Symmetry breaking and the Gross Prasad conjectures for...
Transcript of Symmetry breaking and the Gross Prasad conjectures for...
Symmetry breaking and the Gross Prasadconjectures for orthogonal groups
Puerto Rico, January 2014
Birgit Speh
Cornell University
1
I would like to discuss some work on the Gross Prasad conjec-
tures for tempered representations of real orthogonal groups and
also examples extensions of these conjectures to Arthur-Vogan
packets of orthogonal groups.
2
I would like to discuss some work on the Gross Prasad conjec-
tures for tempered representations of real orthogonal groups and
also examples extensions of these conjectures to Arthur-Vogan
packets of orthogonal groups.
G = SO(p, q), UG an irreducible unitary representation of G on
Hilbert space
H = S0(p − 1, q) ⊂ G, UH an irreducible unitary representations
of H on a Hilbert space
2
I would like to discuss some work on the Gross Prasad conjec-
tures for tempered representations of real orthogonal groups and
also examples extensions of these conjectures to Arthur-Vogan
packets of orthogonal groups.
G = SO(p, q), UG an irreducible unitary representation of G on
Hilbert space
H = S0(p − 1, q) ⊂ G, UH an irreducible unitary representations
of H on a Hilbert space
Gross Prasad conjectures concern
dim HomH(UG ⊗ UH ,C)
for tempered representations UG and UH.
2
HomH(UG, U∨H) = HomH(UG⊗UH ,C) is of interest for automor-
phic representations
3
HomH(UG, U∨H) = HomH(UG⊗UH ,C) is of interest for automor-
phic representations
Periods of automorphic representations
If πA is an automorphic representation of GA and f ∈ πA then∫HA
f(HA)dhA is defines a period of πA if it converges and in
particular an element in HomH(πA,C).
If πA =∏πν then it defines an H- invariant linear functional on
all representations πν i.e an element in HomH(πν, UH) for a trivial
representation UH.
3
HomH(UG, U∨H) = HomH(UG⊗UH ,C) is of interest for automor-
phic representations
Periods of automorphic representations
If πA is an automorphic representation of GA and f ∈ πA then∫HA
f(HA)dhA is defines a period of πA if it converges and in
particular an element in HomH(πA,C).
If πA =∏πν then it defines an H- invariant linear functional on
all representations πν i.e an element in HomH(πν, UH) for a trivial
representation UH.
The integral is related to special values of L-functions and thelinear invariant linear functionals are important in the relativetrace formula of Jacquet.
3
HomH(UG, UH) is also of interest in other areas of mathe-
matics
4
HomH(UG, UH) is also of interest in other areas of mathe-
matics
Analysis on G/H
We get a function on G/H by
v → T (U(g)v).
This function is an eigenfunction of the invariant differential op-
erators. Classical problem is determine the spectrum of the in-
variant differential operators on G/K. More generally determine
the spectrum of invariant operators on G/H where H is fix points
of an involution. Considered and solved by H.Schlichtkrull, E.
van den Ban, T. Oshima and many others ...
4
For real orthogonal groups:
5
For real orthogonal groups:
Sun-Zhou and Aizenbud -Gurevich proved that for irreduciblerepresentations
HomH(UG, UH) has at most dimension 1.
B.Gross and D.Prasad have a precise conjecture for the dimen-sion of HomH(UG, UH) if both UG and UH are tempered involvingarithmetic invariants (epsilon factors)
5
A special case :
DISCRETE Branching laws at the real place : We consider
irreducible unitary representations U of G, whose restriction U|Hto H is a direct sum of irreducible representations of H and
determine explicitly the irreducible representations of H in U|H.
6
A special case :
DISCRETE Branching laws at the real place : We consider
irreducible unitary representations U of G, whose restriction U|Hto H is a direct sum of irreducible representations of H and
determine explicitly the irreducible representations of H in U|H.
For tempered representations (discrete series) discrete branching
laws due to Gross-Wallach, T. Kobayashi confirm part of the
conjectures of Gross-Prasad,
6
A special case :
DISCRETE Branching laws at the real place : We consider
irreducible unitary representations U of G, whose restriction U|Hto H is a direct sum of irreducible representations of H and
determine explicitly the irreducible representations of H in U|H.
For tempered representations (discrete series) discrete branching
laws due to Gross-Wallach, T. Kobayashi confirm part of the
conjectures of Gross-Prasad,
Complete description of all discrete branching laws for unitary
representations with nontrivial cohomology for pairs G=SO(n+1,1),
H=SO(n,1) which are usually not tempered is due to T.Kobayashi,
Y.Oshima, but no arithmetic conjectures about them yet.
6
Symmetry breaking Integral operators and Differential op-
erators for real groups
They are explicit operators in HomH(UG, UH). For principal se-
ries representations this is an analysis problem related to the
construction of intertwining operators.
In this case the restriction of UG to H is not a discrete sum.
7
Symmetry breaking Integral operators and Differential op-
erators for real groups
They are explicit operators in HomH(UG, UH). For principal se-
ries representations this is an analysis problem related to the
construction of intertwining operators.
In this case the restriction of UG to H is not a discrete sum.
So we proceed as follows:
We consider the space of C∞-vectors of UG and UH. It is a
Frechet space and UG and UH act continuously on this Frechet
space and we obtain the smooth representations U∞G and U∞H .
7
Then
8
Then
HomH(UG, UH)
denotes the space of continuous (with respect to the Frechet
topology) intertwining operators
8
Then
HomH(UG, UH)
denotes the space of continuous (with respect to the Frechet
topology) intertwining operators
We have
HomH(UG ⊗ U∨H ,C) = HomH(UG, UH)
8
Results about symmetry breaking operators:
Let P = MAN and P ′ = M ′A′N ′ be Langlands decompositions
of parabolic subgroups G = SO(n + 1,1) and G′ = SO(n,1),
respectively, satisfying
M ′ = M ∩G′, A′ = A ∩G′, N ′ = N ∩G′, P ′ = P ∩G′.
Fix finite-dimensional representations
σ : M → GLC(V )
and
τ : M ′ → GLC(W ).
9
For λ ∈ a∗C and ν ∈ (a′C)∗, we define Vλ of P and Wν of P ′ by
P = MAN 3 meHn 7→ σ(m)eλ(H) ∈ GLC(V )
P ′ = M ′A′N ′ 3 m′eH′n′ 7→ τ(m′)eν(H ′) ∈ GLC(W ),
Consider the induced representations
I(P, Vλ) on the C∞-sections of Vλ := G×P Vλ.
I(P ′,Wν) on the C∞-sections of Wν := G×P Wν.
10
For λ ∈ a∗C and ν ∈ (a′C)∗, we define Vλ of P and Wν of P ′ by
P = MAN 3 meHn 7→ σ(m)eλ(H) ∈ GLC(V )
P ′ = M ′A′N ′ 3 m′eH′n′ 7→ τ(m′)eν(H ′) ∈ GLC(W ),
Consider the induced representations
I(P, Vλ) on the C∞-sections of Vλ := G×P Vλ.
I(P ′,Wν) on the C∞-sections of Wν := G×P Wν.
A example of a symmetry breaking operator in
HomH(I(P, Vλ)), I(P ′, (Vλ)|P ′) is the restriction of a section f in
I(P, Vλ) to the subvariety G′/P ′.
10
Consider now the case V = W trivial and the spherical principal
series representations I(λ) and J(ν) of O(n+1,1) and O(n,1)
using normalized induction.
11
Consider now the case V = W trivial and the spherical principal
series representations I(λ) and J(ν) of O(n+1,1) and O(n,1)
using normalized induction.
In joint work with T. Kobayashi
dim HomH(I(λ), J(ν)) ≤ 2
11
Consider now the case V = W trivial and the spherical principal
series representations I(λ) and J(ν) of O(n+1,1) and O(n,1)
using normalized induction.
In joint work with T. Kobayashi
dim HomH(I(λ), J(ν)) ≤ 2
If I(λ) and I(ν) are tempered then dim HomH(I(λ), J(ν)) = 1.
11
Remarks about the (symmetry breaking) intertwining operators
This G′ intertwining operator has similar form to the usual in-
tertwining operator. Consider the spherical principal series rep-
resentations as functions on the large Bruhat cell N ∼ Rn.
12
Remarks about the (symmetry breaking) intertwining operators
This G′ intertwining operator has similar form to the usual in-
tertwining operator. Consider the spherical principal series rep-
resentations as functions on the large Bruhat cell N ∼ Rn.
Theorem 1. Joint with T.KobayashiThere exists an analytic family
Aλ,ν ∈ HomG′(I(λ), J(ν)
of symmetry breaking operators with the distribution kernel
π∗
L(1/2, eλ1+ν1)L(1/2, eλ1−ν1)|xn|λ+ν−1/2(|x|2 + x2
n)−ν−(n−1)/2.
12
Remark:
If the representations are tempered then the normalization factor
is closely related to the
L-function of the representation I(λ)× J(ν).
13
Remark:
If the representations are tempered then the normalization factor
is closely related to the
L-function of the representation I(λ)× J(ν).
In this case λ and ν are are purely imaginary and I(λ) × J(ν)
and I(−λ) × J(−ν) are isomorphic and the L-function of the
representation I(λ)× J(ν) contains the factor
L(1/2, eλ1+ν1)L(1/2, eλ1−ν1)L(1/2, e−λ1+ν1)L(1/2, e−λ1−ν1)
The L-factor doesn’t have a pole and so the intertwining operator
is nonzero for tempered representations.
13
Outline of ”a naive version” of the Gross Prasad conjectures for
orthogonal groups
Idea
Consider the complex dual group L(G)C of G. The tempered
representations of orthogonal groups are parametrized by homo-
morphisms τ of the Weil group WR into L(G)C.
14
Outline of ”a naive version” of the Gross Prasad conjectures for
orthogonal groups
Idea
Consider the complex dual group L(G)C of G. The tempered
representations of orthogonal groups are parametrized by homo-
morphisms τ of the Weil group WR into L(G)C.
Unfortunately if we look at a homomorphism
τ : WR →L (G)Cwe cannot easily read of the inner form for which this definesa representation. It defines packet AG(τ) of representations ofseveral forms of orthogonal groups.
Similarly we can define a packet AH(τH).
14
We now to define a subset A of compatible representations of
AG(τ)×AH(τH).
15
We now to define a subset A of compatible representations of
AG(τ)×AH(τH).
Gross Prasad Conjecture (simple version)
There exists exactly one pair πG × πH ∈ A so that
HomH(πG, πH) = 1
15
We now to define a subset A of compatible representations of
AG(τ)×AH(τH).
Gross Prasad Conjecture (simple version)
There exists exactly one pair πG × πH ∈ A so that
HomH(πG, πH) = 1
Furthermore an ε-factors determines the pair πG × πH.
15
A simple first example:
G = SO(2,1) = GL(2,R)/D+, H = SO(2). Every representation
UH of H is tempered and are parametrized by the even integers.
The tempered representations of G are spherical principal series
representations and discrete series representations.
16
A simple first example:
G = SO(2,1) = GL(2,R)/D+, H = SO(2). Every representation
UH of H is tempered and are parametrized by the even integers.
The tempered representations of G are spherical principal series
representations and discrete series representations.
If UG is a spherical principal series representation and UH is any
representations of H, then the set A = {UG × UH} satisfies the
G-P conjecture.
16
A simple first example:
G = SO(2,1) = GL(2,R)/D+, H = SO(2). Every representation
UH of H is tempered and are parametrized by the even integers.
The tempered representations of G are spherical principal series
representations and discrete series representations.
If UG is a spherical principal series representation and UH is any
representations of H, then the set A = {UG × UH} satisfies the
G-P conjecture.
If π is a discrete series representation and Fπ the finite dimen-sional representation with the same infinitesimal character as πthen for every representation UH the set
A = {π × UH , Fπ × UH}satisfies the G-P conjecture.
16
Problem: Fπ is not tempered
17
Problem: Fπ is not tempered
Solution: Consider tempered representations of SO(2,1)×SO(2)
and of SO(3)×SO(2) instead and consider the final dimensional
representation Fπ as a representation of SO(3). . The set Fnow contains 2 irreducible tempered representations of SO(2,1)×SO(2) and one representation of SO(3)× SO(2) and
⊕U∈F dim HomH(U,C) = 1
17
Second Example
Assume that both SO(p,q) and SO(p-1,q) has discrete series
representations
In this case A = AG(τ)×AH(τH)
Gross-Wallach proved about 20 years ago that under some ad-
ditional assumption on the function there exists a pair πG, πHAso that
HomH(πG, πH) = 1
18
Second Example
Assume that both SO(p,q) and SO(p-1,q) has discrete series
representations
In this case A = AG(τ)×AH(τH)
Gross-Wallach proved about 20 years ago that under some ad-
ditional assumption on the function there exists a pair πG, πHAso that
HomH(πG, πH) = 1
Open Problem: There are no other such pairs in A.
18
I am interested here in the case of tempered representations
which are not discrete series representations and also to
see if there is an extension of this conjecture to cohomological
relevant representations.
19
I am interested here in the case of tempered representations
which are not discrete series representations and also to
see if there is an extension of this conjecture to cohomological
relevant representations.
To state the conjecture I have to introduce Vogan ”packets” ofrepresentations of a collection of orthogonal groups . We needto discuss
• Inner forms of orthogonal groups
• Relevant inner forms (and relevant parabolic subgroups)
• L-groups and Vogan packets of representations
19
Inner forms:
We consider the families of real forms of SO(p+q,C). We are
interested in families which contain an orthogonal group of real
rank 1. ,
20
Inner forms:
We consider the families of real forms of SO(p+q,C). We are
interested in families which contain an orthogonal group of real
rank 1. ,
First case
p+q =2n+1, Consider G0 = SO(n,n+1), i.e that the form defin-
ing G0 has signature n,n+1
Consider the family SO(n-2r, n+1+2r) for an integer r with |r| ≤n Note this family contains the pure inner form G0, a compact
form is always in this family and either SO(2n,1) or S0(1,2n) is
in this family.
20
Second case:
p+q =2n ∼= 0 mod 4. We may take G0 = SO(n,n), i.e the
form defining GO has signature n,n. consider the family S0(n-2r,
n+2r) |r| ≤ n. Note the family contains G0, the compact form
is always in this family and that SO(2n-1,1) or S0(1,2n-1) is not
in this family,
Since we are interested in families which contain rank one or-
thogonal groups we can ignore this family.
21
Second case:
p+q =2n ∼= 0 mod 4. We may take G0 = SO(n,n), i.e the
form defining GO has signature n,n. consider the family S0(n-2r,
n+2r) |r| ≤ n. Note the family contains G0, the compact form
is always in this family and that SO(2n-1,1) or S0(1,2n-1) is not
in this family,
Since we are interested in families which contain rank one or-
thogonal groups we can ignore this family.
Instead we consider G0 =SO(n-1,n+1) and the family S0(n-2-1r,
n+2r+1) |r| ≤ n which contains SO(2n-1,1) or S0(1,2n-1).
21
Third case:
p+q =2n 6∼= 0 mod 4. We may take G0= SO(n,n). We consider
the family S0(n-2r, n+2r) |r| ≤ n. Note the compact form is
not in this family and either SO(2n-1,1) or S0(1,2n-1) is in this
family.
22
Third case:
p+q =2n 6∼= 0 mod 4. We may take G0= SO(n,n). We consider
the family S0(n-2r, n+2r) |r| ≤ n. Note the compact form is
not in this family and either SO(2n-1,1) or S0(1,2n-1) is in this
family.
Relevant families of pairs of inner forms
Consider a family G of real forms (G,H) of SO(n+1,C)×SO(n,C)
where both groups are in one of the families defined above.
22
Third case:
p+q =2n 6∼= 0 mod 4. We may take G0= SO(n,n). We consider
the family S0(n-2r, n+2r) |r| ≤ n. Note the compact form is
not in this family and either SO(2n-1,1) or S0(1,2n-1) is in this
family.
Relevant families of pairs of inner forms
Consider a family G of real forms (G,H) of SO(n+1,C)×SO(n,C)
where both groups are in one of the families defined above.
We are interested in the situation H ⊂ G. So we say that a pairin G is relevant if this is satisfied.We denote this subset of G by Gr.
22
Examples of relevant families of pairs of inner forms;
First example:
Gr={ S0(n-2r,n+1+2r), Hr=S0(n-2r,n+2r) }
Remark: If n is odd then this family does contains a pair of rank
one orthogonal groups
23
Examples of relevant families of pairs of inner forms;
First example:
Gr={ S0(n-2r,n+1+2r), Hr=S0(n-2r,n+2r) }
Remark: If n is odd then this family does contains a pair of rank
one orthogonal groups
Second example :
Gr={ S0(n-2r,n+1+2r), Hr=S0(n-2r-1,n+2r +1) }
Remark: n is even then this family does contain a pair of rank
one orthogonal groups
23
Third example:
Gr={ SO(n+2r,n-2r), Hr = S0(n-1+2r, n-2r) }
Remark: If n is odd then this family contains a pair of rank one
orthogonal subgroups.
24
Third example:
Gr={ SO(n+2r,n-2r), Hr = S0(n-1+2r, n-2r) }
Remark: If n is odd then this family contains a pair of rank one
orthogonal subgroups.
A choice of compatible maximal parabolic subgroups PG, PH with
PH ⊂ PG in the rank one pairs in Gr defines pairs of relevant max-
imal parabolic subgroups for all members of a family of relevant
inner forms.
24
An example:
Consider the case SO(4,1) ,SO(3,1)
25
An example:
Consider the case SO(4,1) ,SO(3,1)
Relevant pairs of inner forms:
G ={ ( SO(4,1) , SO(3,1)) and (SO(2,3), SO(1,3))}
25
An example:
Consider the case SO(4,1) ,SO(3,1)
Relevant pairs of inner forms:
G ={ ( SO(4,1) , SO(3,1)) and (SO(2,3), SO(1,3))}
Relevant parabolic subgroups have Levi subgroups
(SO(3)R∗, S0(2,1)R∗) and (SO(1,2)R∗,SO(2)R∗).
25
The Langlands dual group of SO(4,1) is Sp(4,C)
The Langlands dual group of SO(3,1) is SO(4,C)
26
The Langlands dual group of SO(4,1) is Sp(4,C)
The Langlands dual group of SO(3,1) is SO(4,C)
The Weil group WR is a non split extension of the Galois group
{1, τ} by C∗.
26
The Langlands dual group of SO(4,1) is Sp(4,C)
The Langlands dual group of SO(3,1) is SO(4,C)
The Weil group WR is a non split extension of the Galois group
{1, τ} by C∗.
Consider 4 dimensional symplectic and orthogonal representa-
tions of WR defining a tempered representations of SO(4,1)×SO(3,1)
induced from relevant parabolic subgroups. These representa-
tions of WR define also tempered representations of SO(2,3)×SO(1,3).
26
The Langlands dual group of SO(4,1) is Sp(4,C)
The Langlands dual group of SO(3,1) is SO(4,C)
The Weil group WR is a non split extension of the Galois group
{1, τ} by C∗.
Consider 4 dimensional symplectic and orthogonal representa-
tions of WR defining a tempered representations of SO(4,1)×SO(3,1)
induced from relevant parabolic subgroups. These representa-
tions of WR define also tempered representations of SO(2,3)×SO(1,3).
We obtain a set A = with 2 representations of groups G× H
26
Gross Prasad conjectures in this case:
If L(1/2, τsp× τorth) 6= 0 exactly one of these representations has
an H-invariant linear functional
Theorem 2. Consider a family G which contains the pair G=SO(4,1),
H=SO(3,1) and a family A of representations of the groups in
G so that the tempered spherical principal series representations
of G× H are in A. This family satisfies the G-P conjecture
27
Theorem 3. Consider a family G which contains the pair G
=SO(5,1) , H=SO(4,1) and and a family A of representations
of the groups in G so that the tempered spherical principal series
representation on G × H are in A. This family A satisfies the
G-P conjectures.
28
Theorem 3. Consider a family G which contains the pair G
=SO(5,1) , H=SO(4,1) and and a family A of representations
of the groups in G so that the tempered spherical principal series
representation on G × H are in A. This family A satisfies the
G-P conjectures.
A consequence of the work with T. Kobayashi
Theorem 4. Consider a family G which contains the pairs G
=SO(n+1,1) , H=SO(n,1) and a family A of the groups in Gso that the tempered spherical principal series representations of
G×H are in A. This family contains at least one representation
with an H invariant linear functional.
28
Is is possible to generalize this idea to representations which are
not tempered?
29
Is is possible to generalize this idea to representations which are
not tempered?
Possible interesting families :
Find families of representations of cohomologically induced rep-
resentations Aq(λ) (generalization of Adams/Johnson packets)
29
Is is possible to generalize this idea to representations which are
not tempered?
Possible interesting families :
Find families of representations of cohomologically induced rep-
resentations Aq(λ) (generalization of Adams/Johnson packets)
Families of representations which contain representations of arith-
metic interest, i.e. reps in Arthur packets
29
Some experimental results:
Consider the complementary series representation I(3) of SO(5,1)
which is isomorphic to a subrepresentation of L2(SO(5,1)/SO(4,1)).
The complementary series representation I(2) of SO(4,1) is iso-
morphic to a subrepresentation of L2(SO(4,1)/SO(3,1)). For
these orthogonal groups the Ramanujan conjecture doesn’t hold
and so these representation are infinity components of automor-
phic representations.
The infinitesimal character of these representations is singular
and there are only 2 irreducible unitary representations in the
family A which contains I(3)×I(2).
30
Some experimental results:
Consider the complementary series representation I(3) of SO(5,1)
which is isomorphic to a subrepresentation of L2(SO(5,1)/SO(4,1)).
The complementary series representation I(2) of SO(4,1) is iso-
morphic to a subrepresentation of L2(SO(4,1)/SO(3,1)). For
these orthogonal groups the Ramanujan conjecture doesn’t hold
and so these representation are infinity components of automor-
phic representations.
The infinitesimal character of these representations is singular
and there are only 2 irreducible unitary representations in the
family A which contains I(3)×I(2).
Joint work with Kobayashi implies In this packet G-P is true.
30
We consider the unitary infinite dimensional subquotients TG and
TH of the spherical principal series representations of SO(5,1)
respectively SO(4,1) with infinitesimal character ρ. These rep-
resentations are automorphic representations with nontrivial co-
homology, are holomorphically induced from a θ stable parabolics
with L = T1SO(3,1) respectively L = T1SO(2,1).
31
We consider the unitary infinite dimensional subquotients TG and
TH of the spherical principal series representations of SO(5,1)
respectively SO(4,1) with infinitesimal character ρ. These rep-
resentations are automorphic representations with nontrivial co-
homology, are holomorphically induced from a θ stable parabolics
with L = T1SO(3,1) respectively L = T1SO(2,1).
Joint with Kobayashi
dim HomSO(4,1)(TG, TH) = 1
TG and TH are both members of an enlarged Arthur packet Acohoof representations in which G-P holds.
31