A short introduction to the Gan–Gross–Prasad conjectures

Xenia Dimitrakopoulou

Abstract
The Gan–Gross–Prasad conjectures are a series of conjectures in the theory of automorphic
representations. When given a linear algebraic group G and a subgroup of it H, a classical
question to ask is how do the G-irreducible representations decompose when restricted to H.
An answer to this question is called a branching law and in the case of the classical groups it is
known that this restriction problem is multiplicity free. The GGP conjectures give an explicit
description of this multiplicity.

Contents
1 From Hecke to Waldspurger 1
1.1 Hecke’s central value formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Waldspurger’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 The GGP conjectures 4

2.1 Basics on automorphic forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Statement of the conjectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 An RTF approach to the GGP conjectures 7

1 From Hecke to Waldspurger

This section is following Chao Li’s notes of Beuzart-Plessis’ course at Columbia in Spring 2019. We
have made very small alterations, since the theory was described beautifully there.

1.1 Hecke’s central value formula

Our story begins with the classical work of Hecke, who described the integral representation of
L-functions of modular forms. Let f ∈ Sk (Γ(1)) be a cusp form of level k and consider its Fourier
expansion X
f= an q n .
n≥1

Let X an
L(s, f ) = , Λ(s, f ) = (2π)−s Γ(s)L(s, f )
ns

1
be its L-function and the completed L-function respectively. Hecke showed that the completed
L-function is equal to the Mellin transform of the modular form
Z ∞
Λ(s, f ) = f (iy)y s−1 dy,
0

and use it to show the analytic continuation and functional equation of Λ(s, f ). Evaluating at the
center of functional equation s = k/2 (which lies outside the range of convergence of Λ(s, f )), we
obtain a central value formula
Z ∞
k−1
Λ(k/2, f ) = f (iy)y 2 dy.
0

This is the identity which the Gan-Gross-Prasad conjectures aim to generalize. In order to pass to
a more general setting, first we need to interpret Hecke’s central value formula in the adelic setting.
Consider the bijection
SO(2)(R)\SL2 (R) → H
 
a b ai + b
7→
c d ci + d
which maps the torus t, t−1 , t ∈ R× ⊂ SL2 (R) to the vertical line {iy | y > 0} ⊂ H. We


can now consider Λ(f, s) as an integral along the image of the split torus in SL2 (R). To make
this precise, recall that f gives rise to a vector φf ∈ π, where π is the cuspidal automorphic
representation of GL2 attached to f . The representation π has an L-function L(π, s), which equals
(up to a linear change of variables) Λ(f, s). By Hecke’s work, we can consider the following integral
representation of this L-function
Z  
t 0 1
L(π, s) = φf |t|s− 2 dt.
A× /Q× 0 1

Note that the extra 1/2 comes from normalization of measures. The RHS is known as an auto-
morphic period PA (φf ) along the subgroup A = Gm ,→ G (a real number since f is a normalized
eigenform).
Recall also that Rankin-Selberg expressed the Petersson inner product
Z
dxdy
hf, f i = |f (z)|2 y k 2
Γ(1)\H y

in terms of the Rankin-Selberg L-function

hf, f i = 2π k−1 ress=k Λ(s, f × f )

The RHS is equal to an adjoint L-value at the edge of critical strip

2−k L(1, π, Ad),

2
 π
while the LHS is also equal to 6 (again due to normalization of measures) times
Z
2
hφf , φf i = |φf (g)| dtam g.
[G]

Squaring Hecke’s central value formula and dividing the Rankin-Selberg identity, we obtain
2
PA (φf ) L(1/2, π)2
= 2k−2 ξ(2) ,
hφf , φf i L(1, π, Ad )

where ξ(s) = π −s/2 Γ(s/2)ζ(s) is the completed Riemann zeta function. This new identity general-
izes to any vector φ ∈ π of any cuspidal automorphic representation of G(A) over any number field
F . Let φ = ⊗v φv ∈ π = ⊗0v πv . Then
2
|PA (φ)| ζ S (2) LS (1/2, π)2 Y
= S S
· αv (φv , φv ) ,
hφ, φi 2 ress=1 ζ (s)L (1, π, Ad)
v∈S

for S a sufficiently large finite set of places including the archimedean places, where the local period
R
A(Fv )
hπv (av ) (φv ) , φv i dav
αv (φv , φv ) = .
hφv , φv iv

Here we choose the Tamagawa measures on A(A) and G(A), and the local measures are chosen
such that Y
dav = dtam a
v

and the local period

2
ζv (2) L (1/2, πv )
αv (φv , φv ) =
ζv (1) L (1, πv , Ad)
for almost all v ’s. One can show directly that αv is not identically zero and L(s, π, Ad) has no
pole or zero at s = 1, hence the new identity implies the equivalence that there exists φ ∈ π such
that PA (φ) 6= 0 ⇐⇒ L(1/2, π) 6= 0.

1.2 Waldspurger’s theorem

Waldspurger in [5] and [6] proved a remarkable generalization by replacing A with any nontrivial
torus T in G. Such a torus is isomorphic to ResF 0 /F Gm /Gm for a quadratic extension of number
fields F 0 /F (and the embedding A ,→ G is unique up to conjugation). Let η : A× /F × → {±1} be
the quadratic character associated to F 0 /F by global class field theory.

Theorem 1.1 (Waldspurger). For every φ = ⊗v φv ∈ π, we have

2
|PT (φ)| ζ S (2) LS (1/2, π)LS (1/2, π × η) Y
= F αv (φv , φv )
hφ, φi 4 LS (1, η)LS (1, π, Ad )
v∈S

where the local periods are defined similarly via integration over Tv instead of Av .

Waldspurger’s formula looks exactly the same as the split torus case, but it is much harder to prove.

3
This is due to the fact that there is no direct relation between the toric period PT (φ) with integral
representations of L(s, π)L(s, π × η). Moreover, unlike the split torus case, the local periods can be
identically zero. It turns out that

αv 6= 0 ⇐⇒ HomTv (πv , C) 6= 0.

This implies that

PT |π 6= 0 ⇐⇒ L(1/2, π)L(1/2, π × η) 6= 0 and HomTv (πv , C) 6= 0 for all v.

2 The GGP conjectures

2.1 Basics on automorphic forms
In order to understand a more general version of Waldspurger’s formula, we will need to understand
what an automorphic form looks like on a more general algebraic group. This section is a basic
introduction to the theory of automorphic forms.
Let F be a number field with local completions Fv for all places v of F . Let G be a connected
reductive linear algebraic group over F . Fixing any faithful algebraic representation ρ : G ,→ GLn
over F , one obtains a system of open compact subgroups Kv = ρ−1 (GLn (Ov )) ⊂ G (Fv ), for almost
Q0
all v, where Ov is the ring of integers of Fv ; for almost all v, Kv is hyperspecial. G(A) = v G (Fv )
be the adelic group, which is a restricted direct product of G (Fv ) relative to the family {Kv } of
open compact subgroups for almost all v; [G] = G(F )\G(A) be the automorphic quotient; the
locally compact group G(A) acts on [G] by right translation and there is a G(A)-invariant measure
(unique up to scaling).

Definition 2.1. An automoprhic form on G is a function

f : [G] −→ C

such that

1. f is smooth
Q
2. f is right Kf -finite, where Kf = v<∞ Kv

3. f is of uniform moderate growth

4. f is Z(g)-finite, where g is the Lie algebra of G.

We denote the space of automorphic forms on G by A(G).

Note that G(A) acts on the vector space A(G) by right translation, giving it a G(A)-module
structure.

4
Definition 2.2. An automorphic representation π of G is an irreducible subquotient of the G(A)-
Q0
module A(G). As an irreducible abstract representation of G(A) = v G (Fv ), π is of the form:

π = ⊗0v πv ,

a restricted tensor product of irreducible smooth representations πv of G (Fv ).

Definition 2.3. An automorphic form f on G is called a cusp form if, for any parabolic subgroup
P = M N of G, the N -constant term
Z
fN (g) = f (ng)dn
N (F )\N (A)

is zero as a function on G(A). We denote the space of cuspidal automorphic forms on G by

A0 (G) ⊂ A(G).

It turns out that, if G is semisimple or has anisotropic center, a cusp form f is rapidly decreasing
as a function on [G] (like a Schwarz function) and hence is square-integrable on [G], i.e
Z
|f (g)|2 dg < ∞.
[G]

L
Moreover, it is known that A0 (G) decomposes as a direct sum A0 (G) = π m0 (π) · π with finite
multiplicities. An irreducible summand π of A0 (G) is called a cuspidal automorphic representation.
Let H be a subgroup of G. As we saw in the previous lecture, we are interested in studying the
automorphic H-period integral
PH :A0 (G) −→ C
Z
φ −→ φ(h)dh.
Z(A)\[H]

It is also natural to consider a twisted version. Let χ be a character of Z(A)H(F )\H(A). We define
the automorphic (H, χ)-period integral PH,χ in a similar manner,
Z
PH,χ (φ) = φ(h)χ(h)dh.
Z(A)\[H]

Let π be a unitary cuspidal automorphic representation of G(A). We consider the restriction of

PH to π, which defines an element in PH,π ∈ HomH(A) (π, C).
Definition 2.4. A cuspidal automorphic representation π is (globally) distinguished by H if the
linear functional PH,π ∈ HomH(A) (π, C) does not vanish, i.e., there exists some φ ∈ π such that
PH (φ) 6= 0. We say that πv is (locally) distinguished by H (Fv ) if HomH(Fv ) (πv , C) 6= 0.

If π is distinguished by H, then HomH(A) (π, C) 6= 0, and in particular, HomH(Fv ) (πv , C) =

6 0 for
every place v. However, the converse statement is not always true, and this observation is the key
part of the GGP conjectures.
It is expected that the automorphic period integral PH behaves nicely only when the pair (H, G)
satisfies certain nice properties, such as

1. the multiplicity-one property dim HomH(Fv ) (πv , C) ≤ 1 holds for all v, or the multiplicity can
be described in a certain nice way, and

5
 2. the locally distinguished representations can be characterized in terms of L-parameters.

A large class of (H, G) called spherical pairs are expected to have the above properties. If F is an
algebraically closed field, we say that the pair (H, G) is spherical if a Borel subgroup B of G has
an open dense orbit on G/H. Over a general field F , the pair (H, G) is said to be spherical if its
base change to an algebraic closure of F is spherical. We then call H a spherical subgroup of G.
Here are some examples of spherical pairs:

Example 2.5.

1. The Whittaker pair (N, G), where G is quasi-split and N is a maximal nilpotent subgroup of
G.

2. The pair (G, G×G), where we view G as a subgroup of G×G via its image under the diagonal
embedding.

3. The Rankin–Selberg pair (GLn−1 , GLn × GLn−1 ).

4. The GGP pairs (SOn−1 , SOn × SOn−1 ) and (Un−1 , Un × Un−1 ).

2.2 Statement of the conjectures

We will now proceed to state the global Gan-Gross-Prasad conjectures in the orthogonal and
Hermitian cases. We follow [1]. Consider F and G as in the previous section and let the subgroup
H of G be reductive. For the number field F consider and extension F 0 /F such that F 0 = F
in the orthogonal case and F 0 a quadratic extension of F in the Hermitian case. Let Wn be a
non-degenerate orthogonal space or Hermitian space with F 0 -dimension n. Let Wn−1 ⊂ Wn be a
non-degenerate subspace of codimension one. Let Gi be SO (Wi ) or U (Wi ) for i = n − 1, n. Let

G = Gn−1 × Gn , H = Gn−1

where we view H as a subgroup of G via the diagonal embedding ∆ : H ,→ G. Note that the pair
(H, G) is spherical.
Let π = πn−1 ⊗ πn be a tempered cuspidal automorphic representation of G(A). The central
L-values of certain automorphic L-functions L(s, π, R) show up in their conjecture, where R is
a finite dimensional representation of the L-group L G. We can describe the L-function as the
Rankin–Selberg convolution of suitable automorphic representations on general linear groups.
For i ∈ {n − 1, n}, let Πi,F 0 be the endoscopic functoriality transfer of πi from Gi to suitable
GLN (AF 0 ) . In the Hermitian case, this is the base change of πi to GLi (AF 0 ); and in the orthogonal
case, this is the endoscopic transfer from Gi (A) to GLi (A) (resp.GLi−1 (A)) if i is even (resp.
odd). The L-function L(s, π, R) can be defined as the Rankin–Selberg convolution L-function
L (s, Πn−1,F 0 × Πn,F 0 ).
We are ready to state the global Gan-Gross-Prasad conjecture.

Conjecture 2.6 (Gan–Gross–Prasad). Let π be a tempered cuspidal automorphic representation

of G(A). The following statements are equivalent.

1. The automorphic H-period integral does not vanish on π, i.e. PH (φ) 6= 0 for some φ ∈ π.

6
 1


2. The space HomH(A) (π, C) 6= 0 and the central value L 2 , π, R 6= 0.

There is a refinment of the conjecture, due to Ichino and Ikeda [2], which resembles Waldspurger’s
formula in this case. In order to state it, define the local periods:
Z
PH,v : (φv , φ0v ) 7→ hπv (hv )φv , φ0v iv dhv .
Hv

The Ichino–Ikeda refinement of the GGP conjecture links the square of the automorphic H-period
N
with central L-values. In particular, for φ = v φv ∈ π, the conjecture states:

|PH (φ)|2 ∆ LS (1/2, BC(π)) Y PH,v (φv , φv )

= ,
hφ, φi Sπ LS (1, π, Ad) hφv , φv iv
v∈S
Qn+1
where ∆ is a product of L-values attached to the quadratic character η, ∆ = i=1 L(i, η i ), and Sπ
is the centralizer of the Langlands parameter of π.
Due to the incredible work of many authors, including Beuzart-Plessis, Gordon, Xue, Yun, Zhang
and others, the GGP conjecture and its Ichino–Ikeda refinement are now theorems in the unitary
case.

Theorem 2.7. Let G = U (Wn−1 ) × U (Wn ) for Hermitian spaces Wn−1 ⊂ Wn over a quadratic
extension F 0 of F . Let π be a tempered cuspidal automorphic representation of G(A). Then
Conjecture 2.6 and its Ichino–Ikeda refinement hold.

3 An RTF approach to the GGP conjectures

In this final section, we will present a very brief introduction to the main ideas that led to the proof
of the unitary GGP conjecture. In 1985, when Waldspurger first proved his theorem, he used a
direct approach via the Shimura correspondance. A year later, Jacquet [3] published a proof that
was using a different method, the trace formula. Since then, trace formuli have been extremely
useful in the study of automorphic forms. In [4], Jacquet and Rallis proposed a relative trace
formula (RTF) approach to the conjecture. It was indeed this recipe that led to the proof of the
conjecture.
Recall that in section 1.1 we observed that even though Hecke’s central value formula was easy
to understand, it was much harder to find a connection between the non-trivial toral automprphic
period integrals and the central L-value in Waldspurger’s formula. A similar issue arises in the
GGP setting. In particular, due to the work of Jacquet–Pietski-Shapiro–Shalika, we know that the
Rankin-Selberg period satisfies:
Y
PGLn,K (φ) = LS (1/2, πK ) PGLn,Kv (φv , φv )
v∈S

for some local periods PGLn,Kv . The rough idea now is that if we can relate the period on the
left hand side with the unitary periods we care about, we would have a relation simlar to the
conjecture’s.
We will know proceed to explain what a relative trace formula is.

7
Fix a number field K and a linear reductive algebraic group G over K. We will study the space
L2 ([G]) with its G(AK )-module structure.
We consider the action of the Hecke algebra Cc∞ (G(AK )) on L2 ([G]). Take f ∈ Cc∞ (G(AK )) and
φ ∈ L2 ([G]) and define the integral operator
Z
R(f )(φ)(x) = f (y)φ(xy)dy.
G(AK )

This operator satisfies

Z Z
R(f )(φ)(x) = f (y)φ(xy)dy = f (x−1 y)φ(y)dy
G(AK ) G(AK )
Z X
−1
= f (x γy)φ(y)
[G] γ∈G(K)
Z
= Kf (x, y)φ(y)
[G]

(1)

f (x−1 γy), the kernel of R(f ).

P
where Kf (x, y) = γ∈G(K)
We will now make an oversimplifying assumption in order to be able to give an elementary treatment
of the relative trace formula. We will assume that we can write our automorphic form φ as φ =
P P
π ψ∈OB(π) hφ, ψiψ, where h, i is the Peterson inner product, the first sum is over irreducible
automorphic representations of G and the second one is over an orthonormal basis of π. It is very
crucial to note that this sum doesn’t make sense unless we take care of the continuous spectrum of
L2 ([G]) (and not just the discrete part which appears here). However, assuming that this expansion
makes sense we have:

X X
R(f )(φ)(x) = hφ, ψiR(f )(ψ)(x)
π ψ∈OB(π)
X X
= hφ, ψiπ(f )ψ(x)
π ψ∈OB(π)
X X Z
= φ(y)ψ(y)π(f )ψ(x)dy
π ψ∈OB(π) [G]
 
Z X X
=  π(f )ψ(x)ψ(y) φ(y)dy
[G] π ψ∈OB(π)
Z X
= Kf,π (x, y)φ(y),
[G] π

(2)
P
where we have set Kf,π (x, y) = ψ∈OB(π) π(f )ψ(x)ψ(y) and so we have obtained by (1) and (2)
P
that Kf (x, y) = π Kf,π (x, y) which is called the spectral expansion of the kernel.
We will now fix two G-subgroups H1 , H2 and a character η : [H2 ] → C× .

8
Define for f ∈ Cc∞ (G(AK )) a distribution
Z
I(f ) = Kf (h1 , h2 ) η (h2 ) dh1 dh2 .
[H1 ]×[H2 ]

Applying the spectral expansion of the kernel on this distribution we get

Z X X
I(f ) = π(f )ψ(h1 )ψ(h2 )η (h2 ) dh1 dh2
[H1 ]×[H2 ] π ψ∈OB(π)

X X Z Z
= π(f )ψ(h1 )dh1 η(h2 )ψ(h2 )dh2
π ψ∈OB(π) [H1 ] [H2 ]
X X
= PH1 (π(f )ψ)PH2 ,η (ψ).
π ψ∈OB(π)

We can start understanding now why the spectral expansion is useful, since the automorphic period
integrals we are interested in have appeared.
The distribution I(f ) also has a different expansion, called the geometric expansion. It stems from
the fact that we have an action of H1 × H2 on G given by (h1 , h2 ) · γ = h−1 1 γh2 . We will denote
the stabilizer of an element γ ∈ G by (H1 , H2 )γ . We now have
 
Z X
f h−1


I(f ) =  1 γh2
 η (h2 ) dh1 dh2
[H1 ]×[H2 ] γ∈G(K)
X Z
f h−1


= 1 γh2 η (h2 ) dh1 dh2
γ∈H1 (K)\G(K)/H2 (K) (H1 ×H2 )(AK )/(H1 ×H2 )γ (K)

X h i Z
f h−1


= vol (H1 × H2 )γ 1 γh2 η (h2 ) dh1 dh2
γ∈H1 (K)\G(K)/H2 (K) (H1 ×H2 )(AK )/(H1 ×H2 )γ (AK )
X h i
= vol (H1 × H2 )γ O(γ, f ),
γ∈H1 (K)\G(K)/H2 (K)

f h−1
R

where we have defined O(γ, f ) = (H1 ×H2 )(AK )/(H1 ×H2 )γ (AK ) 1 γh2 η (h2 ) dh1 dh2 , the orbital
integral of f at γ.
The relative trace formula (RTF) for (G, H1 , H2 ) is
X h i X X
vol (H1 × H2 )γ O(γ, f ) = PH1 (π(f )ψ)PH2 ,η (ψ).
γ∈H1 (K)\G(K)/H2 (K) π ψ∈OB(π)

To prove the unitary GGP conjecture, we need to consider two different RTFs:

1. Set H = U (Wn )\G = U (Wn ) × U (Wn+1 ) /H and let η be trivial. Then, for a sufficiently nice
test function f ∈ Cc∞ (G(A)) we have
X X
PH (R(f )φ)PH (φ) = O(δ, f )
φ∈A0 ([G]) δ∈H(F )\Grs (F )/H(F )

9
 2. Set H1 := GLn,K ,→ G0 = GLn,K × GLn+1,K ←- H2 = GLn,F × GLn+1,F and

n+1
n
η := ηK/F ◦ det ⊗ ηK/F ◦ det : [H2 ] → {±1}.

Then, for a sufficiently nice test function f 0 ∈ Cc∞ (G0 (A)) we have
X X
PH1 (R (f 0 ) φ) PH2 ,η (φ) = O (γ, f 0 )
φ∈A0 ([G0 ]) γ∈H1 (F )\G0rs (F )/H2 (F )

Note that both sums in the RTFs run over regular semisimple elements, hence the notation Grs , G0rs .
The RHS of (2) is related with specific L-values. In particular, PH1 (R (f 0 ) φ) detects the vanishing
of the Rankin–Selberg central L-value and PH2 ,η (φ) detects the same for the image of base change.
In order to prove GGP, one must match the orbits on the geometric sides of the two RTFs.
The recipe for matching the geometric sides uses the isomorphism
G
H V (k)\GVrs (k)/H V (k) ' H1 (k)\G0rs (k)/H2 (k),
V

where k = F or Fv and V runs over isomorphism classes of n-dimensional Hermitian spaces. Noting
that the orbital integrals are local, that is if f = v fv , f 0 = v fv0 then O(δ, f ) = v O (δ, fv ),
Q Q Q

O (γ, f 0 ) = 0


We say that fv = fvV ∈
Q
v O
(γ, fv ) , we aim to perform a local matching.
∞
GV (Fv ) matches fv0 ∈ Cc∞ (G0v ) (fv ↔ fv0 ) if
L
V Cc

O δ, fvV = Ωv (γ)O (γ, fv0 ) , for δ ∈ GVrs (Fv ) ↔ γ ∈ G0rs (Fv )

Ωv (γ) = 1 for γ ∈ G0rs (F ).

Q
where (Ωv )v are local transfer factors such that v
To make this matching work, there are two key steps:

1. (Smooth transfer) For every fv there exists fv0 st fv ↔ fv0 and conversely

2. (Fundamental lemma) For 1G(Oy ) ↔ 1G0 (Oy ) for almost all places v.

Smooth transfer was shown by Zhang in the p-adic case and by Xue in the Archimedean case. The
Fundamental lemma was proved by Beuzart-Plessis, Gordon and Yun.

References
[1] W. T. Gan, B. H. Gross, D. Prasad, and J.-L. Waldspurger. Sur les conjectures de Gross et
Prasad. I. Société Mathématique de France, Paris, 2012. Astérisque No. 346 (2012) (2012).

[2] Atsushi Ichino and Tamutsu Ikeda. On the periods of automorphic forms on special orthogonal
groups and the Gross-Prasad conjecture. Geom. Funct. Anal., 19(5):1378–1425, 2010.

[3] Hervé Jacquet. Sur un résultat de Waldspurger. Ann. Sci. École Norm. Sup. (4), 19(2):185–229,
1986.

10
[4] Hervé Jacquet and Stephen Rallis. On the Gross-Prasad conjecture for unitary groups. In
On certain L-functions, volume 13 of Clay Math. Proc., pages 205–264. Amer. Math. Soc.,
Providence, RI, 2011.

[5] J.-L. Waldspurger. Correspondance de Shimura. J. Math. Pures Appl. (9), 59(1):1–132, 1980.

[6] J.-L. Waldspurger. Sur les coefficients de Fourier des formes modulaires de poids demi-entier.
J. Math. Pures Appl. (9), 60(4):375–484, 1981.

11