CONJECTURE
ATSUSHI ICHINO AND TAMOTSU IKEDA
1. Formulation of the conjecture
We would like to formulate a relation between a certain period of automorphic forms on orthogonal group and some L-value. Our con- jecture can be considered as a refinement of global Gross-Prasad con- jecture [9].
Letk be a global field with char.k = 2. Let (V1, Q1) and (V0, Q0) be quadratic forms with rank n+ 1 and n, respectively. In the following, the subscriptiis either 0 or 1. We denote the special orthogonal group of (Vi, Qi) by Gi. We assume there is an embeddingι:V0 →V1. Then we have an embedding of corresponding orthogonal groups ι : G0 → G1. When n= 2, we assume V0 is not split.
Let πi ⊗vπi,v be an irreducible square integrable automorphic representation of Gi. There is a canonical inner product ∗,∗ on πi defined by
ϕi, ϕi=
Gi(k)\Gi( )
ϕi(gi)ϕi(gi)dgi,
where dgi is the Tamagawa measure on Gi(A). We choose a Haar measure dgi,v for each v. There exist positive number Ci such that dgi = Ci
vdgi,v. The character χQi is the quadratic character of A×k/k× corresponding to k(
(−1)mdetQi)/k.
Put ΔG1,v =
ζv(2)ζv(4)· · ·ζv(2m), dimkV1 = 2m+ 1, ζv(2)ζv(4)· · ·ζv(2m−2)·Lv(m, χQ1), dimkV1 = 2m, ΔG1 =
v
ΔG1,v
Choose decomposable vectors ϕi = ⊗vϕi,v ∈ ⊗vπi,v. Since πi,v is a unirary representation, there is a norm ∗ on πi,v for any place v of k.
We are interested in the periodϕ1|G0, ϕ0whereϕ1|G0 is the restric- tion of ϕ1 toG0(A).
1
The L-group LGi of Gi is a semi-direct product ˆGiWk. Here, Wk is the Weil group of k and
Gˆi =
SO(m,C) if m = dimVi is even, Sp(m−1)/2(C) if m = dimVi is odd.
We denote by st the standard representation of LGi. The (complete) standardL-function is denoted byL(s, πi,st). For simplicity, we some- times denote L(s, πi,st) by L(s, πi). For v /∈ S, the Euler factor for L(s, πi,st) is given by det(1−st(Aπi,v)·qv−s)−1. Here, Aπi,v is the Sa- take parameter of πi,v. We consider the tensor product L-function L(s, π1 ⊗π0). The Euler factor of L(s, π1⊗π0) for v /∈ S is given by det(1−st1(Aπ1,v)⊗st0(Aπ0,v)·qv−s)−1, where st1 and st0 are the standard representations of LG1 and LG0, respectively.
Consider the adjoint representation Ad : LGi →GL(Lie(LGi)). The associated L-functionL(s, πi,Ad) is called the adjointL-function. We put
Pπ1,π0(s) = L(s, π1π0)
L(s+ 12, π1,Ad)L(s+ 12, π0,Ad).
LetS be a set of bad primes containing all archimedean places. We may and do assume the following conditions:
(UR1) G1 andG0 are unramified over kv outside S.
(UR2) π1 and π0 are unramified outsideS.
(UR3) Ifv /∈S, thenϕ1,v andϕ0,v are unramified vectors andϕ1,v= ϕ0,v= 1.
(UR4) For v /∈ S, the Haar measures dg1,v and dg0,v are the standard measures. (By the standard measure, we mean the Haar mea- sure for which the volume of the hyperspecial maximal compact subgroup is 1).
Rallis and Bernstein announced (independently) that they proved that
dim HomG0,v(π1,v⊗π˜0,v,C)≤1
for non-archimedean place v of k. Although their proofs are not pub- lished yet, we admit this as a hypothesis. For archimedean prime, this hypothesis is verified in many cases (but not in general).
We consider the matrix coefficient Φϕ1,v,ϕ
1,v(g1) =ϕ1,v, π1,v(g1)ϕ1,vv, g1 ∈G1,v, Φϕ0,v,ϕ
0,v(g0) =ϕ0,v, π0,v(g0)ϕ0,vv, g0 ∈G0,v.
Put
I(ϕ1,v, ϕ0,v) =
G0,v
Φϕ1,v,ϕ1,v(ι(g0))Φϕ0,v,ϕ0,v(g0)dg0, αv(ϕ1,v, ϕ0,v) =Δ−1G
1,vPπ1,v,π0,v(12)−1I(ϕ1,v, ϕ0,v).
Proposition 1. If both π1,v and π0,v are tempered, then the integral I(ϕ1,v, ϕ0,v) is absolutely convergent.
In fact, it is enough to assume that π1,v and π0,v are sufficiently
“close” to tempered. We expect that the quantity αv(ϕ1,v, ϕ0,v) is meaningful whenever the irreducible admissible representations π1,v and π0,v are unitary and dim HomG0,v(π1,v⊗π˜0,v,C) = 1.
Proposition 2. Let v be a non-archimedean place. Assume that the conditions (UR1), (UR2) (UR3) and (UR4) hold. Assume, moreover, v 2. Then we have αv(ϕ1,v, ϕ0,v) = 1 in the range of absolute conver- gence of I(ϕ1,v, ϕ0,v).
The proof uses the result of [17], which treats only the split case. We can prove an analogue of [17] for unramified quasi-split case.
Conjecture 1.1. Assume that both π1,v and π0,v are tempered. Then (1) The integral αv(ϕ1,v, ϕ0,v) is non-negative for any ϕ1,v ∈ π1,v
and ϕ0,v ∈π0,v.
(2) There existϕ1,v ∈π1,v andϕ0,v ∈π0,v such thatαv(ϕ1,v, ϕ0,v)>
0 if and only if dim HomG0,v(π1,v, π0,v) = 1.
Now our global conjecture can be formulated as follows.
Conjecture 1.2. Assume that π1 ⊗vπ1,v and π0 ⊗vπ0,v are tem- pered cuspidal automorphic representation. Then there should be an integer β such that
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 = 2βΔG1C0Pπ1,π0(12)
v∈S
αv(ϕ1,v, ϕ0,v) ϕ1,v2· ϕ0,v2 holds.
We expect that ifπ1 andπ0 are generic cuspidal automorphic repre- sentations, then the same relation holds even when π1 and π0 are not (known to be) tempered.
Remark 1. When π1 and π0 are tempered, the L-factors L(s, π1,v,Ad), L(s, π0,v,Ad), andL(s, π1,vπ0,v) should be holomorphic for Re(s)>0.
Therefore our conjecture is equivalent to
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 = 2βΔSG1C0PπS1,π0(12)
v∈S
I(ϕ1,v, ϕ0,v) ϕ1,v2· ϕ0,v2.
Here, ΔSG1 and PπS1,π0(s) are the partial Euler products. In particular, the definition of theL-factors for bad primes plays no role in this case.
Now we consider the case whenπ1orπ0are not necessarily tempered.
We assume thatαv(ϕ1,v, ϕ0,v) is meaningful for anyv. Note that in this case the definition of theL-factors for bad primes are crucial. Note also that in this case theL-factors depends not only onπi,v, but also on the Arthur packets whichπi,v belong to. By Proposition 2, we may assume αv(ϕ1,v, ϕ0,v) = 1 for v /∈S. In this case, we conjecture as follows.
Conjecture 1.3. Assume that ϕ1 ∈ π1 and ϕ0 ∈ π0 are square- integrable automorphic forms. Then
(1) The integral ϕ1|G0, ϕ0 should be finite.
(2) Assume that dim HomG0,v(π1,v⊗π˜0,v,C) = 1 for any v. Then exists an integer β such that
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 = 2βΔG1C0Pπ1,π0(12)
v∈S
αv(ϕ1,v, ϕ0,v) ϕ1,v2· ϕ0,v2. Remark 2. At the conference, we claimed that the integralϕ1|G0, ϕ0 is finite, but there was a gap in our proof. Therefore we include this in the conjecture.
2. Relation to the Arthur conjecture
We recall the conjecture of Langlands and Arthur [1] for automorphic representation of reductive algebraic groups. We assume, for simplicity, G is a semi-simple algebraic groups defined over k. We put
Lv =
Wkv ×SU(2) if v is non-archimedean, Wkv if v is archimedean.
A Langlands parameter is a homomorphism ϕv :Lv → LGwhich satis- fies certain additional conditions. Two Langlands parameters are said to be equivalent if they are conjugate by an element of ˆG. Langlands conjectured that for each Langlands parameter, there exists a finite set Πϕv(G) of irreducible admissible representations of G(kv). The finite set Πϕv(G) is called the L-packet of ϕv. The set Π(Gv) of equiva- lence classes of irreducible admissible representations of Gv should be decomposed into a disjoint union
Π(Gv) =
ϕv
Πϕv(G).
The L-packet Πϕv(G) should contain a tempered representation if and only if the Langlands parameterϕv has a bounded image, in which case ϕv is called tempered. If ϕv is tempered, then all members of Πϕv(G) should be tempered.
A homomorphism ψv : Lv ×SL2(C) → LG is called an Arthur pa- rameter if ψv|Lv is a tempered Langlands parameter and if ψv|SL2()
is holomorphic. Arthur conjectured that for each Arthur parameter ψv, there exists a finite set of unitary representations Πψv(G). The set Πψv(G) is called the A-packet of ψv. A-packets are not necessarily disjoint.
Langlands conjectured that there exists a locally compact group Lk such that the equivalence classes of irreduciblen-dimensional represen- tation of Lk is in one-to-one correspondence with the set of irreducible cuspidal automorphic representations of GLn(Ak). There should be a homomorphism ιv :Lv → Lk for eachv.
Arthur conjectured that for each irreducible discrete series automor- phic representations π ⊗vπv of G(A) is associated to a homomor- phism
ψ :Lk×SL2(C)→ LG
in such a way that πv belongs to Πψ◦ιv(G) for each v. In this case, ψ is called the Arthur parameter of π. The set Πψ(G) of irreducible discrete series automorphic representations with Arthur parameter ψ is called the A-packet of ψ.
It is known that the Arthur parameter ψ : Lk × SL2(C) → LG associated with a discrete series automorphic representation should be elliptic in the sense that Im(ψ) is not contained in any proper Levi subgroup of LG. This is the case if and only if CentGˆ(Im(ψ)) is finite.
We put
Sψ = CentGˆ(Im(ψ)).
Now we go back to the situation that G1 = SO(n + 1) and G0 = SO(n). Assume thatπ1 (resp. π0) is a tempered cuspidal automorphic representation of G1 (resp. G0) with Arthur parameter ψ1 (resp. ψ0).
Note that in this case the groups Sψ1 andSψ0 are elementary 2-abelian groups.
Conjecture 2.1. Assume that both π1 and π0 are tempered cuspidal automorphic representations. Then the constant 2β in Conjecture 1.2 is equal to 1/(|Sψ1| · |Sψ0|), i.e., we have
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 = ΔG1C0
|Sψ1|·|Sψ0|Pπ1,π0(12)
v∈S
αv(ϕ1,v, ϕ0,v) ϕ1,v2· ϕ0,v2.
Remark 3. At the conference, we conjectured the the constant 2β de- pends only n, but it seems the constant 2β depends on the full Arthur parameter ψ.
3. Examples
• The case when π1 and π0 are trivial.
Assume, for simplicity, k is a function filed with Discriminant Dk. Assume that both G1 and G0 are quasi-split and that both π1 and π0 are trivial. Obviously, we have
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 = 1.
Consider first the case n = dimV0 = 2m. Note that we excluded the case when V0 is the split hyperbolic plane. Let K be the extension of k which corresponds to χQ0. Then we have
ΔG1 = m j=1
ζ(2j),
C0 =D−(2m−1)/2K/k D−m(2m−1)/2k L(m, χQ0)−1
m−1
j=1
ζ(2j)−1,
Pπ1,π0(12) =L(1−m, χ) m−1
j=1 ζ(−2j+ 1) m
j=1ζ(2j) .
From the functional equationζ(s) =D(1−2s)/2k ζ(1−s), and L(s, χQ0) = (DK/kDk)(1−2s)/2L(1−s, χQ0), the conjecture holds with β= 0 in this case.
Now consider the case n= 2m+ 1. In this case, ΔG1 =L(m+ 1, χQ1)
m j=1
ζ(2j),
C0 =D−m(2m+1)/2k m j=1
ζ(2j),
Pπ1,π0(12) =L(m+ 1, χQ1)−1 m
j=1ζ(−2j+ 1) m
j=1ζ(2j) . In this case, we see the conjecture also holds with β = 0.
• The case n= 2.
This example is due to Waldspurger [25]. PutG1 = PGL2. Then an irreducible cuspidal automorphic representationπ ofG1 can be consid- ered as a representation of GL2(A) with trivial central character. Let T be an anisotropic torus ofG1 corresponding to a quadratic extension K/k. Then a character ω of T(A)/T(k) can be regarded as a charac- ter of A×K whose restriction to A×k is trivial. The base change of π to GL2(AK) is denoted by Π. Choose a cusp formϕ=⊗vϕv ∈π ⊗vπv. Then among other things, Waldspurger [25] proved that
|
T(k)\T( )ϕ(ι(t))ω−1(t)dt|2
ϕ, ϕ = ζ(2)L(12,Π⊗ω−1) 2L(1, π,Ad)
v∈S
α(ϕv, Tv, ωv) ϕv2 . (cf. [25], p. 222, Proposition 7). Here,
α(ϕv, Tv, ωv) = L(1, χTv)L(1, πv,Ad) ζv(2)L(12,Πv⊗ωv−1)
Tv
Φϕv,ϕv(ι(tv))ω−1(tv)dtv. Note that in this equation, the Haar measure of T(A) is the prod- uct measure
vdtv. It is the measure such that the total volume of T(k)\T(A) is equal to 2L(1, χK/k). Now take the Tamagawa measure for T(A). Then we have C0 = L(1, χK/k)−1. Note that ΔG1 = ζ(2).
Therefore Waldspurger’s theorem implies
|ϕ|T( ), ω|2 ϕ, ϕω, ω = 1
4ΔG1C0 L(12,Π⊗ω−1) L(1, π,Ad)L(1, χK/k)
v∈S
αv(ϕv, ωv) ϕv2 . Note that in this case, we have|Sψ1|=|Sψ0|= 2.
• The Jacquet conjecture.
We consider the case n = 3 andk =Q. We assume that G1 and G0 are split overQ. In this case, the Jacquet conjecture is compatible with our conjecture. For example, we consider the case every finite place is unramified. Put G1 = SO(2,2) and G0 = SO(2,1) = PGL2. The local measures dg0,v is defined byd×a dn dk, whereg0 =ank is the standard Iwasawa decomposition d×a andn is the standard measures onkv× and kv, respectively, and dk is the Haar measure of the standard maximal compact subgroup of PGL2(kv) such that the total measure is 1. We denote by Λ(s, πi) the complete L-function for L(s, πi) and so on. We put ξ(s) = π−s/2Γ(2/s)ζ(s). Then ΔG1 = ξ(2)2 = π2/36, and C0 = 2Vol(SL2(Z)\H1)−1 = 6/π. Take Hecke eigenforms fi ∈ Sκi(SL2(Z)) (i = 1,2,3). We assume κ1 +κ2 = κ3. We denote the lift of fi to GL2(A) byfi. Let τi be the irreducible automorphic representation of PGL2(A) generated by fi. Note that f1 ×f2 induces a cusp form on
SO(2,2)(A) and its restriction to SO(2,1) isf1f2. Putπ1=τ1τ2 and π0 =τ3. Then
Pπ1,π0(12) = Λ(12, τ1×τ2×τ3) 3
i=1Λ(1, τi,Ad). Then the main theorem of Harris-Kudla [13] implies
Λ(12, τ1 ×τ2×τ3) = 22κ3+2|f1f2, f3|2. It is well known that
Λ(1, τi,Ad) = 2κifi, fi.
Here the left hand side is the completeL-function, and the right hand side is the usual Petersson inner product.
As both the Tamagawa numbers of SO(2,2) and SO(2,1) are equal to 2, we have
|f1f2,f3|2
f1 ×f2,f1×f2f3,f3 = π 3
|f1f2, f3|2 3
i=1fi, fi. After a little calculation, we have
α∞(ϕ1,∞×ϕ2,∞, ϕ3,∞) = 2.
Putting altogether, we have
|f1f2,f3|2 f1×f2,f1 ×f2f3,f3
= 1
4ΔG1C0Λ(12, τ1×τ2×τ3) 3
i=1Λ(1, τi,Ad) ·α∞(ϕ1,∞×ϕ2,∞, ϕ3,∞).
Therefore in this case it seems our conjecture also holds for β = −2.
Note that in this case we have |Sψ1| = |Sψ0| = 2. We can carry out a similar calculation for holomorphic modular forms with square-free levels by using the result of Watson [26].
• Restriction of the Saito-Kurokawa lifting to the diagonal subset.
In this example, n = 4 and k = Q. Let κ be an odd number. Let f ∈ S2κ(SL2(Z)) and g ∈ Sκ+1(SL2(Z)) be normalized Hecke eigen- forms. We denote the Kohnen plus subspace by Sκ+(1/2)+ (Γ0(4)) ⊂ Sκ+(1/2)(Γ0(4)). Let h ∈ Sκ+(1/2)+ (Γ0(4)) a Hecke eigenform associated tof. LetF ∈Sκ+1(Sp2(Z)) Saito-Kurokawa lift of h. Then it is shown in Ichino [14] that
Λ(2κ,Sym2(g)⊗f) = 2κ+1f, f h, h
|F| 1× 1, g×g|2 g, g2 .
We interpret this result in terms of automorphic representations. Let ϕ1 be the automorphic form onG1(A) = SO(3,2)(A) corresponding to F. Similarly, let ϕ0 be the automorphic form on G0(A) = SO(2,2)(A) corresponding to g ×g. As in the last case, let dg0,v be the Haar measure of G0(Qv) defined in terms of the Iwasawa decomposition.
Then we have , and
ΔG1 =ξ(2)ξ(4), C0 =ξ(2)−2, ϕ1, ϕ1= F,F
ξ(2)ξ(4), ϕ0, ϕ0=g, g2
2ξ(2)2,
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 = ξ(4) 2ξ(2)
|F| 1× 1, g×g|2 g, g .
Let τ and σ be the automorphic representations of GL2(A) generated by f and g, respectively. Then we have
Λ(2κ,Sym2(g)⊗f) =Λ(12,Ad(σ)τ), f, f=2−2κΛ(1, τ,Ad) F,F
h, h =2κ−2
π ξ(2)Λ(32, τ) It follows that
|ϕ1|G0, ϕ0|2
ϕ1, ϕ1ϕ0, ϕ0 =π· ξ(4)
ξ(2) · Λ(12,Ad(σ)τ) ξ(2)Λ(32, τ)Λ(1, τ,Ad). By easy calculation, we have
Pπ1,π0(s) = Λ(s− 12, σ,Ad)Λ(s,Ad(σ)τ)
ξ(s+ 32)Λ(s+ 1, τ)Λ(s+12, σ,Ad)Λ(s+ 12, τ,Ad). It follows that
Pπ1,π0(12) = Λ(0, σ,Ad)Λ(12,Ad(σ)τ) ξ(2)Λ(32, τ)Λ(1, σ,Ad)Λ(1, τ,Ad)
= Λ(12,Ad(σ)τ) ξ(2)Λ(32, τ)Λ(1, τ,Ad)
by the functional equation. After a little calculation, we can show that α∞(ϕ1,∞, ϕ0,∞) = 2π. Putting together, we have
|ϕ1|G0, ϕ0|2 ϕ1, ϕ1ϕ0, ϕ0 = 1
2ΔG1C0Pπ1,π0(12)· α∞(ϕ1,∞, ϕ0,∞) ϕ1,∞2· ϕ0,∞2.
Therefore in this case our conjecture seems to hold withβ =−1.
• Restriction of the hermitian Maass lift.
Now we discuss the case n = 5 and k = Q. We assume that G1 is quasi-split, but not split. We denote the splitting field ofG1 byK. Let
−D and χ be the discriminant and the associated Dirichlet character for K/Q. Note that G1 = SO(4,2)SU(2,2)/{±1}.
We putG0 = SO(3,2). Note that G0 is isogenous to Sp2. Letκ >0 be an even integer andf ∈S2κ−2(SL2(Z)) be a normalized Hecke eigen- form. We denote the irreducible cuspidal automorphic representation of SL2(A) generated byf byσ. Then we haveL(s, σ) =L(s+κ−32, f).
Let
h=
N>0 N≡0,1(4)
c(N)qN ∈Sκ−(1/2)+ (Γ0(4))
is a Hecke eigenform corresponding to f, and F the Saito-Kurokawa lift of h.
It is well-known (cf. [18]) that F,F
h, h = 2κ−3π−1ξ(2)Λ(32, σ).
For simplicity, we assume that c(D)= 0.
Now let ΓK = SU(2,2)(Q)∩GL4(OK) be the special hermitian modu- lar group. Let g ∈Sκ−1(Γ0(D), χ) be a primitive form andG ∈Sκ(ΓK) be the hermitian Maass lift of g. Thus G is a modular form on the hermitian upper half space
H2 ={Z ∈M2(C)|2√1−1(Z− tZ)¯ >0} of degree 2 with respect to ΓK. We may assume that G = 0.
We denote the irreducible cuspidal automorphic representation of GL2(A) generated byg (resp.f) byτ (resp.σ). Letπ1 (resp.π0) be the irreducible cuspidal automorphic representation ofG1(A) (resp.G0(A)) generated by G (resp. F). Then both π1 and π0 are non-tempered. It is easy to check that
L(s, π1) = Λ(s,Sym2(τ))ξ(s+ 1)ξ(s)ξ(s−1), L(s, π0) = Λ(s, σ)ξ(s+ 12)ξ(s− 12),
and
Pπ1,π0(s) = Λ(s,Sym2(τ)σ)Λ(s−1, σ)ξ(s− 32)
Λ(s+32,Sym2(τ))Λ(s+ 12, τ,Ad)Λ(s+ 12, σ,Ad)ξ(s+ 32).
Observe that Pπ1,π0
1
2
= Λ(12,Sym2(τ)σ)Λ(−12, σ)ξ(−1) Λ(2,Sym2(τ))Λ(1, τ,Ad)Λ(1, σ,Ad)ξ(2)
=− Λ(12,Sym2(τ)σ)Λ(32, σ) Λ(2,Sym2(τ))Λ(1, τ,Ad)Λ(1, σ,Ad) The main theorem of Ichino and Ikeda [15] says
|c(D)|2|G| 2,F|2
F,F2 = 2−4κ+2D2κ−3Λ(12, τ ×τ ×σ) f, f2 By using the Kohnen-Zagier formula [19]
|c(D)|2f, f
h, h = 2κ−2Dκ−(3/2)Λ(12, σ, χ), we have
|G| 2,F|2
F,F2 = 2−5κ+4Dκ−(3/2)Λ(12,Sym2(τ)σ) f, fh, h . One can show that
G,G= 2−2κ−3Dκ+1π−2(4, wK)ξ(2)Λ(2,Sym2(τ))Λ(1, f,Ad).
It is well-known that
f, f=2−2κ+2Λ(1, σ,Ad).
The volumes of the fundamental domains are Vol(Sp2(Z)\H2) =2ξ(2)ξ(4),
Vol(Γ(2)K \H2) =2−1D5/2(4, wK)ξ(2)Λ(3, χ)ξ(4).
One can showα∞=−2π. Therefore in this case, our conjecture holds with β =−1.
References
[1] J. Arthur, Unipotent automorphic representations: conjectures, Ast´erisque 171-172(1989), 13–71.
[2] S. B¨ocherer, M. Furusawa, and R. Schulze-Pillot, On the global Gross-Prasad conjecture for Yoshida liftings, Contributions to automorphic forms, geometry, and number theory, Johns Hopkins Univ. Press, 2004, pp. 105–130.
[3] S. B¨ocherer and R. Schulze-Pillot, On the central critical value of the triple product L-function, Number theory (Paris, 1993–1994), London Math. Soc.
Lecture Note Ser.235, Cambridge Univ. Press, 1996, pp. 1–46.
[4] P. Deligne,Valeurs de fonctionsLet p´eriodes d’int´egrales, Automorphic forms, representations andL-functions, Proc. Sympos. Pure Math.33, Part 2, Amer.
Math. Soc., 1979, pp. 313–346.
[5] M. Eichler and D. Zagier, The theory of Jacobi forms, Progr. Math. 55, Birkh¨auser Boston Inc., 1985.
[6] P. B. Garrett,Decomposition of Eisenstein series: Rankin triple products, Ann.
of Math.125(1987), 209–235.
[7] P. B. Garrett and M. Harris,Special values of triple productL-functions, Amer.
J. Math.115(1993), 161–240.
[8] D. Ginzburg, I. I. Piatetski-Shapiro, and S. Rallis,Lfunctions for the orthog- onal group, Mem. Amer. Math. Soc.611(1997).
[9] B. H. Gross and D. Prasad,On the decomposition of a representation ofSOn when restricted toSOn−1, Canad. J. Math.44 (1992), 974–1002.
[10] , On irreducible representations of SO2n+1×SO2m, Canad. J. Math.
46(1994), 930–950.
[11] M. Harris and S. S. Kudla, The central critical value of a triple product L- function, Ann. of Math.133(1991), 605–672.
[12] ,Arithmetic automorphic forms for the nonholomorphic discrete series ofGSp(2), Duke Math. J.66(1992), 59–121.
[13] , On a conjecture of Jacquet, Contributions to automorphic forms, ge- ometry, and number theory, Johns Hopkins Univ. Press, 2004, pp. 355–371.
[14] A. Ichino,Pullbacks of Saito-Kutokawa lifts, to appear.
[15] A. Ichino and T. Ikeda,On Maass lifts and the central critical values of triple productL-functions, preprint.
[16] T. Ikeda,On the lifting of hermitian modular forms, preprint.
[17] S. Kato, A. Murase and T. Sugano,Whittaker-Shintani functions for orthogo- nal groups, Tohoku Math. J.55(2003), 1–64.
[18] W. Kohnen and N.-P. Skoruppa,A certain Dirichlet series attached to Siegel modular forms of degree two, Invent. Math.95(1989), 541–558.
[19] W. Kohnen and D. Zagier,Values of L-series of modular forms at the center of the critical strip, Invent. Math.64(1981), 175–198.
[20] H. Kojima, An arithmetic of Hermitian modular forms of degree two, Invent.
Math.69(1982), 217–227.
[21] A. Krieg, The Maaß spaces on the Hermitian half-space of degree 2, Math.
Ann.289(1991), 663–681.
[22] I. I. Piatetski-Shapiro and S. Rallis, Rankin triple L functions, Compositio Math.64(1987), 31–115.
[23] D. Prasad, Trilinear forms for representations of GL(2) and local -factors, Compositio Math.75(1990), 1–46.
[24] S. Raghavan and J. Sengupta,A Dirichlet series for Hermitian modular forms of degree2, Acta Arith.58(1991), 181–201.
[25] J.-L. Waldspurger, Sur les valeurs de certaines fonctions L automorphes en leur centre de sym´etrie, Compositio Math.54(1985), 173–242.
[26] T. Watson,Rankin triple products and quantum chaos, to appear in Ann. Math.