TRACE FORMULAE FOR SL(2, R)
DORIAN GOLDFELD
Contents 1. Trace Formula for finite groups 3 2. Trace Formula for the infinite additive group R and subgroup Z 5 3. The Selberg Trace Formula for SL(2, R) (Spectral Side) 7 4. The Selberg Trace Formula for SL(2, R) (Geometric Side) 24 4.1. The Identity Contribution C(Id) 24 4.2. The Contribution C(∞) 25 4.3. Contribution of the Hyperbolic Conjugacy Classes C(P ) 31 4.4. Contribution of the Elliptic Conjugacy Class C(R) 35 5. The Selberg Trace Formula for SL(2, R) 39 6. The Kuznetsov Trace Formula for SL(2, R) 42 7. The SL(2, R) Kuznetsov Trace Formula (Spectral Side) 43 8. The SL(2, R) Kuznetsov Trace Formula (Geometric Side) 45 9. The Kuznetsov Trace Formula for SL(2, R) 48 9.1. Kontorovich-Lebedev Transform 48 9.2. Explicit choice of the pT,R test function 49 9.3. Bounds for the pT,R test function 50 9.4. Bounding the Geometric Side of the KTF 54 9.5. Computation of the Main Term M in the KTF 56 9.6. Explicit KTF for GL(2, R) with Error Term 58 9.7. Bounding the Eisenstein term in the KTF 58 10. The Kuznetsov Trace Formula for GL(3, R) 60 10.1. Whittaker functions and Maass forms on h3 63 11. The SL(3, R) Kuznetsov Trace Formula (Spectral Side) 64 11.1. Eisenstein contribution 66 References 69 1 2 DORIAN GOLDFELD
1. Trace Formula for finite groups Let G be a finite group and let Γ be a subgroup of G. We consider the C-vector space of functions n o
V := φ : G → C φ(γx) = φ(x), ∀γ ∈ Γ, ∀x ∈ G . We can think of V as the C-vector space of automorphic forms on Γ\G.
Definition 1.1. (Inner product on V ) For φ1, φ2 ∈ V we define X φ1, φ2 := φ1(x)φ2(x). x∈Γ\G Definition 1.2. (Orthonormal basis for V ) Let Γz ∈ Γ\G and define 1Γz ∈ V by ( 1 if x ∈ Γz, 1 (x) := Γz 0 otherwise.
The functions 1Γz for distinct cosets Γz form an orthonormal basis for V with respect to the above inner product.
Definition 1.3. (The kernel function Kf (x, y)) Let G be a finite group and f : G → C. Let Γ be a subgroup of G. For x, y ∈ G we define the kernel function X −1 Kf (x, y) := f x γy . γ∈Γ
Definition 1.4. (The linear map Kf ) Let φ ∈ V . We define a linear map Kf : V → V where for φ ∈ V we have Kf (φ) ∈ V where X (Kf φ)(x) := Kf (x, y) · φ(y). y∈Γ\G Definition 1.5. (Trace of a linear map on a finite dimensional vector space) Let V be a complex vector space with basis v1, v2, . . . , vn. A linear map L : V → V satisfies n n X X L(vi) = ai,jvj, for all 1 ≤ i ≤ n and ai,j ∈ C . i=1 j=1 Then associated to L we may define the matrix AL := ai,j 1≤i,j≤n. The trace of L (denoted Tr(L)) is defined to be n X Tr(L) = ai,i, i=1 i.e., it is the trace of the matrix AL. TRACE FORMULAE FOR SL(2, R) 3 Remarks on eigenvalues of a matrix: It is well known that for a matrix A with complex coefficients ai,j, (1 ≤ i, j ≤ n) that • Tr(A) = sum of the eigenvalues of A; • Det(A) = product of the eigenvalues of A.
We now compute the action of the linear map Kf : V → V on the orthonormal basis given in definition 1.2. Let z ∈ Γ\G. We have
X Kf 1Γz (x) = Kf (x, y) · 1Γz(y) y∈Γ\G
= Kf (x, z) X = Kf (x1, z) · 1Γx1 (x).
x1∈Γ\G
It follows that the matrix AK associated to the linear map Kf is f given by Kf (x1, z) , and x1,z∈Γ\G
X Tr(Kf ) = Kf (z, z). z∈Γ\G
Proposition 1.6. (Trace formula for a finite group G) Let G be a finite group and let Γ be a subgroup of G. Let Cl[Γ] denote the set −1 of conjugacy classes σ γσ σ ∈ Γ with γ ∈ Γ. For γ ∈ Γ we also define
−1 Γγ := σ ∈ Γ σ γσ = γ , −1 Gγ := g ∈ G g γg = γ .
It follows that
X #(Gγ) X −1 Tr(Kf ) = f z γz . # (Γγ) γ∈Cl[Γ] z∈Gγ \G 4 DORIAN GOLDFELD
Proof. We have X Tr(Kf ) = Kf (z, z) z∈Γ\G X X = f z−1γz z∈Γ\G γ∈Γ X X X = f z−1 · σ−1γσ · z
z∈Γ\G γ∈Cl[Γ] σ∈Γγ \Γ X X X = f z−1 · σ−1γσ · z
γ∈Cl[Γ] z∈Γ\G σ∈Γγ \Γ X X = f z−1γz
γ∈Cl[Γ] z∈Γγ \G X X X = f (gz)−1 · γ · gz
γ∈Cl[Γ] z∈Gγ \G g∈Γγ \Gγ X X X = f z−1 · g−1γg · z
γ∈Cl[Γ] z∈Gγ \G g∈Γγ \Gγ X X X = f z−1γz
γ∈Cl[Γ] z∈Gγ \G g∈Γγ \Gγ
X #(Gγ) X = f z−1γz . # (Γγ) γ∈Cl[Γ] z∈Gγ \G The trace formula has two sides:
X #(Gγ) X −1 Tr(Kf ) = f z γz . | {z } # (Γγ) spectral side γ∈Cl[Γ] z∈Gγ \G | {z } geometric side where the geometric side consists of a sum over conjugacy classes.
2. Trace Formula for the infinite additive group R and subgroup Z
Consider the additive group G = R and subgroup of rational integers Γ = Z. Following the recipe in §1, we define the C vector space of smooth functions n o
V := φ : R → C φ(x + n) = φ(x), ∀n ∈ Z, ∀x ∈ R . TRACE FORMULAE FOR SL(2, R) 5
Definition 2.1. (Inner product on V ) Let φ1, φ2 ∈ V. We define the inner product Z Z 1
φ2, φ2 := φ1(x) φ2(x) dx = φ1(x) φ2(x) dx. 0 Z\R 2πinx Definition 2.2. (Orthonormal basis for V ) Let en(x) := e . It is well known that every periodic function is a linear combination of en(x) with n ∈ Z. Furthermore Z 1 ( 2πi(m−n)x 1 if m = n, em, en = e dx = 0 0 otherwise. This establishes that the en is an othonormal basis of V . n∈Z Definition 2.3. (The kernel function Kf (x, y)) Let f : R → C be a Schwartz function, i.e., a smooth function all of whose derivatives have rapid decay. For x, y ∈ R we define the kernel function X Kf (x, y) := f (−x + n + y) . n∈Z
Definition 2.4. (The linear map Kf ) Let φ ∈ V . We define the linear map Kf : V → V by Z 1 Kf φ (x) := Kf (x, y) · φ(y) dy. 0 Proposition 2.5. (The en are eigenfunctions of Kf ) Let n ∈ Z. Then Kf en (x) = fb(−n) · en(x), R ∞ −2πixξ where fb is Fourier transform of f given by fb(ξ) := −∞ f(x)·e dx. Proof. Z 1 Kf en (x) = Kf (x, y) · en(y) dy 0 1 Z X = f (−x + n + y) · en(y) dy 0 n∈Z ∞ ∞ Z Z = f(−x + y) · en(y) dy = f(y) · en(x + y) dy −∞ −∞
= fb(−n) · en(x), since en(x + y) = en(x) · en(y). 6 DORIAN GOLDFELD
Proposition 2.6. (The trace formula for the additive group R) The trace formula for the group R and subgroup Γ = Z is given by X X fb(n) = f(n) . n∈Z n∈Z | {z } | {z } spectral side geometric side
Remark: The trace formula is the well known Poisson summation formula.
Proof. We have shown that en is an eigenfunction of the linear map Kf with the eigenvalue fb(n). The trace of Kf is given by the sum of the eigenvalues which is just P fb(n). To find the geometric side we must n∈Z compute 1 Z 1 Z X X K(x, x) dx = f(n) dx = f(n). 0 0 n∈Z n∈Z Note that since the additive group Z is abelian there is only one con- jugacy class, namely Z itself.
3. The Selberg Trace Formula for SL(2, R) (Spectral Side) Let G = SL(2, R) and Γ = SL(2, Z). We also require the maximal compact subgroup cos(θ) sin(θ) K = SO(2, R) = 0 ≤ θ < 2π . − sin(θ) cos(θ) In this case we define the vector space V as a space of smooth functions φ given by n o
V := φ : G → C φ(γgk) = φ(g), ∀γ ∈ Γ, ∀k ∈ K, ∀g ∈ G . By the Iwasawa decomposition it is known that every g ∈ G can be uniquely expressed in the form 1 − 1 y 2 xy 2 cos(θ) sin(θ) g = − 1 · 0 y 2 − sin(θ) cos(θ) for some 0 ≤ θ < 2π, x ∈ R, y > 0. The action of SL(2, R) on the upper half plane h := {x + iy | x ∈ R, y > 0} given by a b az + b z := , ∀ ( a b ) ∈ SL(2, ), ∀z ∈ h , c d cz + d c d R TRACE FORMULAE FOR SL(2, R) 7 establishes a one-to-one correspondence between G/K and h given by
1 − 1 1 − 1 y 2 xy 2 cos(θ) sin(θ) y 2 xy 2 − 1 · i = − 1 i = x + iy. 0 y 2 − sin(θ) cos(θ) 0 y 2
1 − 1 y 2 xy 2 Notation involving z: We shall use the notation z = − 1 ∈ 0 y 2 G/K and also z = x + iy ∈ h interchangeably. The usage will depend on the context of the discussion.
Definition 3.1. (Petersson inner product on V ) For φ1, φ2 ∈ V we define the inner product Z × φ1, φ2 := φ1(g) φ2(g) d g
Γ\G/K where d×g denotes the left invariant measure on the coset space Γ\G/K. 1 1 y 2 xy− 2 If z = 1 ∈ G/K then the Petersson inner product can be 0 y− 2 written in a simple explicit manner as ZZ dxdy φ , φ := φ (z) φ (z) . 1 2 1 2 y2 z∈Γ\h We may then define L2 (Γ\G/K) as the Hilbert space completion of V. This space had been studied by Maass and elements of L2 (Γ\G/K) which are eigenfunctions of the Laplacian with rapid decay at the cusp are termed Maass forms for SL(2, Z). 2 2 Definition 3.2. (The space Lcusp of cusp forms) We let Lcusp denote the Hilbert space of Maass forms for SL(2, Z). It can be shown that each Maass form φ vanishes at the cusp, i.e., lim φ(x + iy) = 0. y→∞ It was not known before Selberg’s work on the trace formula whether infinitely many such Maass forms existed or not. In addition to the Maass forms there are are also Eisenstein series which are eigenfunc- tions of the Laplacian but are not in L2. 1 m Definition 3.3. Let Γ := . Let s ∈ with Re(s) > 1. ∞ 0 1 C m∈Z We define the Eisenstein series 1 X E(z, s) := Im(γz)s. 2 γ∈Γ∞\Γ 8 DORIAN GOLDFELD
Theorem 3.4. (Fourier expansion of Eisenstein series) The Eisen- stein series E(z, s) has the Fourier expansion
s√ 2π y X 1 s 1−s s− 2 2πinx E(z, s) = y +M(s)y + σ1−2s(n)|n| Ks− 1 (2π|n|y)e Γ(s)ζ(2s) 2 n6=0 where 1 √ Γ s − ζ(2s − 1) X M(s) = π 2 , σ (n) = ds, Γ(s) ζ(2s) s d|n d>0 Z ∞ 1 − 1 y(u+ 1 ) s du Ks(y) = e 2 u u . 2 0 u Proof. See [Gol06]. Corollary 3.5. (Growth of the Eisenstein series at the cusp) Let s ∈ C with with Re(s) > 1. Then |E(z, s)| yRe(s) and |E(z, s)−ys| y1−Re(s) as y → ∞.
Definition 3.6. (The kernel function Kf ) Let f : K\G/K → C. For z, z0 ∈ SL(2, R) we define the kernel function
0 X −1 0 Kf (z, z ) := f z γz , γ∈Γ provided the sum converges absolutely.
0 Proposition 3.7. (Properties of Kf ) For all z, z ∈ SL(2, R) the Kernel function Kf (given in the above definition) satisfies the follow- ing properties:
0 0 0 0 • Kf (zk, z k ) = Kf (z, z ), (∀ k, k ∈ K), 0 0 0 • Kf (γz, γ z) = Kf (z, z ), (∀ γ, γ ∈ Γ).
Proof. Exercise for the reader.
To facilitate the computation of the trace of the linear operator Kf Selberg chose a class of especially nice test functions f which we now describe.
Definition 3.8. (The function τ) We define the function τ : G → C by
2 2 2 2 a b τ(g) := a + b + c + d − 2, for g = ( c d ) ∈ G . TRACE FORMULAE FOR SL(2, R) 9
1 − 1 y 2 xy 2 Proposition 3.9. (Properties of τ) Let z = − 1 , and let 0 y 2 1 1 ! 0 2 0 0− 2 0 y x y z = − 1 ∈ G/K. Then we have 0 y0 2 (i) τ(k) = 0, (∀ k ∈ K), (ii) τ(kz) = τ(zk) = τ(z), (∀ k ∈ K), (x − x0)2 + (y − y0)2 |z − z0|2 (iii) τ z−1z0 = = , yy0 yy0 |σz − σz0|2 |z − z0|2 (iv) = for all σ ∈ SL(2, ). Im(σz) · Im(σz0) yy0 R Proof. Exercise for the reader.
Definition 3.10. (Selberg’s kernel function) Let f : R+ → C be −1− 0 a smooth function satisfying f(t) |t + 2| . Then for g, g ∈ SL(2, R), we define 0 X −1 0 Kf (g, g ) = f τ g γg . γ∈Γ Theorem 3.11. (Growth of Selberg’s kernel function at the 1 1 1 1 ! − 0 2 0 0− 2 y 2 xy 2 0 y x y cusps ) Let z = − 1 , z = − 1 ∈ G/K. Then, 0 y 2 0 y0 2 for every > 0 and y, y0 > 1, we have 0 X −1 0 − 0 1+ • |Kf (z, z )| ≤ f τ z γz y (y ) γ∈Γ
X −1 0 0 − • f τ z γz (yy ) . γ ∈ Γ−Γ∞ Proof. We compute (x − x0)2 + (y − y0)2 f τ z−1z0 = f yy0 (x − x0)2 y y0 −1− + + yy0 y0 y (x − x0)2 y −1− + yy0 y0 10 DORIAN GOLDFELD
0 Next, for α ∈ Γ, we adopt the notation that zα = αz = xα + iyα. It follows that 2 ! 0 X X |z − zδα| Kf (z, z ) = f y yδα α∈Γ∞\Γ δ∈Γ∞ 2 ! X X |z − zα − m| = f y yα α∈Γ∞\Γ m∈Z 2 2 X X (x − xα − m) + (y − yα) = f y yα α∈Γ∞\Γ m∈Z 2 2 −1− X X (x − xα − m) y + y yα yyα α∈Γ∞\Γ m∈Z −1− 1+ X 1+ X 2 2 y yα (x − xα − m) + y α∈Γ∞\Γ m∈Z Now, we can bound the latter sum above as follows. −1− Z ∞ X 2 2 −2−2 2 2 −1− (x − xα − m) + y y + (u + y ) du 0 m∈Z y−1−2. It immediately follows from corollary 3.5 and the above calculations that 0 − 0 − 0 1+ Kf (z, z ) y E(z , 1 + ) y (y ) . This proves the first assertion. For the second assertion we calculate the partial sums over Γ∞\Γ X −1 0 − X 1+ 0 − f τ z γz y yγ (yy ) γ ∈ Γ−Γ∞ Γ∞γ 6= Γ∞ 1+ − since E(z, 1 + ) − y y by corollary 3.5.
Definition 3.12. (The integral operator Kf ) Let z ∈ h. Then for 2 2 2 φ ∈ L (Γ\h) we define the integral operator Kf : L (Γ\h) → L (Γ\h) by Z dx0dy0 K φ(z) = K (z, z0) φ(z0) . f f (y0)2 Γ\h