemes whose
ideals are spanned by monomials. As a set, Sµ has only one point, the origin,
with multiplicity n.
The set of all 0-dimensional length-n subschemes of C 2 , or equivalently, the
set of ideals I such that dimC C[x, y]/I = n, has the structure of an algebraic
variety, the Hilbert scheme of n points in the plane, or Hilbn (C 2 ). We will prefer
the point of view that the (closed) points of Hilbn (C 2 ) are the ideals I. Its
variety structure may be described in either of two ways.
First, for every I ∈ Hilbn (C 2 ), at least one of the sets Bµ spans modulo I
[Gordan 1900]. In particular, the set MN = {xr ys : r + s < N } always spans
mod I, for N ≥ n. Thus C[x, y]/I may be identified with an element of the
Grassmann variety Gn (W ) of n-dimensional quotients of the space W = CMN ,
giving a map from Hilbn (C 2 ) into Gn (W ). For N sufficiently large — in fact,
for N ≥ n + 1 [Gotzmann 1978] — this map is injective, its image is a locally
closed subvariety of Gn (W ), and the induced variety structure on Hilbn (C 2 ) is
independent of N .
The second description is via Grothendieck’s universal property [Grothendieck
1961], as follows. There exists a subscheme U ⊆ Hilbn (C 2 ) × C 2 ,

U −−−−→ C 2


πy (4.2)

Hilbn (C 2 ),

called the universal family, whose scheme-theoretic fiber over a point I of the
scheme Hilbn (C 2 ) is the corresponding subscheme S ⊆ C 2 . In particular, U is
flat and finite of degree n over Hilbn (C 2 ). The universal property is that for any
scheme T and any family F ⊆ T × C 2 of subschemes of C 2 ,

F −−−−→ C 2


y (4.3)

T,
 MACDONALD POLYNOMIALS AND GEOMETRY 223

flat and finite of degree n over T , there is a unique morphism φ : T → Hilbn (C 2 ),

such that F is the fiber product F = T ×Hilbn (C 2 ) U . The universal property,
as usual, characterizes Hilbn (C 2 ) and the universal family U up to canonical
isomorphism.
The following result, which is unique to C 2 and fails in more than two dimen-
sions, is one of the true miracles of the mathematical universe.

Theorem 4.1 [Fogarty 1968]. The Hilbert scheme Hilbn (C 2 ) is irreducible and
nonsingular , of dimension 2n.

Given I ∈ Hilbn (C 2 ), the operators X and Y of multiplication by x and y,

respectively, are commuting endomorphisms of the vector space
Y
C[x, y]/I ∼
= OS,P
P ∈S

which preserve the direct factors OS,P . For P = (x0 , y0 ), X − x0 and Y − y0 are
nilpotent on OS,P , so that OS,P is the characteristic subspace associated with
the joint eigenvalue (x0 , y0 ) of (X, Y ). Thus the points (x1 , y1 ), . . . , (xn , yn ) of
S, each included with its multiplicity in S, are the joint spectrum of (X, Y ). In
Pn
particular, the polarized power sums ph,k (x, y) = i=1 xhi yik satisfy the identity

ph,k (x, y) = tr X h Y k . (4.4)

Note that trace map associated to the finite morphism π : U → Hilbn (C 2 ) sends
the regular function xh yk on U (coming from U → C 2 ) to tr X h Y k , so the latter
is a regular function on Hilbn (C 2 ).
Now the symmetric group Sn acts on the variety

(C 2 )n = Spec C[x1 , y1 , . . . , xn , yn ]

of ordered n-tuples of points in the plane, and we may identify Spec of the
ring of invariants as the variety of unordered n-tuples, or n-element multisets,
S n C 2 = Spec C[x, y]Sn . By a theorem of Weyl [1946], the polarized power-
sums ph,k generate this ring of invariants. It follows from this and the preceding
paragraph that the map

τ : Hilbn (C 2 ) → S n C 2

sending I to the multiset of points of the corresponding subscheme S, with their

multiplicities in S, is a morphism. It extends to a morphism τ̂ : Hilbn (P 2 ) →
S n P 2 , with τ being the restriction to τ̂ −1 (S n C 2 ). Since Hilbn (P 2 ) is a projective
variety, τ is a projective morphism, called the Chow morphism. Note that τ
restricts to an isomorphism of the open sets where the n points are distinct, and
so is birational.
224 MARK HAIMAN

Definition. The isospectral Hilbert scheme Xn is the reduced fiber product

Xn −−−−→ (C 2 )n
 

σy

y (4.5)
τ
Hilbn (C 2 ) −−−−→ S n C 2 .
In other words, a point of Xn is a point I of the Hilbert scheme, together
with an ordered n-tuple of points (x1 , y1 ), . . . , (xn , yn ) whose underlying un-
ordered multiset is τ (I), that is, the joint spectrum of the operators (X, Y )
on C[x, y]/I. We should stress here that the scheme-theoretic fiber product
indicated by the above diagram is not reduced, but our definition is that Xn
is the underlying reduced subscheme. What this reflects is that the equations
ph,k (x1 , y1 , . . . , xn , yn ) = tr X h Y k , which define Xn set-theoretically, do not gen-
erate its ideal. Indeed we do not know a fully explicit set of generators for the
ideal of Xn . Such a set of generators will necessarily be complicated, since it
must specialize to give generators of all the ideals Jµ of Proposition 3.3.
It is possible to describe the ideal of Xn implicitly, as we shall do next.
Let U ×n denote the n-fold fiber product of the universal family U over the
Hilbert scheme Hilbn (C 2 ). It is a subscheme of Hilbn (C 2 ) × (C 2 )n , flat and
finite of degree nn , and generically reduced, since it has reduced fibers whenever
S consists of n distinct points. As the Hilbert scheme is irreducible, this implies
that U ×n is reduced.2 Thus U ×n is just the set of tuples I, (x1 , y1 ), . . . , (xn , yn )
with all (xi, yi ) ∈ S = V (I), irrespective of multiplicities. Xn is the subscheme
of U ×n consisting of tuples for which the points occur with their correct multi-
plicities in S. Generically, when S has n distinct points, this means the points
(xi, yi ) are a permutation of S.
Now let B = π∗ OU , the sheaf of O-algebras on Hilbn (C 2 ) such that U =
Spec B as a scheme over Hilbn (C 2 ). Since π is flat of degree n, B is locally free
of rank n, that is, it is the sheaf of sections of a rank-n vector bundle over the
Hilbert scheme. Indeed, B is the tautological bundle whose fiber over a point
I ∈ Hilbn (C 2 ) is C[x, y]/I. We have U ×n = Spec B ⊗n , and we seek to identify
the ideal sheaf of Xn in U ×n as a submodule sheaf of the sheaf of O-algebras
B ⊗n . Note that the Sn action here permutes the tensor factors.
Proposition 4.2. Let
Vn
B ⊗n ⊗ B ⊗n → B ⊗n → B (4.6)

be the map given by multiplication, followed by the operator Alt. Then the ideal
sheaf of Xn is the kernel of the map
V
φ : B ⊗n → (B ⊗n )∗ ⊗ n B (4.7)

2 By the same reasoning, the universal family U itself is reduced. This fails in higher

dimension, when the Hilbert scheme is not irreducible.

 MACDONALD POLYNOMIALS AND GEOMETRY 225

induced by (4.6).
Proof. Since Xn is reduced, a section of B ⊗n belongs to the ideal sheaf of Xn
if and only if, regarded as a regular function on Xn , it vanishes on any dense
open subset. Hence it is enough to check the proposition generically, over the
locus where S consists of n distinct points. Suppose s is a section that vanishes
on Xn . Then so do gs and Alt gs, for any section g of B ⊗n . But since it is
alternating, Alt gs also vanishes at any point I, P1 , . . . , Pn of U ×n for which two
of the points Pi , Pj coincide. Hence it vanishes on U ×n \ Xn and thus on all of
U ×n , which means it is zero in B ⊗n . The condition that s belongs to the kernel
of (4.7) is precisely that Alt gs = 0 for all g.
Conversely, suppose s does not vanish on Xn , and choose a point Q =
(I, P1 , . . . , Pn ) ∈ Xn outside the vanishing locus V (s), with the Pi all distinct.
After multiplying by a suitable g we can arrange that gs vanishes at all points of
the Sn orbit of Q, except Q. Then Alt gs does not vanish at Q, so Alt gs 6= 0. 
Now consider the situation over one of the distinguished points Iµ . The fiber
B(Iµ ) of the vector bundle B is C[x, y]/Iµ, and that of B ⊗n is


C[x, y]/ Iµ (x1 , y1 ) + · · · + Iµ (xn , yn ) .

Notice that every alternating polynomial ∆D as in (3.1) vanishes modulo

(Iµ (x1 , y1 ) + · · · + Iµ (xn , yn )),

Vn
except for ∆µ . In other words, the 1-dimensional space B(Iµ ) is spanned
Vn
by the image of ∆µ , and the linear functional Alt : B ⊗n (Iµ ) → B(Iµ ) ∼
=
C{∆µ }, composed with the natural projection C[x, y] → B ⊗n (Iµ ), is just the
map sending f to the coefficient of ∆µ in Alt f. Together with Proposition 3.3,
this proves the next result:
Proposition 4.3. The ideal Jµ of polynomials that as differential operators
annihilate ∆µ is the kernel of the map
Vn
C[x, y] → B ⊗n (Iµ ) → (B ⊗n )∗ (Iµ ) ⊗ B(Iµ )

induced by the map φ in (4.7).

Comparing this with Proposition 4.2 we might well expect the fiber of the ideal
sheaf of Xn over Iµ to be Jµ , and the scheme-theoretic fiber of Xn over Iµ to be
Spec Rµ . We cannot yet draw this conclusion, however, since the map φ is only
a sheaf homomorphism, not necessarily a homomorphism of vector bundles, and
thus the fiber of its kernel at Iµ need not equal the kernel of its fiber. What we
can say is is that the fiber of the kernel factors through the kernel of the fiber,
which gives the following result.
Proposition 4.4. The image of the ideal Jµ in B ⊗n (Iµ ) contains the image of
the fiber map J(Iµ ) → B ⊗n (Iµ ), where J(Iµ ) is the fiber of the ideal sheaf J of
226 MARK HAIMAN

Xn at Iµ . As a consequence dim Rµ ≤ n!, and Rµ is isomorphic to a submodule

of the regular representation of Sn .
Proof. The first part is clear, by the previous propositions. It only remains to
prove the consequence.
By Proposition 4.2, we have an exact sequence of sheaves on Hilbn (C 2 ),
0 → J → B ⊗n → σ∗ OXn → 0, (4.8)
in which σ∗ OXn is the image of the sheaf homomorphism φ given by (4.7). At
generic ideals I ∈ Hilbn (C 2 ) corresponding to sets of n distinct points S ⊆ C 2 ,
the fibers σ∗ OXn (I) have constant dimension n!. This implies that the rank of
the fiber map
V
φ(I): B ⊗n (I) → (B ⊗n )∗ (I) ⊗ n B(I)
is generically n!.
At the special ideal Iµ the rank of φ(Iµ ) cannot exceed the generic rank, and by
Proposition 4.3 this implies dim Rµ ≤ n!. Since Sn acts equivariantly on every-
thing, the same considerations apply to the isotypic components corresponding
to each irreducible character of Sn , to show that the character multiplicities in
Rµ cannot exceed those in a generic fiber σ∗ OXn (I). When I = I(S), this fiber
is the coordinate ring of the set of all permutations of S, and thus affords the
regular representation of Sn . 
Continuing with this argument, we see that if the n! conjecture holds for µ then
the rank of φ(I) does not decrease at Iµ , and so is constant on a neighborhood
of Iµ . This is the criterion for φ to be locally a homomorphism of vector bundles
on a neighborhood of Iµ . When this holds, the sheaves in (4.8) are locally free,
and the scheme Xn is flat of degree n! over Hilbn (C 2 ) at Iµ .
Conversely, if Xn is flat over Hilbn (C 2 ) at Iµ , then B ⊗n and B ⊗n /J are both
locally free, which implies that J is locally free and φ is a homomorphism of
vector bundles. By Propositions 4.2 and 4.3 it then follows that J(Iµ ) = Jµ
and dim Rµ = n!. Recall that a finite, surjective morphism X → H with H
nonsingular is flat if and only if X is Cohen–Macaulay. We have proved:
Theorem 4.5. The following statements are equivalent :
(1) The n! conjecture holds for the partition µ.
(2) The map Xn → Hilbn (C 2 ) is flat over a neighborhood of Iµ .
(3) The isospectral Hilbert scheme Xn is Cohen–Macaulay in a neighborhood of
the point Qµ = (Iµ , 0, . . . , 0).
This theorem has an interesting connection with the Hilbert scheme of regular
Sn orbits in (C 2 )n , which we shall discuss briefly. Given a set S of n distinct
points in the plane, its n! permutations describe a regular

orbit of Sn in (C 2 )n .
n! 2 n
The ideal J of the orbit is a point of Hilb (C ) . Let Zn be the closure in


Hilbn! (C 2 )n of the set of such points. It is not hard to show that the set
 MACDONALD POLYNOMIALS AND GEOMETRY 227

of ideals J which are Sn
stable and for which C[x, y]/J has a given character
is closed in Hilbn! (C 2 )n , so for every J ∈ Zn , C[x, y]/J affords the regular
representation of Sn .
Now there is a natural map from Zn to Hilbn (C 2 ), which may be described
as follows. Since C[x, y]/J affords the regular representation, its only invariants
are the constants. This means that modulo J we have ph,k (x, y) = ch,k for
some constant ch,k , for all h, k. By Weyl’s theorem, the Sn−1 -invariants in
C[x, y], for the action of Sn−1 on x2 , y2 through xn , yn , are generated by x1 , y1 ,
and the polarized power-sums ph,k (x2 , y2 , . . . , xn , yn ). Modulo J, the latter are
congruent to ch,k −xh1 y1k , so the Sn−1 invariants of C[x, y]/J are generated by x1
and y1 . In other words, C[x1 , y1 ]/(J ∩ C[x1 , y1 ]) = (C[x, y]/J)Sn−1 . It follows
that J ∩ C[x1 , y1 ] belongs to Hilbn (C 2 ), after identifying x1 , y1 with x, y.
The above construction defines the map Zn → Hilbn (C 2 ), which also has the
follow geometric description. Let W be the universal family over Zn . It has a
natural Sn -action, in which every fiber affords the regular representation. Then
W/Sn−1 is flat and finite of degree n over Zn , and by the above calculation can
be identified with a family of subschemes of C 2 . The map Zn → Hilbn (C 2 ) is
the one given by the universal property of Hilbn (C 2 ), for the family W/Sn−1 .
If the equivalent conditions of Theorem 4.5 hold, then Xn is a flat family of
subschemes of (C 2 )n , of degree n! over Hilbn (C 2 ). The universal property of



Hilbn! (C 2 )n then yields a map Hilbn (C 2 ) → Hilbn! (C 2 )n , whose image lies
in Zn . Generically, for sets S of n distinct points in C 2 and their corresponding
regular orbits in (C 2 )n , these two maps are mutually inverse. Hence, assuming
the n! conjecture, they are inverse everywhere, and the natural map Zn →
Hilbn (C 2 ) is an isomorphism.
Conversely, if Zn → Hilbn (C 2 ) is an isomorphism, then its inverse defines a
family X 0 ⊆ Hilbn (C 2 ) × (C 2 )n that is flat over Hilbn (C 2 ) and coincides with
Xn generically. But then X 0 is reduced and hence equal to Xn , so Xn is flat,
and the n! conjecture holds.
It would even suffice to show that Zn → Hilbn (C 2 ) is injective. For it is proper
and birational, hence surjective, and a bijective morphism onto a nonsingular
variety (or any normal variety) is an isomorphism, by Zariski’s theorem. To
summarize, we have
Proposition 4.6. The equivalent conditions of Theorem 4.5, for all partitions
µ of n, are also equivalent to:
(4) The natural map Zn → Hilbn (C 2 ) is injective.
(5) This map is an isomorphism.
Proposition 4.6 implies that the n! conjecture is equivalent to an instance of a
conjecture of Nakamura [1996], cited in [Reid 1997], connected with the McKay
correspondence, as we now explain.
Let G be a finite subgroup of SL(V ), where V = C n . For any finite linear
group action on V , V /G = Spec O(V )G is Cohen–Macaulay, and its canonical
228 MARK HAIMAN

δ
module is the isotypic component O(V ) of O(V ), where δ is the determinant
Vn
representation of G on (V ). For G ⊆ SL(V ) the determinant is trivial and
the canonical sheaf ωV /G is equal to OV /G , that is, V /G is Gorenstein.
A resolution of singularities H→V /G is said to be crepant (from “not dis-
crepant”) if ωH = OH . The Gorenstein condition on V /G is necessary, but
not sufficient, for a crepant resolution to exist. When they do exist, crepant
resolutions need not be unique, but they do enjoy a good minimality property:
given
Z −→ H −→ V /G,
f
if Z and H are both crepant resolutions, then f is an isomorphism.
The McKay correspondence is a conjecture to the effect that if H is a crepant
resolution of V /G, then the dimension of the cohomology ring of the complex
analytic manifold H is equal to the number of conjugacy classes (or irreducible
representations) of G. A refinement of this given in [Batyrev and Dais 1996]
specifies the dimension of each cohomology group separately. For V = (C 2 )n
and G = Sn , the Hilbert scheme Hilbn (C 2 ) is a crepant resolution, by Lemma
6.16, and the computation of its cohomology in [Ellingsrud and Strømme 1987]
(see Lemma 6.7) verifies that the McKay correspondence holds in this case.
Returning to the general G, if x ∈ V is chosen generically, then its G-orbit has
N = |G| elements, and thereby defines a point of the Hilbert scheme HilbN (V ).
Nakamura defines the G-Hilbert scheme HilbG (V ) to be the closure of the locus in
HilbN (V ) consisting of such orbits. There is a canonical morphism HilbG (V ) →
V /G. Nakamura’s conjecture is as follows.
Conjecture 4.7. HilbG(V ) is a crepant resolution of V /G whenever one exists.
The relevance of this to the McKay correspondence is that there are vector
bundles on HilbG (V ) canonically associated to the irreducible representations of
G. It is expected that when the conjecture applies, they will form a basis of the
Grothendieck group, and the latter will be isomorphic to the cohomology ring,
establishing the McKay correspondence for the resolution HilbG (V ).
In our situation, Nakamura’s HilbG (V ) is our Zn , and we have already estab-
lished the factorization
Zn → Hilbn (C 2 ) → S n C 2 = V /G.
If Nakamura’s conjecture holds in this case, then both Zn and Hilbn (C 2 ) are
crepant resolutions, hence they are isomorphic. Conversely if Zn ∼ n
= Hilb (C 2 ),
then obviously Zn is a crepant resolution. The equivalent conditions of Theo-
rem 4.5 and Proposition 4.6 are therefore also equivalent to Conjecture 4.7 in
the case V = (C 2 )n , G = Sn .
 MACDONALD POLYNOMIALS AND GEOMETRY 229

5. Frobenius Series
Let T denote the two dimensional algebraic torus group, i.e., the multiplicative
group C ∗ × C ∗ . It acts algebraically on C 2 by the rule

(t, q) · (x, y) = (tx, qy), for (t, q) ∈ T, (x, y) ∈ C 2 .

The universal construction of the Hilbert scheme is functorial with respect to

automorphisms of C 2 , so T also acts on Hilbn (C 2 ). This action sends a sub-
scheme S ⊆ C 2 to (t, q) · S, so on ideals I ⊆ C[x, y] it is given by the pullback
through the ring endomorphism (t, q), that is,

(t, q) · I = I(x/t, y/q).

Similarly T acts on the the schemes U , U ×n , (C 2 )n and Xn , and the various

maps between these schemes are T-equivariant.
Observe that a polynomial p(x, y) is doubly homogeneous of degree (r, s) if
and only if it is an eigenfunction for the T action with eigenvalue tr q s . Hence the
bivariate Hilbert series of a finite-dimensional doubly graded space of polynomi-
als D is simply the character of the T-action, as a function of q and t. Similar
considerations apply to the Frobenius series when Sn acts, as it is just a gener-
ating function for the Hilbert series of the the various Sn -isotypic subspaces.
In the geometric situation we have to deal with T actions on local rings and
sheaves of modules that may be neither graded nor finite-dimensional. For this
we need recourse to a “formal” Hilbert series that captures the naive T character
in the finite dimensional case, and extends to the more general setting.
Definition. Let R be the local ring of a scheme X of finite type over C at
a closed point x with maximal ideal m. Assume X is nonsingular at x and
that x is an isolated fixed point for an algebraic action of T on X. Let M be
a finitely generated R-module with an equivariant T action. Then the formal
Hilbert series of M is given by
P
(−1)i tr(TorRi (M, C), λ)
HM (q, t) = i , with λ = (t, q) ∈ T. (5.1)
det(m/m2 , 1 − λ)

i (M, C)
To see that the definition is sound, observe first that the modules TorR
and the cotangent space m/m2 are finite dimensional representations of T, so
the trace and determinant in the formula make sense. Since R is regular, the
syzygy theorem implies that the sum in the numerator is finite. Since x is an
isolated fixed point, the T action on the cotangent space does not have 1 as an
eigenvalue, so the denominator does not vanish identically, but is a product of
factors of the form (1−tr q s ) with r, s not both zero. Note that HM (q, t) is a
rational function of q and t.
Proposition 5.1. (1) If 0 → M → N → P → 0 is an exact sequence, then
HN = HM + HP .
230 MARK HAIMAN

(2) If M has finite length then HM is its ordinary bivariate Hilbert series as a
doubly graded space.
Proof. (1) follows from the long exact sequence for Tor and the additivity of
the trace on exact sequences. In view of (1), (2) reduces to the case M = C.
Since R is regular, we have in this case from the Koszul resolution of C that
Vi
Tori (M, C) ∼= M ⊗ Tx∗ , where Tx∗ = m/m2 is the cotangent space. Here we
have kept the one-dimensional factor M explicit, since T might not act trivially
on it. Let tr1 q s1 , . . . , trd q sd be the eigenvalues of T on Tx∗ , repeated according to
their multiplicities. Then the denominator is
Y X
(1−trj q sj ) = (−1)i ei (tr1 q s1 , . . . , trd q sd ),
j i

while the numerator is the same thing multiplied by the T character of M . 

When M has an Sn action commuting with the T action, we can define a formal
Frobenius series in an analogous manner. The character of a finite dimensional
Sn × T module V is now a function tr(V, (w, λ)) of both w ∈ Sn and λ =
(t, q) ∈ T, which we can regard as a Q(q, t)-valued character on Sn . With the
same definition of the Frobenius map Φ as in (3.3), Φ ch V is now a symmetric
function with coefficients in Q(q, t). Indeed if we identify the T action with
a double grading of V then Φ ch V in this sense is our earlier Frobenius series
FV (X; q, t).
Definition. With R, X, x, and M as in the definition of formal Hilbert series,
and an action of Sn by R-module automorphisms of M commuting with the T
action, the formal Frobenius series of M is given by
P
(−1)i Φ ch(TorRi (M, C))
FM (X; q, t) = i , with λ = (t, q) ∈ T.
det(m/m2 , 1 − λ)
The formula makes sense for the same reasons as in the Hilbert series case and
the analog of Proposition 5.1 holds.
Proposition 5.2. (1) If 0 → M → N → P → 0 is an exact sequence, then
FN = FM + FP .
(2) If M has finite length then FM is its ordinary Frobenius series as a doubly
graded Sn representation.
Proof. The same as the previous Proposition, except that for (2) we reduce to
the case that M is an irreducible Sn × T-equivariant R module. This means M
is an irreducible representation of Sn (with trivial T action) tensored over C by
a trivial R module C (with some one-dimensional T action). 
Further properties of the formal Frobenius series are given by the next proposi-
tion. The last one is especially important, as it provides a geometric interpreta-
tion of plethystic substiution.
 MACDONALD POLYNOMIALS AND GEOMETRY 231

Proposition 5.3. The formal Frobenius series FM has the following properties:
(1) If the Sn character χλ has multiplicity zero in M/mM , then hsλ , FM i = 0.
(2) Let V be a finite-dimensional doubly graded Sn module. Then (tensoring
over C) we have FV ⊗M = FV ∗ FM , where ∗ is the internal product of sym-
metric functions.
(3) Suppose M is an Sn -equivariant S-module, where S is a finite R-algebra
with an Sn action. Suppose x1 , . . . , xn ∈ S is an M -regular sequence, that T
acts on the elements xi by (t, q) · xi = txi , and that Sn acts on these elements
by permuting them. Then FM/(x)M (X; q, t) = FM [X(1−t); q, t].
Proof. (1) Let Θλ be the Reynolds operator, the central idempotent in the
group algebra of Sn that acts in each representation as the projection on the Sn
isotypic component with character χλ . If M is a finitely generated R-module
with Sn × T action then it has a canonical isotypic decomposition
M
M= Θλ M,
|λ|=n

and each Θλ M is also a finitely generated R module with Sn × T action. The

decomposition means that the identity functor is naturally isomorphic to the
direct sum of the functors M 7→ Θλ M . In particular the functors M 7→ Θλ M
are exact, and commute with Tor, since the Sn action does.
Now, comparing the definitions of the formal Hilbert and Frobenius series,
and using the fact that Φχλ = sλ , we see that
1
hsλ , FM i = HΘλ M .
χλ (1)
By Nakayama’s Lemma, if Θλ (M/mM ) = Θλ M/mΘλ M = 0, then Θλ M = 0.
(2) Recall that the internal product ∗ is defined by pλ ∗ pµ = hpλ, pµ ipλ , and
satisfies the identity
Φ(χ ⊗ µ) = Φ(χ) ∗ Φ(µ)
for all characters χ, µ. From this it is clear that (2) holds in the case where M has
finite length, by Proposition 5.2. For the general case we have Tori (V ⊗ M, C) =
V ⊗ Tori (M, C), which reduces the identity to the corresponding one with the
finite-length modules Tori (M, C) in place of M .
(3) Let V denote the space spanned by elements of the regular sequence x;
as an Sn × T module it affords the permutation representation of Sn tensored
by the one-dimensional representation of T with character t. Since x is a reg-
ular sequence we have the following exact sequence, the Koszul resolution of
M/(x)M :
Vn V1 V0
0→M⊗ V → · · · → M ⊗ V → M ⊗ V → M/(x)M → 0, (5.2)
where the tensor products are over C. The maps in the Koszul complex are
Vk
Sn × T equivariant if we take the T action on V to be multiplication by
232 MARK HAIMAN

tk . From part (1) of Proposition 5.2, together with (2) above, we deduce that
P V
FM/(x)M = FM ∗ g(X; t), where g(X; t) = k (−1)k tk Φ ch k V .
Now I claim that g(X; t) = hn [X(1−t)]. To prove this it suffices to show that
for each power-sum pλ, |λ| = n, we have

hg(X; t), pλ i = hn [X(1−t)], pλ = pλ [1−t].

The second equality here comes from (2.6). For any character χ, it follows
from the definition of the Frobenius map that hΦχ, pλi = χ(w), where w is a
permutation whose cycle lengths are the parts of λ. The symmetric group acts
Vk
on V by signed permutations of the basis of monomials xi1 ∧ · · · ∧ xik , and
such a monomial is stabilized by w if and only if, for each cycle of w, either all
or none of the corresponding variables appear in the monomial. In that case
w(xi1 ∧ · · · ∧ xik ) = ±xi1 ∧ · · · ∧ xik , the sign being (−1)k+r if the variables for r
Vk
of the cycles appear. Hence the trace of w on V is the sum of (−1)k+r over
all collections of the parts of λ that add up to k, where r is the number of parts
included. It follows that
X V Y
(−1)k tk (ch k V )(w) = (1−tλi ) = pλ[1−t].
k i

Finally, we have

hn [X(1−t)] ∗ pλ = hn [X(1−t)], pλ pλ = pλ[1−t]pλ = pλ [X(1−t)],

where the last equality holds because of the identity pk [AB] = pk [A]pk [B]. Since
both sides are linear in f it follows that

hn [X(1−t)] ∗ f = f[X(1−t)]

for all symmetric functions f. 

The remainder of this section is devoted to the proof of the following theorem.
Theorem 5.4. If the n! conjecture holds for µ, then the Frobenius series of Rµ
is given by
FRµ = H̃µ ,
so the Macdonald positivity conjecture holds for Kλµ (q, t), for all λ.
For the rest of the discussion we assume the n! conjecture holds for µ, so that
Xn is locally Cohen–Macaulay at the point Qµ .
We take R to be the local ring of Hilbn (C 2 ) at Iµ . Note that Hilbn (C 2 ) is
nonsingular, and Iµ is a T fixed point, since this is equivalent to Iµ ⊆ C[x, y] be-
ing doubly homogeneous, and thus spanned by monomials. As remarked earlier,
the ideals Iµ are the only such ideals, so Iµ is an isolated fixed point.
The local ring S of Xn at Qµ is a finite R-algebra on which T acts equiv-
ariantly. The symmetric group Sn also acts on S by R-algebra automorphisms,
commuting with the T action.
 MACDONALD POLYNOMIALS AND GEOMETRY 233

Via the map Xn → (C 2 )n , the coordinates x1 , y1 , . . . , xn , yn on (C 2 )n define

global regular functions on Xn and thus elements of the local ring S.
Lemma 5.5. If Xn is Cohen–Macaulay at Qµ then y1 , . . . , yn is a regular se-
quence.
Proof. We are to show that y = y1 , . . . , yn cut out a complete intersection in
Xn . Since dim(Xn ) = 2n we must show that dim V (y) = n. Now V (y) consists
of those points (I, P1 , . . . , Pn ) ∈ Xn for which all the points Pi lie on the x-axis,
which is the same as saying that V (I) lies on the x-axis.
Ellingsrud and Strömme [Ellingsrud and Strømme 1987], studying the coho-
mology of Hilbn (P 2 ), constructed a cell decomposition that contains within it a
cell decomposition of the subset HX ⊆ Hilbn (C 2 ) consisting of points I for which
V (I) lies on the x-axis. Every cell in their decomposition of HX has dimension n.
Since the locus V (y) ⊆ Xn is finite over HX its dimension is n as well. 
∼ Rµ ,
The local ring S of Xn at Qµ is not only finite over R, it is free, and S/mS =
by the proof of Theorem 4.5, where m is the maximal ideal of R. By Nakayama’s
lemma, S is freely generated as an R module by any subspace D complementary
to the ideal mS. We may choose D to be Sn × T stable (in fact, we can choose
D to be the space of derivatives Dµ ), and then D will have the same Frobenius
series as Rµ . This shows that
FS (X; q, t) = FRµ (X; q, t)HR (q, t). (5.3)
Incidentally, the quantity HR (q, t) has a tantalizing explicit value. In [Haiman
1998] we constructed an explicit system of doubly homogeneous regular local
parameters for R. From their degrees one obtains
1
HR (q, t) = Q Q ,
s∈D(µ) (1−q −a(s) t1+l(s) ) s∈D(µ) (1−q
1+a(s) t−l(s) )

where the arms a(s) and legs l(s) are as in (2.16). After the replacement q 7→
q −1 the first factor in the denominator is exactly the normalizing factor for the
definition of Macdonald’s integral forms. This coincidence is a typical example
of the links between the numerology associated with Macdonald polynomials and
geometrically significant quantities attached to the Hilbert scheme.
Now consider the ring S/(y). Just as S is generated as an R module by any
subspace representing S/mS, S/(y) is generated (no longer freely, however) by
representatives of S/((y)+mS) = Rµ /(y). This last space can be identified with
the component of Rµ, or of Dµ , homogeneous of degree zero in y. As mentioned
in the proof of Proposition 3.7, this space is well-understood, and its Frobenius
series is
  X n(µ)
tn(µ) Qµ X/(1−t−1 ); t−1 = t Kλµ (t−1 )sλ ,
λ
where Qµ is a Hall–Littlewood polynomial and Kλµ (t) denotes the classical one-
variable Kostka coefficient. In particular, since Kλµ (t) = 0 unless λ ≥ µ, the
234 MARK HAIMAN

space Rµ /(y) contains only those Sn representations χλ with λ ≥ µ. It follows,

by Proposition 5.3, part (1), that

FS/(y) ∈ Q(q, t){sλ : λ ≥ µ}.

Now using Proposition 5.3, part (3), with q in place of t, this implies

FS [X(1−q)] ∈ Q(q, t){sλ : λ ≥ µ},

and hence, by (5.3),

FRµ [X(1−q)] ∈ Q(q, t){sλ : λ ≥ µ}.

Everything we have done applies symmetrically, with x in place of y and t in

place of q, to show that also

FRµ [X(1−t)] ∈ Q(q, t){sλ : λ ≥ µ0 }.

Finally, since Rµ affords the regular representation (this follows from the n!
conjecture by Proposition 4.4), its only Sn invariants are the constants, so

hFRµ , s(n)i = 1.

By Proposition 2.6 these three conditions imply that FRµ = H̃µ , and the proof
of Theorem 5.4 is complete.

6. The Ideals J and J m

Let R = C[x, y] = C[x1 , y1 , . . . , xn , yn ], and let J denote the ideal in R gener-
ated by all Sn -alternating polynomials, that is, by the polynomials ∆D (x, y) of
(3.1). Since any alternating polynomial must vanish whenever two of the points
(xi, yi ) and (xj , yj ) coincide, it follows that
\
J⊆ (xi − xj , yi − yj ), (6.1)
i<j

and more generally

\
Jm ⊆ (xi − xj , yi − yj )m . (6.2)
i<j

We shall denote the ideal on the right hand side of (6.2) by J (m) ; it is the m-th
symbolic power of J (1) .
Conjecture 6.1. We have J m = J (m) for all m, i .e., we have equality in (6.2).
Here is a result that at least gives the impression of reducing the conjecture to
something simpler.
Proposition 6.2. Suppose that for all n ≥ 3, (x1 − x2 , x2 − x3 ) is a regular
sequence for the R module J m . Then J m = J (m) .
 MACDONALD POLYNOMIALS AND GEOMETRY 235

Proof. The proof is by induction on n. Note that for n = 1 and n = 2 we

trivially have J m = J (m) , and we have the remaining cases through n − 1 by
induction.
First consider the situation locally at a point P ∈ C[x, y] where the (xi , yi)
are not all equal. Without loss of generality we can assume that none of
(x1 , y1 ), . . . , (xr , yr ) is equal to any of (xr+1 , yr+1 ), . . . , (xn , yn ). In the local
ring RP , the differences xi − xj and yi − yj are invertible whenever i is in the
first group and j in the second. Hence J (m) reduces locally to the product of
the ideals J (m) in the first r indices and the last n − r separately.
Less obvious, but still true, is that J, and hence J m , decomposes similarly. To
see this, let g be a generator of J 0 = J(x1 , y1 , . . . , xr , yr )J(xr+1 , yr+1 , . . . , xn , yn ),
alternating in the first r and last n − r indices, i.e., the subgroup Sr × Sn−r ⊆ Sn
acts on g by the sign character. Let h be any polynomial that belongs to the
localization JQ at every point Q 6= P in the Sn orbit of P , but does not vanish
at P . Now f = Alt gh belongs to J. The terms in the alternation corresponding
to elements w ∈ Sn that do not stabilize P belong to JP , by construction of h.
Since g is already alternating with respect to the stabilizer of P , the remaining
P
terms sum to g wP =P wh, and the sum here is invertible in RP . This shows
g ∈ JP , and so JP0 ⊆ JP . The reverse inclusion J ⊆ J 0 is clear.
Using the induction hypothesis, we conclude that J m = J (m) locally outside
the locus V where all n points coincide. Now since x1 − x2 and x2 − x3 belong
the ideal of V , our hypothesis implies depthV J m ≥ 2, and the local cohomology
exact sequence for the sheaf of ideals J˜m associated to J m gives

0 = HV0 (J˜m ) → H 0 (C 2 , J˜m ) = J m → H 0 (U, J˜m ) → HV1 (J˜m ) = 0, (6.3)

where U = C 2 \ V . Thus J m = H 0 (U, J˜m ) = H 0 (U, J f

(m) ). The latter is the

ideal of all polynomials whose restrictions to U belong locally to J (m) , so we

have shown J m ⊇ J (m) . As we had J m ⊆ J (m) to begin with, we must have
J m = J (m) . 

There is of course nothing special about the choice of x1 − x2 and x2 − x3 in the

above Proposition; it’s just an explicit way to guarantee that depthV J m ≥ 2.
This would also follow if depthV (R/J m ) ≥ 1, which means the ideal of V contains
an element that is a non-zero-divisor modulo J m . Note, by the way, that the
proof of Proposition 6.2 works equally well with more than two sets of variables.
Some explorations we have done for small values of n using the computer alge-
bra system Macaulay [Bayer and Stillman 1989] suggest the following conjecture.

Conjecture 6.3. If J denotes the ideal generated by the Sn alternants in

C[x, y, . . . , z], for any number of sets of n variables, and V is the locus where
all the points (xi, yi , . . . , zi), (xj , yj , . . . , zj ) coincide, then x1 − x2 , x2 − x3 , . . . ,
xn−1 − xn is a maximal J m -regular sequence in the ideal of V , for all m. In
particular , depthV J m = n − 1.
236 MARK HAIMAN

We remark that if the sequence x in question is regular then it is maximal:

modulo (x)J, the Vandermonde determinant v(x) is annihiliated by the ideal of
V , so depthV J/(x)J = 0.
The relevance of all this to Xn and the n! conjecture is given by the following
proposition.

Proposition 6.4. The isospectral Hilbert scheme Xn is the blowup Proj R[tJ]
of (C 2 )n at the ideal J.

Proof. We only outline the proof, as the analogous result for the ordinary
Hilbert scheme Hilbn (C 2 ) was given in [Haiman 1998], and the proof carries
over to Xn with only superficial modifications.
First, one shows that the pullback of the ideal J to Xn becomes locally prin-
Vn
cipal. In fact, it can be identified with B, where B is the tautological bundle
n 2
(of Hilb (C ), lifted to Xn ). By the universal property of the blowup, this gives
a morphism from Xn to Proj R[tJ]. This map is projective and generically an
isomorphism, so it’s surjective. To show it’s also a closed embedding, we have
to show that all regular functions on Xn are pulled back from Proj R[tJ].
The regular functions on Xn are the coodinates xi , yi, which come from R, and
the lifts of regular functions on the Hilbert scheme. But the proof of the result
for Hilbn (C 2 ) shows the latter are generated by fractions of the form ∆D /∆µ ,
that are (local) regular functions on Proj R[tJ]. 

In d sets of variables, the above proposition applies as well to the isospectral

Hilbert scheme of points in C d , with an important qualification. For general d
these Hilbert schemes are not irreducible and even have components of dimension
greater than dn [Iarrobino 1972]. What the blowup construction gives is the
component that is the closure of the locus corresponding to reduced subschemes
of n distinct points in C d . We suspect that this generic component may have
good geometric properties, indeed may be Cohen–Macaulay and even Gorenstein
for all d. Note that this would not imply the n! conjecture in more sets of
variables, since the generic component of Hilbn (C d ) may be singular, and thus
the isospectral Hilbert scheme need not be flat over it.
In the remainder of this section we prove the following reduction of the n!
conjecture to Conjecture 6.3.

Theorem 6.5. If Conjecture 6.3 holds in two sets of variables for all m and n,
then Xn is Cohen–Macaulay and normal for all n.

The proof is by induction on n, using geometric properties of the nested Hilbert

scheme H n−1,n to be defined shortly. We employ a local cohomology argument
in the same spirit as the proof of Proposition 6.2. For this we need a large
open set where we may assume the result by induction, which the next lemma
provides.
 MACDONALD POLYNOMIALS AND GEOMETRY 237

Lemma 6.6. Let P = (I, P1 , . . . , Pn) be a point of Xn . Let the distinct points
among P1 , . . . , Pn be Q1 , . . . , Qk , with multiplicities r1 , . . . , rk . Then in Xn there
is a neigborhood of P isomorphic to an open set in the product Xr1 × · · · × Xrk .
Proof. Without loss of generality we can assume Q1 = P1 = · · · = Pr1 ,
Q2 = Pr1 +1 = · · · = Pr1 +r2 , and so on. For our neighborhood of P we can
take the preimage in Xn of the open set U ⊆ (C 2 )n of points where the only
coincidences Pi = Pj that occur have i, j within one of these k consecutive
blocks. Then the result is clear from Proposition 6.4, together with the product
decomposition, valid on U , of the ideal J as J1,...,r1 Jr1 +1,...,r1 +r2 · · · from the
proof of Proposition 6.2. 
For some dimension arguments below we will need the following results.
Lemma 6.7. There is a decomposition of Hilbn (C 2 ) into locally closed affine
cells Cµ , such that every point of Cµ contains Iµ in the closure of its T orbit ,
and dim Cµ = n + l(µ), where l(µ) is the number of parts of µ.
Lemma 6.8. There is a decomposition of the zero-fiber H0n = τ −1 (0) ⊆ Hilbn (C 2 )
into locally closed affine cells Cµ0 , such that every point of Cµ0 contains Iµ in the
closure of its T orbit , and dim Cµ0 = l(µ) − 1.
Proof. See [Ellingsrud and Strømme 1987]. 
Lemma 6.9. Let Gr be the (closed) locus of ideals I ∈ Hilbn (C 2 ) for which some
point of V (I) has multiplicity at least r. Then Gr has codimension r − 1, and
has only one irreducible component of maximal dimension.
Proof. It is known [Briançon 1977] that the zero-fiber H0n = τ −1 (0) is irre-
ducible of dimension n − 1. The locus where all the points coincide is just the
product of C 2 (for the choice of origin) by H0n , so it is irreducible of dimension
n + 1.
By Lemma 6.6, it follows that the (locally closed) locus where the multiplicities
P
are r1 , . . . , rk has dimension i (ri + 1) = n + k and codimension n − k. If
one multiplicity is at least r, this codimension is at least r − 1, with equality
only for multiplicities r, 1, . . ., 1. Again by Lemma 6.6, the locus in Xn where
P1 = · · · = Pr , and the other points are distinct from P1 and each other, is
irreducible. It surjects on the locus in Hilbn (C 2 ) where the multiplicities are
r, 1, . . ., 1, so the latter is irreducible as well. 
As a step toward the Cohen–Macaulay property we need normality results for
Xn and Un .
Definition. An ideal I ∈ Hilbn (C 2 ) is curvilinear if the local rings OS,P have
embedding dimension 1, i.e., their maximal ideals are principal.
This is equivalent to S = V (I) being a subscheme of a smooth curve in C 2 ,
whence the name.
238 MARK HAIMAN

Lemma 6.10. The locus W of curvilinear ideals I ∈ Hilbn (C 2 ) is open and equal
S
to z Wz , where z = ax + by is a linear form, and Wz is the open set of ideals
I such that {1, z, . . . , z n−1 } is a basis of C[x, y]/I.
Proof. If I is curvilinear, then for a generically chosen linear form z, the values
z(P ) at distinct points P ∈ V (I) will be distinct, and for each P , z − z(P ) will
be a local parameter generating the maximal ideal mP ⊆ OS,P . This implies
that, as a C[z] module and hence as a ring,
.Y
C[x, y]/I ∼
= C[z] (z − z(P ))rP , (6.4)
P

where rP is the multiplicity of P , and therefore {1, z, . . ., z n−1 } is a basis of

C[x, y]/I. Conversely, if {1, z, . . . , z n−1 } is a basis, then z generates C[x, y]/I,
and I contains a monic polynomial of degree n in z, so (6.4) holds and I is
curvilinear. 
Lemma 6.11. The universal scheme U over Hilbn (C 2 ) is Cohen–Macaulay and
normal .
Proof. U is Cohen–Macaulay because it is flat over Hilbn (C 2 ). Hence it is
normal if its singular locus has codimension at least 2.
Now I claim that over the curvilinear locus W , U is nonsingular. After a
linear transformation of C 2 , we can restrict to Wx . For I ∈ Wx , we have y and
xn congruent mod I to unique polynomials of degree at most n − 1 in x, so I
contains elements


xn − e1 xn−1 + e2 xn−2 − · · · + (−1)n en , y − (an−1 xn−1 + · · · + a1 x + a0 ) , (6.5)
where the parameters ei and ai are regular functions of I on Wx . On the
other hand these two equations clearly generate a complete intersection ideal
I ⊆ C[x, y] modulo which 1, x, . . . , xn−1 are a basis, so they determine I. This
exhibits Wx explicitly as an affine cell with coordinates ei , ai . (As a matter of
fact, Wx is the cell C(1n) in Lemma 6.7.) Moreover, regarded as equations on
Wx × C 2 = Spec C[e, a, x, y], equations (6.5) define the universal family U .
Viewing the first equation as eliminating en and the second as eliminating y
we conclude that the open subset of U lying over Wx is
Spec C[x, e1 , . . . , en−1 , a0 , . . . , an−1],
and in particular is nonsingular. It follows that U is nonsingular over the whole
curvilinear locus.
Finally, if I is not curvilinear, then some point of V (I) has to have multiplicity
at least 3. By Lemma 6.9 this occurs only on a locus of codimension 2. Since U
is finite over Hilbn (C 2 ), its singular locus also has codimension at least 2. 
Lemma 6.12. If Conjecture 6.3 holds for all m and any given n, then Xn is
normal .
 MACDONALD POLYNOMIALS AND GEOMETRY 239

Proof. Recall that an ideal J in a normal domain R is said to be integrally

closed if every element x ∈ R satisfying
xn ∈ Jxn−1 + J 2 xn−1 + · · · + J n (6.6)
already belongs to J. The above condition for J m is equivalent to saying that
tm x belongs to the integral closure of R[tJ] in R[t], so all the ideals J m are
integrally closed if and only if R[tJ] is normal.
For our R and J, Conjecture 6.3 implies J m = J (m) , by Proposition 6.2. It
is well-known that the powers of an ideal generated by a regular sequence are
integrally closed, and it is obvious that an intersection of integrally closed ideals
is integrally closed, so J (m) is integrally closed.
This shows that R[tJ] is normal, so Xn = Proj R[tJ] is, by definition, arith-
metically normal in the given projective embedding. In particular it is normal.

Now we come to the geometric construction that supplies the inductive machin-
ery.
Definition. The nested Hilbert scheme H n−1,n is the subvariety of pairs
H n−1,n = {(In−1 , In ) : In−1 ⊇ In } ⊆ Hilbn−1 (C 2 ) × Hilbn (C 2 ).
Proposition 6.13 [Cheah 1998; Tikhomirov 1992]. The nested Hilbert scheme
H n−1,n is irreducible of dimension 2n and nonsingular .
If (In−1 , In ) is a point of Hn−1,n then the corresponding subscheme V (In−1 ) ⊆
C 2 is a subscheme of V (In ), so the multiset τ (In ) contains τ (In−1 ) along with
one additional point, or else with the multiplicity of one of the original points
increased by 1. So if the spectrum of In−1 is (x1 , y1 ), . . . , (xn−1 , yn−1 ) then
that of In is (x1 , y1 ), . . . , (xn−1, yn−1 ), (xn , yn ) for a distinguished point (xn , yn ).
Now both the Sn−1 invariants ph,k (x1 , y1 , . . . , xn−1 , yn−1 ) and the Sn invari-
ants ph,k (x1 , y1 , . . . , xn, yn ) are regular functions on H n−1,n, hence so are xn =
p1 (x1 , . . . , xn) − p1 (x1 , . . . , xn−1) and similarly yn . This means we have a mor-
phism
H n−1,n → C 2 = Spec C[xn, yn ],
mapping a pair to its distinguished point. Of course (xn , yn ) ∈ V (In ), and by
suitable choice of In−1 , given In , the distinguished point can be any point of
V (In ). Hence the combined map H n−1,n → C 2 × Hilbn (C 2 ) factors H n−1,n →
Hilbn (C 2 ) through a surjective morphism
α: H n−1,n → U (6.7)
to the universal scheme U over Hilbn (C 2 ). Where the n points are distinct, this
map is locally an isomorphism, so it is birational.
The above map and the map H n−1,n → Hilbn (C 2 ) are projective. In fact,
given In , In−1 is determined by its single generator mod In , so H n−1,n is a
240 MARK HAIMAN

subvariety of the projective space bundle P (B), where B is the tautological

bundle over Hilbn (C 2 ). More precisely, given In and P = (xn , yn ), the possible
ideals In−1 correspond one-to-one with length-1 ideals in the local ring OS,P of
V (In ) at P . Such ideals are simply the 1-dimensional subspaces of the socle,
soc OS,P = (0 : mP ). Thus each fiber of the map (6.7) is a projective space
P(soc OS,P ), of dimension dim soc OS,P − 1.
Lemma 6.14. If the dimension of the fiber of α: H n−1,n
→ U over a point
(I, P ) ∈ U is d, then the multiplicity of P is at least d+2
2 .

Proof. We are to show that if T = C[x, y]/I is a local ring
of finite length,
with d + 1 = dim soc T , then the length of T is at least d+2
2
. This is equivalent
to showing that if there exists a fiber of the map H n−1,n → U with dimension


at least d, then n ≥ d+2
2
. Now by the upper-semicontinuity of fiber dimension,
and the fact (Lemma 6.7) that every I has one of the ideals Iµ in the closure of
its T orbit, the maximal fiber dimension must occur at some Iµ . There we see
immediately that the dimension of the socle is the number of corners of µ, so the
result reduces to the fact that if the diagram of a partition of n has k corners


then n ≥ k+1 2
. 
Lemma 6.15. The map α: H n−1,n → U restricts to an isomorphism outside a
locus of codimension 2 in H n−1,n .
Proof. First note that the 2-dimensional and higher fibers of α form a locus
of codimension

at least 3, by Lemmas 6.9 and 6.14, since for d ≥ 2 we have
d+2
2 − 1−d ≥ 3.
For In curvilinear, soc OS,P is always 1-dimensional, so α restricts to an bijec-
tive morphism on the curvilinear locus, which is then an isomorphism by Zariski’s
theorem and Lemma 6.11.
As noted in the proof of Lemma 6.11, the noncurvilinear locus in Hilbn (C 2 )
is contained in G3 , so its codimension is at least 2. If it had codimension ex-
actly 2 then it would contain the whole codimension 2 component of G3 , and
in particular, every point where the multiplicities are 3, 1, . . . , 1. But there are
clearly curvilinear subschemes with these multiplicities, so the co-dimension of
the noncurvilinear locus is at least 3. If its preimage in H n−1,n had a component
of codimension 1, then every fiber in that component would have dimension at
least 2, contradicting the observation made at the outset. Hence the locus where
In is noncurvilinear has codimension at least 2 in H n−1,n , and α restricts to an
isomorphism outside it. 
Lemma 6.16. The canonical sheaf ω of regular 2n-forms on Hilbn (C 2 ) is trivial ,
i .e., isomorphic to the structure sheaf O.
Proof. We again use the description in the proof of Lemma 6.11 of the open
set Wx of ideals I modulo which 1, x, . . ., xn−1 is a basis: Wx is an affine 2n-
cell with coordinates e1 , . . . , en , a0 , . . . , an−1, in terms of which I is generated at
 MACDONALD POLYNOMIALS AND GEOMETRY 241

each point by equations (6.5). On the locus where I is the ideal of a reduced
subscheme S = {(x1 , y1 ), . . . , (xn , yn )}, the first equation
xn − e1 xn−1 + e2 xn−2 − · · · + (−1)n en
Q
must be the polynomial i (x − xi ), so the parameters ei are the elementary
symmetric functions ei (x). From the second equation,
y − (an−1 xn−1 + · · · + a1 x + a0 ),
the ai are the coefficients of the interpolating polynomial φa (x) that satisfies
yi = φa (xi ) when the xi’s are all distinct.
It is well-known that the elementary symmetric functions satisfy de1 ∧ · · · ∧
den = v(x)dx1 ∧ · · · ∧ dxn , where v(x) is the Vandermonde determinant. In
particular, since the ei (x) and ei (y) are global regular functions on Hilbn (C 2 ),
and v(x)v(y) is Sn invariant, dx dy = dx1 ∧ · · · ∧ dxn ∧ dy1 ∧ · · · ∧ dyn =
v(x)−1 v(y)−1 de1 (x) ∧ · · · den (x)de1 (y) ∧ · · · den (y) makes sense as a rational
2n-form on Hilbn (C 2 ).
Moreover, the equations yi = φa(xi ) say that the vector (y1 , . . . , yn ) is the
product of (an−1 , . . . , a0 ) by the Vandermonde matrix, so da0 ∧ · · · ∧ dan−1 =
dy1 ∧ · · · ∧ dyn /v(x), and hence
de1 ∧ · · · ∧ den ∧ dan−1 ∧ · · · ∧ da0 = dx dy.
In particular, the rational 2n-form dx dy is regular and has no zeroes on Wx .
But dx dy is invariant under the action of SL2 on C 2 , so it follows that dx dy
is regular and nowhere vanishing on every Wz . Since we have already seen that
S
the complement of z Wz has codimension greater than 1, it follows that dx dy
is regular everywhere and vanishes nowhere, which shows that ω = O. 
Lemma 6.17. The canonical sheaf ω of regular 2n-forms on H n−1,n is isomor-
phic to L−1 , where L is the line bundle defined by the exact sequence
0 → L → Bn → Bn−1 → 0 (6.8)
induced on the tautological bundles over H n−1,n by the containment In−1 ⊇ In .
Proof. We are to show that the line bundle Lω is trivial, and it suffices to do
this on an open set whose complement has codimension ≥ 2. By Lemma 6.15, we
can use the open set where In is curvilinear and the map H n−1,n → U restricts
to an isomorphism, which means we can verify it on the curvilinear locus in U .
By Lemma 6.16 and duality for the finite, flat morphism π: U → Hilbn (C 2 ), we
have π∗(LωU ) ∼= (π∗ L−1 )∗ ωH = (π∗ L−1 )∗ , where ωH = O is the canonical sheaf
on the Hilbert scheme. So we have to show that π∗ L−1 ∼ = B ∗ as a B-module,
since π∗ OU = B.
Now let’s examine L as a subbundle of π ∗ B on the curvilinear locus in U . To
avoid confusion, since we already have regular functions x, y on U , we write x0 , y0
for the variables of B, so the fiber of π ∗ B at (I, P ) is C[x0 , y0 ]/I(x0 , y0 ). Then
242 MARK HAIMAN

the fiber of L at (I, P ) is the socle of the summand OS,P in C[x0, y0 ]/I(x0 , y0 ),
since the generator of In−1 mod I belongs to this socle, which is 1-dimensional.
Equivalently, the fiber of L is the ideal (0 : mP (x0 , y0 )) = (0 : (x0 − x, y0 − y))
in π ∗ B(I). Dualizing this, we see that L−1 = π∗ B ∗ /(x0 −x, y0 −y)π ∗ B ∗ , or
π ∗ B ∗ ⊗C[x0 ,y0 ] OU , where C[x0 , y0 ] acts on OU through the homomorphism x0 7→
x, y0 7→ y.
Now π∗ π ∗ B ∗ = B ⊗ B ∗ , so π∗L−1 = (B ⊗ B ∗ )/(x0 −x, y0 −y)(B ⊗ B ∗ ), where
x, y act through B and x0 , y0 act through B ∗ . But this is just another way of
writing π∗ L−1 = B ⊗B B ∗ = B ∗ . 
Lemma 6.18. If Xn is Cohen–Macaulay, then it is Gorenstein, and its canonical
Vn
sheaf ω is the line bundle O(−1), where O(1) = B.
Proof. By Proposition 4.2, Xn = Spec P as a scheme affine over Hilbn (C 2 ),
where P = σ∗ OXn is the image of the sheaf homomorphism φ: B ⊗n → (B ⊗n )∗ ⊗
O(1) in (4.7). If Xn is Cohen–Macaulay this is a vector-bundle homorphism. By
construction, this description of P means the bilinear pairing of vector bundles
P ⊗P → O(1), given by multiplication followed by alternation, is nondegenerate,
so P ∗ ∼
= P ⊗ O(−1). By duality for the flat, finite morphism σ: Xn → Hilb (C 2 ),
n

the canonical sheaf ωXn is the sheaf associated to the P module sheaf P ∗ ⊗ ωH ,
which is the same as P ⊗ O(−1), by Lemma 6.16. This shows ωXn = O(−1), and
since this is a line bundle, Xn is Gorenstein (by definition). 
We now have all the technical ingredients we need to prove Theorem 6.5. From
this point on we assume Conjecture 6.3 holds, and we assume Xn−1 is Cohen–
Macaulay by induction. We shall also assume n ≥ 4. There is no harm in this
since it is trivial to verify the n! conjecture for n ≤ 3.
To carry the induction forward we introduce the fiber product Yn indicated
by the diagram:
Yn −−−−→ H n−1,n
 
 
y y (6.9)

Xn−1 −−−−→ Hilbn−1 (C 2 ).

Since the bottom arrow is flat (by induction), so is the top one. Since Yn is
flat over H n−1,n and generically reduced, it is reduced. A point of Yn is a
pair of ideals (In−1 , In ) ∈ H n−1,n, together with the spectrum (x1 , xn ), . . . ,
(xn−1 , xn−1) of In−1 in some order. The coordinates of the remaining point
(xn , yn ) in the spectrum of In are regular functions on H n−1,n and hence on Yn ,
so we obtain a morphism
f: Yn → Xn (6.10)
sending (In−1 , In , x, y) to (In , x, y). Note that f is projective, since the map
H n−1,n → Hilbn (C 2 ) is. Over the locus where the points (xi , yi ) are all distinct,
Xn−1 → (C 2 )n−1 and H n−1,n → C 2 × Hilbn−1 (C 2 ) restrict to isomorphisms,
hence so does Yn → Xn → (C 2 )n . The locus where some two points coincide
 MACDONALD POLYNOMIALS AND GEOMETRY 243

is a proper closed subvariety of the irreducible variety H n−1,n. Since Yn is flat

over H n−1,n it cannot have a component contained in the coincidence locus, so
the noncoincidence locus is dense in Yn . This shows Yn is irreducible and the
morphism f is birational.
On Yn we have, by pullback from H n−1,n , the two tautological bundles Bn−1
Vn−1 Vn
and Bn . Set O(k, l) = ( Bn−1 )k ( Bn )l . By Lemma 6.17 the canonical
sheaf on H n−1,n is O(1, −1) in this notation. By Lemmas 6.16 and 6.18 the
relative canonical sheaf of Yn over H n−1,n , which is pulled back from that of
Xn−1 over Hilbn−1 (C 2 ), is O(−1, 0). Hence the canonical sheaf ωYn is O(0, −1).
In particular, it is a pullback from Xn .
Now we are going to prove that for the derived functor of the pushforward
we have Rf∗ OYn = OXn . Since ωYn = f ∗ OXn (−1) this also proves Rf∗ ωYn =
OXn (−1). By duality for the projective morphism f∗ we conclude that the sheaf
OXn (−1) is the dualizing complex on Xn , so Xn is Cohen–Macaulay. (This also
shows Xn is Gorenstein with ωXn = O(−1), so we could have made Lemma 6.18
part of the induction.)
Since f is proper and birational, and Xn is normal by Lemma 6.12, we have
f∗ OYn = OXn . We have to prove that Ri f∗ OYn = 0 for all i > 0. Now the fibers
of f are also fibers of the map H n−1,n → U , and thus the fiber dimensions d are


bounded by d+2 2 ≤ n, by Lemma 6.14. In particular, since we are assuming
n ≥ 4, this implies d < n − 2 (exercise for the reader). It follows that Rif∗ O = 0
for i ≥ n − 2.
For i < n − 2 we use the following lemma.
Lemma 6.19. Let f: Y → X be a morphism and let x1 , . . . , xk be global regular
functions on X (and so also on Y ). Suppose that x is an O-regular sequence at
every point of V (x), both in X and in Y . Let U = X \ V (x), W = f −1 (U ), and
f 0 = f|W . Then Rf∗0 OY = OX implies Ri f∗ OY = 0 for 0 < i < k − 1.
Proof. The hypothesis and conclusion are both local with respect to X, so
we can assume X is affine. Then we are to show H i(Y, O) = 0 for 0 < i <
k − 1. Let V = V (x) (in both X and Y , by abuse of notation). The regular
sequence condition implies HVi (O) = 0 for i < k, on both X and Y . The
hypothesis Rf∗0 OY = OX implies that H i (W, OY ) ∼
= H i (U, OX ). Then from the
local cohomology exact sequences

· · · → HVi (OY ) → H i (Y, O) → H i (W, OY ) → HVi+1 (OY ) → · · ·

and
· · · → HVi (OX ) → H i (X, O) → H i (U, OX ) → HVi+1 (OX ) → · · ·
we obtain H i (Y, O) =∼ H i (W, OY ) ∼
= H i(U, OX ) ∼
= H i (X, O) = 0, for 0 < i <
k − 1. 
Conjecture 6.1 implies that (x1 − x2 , . . . , xn−1 − xn ) is a regular sequence on
R[tJ], and hence, by Proposition 6.4, on Xn . To apply the Lemma using this
244 MARK HAIMAN

sequence on Yn and Xn , it remains to prove that the sequence is regular on Yn ,

and that Rf∗ OY = OX outside the locus where all the xi’s coincide.
Note that Yn can be described directly in terms of Xn as the subscheme of
Hilbn−1 (C 2 ) × Xn whose fiber over a point (I, P1 , . . . , Pn) of Xn consists of all
the ideals In−1 ⊆ I for which In−1 /I is a length-1 ideal of the local ring OS,Pn .
About a point where the Pi are not all equal, it follows from Lemma 6.6 that
there is a neighborhood on which Yn is locally isomorphic to Xr1 ×· · · Xrk−1 ×Yrk ,
where of the distinct points Q1 , . . . , Qk , Qk is the one equal to Pn . It also follows
that on such a neighborhood the map f: Yn → Xn is locally given by the identity
on the factors Xri , times the map f: Yrk → Xrk . Hence we have Rf∗ OY = OX
by induction on the locus where the points Pi are not all equal, and, a fortiori,
on the locus where the coordinates xi are not all equal.
To conclude, I claim that x1 , . . . , xn defines a complete intersection in Yn ,
which is Cohen–Macaulay by induction, and hence x is a regular sequence at
each point of V (x). By shifting the origin of coordinates in C 2 , this implies that
(x1 −x2 , . . . , xn−1 −xn ) is a regular sequence at every point where the xi’s are all
equal. Thus we have to show that V (x) has dimension n. Recall that we already
have the analogous result for Xn , Lemma 5.5. By the local product structure
described in the preceding paragraph we can assume the result by induction on
the open set where the yi ’s are not all equal. This reduces us to showing that
the dimension of H0n−1,n = V (x, y) is n − 1, since the locus where all the yi ’s
are equal and all the xi ’s are zero is C 1 × H0n−1,n.
Now we apply Lemma 6.8. By upper-semicontinuity, the maximal fiber di-
mension of H0n−1,n → H0n over the cell Cµ0 occurs at Iµ . There, by the remarks
preceding Lemma 6.14, the fiber dimension is one less than the number of cor-
ners of the diagram of µ. Thus to show that the preimage of Cµ0 has dimen-
sion at most n − 1, we just have to check that for any partition of n, we have
(number of parts) + (number of corners) ≤ n + 1. But this is clear, since the
number of cells in the first column of the diagram is the number of parts, and
at most one of them can be a corner.

7. Diagonal Harmonics
A polynomial function f on a vector space V is said to be harmonic with
respect to a group G of linear endomorphisms of V if f is annihilated by all G-
invariant partial differential operators without constant term. For V = (Q 2 )n =
Q n ⊕ Q n , and G the symmetric group Sn acting “diagonally” by simultaneous
coordinate permutations in each summand, we refer to the space Dn of harmonic
polynomials as the diagonal harmonics.
By Weyl’s theorem on the ring of invariants Q[x, y]Sn , we may equivalently
define Dn as the solution space of the system of differential equations
X
ph,k (∂x, ∂y)f = ∂xhi ∂yik f = 0, for 1 ≤ h + k ≤ n.
i
 MACDONALD POLYNOMIALS AND GEOMETRY 245

In particular the diagonal harmonics are solutions of the Laplace equation

X
(∂x2i + ∂yi2 )f = 0,
i

so they are harmonic polynomials in the classical sense. It is easy to see that
the polynomials ∆µ of Section 3 are diagonal harmonics, and hence the spaces
Dµ are subspaces of Dn .
Let In ⊆ Q[x, y] be the ideal generated by the polarized power sums ph,k for
h + k > 0. We may describe Dn in derivative-free terms as follows.

Proposition 7.1 [Haiman 1994]. The quotient ring Q[x, y]/In is isomorphic
as a doubly graded Sn module to Dn .

For geometric purposes we will work instead with the ring

Rn = C[x, y]/In ,

In being again generated by the polarized power sums, which of course has the
same Frobenius series as Q[x, y]/In or Dn .
Computations have suggested a series of surprising combinatorial conjectures
concerning the Hilbert and Frobenius series of the rings Rn . As these are treated
at length in [Haiman 1994], we here mention only three that are simple to state.

Conjecture 7.2. The dimension of Rn as a vector space, or HRn (1, 1), is

n
equal to (n + 1)n−1 . Moreover q ( 2 ) HRn (q, q −1 ) = (1 + q + q 2 + · · · + q n )n−1 .

Conjecture 7.3. The specialization HRn (q, 1) enumerates spanning trees T

on the vertex set {0, 1, . . ., n}, each counted with weight q i(T ), where i(T ) is the
number of inversions in T . An inversion is a pair i < j for which vertex j lies
on the unique path in T from vertex 0 to vertex i.

Conjecture 7.4. As an Sn module, Rn is isomorphic to the sign character

tensored with the permutation representation of Sn on the finite Abelian group
(Z/(n+1)Z)n /H, where Sn acts by permuting the factors, and H = (Z/(n+1)Z)·
(1, 1, . . . , 1) is the subgroup of Sn -invariant elements.

The above conjectures are corollaries to a pair of more general conjectures giv-
n
ing the Frobenius series specializations FRn (X; q, 1) and q ( 2 ) FRn (X; q, 1/q). In
[Garsia and Haiman 1996a] we showed that these specializations are in turn
corollaries to the following master formula.

Conjecture 7.5. The Frobenius series of Rn is given by

X (1−q)(1−t)Bµ (q, t)Πµ (q, t)H̃µ (X; q, t)

FRn (X; q, t) = Q −a(s) t1+l(s) )(1−q 1+a(s) t−l(s) )
, (7.1)
|µ|=n s∈D(µ) (1−q
246 MARK HAIMAN

where µ ranges over partitions of n, the arms and legs a(s), l(s) are as in (2.16),
Bµ is given by (2.12), and
Y
Πµ = Ω[1 − Bµ ] = (1−q k th ).
(h,k)∈D(µ)
(h,k)6=(0,0)

In the remainder of this section we show how formula (7.1) comes about, and
prove that it holds if the n! conjecture and a suitable cohomology vanishing
hypothesis on Xn are true. The development parallels that in [Haiman 1998], to
which we refer for some geometric results. There we studied the specialization
of (7.1) to the Hilbert series for the Sn -alternating component, which can be
expressed without recourse to Macdonald polynomials as
X (1−q)(1−t)Bµ (q, t)Πµ (q, t)tn(µ) q n(µ )
0

Cn (q, t) = Q −a(s) t1+l(s) )(1−q 1+a(s) t−l(s) )

.
|µ|=n s∈D(µ) (1−q

This turns

out to be a two-parameter analog of the Catalan number Cn =
1 2n
n+1 n . We proved that Cn (q, t) is a polynomial in q and t, and that un-
der certain cohomology vanishing hypotheses it is the Hilbert series of the Sn -
alternating diagonal harmonics. Because we examined only the alternating com-
ponent we could work on Hilbn (C 2 ) without introducing the isospectral variety
Xn . In essence, what we will now do is to lift these results to Xn .
Proposition 7.6. Spec Rn is the scheme theoretic fiber ρ−1 (0) over the origin,
under the canonical map
ρ: (C 2 )n → S n C 2 .
Proof. The coordinate ring of S n C 2 is C[x, y]Sn and the ideal of the origin is
the homogeneous maximal ideal m = (ph,k : h + k > 0). By definition, the ideal
of ρ−1 (0) is generated by the image of m in C[x, y], or In . 
In what follows, we assume the n! conjecture holds for all µ, so Xn is flat over
Hilbn (C 2 ).
Consider the fiber square
ψ
Xn −−−−→ (C 2 )n
 

σy
ρ
y
τ
Hilbn (C 2 ) −−−−→ S n C 2 .
Define Xn0 = (ρψ)−1 (0) = (τ σ)−1 (0) to be the scheme-theoretic fiber of Xn over
0 ∈ S n C 2 . This is a nonreduced subscheme of Xn . By Proposition 7.6, ψ induces
a morphism
ψ
Xn0 −→ ρ−1 (0) = Spec Rn ,
corresponding to a ring homomorphism
ψ] : Rn → H 0 (Xn0 , O).
 MACDONALD POLYNOMIALS AND GEOMETRY 247

As a scheme finite over Hilbn (C 2 ), we have Xn = Spec σ∗ OXn , and since we are
assuming the n! conjecture, σ∗ OXn is locally free of rank n!, that is, it is the
sheaf of sections of a vector bundle P , the image of the homomorphism φ in
(4.7). Then Xn0 = σ−1 (H0n ) = Spec P |H0n as a scheme over H0n = τ −1 (0). Hence
we can identify the global sections H 0 (Xn0 , O) with H 0 (H0n , P ).

Proposition 7.7 [Haiman 1998]. The scheme theoretic zero fiber H0n = τ −1 (0)
in the Hilbert scheme is reduced, Cohen–Macaulay, and has a T-equivariant
resolution by locally free sheaves on Hilbn (C 2 )
Vn+1 V1
0→B⊗ V → ··· → B ⊗ V → B → OH0n → 0, (7.2)

where V = B 0 ⊕ Ot ⊕ Oq , B 0 is a summand of the tautological bundle B =

B 0 ⊕ O, and Ot , Oq denote the trivial bundle O tensored by the 1-dimensional
representation of T with character t or q, respectively.

To proceed further we will need to assume the validity of the following conjecture.

Conjecture 7.8. For all i > 0 and k ≥ 0 we have H i(Xn , B ⊗k ) = 0, and for
i = 0, the canonical map

C[x01 , y10 , . . . , x0k , yk0 , x, y] → H 0 (Xn , B ⊗k ) (7.3)

is surjective.

To clarify, recall that B = C[x0 , y0 ]/I, where C[x0 , y0 ] really means the trivial
bundle O ⊗C C[x0 , y0 ], and we use primes to avoid confusion with the vari-
ables x, y. The map in (7.3) is induced by the maps C[x0 , y0 ] → B, with
the identifications C[x0 , y0 ]⊗k = C[x01 , y10 , . . . , x0k , yk0 ] and H 0 (Xn , C[x0 , y0 ]⊗k ) =
C[x0 , y0 ] ⊗ H 0 (Xn , O) = C[x0 , y0 , x, y]. Note that H 0 (Xn , O) = C[x, y] because
the map ψ: Xn → (C 2 )n is proper and birational, and (C 2 )n is obviously normal.
Vk
Note also that the exterior power B is a summand of B ⊗k , so the conjecture
extends to tensors of exterior powers as well. We do not use the full strength of
Vk
Conjecture 7.8 below, only the vanishing property for bundles B ⊗ B and the
surjectivity property for B and B ⊗ B.

Proposition 7.9. Assume that Conjecture 7.8 holds. Then the canonical ho-
momorphism
ψ] : Rn → H 0 (Xn0 , O) = H 0 (H0n , P )

is an isomorphism.

Proof. Since we are assuming Xn is flat over Hilbn (C 2 ), the pullback functor
σ∗ on sheaves is exact. Applying σ∗ to the resolution (7.2) we get a resolution
on Xn
Vn+1 V1
0→B⊗ V → · · · → B ⊗ V → B → OXn0 → 0. (7.4)
248 MARK HAIMAN

By our vanishing hypothesis, (7.4) is an acyclic resolution of OXn0 , and therefore,

applying H 0 , we get an exact sequence
Vn+1 V1
0 → H 0 (Xn , B ⊗ V ) → · · · → H 0 (Xn , B ⊗ V ) →
→ H 0 (Xn , B) → H 0 (Xn0 , O) → 0. (7.5)

There is a trace map of O-module sheaves

tr: B → O

which sends a section f of B to the function whose value at a point Q is the trace
of multiplication by f on the fiber B(Q), divided by n. On Xn , the joint spectrum
of the multiplication operators X, Y is (x1 , y1 ), . . . , (xn , yn ), and therefore

1X
n
tr f(x0 , y0 ) = f(xi , yi ).
n
i=1

By the construction of (7.2) in [Haiman 1998], the map B → OXn0 factors as

B −→ O −→ OXn0 .
tr

Hence H 0 (Xn , B) → H 0 (Xn0 , O) factors through H 0 (Xn , O) = C[x, y], which

shows that ψ] is surjective.
To prove that ψ is injective, we must show that if f(x0 , y0 , x, y) represents a
global section in the kernel of H 0 (Xn , B) → H 0 (Xn0 , O), then tr f ∈ In . We are
using the surjectivity property in Conjecture 7.8 to assume that such a repre-
sentative polynomial f exists. By (7.5), the kernel in question is the sum of the
images of three maps

H 0 (Xn , B ⊗ B 0 ) → H 0 (Xn , B), (7.6)

H (Xn , B ⊗ Ot ) → H (Xn , B),
0 0

H 0 (Xn , B ⊗ Oq ) → H 0 (Xn , B).

By the construction of (7.2), the second and third maps are multiplication by x0
and y0 , respectively. For any f(x0 , y0 , x, y), the trace map satisfies the identity

tr f − f(0, 0, x, y) ∈ In . (7.7)

To prove this it is sufficient to take f = (x0 )h (y0 )k , since these monomials gener-
ate C[x0 , y0 , x, y] as a C[x, y] module, and the operation we are performing on
f is C[x, y]-linear. We obtain

0 h 0 k 0 if h + k = 0,
tr(x ) (y ) − 0 0 =
h k
ph,k (x, y) if h + k > 0.

In particular if f belongs to the ideal (x0 , y0 ) then f(0, 0, x, y) = 0 and tr f ∈ In .

 MACDONALD POLYNOMIALS AND GEOMETRY 249

This leaves us only to consider the first map (7.6). The summand B 0 is defined
to be the kernel of the trace map. Hence if f(x0 , y0 , x00, y00 , x, y) represents a
section in H 0 (Xn , B ⊗ B 0 ), then
X
f(x0 , y0 , xi, yi , x, y)
i

is the zero section in H 0 (Xn , B). Now for each j there is a homomorphism of
sheaves of OXn algebras B → OXn mapping (x0 , y0 ) to (xi , yi ). This is so because
Xn ⊆ U ×n , and the homomorphism B → OXn corresponds to the projection
Xn → U on the j-th factor. Applying these homomorphisms to the sum above,
we see that
X
f(xj , yj , xi, yi , x, y) = 0 in H 0 (Xn , O) = C[x, y],
i

for all j. By (7.7) this implies that f(xj , yj , 0, 0, x, y) ∈ In for each j. Summing
over j and using (7.7) again we find that f(0, 0, 0, 0, x, y) ∈ In .
The map H 0 (B ⊗ B 0 ) → H 0 (X, B) is multiplication in B, which sends
f(x0 , y0 , x00, y00 , x, y) to f(x0 , y0 , x0 , y0 , x, y). Modulo In , the trace map

H 0 (Xn , B) → H 0 (Xn , O)

is the same as evaluation at (x0 , y0 ) = (0, 0), again by (7.7). Hence the image of
f(x0 , y0 , x00, y00 , x, y) in H 0 (Xn , O) is given modulo In by f(0, 0, 0, 0, x, y), and
since the latter belongs to In the proof is complete. 
Theorem 7.10. Assuming the n! conjecture and Conjecture 7.8 hold, the Frobe-
nius series of Rn is given by the master formula (7.1) in Conjecture 7.5.
Proof. In [Haiman 1998] we derived an Atiyah–Bott type Lefschetz formula
for T-equivariant vector bundles on H0n , using the resolution (7.2) and explicit
local parameters for Hilbn (C 2 ) at the T-fixed points Iµ . This formula takes the
form
X
(−1)i FH i (H0n ,V ) (X; q, t)
i
X (1−q)(1−t)Bµ (q, t)Πµ (q, t)FV (Iµ ) (X; q, t)
= Q −a(s) t1+l(s) )(1−q 1+a(s) t−l(s) )
. (7.8)
s∈D(µ) (1−q
|µ|=n

Actually, this formula was derived for Hilbert series, but when V is a bundle
of Sn modules it generalizes immediately to Frobenius series. The q, t-Catalan
numbers studied in [Haiman 1998] correspond to the line bundle V = O(1).
If the n! conjecture and Conjecture (7.8) hold, then by Proposition 7.9, the
Frobenius series of Rn is equal to FH 0 (H0n ,P ) , where P = σ∗ OXn . Moreover,
using the resolution (7.4), we see that Conjecture 7.5 implies H i (Xn0 , O) = 0 for
i > 0, or equivalently, since σ is finite, H i (H0n , P ) = 0. Therefore the Euler
characteristic on the left-hand side of (7.8) reduces to FRn (X; q, t).
250 MARK HAIMAN

By Theorem 5.4, the n! conjecture implies that FP (Iµ ) (X; q, t) = H̃µ (X; q, t),
and the result follows. 
We conclude with some remarks on the vanishing hypothesis, Conjecture 7.8.
Strong vanishing theorems such as this are a relatively rare phenomenon. The
conjecture is the analog, for the tautological bundle B on the isospectral Hilbert
scheme, of a theorem that does hold for the tautological (quotient) bundle on a
Grassmann variety.
In the case of the Hilbert scheme there is some favorable computational evi-
dence. Namely, assuming the n! conjecture — which has been verified for n ≤ 8 —
one can use (7.8) to compute the Frobenius series Euler characteristic of any ex-
plicit enough bundle V . If V has nonvanishing higher cohomology, we should
expect to see some negative terms. For the bundles referred to in the conjecture,
and reasonable values of n and k, we have done a number of these computations
and the results invariably have positive coefficients. Note also that if the n! con-
jecture were to fail, we should not even expect to obtain a polynomial in (7.8).
For the specialization to Hilbert series, the formula can be evaluated for values
of n much larger than those for which we can check the n! conjecture. A. Garsia
and I have done some of these computations for n as large as 20, always obtain-
ing polynomials with positive coefficients. I regard this as strong evidence for
both the n! conjecture and Conjecture 7.8.

8. The Commuting Variety

The material in this section is based on my conversations with I. Grojnowski,
and represents joint work in progress. At present our results are not definitive,
but we have made some observations and conjectures that I will discuss briefly.
Definition. The commuting variety Cn is the variety of pairs of n × n matrices
(X, Y ) such that XY = Y X.
Little is known about Cn , except that it is irreducible of dimension n2 + n
[Motzkin and Taussky 1952; Richardson 1979]. It is not even known whether
the equations XY = Y X generate its ideal, although this is conjectured to be
true. There is also a conjecture, generally attributed to Hochster, that Cn is
Cohen–Macaulay.
We will be interested in the open set Cn0 of pairs for which the vectors X h Y k e1
span C n , where e1 is the first unit coordinate vector. This is an open set, since
its complement is defined by the vanishing of the n × n minors of the matrix
whose columns are X h Y k e1 .
Given an ideal I ⊆ C[x, y] belonging to Hilbn (C 2 ), fix a basis {1, v2 , . . . , vn } of
C[x, y]/I. With respect to this basis, the operators X and Y of multiplication by
x and y are represented by commuting matrices, and the pair (X, Y ) belongs to
Cn0 because we took our first basis vector to be 1. Conversely, given (X, Y ) ∈ Cn0 ,
we have a surjective map θ: C[x, y] → C n sending p(x, y) to p(X, Y )e1 , whose
 MACDONALD POLYNOMIALS AND GEOMETRY 251

kernel is an ideal I ∈ Hilbn (C 2 ). Then θ induces an isomorphism C[x, y]/I →

C n , under which the unit coordinate basis e1 , . . . , en of C n corresponds to a
basis 1, v2 , . . . , vn of C[x, y]/I. It is easy to see that these two constructions are
mutually inverse and so define a smooth fibration

Cn0 → Hilbn (C 2 )

with fiber G, where G ⊆ GLn is the stabilizer of e1 (so G parametrizes ordered

bases of C n whose first vector is given). In particular this shows that Cn0 is
nonsingular.
Definition. The isospectral commuting variety ICn is the variety of tuples
(X, Y, a, b) ∈ Cn × C 2n such that (a1 , b1 ), . . . , (an , bn ) is the joint spectrum of X
and Y in some order. In other words, we have the identity
Y
n
det(I + rX + sY ) = (1 + rai + sbi ), (8.1)
i=1

where r, s are indeterminates.

Note that if X and Y commute there is a g ∈ GLn such that g−1 Xg and g−1 Y g
are both upper triangular, by Lie’s theorem. In particular they have a joint
spectrum as defined above, given by the diagonal entries of the triangular form.
Note also that there is an action of Sn on ICn , permuting the pairs (ai , bi).
Under this action we have ICn /Sn = Cn , since the invariants ph,k (a, b) are
equal to tr X h Y k and so reduce to functions on Cn .
Let ICn0 denote the open subset of ICn lying over Cn0 . From the definition
of the isospectral Hilbert scheme Xn it follows immediately that we have (set-
theoretically) a fiber square
ICn0 −−−−→ Xn
 
 
y y
Cn0 −−−−→ Hilbn (C 2 ).
Since the bottom arrow is a smooth morphism, so is the top arrow in the scheme-
theoretic fiber square. Hence the scheme-theoretic fiber product is reduced and
therefore equal to the set-theoretic fiber product. This proves
Proposition 8.1. The open set ICn0 in ICn is Cohen–Macaulay (and hence
Gorenstein) if and only if Xn is.
Conjecture 8.2. The isospectral commuting variety ICn is Gorenstein.
Note that this implies the conjecture that the commuting variety Cn = ICn /Sn
is Cohen–Macaulay, as well as the n! conjecture. We should point out here that
the ideal of ICn is certainly not generated by the ideal of Cn (conjecturally
XY = Y X) together with equations (8.1). This fails even for n = 2.
252 MARK HAIMAN

References

[Batyrev and Dais 1996] V. V. Batyrev and D. I. Dais, “Strong McKay correspondence,
string-theoretic Hodge numbers and mirror symmetry”, Topology 35:4 (1996), 901–
929.
[Bayer and Stillman 1989] D. Bayer and M. Stillman, Macaulay (a computer algebra
system for algebraic geometry), version 3.0, 1989. See ftp://math.columbia.edu/pub/
bayer/Macaulay or ftp://math.harvard.edu/Macaulay/.
[Bergeron and Garsia 1992] N. Bergeron and A. M. Garsia, “On certain spaces
of harmonic polynomials”, pp. 51–86 in Hypergeometric functions on domains of
positivity, Jack polynomials, and applications (Tampa, FL, 1991), edited by D. S. P.
Richards, Contemporary Mathematics 138, Amer. Math. Soc., Providence, RI, 1992.
[Briançon 1977] J. Briançon, “Description de Hilbn C{x, y}”, Invent. Math. 41:1
(1977), 45–89.
[Cheah 1998] J. Cheah, “Cellular decompositions for nested Hilbert schemes of points”,
Pacific J. Math. 183:1 (1998), 39–90.
[De Concini and Procesi 1981] C. De Concini and C. Procesi, “Symmetric functions,
conjugacy classes and the flag variety”, Invent. Math. 64:2 (1981), 203–219.
[Ellingsrud and Strømme 1987] G. Ellingsrud and S. A. Strømme, “On the homology
of the Hilbert scheme of points in the plane”, Invent. Math. 87:2 (1987), 343–352.
[Fogarty 1968] J. Fogarty, “Algebraic families on an algebraic surface”, Amer. J. Math
90 (1968), 511–521.
[Garsia and Haiman 1993] A. M. Garsia and M. Haiman, “A graded representation
model for Macdonald’s polynomials”, Proc. Nat. Acad. Sci. U.S.A. 90:8 (1993),
3607–3610.
[Garsia and Haiman 1996a] A. M. Garsia and M. Haiman, “A remarkable q, t-Catalan
sequence and q-Lagrange inversion”, J. Algebraic Combin. 5:3 (1996), 191–244.
[Garsia and Haiman 1996b] A. M. Garsia and M. Haiman, “Some natural bigraded
Sn -modules and q, t-Kostka coefficients”, Electron. J. Combin. 3:2 (1996), RP24.
[Garsia and Haiman 1998] A. M. Garsia and M. Haiman, “A random q, t-hook walk
and a sum of Pieri coefficients”, J. Combin. Theory Ser. A 82:1 (1998), 74–111.
[Garsia and Procesi 1992] A. M. Garsia and C. Procesi, “On certain graded Sn -modules
and the q-Kostka polynomials”, Adv. Math. 94:1 (1992), 82–138.
[Garsia and Remmel 1998] A. M. Garsia and J. Remmel, “Plethystic formulas and
positivity for q, t-Kostka coefficients”, pp. 245–262 in Mathematical essays in honor
of Gian-Carlo Rota (Cambridge, MA, 1996), edited by B. E. Sagan and R. P. Stanley,
Progress in Mathematics 161, Birkhäuser, Boston, 1998.
[Garsia and Tesler 1996] A. M. Garsia and G. Tesler, “Plethystic formulas for
Macdonald q, t-Kostka coefficients”, Adv. Math. 123:2 (1996), 144–222.
[Gordan 1900] M. Gordan, “Les invariants des formes binaires”, J. Math. Pures Appl.
6 (1900), 141–156.
[Gotzmann 1978] G. Gotzmann, “Eine Bedingung für die Flachheit und das Hilbert-
polynom eines graduierten Ringes”, Math. Z. 158:1 (1978), 61–70.
 MACDONALD POLYNOMIALS AND GEOMETRY 253

[Grothendieck 1961] A. Grothendieck, “Techniques de construction et théorèmes

d’existence en géométrie algébrique, IV: Les schémas de Hilbert”, pp. 249–276
(exposé no. 221) in Séminaire Bourbaki 1960/1961 (exposés 205–222), IHP, Paris,
1961. Reprinted by Benjamin, New York, 1966, and Soc. Math. France, Paris, 1995.
[Haiman 1994] M. D. Haiman, “Conjectures on the quotient ring by diagonal invari-
ants”, J. Algebraic Combin. 3:1 (1994), 17–76.
[Haiman 1998] M. Haiman, “t, q-Catalan numbers and the Hilbert scheme”, Discrete
Math. 193:1-3 (1998), 201–224.
[Hotta and Springer 1977] 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 (1977), 113–127.
[Iarrobino 1972] A. Iarrobino, “Reducibility of the families of 0-dimensional schemes
on a variety”, Invent. Math. 15 (1972), 72–77.
[Kirillov and Noumi 1998] A. N. Kirillov and M. Noumi, “Affine Hecke algebras and
raising operators for Macdonald polynomials”, Duke Math. J. 93:1 (1998), 1–39.
[Knop 1997] F. Knop, “Integrality of two variable Kostka functions”, J. Reine Angew.
Math. 482 (1997), 177–189.
[Kraft 1981] H. Kraft, “Conjugacy classes and Weyl group representations”, pp. 191–
205 in Young tableaux and Schur functions in algebra and geometry (Toruń, Poland,
1980), Astérisque 87–88, Soc. Math. France, Paris, 1981.

[Macdonald 1988] I. G. Macdonald, “A new class of symmetric functions”, pp. 131–171

in Actes du 20e Séminaire Lotharingien, I.R.M.A. Publ. 372/S–20, Strasbourg, 1988.
[Macdonald 1995] I. G. Macdonald, Symmetric functions and Hall polynomials, 2nd
ed., Oxford Univ. Press, 1995.
[Motzkin and Taussky 1952] T. S. Motzkin and O. Taussky, “Pairs of matrices with
property L”, Trans. Amer. Math. Soc. 73 (1952), 108–114.
[Nakamura 1996] I. Nakamura, “Simple singularities, McKay correspondence, and
Hilbert schemes of G-orbits”, preprint, Hokkaido University, 1996.
[Reid 1997] M. Reid, “McKay correspondence”, preprint, 1997. Available at http://
xxx.lanl.gov/abs/math.AG/9702016.
[Richardson 1979] R. W. Richardson, “Commuting varieties of semisimple Lie algebras
and algebraic groups”, Compositio Math. 38:3 (1979), 311–327.
[Sahi 1996] S. Sahi, “Interpolation, integrality, and a generalization of Macdonald’s
polynomials”, Internat. Math. Res. Notices 10 (1996), 457–471.
[Springer 1978] T. A. Springer, “A construction of representations of Weyl groups”,
Invent. Math. 44:3 (1978), 279–293.
[Tikhomirov 1992] A. S. Tikhomirov, “On Hilbert schemes and flag varieties of points
on algebraic surfaces”, preprint, 1992.

[Weyl 1946] H. Weyl, The classical groups: their invariants and representations, 2nd
ed., Princeton University Press, Princeton, NJ, 1946.
254 MARK HAIMAN

Mark Haiman
Department of Mathematics
University of California, San Diego
La Jolla, CA, 92093-0112
United States
[email protected]