Upload
nguyen-dan
View
223
Download
0
Embed Size (px)
Citation preview
8/8/2019 bo de Co ban
1/38
The fundamental lemma
by Bill Casselman
8/8/2019 bo de Co ban
2/38
Contents
1. The naked lemma .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3. The program for this seminar .............. 14
4. Unramified groups ........................... 185. Satake transform and Hecke algebra . . . . . 22
6. The L- g r o u p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 87. Endoscopic groups . . . . . . . . . . . . . .. . . . . . .. . . . . 31
2
8/8/2019 bo de Co ban
3/38
1. The naked lemma
3
8/8/2019 bo de Co ban
4/38
Suppose G to be an unramified reductive group defined over
a p-adic field k, H an unramified endoscopic group of G. Thetwo have the same rank.
Given a strongly regular semi-simple element t of H, let Tt
be its centalizer in H, a maximal torus. There exists a specialembedding of Tt in G. The element t gives rise to an elementtG of G, which we assume to be strongly regular in G.
Define orbital integrals of a locally constant function of com-
pact support on any reductive groupfor any strongly regular
t let
(f, t) = G
(t) 11/2vol T(o)vol G(o) CG(t)\G f(g1tg)dg
dt .
4
8/8/2019 bo de Co ban
5/38
(f, t) =
G (t) 1
1/2
vol T(o)
vol G(o)
CG(t)\G
f(g1tg)dg
dt.
The ratio of volumes is to eliminate the effect of choice of
measures on G and T.
Orbital integrals have something to with fixed-points, since
CG(t) is the set of points on G fixed under conjugation by t.We shall see that this connection extends to something very
deep. The product is over roots
with respect toT
. Such
factors are familiar in fixed point formulas.
5
8/8/2019 bo de Co ban
6/38
The data determining H have something to do a set of ele-ments s in the torus
T in the L-group LG. They all give rise
to the same map from the stable conjugacy class of tG,constant on the ordinary conjugacy classes. The -orbitalintegrals are linear combinations of the ordnary orbital inte-
grals:
G,H(f, t) = tt
(t
)G(f, t
) .
If 1 this is the stable sum stG(f, t).
Stable conjugacy means conjugacy in the algebrac closure.
6
8/8/2019 bo de Co ban
7/38
Any element f of the Hecke algebra HG//G(o) gives riseto an fH in H(H//H(o)
.
Fundamental Lemma:
G/H(f, tG) = stH(f
H, t)
for all strongly G-regular semi-simple elements t of H.
Very roughly, this says that analysis on G invariant underconjugation can be reduced to stable analysis on G and allits endoscopic groups.
Implicit in this are assertions about character sums in G match-ing stable character sums in H.
7
8/8/2019 bo de Co ban
8/38
2. History
8
8/8/2019 bo de Co ban
9/38
Bill Casselman, Notes on p-adic spherical functions, written
as a preface to a new printing of Ian Macdonalds book on
spherical functions.
Mark Goresky, Robert Kottwitz, and Robert MacPherson, Ho-
mology of affine Springer fibres in the unramified case, avail-
able on the http://www.arxiv.org.
Tom Hales, The Fundamental Lemma for Sp(4), SLProceedingsof the American Mathematical Society 125 (1997), 301-308
http://www.arxiv.org/math/0312227v2/
Tom Hales, A statement of the Fundamental Lemma, avail-
able on the http://www.arxiv.org.
http://www.arxiv.org/math/0312227v2/
David Kazhdan and George Lusztig, Fixed point varieties on
affine flag manifolds, Israel Journal of Mathematics 62 (1988),129168; appendix by Joseph Bernstein
9
8/8/2019 bo de Co ban
10/38
Robert Edward Kottwitz, Calculation of some orbital integral-
s, pages 349362 in The zeta functions of Picard modular
surfaces, CRM.
Jean-Pierre Labesse and Robert Langlands, L-indistinguishabilityfor SL(2), Canadian Journal of Mathematics 31 (1979), pages
726785Robert P. Langlands, Les debuts dune formule des traces
stables, Publications mathematiques de lUniversit e Paris VII,1983
10
8/8/2019 bo de Co ban
11/38
Gerard Laumon, On the Fundamental Lemma for unitary group-
s, audio file at
http://www.fields.utoronto.ca/audio/02-03/shimura/laumon/
Ngo Bao Chau, Hitchin fibration and the Fundamental Lem-ma, talk at the IAS (October 9, 2006):
http://www.math.ias.edu/lg40/ngo.pdf
Jonathan Rogawski, Automorphic representations of unitarygroups in three variables, Princeton University Press, 1990.
Jonathan Rogawski and Don Blasius, Fundamental Lemmas
for U(3) and related groups, Pages 363394 in The zeta func-tions of Picard modular surfaces, CRM.
11
8/8/2019 bo de Co ban
12/38
The first result of this type was in the paper by Labesse and
Langlands on SL(2). It was first formulated in complete gen-erality in Langlands Paris VII booklet. Both of these were
motivated by the problems of stable conjugacy arising natu-rally in the trace formula.
It was proven for SL(3) by Kottwitz. He calculated orbital in-tegrals by means of geometry on the Bruhat-Tits building of
G, following Langlands work on base change.
The group SL(n) was dealt with by Waldspurger. Rogawskiand others proved it for U(3), Hales for Sp(4). I do not know
what techniques they used.
12
8/8/2019 bo de Co ban
13/38
A paper by Kazhdan & Lusztig, with an appendix by Bern-
stein, changed the situation drastically. In this it was pointedout that orbital integrals count points on algebraic varieties
over finite fields. Goresky, Kottwitz, and MacPherson used
this idea and formulated the Fundamental Lemma in terms of
equivariant cohomology. They proved it for unramified conju-
gacy classes.
Laumon, working at first by himself and then with Ngo, proved
it for unitary groups U(n). It appears that recently Ngo, usingsomewhat related techniques, has proved it in all cases.
All along, various important technical results and reductions
have been added by various people, including Labesse, Lang-lands, Shelstad, Kottwitz, Waldspurger, Hales, Cunningham,
Kazhdan, Lusztig, and Bernstein.
13
8/8/2019 bo de Co ban
14/38
3. The program for this seminar
14
8/8/2019 bo de Co ban
15/38
There are many relatively elementary items to be explained.
These include
the structure of unramified groups
the Hecke algebras
the L-group (which contains the torus T) the map from the Hecke algebra of G to that of H
characterization and properties of (unramified) endoscopic
groups stable conjugacy and Galois cohomology
the characters
orbital integrals and the geometry of buildings the work of Kazhdan, Lusztig & Bernstein relating orbitalintegrals to the geometry of the affine Grassmannian
15
8/8/2019 bo de Co ban
16/38
There are many nastier items to be explained. These include
why LH does not always embed into LG
specification of the embedding of TH into G (transfer fac-tors)
reduction of the general case to that in which f and fH areHecke units
relationship very generally between orbital integrals and
varieties over finite fields role of equivariant cohomology
Ngos use of what he calls the Hitchin fibration
the transfer from the geometric fields to p-adic groups the geometric version of the Satake transform in terms ofsheaves on the affine Grassmannian
16
8/8/2019 bo de Co ban
17/38
It would be nice to exhibit some applications, but thats too
much to expect.
17
8/8/2019 bo de Co ban
18/38
4. Unramified groups
18
8/8/2019 bo de Co ban
19/38
Split reductive groups over k are classified up to isomorphis-
m by root dataquadruples
(L,, L,)
where L is a lattice, a root system in L, to coroots inthe dual lattice L. Here L is to be identified with the groupof characters of a maximal split torus T in G. Given a Borelsubgroup B we get a based root datum
(L,, L,) .
The lattice L is the character group X
(T) = Hom(T,Gm),L the cocharacter group X(T) = Hom(Gm, T). We may
identify T(k) with Hom(L, k) or L C.
19
8/8/2019 bo de Co ban
20/38
An unramified group has two characterizations. (1) It is ob-
tained by base extension from a smooth reductive group
scheme over o. (2) It is one that splits over an unramfied ex-
tension of k. Essentially because reductive groups over finitefields possess Borel subgroups, so do G(o) and G. Let it beB, T a torus T contained in B, and A a maximal split torusin T. Such groups are classified through Galois descent the-
ory by quintuples(L,, L,, )
where is an automorphism of the based root datum, giving
rise to the Galois action on roots.
20
8/8/2019 bo de Co ban
21/38
The lattice L and determine the unramified torus T. The
embedding of A in T induces an isomorphism of A = A/A(o)with T = T/T(o). These may both be identified with X(A) =(L) via images of the generator of p.
The relative Weyl group is the subgroup of the split Weyl
group that takes X(A) to itself. Elements of W may alsobe characterized as elements of the split Weyl group fixed by
the Frobenius.
21
8/8/2019 bo de Co ban
22/38
5. Satake transform and Hecke algebra
22
8/8/2019 bo de Co ban
23/38
The Hecke algebra is the convolution ring HZG//G(o). offunctions on G of compact support, bi-invariant under G(o).If (, V) is an admissible representation of G, it is called un-ramfied if it subspace of vectors fixed by G(o) is not 0. A
function in the Hecke algebra acts by convolution on thissubspace.
23
8/8/2019 bo de Co ban
24/38
There is a well known way to obtain all such representations.
Let be a character of T/T(o) = A/A(o) with values in C,and let
I() = {f:G R | f(bg) = 1/2
(b)(b)f(g)}
where
1/2 : b | det Ad(b)|
is the modulus character of B. It is there because B is nota unimodular group. Because of the Iwasawa decomposition
G = B G(o), the subspace of vectors fixed by G(o) has
dimension one. The Hecke algebra acts on it by scalar multi-plication, so we get a ring homomorphism from H to C.
24
8/8/2019 bo de Co ban
25/38
Choose a basis (ti) for the lattice T/T(o). Characters of T
are parametrized by an array (si), where (ti) = si. Neces-sarily, each si is invertible. For a given f in H, the functiontaking f to as a function of is a polynomial in the vari-ables si . Let R be the ring of all such polynomials. The rep-
resentations I() and I(w) are generically isomorophic forw in the Weyl group. We therefore have a map from H to thering of invariants RW. It is an isomorphism.
25
8/8/2019 bo de Co ban
26/38
The points (si) make up a copy of (C)n, and can be seen
as the complex points on a torus T. It may be canonicallyidentified with Hom(L,C). The character group of T is thecocharacter group of T.
This torus sits in turn in a complex reductive group called theL-group.
26
8/8/2019 bo de Co ban
27/38
Let me summarize. If F is a field, T a torus defined over F,and t in T(F), then the map taking an algebraic character
to (t) identifies T(F) with Hom(X
(T), F|times
).An unramified character of T is a homomorphism from thelattice L = T/T(o) to C. It may be identified with the
complex points on a torus T whose character lattice is L,whereas the character lattice of T is L. Principal series areparametrized by W-orbits on T, and also by homomorphismsfrom H
G//G(o)
to C.
27
8/8/2019 bo de Co ban
28/38
6. The L-group
28
8/8/2019 bo de Co ban
29/38
The complex group dual to G is the reductive group G de-fined over C associated to the dual root datum(L,, L,) .
The automorphism corresponds to an automorphism of thisgroup preserving a Borel subgroup B containing T. Thence asemi-direct product LG =
G .
The lattice L is now the character group of a complex torusT. Unramified characters of T correspond to elements of T.If G is split and is trivial, a Weyl group orbit of unramifiedcharacters corresponds to a semi-simple conjugacy class in
the coset G. In the general case, it turns out that they corre-spond to conjugacy classes in the coset G .
29
3
8/8/2019 bo de Co ban
30/38
Sem
i-simpleconjugacyclassesinL
G
correspo
ndtoho
phis
msfromH
toC.
A
co
njugacyclassin
G
correspondstoa
homom
from
HtoC.S
odoesthecharacterofarepre
sentatio
LG.
The
integralH
eckealgeb
ramaybe
identified
withari
certainreprese
ntationsofLG
.
30
8/8/2019 bo de Co ban
31/38
7. Endoscopic groups
31
8/8/2019 bo de Co ban
32/38
An unramified endoscopic group of G is an unramified reduc-
tive group H with datum
(L,H, L,H, H)
subject to the condition that there exist s in T = Hom(L,C)with
(a) H = Ker(s) G
(b) H = w G for some w in W
(c) H(s) = s
The element s is by no means unique it is not actually partof the data describing an endoscopic group. (Maybe it should
be?)
32
8/8/2019 bo de Co ban
33/38
The group LH often embeds into LG. When it does, eachemi-simpe class in H givs rise to one of G , and themap backwards is the restriction of a representation of LG toone of LH, or in othr words a map from the Hecke algebra of
G to that of H.But sometimes LH does not embed into LG . . .
33
8/8/2019 bo de Co ban
34/38
Example. G = SL(2). The group is split, so G = I.
The lattices L and L may be identified with Z, with 1, with 2. Then T = Hom(L,C) is C.The only s annihilating any roots is s = 1. So H = if s = 1and all of
Gotherwise.
If w = 1, so is H. (c) makes is no condition on s. So thereare two cases: (i) s = 1 when H = G, and (ii) s = 1 when His the split one-dimensional torus.
If w = 1, then H is non-trivial, and the torus T is the norm-one subgroup of the multiplicative group of the unramified
quadratic extension. It must stabilize positive roots if there
are any, amd this does not happen, and s must be 1. Condi-tion (c) is vacuous. Therefore H itself is (iii the unramifiednon-split torus.
34
8/8/2019 bo de Co ban
35/38
Example. G = PGL(2). The group is again split and G = I.Here L = Z but = 2. So both s = 1 fix all roots.
When w = 1 there are again two possibilities, (i) the splittorus and (ii) PGL(2) itself. In the second case, both s = 1will be compatible.
If w = 1, then (c) tells us that w(s) = s, so s = 1 andH = G. But in this case w does not fix the positive roots,so this case is disallowed.
35
8/8/2019 bo de Co ban
36/38
Example. G = U(3)?
36
8/8/2019 bo de Co ban
37/38
It would be nice to have a simple algorithm for listing all pos-
sible endoscopic groups, along with information about em-beddings of LH into LG.
It is not s, but at most the W-orbit of s that matters. This or-bit corresponds to an unramified representation, and presum-
ably endoscopy says something about its character. What is
conjectured and what proven?
37
8/8/2019 bo de Co ban
38/38
38