LECTURE NOTES ON CHEREDNIK ALGEBRAS

PAVEL ETINGOF AND XIAOGUANG MA

arXiv:1001.0432v4 [math.RT] 19 Apr 2010

Contents
1. Introduction 4
2. Classical and quantum Olshanetsky-Perelomov systems for finite Coxeter groups 6
2.1. The rational quantum Calogero-Moser system 6
2.2. Complex reflection groups 6
2.3. Parabolic subgroups 7
2.4. Olshanetsky-Perelomov operators 7
2.5. Dunkl operators 8
2.6. Proof of Theorem 2.9 9
2.7. Uniqueness of the operators Lj 11
2.8. Classical Dunkl operators and Olshanetsky-Perelomov Hamiltonians 11
2.9. Rees algebras 12
2.10. Proof of Theorem 2.24 12
2.11. Notes 13
3. The rational Cherednik algebra 14
3.1. Definition and examples 14
3.2. The PBW theorem for the rational Cherednik algebra 15
3.3. The spherical subalgebra 15
3.4. The localization lemma 16
3.5. Category O for rational Cherednik algebras 16
3.6. The grading element 17
3.7. Standard modules 18
3.8. Finite length 19
3.9. Characters 19
3.10. Irreducible modules 20
3.11. The contragredient module 20
3.12. The contravariant form 20
3.13. The matrix of multiplicities 21
3.14. Example: the rank 1 case 21
3.15. The Frobenius property 22
3.16. Representations of H1,c of type A 23
3.17. Notes 25
4. The Macdonald-Mehta integral 26
4.1. Finite Coxeter groups and the Macdonald-Mehta integral 26
4.2. Proof of Theorem 4.1 26
4.3. Application: the supports of Lc (C) 31
4.4. Notes 35
1
5. Parabolic induction and restriction functors for rational Cherednik algebras 36
5.1. A geometric approach to rational Cherednik algebras 36
5.2. Completion of rational Cherednik algebras 36
5.3. The duality functor 37
5.4. Generalized Jacquet functors 37
5.5. The centralizer construction 37
5.6. Completion of rational Cherednik algebras at arbitrary points of h/G 38
5.7. The completion functor 39
5.8. Parabolic induction and restriction functors for rational Cherednik algebras 40
5.9. Some evaluations of the parabolic induction and restriction functors 41
5.10. Dependence of the functor Resb on b 41
5.11. Supports of modules 43
5.12. Notes 44
6. The Knizhnik-Zamolodchikov functor 45
6.1. Braid groups and Hecke algebras 45
6.2. KZ functors 45
6.3. The image of the KZ functor 47
6.4. Example: the symmetric group Sn 48
6.5. Notes 48
7. Rational Cherednik algebras and Hecke algebras for varieties with group actions 49
7.1. Twisted differential operators 49
7.2. Some algebraic geometry preliminaries 49
7.3. The Cherednik algebra of a variety with a finite group action 49
7.4. Globalization 51
7.5. Modified Cherednik algebra 52
7.6. Orbifold Hecke algebras 52
7.7. Hecke algebras attached to Fuchsian groups 53
7.8. Hecke algebras of wallpaper groups and del Pezzo surfaces 55
7.9. The Knizhnik-Zamolodchikov functor 55
7.10. Proof of Theorem 7.15 56
7.11. Example: the simplest case of double affine Hecke algebras 57
7.12. Affine and extended affine Weyl groups 58
7.13. Cheredniks double affine Hecke algebra of a root system 58
7.14. Algebraic flatness of Hecke algebras of polygonal Fuchsian groups 59
7.15. Notes 61
8. Symplectic reflection algebras 62
8.1. The definition of symplectic reflection algebras 62
8.2. The PBW theorem for symplectic reflection algebras 62
8.3. Koszul algebras 63
8.4. Proof of Theorem 8.6 63
8.5. The spherical subalgebra of the symplectic reflection algebra 64
8.6. The center of the symplectic reflection algebra Ht,c 65
8.7. A review of deformation theory 66
8.8. Deformation-theoretic interpretation of symplectic reflection algebras 69
8.9. Finite dimensional representations of H0,c . 71

2
8.10. Azumaya algebras 72
8.11. Cohen-Macaulay property and homological dimension 73
8.12. Proof of Theorem 8.32 74
8.13. Notes 75
9. Calogero-Moser spaces 76
9.1. Hamiltonian reduction along an orbit 76
9.2. The Calogero-Moser space 76
9.3. The Calogero-Moser integrable system 77
9.4. Proof of Wilsons theorem 79
9.5. The Gan-Ginzburg theorem 80
9.6. The space Mc for Sn and the Calogero-Moser space. 81
9.7. The Hilbert scheme Hilbn (C2 ) and the Calogero-Moser space 82
9.8. The cohomology of Cn 83
9.9. Notes 84
10. Quantization of Claogero-Moser spaces 85
10.1. Quantum moment maps and quantum Hamiltonian reduction 85
10.2. The Levasseur-Stafford theorem 85
10.3. Corollaries of Theorem 10.1 87
10.4. The deformed Harish-Chandra homomorphism 88
10.5. Notes 89
References 90

3
 1. Introduction
Double affine Hecke algebras, also called Cherednik algebras, were introduced by Chered-
nik in 1993 as a tool in his proof of Macdonalds conjectures about orthogonal polynomials for
root systems. Since then, it has been realized that Cherednik algebras are of great indepen-
dent interest; they appeared in many different mathematical contexts and found numerous
applications.
The present notes are based on a course on Cherednik algebras given by the first author
at MIT in the Fall of 2009. Their goal is to give an introduction to Cherednik algebras, and
to review the web of connections between them and other mathematical objects. For this
reason, the notes consist of many parts that are relatively independent of each other. Also,
to keep the notes within the bounds of a one-semester course, we had to limit the discussion
of many important topics to a very brief outline, or to skip them altogether. For a more
in-depth discussion of Cherednik algebras, we refer the reader to research articles dedicated
to this subject.
The notes do not contain any original material. In each section, the sources of the expo-
sition are listed in the notes at the end of the section.
The organization of the notes is as follows.
In Section 2, we define the classical and quantum Calogero-Moser systems, and their
analogs for any Coxeter groups introduced by Olshanetsky and Perelomov. Then we intro-
duce Dunkl operators, prove the fundamental result of their commutativity, and use them to
establish integrability of the Calogero-Moser and Olshanetsky-Perelomov systems. We also
prove the uniqueness of the first integrals for these systems.
In Section 3, we conceptualize the commutation relations between Dunkl operators and
coordinate operators by introducing the main object of these notes - the rational Cherednik
algebra. We develop the basic theory of rational Cherednik algebras (proving the PBW
theorem), and then pass to the representation theory of rational Cherednik algebras, more
precisely, study the structure of category O. After developing the basic theory (parallel to
the case of semisimple Lie algebras), we completely work out the representations in the rank
1 case, and prove a number of results about finite dimensional representations and about
representations of the rational Cherednik algebra attached to the symmetric group.
In Section 4, we evaluate the Macdonald-Mehta integral, and then use it to find the sup-
ports of irrieducible modules over the rational Cherednik algebras with the trivial lowest
weight, in particular giving a simple proof of the theorem of Varagnolo and Vasserot, classi-
fying such representations which are finite dimensional.
In Section 5, we describe the theory of parabolic induction and restriction functors for
rational Cherednik algebras, developed in [BE], and give some applications of this theory,
such as the description of the category of Whittaker modules and of possible supports of
modules lying in category O.
In Section 6, we define Hecke algebras of complex reflection groups, and the Knizhnik-
Zamolodchikov (KZ) functor from the category O of a rational Cherednik algebra to the
category of finite dimensional representations of the corresponding Hecke algebra. We use
this functor to prove the formal flatness of Hecke algebras of complex reflection groups (a
theorem of Broue, Malle, and Rouquier), and state the theorem of Ginzburg-Guay-Opdam-
Rouquier that the KZ functor is an equivalence from the category O modulo its torsion part
to the category of representations of the Hecke algebra.
4
 In Section 7, we define rational Cherednik algebras for orbifolds. We also define the
corresponding Hecke algebras, and define the KZ functor from the category of modules over
the former to that over the latter. This generalizes to the curved case the KZ functor
for rational Cherednik algebras of complex reflection groups, defined in Section 6. We then
apply the KZ functor to showing that if the universal cover of the orbifold in question has
a trivial H 2 (with complex coefficients), then the orbifold Hecke algebra is formally flat,
and explain why the condition of trivial H 2 cannot be dropped. Next, we list examples of
orbifold Hecke algebras which satisfy the condition of vanishing H 2 (and hence are formally
flat). These include usual, affine, and double affine Hecke algebras, as well as Hecke algebras
attached to Fuchsian groups, which include quantizations of del Pezzo surfaces and their
Hilbert schemes; we work these examples out in some detail, highlighting connections with
other subjects. Finally, we discuss the issue of algebraic flatness, and prove it in the case of
algebras of rank 1 attached to Fuchsian groups, using the theory of deformations of group
algebras of Coxeter groups developed in [ER].
In Section 8, we define symplectic reflection algebras (which inlude rational Cherednik al-
gebras as a special case), and generalize to them some of the theory of Section 3. Namely, we
use the theory of deformations of Koszul algebras to prove the PBW theorem for symplectic
reflection algebras. We also determine the center of symplectic reflection algebras, showing
that it is trivial when the parameter t is nonzero, and is isomorphic to the shperical subal-
gebra if t = 0. Next, we give a deformation-theoretic interpretation of symplectic reflection
algebras as universal deformations of Weyl algebras smashed with finite groups. Finally, we
discuss finite dimensional representations of symplectic reflection algebras for t = 0, show-
ing that the Azumaya locus on the space of such representations coincides with the smooth
locus. This uses the theory of Cohen-Macaulay modules and of homological dimension in
commutative algebra. In particular, we show that for Cherednik algebras of type An1 , the
whole representation space is smooth and coincides with the spectrum of the center.
In Section 9, we give another description of the spectrum of the center of the rational
Cherednik algebra of type An1 (for t = 0), as a certain space of conjugacy classes of pairs
of matrices, introduced by Kazhdan, Kostant, and Sternberg, and called the Calogero-Moser
space (this space is obtained by classical hamiltonian reduction, and is a special case of a
quiver variety). This yields a new construction of the Calogero-Moser integrable system.
We also sketch a proof of the Gan-Ginzburg theorem claiming that the quotient of the
commuting scheme by conjugation is reduced, and hence isomorphic to C2n /Sn . Finally, we
explain that the Calogero-Moser space is a topologically trivial deformation of the Hilbert
scheme of the plane, we use the theory of Cherednik algebras to compute the cohomology
ring of this space.
In Section 10, we generalize the results of Section 9 to the quantum case. Namely, we
prove the quantum analog of the Gan-Ginzburg theorem (the Harish Chandra-Levasseur-
Stafford theorem), and explain how to quantize the Calogero-Moser space using quantum
Hamiltonian reduction. Not surprisingly, this gives the same quantization as was constructed
in the previous sections, namely, the spherical subalgebra of the rational Cherednik algebra.
Acknowledgements. We are grateful to the participants of the course on Cherednik
algebras at MIT, especially to Roman Travkin and Aleksander Tsymbalyuk, for many useful
comments and corrections. This work was partially supported by the NSF grants DMS-
0504847 and DMS-0854764.

5
 2. Classical and quantum Olshanetsky-Perelomov systems for finite
Coxeter groups
2.1. The rational quantum Calogero-Moser system. Consider the differential operator
Xn X
2 1
H= 2
c(c + 1) .
i=1
xi (xi xj )2
i6=j

This is the quantum Hamiltonian for a system of n particles on the line of unit mass and the
interaction potential (between particle 1 and 2) c(c + 1)/(x1 x2 )2 . This system is called
the rational quantum Calogero-Moser system.
It turns out that the rational quantum Calogero-Moser system is completely integrable.
Namely, we have the following theorem.
Theorem 2.1. There exist differential operators Lj with rational coefficients of the form
X n
j
Lj = ( ) + lower order terms, j = 1, . . . , n,
i=1
xi

which are invariant under the symmetric group Sn , homogeneous of degree j, and such
that L2 = H and [Lj , Lk ] = 0, j, k = 1, . . . , n.
We will prove this theorem later.
P
Remark 2.2. L1 = i .
xi
2.2. Complex reflection groups. Theorem 2.1 can be generalized to the case of any finite
Coxeter group. To describe this generalization, let us recall the basic theory of finite Coxeter
groups and, more generally, complex reflection groups.
Let h be a finite-dimensional complex vector space. We say that a semisimple element
s GL(h) is a (complex) reflection if rank (1 s) = 1. This means that s is conjugate to
the diagonal matrix diag(, 1, . . . , 1) where 6= 1.
Now assume h carries a nondegenerate inner product (, ). We say that a semisimple
element s O(h) is a real reflection if rank (1 s) = 1; equivalently, s is conjugate to
diag(1, 1, . . . , 1).
Now let G GL(h) be a finite subgroup.
Definition 2.3. (i) We say that G is a complex reflection group if it is generated by
complex reflections.
(ii) If h carries an inner product, then a finite subgroup G O(h) is a real reflection
group (or a finite Coxeter group) if G is generated by real reflections.
For the complex reflection groups, we have the following important theorem.
Theorem 2.4 (The Chevalley-Shepard-Todd theorem, [Che]). A finite subgroup G of GL(h)
is a complex reflection group if and only if the algebra (Sh)G is a polynomial (i.e., free)
algebra.
By the Chevalley-Shepard-Todd theorem, the algebra (Sh)G has algebraically independent
generators Pi , homogeneous of some degrees di for i = 1, . . . , dim h. The numbers di are
uniquely determined, and are called the degrees of G.
6
Example 2.5. If G = Sn , h = Cn1 (the space of vectors in Cn with zero sum of co-
ordinates), then one can take Pi (p1 , . . . , pn ) = pi+1
P 1 + + pi+1
n , i = 1, . . . , n 1 (where
i p i = 0).
2.3. Parabolic subgroups. Let G GL(h) be a finite subgroup.
Definition 2.6. A parabolic subgroup of G is the stabilizer Ga of a point a h.
Note that by Chevalleys theorem, a parabolic subgroup of a complex (respectively, real)
reflection group is itself a complex (respectively, real) reflection group.
Also, if W is a real reflection group, then it can be shown that a subgroup W W is
parabolic if and only if it is conjugate to a subgroup generated by a subset of simple reflections
of W . In this case, the rank of W , i.e. the number of generating simple reflections, equals

the codimension of the space hW .
Example 2.7. Consider the Coxeter group of type E8 . It has the Dynkin diagram:

The parabolic subgroups will be Coxeter groups whose Dynkin diagrams are obtained by
deleting vertices from the above graph. In particular, the maximal parabolic subgroups are
D7 , A7 , A1 A6 , A2 A1 A4 , A4 A3 , D5 A2 , E6 A1 , E7 .
Suppose G G is a parabolic subgroup, and b h is such that Gb = G . In this case,

we have a natural G -invariant decomposition h = hG (h G ) , and b hG . Thus we have
G
a nonempty open set hG reg of all a h for which Ga = G ; this set is nonempty because it

contains b. We also have a G -invariant decomposition h = h G (hG ) , and we can define
G
the open set hG
reg of all h for which G = G . It is clear that this set is nonempty. This
implies, in particular, that one can make an alternative definition of a parabolic subgroup
of G as the stabilizer of a point in h .
2.4. Olshanetsky-Perelomov operators. Let s GL(h) be a complex reflection. Denote
by s h an eigenvector in h of s with nontrivial eigenvalue.
Let W O(h) be a real reflection group and S W the set of reflections. Clearly, W
acts on S by conjugation. Let c : S C be a conjugation invariant function.
Definition 2.8. [OP] The quantum Olshanetsky-Perelomov Hamiltonian attached to W is
the second order differential operator
X cs (cs + 1)(s , s )
H := h ,
sS
s2
where h is the Laplace operator on h.
Here we use the inner product on h which is dual to the inner product on h.
Let us assume that h is an irreducible representation of W (i.e. W is an irreducible finite
Coxeter group, and h is its reflection representation.) In this case, we can take P1 (p) = p2 .
Theorem 2.9. The system defined by the Olshanetsky-Perelomov operator H is completely
integrable. Namely, there exist differential operators Lj on h with rational coefficients and
symbols Pj , such that Lj are homogeneous (of degree dj ), L1 = H, and [Lj , Lk ] = 0, j, k.
7
 This theorem is obviously a generalization of Theorem 2.1 about W = Sn .
To prove Theorem 2.9, one needs to develop the theory of Dunkl operators.
Remark 2.10. 1. We will show later that the operators Lj are unique.
2. Theorem 2.9 for classical root systems was proved by Olshanetsky and Perelomov
(see [OP]), following earlier work of Calogero, Sutherland, and Moser in type A. For a
general Weyl group, this theorem (in fact, its stronger trigonometric version) was proved by
analytic methods in the series of papers [HO],[He3],[Op3],[Op4]. A few years later, a simple
algebraic proof using Dunkl operators, which works for any finite Coxeter group, was found
by Heckman, [He1]; this is the proof we will give below.
For the trigonometric version, Heckman also gave an algebraic proof in [He2], which used
non-commuting trigonometric counterparts of Dunkl operators. This proof was later im-
proved by Cherednik ([Ch1]), who defined commuting (although not Weyl group invariant)
versions of Heckmans trigonometric Dunkl operators, now called Dunkl-Cherednik opera-
tors.
2.5. Dunkl operators. Let G GL(h) be a finite subgroup. Let S be the set of reflections
in G. For any reflection s S, let s be the eigenvalue of s on s h (i.e. ss = s s ),
and let s h be an eigenvector such that ss = 1
s s . We normalize them in such a way
that hs , s i = 2.
Let c : S C be a function invariant with respect to conjugation. Let a h.
The following definition was made by Dunkl for real reflection groups, and by Dunkl and
Opdam for complex reflection groups.
Definition 2.11. The Dunkl operator Da = Da (c) on C(h) is defined by the formula
X 2cs s (a)
Da = Da (c) := a (1 s).
sS
(1 s ) s

Clearly, Da CG D(hreg ), where hreg is the set of regular points of h (i.e. not preserved
by any reflection), and D(hreg ) denotes the algebra of differential operators on hreg .
Example 2.12. Let G = Z2 , h = C. Then there is only one Dunkl operator up to scaling,
and it equals to
c
D = x (1 s),
x
where the operator s is given by the formula (sf )(x) = f (x).
Remark 2.13. The Dunkl operators Da map the space of polynomials C[h] to itself.
Proposition 2.14. (i) For any x h , one has
X
[Da , x] = (a, x) cs (a, s )(x, s )s.
sS
1
(ii) If g G then gDa g = Dga .
Proof. (i) The proof follows immediately from the identity
1 s
x sx = (x, s )s .
2
(ii) The identity is obvious from the invariance of the function c. 
8
 The main result about Dunkl operators, on which all their applications are based, is the
following theorem.
Theorem 2.15 (C. Dunkl, [Du1]). The Dunkl operators commute:
[Da , Db ] = 0 for any a, b h.
Proof. Let x h . We have
[[Da , Db ], x] = [[Da , x], Db ] [[Db , x], Da ].
Now, using Proposition 2.14, we obtain:
X
[[Da , x], Db ] = [ cs (a, s )(x, s )s, Db ]
sS
X 1 1
s
= cs (a, s )(x, s )(b, s )sDs .
sS
2

Since a and b occur symmetrically, we obtain that [[Da , Db ], x] = 0. This means that for any
f
f C[h], [Da , Db ]f = f [Da , Db ]1 = 0. So for f, g C[h], g [Da , Db ] = [Da , Db ]f = 0. Thus
g
f
[Da , Db ] = 0 which implies [Da , Db ] = 0 in the algebra CG D(hreg ) (since this algebra
g
acts faithfully on C(h)). 

2.6. Proof of Theorem 2.9. For any element B CW D(hregP ), define m(B) to be the
W
differential operator C(h) C(h), defined by B. That is, if B = gW Bg g, Bg D(hreg ),
P
then m(B) = gW Bg . It is clear that if B is W -invariant, then A CW D(hreg ),
m(AB) = m(A)m(B).
Proposition 2.16 ([Du1], [He1]). Let {y1 , . . . , yr } be an orthonormal basis of h. Then we
have
Xr
m( Dy2i ) = H,
i=1
P cs (s , s )
where H = h sS s .
s
Proof. For any y h, we have m(Dy2 ) = m(Dy y ). A simple computation shows that
X cs s (y)
Dy y = y2 (1 s)y
sS
s
X cs s (y)
= y2 (y (1 s) + s (y)s s).
sS
s

This means that

X cs s (y)2
m(Dy2 ) = y2 s .
sS
s
9
So we get
r
X r
X X r
X s (yi )2
m( Dy2i ) = y2i cs s = H,
i=1 i=1 sS i=1
s
Pr
since i=1 s (yi )2 = (s , s ). 
Recall that by the Chevalley-Shepard-Todd theorem, the algebra (Sh)W is free. Let P1 =
p , P2 , . . . , Pr be homogeneous generators of (Sh)W .
2

Corollary 2.17. The differential operators Lj = m(Pj (Dy1 , . . . , Dyr )) are pairwise commu-
tative, have symbols Pj , homogeneity degree dj , and L1 = H.
Proof. Since Dunkl operators commute, the operators Lj are well defined. Since m(AB) =
m(A)m(B) when B is invariant, the operators Lj are pairwise commutative. The rest is
clear. 
Now to prove Theorem 2.9, we will show that the operators H and H are conjugate to
each other by a certain function; this will complete the proof.
Q
Proposition 2.18. Let c (x) := sS s (x)cs . Then we have
c1 H c = H.
Remark 2.19. The function c (x) is not rational. It is a multivalued analytic function.
Nevertheless, it is easy to see that for any differential operator L with rational coefficients,
c1 L c also has rational coefficients.
Proof of Proposition 2.18. We have
Xr X cs (s , s )
yi (log c )yi = s .
i=1 sS
2s
Therefore, we have
X cs (s , s )
c H c1 = h s + U,
sS
s
where
X cs (cs + 1)(s , s )
U = c (h c1 ) .
sS
s2
Let us compute U. We have
X cs (cs + 1)(s , s ) X cs cu (s , u )
c (h c1 ) = + .
sS
s2 s6=uS
s u

We claim that
Q the last sum is actually zero. Indeed, this sum is invariant under the Coxeter
group, so sS s is a regular anti-invariant function of degree |S| 2. But the smallest
degree of a nonzero anti-invariant is |S|, so = 0, U = 0, and we are done (Proposition 2.18
and Theorem 2.9 are proved). 
Remark 2.20. A similar method works for any complex reflection group G. Namely, the
operators Li = m(Pi (Dy1 , . . . , Dyr )) form a quantum integrable system. However, if G is not
a real reflection group, this system does not have a quadratic Hamiltonian in momentum
variables (so it does not have a physical meaning).
10
2.7. Uniqueness of the operators Lj .
Proposition 2.21. The operators Lj are unique.
Proof. Assume that we have two choices for Lj : Lj and Lj . Denote Lj Lj by M.
Assume M 6= 0. We have
(i) M is a differential operator on h with rational coefficients, of order smaller than dj
and homogeneity degree dj ;
(ii) [M, H] = 0.
Let M0 be the symbol of M. Then M0 is a polynomial of p h with coefficients in C(h).
We have, from (ii),
{M0 , p2 } = 0, p h ,
and from (i) we see that the coefficients of M0 are not polynomial (as they have negative
degree).
However, we have the following lemma.
Lemma 2.22. Let h be a finite dimensional vector space. Let : (x, p) 7 (x, p) be a
rational function on h h which is a polynomial in p h . Let f : h C be a polynomial
such that the differentials df (p) for p h span h (e.g., f (p) = p2 ). Suppose that the
Poisson bracket of f and vanishes: {, f } = 0. Then is a polynomial.
Proof. (R. Raj) Let Z h be the pole divisor of . Let x0 h be a generic point in Z. Then
1 is regular and vanishes at (x0 , p) for generic p h . Also from { 1 , f } = 0, we have
1 vanishes along the entire flowline of the Hamiltonian flow defined by f and starting at
x0 . This flowline is defined by the formula
x(t) = x0 + tdf (p), p(t) = p,
and it must be contained in the pole divisor of near x0 . This implies that df (p) must be
in Tx0 Z for almost every, hence for every p h . This is a contradiction with the assumption
on f , which implies that in fact has no poles. 


2.8. Classical Dunkl operators and Olshanetsky-Perelomov Hamiltonians. We con-

tinue to use the notations in Section 2.4.
Definition 2.23. The classical Olshanetsky-Perelomov Hamiltonian corresponding to W is
the following classical Hamiltonian on hreg h = T hreg :
X c2 (s , s )
s
H0 (x, p) = p2 .
sS
s2 (x)

Theorem 2.24 ([OP],[HO, He3, Op3, Op4],[He1]). The Hamiltonian H0 defines a classical
integrable system. Namely, there exist unique regular functions L0j on hreg h , where highest
terms in p are Pj , such that L0j are homogeneous of degree dj (under x 7 x, x h , p 7
1 p, p h), and such that L01 = H0 and {L0j , L0k } = 0, j, k.
Proof. The proof is given in the next subsection. 
11
Example 2.25. Let W = Sn , h = Cn1 . Then
Xn X 1
H0 = p2i c2 2
( the classical Calogero-Moser Hamiltonian).
i=1 i6=j
(xi xj )

So the theorem says that there are functions L0j , j = 1, . . . , n 1,

X j+1
L0j = pi + lower terms,
i

homogeneous of degree zero, such that L01 = H0 and {L0j , L0k } = 0.

2.9. Rees algebras. Let A be a filtered algebra over a field k: k = F 0 A F 1 A ,
i F i A = A. Then the Rees algebra A = Rees(A) is defined by the formula A = n
n=0 F A.
This is an algebra over k[~], where ~ is the element 1 of the summand F 1 A.
2.10. Proof of Theorem 2.24. The proof of Theorem 2.24 is similar to the proof of its
quantum analog. Namely, to construct the functions L0j , we need to introduce classical
Dunkl operators. To do so, we introduce a parameter ~ (Plancks constant) and define
Dunkl operators Da (~) = Da (c, ~) with ~:
X 2cs s (a)
Da (c, ~) = ~Da (c/~) = ~a (1 s), where a h.
sS
(1 s )s

These operators can be regarded as elements of the Rees algebra A = Rees(CW D(hreg )),
where the filtration is by order of differential operators (and W sits in degree 0). Reducing
these operators modulo ~, we get classical Dunkl operators Da0 (c) A0 := A/~A = CW
O(T hreg ). They are given by the formula
X 2cs s (a)
Da0 (c) = pa (1 s),
sS
(1 s ) s

where pa is the classical momentum (the linear function on h corresponding to a h).

It follows from the commutativity of the quantum Dunkl operators Da (c) that the Dunkl
operators Da (c, ~) commute. Hence, so do the classical Dunkl operators Da0 :
[Da0 , Db0] = 0.
We also have the following analog of Proposition 2.14:
Proposition 2.26. (i) For any x h , one has
X
[Da0 , x] = cs (a, s )(x, s )s.
sS

(ii) If g W then gDa0 g 1 = Dga

0
.
Now let us construct the classical Olshanetsky-Perelomov Hamiltonians.
P As in the P
quan-
tum case, we have the operation m(), which is given by the formula gW Bg g 7 Bg ,
B O(T hreg ). We define the Hamiltonian
r
X
H 0 := m( (Dy0i )2 ).
i=1
12
By taking the limit of quantum situation, we find
X cs (s , s )
H 0 = p2 ps .
sS
s (x)
Unfortunately, this is no longer conjugate to H0 . However, consider the (outer) automor-
phism c of the algebra CW O(T hreg ) defined by the formulas
c (x) = x, c (s) = s, c (pa ) = pa + a log c ,
for x h , a h, s W . It is easy to see that if b0 A0 and b A is a deformation of
1
b0 then c (b0 ) = lim~0 c/~ bc/~ . Therefore, taking the limit ~ 0 in Proposition 2.16, we
find that H0 = c (H 0 ).
Now set L0j = m(c (Pj (Dy01 , . . . , Dy0r ))). These functions are well defined since Da0 com-
mute, are homogeneous of degree zero, and L01 = H0 .
Moreover, we can define the operators Lj (~) in Rees(D(hreg )W ) in the same way as Lj , but
using the Dunkl operators Dyi (~) instead of Dyi . Then [Lj (~), Lk (~)] = 0, and Lj (~)|~=0 =
L0j . This implies that L0j Poisson commute: {L0j , L0k } = 0.
Theorem 2.24 is proved.
Remark 2.27. As in the quantum situation, Theorem 2.24 can be generalized to complex
reflection groups, giving integrable systems with Hamiltonians which are non-quadratic in
momentum variables.
2.11. Notes. Section 2.1 follows Section 5.4 of [E4]; the definition of complex reflection
groups and their basic properties can be found in [GM]; the definition of parabolic subgroups
and the notations are borrowed from Section 3.1 of [BE]; the remaining parts of this section
follow Section 6 of [E4].

13
 3. The rational Cherednik algebra
3.1. Definition and examples. Above we have made essential use of the commutation
relations between operators x h , g G, and Da , a h. This makes it natural to consider
the algebra generated by these operators.
Definition 3.1. The rational Cherednik algebra associated to (G, h) is the algebra Hc (G, h)
generated inside A = Rees(CG D(hreg )) by the elements x h , g G, and Da (c, ~), a h.
If t C, then the algebra Ht,c (G, h) is the specialization of Hc (G, h) at ~ = t.
Proposition 3.2. The algebra Hc is the quotient of the algebra CG T(h h )[~] (where
T denotes the tensor algebra) by the ideal generated by the relations
X
[x, x ] = 0, [y, y ] = 0, [y, x] = ~(y, x) cs (y, s )(x, s)s,
sS

where x, x h , y, y h.
Proof. Let us denote the algebra defined in the proposition by Hc = Hc (G, h). Then accord-
ing to the results of the previous sections, we have a surjective homomorphism : Hc Hc
defined by the formula (x) = x, (g) = g, (y) = Dy (c, ~).
Let us show that this homomorphism is injective. For this purpose assume that yi is a
basis of h, and xi is the dual basis of h . Then it is clear from the relations of Hc that Hc is
spanned over C[~] by the elements
r
Y r
Y
(3.1) g yimi xni i .
i=1 i=1

Thus it remains to show that the images of the elements (3.1) under the map , i.e. the
elements
r
Y r
Y
g Dyi (c, ~) mi
xni i .
i=1 i=1

are linearly independent. But this follows from the obvious fact that the symbols of these
elements in CG C[h hreg ][~] are linearly independent. The proposition is proved. 
Remark 3.3. 1. Similarly, one can define the universal algebra H(G, h), in which both ~
and c are variables. (So this is an algebra over C[~, c].) It has two equivalent definitions
similar to the above.
2. It is more convenient to work with algebras defined by generators and relations than
with subalgebras of a given algebra generated by a given set of elements. Therefore, from
now on we will use the statement of Proposition 3.2 as a definition of the rational Cherednik
algebra Hc . According to Proposition 3.2, this algebra comes with a natural embedding
c : Hc Rees(CG D(hreg )), defined by the formula x x, g g, y Dy (c, ~). This
embedding is called the Dunkl operator embedding.
Example 3.4. 1. Let G = Z2 , h = C. In this case c reduces to one parameter, and the
algebra Ht,c is generated by elements x, y, s with defining relations
s2 = 1, sx = xs, sy = ys, [y, x] = t 2cs.
14
 2. Let G = Sn , h = Cn . In this case there is also only one complex parameter c, and the
algebra Ht,c is the quotient of Sn Chx1 , . . . , xn , y1 , . . . , yn i by the relations
X
[xi , xj ] = [yi , yj ] = 0, [yi , xj ] = csij , [yi , xi ] = t c sij .
j6=i

Here ChEi denotes the free algebra on a set E, and sij is the transposition of i and j.
3.2. The PBW theorem for the rational Cherednik algebra. Let us put a filtration
on Hc by setting deg y = 1 for y h and deg x = deg g = 0 for x h , g G. Let gr(Hc )
denote the associated graded algebra of Hc under this filtration, and similarly for Ht,c . We
have a natural surjective homomorphism
: CG C[h h ][~] gr(Hc ).
For t C, it specializes to surjective homomorphisms
t : CG C[h h ] gr(Ht,c ).
Proposition 3.5 (The PBW theorem for rational Cherednik algebras). The maps and t
are isomorphisms.
Proof. The statement is equivalent to the claim that the elements (3.1) are a basis of Ht,c ,
which follows from the proof of Proposition 3.2. 
Remark 3.6. 1. We have
H0,0 = CG C[h h ] and H1,0 = CG D(h).
2. For any C , the algebra Ht,c is naturally isomorphic to Ht,c .
3. The Dunkl operator embedding c specializes to embeddings
0,c : H0,c CG C[h hreg ],
given by x 7 x, g 7 g, y 7 Da0 , and
1,c : H1,c CG D(hreg ),
given by x 7 x, g 7 g, y 7 Da . So H0,c is generated by x, g, Da0, and H1,c is generated by
x, g, Da .
Since Dunkl operators map polynomials to polynomials, the map 1,c defines a represen-
tation of H1,c on C[h]. This representation is called the polynomial representation of H1,c .
P
3.3. The spherical subalgebra. Let e CG be the symmetrizer, e = |G|1 gG g. We
have e2 = e.
Definition 3.7. Bc := eHc e is called the spherical subalgebra of Hc . The spherical subalgebra
of Ht,c is Bt,c := Bc /(~ t) = eHt,c e.
Note that
e (CG D(hreg )) e = D(hreg )G , e (CG C[hreg h ]) e = C[hreg h ]G .
Therefore, the restriction gives the embeddings: 1,c : B1,c D(hreg )G , and 0,c : B0,c
C[h hreg ]G . In particular, we have
Proposition 3.8. The spherical subalgebra B0,c is commutative and does not have zero
divisors. Also B0,c is finitely generated.
15
Proof. The first statement is clear from the above. The second statement follows from the
fact that gr(B0,c ) = B0,0 = C[h h ]G , which is finitely generated by Hilberts theorem. 
Corollary 3.9. Mc = SpecB0,c is an irreducible affine algebraic variety.
Proof. Directly from the definition and the proposition. 
We also obtain
Proposition 3.10. Bc is a flat quantization (non-commutative deformation) of B0,c over
C[~].
So B0,c carries a Poisson bracket {, }(thus Mc is a Poisson variety), and Bc is a quanti-
zation of the Poisson bracket, i.e. if a, b Bc and a0 , b0 are the corresponding elements in
B0,c , then
[a, b]/~ {a0 , b0 } (mod ~).
Definition 3.11. The Poisson variety Mc is called the Calogero-Moser space of G, h with
parameter c.
loc
3.4. The localization lemma. Let Ht,c = Ht,c [ 1 ] be the localization of Ht,c as a module
over C[h] with respect to the discriminant (a polynomial vanishing to the first order on
loc loc
each reflection plane). Define also Bt,c = eHt,c e.
Proposition 3.12. (i) For t 6= 0 the map t,c induces an isomorphism of algebras
loc loc
Ht,c CG D(hreg ), which restricts to an isomorphism Bt,c D(hreg )G .
loc
(ii) The map 0,c induces an isomorphism of algebras H0,c CG C[h hreg ], which
loc
restricts to an isomorphism B0,c C[h hreg ]G .
Proof. This follows immediately from the fact that the Dunkl operators have poles only on
the reflection hyperplanes. 
Since gr(B0,c ) = B0,0 = C[h h]G , we get the following geometric corollary.
Corollary 3.13. (i) The family of Poisson varieties Mc is a flat deformation of the
Poisson variety M0 := (h h )/G. In particular, Mc is smooth outside of a subset of
codimension 2.
(ii) We have a natural map c : Mc h/G, such that c1 (hreg /G) is isomorphic to
(hreg h )/G. The Poisson structure on Mc is obtained by extension of the symplectic
Poisson structure on (hreg h )/G.
Example 3.14. Let W = Z2 , h = C. Then B0,c = hx2 , xp, p2 c2 /x2 i. Let X := x2 , Z := xp
and Y := p2 c2 /x2 . Then Z 2 XY = c2 . So Mc is isomorphic to the quadric Z 2 XY = c2
in the 3-dimensional space and it is smooth for c 6= 0.
3.5. Category O for rational Cherednik algebras. From the PBW theorem, we see that
H1,c = Sh CG Sh. It is similar to the structure of the universal enveloping algebra of a
simple Lie algebra: U(g) = U(n )U(h)U(n+ ). Namely, the subalgebra CG plays the role
of the Cartan subalgebra, and the subalgebras Sh and Sh play the role of the positive and
negative nilpotent subalgebras. This similarity allows one to define and study the category
O analogous to the Bernstein-Gelfand-Gelfand category O for simple Lie algebras.
16
Definition 3.15. The category Oc (G, h) is the category of modules over H1,c (G, h) which
are finitely generated over Sh and locally finite under Sh (i.e., for M Oc (G, h), v M,
(Sh)v is finite dimensional).
If M is a locally finite (Sh)G -module, then
M = h /G M ,
where
M = {v M|p (Sh)G , N s.t. (p (p))N v = 0},
(notice that h /G = Specm(Sh)G ).
Proposition 3.16. M are H1,c -submodules.
Proof. Note first that we have an isomorphism : H1,c (G, h) = H1,c (G, h ), which is given
by xa 7 ya , yb 7 xb , g 7 g. Now let x1 , . . . , xr be a basis of h and y1 , . . . , yr a basis of h.
Suppose P = P (x1 , . . . , xr ) (Sh )G . Then we have

[y, P ] = P Sh , where y h,
y
(this follows from the fact that both sides act in the same way in the polynomial represen-
tation, which is faithful). So using the isomorphism , we conclude that if Q (Sh)G , Q =
Q(y1 , . . . , yr ), then [x, Q] = x Q for x h .
Now, to prove the proposition, the only thing we need to check is that M is invariant
under x h . For any v M , we have (Q (Q))N v = 0 for some N. Then
(Q (Q))N +1 xv = (N + 1)x Q (Q (Q))N v = 0.
So xv M .

Corollary 3.17. We have the following decomposition:
M
Oc (G, h) = Oc (G, h) ,
h /G

where Oc (G, h) is the subcategory of modules where (Sh)G acts with generalized eigenvalue
.
Proof. Directly from the definition and the proposition. 
Note that Oc (G, h) is an abelian category closed under taking subquotients and exten-
sions.

3.6. The grading element. Let

X 1 X 2cs
(3.2) h= xi yi + dim h s.
i
2 sS
1 s

Proposition 3.18. We have

[h, x] = x, x h , [h, y] = y, y h.
17
Proof. Let us prove the first relation; the second one is proved similarly. We have
X X 2cs s 1
[h, x] = xi [yi , x] (s , x)s s
i sS
1 s 2
X X X X
= xi (yi , x) xi cs (s, x)(s , yi )s + cs (s , x)s s.
i i sS sS
P P
The last two terms cancel since i xi (s , yi ) = s , so we get i xi (yi , x) = x. 
Proposition 3.19. Let G = W be a real reflection group. Let
X 1 X 1X 2 1X 2
h= xi yi + dim h cs s, E = x, F= y .
i
2 sS
2 i i 2 i i
Then
P
(i) h = i (xi yi + yi xi )/2;
(ii) h, E, F form an sl2 -triple.
Proof. A direct calculation. 
Theorem 3.20. Let M be a module over H1,c (G, h).
(i) If h acts locally nilpotently on M, then h acts locally finitely on M.
(ii) If M is finitely generated over Sh , then M Oc (G, h)0 if and only if h acts locally
finitely on M.
Proof. (i) Assume that Sh acts locally nilpotently on M. Let v M. Then Sh v is a finite
dimensional vector space and let d = dim Sh v. We prove that v is h-finite by induction
in dimension d. We can use d = 0 as base, so only need to do the induction step. The
space Sh v must contain a nonzero vector u such that y u = 0, y h. Let U M be
the subspace of vectors with this property. h acts on U by an element of CG, hence locally
finitely. So it is sufficient to show that the image of v in M/hUi is h-finite (where hUi is
the submodule generated by U). But this is true by the induction assumption, as u = 0 in
M/hUi.
(ii) We need to show that if h acts locally finitely on M, then h acts locally nilpotently
on M. Assume h acts locally finitely on M. Then M = B M[], where B C. Since M
is finitely generated over Sh , B is a finite union of sets of the form z + Z0 , z C. So Sh
must act locally nilpotently on M. 
We can obtain the following corollary easily.
Corollary 3.21. Any finite dimensional H1,c (G, h)-module is in Oc (G, h)0 .
We see that any module M Oc (G, h)0 has a grading by generalized eigenvalues of h:
M = M[].
3.7. Standard modules. Let be a finite dimensional representation of G. The standard
module over H1,c (G, h) corresponding to (also called the Verma module) is
Mc (G, h, ) = Mc ( ) = H1,c (G, h) CGSh Oc (G, h)0 ,
where Sh acts on by zero.
So from the PBW theorem, we have that as vector spaces, Mc ( )
= Sh .
18
Remark 3.22. More generally, h , let G = Stab(), and be a finite dimensional
representation of G . Then we can define Mc, (G, h, ) = H1,c (G, h) CG Sh , where Sh
acts on by . These modules are called the Whittaker modules.
Let be irreducible, and let hc ( ) be the number given by the formula
dim h X 2cs
hc ( ) = s| .
2 sS
1 s

Then we see that h acts on S m h by the scalar hc ( ) + m.

Definition 3.23. A vector v in an H1,c -module M is singular if yi v = 0 for all i.
Proposition 3.24. Let U be an H1,c (G, h)-module. Let U be a G-submodule consisting
of singular vectors. Then there is a unique homomorphism : Mc ( ) U of C[h]-modules
such that | is the identity, and it is an H1,c -homomorphism.
Proof. The first statement follows from the fact that Mc ( ) is a free module over C[h] gen-
erated by . Also, it follows from the Frobenius reciprocity that there must exist a map
which is an H1,c -homomorphism. This implies the proposition. 
3.8. Finite length.
Proposition 3.25. K R such that for any M N in Oc (G, h)0 , if M[] = N[] for
Re () K, then M = N.
Proof. Let K = max Re hc ( ). Then if M 6= N, M/N begins in degree 0 with Re 0 > K,
which is impossible since by Proposition 3.24, 0 must equal hc ( ) for some . 
Corollary 3.26. Any M Oc (G, h)0 has finite length.
Proof. Directly from the proposition. 
3.9. Characters. For M Oc (G, h)0 , define the character of M as the following formal
series in t: X
ch M (g, t) = t Tr M [] (g) = Tr M (gth ), g G.

Proposition 3.27. We have

(g)thc ( )
ch Mc ( ) (g, t) = .
deth (1 tg)
Proof. We begin with the following lemma.
Lemma 3.28 (MacMahons Master theorem). Let V be a finite dimensional space, A : V
V a linear operator. Then
X 1
tn Tr (S n A) = .
n0
det(1 tA)
Proof of the lemma. If A is diagonalizable, this is obvious. The general statement follows by
continuity. 
1
The lemma implies that Tr Sh (gtD ) = where D is the degree operator. This
det(1 gt)
implies the required statement. 
19
3.10. Irreducible modules. Let be an irreducible representation of G.
Proposition 3.29. Mc ( ) has a maximal proper submodule Jc ( ).
Proof. The proof is standard. Jc ( ) is the sum of all proper submodules of Mc ( ), and it is
not equal to Mc ( ) because any proper submodule has a grading by generalized eigenspaces
of h, with eigenvalues such that hc ( ) > 0. 
We define Lc ( ) = Mc ( )/Jc ( ), which is an irreducible module.
Proposition 3.30. Any irreducible object of Oc (G, h)0 has the form Lc ( ) for an unique .
Proof. Let L Oc (G, h)0 be irreducible, with lowest eigenspace of h containing an irreducible
G-module . Then by Proposition 3.24, we have a nonzero homomorphism Mc ( ) L, which
is surjective, since L is irreducible. Then we must have L = Lc ( ). 
Remark 3.31. Let be a character of G. Then we have an isomorphism H1,c (G, h)
H1,c (G, h), mapping g G to 1 (g)g. This automorphism maps Lc ( ) to Lc (1 )
isomorphically.
3.11. The contragredient module. Set c(s) = c(s1 ). We have a natural isomorphism
: H1,c (G, h )op H1,c (G, h), acting trivially on h and h , and sending g G to g 1.
Thus if M is an H1,c (G, h)-module, then the full dual space M is an H1,c (M, h )-module.
If M Oc (G, h)0 , then we can define M , which is the h-finite part of M .
Proposition 3.32. M belongs to Oc (G, h )0 .
Proof. Clearly, if L is irreducible, then so is L . Then L is generated by its lowest h-
eigenspace over H1,c (G, h ), hence over Sh . Thus, L Oc (G, h )0 . Now, let M Oc (G, h)0
be any object. Since M has finite length, so does M . Moreover, M has a finite filtration
with successive quotients of the form L , where L Oc (G, h)0 is irreducible. This implies
the required statement, since Oc (G, h)0 is closed under taking extensions. 
Clearly, M = M. Thus, M 7 M is an equivalence of categories Oc (G, h) Oc (G, h )op .
3.12. The contravariant form. Let be an irreducible representation of G. By Propo-
sition 3.24, we have a unique homomorphism : Mc (G, h, ) Mc (G, h , ) which is the
identity in the lowest h-eigenspace. Thus, we have a pairing
c : Mc (G, h, ) Mc (G, h , ) C,
which is called the contravariant form.
Remark 3.33. If G = W is a real reflection group, then h = h , c = c, and
= via a
symmetric form. So for real reflection groups, c is a symmetric form on Mc ( ).
Proposition 3.34. The maximal proper submodule Jc ( ) is the kernel of (or, equivalently,
of the contravariant form c ).
Proof. Let K be the kernel of the contravariant form. It suffices to show that Mc ( )/K is
irreducible. We have a diagram:

Mc (G, h, ) / M (G, h , )
c
PPP O
PPP
PPP
PPP
 ( ?
/
Lc (G, h, ) L (G, h , )
c
20
Indeed, a nonzero map exists by Proposition 3.24, and it factors through Lc (G, h, ),
with being an isomorphism, since Lc (G, h , ) is irreducible. Now, by Proposition 3.24
(uniqueness of ), the diagram must commute up to scaling, which implies the statement. 
Proposition 3.35. Assume that hc ( ) hc ( ) never equals a positive integer for any ,
IrrepG. Then Oc (G, h)0 is semisimple, with simple objects Mc ( ).
Proof. It is clear that in this situation, all Mc ( ) are simple. Also consider Ext1 (Mc ( ), Mc ( )).
/ Z, it is clearly 0. Otherwise, hc ( ) = hc ( ), and again Ext1 (Mc ( ), Mc ( )) =
If hc ( )hc ( )
0, since for any extension
0 Mc ( ) N Mc ( ) 0,
by Proposition 3.24 we have a splitting Mc ( ) N. 
Remark 3.36. In fact, our argument shows that if Ext1 (Mc ( ), Mc ( )) 6= 0, then hc ( )
hc ( ) N.
3.13. The matrix of multiplicities. For , IrrepG, write < if
Re hc () Re hc ( ) N.
Proposition 3.37. There exists a matrix of integers N = (n, ), with n, 0, such that
n, = 1, n, = 0 unless < , and
X
Mc () = n, Lc ( ) K0 (Oc (G, h)0 ).
Proof. This follows from the Jordan-Holder theorem and the fact that objects in Oc (G, h)0
have finite length. 
Corollary 3.38. Let N 1 = (n, ). Then
X
Lc ( ) = n, Mc ().
Corollary 3.39. We have
P
n, (g)thc ( )
ch Lc ( ) (g, t) = .
deth (1 tg)
Both of the corollaries can be obtained from the above proposition easily.
One of the main problems in the representation theory of rational Cherednik algebras is
the following problem.
Problem: Compute the multiplicities n, or, equivalently, ch Lc ( ) for all .
In general, this problem is open.
3.14. Example: the rank 1 case. Let G = Z/mZ and be an m-th primitive root of 1.
Then the algebra H1,c (G, h) is generated by x, y, s with relations
m1
X
[y, x] = 1 2 cj sj , sxs1 = x, sys1 = 1 y.
j=1

Consider the one-dimensional space C and let y act by 0 and g G act by 1. We have
Mc (C) = C[x]. The contravariant form c,C on Mc (C) is defined by

c,C (xn , xn ) = an ; c,C (xn , xn ) = 0, n 6= n .
21
Recall that c,C satisfies c,C (xn , xn ) = c,C (xn1 , yxn ), which gives
an = an1 (n bn ),
where bn are new parameters:
m1
X 1 jn
bn := 2 cj (b0 = 0, bn+m = bn ).
j=1
1 j
Thus we obtain the following proposition.
Proposition 3.40. (i) Mc (C) is irreducible if only if n bn 6= 0 for any n 1.
(ii) Assume that r is the smallest positive integer such that r = br . Then Lc (C) has
dimension r (which can be any number not divisible by m) with basis 1, x, . . . , xr1 .
Remark 3.41. According to Remark 3.31, this proposition in fact describes all the irre-
ducible lowest weight modules.
Example 3.42. Consider the case m = 2. The Mc (C) is irreducible unless c 1/2 + Z0 .
If c = (2n + 1)/2 1/2 + Z, n 0, then Lc (C) has dimension 2n + 1. A similar answer is
obtained for lowest weight C , replacing c by c.
3.15. The Frobenius property. Let A be a Z+ -graded commutative algebra. The algebra
A is called Frobenius if the top degree A[d] of A is 1-dimensional, and the multiplication
map A[m] A[d m] A[d] is a nondegenerate pairing for any 0 m d. In particular,
the Hilbert polynomial of a Frobenius algebra A is palindromic.
Now, let us go back to considering modules over the rational Cherednik algebra H1,c . Any
submodule J of the polynomial representation Mc (C) = Mc = C[h] is an ideal in C[h], so
the quotient A = Mc /J is a Z+ -graded commutative algebra.
Now suppose that G preserves an inner product in h, i.e., G O(h).
Theorem 3.43. If A = Mc (C)/J is finite dimensional, then A is irreducible (A = Lc (C))
A is a Frobenius algebra.
Proof. 1) Suppose A is an irreducible H1,c -module, i.e., A = Lc (C). By Proposition 3.19, A
is naturally a finite dimensional sl2 -module (in particular, it integrates to the group SL2 (C)).
Hence, by the representation theory of sl2 , the top degree of A is 1-dimensional. Let A
denote a nonzero linear function on the top component. Let c be the contravariant form
on Mc (C). Consider the form
 
0 1
(v1 , v2 ) 7 E(v1 , v2 ) := c (v1 , gv2 ), where g = SL2 (C).
1 0
Then E(xv1 , v2 ) = E(v1 , xv2 ). So for any p, q Mc (C) = C[h], E(p, q) = (p(x)q(x)) (for a
suitable normalization of ).
Since E is a nondegenerate form, A is a Frobenius algebra.
2) Suppose A is Frobenius. Then the highest component is 1-dimensional, and
E : A A C, E(a, b) = (ab) is nondegenerate. We have E(xa, b) = E(a, xb). So
set (a, b) = E(a, g 1b). Then satisfies (a, xi b) = (yi a, b). Thus, for all p, q C[h],
(p(x), q(x)) = (q(y)p(x), 1). So = c up to scaling. Thus, c is nondegenerate and A is
irreducible. 
Remark 3.44. If G * O(h), this theorem is false, in general.
22
 Now consider the Frobenius property of Lc (C) for any G GL(h).
Theorem 3.45. Let U Mc (C) = C[h] be a G-subrepresentation of dimension l = dim h,
sitting in degree r, which consists of singular vectors. Let J = hUi. Assume that A = Mc /J
is finite dimensional. Then
(i) A is Frobenius.
(ii) A admits a BGG type resolution:
A Mc (C) Mc (U) Mc (2 U) Mc (l U) 0.

(iii) The character of A is given by the formula

l P 2cs detU (1 gtr )
A (g, t) = t 2 sS 1s .
deth (1 gt)
In particular, dim A = r l .
(iv) If G preserves an inner product, then A is irreducible.
Proof. (i) Since Spec A is a complete intersection, A is Frobenius.
(ii) We will use the following theorem:
Theorem 3.46 (Serre). Let f1 , . . . , fn C[t1 , . . . , tn ] be homogeneous polynomials, and
assume that C[t1 , . . . , tn ] is a finitely generated module over C[f1 , . . . , fn ]. Then this is a free
module.
Consider SU Sh . Then Sh is a finitely generated SU-module (as Sh /hUi is finite
dimensional). By Serres theorem, we know that Sh is a free SU-module. The rank of this
module is r l . Consider the Koszul complex attached to this module. Since the module is
free, the Koszul complex is exact (i.e., it is a resolution of the zero fiber). At the level of
SU-modules, it looks exactly like we want in (3.45).
So we only need to show that the maps of the resolution are morphisms over H1,c . This
is shown by induction. Namely, let j : Mc (j U) Mc (j1 U) be the corresponding
differentials (so that 0 : Mc (C) A is the projection). Then 0 is an H1,c -morphism, which
is the base of induction. If j is an H1,c -morphism, then the kernel of j is a submodule
Kj Mc (j U). Its lowest degree part is j+1 U sitting in degree (j + 1)r and consisting of
singular vectors. Now, j+1 is a morphism over Sh which maps j+1 U identically to itself.
By Proposition 3.24, there is only one such morphism, and it must be an H1,c -morphism.
This completes the induction step.
(iii) follows from (ii) by the Euler-Poincare formula.
(iv) follows from Theorem 3.43.

3.16. Representations of H1,c of type A. Let us now apply the above results to the case
of type A. We will follow the paper [CE].
Let G = Sn , and h be its reflection representation. In this case the function c reduces
to one number. We will denote the rational Cherednik algebra H1,c (Sn ) by Hc (n). It is
generated by x1 , . . . , xn , y1 , . . . , yn and CSn with the following relations:
X X 1
yi = 0, xi = 0, [yi , xj ] = + csij , i 6= j,
n
23
 n1 X
[yi , xi ] = c sij .
n j6=i

The polynomial representation Mc (C) of this algebra is the space of C[x1 , . . . , xn ]T of poly-
nomials of x1 , . . . , xn , which are invariant under simultaneous translation T : xi 7 xi + a.
In other words, it is the space of regular functions on h = Cn /, where is the diagonal.
Proposition 3.47 (C. Dunkl). Let r be a positive integer not divisible by n, and c = r/n.
Then Mc (C) contains a copy of the reflection representation h of Sn , which consists of
singular vectors (i.e. those killed by y h). This copy sits in degree r and is spanned by the
functions
r dz
fi (x1 , . . . , xn ) = Res [(z x1 ) (z xn )] n .
z xi
(the symbol Res denotes the residue at infinity).
P
Remark 3.48. The space spanned by fi is (n 1)-dimensional, since i fi = 0 (this sum
is the residue of an exact differential).
Proof. This proposition can be proved by a straightforward computation. The functions fi
are a special case of Jack polynomials. 
Let Ic be the submodule of Mc (C) generated by fi . Consider the Hc (n)-module Vc =
Mc (C)/Ic , and regard it as a C[h]-module. We have the following results.
Theorem 3.49. Let d = (r, n) denote the greatest common divisor of r and n. Then the
(set-theoretical) support of Vc is the union of Sn -translates of the subspaces of Cn /, defined
by the equations
x1 = x2 = = x nd ; x nd +1 = = x2 nd ; ... x(d1) nd +1 = = xn .
In particular, the Gelfand-Kirillov dimension of Vc is d 1.
Corollary 3.50 ([BEG]). If d = 1 then the module Vc is finite dimensional, irreducible,
admits a BGG type resolution, and its character is
det |h (1 gtr )
Vc (g, t) = t(1r)(n1)/2 .
det |h (1 gt)
Proof. For d = 1 Theorem 3.49 says that the support of Mc (C)/Ic is {0}. This implies that
Mc (C)/Ic is finite dimensional. The rest follows from Theorem 3.45. 
Proof of Theorem 3.49. The support of Vc is the zero set of Ic , i.e. the common zero set of
n
X
fi . Fix x1 , . . . , xn C. Then fi (x1 , . . . , xn ) = 0 for all i iff i fi = 0 for all i , i.e.
i=1
n n
!
Y r
X i
Res (z xj ) n dz = 0.
j=1 i=1
z xi
Assume that x1 , . . . xn take distinct values y1 , . . . , yp with positive multiplicities m1 , . . . , mp .
The previous equation implies that the point (x1 , . . . , xn ) is in the zero set iff
p p
!
Y X
i (z y1 ) (z\
r
Res (z yj )mj n 1 yi ) (z yp ) dz = 0 i .
j=1 i=1
24
Since i are arbitrary, this is equivalent to the condition
p
Y r
Res (z yj )mj n 1 z i dz = 0, i = 0, . . . , p 1.
j=1

We will now need the following lemma.

p
Y P P
Lemma 3.51. Let a(z) = (z yj )j , where j C, j j Z and j j > p. Suppose
j=1

Res a(z)z i dz = 0, i = 0, 1, . . . , p 2.
Then a(z) is a polynomial.
Proof. Let g(z) be a polynomial. Then
0 = Res d(g(z) a(z)) = Res (g (z)a(z) + a (z)g(z))dz,
and hence !
X j
Res g (z) + g(z) a(z)dz = 0.
i
z y j
Y X j
l
Let g(z) = z (z yj ). Then g (z) + g(z) is a polynomial of degree l + p 1
j j P
z y j
P
with highest coefficient l + p + j 6= 0 (as j > p). This means that for every l 0,
l+p1
Res z a(z)dz is a linear combination of residues of z q a(z)dz with q < l + p 1. By
the assumption of the lemma, this implies by induction in l that all such residues are 0 and
hence a is a polynomial. 
P P
In our case (mj r/n 1) = r p (since mj = n) and the conditions of the lemma are
p
Y r
satisfied. Hence (x1 , . . . , xn ) is in the zero set of Ic iff (z yj )mj n 1 is a polynomial. This
j=1
is equivalent to saying that all mj are divisible by n/d.
We have proved that (x1 , . . . , xn ) is in the zero set of Ic if and only if (z x1 ) (z xn )
is the (n/d)-th power of a polynomial of degree d. This implies the theorem. 
Remark 3.52. For c > 0, the above representations are the only irreducible finite di-
mensional representations of H1,c (Sn ). Namely, it is proved in [BEG] that the only finite
dimensional representations of H1,c (Sn ) are multiples of Lc (C) for c = r/n, and of Lc (C )
(where C is the sign representation) for c = r/n, where r is a positive integer relatively
prime to n.
3.17. Notes. The discussion of the definition of rational Cherednik algebras and their basic
properties follows Section 7 of [E4]. The discussion of the category O for rational Cherednik
algebras follows Section 11 of [E4]. The material in Sections 3.14-3.16 is borrowed from [CE].

25
 4. The Macdonald-Mehta integral
4.1. Finite Coxeter groups and the Macdonald-Mehta integral. Let W be a finite
Coxeter group of rank r with real reflection representation hR equipped with a Euclidean
W -invariant inner product (, ). Denote by h the complexification of hR . The reflection
hyperplanes subdivide hR into |W | chambers; let us pick one of them to be the dominant
chamber and call its interior D. For each reflection hyperplane, pick the perpendicular vector
hR with (, ) = 2 which has positive inner products with elements of D, and call it
the positive root corresponding to this hyperplane. The walls of D are then defined by the
equations (i , v) = 0, where i are simple roots. Denote by S the set of reflections in W ,
and for a reflection s S denote by s the corresponding positive root. Let
Y
(x) = (s , x)
sS

be the corresponding discriminant polynomial. Let Q di , i = 1, . . . , r, be the degrees of the

generators of the algebra C[h]W . Note that |W | = i di .
Let H1,c (W, h) be the rational Cherednik algebra of W . Here we choose c = k as a
constant function. Let Mc = Mc (C) be the polynomial representation of H1,c (W, h), and c
be the contravariant form on Mc defined in Section 3.12. We normalize it by the condition
c (1, 1) = 1.
Theorem 4.1. (i) (The Macdonald-Mehta integral) For Re (k) 0, one has
Z Yr
r/2 (x,x)/2 2k (1 + kdi )
(4.1) (2) e |(x)| dx = .
hR i=1
(1 + k)
(ii) Let b(k) := c (, ). Then
r dY
Y i 1

b(k) = |W | (kdi + m).

i=1 m=1

For Weyl groups, this theorem was proved by E. Opdam [Op1]. The non-crystallographic
cases were done by Opdam in [Op2] using a direct computation in the rank 2 case (reducing
(4.1) to the beta integral by passing to polar coordinates), and a computer calculation by F.
Garvan for H3 and H4 .
Example 4.2. In the case W = Sn , we have the following integral (the Mehta integral):
Z Y n
Y
(n1)/2 (x,x)/2 2k (1 + kd)
(2) P
e |xi xj | dx = .
{xRn | i xi =0} (1 + k)
i6=j d=2

In the next subsection, we give a uniform proof of Theorem 4.1 which is given in [E2]. We
emphasize that many parts of this proof are borrowed from Opdams previous proof of this
theorem.

4.2. Proof of Theorem 4.1.

Proposition 4.3. The function b is a polynomial of degree at most |S|, and the roots of b
are negative rational numbers.
26
Proof. Since has degree |S|, it follows from the definition of b that it is a polynomial of
degree |S|.
Suppose that b(k) = 0 for some k C. Then c (, P ) = 0 for any polynomial P . Indeed,
if there exists a P such that c (, P ) 6= 0, then there exists such a P which is antisymmetric
of degree |S|. Then P must be a multiple of which contradicts the equality c (, ) = 0.
Thus, Mc is reducible and hence has a singular vector, i.e. a nonzero homogeneous poly-
nomial f of positive degreePd living in an irreducible
P representation of W killed by ya .
Applying the element h = i xai yai + r/2 + k sS s to f , we get
d
k= ,
m
P
where m is the eigenvalue of the operator T := sS (1 s) on . But it is clear (by
computing the trace of T ) that m 0 and m Q. This implies that any root of b is
negative rational. 

Denote the Macdonald-Mehta integral by F (k).

Proposition 4.4. One has
F (k + 1) = b(k)F (k).
P
Proof. Let F = i ya2i /2. Introduce the Gaussian inner product on Mc as follows:
Definition 4.5. The Gaussian inner product c on Mc is given by the formula
c (v, v ) = c (exp(F)v, exp(F)v ).
This makes sense because the operator F is locally nilpotent on Mc . Note
P 2 that is a
nonzero W -antisymmetric polynomial of the smallest possible degree, so ( yai ) = 0 and
hence
(4.2) c (, ) = c (, ) = b(k).
For a h, let xa h H1,c (W, h), ya h H1,c (W, h) be the corresponding generators
of the rational Cherednik algebra.
Proposition 4.6. Up to scaling, c is the unique W -invariant symmetric bilinear form on
Mc satisfying the condition
c ((xa ya )v, v ) = c (v, ya v ), a h.
Proof. We have
c ((xa ya )v, v ) = c (exp(F)(xa ya )v, exp(F)v ) = c (xa exp(F)v, exp(F)v )
= c (exp(F)v, ya exp(F)v ) = c (exp(F)v, exp(F)ya v ) = c (v, yav ).
Let us now show uniqueness. If is any W -invariant symmetric bilinear form satisfying
the condition of the Proposition, then let (v, v ) = (exp(F)v, exp(F)v ). Then is
contravariant, so its a multiple of c , hence is a multiple of c . 

Now we will need the following known result (see [Du2], Theorem 3.10).
27
Proposition 4.7. For Re (k) 0 we have
Z
1
(4.3) c (f, g) = F (k) f (x)g(x)dc (x)
hR

where
dc (x) := e(x,x)/2 |(x)|2k dx.
Proof. It follows from Proposition 4.6 that c is uniquely, up to scaling, determined by the
condition that it is W -invariant, and ya = xa ya . These properties are easy to check for the
right hand side of (4.3), using the fact that the action of ya is given by Dunkl operators. 
Now we can complete the proof of Proposition 4.4. By Proposition 4.7, we have
F (k + 1) = F (k)c (, ),
so by (4.2) we have
F (k + 1) = F (k)b(k).

Let Y
b(k) = b0 (k + ki )ni .
We know that ki > 0, and also b0 > 0 (because the inner product 0 on real polynomials is
positive definite).
Corollary 4.8. We have
Y  (k + ki ) ni
F (k) = bk0 .
i
(ki )

Proof. Denote the right hand side by F (k) and let (k) = F (k)/F (k). Clearly, (0) = 1.
Proposition 4.4 implies that (k) is a 1-periodic positive function on [0, ). Also by the
Cauchy-Schwarz inequality,
F (k)F (k ) F ((k + k )/2)2 ,
so log F (k) is convex for k 0. This implies that = 1, since (log F (k)) 0 as k
+. 
Remark 4.9. The proof of this corollary is motivated by the standard proof of the following
well known characterization of the function.
Proposition 4.10. The function is determined by three properties:
(i) (x) is positive on [1, +) and (1) = 1;
(ii) (x + 1) = x(x);
(iii) log (x) is a convex function on [1, +).
R
Proof. It is easy to see from the definition (x) = 0 tx1 et dt that the function has
properties (i) and (ii); property (iii) follows from this definition and the Cauchy-Schwarz
inequality.
Conversely, suppose we have a function F (x) satisfying the above properties, then we have
F (x) = (x)(x) for some 1-periodic function (x) with (x) > 0. Thus, we have
(log F ) = (log ) + (log ) .
28
Since limx+ (log ) = 0, (log F ) 0, and is periodic, we have (log ) 0. Since
R n+1
n
(log ) dx = 0, we see that (log ) 0. So we have (x) 1. 

In particular, we see from Corollary 4.8 and the multiplication formulas for the function
that part (ii) of Theorem 4.1 implies part (i).
It remains to establish (ii).
Proposition 4.11. The polynomial b has degree exactly |S|.
Proof. By Proposition 4.3, b is a polynomial of degree at most |S|. To see that the degree is
precisely |S|, let us make the change of variable x = k 1/2 y in the Macdonald-Mehta integral
and use the steepest descent method. We find that the leading term of the asymptotics of
log F (k) as k + is |S|k log k. This together with the Stirling formula and Corollary 4.8
implies the statement. 
Proposition 4.12. The function
r
Y 1 e2ikdj
G(k) := F (k)
j=1
1 e2ik
analytically continues to an entire function of k.
Proof. Let D be an element. Consider the real hyperplane Ct = it + hR , t > 0. Then
Ct does not intersect reflection hyperplanes, so we have a continuous branch of (x)2k on
Ct which tends to the positive branch in D as t 0. Then, it is easy to see that for any
w W , the limit of this branch in the chamber w(D) will be e2ik(w) |(x)|2k , where (w) is
the length of w. Therefore, by letting t = 0, we get
Z X
r/2 1
(2) e(x,x)/2 (x)2k dx = F (k)( e2ik(w) )
Ct |W | wW

(as this integral does not depend on t by Cauchys theorem). But it is well known that
X Yr
2ik(w) 1 e2ikdj
e = 2ik
,
wW j=1
1 e
([Hu], p.73), so Z
r/2
(2) |W | e(x,x)/2 (x)2k dx = G(k).
Ct
R (x,x)/2 2k
Since Ct
e (x) dx is clearly an entire function, the statement is proved.

Corollary 4.13. For every k0 [1, 0] the total multiplicity of all the roots of b of the
form k0 p, p Z+ , equals the number of ways to represent k0 in the form m/di , m =
1, . . . , di 1. In other words, the roots of b are ki,m = m/di pi,m , 1 m di 1, where
pi,m Z+ .
Proof. We have
Yr
F (k) 1 e2ikdj
G(k p) = ,
b(k 1) b(k p) j=1 1 e2ik
29
Now plug in k = 1 + k0 and a large positive integer p. Since by Proposition 4.12 the left
hand side is regular, so must be the right hand side, which implies the claimed upper bound
for the total multiplicity, as F (1 + k0 ) > 0. The fact that the bound is actually attained
follows from the fact that the polynomial b has degree exactly |S| (Proposition 4.11), and
the fact that all roots of b are negative rational (Proposition 4.3). 
It remains to show that in fact in Corollary 4.13, pi,m = 0 for all i, m; this would imply
(ii) and hence (i).
Proposition 4.14. Identity (4.1) of Theorem 4.1 is satisfied in C[k]/k 2 .
Proof. Indeed, we clearly have F (0) = 1. Next, a rank 1 computation gives F (0) = |S|,
where is the Euler constant (i.e. = limn+ (1 + + 1/n log n)), while the derivative
of the right hand side of (4.1) at zero equals to
r
X
(di 1).
i=1

But it is well known that

r
X
(di 1) = |S|,
i=1

([Hu], p.62), which implies the result. 

Proposition 4.15. Identity (4.1) of Theorem 4.1 is satisfied in C[k]/k 3 .
Note that Proposition 4.15 immediately implies (ii), and hence the whole theorem. Indeed,
it yields that
r dX
X i 1

(log F ) (0) = (log ) (m/di ),
i=1 m=1
so by Corollary 4.13
r dX
X i 1 r dX
X i 1

(log ) (m/di + pi,m ) = (log ) (m/di ),
i=1 m=1 i=1 m=1

which implies that pi,m = 0 since (log ) is strictly decreasing on [0, ).
To prove PropositionP 4.15, we will need the following result about finite Coxeter groups.
Let (W ) = 3|S|2 ri=1 (d2i 1).
Lemma 4.16. One has
X
(4.4) (W ) = (G),
GPar 2 (W )

where Par 2 (W ) is the set of parabolic subgroups of W of rank 2.

Proof. Let
Yr
1q
Q(q) = |W | .
i=1
1 q di
30
It follows from Chevalleys theorem that
X
Q(q) = (1 q)r det(1 qw|h )1 .
wW

Let us subtract the terms for w = 1 and w S from both sides of this equation, divide both
sides by (q 1)2 , and set q = 1 (cf. [Hu], p.62, formula (21)). Let W2 be the set of elements
of W that can be written as a product of two different reflections. Then by a straightforward
computation we get
1 X 1
(W ) = .
24 wW
r Tr h (w)
2

In particular, this is true for rank 2 groups. The result follows, as any element w W2
belongs to a unique parabolic subgroup Gw of rank 2 (namely, the stabilizer of a generic
point hw , [Hu], p.22). 
Proof of Proposition 4.15. Now we are ready to prove the proposition. By Proposition 4.14,
it suffices to show the coincidence of the second derivatives of (4.1) at k = 0. The second
derivative of the right hand side of (4.1) at zero is equal to
r
2 X 2
(d 1) + 2 |S|2 .
6 i=1 i
On the other hand, we have
X Z
r/2
F (0) = (2) e(x,x)/2 log 2 (x) log 2 (x)dx.
,S hR

Thus, from a rank 1 computation we see that our job is to establish the equality
X Z 2 (x)
r
2 X 2
r/2
(2) e(x,x)/2 log 2 (x) log 2 dx = ( (d2i 1) 3|S|2 ) = (W ).
6=S hR
(x) 6 i=1 6

Since this equality holds in rank 2 (as in this case (4.1) reduces to the beta integral), in
general it reduces to equation (4.4) (as for any 6= S, s and s are contained in a
unique parabolic subgroup of W of rank 2). The proposition is proved. 
4.3. Application: the supports of Lc (C). In this subsection we will use the Macdonald-
Mehta integral to computation of the support of the irreducible quotient of the polynoamial
representation of a rational Cherednik algebra (with equal parameters). We will follow the
paper [E3].
First note that the vector space h has a stratification labeled by parabolic subgroups of W .

Indeed, for a parabolic subgroup W W , let hW reg be the set of points in h whose stabilizer
is W . Then a
h= hW
reg ,
W Par(W )

where Par(W ) is the set of parabolic subgroups in W .

For a finitely generated module M over C[h], denote the support of M by supp (M).
The following theorem is proved in [Gi1], Section 6 and in [BE] with different method. We
will recall the proof from [BE] later.
31
Theorem 4.17. Consider the stratification of h with respect to stabilizers of points in W .
Then the support supp (M) of any object M of Oc (W, h) in h is a union of strata of this
stratification.
This makes one wonder which strata occur in supp (Lc ( )), for given c and . In [VV],
Varagnolo and Vasserot gave a partial answer for = C. Namely, they determined (for W
being a Weyl group) when Lc (C) is finite dimensional, which is equivalent to supp (Lc (C)) =
0. For the proof (which is quite complicated), they used the geometry affine Springer fibers.
Here we will give a different (and simpler) proof. In fact, we will prove a more general result.
Recall that for any Coxeter group W , we have the Poincare polynomial:
X Yr
(w) 1 q di (W )
PW (q) = q = , where di (W ) are the degrees of W.
wW i=1
1q

Lemma 4.18. If W W is a parabolic subgroup of W , then PW is divisible by PW .

Proof. By Chevalleys theorem, C[h] is a free module over C[h]W and C[h]W is a direct

summand in this module. So C[h]W is a projective module, thus free (since it is graded).
Hence, there exists a polynomial Q(q) such that we have
Q(q)hC[h]W (q) = hC[h]W (q),
where hV (q) denotes the Hilbert series of a graded vector space V . Notice that we have
1
hC[h]W (q) = , so we have
PW (q)(1 q)r
Q(q) 1
= , i.e. Q(q) = PW (q)/PW (q).
PW (q) PW (q)

Corollary 4.19. If m 2 then we have the following inequality:
#{i|m divides di (W )} #{i|m divides di (W )}.
Proof. This follows from Lemma 4.18 by looking at the roots of the polynomials PW and
PW . 
Our main result is the following theorem.
Theorem 4.20. [E3] Let c 0. Then a supp (Lc (C)) if and only if
PW 2ic
(e ) 6= 0.
PW a
We can obtain the following corollary easily.
Corollary 4.21. (i) Lc (C) 6= Mc (C) if and only if c Q>0 and the denominator m of
c divides di for some i;
PW 2ic
(ii) Lc (C) is finite dimensional if and only if (e ) = 0, i.e., iff
PW
#{i|m divides di (W )} > #{i|m divides di (W )}.
for any maximal parabolic subgroup W W .
32
Remark 4.22. Varagnolo and Vasserot prove that Lc (C) is finite dimensional if and only if
there exists a regular elliptic element in W of order m. Case-by-case inspection shows that
this condition is equivalent to the combinatorial condition of (2). Also, a uniform proof of
this equivalence is given in the appendix to [E3], written by S. Griffeth.
Example 4.23. For type An1 , i.e., W = Sn , we get that Lc (C) is finite dimensional if and
only if the denominator of c is n. This agrees with our previous results in type An1 .
Example 4.24. Suppose W is the Coxeter group of type E7 . Then we have the following
list of maximal parabolic subgroups and the degrees (note that E7 itself is not a maximal
parabolic).

Subgroups E7 D6 A3 A2 A1 A6
Degrees 2,6,8,10,12,14,18 2,4,6,6,8,10 2,3,4,2,3,2 2,3,4,5,6,7
Subgroups A4 A2 E6 D5 A1 A5 A1
Degrees 2,3,4,5,2,3 2,5,6,8,9,12 2,4,5,6,8,2 2,3,4,5,6,2

So Lc (C) is finite dimensional if and only if the denominator of c is 2, 6, 14, 18.

The rest of the subsection is dedicated to the proof of Theorem 4.20. First we recall some
basic facts about the Schwartz space and tempered distributions.
Let S(Rn ) be the set of Schwartz functions on Rn , i.e.
S(Rn ) = {f C (Rn )|, , sup |x f (x)| < }.
This space has a natural topology.
A tempered distribution on Rn is a continuous linear functional on S(Rn ). Let S (Rn )
denote the space of tempered distributions.
We will use the following well known lemma.
2
Lemma 4.25. (i) C[x]ex /2 S(Rn ) is a dense subspace.
(ii) Any tempered distribution has finite order, i.e., N = N() such that if f S(Rn )
satisfying f = df = = dN 1 f = 0 on supp , then h, f i = 0.
Proof of Theorem 4.20. Recall that on Mc (C), we have the Gaussian form c from Section
4.2. We have for Re (c) 0,
Z
(2)r/2 2
c (P, Q) = ex /2 |(x)|2c P (x)Q(x)dx,
FW (c) hR
where P, Q are polynomials and
Z
r/2 2 /2
FW (k) = (2) ex |(x)|2k dx
hR
is the Macdonald-Mehta integral.
Consider the distribution:
(2)r/2
cW = |(x)|2c .
FW (c)
It is well-known that this distribution is meromorphic in c (Bernsteins theorem). Moreover,
since c (P, Q) is a polynomial in c for any P and Q, this distribution is in fact holomorphic
in c C.
33
Proposition 4.26.
FWa PW 2ic
supp (cW ) = {a hR | (c) 6= 0} = {a hR | (e ) 6= 0}
FW PW a
= {a hR |#{i|denominator of c divides di (W )}
= #{i|denominator of c divides di (Wa )}}.
Proof. First note that the last equality follows from the product formula for the Poincare
polynomial, and the second equality from the Macdonald-Mehta identity. Now let us prove
the first equality.
Look at cW near a h. Equivalently, we can consider
(2)r/2
cW (x + a) = |(x + a)|2c
FW (c)
with x near 0. We have
Y Y
W (x + a) = s (x + a) = (s (x) + s (a))
sS sS
Y Y
= s (x) (s (x) + s (a))
sSWa sS\SWa

= Wa (x) (x),
where is a nonvanishing function near a (since s (a) 6= 0 if s / S Wa ).
So near a, we have
FWa
cW (x + a) = (c) cWa (x) ||2c ,
FW
and the last factor is well defined since is nonvanishing. Thus cW (x) is nonzero near a if
FWa
and only if (c) 6= 0 which finishes the proof. 
FW
Proposition 4.27. For c 0,
supp (cW ) = supp Lc (C)R ,
where the right hand side stands for the real points of the support.
Proof. Let a / supp Lc (C) and assume a supp cW . Then we can find a P Jc (C) = ker c
such that P (a) 6= 0. Pick a compactly supported test function Cc (hR ) such that P does
not vanish anywhere on supp , and hcW , i =
6 0 (this can be done since P (a) 6= 0 and cW
is nonzero near a). Then we have /P S(hR ). Thus from Lemma 4.25 (i) it follows that
there exists a sequence of polynomials Pn such that
2
Pn (x)ex /2 in S(hR ), when n .
P
2
So P Pn ex /2 in S(hR ), when n .
2
But we have hcW , P Pn ex /2 i = c (P, Pn ) = 0 which is a contradiction. This implies that
supp cW (supp Lc (C))R .
To show the opposite inclusion, let P be a polynomial on h which vanishes identically on
2
supp cW . By Lemma 4.25 (ii), there exists N such that hcW , P N (x)Q(x)ex /2 i = 0. Thus,
34
for any polynomial Q, c (P N , Q) = 0, i.e. P N Ker c . Thus, P |supp Lc (C) = 0. This implies
the required inclusion, since supp cW is a union of strata. 
Theorem 4.20 follows from Proposition 4.26 and Proposition 4.27. 
4.4. Notes. Our exposition in Sections 4.1 and 4.2 follows the paper [E2]; Section 4.3 follows
the paper [E3].

35
 5. Parabolic induction and restriction functors for rational Cherednik
algebras
5.1. A geometric approach to rational Cherednik algebras. An important property
of the rational Cherednik algebra H1,c (G, h) is that it can be sheafified, as an algebra, over
h/G (see [E1]). More specifically, the usual sheafification of H1,c (G, h) as a Oh/G -module
is in fact a quasicoherent sheaf of algebras, H1,c,G,h. Namely, for every affine open subset
U h/G, the algebra of sections H1,c,G,h (U) is, by definition, C[U] C[h]G H1,c (G, h).
The same sheaf can be defined more geometrically as follows (see [E1], Section 2.9). Let
e
U be the preimage of U in h. Then the algebra H1,c,G,h(U) is the algebra of linear operators
on O(U e ) generated by O(Ue ), the group G, and Dunkl operators
X 2cs s (a)
a (1 s), where a h.
sS
1 s s

5.2. Completion of rational Cherednik algebras. For any b h we can define the
completion H d1,c (G, h)b to be the algebra of sections of the sheaf H1,c,G,h on the formal neigh-
borhood of the image of b in h/G. Namely, H d1,c (G, h)b is generated by regular functions on
the formal neighborhood of the G-orbit of b, the group G, and Dunkl operators.
The algebra H d
1,c (G, h)b inherits from H1,c (G, h) the natural filtration F by order of dif-
ferential operators, and each of the spaces F n H d 1,c (G, h)b has a projective limit topology; the
whole algebra is then equipped with the topology of the nested union (or inductive limit).
Consider the completion of the rational Cherednik algebra at zero, H d1,c (G, h)0 . It naturally
b
contains the algebra C[[h]]. Define the category Oc (G, h) of representations of H d 1,c (G, h)0
which are finitely generated over C[[h]]0 = C[[h]].
We have a completion functor b: Oc (G, h) O bc (G, h), defined by
c= H
M d 1,c (G, h)0 H1,c (G,h) M = C[[h]] C[h] M.

bc (G, h), let E(N) be the subspace spanned by generalized eigenvectors of

Also, for N O
h in N where h is defined by (3.2). Then it is easy to see that E(N) Oc (G, h)0 .
Theorem 5.1. The restriction of the completion functor b to Oc (G, h)0 is an equivalence
bc (G, h). The inverse equivalence is given by the functor E.
of categories Oc (G, h)0 O

Proof. It is clear that M M c, so M E(M)c (as M is spanned by generalized eigenvectors

of h). Let us demonstrate the opposite inclusion. Pick generators m1 , . . . , mr of M which
c). Then v =
are generalized eigenvectors of h with eigenvalues 1 , . . . , r . Let 0 6= v E(M
P N
P (i )
i fi mi , where fi C[[h]]. Assume that (h ) v = 0 for some N. Then v = i fi mi ,
(d)
where for f C[[h]] we denote by f the degree d part of f . Thus v M, so M = E(M ). c
It remains to show that E(N) \ = N, i.e. that N is the closure of E(N). In other words,
letting m denote the maximal ideal in C[[h]], we need to show that the natural map E(N)
N/mj N is surjective for every j.
To do so, note that h preserves the descending filtration of N by subspaces mj N. On
the other hand, the successive quotients of these subspaces, mj N/mj+1 N, are finite dimen-
sional, which implies that h acts locally finitely on their direct sum grN, and moreover each
36
generalized eigenspace is finite dimensional. Now for each C denote by Nj, the general-
ized -eigenspace of h in N/mj N. We have surjective homomorphisms Nj+1, Nj, , and
for large enough j they are isomorphisms. This implies that the map E(N) N/mj N is
surjective for every j, as desired. 
Example. Suppose that c = 0. Then Theorem 5.1 specializes to the well known fact that
the category of G-equivariant local systems on h with a locally nilpotent action of partial
differentiations is equivalent to the category of all G-equivariant local systems on the formal
neighborhood of zero in h. In fact, both categories in this case are equivalent to the category
of finite dimensional representations of G.
We can now define the composition functor J : Oc (G, h) Oc (G, h)0 , by the formula
J (M) = E(M c). The functor J is called the Jacquet functor ([Gi2]).

5.3. The duality functor. Recall that in Section 3.11, for any H1,c (G, h)-module M, the
full dual space M is naturally an H1,c (G, h )-module, via M (a) = M ((a)) .
It is clear that the duality functor defines an equivalence between the category Oc (G, h)0
and O bc (G, h )op , and that M = E(M ) (where M is the contragredient, or restricted dual
module to M defined in Section 3.11).
5.4. Generalized Jacquet functors.
bc (G, h), a vector v M is h-finite if and only if it is
Proposition 5.2. For any M O
h-nilpotent.
Proof. The if part follows from Theorem 3.20. To prove the only if part, assume that
(h )N v = 0. Then for any u S r h v, we have (h + r)N u = 0. But by Theorem 5.1,
the real parts of generalized eigenvalues of h in M are bounded below. Hence S r h v = 0 for
large enough r, as desired. 
According to Proposition 5.2, the functor E can be alternatively defined by setting E(M)
to be the subspace of M which is locally nilpotent under the action of h.
This gives rise to the following generalization of E: for any h we define the functor
E : O bc (G, h) Oc (G, h) by setting E (M) to be the space of generalized eigenvectors of
G
C[h ] in M with eigenvalue . This way, we have E0 = E.
We can also define the generalized Jacquet functor J : Oc (G, h) Oc (G, h) by the
formula J (M) = E (M).c Then we have J0 = J , and one can show that the restriction of
J to Oc (G, h) is the identity functor.
P
5.5. The centralizer construction. For a finite group H, let eH = |H|1 gH g be the
symmetrizer of H.
If G H are finite groups, and A is an algebra containing C[H], then define the algebra
Z(G, H, A) to be the centralizer EndA (P ) of A in the right A-module P = FunH (G, A) of
H-invariant A-valued functions on G, i.e. such functions f : G A that f (hg) = hf (g).
Clearly, P is a free A-module of rank |G/H|, so the algebra Z(G, H, A) is isomorphic to
Mat|G/H| (A), but this isomorphism is not canonical.
The following lemma is trivial.
Lemma 5.3. The functor N 7 I(N) := P A N = FunH (G, N) defines an equivalence of
categories A mod Z(G, H, A) mod.
37
5.6. Completion of rational Cherednik algebras at arbitrary points of h/G. The
following result is, in essence, a consequence of the geometric approach to rational Cherednik
algebras, described in Subsection 5.1. It should be regarded as a direct generalization to the
case of Cherednik algebras of Theorem 8.6 of [L] for affine Hecke algebras.
Let b h. Abusing notation, denote the restriction of c to the set Sb of reflections in Gb
also by c.
Theorem 5.4. One has a natural isomorphism
d
:H d
1,c (G, h)b Z(G, Gb , H1,c (Gb , h)0 ),

d
defined by the following formulas. Suppose that f P = FunGb (G, H 1,c (Gb , h)0 ). Then

((u)f )(w) = f (wu), u G;

for any h ,
((x )f )(w) = (x(b)
w + (w, b))f (w),
(b)
where x h H1,c (G, h), x h H1,c (Gb , h) are the elements corresponding to ; and
for any a h,
X 2cs s (wa)
(b)
(5.1) ((ya )f )(w) = ywa f (w) (b)
(f (w) f (sw)).
1 s xs + s (b)
sS:sG
/ b

(b)
where ya h H1,c (G, h), ya h H1,c (Gb , h).
Proof. The proof is by a direct computation. We note that in the last formula, the fraction
(b)
s (wa)/(xs + s (b)) is viewed as a power series (i.e., an element of C[[h]]), and that only
the entire sum, and not each summand separately, is in the centralizer algebra. 
Remark. Let us explain how to see the existence of without writing explicit formulas,
and how to guess the formula (5.1) for . It is explained in [E1] (see e.g. [E1], Section
2.9) that the sheaf of algebras obtained by sheafification of H1,c (G, h) over h/G is generated
(on every affine open set in h/G) by regular functions on h, elements of G, and Dunkl
operators. Therefore, this statement holds for formal neighborhoods, i.e., it is true on the
formal neighborhood of the image in h/G of any point b h. However, looking at the formula
for Dunkl operators near b, we see that the summands corresponding to s S, s / Gb are
actually regular at b, so they can be safely deleted without changing the generated algebra
(as all regular functions on the formal neighborhood of b are included into the system of
generators). But after these terms are deleted, what remains is nothing but the Dunkl
operators for (Gb , h), which, together with functions on the formal neighborhood of b and
the group Gb , generate the completion of H1,c (Gb , h). This gives a construction of without
using explicit formulas.
Also, this argument explains why should be defined by formula (5.1) of Theorem 5.4.
Indeed, what this formula does is just restores the terms with s / Gb that have been
previously deleted.
The map defines an equivalence of categories
d
: H d
1,c (G, h)b mod Z(G, Gb , H1,c (Gb , h)0 ) mod.
38
Corollary 5.5. We have a natural equivalence of categories

: Oc (G, h) Oc (G , h/hG )0 .

Proof. The category Oc (G, h) is the category of modules over H1,c (G, h) which are finitely
generated over C[h] and extend by continuity to the completion of the algebra H1,c (G, h)
at . So it follows from Theorem 5.4 that we have an equivalence Oc (G, h) Oc (G , h)0 .
Composing this equivalence with the equivalence : Oc (G , h)0 Oc (G , h/hG )0 , we obtain
the desired equivalence . 

Remark 5.6. Note that in this proof, we take the completion of H1,c (G, h) at a point of
h rather than b h.

5.7. The completion functor. Let O bc (G, h)b be the category of modules over H
d 1,c (G, h)b
d
which are finitely generated over C[h]b .

Proposition 5.7. The duality functor defines an anti-equivalence of categories Oc (G, h)

bc (G, h ) .
O

Proof. This follows from the fact (already mentioned above) that Oc (G, h) is the category
of modules over H1,c (G, h) which are finitely generated over C[h] and extend by continuity
to the completion of the algebra H1,c (G, h) at . 

Let us denote the functor inverse to also by ; it is the functor of continuous dual (in
the formal series topology).
We have an exact functor of completion at b, Oc (G, h)0 O bc (G, h)b , M 7 M
cb . We also
b
have a functor E : O bc (G, h) Oc (G, h)0 in the opposite direction, sending a module N to
b
b
the space E (N) of h-nilpotent vectors in N.

Proposition 5.8. The functor E b is right adjoint to the completion functor cb .

Proof. We have

HomHd cb , N) = Hom d
(M d
1,c (G,h)b H1,c (G,h)b (H1,c (G, h)b H1,c (G,h) M, N)

= HomH1,c (G,h) (M, N|H1,c (G,h) ) = HomH1,c (G,h) (M, E b (N)).



Remark 5.9. Recall that by Theorem 5.1, if b = 0 then these functors are not only adjoint
but also inverse to each other.

Proposition 5.10. c) in Oc (G, h)0 .

(i) For M Oc (G, h )b , one has E b (M ) = (M
(ii) For M Oc (G, h)0 , (Mcb ) = Eb (M ) in Oc (G, h )b .
b
(iii) The functors Eb , E are exact.

Proof. (i),(ii) are straightforward from the definitions. (iii) follows from (i),(ii), since the
completion functors are exact. 
39
5.8. Parabolic induction and restriction functors for rational Cherednik algebras.
Theorem 5.4 allows us to define analogs of parabolic restriction functors for rational Chered-
nik algebras.

Namely, let b h, and Gb = G . Define a functor Resb : Oc (G, h)0 Oc (G , h/hG )0 by
the formula
cb ).
Resb (M) = ( E I 1 )(M
We can also define the parabolic induction functors in the opposite direction. Namely, let

N Oc (G , h/hG )0 . Then we can define the object Indb (N) Oc (G, h)0 by the formula
Indb (N) = (E b 1 I)(\
1 (N) ).
0

Proposition 5.11. (i) The functors Indb , Resb are exact.

cb ).
(ii) One has Indb (Resb (M)) = E b (M
Proof. Part (i) follows from the fact that the functor E b and the completion functor b b are
exact (see Proposition 5.10). Part (ii) is straightforward from the definition. 
Theorem 5.12. The functor Indb is right adjoint to Resb .
Proof. We have
Hom(Resb (M), N) = Hom(( E I 1 )(M cb ), N) = Hom((E I 1 )(M
cb ), 1 (N))
cb ), \
= Hom((I 1 )(M cb , (1 I)(\
1 (N) ) = Hom(M 1 (N) ))
0 0
b 1 \1
= Hom(M, (E I)( (N) )) = Hom(M, Indb (N)).
0

At the end we used Proposition 5.8. 

Then we can obtain the following corollary easily.
Corollary 5.13. The functor Resb maps projective objects to projective ones, and the functor
Indb maps injective objects to injective ones.

We can also define functors res : Oc (G, h)0 Oc (G , h/hG )0 and ind : Oc (G , h/hG )0
G
Oc (G, h)0 , attached to hreg , by
res := Res , ind := Ind ,
where is as in Subsection 5.3.
Corollary 5.14. The functors res , ind are exact. The functor ind is left adjoint to res .
The functor ind maps projective objects to projective ones, and the functor res injective
objects to injective ones.
Proof. Easy to see from the definition of the functors and the Theorem 5.12. 
We also have the following proposition, whose proof is straightforward.
Proposition 5.15. We have
ind (N) = (J 1 )(N), and c),
res (M) = ( E )(M
where is defined in Corollary 5.5.
40
5.9. Some evaluations of the parabolic induction and restriction functors. For
generic c, the category Oc (G, h) is semisimple, and naturally equivalent to the category
RepG of finite dimensional representations of G, via the functor 7 Mc (G, h, ). (If G is
a Coxeter group, the exact set of such c (which are called regular) is known from [GGOR]
and [Gy]).
Proposition 5.16. (i) Suppose that c is generic. Upon the above identification, the
functors Indb , ind and Resb , res go to the usual induction and restriction functors
between categories RepG and RepG . In other words, we have
G
c n Mc (G , h/h , ),
Resb (Mc (G, h, )) = G
and

Indb (Mc (G , h/hG , )) = Gb n Mc (G, h, ),
where n is the multiplicity of occurrence of in |G , and similarly for res , ind .
(ii) The equations of (i) hold at the level of Grothendieck groups for all c.
Proof. Part (i) is easy for c = 0, and is obtained for generic c by a deformation argument.
Part (ii) is also obtained by deformation argument, taking into account that the functors
Resb and Indb are exact and flat with respect to c. 
Example 5.17. Suppose that G = 1. Then Resb (M) is the fiber of M at b, while Indb (C) =
PKZ , the object defined in [GGOR], which is projective and injective (see Remark 5.22). This
shows that Proposition 5.16 (i) does not hold for special c, as PKZ is not, in general, a direct
sum of standard modules.
5.10. Dependence of the functor Resb on b. Let G G be a parabolic subgroup. In
the construction of the functor Resb , the point b can be made a variable which belongs to

the open set hG
reg .
Namely, let hd G be the formal neighborhood of the locally closed set hG in h, and let
reg reg
: hdG h/G be the natural map (note that this map is an etale covering of the image with
reg
d
the Galois group NG (G )/G , where NG (G ) is the normalizer of G in G). Let H 1,c (G, h)hG
reg
be the pullback of the sheaf H1,c,G,h under . We can regard it as a sheaf of algebras over

hG
reg . Similarly to Theorem 5.4 we have an isomorphism

d
:H 1,c (G, h)hG
d
Z(G, G , H
G
1,c (G , h/h )0 )D(h
G
reg reg ),
G
where D(hG
reg ) is the sheaf of differential operators on hreg , and is an appropriate completion
of the tensor product.
Thus, repeating the construction of Resb , we can define the functor

Res : Oc (G, h)0 Oc (G , h/hG )0 Loc(hG
reg ),

where Loc(hG G
reg ) stands for the category of local systems (i.e. O-coherent D-modules) on hreg .
This functor has the property that Resb is the fiber of Res at b. Namely, the functor Res is
defined by the formula
Res(M) = (E I 1 )(Mc G ),
hreg

c G is the restriction of the sheaf M on h to the formal neighborhood of hG

where M

reg .
hreg
41
Remark 5.18. If G is the trivial group, the functor Res is just the KZ functor from [GGOR],
which we will discuss later. Thus, Res is a relative version of the KZ functor.
Remark 5.19. Note that the object Res(M) is naturally equivariant under the group
NG (G )/G .
Thus, we see that the functor Resb does not depend on b, up to an isomorphism. A similar
statement is true for the functors Indb , res , ind .
Conjecture 5.20. For any b h, h such that Gb = G , we have isomorphisms of
= res , Indb
functors Resb = ind .
Remark 5.21. Conjecture 5.20 would imply that Indb is left adjoint to Resb , and that Resb
maps injective objects to injective ones, while Indb maps projective objects to projective
ones.
Remark 5.22. If b and are generic (i.e., Gb = G = 1) then the conjecture holds. Indeed,
in this case the conjecture reduces to showing that we have an isomorphism of functors
Fiber b (M)
= Fiber (M ) (M Oc (G, h)). Since both functors are exact functors to the
category of vector spaces, it suffices to check that dim Fiber b (M) = dim Fiber (M ). But this
is true because both dimensions are given by the leading coefficient of the Hilbert polynomial
of M (characterizing the growth of M).
It is important to mention, however, that although Resb is isomorphic to Resb if Gb = Gb ,
this isomorphism is not canonical. So let us examine the dependence of Resb on b a little
more carefully.
Theorem 5.16 implies that if c is generic, then

Res(Mc (G, h, )) = Mc (G , h/hG , ) L ,

where L is a local system on hG
reg of rank n . Let us characterize the local system L
explicitly.
Proposition 5.23. The local system L is given by the connection on the trivial bundle
given by the formula
X 2cs ds
=d (1 s).

1 s s
sS:sG
/

with values in HomG (, |G ).

Proof. This follows immediately from formula (5.1). 
Definition 5.24. We will call the connection of Proposition 5.23 the parabolic KZ (Knizhnik-
Zamolodchikov) connection.
Example 5.25. Let G = Sn and G = Sn1 Snk with n1 + + nk = n. In this case,
there is only one parameter c.
Let h = Cn be the permutation representation of G. Then

hG = (Cn )G = {v h|v = (z1 , . . . , z1 , z2 , . . . , z2 , . . . , zk , . . . , zk )}.
| {z } | {z } | {z }
n1 n2 nk
42
Thus, the parabolic KZ connection on the trivial bundle with fiber being a representation
of Sn has the form
n1 ++np n1 ++nq
X X X dzp dzq
dc (1 sij ).
1p<qk i=n1 ++np1 +1 j=n1 ++nq1 +1
zp zq

So the differential equations for a flat section F (z) of this bundle have the form
F X n1 ++n
X p n1 ++nq
X (1 sij )F
=c .
zp q6=p i=n ++n
zp z q
1 p1 +1 j=n1 ++nq1 +1
Q cnp nq
So F (z) = G(z) p<q (zp zq ) , where the function G satisfies the differential equation

G X n1 ++n
X p n1 ++nq
X sij G
= c .
zp q6=p i=n ++n
z zq
+1 p
1 p1 +1 j=n1 ++nq1

Let = V n where V is a finite dimensional space with dim V = N (this class of repre-
sentations contains as summands all irreducible representations of Sn ). Let Vp = V np , so
that = V1 Vk . Then the equation for G can be written as
G X pq G
= c , p = 1, . . . , k,
zp q6=p
zp zq
PN
where = s,t=1 Es,t Et,s is the Casimir element for glN (Ei,j is the N by N matrix with
the only 1 at the (i, j)-th place, and the rest of the entries being 0).
This is nothing but the well known Knizhnik-Zamolodchikov system of equations of the
WZW conformal field theory, for the Lie algebra glN , see [EFK]. (Note that the repre-
sentations Vi are the most general in the sense that any irreducible finite dimensional
representation of glN occurs in V r for some r, up to tensoring with a character.)
This motivates the term parabolic KZ connection.
5.11. Supports of modules. The following two basic propositions are proved in [Gi1],
Section 6. We will give different proofs of them, based on the restriction functors.
Proposition 5.26. Consider the stratification of h with respect to stabilizers of points in G.
Then the (set-theoretical) support SuppM of any object M of Oc (G, h) in h is a union of
strata of this stratification.
Proof. This follows immediately from the existence of the flat connection along the set of
points b with a fixed stabilizer G on the bundle Resb (M). 
Proposition 5.27. For any irreducible object M in Oc (G, h), SuppM/G is an irreducible
algebraic variety.
Proof. Let X be a component of SuppM/G. Let M be the subspace of elements of M
whose restriction to a neighborhood of a generic point of X is zero. It is obvious that M is
an H1,c (G, h)-submodule in M. By definition, it is a proper submodule. Therefore, by the
irreducibility of M, we have M = 0. Now let f C[h]G be a function that vanishes on X.
Then there exists a positive integer N such that f N maps M to M , hence acts by zero on
M. This implies that SuppM/G = X, as desired. 
43
 Propositions 5.26 and 5.27 allow us to attach to every irreducible module M Oc (G, h),
a conjugacy class of parabolic subgroups, CM Par(G), namely, the conjugacy class of the
stabilizer of a generic point of the support of M. Also, for a parabolic subgroup G G,
denote by X (G ) the set of points b h whose stabilizer contains a subgroup conjugate to
G .
The following proposition is immediate.
Proposition 5.28. (i) Let M Oc (G, h)0 be irreducible. If b is such that Gb CM ,
then Resb (M) is a nonzero finite dimensional module over H1,c (Gb , h/hGb ).
(ii) Conversely, let b h, and L be a finite dimensional module H1,c (Gb , h/hGb ). Then
the support of Indb (L) in h is X (Gb ).
Let FD(G, h) be the set of c for which H1,c (G, h) admits a finite dimensional representation.
Corollary 5.29. Let G be a parabolic subgroup of G. Then X (G ) is the support of some

irreducible representation from Oc (G, h)0 if and only if c FD(G , h/hG ).
Proof. Immediate from Proposition 5.28. 
Example 5.30. Let G = Sn , h = Cn1 . In this case, the set Par(G) is the set of partitions
of n. Assume that c = r/m, (r, m) = 1, 2 m n. By a result of [BEG], finite dimensional
representations of Hc (G, h) exist if and only if m = n. Thus the only possible classes CM
for irreducible modules M have stabilizers Sm Sm , i.e., correspond to partitions into
parts, where each part is equal to m or 1. So there are [n/m] + 1 possible supports for
modules, where [a] denotes the integer part of a.
5.12. Notes. Our discussion of the geometric approach to rational Cherednik algebras in
Section 5.1 follows [E1] and Section 2.2 of [BE]. Our exposition in the other sections follows
the corresponding parts of the paper [BE].

44
 6. The Knizhnik-Zamolodchikov functor
6.1. Braid groups and Hecke algebras. Let G be a complex reflection group and let h
be its reflection representation. For any reflection hyperplane H h, its pointwise stabilizer
is a cyclic group of order mH . Fix a collection of nonzero constants q1,H , . . . , qmH 1,H which
are G-invariant, namely, if H and H are conjugate to each other under some element in G,
then qi,H = qi,H for i = 1, . . . , mH 1.
Let BG = 1 (hreg /G, x0 ) be the braid group of G, and TH BG be a representative of
the conjugacy class defined by a small circle around the image of H in h/G oriented in the
counterclockwise direction.
The following theorem follows from elementary algebraic topology.
Proposition 6.1. The group G is the quotient of the braid group BG by the relations
THmH = 1
for all reflection hyperplanes H.
Proof. See, e.g., [BMR] Proposition 2.17. 
Definition 6.2. The Hecke algebra of G is defined to be
mY
H 1

Hq (G) = C[BG ]/h(TH 1) (TH exp(2ij/mH )qj,H ), for all Hi.

j=1

Thus, by Proposition 6.1 we have an isomorphism

H1 (G)
= CG.
So Hq (G) is a deformation of CG.
Example 6.3 (Coxeter group case). Now let W be a Coxeter group. Let S be the set of
reflections and let s = 0 be the reflection hyperplane corresponding to s S. The Hecke
algebra Hq (W ) is the quotient of C[BW ] by the relations
(Ts 1)(Ts + qs ) = 0,
for all reflections s where Ts is a small counterclockwise circle around the image of the
hyperplane s = 0 in h/W .
6.2. KZ functors. For a complex reflection group G, let Loc(hreg ) be the category of local
systems (i.e., O-coherent D-modules) on hreg , and let Loc(hreg )G be the category of G-
equivariant local systems on hreg , i.e. of local systems on hreg /G.
Suppose that G = 1 is the trivial subgroup in G. Then the restriction functor defined in
Section 5.10 defines a functor Res : Oc (G, h)0 Loc(hreg /G). Also, we have the monodromy
functor Mon : Loc(hreg /G) = Rep(BG ). The composition of these two functors is a functor
from Oc (G, h)0 to Rep(BG ), which is exactly the KZ functor defined in [GGOR]. We will
denote this functor by KZ.
Theorem 6.4 (Ginzburg, Guay, Opdam, Rouquier, [GGOR]). The KZ functor factors
through
RepHq (G), where
mX
H 1
csH (1 e2ij/mH )
qj,H = exp(2ibj,H /mH ), and bj,H = 2 .
1 e2i/mH
=1
45
Proof. Assume first that c is generic. Then the category Oc (G, h)0 is semisimple, with simple
objects Mc ( ), so it is enough to check the statement on Mc ( ). Consider the trivial bundle
over hreg with fiber . The KZ connection on it has the form
X 2cs ds
d (1 s).
sS
1 s s
The residue of the connection form of this connection on the hyperplane H on the j-th
irreducible representation of Z/mH Z is
mX
H 1
csH
2 (1 e2ij/mH ).
=1
1 e2i/mH
Therefore, the monodromy operator around this hyperplane is diagonalizable, and the eigen-
values of this operator are 1 and exp(2ij/mH )qj,H , as desired.
For special c, introduce the generalized Verma module
Mc,n ( ) = Hc (G, h) CGSh ( Sh/mn+1 ),
where m Sh is the maximal ideal of 0, n 0. Clearly, Mc,0 = Mc ( ). Moreover,
Mc,n Oc (G, h)0 for any n, since it has a finite filtration whose successive quotients are
Verma modules.
Theorem 6.5. For large enough n, Mc,n (CG) contains a direct summand which is a projec-
tive generator of Oc (G, h)0 .
Proof. From the definition, Mc,n = Sh CG Sh/mn+1 . Let be the degree operator on
Mc,n (CG) with deg h = 1, deg h = 1, and deg G = 0, i.e., we have
[, x] = x, [, y] = y, where x h , y h.
So h is a module endomorphism of Mc,n (CG) where h is the operator defined in (3.2).
Moreover, h acts locally finitely. In particular, we have a decomposition of Mc,n (CG)
into generalized eigenspaces of h :
M

Mc,n (CG) = Mc,n (CG).
C

We have
Hom(Mc,n (CG), N) = {vectors in N which are killed by mn+1 },
and

Hom(Mc,n (CG), N) = {vectors in N which are killed by mn+1
and are generalized eigenvectors of h with generalized eigenvalue }.
Let P= {hc ( )| is a irreducible representation of G} (recall that
dim h 2cs
hc ( ) = 2
sS 1s s| ), and let
M

Mc,n (CG) = Mc,n (CG).


Claim: for large n, Mc,n (CG) is a projective generator of Oc (G, h)0 .
46

Proof of the claim. First, for any , there exists n such that Mc,n (CG) is projective (since
n+1
the condition of being killed by m is vacuous for large n).

Secondly, consider the functor Hom(Mc,n (CG), ). For any module N Oc (G, h)0 , if

Hom(Mc,n (CG), N) = 0, then N[] = 0. So N = 0. Thus this functor does not kill

nonzero objects, and so Mc,n (CG) is a generator. 
Theorem 6.5 is proved. 
Corollary 6.6. (i) Oc (G, h)0 has enough projectives, so it is equivalent to the category
of modules over a finite dimensional algebra.
(ii) Any object of Oc (G, h)0 is a quotient of a multiple of Mc,n (CG) for large enough n.
Proof. Directly from the definition and the above theorem. 
Now we can finish the proof of Theorem 6.4. We have shown that for generic c,
KZ(Mc,n (CG)) RepHq (G). Hence this is true for any c, since Mc,n (CG) is a flat fam-
ily of modules over Hc (G, h). Then, KZ(M) is a Hq (G)-module for all M, since any M is a
quotient of Mc,n (CG) and the functor KZ is exact. 
Corollary 6.7 (Broue, Malle, Rouquier, [BMR]). Let qj,H = exp(tj,H ) where tj,H s are
formal parameters. Then Hq (G) is a free module over C[[tj,H ]] of rank |G|.
Proof. We have
Hq (G)/(t) = H1 (G) = CG.
So it remains to show that Hq (G) is free. To show this, it is sufficient to show that any
IrrepG admits a flat deformation q to a representation of Hq (G). We can define this
deformation by letting q = KZ(Mc ( )). 
Remark 6.8. 1. The validity of this Corollary in characteristic zero implies that it is also
valid over a field positive characteristic.
2. It is not known in general if the Corollary holds for numerical q (even generically). This
is a conjecture of Broue, Malle, and Rouquier. But it is known for many cases (including all
Coxeter groups).
3. The proof of the Corollary is analytic (it is based on the notion of monodromy). There
is no known algebraic proof, except in special cases, and in the case of Coxeter groups, which
well discuss later.
6.3. The image of the KZ functor. First, let us recall the definition of a quotient category.
Let A be an abelian category and B A a full abelian subcategory.
Definition 6.9. B is a Serre subcategory if it is closed under subquotients and extensions
(i.e., if two terms in a short exact sequence are in B, so is the third one).
If B A is a Serre subcategory, one can define a category A/B as follows:
objects in A/B = objects in A,
HomA/B (X, Y ) = lim HomA (X , Y /Y ).

Y ,X/X B

The category A/B is an abelian category with the following universal property: any exact
functor F : A C that kills B must factor through A/B.
47
 Now let Oc (G, h)tor
0 be the full subcategory of Oc (G, h)0 consisting of modules supported
on the reflection hyperplanes. It is a Serre subcategory, and ker(KZ) = Oc (G, h)tor
0 . Thus
we have a functor:
KZ : Oc (G, h)0 /Oc (G, h)tor
0 RepHq (G).

Theorem 6.10 (Ginzburg, Guay, Opdam, Rouquier, [GGOR]). If dim Hq (G) = |G|, the
functor KZ is an equivalence of categories.
Proof. See [GGOR], Theorem 5.14. 
6.4. Example: the symmetric group Sn . Let h = Cn , G = Sn . Then we have (for
q C ):
Hq (Sn ) = hT1 , . . . , Tn1 i/hthe braid relations and (Ti 1)(Ti + q) = 0i.
The following facts are known:
(1) dim Hq (Sn ) = n!;
(2) Hq (Sn ) is semisimple if and only if ord(q) 6= 2, 3, . . . , n.
Now suppose q is generic. Let be a partition of n. Then we can define an Hq (Sn )-
module S , the Specht module for the Hecke algebra in the sense of [DJ]. This is a certain
deformation of the classical irreducible Specht module for the symmetric group. The Specht
module carries an inner product h, i. Denote D = S /Radh, i.
Theorem 6.11 (Dipper, James, [DJ]). D is either an irreducible Hq (Sn )-module, or 0.
D 6= 0 if and only if is e-regular where e = ord(q) (i.e., every part of occurs less than
e times).
Proof. See [DJ], Theorem 6.3, 6.8. 
Now let Mc () be the Verma module associated to the Specht module for Sn and Lc ()
be its irreducible quotient. Then we have the following theorem.
Theorem 6.12. If c 0, then KZ(Mc ()) = S and KZ(Lc ()) = D .
Proof. See Section 6.2 of [GGOR]. 
Corollary 6.13. If c 0, then SuppLc () = Cn if and only if is e-regular. If c > 0,
then SuppLc () = Cn if and only if is e-regular, or equivalently, is e-restricted (i.e.,
i i+1 < e for i = 1, . . . , n 1).
Proof. Directly from the definition and the above theorem. 
6.5. Notes. The references for this section are [GGOR], [BMR].

48
 7. Rational Cherednik algebras and Hecke algebras for varieties with
group actions
7.1. Twisted differential operators. Let us recall the theory of twisted differential oper-
ators (see [BB], section 2).
Let X be a smooth affine algebraic variety over C. Given a closed 2-form on X, the
algebra D (X) of differential operators on X twisted by can be defined as the algebra
generated by OX and Lie derivatives Lv , v Vect(X), with defining relations
f Lv = Lf v , [Lv , f ] = Lv f, [Lv , Lw ] = L[v,w] + (v, w).
This algebra depends only on the cohomology class [] of , and equals the algebra D(X)
of usual differential operators on X if [] = 0.
An important special case of twisted differential operators is the algebra of differential
operators on a line bundle. Namely, let L be a line bundle on X. Since X is affine, L admits
an algebraic connection with curvature , which is a closed 2-form on X. Then it is easy
to show that the algebra D(X, L) of differential operators on L is isomorphic to D (X).
If the variety X is smooth but not necessarily affine, then (sheaves of) algebras of twisted
differential operators are classified by the space H2 (X, 1 1
X ), where X is the two-step com-
plex of sheaves 1X 2,cl
X , given by the De Rham differential acting from 1-forms to closed
2-forms (sitting in degrees 1 and 2, respectively). If X is projective then this space is
isomorphic to H2,0 (X, C) H1,1 (X, C). We refer the reader to [BB], Section 2, for details.
Remark 7.1. One can show that D (X) is the universal deformation of D(X) (see [E1]).
7.2. Some algebraic geometry preliminaries. Let Z be a smooth hypersurface in a
smooth affine variety X. Let i : Z X be the corresponding closed embedding. Let
N denote the normal bundle of Z in X (a line bundle). Let OX (Z) denote the module
of regular functions on X \ Z which have a pole of at most first order at Z. Then we
have a natural map of OX -modules : OX (Z) i N. Indeed, we have a natural residue
map : OX (Z) OX 1X i OZ (where 1X is the module of 1-forms), hence a map
: OX (Z) i OZ OX T X = i (T X|Z ) (where T X is the tangent bundle). The map
is obtained by composing with the natural projection T X|Z N.
We have an exact sequence of OX -modules:

0 OX OX (Z)
i N 0
Thus we have a natural surjective map of OX -modules Z : T X OX (Z)/OX .

7.3. The Cherednik algebra of a variety with a finite group action. We will now
generalize the definition of Ht,c (G, h) to the global case. Let X be an affine algebraic variety
over C, and G be a finite group of automorphisms of X. Let E be a G-invariant subspace
of the space of closed 2-forms on X, which projects isomorphically to H2 (X, C). Consider
the algebra G OT X , where T X is the cotangent bundle of X. We are going to define a
deformation Ht,c, (G, X) of this algebra parametrized by
(1) complex numbers t,
(2) G-invariant functions c on the (finite) set S of pairs s = (Y, g), where g G, and Y
is a connected component of the set of fixed points X g such that codimY = 1, and
(3) elements E G = H2 (X, C)G .
49
 If all the parameters are zero, this algebra will conicide with G OT X .
Let t, c = {c(Y, g)}, E G be variables. Let D/t (X)r be the algebra (over C[t, t1 , ])
of twisted (by /t) differential operators on X with rational coefficients.
Definition 7.2. A Dunkl-Opdam operator for (X, G) is an element of D/t (X)r [c] given by
the formula
X 2c(Y, g)
(7.1) D := tLv fY (x) (1 g),
1 Y,g
(Y,g)S

where Y,g is the eigenvalue of g on the conormal bundle to Y , v (X, T X) is a vector

field on X, and fY OX (Z) is an element of the coset Y (v) OX (Z)/OX (recall that Y
is defined in Subsection 7.2).
Definition 7.3. The algebra Ht,c, (X, G) is the subalgebra of G D/t (X)r [c] generated
(over C[t, c, ]) by the function algebra OX , the group G, and the Dunkl-Opdam operators.
By specializing t, c, to numerical values, we can define a family of algebras over C, which
we will also denote Ht,c, (G, X). Note that when we set t = 0, the term tLv does not become
0 but turns into the classical momentum.
Definition 7.4. Ht,c, (G, X) is called the Cherednik algebra of the orbifold X/G.
Remark 7.5. One has H1,0, (G, X) = G D (X). Also, if 6= 0 then Ht,c, (G, X) =
Ht,c, (G, X).
Example 7.6. X = h is a vector space and G is a subgroup in GL(h). Let v be a constant
vector field, and let fY (x) = (Y , v)/Y (x), where Y h is a nonzero functional vanishing
on Y . Then the operator D is just the usual Dunkl-Opdam operator Dv in the complex
reflection case (see Section 2.5). ThisPimplies that all the Dunkl-Opdam operators in the
sense of Definition 7.2 have the form fi Dyi + a, where fi C[h], a G C[h], and Dyi
are the usual Dunkl-Opdam operators (for some basis yi of h). So the algebra Ht,c (G, h) =
Ht,c,0 (G, X) is the rational Cherednik algebra for (G, h), see Section 3.1.
The algebra Ht,c, (G, X) has a filtration F which is defined on generators by deg(OX ) =
deg(G) = 0, deg(D) = 1 for Dunkl-Opdam operators D.
Theorem 7.7 (the PBW theorem). We have
grF (Ht,c, (G, X)) = G O(T X)[t, c, ].
Proof. Suppose first that X = h is a vector space and G is a subgroup in GL(h). Then, as
we mentioned, Ht,c, (G, h) = Ht,c (G, h) is the rational Cherednik algebra for G, h. So in this
case the theorem is true.
Now consider arbitrary X. We have a homomorphism of graded algebras
: grF (Ht,c, (G, X)) G O(T X)[t, c, ] (the principal symbol homomorphism).
The homomorphism is clearly surjective, and our job is to show that it is injective (this
is the nontrivial part of the proof). In each degree, is a morphism of finitely generated
G
OX -modules. Therefore, to check its injectivity, it suffices to check the injectivity on the
formal neighborhood of each point z X/G.
Let x be a preimage of z in X, and Gx be the stabilizer of x in G. Then Gx acts on the
formal neighborhood Ux of x in X.
50
Lemma 7.8. Any action of a finite group on a formal polydisk over C is linearizable.
Proof. Let D be a formal polydisk over C. Suppose we have an action of a finite group G
on D. Then we have a group homomorphism:
: G Aut(D) = GLn (C) AutU (D),
where AutU (D) is the group of unipotent automorphisms of D (i.e. those whose derivative
at the origin is 1), which is a prounipotent algebraic group.
Our job is to show that the image of G under can be conjugated into GLn (C). The
obstruction to this is in the cohomology group H1 (G, AutU (D)), which is trivial since G is
finite and AutU (D) is prounipotent over C. 
It follows from Lemma 7.8 that it suffices to prove the theorem in the linear case, which
has been accomplished already. We are done. 
Remark 7.9. The following remark is meant to clarify the proof of Theorem 7.7. In the case
X = h, the proof of Theorem 7.7 is based, essentially, on the (fairly nontrivial) fact that the
usual Dunkl-Opdam operators Dv commute with each other. It is therefore very important
to note that in contrast with the linear case, for a general X we do not have any natural
commuting family of Dunkl-Opdam operators. Instead, the operators (7.1) satisfy a weaker
property, which is still sufficient to validate the PBW theorem. This property says that if
D1 , D2 , D3 are Dunkl-Opdam operators corresponding to vector fields v1 , v2 , v3 := [v1 , v2 ]
and some choices of the functions fY , then [D1 , D2 ] D3 G O(X) (i.e., it has no poles).
To prove this property, it is sufficient to consider the case when X is a formal polydisk, with
a linear action of G. But in this case everything follows from the commutativity of the usual
Dunkl operators Dv .
Example 7.10. (1) Suppose G = 1. Then for t 6= 0, Ht,0, (G, X) = D/t (X).
(2) Suppose G is a Weyl group and X = H the corresponding torus. Then H1,c,0(G, H)
is called the trigonometric Cherednik algebra.
7.4. Globalization. Let X be any smooth algebraic variety, and G Aut(X). Assume
that X admits a cover by affine G-invariant open sets. Then the quotient variety X/G
exists.
For any affine open set U in X/G, let U be the preimage of U in X. Then we can
define the algebra Ht,c,0(G, U ) as above. If U V , we have an obvious restriction map
Ht,c,0 (G, V ) Ht,c,0 (G, U ). The gluing axiom is clearly satisfied. Thus the collection of
algebras Ht,c,0 (G, U ) can be extended (by sheafification) to a sheaf of algebras on X/G. We
are going to denote this sheaf by Ht,c,0,G,X and call it the sheaf of Cherednik algebras on
X/G. Thus, Ht,c,0,G,X (U) = Ht,c,0(G, U ).
Similarly, if H2 (X, 1 G
X ) , we can define the sheaf of twisted Cherednik algebras
Ht,c,,G,X . This is done similarly to the case of twisted differential operators (which is the
case G = 1).
Remark 7.11. (1) The construction of Ht,c, (G, X) and the PBW theorem extend in a
straightforward manner to the case when the ground field is not C but an algebraically
closed field k of positive characteristic, provided that the order of the group G is
relatively prime to the characteristic.
51
 (2) The construction and main properties of the (sheaves of) Cherednik algebras of alge-
braic varieties can be extended without significant changes to the case when X is a
complex analytic manifold, and G is not necessarily finite but acts properly discon-
tinuously. In the following lectures, we will often work in this generalized setting.
7.5. Modified Cherednik algebra. It will be convenient for us to use a slight modification
of the sheaf Ht,c,,G,X . Namely, let be a function on the set of conjugacy classes of Y such
that (Y, g) S. We define Ht,c,,,G,X in the same way as Ht,c,,G,X except that the Dunkl-
Opdam operators are defined by the formula
X 2c(Y, g) X
(7.2) D := tLv + fY (x) (g 1) + fY (x)(Y ).
1 Y,g Y
(Y,g)S

The following result shows that this modification is in fact tautological. Let Y be the class
in H2 (X, 1 1
X ) defined by the line bundle OX (Y ) , whose sections are functions vanishing
on Y .
Proposition 7.12. One has an isomorphism
Ht,c,,,G,X Ht,c,+PY (Y )Y ,G,X .
Proof. Let y Y and z be a function on the formal neighborhood of y such that z|Y = 0
and dzy 6= 0. Extend it to a system of local formal coordinates z1 = z, z2 , . . . , zd near y. A

Dunkl-Opdam operator near y for the vector field z can be written in the form
n1
1 X 2c(Y, g m ) m
D= + ( (g 1) + (Y )).
z z m=1 1 m
Y,g

Conjugating this operator by the formal expression z (Y ) := (z m )(Y )/m , we get

n1
(Y ) (Y ) 1 X 2c(Y, g m) m
z Dz = + (g 1)
z z m=1 1 m
Y,g

This implies the required statement. 

We note that the sheaf H1,c,,0,G,X localizes to G DX on the complement of all the
hypersurfaces Y . This follows from the fact that the line bundle OX (Y ) is trivial on the
complement of Y .
7.6. Orbifold Hecke algebras. Let X be a connected and simply connected complex man-
ifold, and G is a discrete group of automorphisms of X which acts properly discontinuously.
Then X/G is a complex orbifold. Let X X be the set of points with trivial stabilizer. Fix
a base point x0 X . Then the braid group of X/G is defined to be BG = 1 (X /G, x0 ).
We have an exact sequence 1 K BG G 1.
Now let S be the set of pairs (Y, g) such that Y is a component of X g of codimension 1 in
X (such Y will be called a reflection hypersurface). For (Y, g) S, let GY be the subgroup of
G whose elements act trivially on Y . This group is obviously cyclic; let nY = |GY |. Let CY
be the conjugacy class in BG corresponding to a small circle going counterclockwise around
the image of Y in X/G, and TY be a representative in CY .
The following theorem follows from elementary topology:
52
Theorem 7.13. K is defined by relations TYnY = 1, for all reflection hypersurfaces Y (i.e.,
K is the intersection of all normal subgroups of BG containing TYnY ).
Proof. See, e.g., [BMR] Proposition 2.17. 

For any conjugacy class of hypersurfaces Y such that (Y, g) S we introduce formal
parameters 1Y , . . . , nY Y . The entire collection of these parameters will be denoted by .
Let A0 = C[G].
Definition 7.14. We define the Hecke algebra of (G, X), denoted A = H (G, X, x0 ), to be
the quotient of the group algebra of the braid group, C[BG ][[ ]], by the relations
nY
Y
(7.3) (T e2ji/nY ejY ) = 0, T CY
j=1

(i.e., by the closed ideal in the formal series topology generated by these relations).
Thus, A is a deformation of A0 .
It is clear that up to an isomorphism this algebra is independent on the choice of x0 , so
we will sometimes drop x0 form the notation.
The main result of this section is the following theorem.
Theorem 7.15. Assume that H2 (X, C) = 0. Then A = H (G, X) is a flat formal defor-
mation of A0 , which means A = A0 [[ ]] as a module over C[[ ]].
Example 7.16. Let h be a finite dimensional vector space, and G be a complex reflection
group in GL(h). Then H (G, h) is the Hecke algebra of G studied in [BMR]. It follows
from Theorem 7.15 that this Hecke algebra is flat. This proof of flatness is in fact the same
as the original proof of this result given in [BMR] (based on the Dunkl-Opdam-Cherednik
operators, and explained above).
Example 7.17. Let h be a universal covering of a maximal torus of a simply connected
simple Lie group G, Q be the dual root lattice, and G b = G Q be its affine Weyl group.
Then H (h, G) b is the affine Hecke algebra. This algebra is also flat by Theorem 7.15. In
fact, its flatness is a well known result from representation theory; our proof of flatness is
essentially due to Cherednik [Ch].
Example 7.18. Let G, h, Q be as in the previous example, C+ be a complex number
bb
with a positive imaginary part, and G = G (Q Q ) be the double affine Weyl group.
bb
Then H (h, G) is (one of the versions of) the double affine Hecke algebra of Cherednik ([Ch]),
and it is flat by Theorem 7.15. The fact that this algebra is flat was proved by Cherednik,
Sahi, Noumi, Stokman (see [Ch],[Sa],[NoSt],[St]) using a different approach (q-deformed
Dunkl operators).
7.7. Hecke algebras attached to Fuchsian groups. Let H be a simply connected com-
plex Riemann surface (i.e., Riemann sphere, Euclidean plane, or Lobachevsky plane), and
be a cocompact lattice in Aut(H) (i.e., a Fuchsian group). Let = H/. Then is a com-
pact complex Riemann surface. When contains elliptic elements (i.e., nontrivial elements
53
of finite order), we are going to regard as an orbifold: it has special points Pi , i = 1, . . . , m
with stabilizers Zni . Then is the orbifold fundamental group of .1
Let g be the genus of , and al , bl , l = 1, . . . , g, be the a-cycles and b-cycles of . Let cj
be the counterclockwise loops around Pj . Then is generated by al , bl , cj with relations
n
Y
cj j = 1, c1 c2 cm = al bl a1 1
l bl .
l

For each j, introduce formal parameters kj , k = 1, . . . , nj . Define the Hecke algebra H ()

of to be generated over C[[ ]] by the same generators al , bl , cj with defining relations
nj
Y Y
(cj e2ji/nj ekj ) = 0, c1 c2 cm = al bl a1 1
l bl .
k=1 l

Thus H () is a deformation of C[].

This deformation is flat if H is a Euclidean plane or a Lobachevsky plane. Indeed, H () =
H (, H), so the result follows from Theorem 7.15 and the fact that H2 (H, C) = 0.
On the other hand, if H is the Riemann sphere (so that the condition H2 (H, C) = 0 is
violated) and 6= 1 then this deformation is not flat. Indeed, let = (~) be a 1-parameter
subdeformation of H () which is flat. Let us compute the determinant of the product
c1 cm in the regular representation of this algebra (which is finite dimensional if H is the
sphere). On the one hand, it is 1, as c1 cm is a product of commutators. On the other
hand, the eigenvalues of cj in this representation are e2ji/nj ekj with multiplicity ||/nj .
Computing determinants as products of eigenvalues, we get a nontrivial equation on kj (~),
which means that the deformation H is not flat.
Thus, we see that H () fails to be flat in the following forbidden cases:
g = 0, m = 2, (n1 , n2 ) = (n, n);
m = 3, (n1 , n2 , n3 ) = (2, 2, n), (2, 3, 3), (2, 3, 4), (2, 3, 5).
Indeed, the orbifold Euler characteristic of a closed surface of genus g with m special
points x1 , . . . , xm whose orders are n1 , . . . , nm is
Xm
orb 1
(, x1 , . . . , xm ) = 2 2g m + ,
i=1
ni
and above solutions are the solutions of the inequality
orb (CP 1, x1 , . . . , xm ) > 0.
(note that the solutions for m = 1 and solutions (n1 , n2 ) with n1 6= n2 dont arise, since they
dont correspond to any orbifolds).
1Let X be a connected topological space on with a properly discontinuous action of a discrete group G.
Then the orbifold fundamental group of the orbifold X/G with base point x X, denoted 1orb (X/G, x),
is the set of pairs (g, ), where g G and is a homotopy class of paths leading from x to gx, with
multiplication law (g1 , 1 )(g2 , 2 ) = (g1 g2 , ), where is 1 followed by g1 (2 ). Obviously, in this situation
we have an exact sequence
1 1 (X, x) 1orb (X/G, x) G 1.

54
7.8. Hecke algebras of wallpaper groups and del Pezzo surfaces. The case when H
is the Euclidean plane (i.e., is a wallpaper group) deserves special attention. If there are
elliptic elements, this reduces to the following configurations: g = 0 and
m = 3, (n1 , n2 , n3 ) = (3, 3, 3), (2, 4, 4), (2, 3, 6) (cases E6 , E7 , E8 ),
or
m = 4, (n1 , n2 , n3 , n4 ) = (2, 2, 2, 2) (case D4 ).
In these cases, the algebra H (, H) (for numerical ) has Gelfand-Kirillov dimension 2,
so it can be interpreted in terms of the theory of noncommutative surfaces.
Recall that a del Pezzo surface (or a Fano surface) is a smooth projective surface, whose
anticanonical line bundle is ample. It is known that such surfaces are CP1 CP1 , or a blow-
up of CP2 at up to 8 generic points. The degree of a del Pezzo surface X is by definition the
self intersection number K K of its canonical class K. For example, a del Pezzo surface of
degree 3 is a cubic surface in CP3 , and the degree of CP2 with n generic points blown up is
d = 9 n. P
Now suppose is numerical. Let ~ = j,k n1 j kj . Also let n be the largest of nj , and c
be the corresponding cj . Let e C[c] H (, H) be the projector to an eigenspace of c.
Consider the spherical subalgebra B (, H) := eH (, H)e.
Theorem 7.19 (Etingof, Oblomkov, Rains, [EOR]). (i) If ~ = 0 then the algebra B (, H)
is commutative, and its spectrum is an affine del Pezzo surface. More precisely, in
the case (2, 2, 2, 2), it is a del Pezzo surface of degree 3 (a cubic surface) with a tri-
angle of lines removed; in the cases (3, 3, 3), (2, 4, 4), (2, 3, 6) it is a del Pezzo surface
of degrees 3,2,1 respectively with a nodal rational curve removed.
(ii) The algebra B (, H) for ~ 6= 0 is a quantization of the unique algebraic symplectic
structure on the surface from (i) with Plancks constant ~.
Proof. See [EOR]. 
Remark 7.20. In the case (2, 2, 2, 2), H (, ) is the Cherednik-Sahi algebra of rank 1; it
controls the theory of Askey-Wilson polynomials.
Example 7.21. This is a multivariate version of the Hecke algebras attached to Fuchsian
groups, defined in the previous subsection. Namely, letting , H be as in the previous
subsection, and N 1, we consider the manifold X = H N with the action of N = SN N .
If H is a Euclidean or Lobachevsky plane, then by Theorem 7.15 H (N , X N ) is a flat
deformation of the group algebra C[N ]. If N > 1, this algebra has one more essential
parameter than for N = 1 (corresponding to reflections in SN ). In the Euclidean case, one
expects that an appropriate spherical subalgebra of this algebra is a quantization of the
Hilbert scheme of a del Pezzo surface.
7.9. The Knizhnik-Zamolodchikov functor. In this subsection we will define a global
analog of the KZ functor defined in [GGOR]. This functor will be used as a tool of proof of
Theorem 7.15.
Let X be a simply connected complex manifold, and G a discrete group of holomorphic
transformations of X acting on X properly discontinuously. Let X X be the set of points
with trivial stabilizer. Then we can define the sheaf of Cherednik algebras H1,c,,0,G,X on
X/G. Note that the restriction of this sheaf to X /G is the same as the restriction of the
55
sheaf G DX to X /G (i.e. on X /G, the dependence of the sheaf on the parameters c and
disappears). This follows from the fact that the line bundles OX (Y ) become trivial when
restricted to X .
Now let M be a module over H1,c,,0,G,X which is a locally free coherent sheaf when
restricted to X /G. Then the restriction of M to X /G is a G-equivariant D-module on
X which is coherent and locally free as an O-module. Thus, M corresponds to a locally
constant sheaf (local system) on X /G, which gives rise to a monodromy representation of
the braid group 1 (X /G, x0 ) (where x0 is a base point). This representation will be denoted
by KZ(M). This defines a functor KZ, which is analogous to the one in [GGOR].
It follows from the theory of D-modules that any OX/G -coherent H1,c,,0,G,X -module is
locally free when restricted to X /G. Thus the KZ functor acts from the abelian category
Cc, of OX/G -coherent H1,c,,0,G,X -modules to the category of representations of 1 (X /G, x0 ).
It is easy to see that this functor is exact.
For any Y , let gY be the generator of GY which has eigenvalue e2i/nY in the conormal
bundle to Y . Let (c, ) (c, ) be the invertible linear transformation defined by the
formula
nXY 1
1 e2jmi/nY
jY = 2i(2 c(Y, gYm) (Y ))/nY .
m=1
1 e2mi/nY

Proposition 7.22. The functor KZ maps the category Cc, to the category of representations
of the algebra H (c,) (G, X).
Proof. The result follows from the corresponding result in the linear case (which we have
already proved) by restricting M to the union of G-translates of a neighborhood of a generic
point y Y , and then linearizing the action of GY on this neighborhood. 

7.10. Proof of Theorem 7.15. Consider the module M = IndGD DX

X
OX . Then KZ(M)
is the regular representation of G which is denoted by regG. We want to show that M
deforms uniquely (up to an isomorphism) to a module over H1,c,0,,G,X for formal c, . The
obstruction to this deformation is in Ext2GDX (M, M) and the freedom of this deformation
is in Ext1GDX (M, M). Since
ExtiGDX (M, M) = ExtiDX (OX , ResM) = ExtiDX (OX , OX CG)

= ExtiDX (OX , OX ) CG = Hi (X, C) CG,

and X is simply connected, we have
Ext1GDX (M, M) = 0, and Ext2GDX (M, M) = 0 if H2 (X, C) = 0.
Thus such deformation exists and is unique if H2 (X, C) = 0.
Now let Mc, be the deformation. Then KZ(Mc, ) is a H (c,) (G, X)-module from Propo-
sition 7.22 and it deforms flatly the module regG. This implies H (c,) (G, X) is flat over
C[[ ]].
Remark 7.23. When X is not simply connected, the theorem is still true under the as-
e C) = 0, where X
sumption 2 (X) C = 0 (i.e. H2 (X, e is the universal cover of X), and the
proof is contained in [E1].
56
7.11. Example: the simplest case of double affine Hecke algebras. Now let G =
Z2 Z2 acting on C. Then the conjugacy classes of reflection hyperplanes are four points:
0, 1/2, 1/2 + /2, /2, where we suppose the lattice in C is Z Z. Correspondingly, the
presentation of G is as follows:
generators: T1 , T2 , T3 , T4 ; relations: T1 T2 T3 T4 = 1, Ti2 = 1.
Thus, the corresponding orbifold Hecke algebra is the following deformation of CG:
generators: T1 , T2 , T3 , T4 ; relations: T1 T2 T3 T4 = 1, (Ti pi )(Ti qi ) = 0,
where pi , qi (i = 1, . . . , 4), are parameters.
If we renormalize the Ti , these relations turn into
(Ti ti )(Ti + t1
i ) = 0, T1 T2 T3 T4 = q,
and we get the type C C1 double affine Hecke algebra. If we set three of the four Ti s
satisfying the undeformed relation Ti2 = 1, we get the double affine Hecke algebra of type
A1 . More precisely, this algebra is generated by T1 , . . . , T4 with relations
T22 = T32 = T42 = 1, (T1 t)(T1 + t1 ) = 0, T1 T2 T3 T4 = q.
Another presentation of this algebra (which is more widely used) is as follows. Let E =
C/Z2 , an elliptic curve with an Z2 action defined by z 7 z. Define the partial braid group
B = 1orb (E\{0}/Z2 , x),
where x is a generic point. Notice that comparing to the usual braid group, we do not
delete three of the four reflection points. The generators of the group 1 (E \ {0}, x) (the
fundamental group of a punctured 2-torus) are X (corresponding to the a-cycle on the torus),
Y (corresponding to the b-cycle on the torus) and C (corresponding to the loop around 0). In
order to construct B, which is an extension of Z2 by 1 (E \ {0}, x), we introduce an element
T s.t. T 2 = C (the half-loop around the puncture). Then X, Y, T satisfy the following
relations:
T XT = X 1 , T 1 Y T 1 = Y 1 , Y 1 X 1 Y XT 2 = 1.
The Hecke algebra of the partial braid group is then defined to be the group algebra of B
plus an extra relation: (T q1 )(T + q2 ) = 0.
A common way to present this Hecke algebra is to renormalize the generators so that one
has the following relations:
T XT = X 1 , T 1Y T 1 = Y 1 , Y 1 X 1 Y XT 2 = q, (T t)(T + t1 ) = 0.
This is Cheredniks definition for HH(q, t), the double affine Hecke algebra of type A1 which
depends on two parameters q, t.
There are two degenerations of the algebra HH(q, t).
1. The trigonometric degeneration.
Set Y = e~y , q = e~ , t = e~c and T = se~cs , where s Z2 is the reflection. Then s, X, y
satisfy the following relations modulo ~:
s2 = 1, sXs1 = X 1 , sy + ys = 2c, X 1 yX y = 1 2cs.
The algebra generated by s, X, y with these relations is called the type A1 trigonomet-
ric Cherednik algebra. It is easy to show that it is isomorphic to the Cherednik algebra
H1,c (Z2 , C ), where Z2 acts on C by z z 1 .
57
 2. The rational degeneration.
In the trigonometric Cherednik algebra, set X = e~x and y = y/~. Then s, x, y satisfy the
following relations modulo ~:
s2 = 1, sx = xs, sy = ys, yx xy = 1 2cs.
The algebra generated by s, x, y with these relations is the rational Cherednik algebra
H1,c (Z2 , C) with the action of Z2 on C is given by z z.

7.12. Affine and extended affine Weyl groups. Let R = {} Rn be a root system
with respect to a nondegenerate symmetric bilinear form (, ) on Rn . We will assume that
R is reduced. Let {i }ni=1 R be the set of simple roots and R+ (respectively R ) be the

set of positive
Ln (respectively negative) roots. The corootsLn are denoted by = 2/(, ).

Let Q = i=1 Zi be the coroot lattice and P = i=1 Zi the coweight lattice, where
i s are the fundamental coweights, i.e., (i, j ) = ij . Let be the maximal positive root,
and assume that the bilinear form is normalized by the condition (, ) = 2. Let W be the
Weyl group which is generated by the reflections s ( R).
By definition, the affine root system is
Ra = { = [, j] Rn R| where R, j Z}.
a
The set of positive affine roots is R+ = {[, j] | j Z>0 } {[, 0] | R+ }. Define 0 =
[, 1]. We will identify R with = [, 0] Ra .
For an arbitrary affine root = [, j] and a vector z = [z, ] Rn R, the corresponding
affine reflection is defined as follows:
(z, )
s (z) = z 2 = z (z, ) .
(, )
e+ }, and we have
The affine Weyl group W a is generated by the affine reflections {s | R
an isomorphism:
Wa = W Q ,
where the translation Q is naturally identified with the composition s[,1] s W a .
ext
Define the extended affine Weyl group to be W a = W P acting on Rn+1 via b(z) =
ext
[z, (b, z)] for z = [z, ], b P . Then W a W a . Moreover, W a is a normal subgroup of
ext ext
W a and W a /W a = P /Q . The latter group can be identified with the group = {r } of
ext
the elements of W a permuting simple affine roots under their action in Rn+1 . It is a normal
commutative subgroup of Aut = Aut(Dyna ) (Dyna denotes the affine Dynkin diagram).
The quotient Aut/ is isomorphic to the group of the automorphisms preserving 0 , i.e. the
group AutDyn of automorphisms of the finite Dynkin diagram.

7.13. Cheredniks double affine Hecke algebra of a root system. In this subsection,
we will give an explicit presentation of Cheredniks DAHA for a root system, defined in
Example 7.18. This is done by giving an explicit presentation of the corresponding braid
group (which is called the elliptic braid group), and then imposing quadratic relations on the
generators corresponding to reflections.
For a root system R, let m = 2 if R is of type D2k , m = 1 if R is of type B2k , Ck , and
otherwise m = ||. Let mij be the number of edges between vertex i and vertex j in the
58
affine Dynkin diagram of Ra . Let Xi (i = 1, . . . , n) be a family of pairwise commutative and
algebraically independent elements. Set
n
Y Xn
X[b,j] = Xii q j , where b = i i P, j Z/mZ.
i=1 i=1
ext
For an element w Wa , we can define an action on these X[b,j] by wX[b,j] = Xw[b,j].
Definition 7.24 (Cherednik). The double affine Hecke algebra (DAHA) of the root system
R, denoted by HH, is an algebra defined over the field Cq,t = C(q 1/m , t1/2 ), generated by
Ti , i = 0, . . . , n, , Xb , b P , subject to the following relations:
(1) Ti Tj Ti = Tj Ti Tj , mij factors each side;
(2) (Ti ti )(Ti + t1 i ) = 0 for i = 0, . . . , n;
(3) Ti 1 = T(i) , for and i = 0, . . . , n;
(4) Xb 1 = X(b) , for , b P ;
(5) Ti Xb Ti = Xb X1 i
, if i > 0 and (b, i ) = 1; Ti Xb = Xb Ti , if i > 0 and (b, i ) = 0;
(6) T0 Xb T0 = Xb0 if (b, ) = 1; T0 Xb = Xb T0 if (b, ) = 0.
Here ti are parameters attached to simple affine roots (so that roots of the same length
give rise to the same parameters).
The degenerate double affine Hecke algebra (trigonometric Cherednik algebra) HHtrig is
ext P
generated by the group algebra of W a , and pairwise commutative yb = ni=1 (b, )yi + u
for b = [b, u] P Z, with the following relations:
si yb ysi (b) si = ki (b, i ), for i = 1, . . . , n,
s0 yb ys0 (b) s0 = k0 (b, ), r yb r1 = yr (b) for r .
Remark 7.25. This degeneration can be obtained from the DAHA similarly to the case of
A1 , which is described above.
7.14. Algebraic flatness of Hecke algebras of polygonal Fuchsian groups. Let W
be the Coxeter group of rank r corresponding to a Coxeter datum:
mij (i, j = 1, . . . , r, i 6= j), such that 2 mij and mij = mji .
So the group W has generators si i = 1, . . . , r, and defining relations
s2i = 1, (si sj )mij = 1 if mij 6= .
It has a sign character : W {1} given by (si ) = 1. Denote by W+ the kernel of
(the even subgroup of W ). It is generated by aij = si sj with relations:
m
aij = a1
ji , aij ajk aki = 1, aij ij = 1.
We can deform the group algebra C[W ] as follows. Define the algebra A(W ) with invertible
generators si , and tij,k , i, j = 1, . . . , r, k Zmij for (i, j) such that mij < and defining
relations
tij,k = t1
ji,k , s2i = 1, [tij,k , ti j ,k ] = 0, sp tij,k = tji,k sp ,
mij
Y
(si sj tij,k ) = 0 if mij < .
k=1
Notice that if we set tij,k = exp(2ki/mij ), we get C[W ].
59
 Define also the algebra A+ (W ) over R := C[tij,k ] (tij,k = t1
ji,k ) by generators aij , i 6= j
1
(aij = aji ), and relations
mij
Y
(aij tij,k ) = 0 if mij < , aij ajp api = 1.
k=1

If w is a word in letters si , let Tw be the corresponding element of A(W ). Choose a

function w(x) which attaches to every element x W , a reduced word w(x) representing x
in W .
Theorem 7.26 (Etingof, Rains, [ER]). (i) The elements Tw(x) , x W , form a spanning
set in A(W ) as a left R-module.
(ii) The elements Tw(x) , x W+ , form a spanning set in A+ (W ) as a left R-module.
(iii) The elements Tw(x) , x W , are linearly independent if W has no finite parabolic
subgroups of rank 3.
Proof. We only give the proof of (i). Statement (ii) follows from (i). Proof of (iii), which is
quite nontrivial, can be found in [ER] (it uses the geometry of constructible sheaves on the
Coxeter complex of W ).
Let us write the relation
mij
Y
(si sj tij,k ) = 0
k=1
as a deformed braid relation:
sj si sj . . . + S.L.T. = tij si sj si . . . + S.L.T.,
where tij = (1)mij +1 tij,1 tij,mij , S.L.T. mean smaller length terms, and the products
on both sides have length mij . This can be done by multiplying the relation by si sj (mij
factors).
Now let us show that Tw(x) span A(W ) over R. Clearly, Tw for all words w span A(W ).
So we just need to take any word w and express Tw via Tw(x) .
It is well known from the theory of Coxeter groups (see e.g. [B]) that using the braid
relations, one can turn any non-reduced word into a word that is not square free, and any
reduced expression of a given element of W into any other reduced expression of the same
element. Thus, if w is non-reduced, then by using the deformed braid relations we can reduce
Tw to a linear combination of Tu with words u of smaller length than w. On the other hand,
if w is a reduced expression for some element x W , then using the deformed braid relations
we can reduce Tw to a linear combination of Tu with u shorter than w, and Tw(x) . Thus Tw(x)
are a spanning set. This proves (i). 
Thus, A+ (W ) is a deformation of C[W+ ] over R, and similarly A(W ) is a twisted
deformation of C[W ].
Now let = (m1 , . . . , mr ), r 3, be the Fuchsian group defined by generators cj ,
j = 1, . . . , r, with defining relations
r
Y
mj
cj = 1, cj = 1.
j=1

Here 2 mj < .
60
 Suppose acts on H where H is a simply connected complex Riemann surface as in
Section 7.7. We have the Hecke algebra of , H (H, ), defined by the same (invertible)
generators cj and relations
Y r
Y
(cj exp(2ik/nj )qjk ) = 0, cj = 1,
k j=1

where qjk = exp(jk ).

We saw above (Theorem 7.15) that if jk s are formal, the algebra H (, H) is flat in if
|| is infinite (i.e., H is Euclidean or hyperbolic). Here is a much stronger non-formal version
of this theorem.
1
Theorem
P 7.27. The algebra H (, H) is free as a left module over R := C[qjk ] if and only
if j (1 1/mj ) 2 (i.e., H is Euclidean or hyperbolic).
Proof. Let us consider the Coxeter datum: mij , i, j = 1, . . . , r, such that mi,i+1 := mi
(i Z/rZ), and mij = otherwise. Suppose the corresponding Coxeter group is W . Then
we can see that = W+ . Notice that the algebra H (, H) for genus 0 orbifolds is the
algebra A+ (W ), i.e.,
P we have H (, H) = A+ (W ).
The condition j (1 1/mj ) 2 is equivalent to the condition that W has no finite
parabolic subgroups of rank 3. From Theorem 7.26 (ii) and Theorem 7.15, we can see that
A+ (W ) is free as a left module over R. We are done. 
7.15. Notes. Section 7.8 follows Section 6 of the paper [EOR]; Cheredniks definition of the
double affine Hecke algebra of a root system is from Cheredniks book [Ch]; Sections 7.7 and
7.14 follow the paper [ER]; The other parts of this section follow the paper [E1].

61
 8. Symplectic reflection algebras
8.1. The definition of symplectic reflection algebras. Rational Cherednik algebras for
finite Coxeter groups are a special case of a wider class of algebras called symplectic reflection
algebras. To define them, let V be a finite dimensional symplectic vector space over C with
a symplectic form , and G be a finite group acting symplectically (linearly) on V . For
simplicity let us assume that (2 V )G = C (i.e., V is symplectically irreducible) and that
G acts faithfully on V (these assumptions are not important, and essentially not restrictive).
Definition 8.1. A symplectic reflection in G is an element g such that the rank of the
operator 1 g on V is 2.
If s is a symplectic reflection, then let s (x, y) be the form applied to the projections of
x, y to the image of 1 s along the kernel of 1 s; thus s is a skewsymmetric form of rank
2 on V .
Let S G be the set of symplectic reflections, and c : S C be a function which is
invariant under the action of G. Let t C.
Definition 8.2. The symplectic reflection algebra Ht,c = Ht,c [G, V ] is the quotient of the
algebra C[G] T(V ) by the ideal generated by the relation
X
(8.1) [x, y] = t(x, y) 2 cs s (x, y)s.
sS

Example 8.3. Let W be a finite Coxeter group with reflection representation h. We can set
V = h h , (x, x ) = (y, y ) = 0, (y, x) = (y, x), for x, x h and y, y h. In this case
(1) symplectic reflections are the usual reflections in W ;
(2) s (x, x ) = s (y, y ) = 0, s (y, x) = (y, s )(s , x)/2.
Thus, Ht,c [G, h h ] coincides with the rational Cherednik algebra Ht,c (G, h) defined in
Section 3.
Example 8.4. Let be a finite subgroup of SL2 (C), and V = C2 be the tautological
representation, with its standard symplectic form. Then all nontrivial elements of are
symplectic reflections, and for any symplectic reflection s, s = . So the main commutation
relation of Ht,c [, V ] takes the form
X
[y, x] = t 2cg g.
g,g6=1

Example 8.5. (Wreath products) Let be as in the previous example, G = Sn n , and

V = (C2 )n . Then symplectic reflections are conjugates (g, 1, ..., 1), g , g 6= 1, and also
conmjugates of transpositions in Sn (so there is one more conjugacy class of reflections than
in the previous example).
Note also that for any V, G, H0,0 [G, V ] = G SV , and H1,0 [G, V ] = G Weyl(V ), where
Weyl(V ) is the Weyl algebra of V , i.e. the quotient of the tensor algebra T(V ) by the
relation xy yx = (x, y), x, y V .
8.2. The PBW theorem for symplectic reflection algebras. To ensure that the sym-
plectic reflection algebras Ht,c have good properties, we need to prove a PBW theorem for
them, which is an analog of Proposition 3.5. This is done in the following theorem, which
also explains the special role played by symplectic reflections.
62
Theorem 8.6. Let : 2 V C[G] be a linear G-equivariant function. Define the algebra
H to be the quotient of the algebra C[G]T(V ) by the relation [x, y] = (x, y), x, y V . Put
an increasing filtration on H by setting deg(V ) = 1, deg(G) = 0, and define : CG SV
grH to be the natural surjective homomorphism. Then is an isomorphism if and only if
has the form X
(x, y) = t(x, y) 2 cs s (x, y)s,
sS
for some t C and G-invariant function c : S C.
Unfortunately, for a general symplectic reflection algebra we dont have a Dunkl operator
representation, so the proof of the more difficult if part of this Theorem is not as easy
as the proof of Proposition 3.5. Instead of explicit computations with Dunkl operators, it
makes use of the deformation theory of Koszul algebras, which we will now discuss.
8.3. Koszul algebras. Let R be a finite dimensional semisimple algebra (over C). Let A
be a Z+ -graded algebra, such that A[0] = R, and assume that the graded components of A
are finite dimensional.
Definition 8.7. (i) The algebra A is said to be quadratic if it is generated over R by
A[1], and has defining relations in degree 2.
(ii) A is Koszul if all elements of Exti (R, R) (where R is the augmentation module over
A) have grade degree precisely i.
Remark 8.8. (1) Thus, in a quadratic algebra, A[2] = A[1] R A[1]/E, where E is the
subspace (R-subbimodule) of relations.
(2) It is easy to show that a Koszul algebra is quadratic, since the condition to be
quadratic is just the Koszulity condition for i = 1, 2.
Now let A0 be a quadratic algebra, A0 [0] = R. Let E0 be the space of relations for A0 . Let
E A0 [1] R A0 [1][[~]] be a free (over C[[~]]) R-subbimodule which reduces to E0 modulo ~
(deformation of the relations). Let A be the (~-adically complete) algebra generated over
R[[~]] by A[1] = A0 [1][[~]] with the space of defining relations E. Thus A is a Z+ -graded
algebra.
The following very important theorem is due to Beilinson, Ginzburg, and Soergel, [BGS]
(less general versions appeared earlier in the works of Drinfeld [Dr], Polishchuk-Positselski
[PP], Braverman-Gaitsgory [BG]). We will not give its proof.
Theorem 8.9 (Koszul deformation principle). If A0 is Koszul then A is a topologically free
C[[~]] module if and only if so is A[3].
Remark. Note that A[i] for i < 3 are obviously topologically free.
We will now apply this theorem to the proof of Theorem 8.6.
2
Pof Theorem 8.6. Let : V 2 C[G]
8.4. Proof be a linear G-equivariant map. We write
(x, y) = gG g (x, y)g, where g (x, y) V . To apply Theorem 8.9, let us homogenize
our algebras. Namely, let A0 = (CG SV ) C[u] (where u has degree 1). Also let ~ be a
formal parameter, and consider the deformation A = H~u2 of A0 . That is, A is the quotient
of G T(V )[u][[~]] by the relations [x, y] = ~u2 (x, y). This is a deformation of the type
considered in Theorem 8.9, and it is easy to see that its flatness in ~ is equivalent to Theorem
63
8.6. Also, the algebra A0 is Koszul, because the polynomial algebra SV is a Koszul algebra.
Thus by Theorem 8.9, it suffices to show that A is flat in degree 3.
The flatness condition in degree 3 is the Jacobi identity
[(x, y), z] + [(y, z), x] + [(z, x), y] = 0,
which must be satisfied in CG V . In components, this equation transforms into the system
of equations
g (x, y)(z z g ) + g (y, z)(x xg ) + g (z, x)(y y g ) = 0
for every g G (here z g denotes the result of the action of g on z).
This equation, in particular, implies that if x, y, g are such that g (x, y) 6= 0 then for any
z V z z g is a linear combination of x xg and y y g . Thus g (x, y) is identically zero
unless the rank of (1 g)|V is at most 2, i.e. g = 1 or g is a symplectic reflection.
If g = 1 then g (x, y) has to be G-invariant, so it must be of the form t(x, y), where
t C.
If g is a symplectic reflection, then g (x, y) must be zero for any x such that x xg = 0.
Indeed, if for such an x there had existed y with g (x, y) 6= 0 then z z g for any z would
be a multiple of y y g , which is impossible since Im(1 g)|V is 2-dimensional. This implies
that g (x, y) = 2cg g (x, y), and cg must be invariant.
Thus we have shown that if A is flat (in degree 3) then must have the form given in
Theorem 8.6. Conversely, it is easy to see that if does have such form, then the Jacobi
identity holds. So Theorem 8.6 is proved.
8.5. The spherical subalgebra of the symplectic reflection algebra. The properties of
symplectic reflection algebras are similar to the properties of rational Cherednik algebras we
have studied before. The main difference is that we no longer have the Dunkl representation
and localization results, so some proofs are based on different ideas and are more complicated.
The spherical subalgebra of the symplectic reflection P algebra is defined in the same way
as in the Cherednik algebra case. Namely, let e = |G|1 gG g, and Bt,c = eHt,c e.
Proposition 8.10. Bt,c is commutative if and only if t = 0.
Proof. Let A be a Z+ -filtered algebra. If A is not commutative, then we can define a nonzero
Poisson bracket on grA in the following way. Let m be the minimum of deg(a) + deg(b)
deg([a, b]) (over a, b A such that [a, b] 6= 0). Then for homogeneous elements a0 , b0 A0 of
degrees p, q, we can define {a0 , b0 } to be the image in A0 [p + q m] of [a, b], where a, b are
any lifts of a0 , b0 to A. It is easy to check that { , } is a Poisson bracket on A0 of degree
m.
Let us now apply this construction to the filtered algebra A = Bt,c . We have gr(A) =
A0 = (SV )G .
Lemma 8.11. A0 has a unique, up to scaling, Poisson bracket of degree 2, and no nonzero
Poisson brackets of degrees < 2.
Proof. A Poisson bracket on (SV )G is the same thing as a Poisson bracket on the variety
V /G. On the smooth part (V /G)s of V /G, it is simply a bivector field, and we can lift
it to a bivector field on the preimage Vs of (V /G)s in V , which is the set of points in V
with trivial stabilizers. But the codimension on V \ Vs in V is 2 (as V \ Vs is a union
of symplectic subspaces), so the bivector on Vs extends to a regular bivector on V . So if
64
this bivector is homogeneous, it must have degree 2, and if it has degree 2 then it
must be with constant coefficients, so being G-invariant, it is a multiple of . The lemma is
proved. 
Now, for each t, c we have a natural Poisson bracket on A0 of degree 2, which depends
linearly on t, c. So by the lemma, this bracket has to be of the form f (t, c), where is the
unique up to scaling Poisson bracket of degree 2, and f a homogeneous linear function.
Thus the algebra A = Bt,c is not commutative unless f (t, c) = 0. On the other hand, if
f (t, c) = 0, and Bt,c is not commutative, then, as weve shown, A0 has a nonzero Poisson
bracket of degree < 2. But By Lemma 8.11, there is no such brackets. So Bt,c must be
commutative if f (t, c) = 0.
It remains to show that f (t, c) is in fact a nonzero multiple of t. First note that f (1, 0) 6= 0,
since B1,0 is definitely noncommutative. Next, let us take a point (t, c) such that Bt,c is
commutative. Look at the Ht,c -module Ht,c e, which has a commuting action of Bt,c from the
right. Its associated graded is SV as an (CG SV, (SV )G )-bimodule, which implies that
the generic fiber of Ht,c e as a Bt,c -module is the regular representation of G. So we have a
family of finite dimensional representations of Ht,c on the fibers of Ht,c e, all realized in the
regular representation of G. Computing the trace of the main commutation relation (8.1) of
Ht,c in this representation, we obtain that t = 0 (since Tr (s) = 0 for any reflection s). The
proposition is proved. 
Note that B0,c has no zero divisors, since its associated graded algebra (SV )G does not.
Thus, like in the Cherednik algebra case, we can define a Poisson variety Mc , the spectrum
of B0,c , called the Calogero-Moser space of G, V . Moreover, the algebra Bc := B~,c over C[~]
is an algebraic quantization of Mc .
8.6. The center of the symplectic reflection algebra Ht,c . Consider the bimodule Ht,c e,
which has a left action of Ht,c and a right commuting action of Bt,c . It is obvious that
EndHt,c Ht,c e = Bt,c (with opposite product). The following theorem shows that the bimodule
Ht,c e has the double centralizer property (i.e., EndBt,c Ht,c e = Ht,c ).
Note that we have a natural map t,c : Ht,c EndBt,c Ht,c e.
Theorem 8.12. t,c is an isomorphism for any t, c.
Proof. The complete proof is given [EG]. We will give the main ideas of the proof skipping
straightforward technical details. The first step is to show that the result is true in the
graded case, (t, c) = (0, 0). To do so, note the following easy lemma:
Lemma 8.13. If X is an affine complex algebraic variety with algebra of functions OX and
G a finite group acting freely on X then the natural map X : G OX EndOXG OX is an
isomorphism.
Therefore, the map 0,0 : G SV End(SV )G (SV ) is injective, and moreover becomes an
isomorphism after localization to the field of quotients C(V )G . To show its surjective, take
a End(SV )G (SV ). There exists a G C(V ) which maps to a. Moreover, by Lemma
8.13, a can have poles only at fixed points of G on V . But these fixed points form a subset
of codimension 2, so there can be no poles and we are done in the case (t, c) = (0, 0).
Now note that the algebra EndBt,c Ht,c e has an increasing integer filtration (bounded be-
low) induced by the filtration on Ht,c . This is due to the fact that Ht,c e is a finitely generated
65
eHt,c e-module (since it is true in the associated graded situation, by Hilberts theorem about
invariants). Also, the natural map grEndBt,c Ht,c e EndgrBt,c grHt,c e is clearly injective.
Therefore, our result in the case (t, c) = (0, 0) implies that this map is actually an iso-
morphism (as so is its composition with the associated graded of t,c ). Identifying the two
algebras by this isomorphism, we find that gr(t,c ) = 0,0 . Since 0,0 is an isomorphism, t,c
is an isomorphism for all t, c, as desired. 2 
Denote by Zt,c the center of the symplectic reflection algebra Ht,c . We have the following
theorem.
Theorem 8.14. If t 6= 0, the center of Ht,c is trivial. If t = 0, we have grZ0,c = Z0,0 . In
particular, H0,c is finitely generated over its center.
Proof. The t 6= 0 case was proved by Brown and Gordon [BGo] as follows. If t 6= 0, we have
grZt,c Z0,0 = (SV )G . Also, we have a map
t,c : Zt,c Bt,c = eHt,c e, z 7 ze = eze.
The map t,c is injective since gr(t,c ) is injective. In particular, the image of gr(t,c ) is
contained in Z(Bt,c ), the center of Bt,c . Thus it is enough to show that Z(Bt,c ) is trivial. To
show this, note that grZ(Bt,c ) is contained in the Poisson center of B0,0 which is trivial. So
Z(Bt,c ) is trivial.
Now suppose t = 0. We need to show that gr(0,c ) : gr(Z0,c ) Z0,0 is an isomorphism. It
1
suffices to show that 0,c is an isomorphism. To show this, we construct 0,c : B0,c Z0,c as
follows.
For any b B0,c , since B0,c is commutative, we have an element b EndB0,c (H0,c e) which
is defined as the right multiplication by b. From Theorem 8.12, b H0,c . Moreover, b Z0,c
since it commutes with H0,c as a linear operator on the faithful H0,c -module H0,c e. So b Z0,c .
1
It is easy to see that be = b. So we can set b = 0,c (b) which defines the inverse map to
0,c . 

8.7. A review of deformation theory. Now we would like to explain that symplectic
reflection algebras are the most general deformations of algebras of the from G Weyl(V ).
Before we do so, we give a brief review of classical deformation theory of associative algebras.

8.7.1. Formal deformations of associative algebras. Let A0 be an associative algebra with

unit over C. Denote by 0 the multiplication in A0 .
Definition 8.15. A (flat) formal n-parameter deformation of A0 is an algebra A over
C[[~]] = C[[~1 , . . . , ~n ]] which is topologically free as a C[[~]]-module, together with an
algebra isomorphism 0 : A/mA A0 where m = h~1 , . . . , ~n i is the maximal ideal in C[[~]].
When no confusion is possible, we will call A a deformation of A0 (omitting formal).
Let us restrict ourselves to one-parameter deformations with parameter ~. Let us choose
an identification : A A0 [[~]] as C[[~]]-modules, such that = 0 modulo ~. Then the
2Here we use the fact that the filtration is bounded from below. In the case of an unbounded filtration,
it is possible for a map not to be an isomorphism if its associated graded is an isomorphism. An example of
this is the operator of multiplication by 1 + t1 in the space of Laurent polynomials in t, filtered by degree.
66
product in A is completely determined by the product of elements of A0 , which has the form
of a star-product
(a, b) = a b = 0 (a, b) + ~1 (a, b) + ~2 2 (a, b) + ,
where i : A0 A0 A0 are linear maps, and 0 (a, b) = ab.

8.7.2. Hochschild cohomology. The main tool in deformation theory of associative algebras
is Hochschild cohomology. Let us recall its definition.
Let A be an associative algebra. Let M be a bimodule over A. A Hochschild n-cochain of
A with coefficients in M is a linear map An M. The space of such cochains is denoted
by C n (A, M). The differential d : C n (A, M) C n+1 (A, M) is defined by the formula
df (a1 , . . . , an+1 ) = f (a1 , . . . , an )an+1 f (a1 , . . . , an an+1 ) + f (a1 , . . . , an1 an , an+1 )
+ (1)n f (a1 a2 , . . . , an+1 ) + (1)n+1 a1 f (a2 , . . . , an+1 ).
It is easy to show that d2 = 0.
Definition 8.16. The Hochschild cohomology HH (A, M) is defined to be the cohomology
of the complex (C (A, M), d).
Proposition 8.17. One has a natural isomorphism
HHi (A, M) ExtiAbimod (A, M),
where Abimod denotes the category of A-bimodules.
Proof. The proof is obtained immediately by considering the bar resolution of the bimodule
A:
A A A A A A,
where the bimodule structure on An is given by
b(a1 a2 an )c = ba1 a2 an c,
and the map n : An An1 is given by the formula
n (a1 a2 ... an ) = a1 a2 an + (1)n a1 an1 an .


Note that we have the associative Yoneda product

HHi (A, M) HHj (A, N) HHi+j (A, M A N),
induced by tensoring of cochains.
If M = A, the algebra itself, then we will denote HH (A, M) by HH (A). For example,
HH0 (A) is the center of A, and HH1 (A) is the quotient of the Lie algebra of derivations of A
by inner derivations. The Yoneda product induces a graded algebra structure on HH (A); it
can be shown that this algebra is supercommutative.
67
8.7.3. Hochschild cohomology and deformations. Let A0 be an algebra, and let us look for
1-parameter deformations A = A0 [[~]] of A0 . Thus, we look for such series which satisfy
the associativity equation, modulo the automorphisms of the C[[~]]-module A0 [[~]] which are
the identity modulo ~. 3
The associativity equation ( Id) = (Id ) reduces to a hierarchy of linear
equations:
XN N
X
s (N s (a, b), c) = s (a, N s (b, c)).
s=0 s=0
(These equations are linear in N if i , i < N, are known).
To study these equations, one can use Hochschild cohomology. Namely, we have the
following standard facts (due to Gerstenhaber, [Ge]), which can be checked directly.
(1) The linear equation for 1 says that 1 is a Hochschild 2-cocycle. Thus algebra struc-
tures on A0 [~]/~2 deforming 0 are parametrized by the space Z 2 (A0 ) of Hochschild
2-cocycles of A0 with values in M = A0 .
(2) If 1 , 1 are two 2-cocycles such that 1 1 is a coboundary, then the algebra struc-
tures on A0 [~]/~2 corresponding to 1 and 1 are equivalent by a transformation of
A0 [~]/~2 that equals the identity modulo ~, and vice versa. Thus equivalence classes
of multiplications on A0 [~]/~2 deforming 0 are parametrized by the cohomology
HH2 (A0 ).
(3) The linear equation for N says that dN is a certain quadratic expression bN in
1 , . . . , N 1 . This expression is always a Hochschild 3-cocycle, and the equation is
solvable if and only if it is a coboundary. Thus the cohomology class of bN in HH3 (A0 )
is the only obstruction to solving this equation.
8.7.4. Universal deformation. In particular, if HH3 (A0 ) = 0 then the equation for n can be
solved for all n, and for each n the freedom in choosing the solution, modulo equivalences,
is the space H = HH2 (A0 ). Thus there exists an algebra structure over C[[H]] on the space
Au := A0 [[H]] of formal functions from H to A0 , a, b 7 u (a, b) A0 [[H]], (a, b A0 ), such
that u (a, b)(0) = ab A0 , and every 1-parameter flat formal deformation A of A0 is given
by the formula (a, b)(~) = u (a, b)((~)) for a unique formal series ~H[[~]], with the
property that (0) is the cohomology class of the cocycle 1 .
Such an algebra Au is called a universal deformation of A0 . It is unique up to an isomor-
phism (which may involve an automorphism of C[[H]]). 4
Thus in the case HH3 (A0 ) = 0, deformation theory allows us to completely classify 1-
parameter flat formal deformations of A0 . In particular, we see that the moduli space
parametrizing formal deformations of A0 is a smooth space it is the formal neighborhood
of zero in H.
If HH3 (A0 ) is nonzero then in general the universal deformation parametrized by H does
not exist, as there are obstructions to deformations. In this case, the moduli space of
3Note that we dont have to worry about the existence of a unit in A since a formal deformation of an
algebra with unit always has a unit.
4In spite of the universal property of A , it may happen that there is an isomorphism between the algebras
u
A and A2 corresponding to different paths 1 (~), 2 (~) (of course, reducing to a nontrivial automorphism of
1

A0 modulo ~). For this reason, sometimes Au is called a semiuniversal, rather than universal, deformation
of A0 .
68
deformations will be a closed subscheme of the formal neighborhood of zero in H, which
is often singular. On the other hand, even when HH3 (A0 ) 6= 0, the universal deformation
parametrized by (the formal neighborhood of zero in) H may exist (although its existence
may be more difficult to prove than in the vanishing case). In this case one says that the
deformations of A0 are unobstructed (since all obstructions vanish even though the space of
obstructions doesnt).
8.8. Deformation-theoretic interpretation of symplectic reflection algebras. Let V
be a symplectic vector space (over C) and Weyl(V ) the Weyl algebra of V . Let G be a finite
group acting symplectically on V . Then from the definition, we have
A0 := H1,0 [G, V ] = G Weyl(V ).
Let us calculate the Hochschild cohomology of this algebra.
Theorem 8.18 (Alev, Farinati, Lambre, Solotar, [AFLS]). The cohomology space
HHi (G Weyl(V )) is naturally isomorphic to the space of conjugation invariant functions
on the set Si of elements g G such that rank (1 g)|V = i.
Corollary 8.19. The odd cohomology of G Weyl(V ) vanishes, and HH2 (G Weyl(V ))
is the space C[S]G of conjugation invariant functions on the set of symplectic reflections. In
particular, there exists a universal deformation A of A0 = G Weyl(V ) parametrized by
C[S]G .
Proof. Directly from the theorem. 
Proof of Theorem 8.18.
Lemma 8.20. Let B be a C-algebra together with an action of a finite group G. Then
M
HH (G B, G B) = ( HH (B, Bg))G ,
gG

where Bg is the bimodule isomorphic to B as a space where the left action of B is the usual
one and the right action is the usual action twisted by g.
Proof. The algebra G B is a projective B-module. Therefore, using the Shapiro lemma,
we get
HH (G B, G B) = Ext(GG)(BBop ) (G B, G B)
= ExtGdiagonal (BBop ) (B, G B) = ExtBBop (B, G B)G
M M
= ( ExtBBop (B, Bg))G = ( HH (B, Bg))G ,
gG gG

as desired. 
Now apply the lemma to B = Weyl(V ). For this we need to calculate HH (B, Bg),
where g is any element of G. We may assume that g is diagonal in some symplectic basis:
g = diag(1 , 1 1
1 , . . . , n , n ). Then by the Kunneth formula we find that
n
O

HH (B, Bg) = HH (A1 , A1 gi ),
i=1
69
where A1 is the Weyl algebra of the 2-dimensional space, (generated by x, y with defining
relation xy yx = 1), and gi = diag(i , 1
i ).

Thus we need to calculate HH (B, Bg), B = A1 , g = diag(, 1 ).
Proposition 8.21. HH (B, Bg) is 1-dimensional, concentrated in degree 0 if = 1 and in
degree 2 otherwise.
Proof. If B = A1 then B has the following Koszul resolution as a B-bimodule:
B B B C2 B B B B.
Here the first map is given by the formula
b1 b2 7 b1 x yb2 b1 y xb2 b1 y x b2 + b1 x y b2 ,
the second map is given by
b1 x b2 7 b1 x b2 b1 xb2 , b1 y b2 7 b1 y b2 b1 yb2 ,
and the third map is the multiplication.
Thus the cohomology of B with coefficients in Bg can be computed by mapping this
resolution into Bg and taking the cohomology. This yields the following complex C :
(8.2) 0 Bg Bg Bg Bg 0,
where the first nontrivial map is given by bg 7 [bg, y] x [bg, x] y, and the second
nontrivial map is given by bg x 7 [x, bg], bg y 7 [y, bg].
Consider first the case g = 1. Equip the complex C with the Bernstein filtration (deg(x) =
deg(y) = 1), starting with 0, 1, 2, for C 0 , C 1 , C 2 , respectively (this makes the differential

preserve the filtration). Consider the associated graded complex Cgr . In this complex,

brackets are replaced with Poisson brackets, and thus it is easy to see that Cgr is the De
Rham complex for the affine plane, so its cohomology is C in degree 0 and 0 in other degrees.
Therefore, the cohomology of C is the same.
Now consider g 6= 1. In this case, declare that C 0 , C 1 , C 2 start in degrees 2,1,0 respectively
(which makes the differential preserve the filtration), and again consider the graded complex

Cgr . The graded Euler characteristic of this complex is (t2 2t + 1)(1 t)2 = 1.
0
The cohomology in the Cgr term is the set of b C[x, y] such that ab = bag for all a. This
means that HH0 = 0.
2
The cohomology of the Cgr term is the quotient of C[x, y] by the ideal generated by a ag ,
a C[x, y]. Thus the cohomology HH2 of the rightmost term is 1-dimensional, in degree 0.
By the Euler characteristic argument, this implies that HH1 = 0. The cohomology of the
filtered complex C is therefore the same, and we are done. 
The proposition implies that in the n-dimensional case HH (B, Bg) is 1-dimensional, con-
centrated in degree rank (1 g). It is not hard to check that the group G acts on the sum
of these 1-dimensional spaces by simply permuting the basis vectors. Thus the theorem is
proved. 
Remark 8.22. Another proof of Theorem 8.18 is given in [Pi].
Theorem 8.23. The algebra H1,c [G, V ], with formal c, is the universal deformation of
H1,0 [G, V ] = G Weyl(V ). More specifically, the map f : C[S]G HH2 (G Weyl(V ))
induced by this deformation coincides with the isomorphism of Corollary 8.19.
70
Proof. The proof (which we will not give) can be obtained by a direct computation with the
Koszul resolution for G Weyl(V ). Such a proof is given in [Pi]. The paper [EG] proves
a slightly weaker statement that the map f is an isomorphism, which suffices to show that
H1,c (G, V ) is the universal deformation of H1,0 [G, V ]. 

8.9. Finite dimensional representations of H0,c . Let Mc = SpecZ0,c . We can regard

H0,c = H0,c [G, V ] as a finitely generated module over Z0,c = O(Mc ). Let Mc be a central
character, : Z0,c C. Denote by hi the ideal in H0,c generated by the kernel of .
Proposition 8.24. If is generic then H0,c /hi is the matrix algebra of size |G|. In par-
ticular, H0,c has a unique irreducible representation V with central character . This repre-
sentation is isomorphic to CG as a G-module.
Proof. It is shown by a standard argument (which we will skip) that it is sufficient to check
the statement in the associated graded case c = 0. In this case, for generic , G SV /hi =
G Fun(O ), where O is the (free) orbit of G consisting of the points of V that map to
V /G, and Fun(O ) is the algebra of functions on O . It is easy to see that this algebra
is isomorphic to a matrix algebra, and has a unique irreducible representation, Fun(O ),
which is a regular representation of G. 
Corollary 8.25. Any irreducible representation of H0,c has dimension |G|.
Proof. We will use the following lemma.
Lemma 8.26 (The Amitsur-Levitzki identity). For any N N matrices X1 , . . . , X2N with
entries in a commutative ring A,
X
(1) X(1) X(2N ) = 0.
S2n
P
Proof. Consider the ring MatN (A) (1 , . . . , 2n ). Let X = i Xi i R. So we have
X
X2 = [Xi , Xj ]i j MatN (A even (1 , . . . , 2n )).
i<j

It is obvious that Tr X 2 = 0. Similarly, one can easily show that Tr X 4 = 0, . . . , Tr X 2N = 0.

Since the ring A even (1 , . . . , 2n ) is commutative, from the Cayley-Hamilton theorem, we
know that X 2N = 0 which implies the lemma. 
Since for generic the algebra H0,c /hi is a matrix algebra, the algebra H0,c satisfies the
Amitsur-Levitzki identity. Next, note that since H0,c is a finitely generated Z0,c -module (by
passing to the associated graded and using Hilberts theorem), every irreducible representa-
tion of H0,c is finite dimensional. If H0,c had an irreducible representation E of dimension
m > |G|, then by the density theorem the matrix algebra Matm would be a quotient of
H0,c . But one can show that the Amitsur-Levitzki identity of degree |G| is not satisfied for
matrices of bigger size than |G|. Contradiction. Thus, dim E |G|, as desired. 
In general, for special central characters there are representations of H0,c of dimension less
than |G|. However, in some cases one can show that all irreducible representations have
dimension exactly |G|. For example, we have the following result.
71
Theorem 8.27. Let G = Sn , V = h h , h = Cn (the rational Cherednik algebra for Sn ).
Then for c 6= 0, every irreducible representation of H0,c has dimension n! and is isomorphic
to the regular representation of Sn .
Proof. Let E be an irreducible representation of H0,c . Let us calculate the trace in E of any
permutation 6= 1. Let j be an index such that (j) = i 6= j. Then sij (j) = j. Hence in
H0,c we have
[yj , xi sij ] = [yj , xi ]sij = cs2ij = c.
Hence Tr E () = 0, and thus E is a multiple of the regular representation of Sn . But by
Theorem 8.25, dim E n!, so we get that E is the regular representation, as desired. 

8.10. Azumaya algebras. Let Z be a finitely generated commutative algebra over C, M =

SpecZ the corresponding affine scheme, and A a finitely generated Z-algebra.
Definition 8.28. A is said to be an Azumaya algebra of degree N if the completion A of
A at every maximal ideal in Z is the matrix algebra of size N over the completion Z of
Z.
Thus, an Azumaya algebra should be thought of as a bundle of matrix algebras on M. 5
For example, if E is an algebraic vector bundle on M then End(E) is an Azumaya algebra.
However, not all Azumaya algebras are of this form.
Example 8.29. For q C , consider the quantum torus
Tq = ChX 1 , Y 1 i/hXY qY Xi.
If q is a root of unity of order N, then the center of Tq is hX N , Y N i = C[M] where
M = (C )2 . It is not difficult to show that Tq is an Azumaya algebra of degree N, but
Tq C[M ] C(M) 6= MatN (C(M)), so Tq is not the endomorphism algebra of a vector bundle.
Example 8.30. Let X be a smooth irreducible variety over a field of characteristic p. Then
D(X), the algebra of differential operators on X, is an Azumaya algebra with rank pdim X ,
which is not an endomorphism algebra of a vector bundle. Its center is Z = O(T X)F , the
Frobenius twisted functions on T X.
It is clear that if A is an Azumaya algebra (say, over C) then for every central character
of A, A/hi is the algebra MatN (C) of complex N by N matrices, and every irreducible
representation of A has dimension N.
The following important result is due to M. Artin.
Theorem 8.31. Let A be a finitely generated (over C) polynomial identity (PI) algebra of
degree N (i.e. all the polynomial relations of the matrix algebra of size N are satisfied in
A). Then A is an Azumaya algebra if and only if every irreducible representation of A has
dimension exactly N.
Proof. See [Ar] Theorem 8.3. 

5If M is not affine, one can define, in a standard manner, the notion of a sheaf of Azumaya algebras on
M.
72
 Thus, by Theorem 8.27, for G = Sn , the rational Cherednik algebra H0,c (Sn , Cn ) for c 6= 0
is an Azumaya algebra of degree n!. Indeed, this algebra is PI of degree n! because the clas-
sical Dunkl representation embeds it into matrices of size n! over C(x1 , . . . , xn , p1 , . . . , pn )Sn .
Let us say that M is an Azumaya point if for some affine neighborhood U of the
localization of A to U is an Azumaya algebra. Obviously, the set Az(M) of Azumaya points
of M is open.
Now we come back to the study the space Mc corresponding to a symplectic reflection
algebra H0,c .
Theorem 8.32. The set Az(Mc ) coincides with the set of smooth points of Mc .
The proof of this theorem is given in the following two subsections.
Corollary 8.33. If G = Sn and V = h h , h = Cn (the rational Cherednik algebra case)
then Mc is a smooth algebraic variety for c 6= 0.
Proof. Directly from the above theorem. 
8.11. Cohen-Macaulay property and homological dimension. To prove Theorem 8.32,
we will need some commutative algebra tools. Let Z be a finitely generated commutative
algebra over C without zero divisors. By Noethers normalization lemma, there exist ele-
ments z1 , . . . , zn Z which are algebraically independent, such that Z is a finitely generated
module over C[z1 , . . . , zn ].
Definition 8.34. The algebra Z (or the variety SpecZ) is said to be Cohen-Macaulay if Z
is a locally free (=projective) module over C[z1 , . . . , zn ]. 6
Remark 8.35. It was shown by Serre that if Z is locally free over C[z1 , . . . , zn ] for some
choice of z1 , . . . , zn , then it happens for any choice of them (such that Z is finitely generated
as a module over C[z1 , . . . , zn ]).
Remark 8.36. Another definition of the Cohen-Macaulay property is that the dualizing
complex Z of Z is concentrated in degree zero. We will not discuss this definition here.
It can be shown that the Cohen-Macaulay property is stable under localization. Therefore,
it makes sense to make the following definition.
Definition 8.37. An algebraic variety X is Cohen-Macaulay if the algebra of functions on
every affine open set in X is Cohen-Macaulay.
Let Z be a finitely generated commutative algebra over C without zero divisors, and let
M be a finitely generated module over Z.
Definition 8.38. M is said to be Cohen-Macaulay if for some algebraically independent
z1 , . . . , zn Z such that Z is finitely generated over C[z1 , . . . , zn ], M is locally free over
C[z1 , . . . , zn ].
Again, if this happens for some z1 , . . . , zn , then it happens for any of them. We also
note that M can be Cohen-Macaulay without Z being Cohen-Macaulay, and that Z is a
Cohen-Macaulay algebra iff it is a Cohen-Macaulay module over itself.
We will need the following standard properties of Cohen-Macaulay algebras and modules.
6It was proved by Quillen that a locally free module over a polynomial algebra is free; this is a difficult
theorem, which will not be needed here.
73
Theorem 8.39. (i) Let Z1 Z2 be a finite extension of finitely generated commutative
C-algebras, without zero divisors, and M be a finitely generated module over Z2 . Then
M is Cohen-Macaulay over Z2 iff it is Cohen-Macaulay over Z1 .
(ii) Suppose that Z is the algebra of functions on a smooth affine variety. Then a Z-
module M is Cohen-Macaulay if and only if it is projective.
Proof. The proof can be found in the text book [Ei]. 
In particular, this shows that the algebra of functions on a smooth affine variety is Cohen-
Macaulay. Algebras of functions on many singular varieties are also Cohen-Macaulay.
Example 8.40. The algebra of regular functions on the cone xy = z 2 is Cohen-Macaulay.
This algebra can be identified as C[a, b]Z2 by letting x = a2 , y = b2 and z = ab, where the Z2
action is defined by a 7 a, b 7 b. It contains a subalgebra C[a2 , b2 ], and as a module
over this subalgebra, it is free of rank 2 with generators 1, ab.
Example 8.41. Any irreducible affine algebraic curve is Cohen-Macaulay. For example, the
algebra of regular functions on y 2 = x3 is isomorphic to the subalgebra of C[t] spanned by
1, t2 , t3 , . . .. It contains a subalgebra C[t2 ] and as a module over this subalgebra, it is free of
rank 2 with generators 1, t3 .
Example 8.42. Consider the subalgebra in C[x, y] spanned by 1 and xi y j with i+j 2. It is
a finite generated module over C[x2 , y 2 ], but not free. So this algebra is not Cohen-Macaulay.
Another tool we will need is homological dimension. We will say that an algebra A has
homological dimension d if any (left) A-module M has a projective resolution of length
d. The homological dimension of A is the smallest integer having this property. If such
an integer does not exist, A is said to have infinite homological dimension.
It is easy to show that the homological dimension of A is d if and only if for any A-
modules M, N one has Exti (M, N) = 0 for i > d. Also, the homological dimension clearly
does not decrease under taking associated graded of the algebra under a positive filtration
(this is clear from considering the spectral sequence attached to the filtration).
It follows immediately from this definition that homological dimension is Morita invariant.
Namely, recall that a Morita equivalence between algebras A and B is an equivalence of
categories A-mod B-mod. Such an equivalence maps projective modules to projective
ones, since projectivity is a categorical property (P is projective if and only if the functor
Hom(P, ) is exact). This implies that if A and B are Morita equivalent then their homological
dimensions are the same.
Then we have the following important theorem.
Theorem 8.43. The homological dimension of a commutative finitely generated C-algebra
Z is finite if and only if Z is regular, i.e. is the algebra of functions on a smooth affine
variety.
8.12. Proof of Theorem 8.32. First let us show that any smooth point of Mc is an
Azumaya point. Since H0,c = EndB0,c H0,c e = EndZ0,c (H0,c e), it is sufficient to show that
the coherent sheaf on Mc corresponding to the module H0,c e is a vector bundle near . By
Theorem 8.39 (ii), for this it suffices to show that H0,c e is a Cohen-Macaulay Z0,c -module.
To do so, first note that the statement is true for c = 0. Indeed, in this case the claim is
that SV is a Cohen-Macaulay module over (SV )G . But SV is a polynomial algebra, which
is Cohen-Macaulay, so the result follows from Theorem 8.39, (i).
74
 Now, we claim that if Z, M are positively filtered and grM is a Cohen-Macaulay grZ-
module then M is a Cohen-Macaulay Z-module. Indeed, let z1 , . . . , zn be homogeneous
algebraically independent elements of grZ such that grZ is a finite module over the subalge-
bra generated by them. Let z1 , . . . , zn be their liftings to Z. Then z1 , . . . , zn are algebraically
independent, and the module M over C[z1 , . . . , zn ] is finitely generated and (locally) free since
so is the module grM over C[z1 , . . . , zn ].
Recall now that grH0,c e = SV , grZ0,c = (SV )G . Thus the c = 0 case implies the general
case, and we are done.
Now let us show that any Azumaya point of Mc is smooth. Let U be an affine open set
in Mc consisting of Azumaya points. Then the localization H0,c (U) := H0,c Z0,c OU is an
Azumaya algebra. Moreover, for any U, the unique irreducible representation of H0,c (U)
with central character is the regular representation of G (since this holds for generic by
Proposition 8.24). This means that e is a rank 1 idempotent in H0,c (U)/hi for all . In
particular, H0,c (U)e is a vector bundle on U. Thus the functor F : OU -mod H0,c (U)-mod
given by the formula F (Y ) = H0,c (U)e OU Y is an equivalence of categories (the quasi-
inverse functor is given by the formula F 1 (N) = eN). Thus H0,c (U) is Morita equivalent
to OU , and therefore their homological dimensions are the same.
On the other hand, the homological dimension of H0,c is finite (in fact, it equals to dim V ).
To show this, note that by the Hilbert syzygies theorem, the homological dimension of
SV is dim V . Hence, so is the homological dimension of G SV (as ExtGSV (M, N) =
ExtSV (M, N)G ). Thus, since grH0,c = G SV , we get that H0,c has homological dimension
dim V . Hence, the homological dimension of H0,c (U) is also dim V (as the homological
dimension clearly does not increase under the localization). But H0,c (U) is Morita equivalent
to OU , so OU has a finite homological dimension. By Theorem 8.43, this implies that U
consists of smooth points.
Corollary 8.44. Az(Mc ) is also the set of points at which the Poisson structure of Mc is
symplectic.
Proof. The variety Mc is symplectic outside of a subset of codimension 2, because so is M0 .
Thus the set S of smooth points of Mc where the top exterior power of the Poisson bivector
vanishes is of codimension 2. Since the top exterior power of the Poisson bivector is locally
a regular function, this implies that S is empty. Thus, every smooth point is symplectic, and
the corollary follows from the theorem. 
8.13. Notes. Our exposition in this section follows Section 8 Section 10 of [E4].

75
 9. Calogero-Moser spaces
9.1. Hamiltonian reduction along an orbit. Let M be an affine algebraic variety and
G a reductive algebraic group. Suppose M is Poisson and the action of G preserves the
Poisson structure. Let g be the Lie algebra of G and g the dual of g. Let : M g be a
moment map for this action (we assume it exists). It induces a map : Sg C[M].
Let O be a closed coadjoint orbit of G, IO be the ideal in Sg corresponding to O, and let
JO be the ideal in C[M] generated by (IO ). Then JOG is a Poisson ideal in C[M]G , and
A = C[M]G /JOG is a Poisson algebra.
Geometrically, Spec(A) = 1 (O)/G (categorical quotient). It can also be written as
1
(z)/Gz , where z O and Gz is the stabilizer of z in G.

Definition 9.1. The scheme 1 (O)/G is called the Hamiltonian reduction of M with respect
to G along O. We will denote by R(M, G, O).

The following proposition is standard.

Proposition 9.2. If M is a symplectic variety and the action of G on 1 (O) is free, then
R(M, G, O) is a symplectic variety, of dimension dim(M) 2 dim(G) + dim(O).

9.2. The Calogero-Moser space. Let M = T Matn (C), and G = PGLn (C) (so g =
sln (C)). Using the trace form we can identify g with g, and M with Matn (C) Matn (C).
Then a moment map is given by the formula (X, Y ) = [X, Y ], for X, Y Matn (C).
Let O be the orbit of the matrix diag(1, 1, . . . , 1, n 1), i.e. the set of traceless
matrices T such that T + 1 has rank 1.

Definition 9.3 (Kazhdan, Kostant, Sternberg, [KKS]). The scheme Cn := R(M, G, O) is

called the Calogero-Moser space.

Proposition 9.4. The action of G on 1 (O) is free, and thus (by Proposition 9.2) Cn is a
smooth symplectic variety (of dimension 2n).

Proof. It suffices to show that if X, Y are such that XY Y X + 1 has rank 1, then (X, Y )
is an irreducible set of matrices. Indeed, in this case, by Schurs lemma, if B GLn is such
that BX = XB and BY = Y B then B is a scalar, so the stabilizer of (X, Y ) in PGLn is
trivial.
To show this, assume that W = 6 0 is an invariant subspace of X, Y . In this case, the
eigenvalues of [X, Y ] on W are a subcollection of the collection of n 1 copies of 1 and
one copy of n 1. The sum of the elements of this subcollection must be zero, since it is the
trace of [X, Y ] on W. But then the subcollection must be the entire collection, so W = Cn ,
as desired. 

Thus, Cn is the space of conjugacy classes of pairs of n n matrices (X, Y ) such that the
matrix XY Y X + 1 has rank 1.
In fact, one also has the following more complicated theorem.

Theorem 9.5 (G. Wilson, [Wi]). The Calogero-Moser space is connected.

We will give a proof of this theorem later, in Subsection 9.4.

76
9.3. The Calogero-Moser integrable system. Let M be a symplectic variety, and let
H1 , . . . , Hn be regular functions on M such that {Hi , Hj } = 0 and Hi s are algebraically
independent everywhere. Assume that M carries a symplectic action of a reductive algebraic
group G with moment map : M g , which preserves the functions Hi , and let O be
a coadjoint orbit of G. Assume that G acts freely on 1 (O), and so the Calogero-Moser
space R(M, G, O) is symplectic. The functions Hi descend to R(M, G, O). If they are still
algebraically independent and n = dim R(M, G, O)/2, then we get an integrable system on
R(M, G, O).
A vivid example of this is the Kazhdan-Kostant-Sternberg construction of the Calogero-
Moser system. In this case M = T Matn (C) (regarded as the set of pairs of matrices (X, Y )
as in Section 9.2), with the usual symplectic form = Tr (dY dX). Let Hi = Tr (Y i ),
i = 1, . . . , n. Let G = PGLn (C) act on M by conjugation, and let O be the coadjoint
orbit of G considered in Subsection 9.2. Then the system H1 , . . . , Hn descends to a system
of functions in involution on R(M, G, O), which is the Calogero-Moser space Cn . Since
this space is 2n-dimensional, H1 , . . . , Hn form an integrable system on Cn . It is called the
(rational) Calogero-Moser system.
The Calogero-Moser flow is, by definition, the Hamiltonian flow on Cn defined by the
Hamiltonian H = H2 = Tr (Y 2 ). Thus this flow is integrable, in the sense that it can be
included in an integrable system. In particular, its solutions can be found in quadratures
using the inductive procedure of reduction of order. However (as often happens with systems
obtained by reduction), solutions can also be found by a much simpler procedure, since
they can be found already on the non-reduced space M: indeed, on M the Calogero-
Moser flow is just the motion of a free particle in the space of matrices, so it has the form
gt (X, Y ) = (X + 2Y t, Y ). The same formula is valid on Cn . In fact, we can use the same
method to compute the flows corresponding to all the Hamiltonians Hi = Tr (Y i ), i N:
these flows are given by the formulas
(i)
gt (X, Y ) = (X + iY i1 t, Y ).
Let us write the Calogero-Moser system explicitly in coordinates. To do so, we first need
to introduce local coordinates on the Calogero-Moser space Cn .
To this end, let us restrict our attention to the open set Un Cn which consists of
conjugacy classes of those pairs (X, Y ) for which the matrix X is diagonalizable, with distinct
eigenvalues; by Wilsons Theorem 9.5, this open set is dense in Cn .
A point P Un may be represented by a pair (X, Y ) such that X = diag(x1 , . . . , xn ),
xi 6= xj . In this case, the entries of T := XY Y X are (xi xj )yij . In particular, the
diagonal entries are zero. Since the matrix T + 1 has rank 1, its entries ij have the form
ai bj for some numbers ai , bj . On the other hand, ii = 1, so bj = a1 1
j and hence ij = ai aj .
By conjugating (X, Y ) by the matrix diag(a1 , . . . , an ), we can reduce to the situation when
ai = 1, so ij = 1. Hence the matrix T has entries 1 ij (zeros on the diagonal, ones off
the diagonal). Moreover, the representative of P with diagonal X and T as above is unique
up to the action of the symmetric group Sn . Finally, we have (xi xj )yij = 1 for i 6= j, so
the entries of the matrix Y are yij = 1/(xi xj ) if i 6= j. On the other hand, the diagonal
entries yii of Y are unconstrained. Thus we have obtained the following result.
Proposition 9.6. Let Cnreg be the open set of (x1 , . . . , xn ) Cn such that xi 6= xj for i 6= j.
Then there exists an isomorphism of algebraic varieties : T (Cnreg /Sn ) Un given by the
77
formula
(x1 , . . . , xn , p1 , . . . , pn ) 7 (X, Y ),
where X = diag(x1 , . . . , xn ), and Y = Y (x, p) := (yij ),
1
yij = , i 6= j, yii = pi .
xi xj
In fact, we have a stronger result:
Proposition 9.7. is an isomorphism of symplectic varieties (where the cotangent bundle
is equipped with the usual symplectic structure).
For the proof of Proposition 9.7, we will need the following general and important but
easy theorem.
Theorem 9.8 (The necklace bracket formula). Let a1 , . . . , ar and b1 , . . . , bs be either X or
Y . Then on M we have
X
{Tr (a1 ar ), Tr (b1 bs )} = Tr (ai+1 ar a1 ai1 bj+1 bs b1 bj1 )
(i,j):ai =Y,bj =X
X
Tr (ai+1 ar a1 ai1 bj+1 bs b1 bj1 ).
(i,j):ai =X,bj =Y

Proof of Proposition 9.7. Let ak = Tr (X k ), bk = Tr (X k Y ). It is easy to check using the

necklace bracket formula that on M we have
{am , ak } = 0,
{bm , ak } = kam+k1 , {bm , bk } = (k m)bm+k1 .
P P k
On the other hand, ak = xki , bk = xi pi . Thus we see that
{f, g} = { f, g},
where f, g are either ak or bk . But the functions ak , bk , k = 0, . . . , n 1, form a local
coordinate system near a generic point of Un , so we are done. 
Now let us write the Hamiltonian of the Calogero-Moser system in coordinates. It has the
form
X X 1
(9.1) H = Tr (Y (x, p)2 ) = p2i .
i i6=j
(xi xj )2
Thus the Calogero-Moser Hamiltonian describes the motion of a system of n particles on the
line with interaction potential 1/x2 , which we considered in Section 2.
Now we finally see the usefulness of the Hamiltonian reduction procedure. The point is
that it is not clear at all from formula (9.1) why the Calogero-Moser Hamiltonian should be
completely integrable. However, our reduction procedure implies the complete integrability
of H, and gives an explicit formula for the first integrals: 7
Hi = Tr (Y (x, p)i ).
Moreover, this procedure immediately gives us an explicit solution of the system. Namely,
assume that x(t), p(t) is the solution with initial condition x(0), p(0). Let (X0 , Y0 ) =
7Thus,for type A we have two methods of proving the integrability of the Calogero-Moser system - one
using Dunkl operators and one using Hamiltonian reduction.
78
(x(0), p(0)). Then xi (t) are the eigenvalues of the matrix Xt := X0 + 2tY0 , and pi (t) =
xi (t)/2.

9.4. Proof of Wilsons theorem. Let us now give a proof of Theorem 9.5.
We have already shown that all components of Cn are smooth and have dimension 2n.
Also, we know that there is at least one component (the closure of Un ), and that the other
components, if they exist, do not contain pairs (X, Y ) in which X is regular semisimple.
This means that these components are contained in the Q hypersurface (X) = 0, where
(X) stands for the discriminant of X (i.e., (X) := i6=j (xi xj ), where xi are the
eigenvalues of X).
Thus, to show that such additional components dont in fact exist, it suffices to show that
the dimension of the subscheme Cn (0) cut out in Cn by the equation (X) = 0 is 2n 1.
To do so, first notice that the condition rank ([X, Y ] + 1) = 1 is equivalent to the equa-
tion 2 ([X, Y ] + 1) = 0; thus, the latter can be used as the equation defining Cn inside
T Matn /P GLn .
Define Cn0 := Spec(grO(Cn )) to be the degeneration of Cn (we use the filtration on O(Cn )
defined by deg(X) = 0, deg(Y ) = 1). Then Cn0 is a closed subscheme in the scheme Cen0 cut
out by the equations 2 ([X, Y ]) = 0 in T Matn /PGLn .
Let (Cen0 )red be the reduced part of Cen0 . Then (Cen0 )red coincides with the categorical quotient
{(X, Y )|rank ([X, Y ]) 1}/PGLn .
Our proof is based on the following proposition.
Proposition 9.9. The categorical quotient {(X, Y )|rank ([X, Y ]) 1}/PGLn coincides with
the categorical quotient {(X, Y )|[X, Y ] = 0}/PGLn .
Proof. It is clear that {(X, Y )|[X, Y ] = 0}/PGLn is contained in {(X, Y )|rank ([X, Y ])
1}/PGLn . For the proof of the opposite inclusion we need to show that any regular function
on {(X, Y )|rank ([X, Y ]) 1}/PGLn is completely determined by its values on the subvariety
{(X, Y )|[X, Y ] = 0}/PGLn , i.e. that any invariant polynomial on the set of pairs of matrices
with commutator of rank at most 1 is completely determined by its values on pairs of
commuting matrices. To this end, we need the following lemma from linear algebra.
Lemma 9.10. If A, B are square matrices such that [A, B] has rank 1, then there exists
a basis in which both A, B are upper triangular.
Proof. Without loss of generality, we can assume ker A 6= 0 (by replacing A with A if
needed) and that A 6= 0. It suffices to show that there exists a proper nonzero subspace
invariant under A, B; then the statement will follow by induction in dimension.
Let C = [A, B] and suppose rank C = 1 (since the case rank C = 0 is trivial). If ker A
ker C, then ker A is B-invariant: if Av = 0 then ABv = BAv + Cv = 0. Thus ker A is the
required subspace. If ker A * ker C, then there exists a vector v such that Av = 0 but Cv 6= 0.
So ABv = Cv 6= 0. Thus ImC ImA. So ImA is B-invariant: BAv = ABv + Cv ImA.
So ImA is the required subspace.
This proves the lemma. 
Now we are ready to prove Proposition 9.9. By the fundamental theorem of invariant
theory, the ring of invariants of X and Y is generated by traces of words of X and Y :
Tr(w(X, Y )). If X and Y are upper triangular with eigenvalues xi , yi , then any such trace
79
 P m r
has the form xi yi , i.e. coincides with the value of the corresponding invariant on the
diagonal parts Xdiag , Ydiag of X and Y , which commute. The proposition is proved. 
We will also need the following proposition:
Proposition 9.11. The categorical quotient {(X, Y )|[X, Y ] = 0}/PGLn is isomorphic to
(Cn Cn )/Sn , i.e. its function algebra is C[x1 , . . . , xn , y1 , . . . , yn ]Sn .
Proof. Restriction to diagonal matrices defines a homomorphism
: O({(X, Y )|[X, Y ] = 0}/PGLn ) C[x1 , . . . , xn , y1 , . . . , yn ]Sn .
Since (as explained in the proof of Proposition 9.9), any invariant polynomial of entries of
commuting matrices isP determined by its values on diagonal matrices, this map is injective.
Also, (Tr (X m Y r )) = xm r
i yi , where xi , yi are the eigenvalues of X and Y .
Now we use the following well known theorem of H. Weyl (from his book Classical
groups).
Theorem 9.12. Let B be an algebra over C. Then the algebra S n B is generated by elements
of the form
b 1 1 + 1 b 1 + + 1 1 b.
Proof. Since S n B is linearly spanned by elements of the form x x, x B, it suffices
to prove the theorem in the special case B = C[x]. In this case, the result is simply the fact
that the ring of symmetric functions is generated by power sums, which is well known. 
P
Corollary 9.13. The ring C[x1 , . . . , xn , y1, . . . , yn ]Sn is generated by the polynomials xm r
i yi
for m, r 0, m + r > 0.
Proof. Apply Theorem 9.12 in the case B = C[x, y]. 
Corollary 9.13 implies that is surjective. Proposition 9.11 is proved. 
Now we are ready to prove Wilsons theorem. Let Cn (0)0 be the degeneration of Cn (0), i.e.
the subscheme of Cn0 cut out by the equation (X) = 0. According to Propositions 9.9 and
9.11, the reducedQpart (Cn (0)0 )red is contained in the hypersurface in (Cn Cn )/Sn cut out
by the equation i<j (xi xj ) = 0. This hypersurface has dimension 2n 1, so we are done.

9.5. The Gan-Ginzburg theorem. Let Comm(n) be the commuting scheme defined in
T Matn = Matn Matn by the equations [X, Y ] = 0, X, Y Matn . Let G = PGLn , and
consider the categorical quotient Comm(n)/G (i.e., the Hamiltonian reduction 1 (0)/G of
T Matn by the action of G), whose algebra of regular functions is A = C[Comm(n)]G .
It is not known whether the commuting scheme Comm(n) is reduced (i.e. whether the
corresponding ideal is a radical ideal); this is a well known open problem. The underlying
variety is irreducible (as was shown by Gerstenhaber [Ge1]), but very singular, and has a
very complicated structure. However, we have the following result.
Theorem 9.14 (Gan, Ginzburg, [GG]). Comm(n)/G is reduced, and isomorphic to C2n /Sn .
Thus A = C[x1 , . . . , xn , y1, . . . , yn ]Sn . The Poisson bracket on this algebra is induced from
the standard symplectic structure on C2n .
80
Sketch of the proof. Look at the almost commuting variety Mn gln gln Cn (Cn )
defined by
Mn = {(X, Y, v, f)|[X, Y ] + v f = 0}.
Gan and Ginzburg proved the following result.
Theorem 9.15. Mn is a complete intersection. It has n + 1 irreducible components denoted
by Min , labeled by i = dim ChX, Y iv. Also, Mn is generically reduced.
Since Mn is generically reduced and is a complete intersection, by a standard result of
commutative algebra it is reduced. Thus C[Mn ] has no nonzero nilpotents. This implies
C[Mn ]G has no nonzero nilpotents.
However, it is easy to show that the algebra C[Mn ]G is isomorphic to the algebra of
invariant polynomials of entries of X and Y modulo the rank 1 relation 2 [X, Y ] = 0. By
a scheme-theoretic version of Proposition 9.9 (proved in [EG]), the latter is isomorphic to
A. This implies the theorem (the statement about Poisson structures is checked directly in
coordinates on the open part where X is regular semisimple). 
9.6. The space Mc for Sn and the Calogero-Moser space. Let H0,c = H0,c [Sn , V ] be
the symplectic reflection algebra of the symmetric group Sn and space V = h h , where
h = Cn (i.e., the rational Cherednik algebra H0,c (Sn , h)). Let Mc = Spec B0,c [Sn , V ] be the
Calogero-Moser space defined in Section 8.5. It is a symplectic variety for c 6= 0.
Theorem 9.16. For c 6= 0 the space Mc is isomorphic to the Calogero-Moser space Cn as a
symplectic variety.
Proof. To prove the theorem, we will first construct a map f : Mc Cn , and then prove that
f is an isomorphism.
Without loss of generality, we may assume that c = 1. As we have shown before, the
algebra H0,c is an Azumaya algebra. Therefore, Mc can be regarded as the moduli space of
irreducible representations of H0,c .
Let E Mc be an irreducible representation of H0,c . We have seen before that E has
dimension n! and is isomorphic to the regular representation as a representation of Sn . Let
Sn1 Sn be the subgroup which preserves the element 1. Then the space of invariants
E Sn1 has dimension n. On this space we have operators X, Y : E Sn1 E Sn1 obtained
by restriction of the operators x1 , y1 on E to the subspace of invariants. We have
n
X
[X, Y ] = T := s1i .
i=2
Let us now calculate the right hand side of this equation explicitly. Let e be the symmetrizer
of Sn1 . Let us realize the regular representation E of Sn as C[Sn ] with action of Sn by
left multiplication. Then v1 = e, v2 = es12 , . . . , vn = es1n is a basis of E Sn1 . The element
T commutes with e, so we have X
T vi = vj .
j6=i
This means that T + 1 has rank 1, and hence the pair (X, Y ) defines a point on the Calogero-
Moser space Cn . 8
8Note that the pair (X, Y ) is well defined only up to conjugation, because the representation E is well
defined only up to an isomorphism.
81
 We now set (X, Y ) = f (E). It is clear that f : Mc Cn is a regular map. So it remains to
show that f is an isomorphism. This is equivalent to showing that the corresponding map
of function algebras f : O(Cn ) B0,c is an isomorphism.
Let us calculate f and f more explicitly. To do so, consider the open set U in Mc consisting
of representations in which xi xj acts invertibly. These are exactly the representations that
are obtained by restricting representations of Sn C[x1 , . . . , xn , p1 , . . . , pn , (x)1 ] using the
classical Dunkl embedding. Thus representations E U are of the form E = E, (, Cn ,
and has distinct coordinates), where E, is the space of complex valued functions on the
orbit O, C2n , with the following action of H0,c :
X (sij F )(a, b)
(xi F )(a, b) = ai F (a, b), (yi F )(a, b) = bi F (a, b) + .
j6=i
ai aj

(the group Sn acts by permutations).

S
Now let us consider the space E,n1 . A basis of this space is formed by characteristic
functions of Sn1-orbits on O, . Using the above presentation, it is straightforward to
calculate the matrices of the operators X and Y in this basis:
X = diag(1 , . . . , n ),
and
1
Yij = i if j = i, Yij = if j 6= i.
i j
This shows that f induces an isomorphism f |U : U Un , where Un is the subset of Cn
consisting of pairs (X, Y ) for which X has distinct eigenvalues.
From this presentation, it is straigtforward that f (Tr (X p )) = xp1 + + xpn for every
positive integer p. Also, f commutes with the natural SL2 (C)-action on Mc and Cn (by
(X, Y ) (aX + bY, cX + dY )), so we also get f (Tr (Y p )) = y1p + + ynp , and
p
p 1 X X m pm
f (Tr (X Y )) = x yi xi .
p + 1 m=0 i i
Now, using the necklace bracket formula on Cn and the commutation relations of H0,c , we
find, by a direct computation, that f preserves Poisson bracket on the elements Tr (X p ),
Tr (X q Y ). But these elements are a local coordinate system near a genericP
point, so it
Pfollows
that f is a Poisson map. Since the algebra B0,c is Poisson generated by xpi and yip for

all p, we get that f is a surjective map.
Also, f is injective. Indeed, by Wilsons theorem the Calogero-Moser space is connected,
and hence the algebra O(Cn ) has no zero divisors, while Cn has the same dimension as Mc .
This proves that f is an isomorphism, so f is an isomorphism. 
9.7. The Hilbert scheme Hilbn (C2 ) and the Calogero-Moser space. The Hilbert
scheme Hilbn (C2 ) is defined to be
Hilbn (C2 ) = { ideals I C[x, y]|codimI = n}
= {(E, v)|E is a C[x, y]-module of dimension n, v is a cyclic vector of E}.
The second equality can be easily seen from the short exact sequence
0 I C[x, y] E 0.
82
Let S (n) C2 = |C2 {z
C}2 /Sn , where Sn acts by permutation. We have a natural map
n times
Hilbn (C2 ) S (n) C2 which sends every ideal I to its zero set (with multiplicities). This map
is called the Hilbert-Chow map.
Theorem 9.17 (Fogarty, [F]). (i) Hilbn (C2 ) is a smooth quasiprojective variety.
(ii) The Hilbert-Chow map Hilbn (C2 ) S (n) C2 is projective. It is a resolution of singu-
larities.
Proof. Proof can be found in [Na]. 

Now consider the Calogero-Moser space Cn defined in Section 9.2.

Theorem 9.18 (see [Na]). The Hilbert Scheme Hilbn (C2 ) is C -diffeomorphic to Cn .
Remark 9.19. More precisely there exists a family of algebraic varieties over A1 , say Xt ,
t A1 , such that Xt is isomorphic to Cn if t 6= 0 and X0 is the Hilbert scheme; and also if we
denote by Xt the deformation of C2n /Sn into the Calogero-Moser space, then there exists
a map ft : Xt Xt , such that for t 6= 0, ft is an isomorphism and f0 is the Hilbert-Chow
map.
Remark 9.20. Consider the action of G = PGLn on T Matn . As we have discussed, the
corresponding moment map is (X, Y ) = [X, Y ], so 1 (0) = {(X, Y )|[X, Y ] = 0} is the
commuting variety. We can consider two kinds of quotient 1 (0)/G (i.e., of Hamiltonian
reduction):
(1) The categorical quotient, i.e.,
Spec(C[xij , yij ]/h[X, Y ] = 0i)G
= (Cn Cn )/Sn .
It is a reduced (by Gan-Ginzburg Theorem 9.14), affine but singular variety.
(2) The GIT quotient, in which the stability condition is that there exists a cyclic vector
for X, Y . This quotient is Hilbn (C2 ), which is smooth but not affine.
Both of these reductions are degenerations of the reduction along the orbit of matrices T
such that T + 1 has rank 1, which yields the space Cn . This explains why Theorem 9.18 and
the results mentioned in Remark 9.19 are natural to expect.
9.8. The cohomology of Cn . We also have the following result describing the cohomology
of Cn (and hence, by Theorem 9.18, of Hilbn (C2 )). Define the age filtration for the symmetric
group Sn by setting
age(transposition) = 1.
Then one can show that for any Sn , age() = rank (1 )|reflection representation . It is easy
to see that 0 age n1. Notice also that the age filtration can be defined for any Coxeter
group.
Theorem 9.21 (Lehn-Sorger, Vasserot). The cohomology ring H (Cn , C) lives in even degrees
only and is isomorphic to gr(Center(C[Sn ])) under the age filtration (with doubled degrees).
Proof. Let us sketch a noncommutative-algebraic proof of this theorem, given in [EG]. This
proof is based on the following result.
83
Theorem 9.22 (Nest-Tsygan, [NT]). If M is an affine symplectic variety and A is a quan-
tization of M, then
HH (A[~1 ], A[~1 ])
= H (M, C((~)))
as an algebra over C((~)).
Now, we know that the algebra Bt,c is a quantization of Cn . Therefore by above theorem,
the cohomology algebra of Cn is the cohomology of Bt,c (for generic t). But Bt,c is Morita
equivalent to Ht,c , so this cohomology is the same as the Hochschild cohomology of Ht,c .
However, the latter can be computed by using that Ht,c is given by generators and relations
(by producing explicit representatives of cohomology classes and computing their product),
which gives the result. 
9.9. Notes. Sections 9.19.6 follow Section 1, 2, 4 of [E4]; the parts about the Hilbert
scheme and its relation to Calogero-Moser spaces follow the book [Na] (see also [GS]); the
other parts follow the paper [EG].

84
 10. Quantization of Claogero-Moser spaces
10.1. Quantum moment maps and quantum Hamiltonian reduction. Now we would
like to quantize the notion of a moment map. Let g be a Lie algebra, and A be an associative
algebra equipped with a g-action, i.e. a Lie algebra map : g DerA. A quantum moment
map for (A, ) is an associative algebra homomorphism : U(g) A such that for any
a g, b A one has [(a), b] = (a)b.
The space of g-invariants Ag , i.e. elements b A such that [(a), b] = 0 for all a g, is
a subalgebra of A. Let J A be the left ideal generated by (a), a g. Then J is not
g g
a 2-sided
P ideal, but J := J A is a 2-sided P ideal in P Ag . Indeed, let c Ag , and b J g ,
b = i bi (ai ), bi A, ai g. Then bc = bi (ai )c = bi c(ai ) J g .
Thus, the algebra A//g := Ag /J g is an associative algebra, which is called the quantum
Hamiltonian reduction of A with respect to the quantum moment map .

10.2. The Levasseur-Stafford theorem. In general, similarly to the classical case, it is

rather difficult to compute the quantum reduction A//g. For example, in this subsection
we will describe A//g in the case when A = D(g) is the algebra of differential operators
on a reductive Lie algebra g, and g acts on A through the adjoint action on itself. This
description is a very nontrivial result of Levasseur and Stafford.
Let h be a Cartan subalgebra of g, and W the Weyl group of (g, h). Let hreg denote the set
of regular points in h, i.e. the complement of the reflection hyperplanes. To describe D(g)//g,
we will construct a homomorphism HC : D(g)g D(h)W , called the Harish-Chandra homo-
morphism (as it was first constructed by Harish-Chandra). Recall that we have the classical
Harish-Chandra isomorphism : C[g]g C[h]W , defined simply by restricting g-invariant
functions on g to the Cartan subalgebra h. Using this isomorphism, we can define an action
of D(g)g on C[h]W , which is clearly given by W -invariant differential operators. However,
these operators will, in general, have poles on the reflection hyperplanes. Thus we get a
homomorphism HC : D(g)g D(hreg )W .
The homomorphism HC is called the radial part homomorphism, as for example for
g = su(2) it computes the radial parts of rotationally invariant differential operators on R3
in spherical coordinates. This homomorphism is not yet what we want, since it does not
actually land in D(h)W (the radial parts have poles).
ThusQ we define the Harish-Chandra homomorphism by twisting HC by the discriminant
(x) = >0 (, x) (x h, and runs over positive roots of g):

HC(D) := HC (D) 1 D(hreg )W .

Theorem 10.1. (i) (Harish-Chandra, [HC]) For any reductive g, HC lands in D(h)W
D(hreg )W .
(ii) (Levasseur-Stafford [LS]) The homomorphism HC defines an isomorphism D(g)//g =
D(h)W .
Remark 10.2. (1) Part (i) of the theorem says that the poles magically disappear after
conjugation by .
(2) Both parts of this theorem are quite nontrivial. The first part was proved by Harish-
Chandra using analytic methods, and the second part by Levasseur and Stafford
using the theory of D-modules.
85
 In the case g = gln , Theorem 10.1 is a quantum analog of Theorem 9.14. The remaining
part of this subsection is devoted to the proof of Theorem 10.1 in this special case, using
Theorem 9.14.
We start the proof with the following proposition, valid for any reductive Lie algebra.
Proposition 10.3. If D (Sg)g is a differential operator with constant coefficients, then
HC(D) is the W -invariant differential operator with constant coefficients on h, obtained from
D via the classical Harish-Chandra isomorphism : (Sg)g (Sh)W .
Proof. Without loss of generality, we may assume that g is simple.
Lemma 10.4. Let D be the Laplacian g of g, corresponding to an invariant form. Then
HC(D) is the Laplacian h .
Proof. Let us calculate HC (D). We have
Xr X
D= x2i + 2 f e ,
i=1 >0

where xi is an orthonormal basis of h, and e , f are root elements such that (e , f ) = 1.

Thus if F (x) is a g-invariant function on g, then we get
Xr X
(DF )|h = x2i (F |h ) + 2 (f e F )|h .
i=1 >0

Now let x h, and consider (f e F )(x). We have

(f e F )(x) = s t |s=t=0 F (x + tf + se ).
On the other hand, we have
1 e
Ad(es(x)
)(x + tf + se ) = x + tf + ts(x)1 h + ,
where h = [e , f ]. Hence, s t |s=t=0 F (x + tf + se ) = (x)1 (h F )(x). This implies that
X
HC (D)F (x) = h F (x) + 2 (x)1 h F (x).
>0
P
Now the statement of the Lemma reduces to the identity 1 h = h +2 >0 (x)1 h .
This identity follows immediately from the identity h = 0. To prove the latter, it suffices
to note that is the lowest degree nonzero polynomial on h, which is antisymmetric under
the action of W . The lemma is proved. 
Now let D be any element of (Sg)g D(g)g of degree d (operator with constant coeffi-
cients). It is obvious that the leading order part of the operator HC(D) is the operator (D)
with constant coefficients, whose symbol is just the restriction of the symbol of D from g
to h . Our job is to show that in fact HC(D) = (D). To do so, denote by Y the difference
HC(D) (D). Assume Y 6= 0. By Lemma 10.4, the operator HC(D) commutes with h .
Therefore, so does Y . Also Y has homogeneity degree d but order m d 1. Let S(x, p) be
the symbol of Y (x h, p h ). Then S is a homogeneous function of homogeneity degree
d under the transformations x t1 x, p tp, polynomial in p of degree m. From these
properties of S it is clear that S is not a polynomial (its degree in x is m d < 0). On
the other hand, since Y commutes with h , the Poisson bracket of S with p2 is zero. Thus
Proposition 10.3 follows from Lemma 2.22. 
86
 Now we continue the proof of Theorem 10.1. Consider the filtration on D(g) in which
deg(g) = 1 deg(g ) = 0 (the order filtration), and the associated graded map grHC : C[g
g ]g C[hreg h ]W , which attaches to every differential operator the symbol of its radial
part. It is easy to see that this map is just the restriction map to h h g g , so it
actually lands in C[h h ]W .
Moreover, grHC is a map onto C[h h ]W . Indeed, grHC is a Poisson map, so the
surjectivity follows from the following Lemma.
Lemma 10.5. P C[h h ]W is generated W W
P m as a Poisson algebra by C[h] and C[h ] , i.e. by
m
functions fm = xi and fm = pi , m 1.
P

Proof. We have {fm , fr } = mr xir1 pm1
i . Thus the result follows from Corollary 9.13. 

Let K0 be the kernel of grHC. Then by Theorem 9.14, K0 is the ideal of the commuting
scheme Comm(g)/G.
Now consider the kernel K of the homomorphism HC. It is easy to see that K J g ,
so gr(K) gr(J)g . On the other hand, since K0 is the ideal of the commuting scheme,
we clearly have gr(J)g K0 , and K0 grK. This implies that K0 = grK = gr(J)g , and
K = J g.
It remains to show that Im HC = D(h)W . Since grK = K0 , we have grIm HC = C[hh ]W .
Therefore, to finish the proof of the Harish-Chandra and Levasseur-Stafford theorems, it
suffices to prove the following proposition.
Proposition 10.6. Im HC D(h)W .
Proof. We will use the following Lemma.
Lemma 10.7 (N. Wallach, [Wa]). D(h)W is generated as an algebra by W -invariant func-
tions and W -invariant differential operators with constant coefficients.
Proof. The lemma follows by taking associated graded algebras from Lemma 10.5. 
Remark 10.8. Levasseur and Stafford showed [LS] that this lemma is valid for any finite
group W acting on a finite dimensional vector space h. However, the above proof does not
apply, since, as explained in [Wa], Lemma 10.5 fails for many groups W , including Weyl
groups of exceptional Lie algebras E6 , E7 , E8 (in fact it even fails for the cyclic group of
order > 2 acting on a 1-dimensional space!). The general proof is more complicated and uses
some results in noncommutative algebra.
Lemma 10.7 and Proposition 10.3 imply Proposition 10.6. 

Thus, Theorem 10.1 is proved.

10.3. Corollaries of Theorem 10.1. Let gR be the compact form of g, and O a regular
coadjoint orbit in gR . Consider the map
Z
O : h C, O (x) = e(b,x) db, x h,
O

where db is the measure on the orbit coming from the Kirillov-Kostant symplectic structure.
87
Theorem 10.9 (Harish-Chandra formula). For a regular element x h, we have
X
O (x) = 1 (x) (1)(w) e(w,x) ,
wW

where is the intersection of O with the dominant chamber in the dual Cartan subalgebra
hR gR , and (w) is the length of an element w W .
Proof. Take D (Sg)g . Then (x)O is an eigenfunction of HC(D) = (D) (Sh)W with
eigenvalue O (D), where O (D) is the value of the invariant polynomial D at the orbit O.
Since the solutions of the equation (D) = O (D) have a basis e(w,x) where w W ,
we have X
(x)O (x) = cw e(w,x) .
wW
Since it is antisymmetric, we have cw = c (1)(w) , where c is a constant. The fact that
c = 1 can be shown by comparing the asymptotics of both sides as x in the regular
chamber (using the stationary phase approximation for the integral). 
From Theorem 10.9 and the Weyl Character formula, we have the following corollary.
Corollary 10.10 (Kirillov character formula for finite dimensional representations, [Ki]). If
is a dominant integral weight, and L is the corresponding representation of G, then
Z
(x)
x
Tr L (e ) = e(b,x) db,
Tr (x) O+
where
Q Tr (x) is the trigonometric version of (x), i.e. the Weyl denominator
(x)/2 (x)/2
>0 (e e ), and O denotes the coadjoint orbit passing through .
10.4. The deformed Harish-Chandra homomorphism. Finally, we would like to ex-
plain how to quantize the Calogero-Moser space Cn , using the procedure of quantum Hamil-
tonian reduction.
Let g = gln , A = D(g) as above. Let k be a complex number, and Wk be the representation
of sln on the space of functions of the form (x1 xn )k f (x1 , . . . , xn ), where f is a Laurent
polynomial of degree 0. We regard Wk as a g-module by pulling it back to g under the
natural projection g sln . Let Ik be the annihilator of Wk in U(g). The ideal Ik is the
quantum counterpart of the coadjoint orbit of matrices T such that T + 1 has rank 1.
Let Bk = D(g)g /(D(g)(Ik ))g where : U(g) A is the quantum momentum map (the
quantum Hamiltonian reduction with respect to the ideal Ik ). Then Bk has a filtration
induced from the order filtration of D(g)g .
Let HCk : D(g)g Bk be the natural homomorphism, and K(k) be the kernel of HCk .
Theorem 10.11 (Etingof-Ginzburg, [EG]). (i) K(0) = K, B0 = D(h)W , HC0 = HC.
(ii) grK(k) = Ker (grHCk ) = K0 for all complex k. Thus, HCk is a flat family of
homomorphisms.
(iii) The algebra grBk is commutative and isomorphic to C[h h ]W as a Poisson algebra.
Because of this theorem, the homomorphism HCk is called the deformed Harish-Chandra
homomorphism.
Theorem 10.11 implies that Bk is a quantization of the Calogero-Moser space Cn (with
deformation parameter 1/k). But we already know one such quantization - the spherical
88
Cherednik algebra B1,k for the symmetric group. Therefore, the following theorem comes as
no surprize.
Theorem 10.12 ([EG]). The algebra Bk is isomorphic to the spherical rational Cherednik
algebra B1,k (Sn , Cn ).
Thus, quantum Hamiltonian reduction provides a Lie-theoretic construction of the spher-
ical rational Cherednik algebra for the symmetric group. A similar (but more complicated)
Lie theoretic construction exists for symplectic reflection algebras for wreath product groups
defined in Example 8.5 (see [EGGO]).
10.5. Notes. Our exposition in this section follows Section 4, Section 5 of [E4].

89
 References
[AFLS] J. Alev, M.A. Farinati, T. Lambre, and A.L. Solotar: Homologie des invariants dune algebre de
Weyl sous laction dun groupe fini. J. of Algebra 232 (2000), 564577.
[Ar] M. Artin: On Azumaya algebras and finite dimensional representations of rings. J. Algebra 11 (1969),
pp. 532563.
[B] N. Bourbaki: Chapitres IV, V, VI of Groupes et algebres de Lie, Hermann, Paris, 1968.
[BB] A. Beilinson, J. Bernstein: Proof of Jantzens conjecture. Advances in Soviet Mathematics 16 (1993),
150.
[BE] R. Bezrukavnikov, P. Etingof: Parabolic induction and restriction functors for rational Cherednik
algebras. Selecta Math. (N.S.) 14 (2009), no. 3-4, 397425.
[BEG] Y. Berest, P. Etingof, V. Ginzburg: Finite-dimensional representations of rational Cherednik algebras.
Int. Math. Res. Not. 2003, no. 19, 10531088.
[BGS] A. Beilinson, V. Ginzburg, W. Soergel: Koszul duality patterns in Representation Theory, J. Amer.
Math. Soc. 9 (1996), 473527.
[BMR] M. Broue, G. Malle, R. Rouquier: Complex reflection groups, braid groups, Hecke algebras. J. Reine
Angew. Math. 500 (1998), 127-190.
[BG] A. Braverman, D. Gaitsgory: Poincare-Birkhoff-Witt theorem for quadratic algebras of Koszul type. J.
Algebra 181 (1996), 315328.
[BGo] K.A. Brown, I. Gordon: Poisson orders, symplectic reflection algebras and representation theory. J.
Reine Angew. Math. 559 (2003), 193216.
[CE] T. Chmutova, P. Etingof: On some representations of the rational Cherednik algebra. Represent. Theory
7 (2003), 641650.
[Ch] I. Cherednik: Double affine Hecke algebras. London Mathematical Society Lecture Note Series, 319,
Cambridge University Press, Cambridge, 2005.
[Ch1] I. Cherednik: A unification of Knizhnik-Zamolodchikov and Dunkl operators via affine Hecke algebras.
Invent. Math. 106 (1991), pp. 411431.
[Che] C. Chevalley: Invariants of finite groups generated by reflections. Amer. J. Of Math. 67 77 (4): 778
782, 1955.
[Dr] V. Drinfeld: On quadratic commutation relations in the quasiclassical case, Mathematical physics,
functional analysis (Russian), 2534, 143, Naukova Dumka, Kiev, 1986. Selecta Math. Sovietica. 11
(1992), 317326.
[DJ] R. Dipper, G. James: Representations of Hecke algebras of general linear groups. Proc. London Math.
Soc. (3), 1986, vol. 52 (1) pp. 2052.
[Du1] C. Dunkl: Differential-difference operators associated to reflection groups. Trans. Amer. Math. Soc.
311 (1989), 167183.
[Du2] C. Dunkl: Integral kernels with reflection group invariance. Can.J.Math. 43 (1991), 12131227.
[E1] P. Etingof: Cherednik and Hecke algebras of varieties with a finite group action. math.QA/0406499.
[E2] P. Etingof: A uniform proof of the Macdonald-Mehta-Opdam identity for finite Coxeter groups.
arXiv:0903.5084.
[E3] P. Etingof: Supports of irreducible spherical representations of rational Cherednik algebras of finite
Coxeter groups. arXiv:0911.3208.
[E4] P. Etingof: Calogero-Moser systems and representation theory. Zurich Lectures in Advanced Mathe-
matics. European Mathematical Society (EMS), Zrich, 2007; arXiv:math/0606233.
[EG] P. Etingof, V. Ginzburg: Symplectic reflection algebras, Calogero-Moser space, and deformed Harish-
Chandra homomorphism. Invent. Math. 147 (2002), 243348.
[EGGO] P. Etingof; W. L. Gan; V. Ginzburg; A. Oblomkov, Harish-Chandra homomorphisms and sym-
plectic reflection algebras for wreath-products, Publ. Math. Inst. Hautes tudes Sci., v.105, 2007, p.
91.
[Ei] D. Eisenbud: Commutative algebra. With a view toward algebraic geometry. Graduate Texts in Math-
ematics, 150. Springer-Verlag, New York, 1995.
[ER] P. Etingof, E. Rains: New deformations of group algebras of Coxeter groups. Int. Math. Res. Not. 2005,
no. 10, 635646.

90
[EOR] P. Etingof, A. Oblomkov, E. Rains: Generalized double affine Hecke algebras of rank 1 and quantized
del Pezzo surfaces. Adv. Math. 212 (2007), no. 2, 749796.
[EFK] Etingof P., Frenkel I., Kirillov A. (jr.) Lectures on representation theory and Knizhnik-Zamolodchikov
equation AMS, 1998.
[F] J. Fogarty, Algebraic families on an algebraic surface. Amer. J. Math. 90 (1968), 511521.
[Ge] M. Gerstenhaber, On the deformation of rings and algebras. Ann. of Math. (2) 79 1964 59103.
[Ge1] M. Gerstenhaber, On dominance and varieties of commuting matrices. Ann. of Math. (2) 73 1961
324348.
[GG] W.L. Gan, V. Ginzburg: Almost-commuting variety, D-modules, and Cherednik algebras. With an
appendix by Ginzburg. IMRP Int. Math. Res. Pap. 2006, 26439, 154.
[GGOR] V. Ginzburg, N. Guay, E. Opdam, R. Rouquier: On the category O for rational Cherednik algebras.
Invent. Math. 154 (2003), no. 3, 617651.
[Gi1] V. Ginzburg, Non-commutative symplectic geometry, quiver varieties, and operads. Math. Res. Lett.
8 (2001), 377400.
[Gi2] V. Ginzburg: On primitive ideals. Selecta Math. (N.S.) 9 (2003), no. 3, 379407.
[GM] M. Geck, G. Malle: Reflection groups. A contribution to the Handbook of Algebra. arXiv:math/0311012.
[GS] I. Gordona, J.T. Stafford: Rational Cherednik algebras and Hilbert schemes. Advances in Mathematics
198 (2005) 222274.
[Gy] A. Gyoja: Modular representation theory over a ring of higher dimension with applications to Hecke
algebras. J. Algebra 174 (1995), no. 2, 553572.
[HC] Harish-Chandra: Invariant differential operators and distributions on a semisimple Lie algebra, Amer.
J. Math, 86 (1964), 534-564.
[HO] G.J. Heckman, E.M. Opdam Root systems and hypergeometric funtions, I. Compositio Math. 64 (1987),
no. 3, 329352.
[He1] G. J. Heckman: A remark on the Dunkl differential-difference operators. Harmonic analysis on reduc-
tive groups, Brunswick, ME, 1989; eds. W. Barker and P. Sally, Progr. Math., vol. 101, Birkhauser
Boston, Boston, MA, 1991, pp. 181191.
[He2] G.J. Heckman, An elementary approach to the hypergeometric shift operators of Opdam. Invent. Math.
103 (1991), no. 2, 341350.
[He3] G.J. Heckman: Root systems and hypergeometic functions, II. Compositio Math. 64 (1987), no. 3,
353373.
[Hu] J.E. Humphreys, Reflection Groups and Coxeter Groups. Cambridge Studies in Advanced. Mathemat-
ics, 29, Cambridge University Press (1990).
[Ki] A.A. Kirillov: Unitary representations of nilpotent Lie groups. (Russian) Uspehi Mat. Nauk 17 1962
no. 4 (106), 57110.
[KKS] D. Kazhdan, B. Kostant, S. Sternberg: Hamiltonian group actions and dynamical systems of Calogero
type. Comm.Pure Apll. Math., 31(1978), 481507.
[L] G. Lusztig: Affine Hecke algebras and their graded version. J. Amer. Math. Soc. 2 (1989), no. 3,
599635.
[LS] T. Levasseur, J. Stafford: Invariant differential operators and a homomorphism of Harish-Chandra. J.
Amer. Math. Soc. 8 (1995), 365372.
[Na] H. Nakajima, Lectures on Hilbert schemes of points on surfaces, University Lecture Series, 18, Provi-
dence, R.I., American Mathematical Society, 1999.
[NT] R. Nest, B. Tsygan, On the cohomology ring of an algebra. Advances in geometry, 337370, Progr.
Math., 172, Birkhauser Boston, Boston, MA, 1999.
[NoSt] M. Noumi, J.V. Stokman, Askey-Wilson polynomials: an affine Hecke algebraic approach. Laredo
Lectures on Orthogonal Polynomials and Special Functions, 111144, Adv. Theory Spec. Funct. Or-
thogonal Polynomials, Nova Sci. Publ., Hauppauge, NY, 2004.
[OP] M.A. Olshanetsky, A.M. Perelomov: Quantum integrable systems related to Lie algebras. Phys. Rep.
94 (1983), no. 6, 313404.
[Op1] E.M. Opdam: Some applications of hypergeometric shift operators. Invent. Math. 98 (1989), no. 1,
118.

91
[Op2] E.M. Opdam, Dunkl operators, Bessel functions and the discriminant of a finite Coxeter group. Com-
positio Math. 85 (1993), 333373.
[Op3] E. M. Opdam: Root systems and hypergeometric functions III. Compositio Math., 67(1988), 21-49.
[Op4] E. M. Opdam: Root systems and hypergeometric functions IV. Compositio Math., 67 (1988), 191-209.
[Pi] G. Pinczon, On two Theorems about Symplectic Reflection Algebras, Lett.Math.Phys., 22, issues 2-3,
2007, pp. 237-253.
[PP] A. Polishchuk, L. Positselski: Quadratic algebras. University Lecture Series, 37. American Mathemat-
ical Society, Providence, RI, 2005.
[Sa] S. Sahi, Nonsymmetric Koornwinder polynomials and duality. Ann. of Math. (2) 150 (1999), no. 1,
267282.
[St] J. Stokman: Koornwinder polynomials and affine Hecke algebras. Internat. Math. Res. Notices 2000,
no. 19, 10051042.
[VV] M. Varagnolo, E. Vasserot, Finite dimensional representations of DAHA and affine Springers fibers :
the spherical case. Duke Math. J. 147 (2009), no. 3, 439540.
[Wa] N. Wallach: Invariant differential operators on a reductive Lie algebra and Weyl group representations.
J. Amer. Math. Soc. 6 (1993), 779816.
[Wi] G. Wilson: Collisions of Calogero-Moser particles and an adelic Grassmannian. Inv. Math. 133 (1998),
141.

P.E.: Department of Mathematics, Massachusetts Institute of Technology, Cambridge,

MA 02139, USA
E-mail address: [email protected]

X.M.: Department of Mathematics, Massachusetts Institute of Technology, Cambridge,

MA 02139, USA
E-mail address: [email protected]

92