Weyl group multiple Dirichlet series

Paul E. Gunnells
UMass Amherst

August 2010

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

1 / 33

Basic problem

Let be an irreducible root system of rank r. Our goal: explain general construction of multiple Dirichlet series in r complex variables s = (s1 , . . . , sr ) Z(s) =
c1 ,...,cr

a(c1 , . . . , cr ) cs1 . . . csr r 1

satisfying a group of functional equations isomorphic to the Weyl group W of . The functional equations intermix all the variables, and are closely related to the usual action of W on the space containing .

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

2 / 33

Example
2 Let = A2 , W = 1 , 2 | i = 1, 1 2 1 = 2 1 2 . The desired functional equations look like

1 : s1 2 s1 , s2 s1 + s2 1,
s1 =1 e 1

2 : s1 s1 + s2 1, s2 2 s2
s1 =s2

s2 =1 2

s1 +2s2 =3

12 21 121 s1 +s2 =2

2s1 +s2 =3
Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series August 2010 3 / 33

Why?
Such series provide tools for certain problems in analytic number theory (moments, mean values, . . . ). Conjecturally these series arise as FourierWhittaker coecients of Eisenstein series on metaplectic groups 1 n G(AF ) G(AF ) 1 This has been proved in some cases (type A and type B (double covers)). The series are built out of arithmetically interesting data, such as Gauss sums, nth power residue symbols, Hilbert symbols, and (sometimes) L-functions. The objects that arise in the construction have interesting relationships with combinatorics, representation theory, and statistical mechanics.
Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series August 2010 4 / 33

Maass and the half-integral weight Eisenstein series

Let E (z, s) be the half-integral weight Eisenstein series on 0 (4): E (z, s) =
\0 (4)

j1/2 (, z)1 (z)s/2 .

Maass showed that its dth Fourier coecient is essentially L(s, d ), where d is the quadratic character attached to Q( d/Q). Essentially means up to the Euler 2-factor, archimedian factors, and certain correction factors that have to be inserted when d isnt squarefree.

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

5 / 33

Siegel, GoldfeldHostein
Siegel (1956), GoldfeldHostein (1985):

Z(s, w) =
0

(E (iy, s/2) const term) y w

dy . y

The result is a Dirichlet series roughly of the form Z(s, w) L(s, d ) . dw

This behaves well in s since its built from the Dirichlet L-functions, and it turns out to have nice analytic properties in w as well. GoldfeldHostein used this to get estimates for sums like L(1, d ),
|d|<X d fund. |d|<X d fund.

1 L( , d ). 2

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

6 / 33

Siegel, GoldfeldHostein
Z(s, w) satises a functional equation in s, again because of the Dirichlet L-functions. But it turns out that it satises extra functional equations. In fact, Z satises a group of 12 functional equations, and is an example of a Weyl group multiple Dirichlet series of type A2 . There is a subgroup of functional equations isomorphic to S3 = W (A2 ), and an extra one swapping s and w that corresponds to the outer automorphism of the Dynkin diagram: s w

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

7 / 33

Connection to A2

Why is this series related to root system A2 (besides the fact that there are two variables and I drew the picture that way)? Imagine expanding the L-functions in the rough denition: Z(s, w) =
d

L(s, d ) = dw

dw
d d c c

d s c = c

d,c

d s w c d . c

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

8 / 33

The general shape

Heuristically, the multiple Dirichlet series looks like Z(s) =
c1 ,...,cr

a(c1 , . . . , cr ) cs1 . . . csr r 1

where a(c1 , . . . , cr ) is a product of nth power residue symbols corresponding to the edges of the Dynkin diagram. For instance D4 , n = 2 leads to a series related to the third moment of quadratic Dirichlet L-functions. c1 c4 c3

c2
Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series August 2010 9 / 33

Setup

number eld with 2nth roots of unity

S set of places of F containing archimedian, ramied, and such that OS is a PID irreducible simply-laced root system of rank r {1 , . . . , r } the simple roots s = (s1 , . . . , sr )

m = (m1 , . . . , mr ) r-tuple of integers in OS

r-tuple of complex variables

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

10 / 33

Setup

FS =

vS

Fv

M() certain nite-dimensional space of complex-valued functions on (FS )r (to deal with Hilbert symbols and units) H(c; m) to be specied later . . . this is the most important part of the denition M()

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

11 / 33

The multiple Dirichlet series

Then the multiple Dirichlet series looks like Z(s; m, ; , n) =

c

H(c; m) (c) , |ci |si

{0})/OS .

where c = (c1 , . . . , cr ) and each ci ranges over (OS

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

12 / 33

The function H
The coecients H have to be carefully dened to guarantee that Z satises the desired group of functional equations. General considerations tell us how to dene H in the following cases: When c1 cr and c c are relatively prime, one uses a r 1 twisted multiplicativity to construct H(cc ; m) from H(c; m) and H(c ; m). One puts H(cc ; m) = (c, c )H(c; m)H(c ; m), where (c, c ) is a root of unity built out of residue symbols and root data: r ci c ci c i i . (c, c ) = c ci c cj i j
i=1 ij

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

13 / 33

The function H

When (c1 cr , m m ) = 1, we can dene H(c; mm ) in terms r 1 of H(c; m) and certain power residue symbols:
r

H(c; mm ) =
j=1

m j H(c; m) cj

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

14 / 33

The function H
So we reduce the denition of H to that of H(k1 , . . . , kr ; l1 , . . . , lr ), where is a prime in OS . This leads naturally to the generating function N = N (x1 , . . . , xr ) =
k1 ,...,kr 0

H(k1 , . . . , kr ; l1 , . . . , lr )xk1 xkr r 1

(m is xed). One can ask what properties this series has to satisfy so that one can prove Z satises the right group of functional equations.

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

15 / 33

The function N

N = N (x1 , . . . , xr ) =
k1 ,...,kr 0

H(k1 , . . . , kr )xk1 xkr . r 1

If one puts xi = q si , where q = |OS /|, then one can see that the global functional equations imply N must transform a certain way under a certain W -action. This leads to a connection with characters of representations of g, the simple complex Lie algebra attached to . In this relationship the monomials correspond to certain weight spaces.

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

16 / 33

Building N
The connection with characters leads to two approaches to dening N : Crystal graphs. These are models for g representations that have various combinatorial incarnations (GelfandTsetlin patterns, tableaux, Proctor patterns, Littlemann path model, . . . ). One tries to extract a statistic from the combinatorial model to dene the coecients of N . (BrubakerBumpFriedberg, BeinekeBrubakerFrechette, ChintaPG) Weyl character formula. This is an explicit expression for a given character as a ratio of two polynomials. We take this approach and dene a deformation of Weyls formula that reects the metaplectacity (metaplectaciousness?) of the setup. (ChintaPG, BucurDiaconu)

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

17 / 33

WCF
w = weight lattice of

{1 , . . . , r } fundamental weights i the Weyl vector 1 1 Z[y1 , . . . , yr ] group ring a dominant weight of the weight lattice (yi i )

Then according to Weyl the character of the irreducible representation of highest weight is = (y) =
wW

sgn(w)yw(+) = ) >0 (1 y

sgn(w)yw(+)
wW

1 . (y)

>0 (1

y ).
August 2010 18 / 33

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

Deformation of the WCF

Our goal now is to dene the W -action leading to H. For the application to multiple Dirichlet series, we normalize things slightly dierently. Thus we work with the root lattice, introduce some q = |OS /| powers, shift the character around, . . . root lattice of d : Z height function on the roots

= + li i a strictly dominant weight (recall that were dening H(c; m) when m = (l1 , . . . , lr ))

A C[x1 , . . . , x1 ] complex group ring of (xi i ) r 1 C(x1 , . . . , xr ) fraction eld of A A

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

19 / 33

The action on monomials

We let the Weyl group act on monomials through a change of variables map. This is essentially the same as the geometric action of W on the root lattice (except for the q power). If f (x) = x , we put f (wx ) = q d(w
1 )

xw

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

20 / 33

Ane action of W

Given any , we put w = w( ) + , where the action on the right hand side is the usual action on the root lattice. This just performs an ane reection of R (the same as the usual w reection but shifted to have center ). If i is a simple reection, we put i () = d(i ). This is just the multiple of i needed to go from to i .

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

21 / 33

Ane action of W

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

22 / 33

Gauss sums

Choose some complex numbers (i), i = 1, . . . , n 1 such that (i)(n i) = 1/q. Put (0) = 1. Ultimately these numbers will be Gauss sums (the same ones appearing in the metaplectic cocycle), but actually any complex numbers satisfying these relations will work. Extend (i) to all integers by reducing i mod n.

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

23 / 33

Homogeneous decomposition
The action on a monomial f (x) = x depends on the congruence class of the monomial mod n. To treat general rational functions, we decompose A into homogeneous parts A . A=
/n

A consists of those rational function f /g where all monomials in g lie in n and those in f map to modulo n. e.g., 1 xy 1 xy = 2 x2 y 2 x y 2 x2 y 2

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

24 / 33

Finally

Theorem (ChintaPG) Suppose f A . Let i be a simple reection and let (k)n be the remainder upon division of k by n. Then (f | i )(x) = (qxi )li +1(i ())n 1 1/q f (i x) 1 q n1 xn i 1 (qxi )n f (i x) (i ()) (qxi )li +1n (1 q n1 xn ) i (P ) (Q)

extends to a W -action on C(x1 , . . . , xr ).

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

25 / 33

The W -action

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

26 / 33

The W -action

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

27 / 33

The W -action

i x

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

28 / 33

The W -action

Q P

i x

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

29 / 33

Making the multiple Dirichlet series

Theorem (ChintaPG) Put (x) = Then
>0 (1

q n xn ) and D(x) = h(x) =

wW

>0 (1

q n1 xn ).

(1| w)(x) (wx)

is a rational function such that hD is a polynomial. Let N = hD, dene H by N=

k1 ,...,kr 0

H(k1 , . . . , kr ; l1 , . . . , lr )xk1 xkr , r 1

and specialize the (i) to the appropriate Gauss sums. Then the resulting multiple Dirichlet series Z(s; m, ; , n) has analytic continuation to Cr and satises a group of functional equations isomorphic to W .
Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series August 2010 30 / 33

A2 examples (n = 2)
Here g1 = q(1) and the notation (a, b) means = (a + 1)1 + (b + 1)2 . (0, 0): 1 + g1 x + g1 y g1 qx2 y g1 qxy 2 q 2 x2 y 2

(1, 0): 1 qx2 + g1 y g1 qx2 y + g1 q 2 x2 y + q 3 x3 y g1 q 3 x2 y 3 q 4 x3 y 3 (2, 1): 1qx2 +q 2 x2 +g1 q 2 x3 qy 2 +q 2 x2 y 2 2q 3 x2 y 2 +q 4 x2 y 2 g1 q 3 x3 y 2 + g1 q 4 x3 y 2 +q 4 x4 y 2 q 5 x4 y 2 g1 q 5 x5 y 2 +q 4 x2 y 4 q 5 x2 y 4 g1 q 5 x3 y 4 + g1 q 6 x3 y 4 q 5 x4 y 4 + q 6 x4 y 4 + g1 q 6 x5 y 4 g1 q 7 x5 y 4 + q 7 x3 y 5 q 8 x5 y 5 (1, 1): 1 qx2 qy 2 + q 2 x2 y 2 q 3 x2 y 2 + q 4 x4 y 2 + q 4 x2 y 4 q 5 x4 y 4

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

31 / 33

Open questions

The WCF method works for all , whereas the crystal graph approach has only been worked out for some (classical) . Can one do the latter for all uniformly? (KimLee, McNamara) Prove that Z is a Whittaker coecient of a metaplectic Eisenstein series. (ChintaOen) Prove that the crystal graph descriptions and the WCF descriptions coincide. (ChintaOen + McNamara) Develop multiple Dirichlet series on ane Weyl groups and crystallographic Coxeter groups (BucurDiaconu, Lee) What is the geometric interpretation of Weyl group multiple Dirichlet series over function elds?

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

32 / 33

References

Gautam Chinta and PG, Weyl group multiple Dirichlet series constructed from quadratic twists, Invent. Math. 167 (2007), no.2, 327353. Gautam Chinta, Sol Friedberg, and PG, On the p-parts of quadratic Weyl multiple Dirichlet series, J. Reine Angew. Math. 623 (2008), 123. Gautam Chinta and PG, Weyl group multiple Dirichlet series of type A2 , to appear in the Lang memorial volume. , Constructing Weyl group multiple Dirichlet series, J. Amer. Math. Soc. 23 (2010), 189215.

Paul E. Gunnells (UMass Amherst) Weyl group multiple Dirichlet series

August 2010

33 / 33