MATH G9905 RESEARCH SEMINAR IN NUMBER THEORY (FALL 2014) LECTURE 4 (SEPTEMBER 30, 2014) XIAOQING LI UPPER BOUNDS FOR A RANKIN–SELBERG TYPE INTEGRAL NOTES TAKEN BY PAK-HIN LEE Abstract. In this talk, we will give an almost sharp upper bound for a Rankin–Selberg type integral involving the Arthur truncated Eisenstein series on GL(n). Langlands–Shahidi constant formula, Maass–Selberg relations and other tools on higher rank groups will be used. This is a joint work with Goldfeld. We will generalize Sarnak’s method to derive a logarithmic zero free region for Rankin– Selberg L-functions on GL(n). His idea is to consider the integral 2 Z ∞Z 1 × 1 2 ˆ d z |ζ(1 + 2it)|. + it |ζ(1 + 2it)| E z, A 2 η 0 How can we generalize this integral to GL(2n)? A natural generalization is 2 Z 1 Z ∞Z ∞ Z ∞ Z 1 1 2 EˆA z, + it d× z ··· |L(1 + it, f × f )| ··· 2 0 η1 η2 η2n−1 0 ˆ involves sums of the form but R1 R 1of EA P this doesn’t work because the Fourier expansion ˆ 2 γ∈U2n−1 \ SL2n−1 . A better solution is to replace 0 · · · 0 |EA | by a certain Rankin–Selberg integral. n×n ∗ Let G = GL(2n, R), Γ = SL(2n, Z) and K = O(2n, R). Let Pn,n (Z) = ⊂ 0 n×n In×n ∗ Γ be the maximal parabolic, N P = be the unipotent radical, and M P = 0 In×n n×n 0 be the standard Levi. 0 n×n The cuspidal Eisenstein series is X | det m1 | ns E(z, s) := f (m1 )f (m2 ) | det m2 | γ γ∈Pn,n \Γ where m1and m2 come from the Iwasawa decomposition of z = nmk with n ∈ N P , m = m1 ∈ M P and k ∈ K, and f is a fixed Hecke–Maass form for GL(n, Z). m2 Last updated: October 6, 2014. Please send corrections and comments to [email protected]. 1 Arthur’s truncated Eisenstein series is EˆA (z, s) := E(z, s) − X cP E(z, s)|γ . γ∈Pn,n \Γ h(z)|γ ≥A m1 , and cP E is the constant term of E along Pn,n . Here h(z) = det det m2 There is no obvious connection between the Fourier expansions of EˆA and E. The idea is to smooth out the sharp truncation, so we introduce a smooth version of Arthur’s truncated Eisenstein series n 1 Λ(2ns − 2n, f × f ) ∗ EˆA (z, s) := E(z, s) − A 2 E z, s − + E(z, 2 − s) 2 Λ(1 + 2ns − 2n, f × f ) n X A2 ns − 1− f (m1 )f (m2 ) h(z) n h(z) 2 γ γ∈P \Γ h(γ·z)≥A Λ(2ns − 2n, f × f ) − Λ(1 + 2ns − 2n, f × f ) X n(2−s) h(z) γ∈P \Γ h(γ·z)≥A n A2 f (m1 )f (m2 ) . 1− n h(z) 2 γ The truncated sum can be rewritten as Z n n X A− 2 w w A2 1 ns E z, s + dw h(z) f (m1 )f (m2 ) = 1− n 2πi (2) w(w + 1) 2 h(z) 2 γ γ∈P \Γ h(γ·z)≥A which is smooth. Note that Z (2) 10 x−w dw = w(w + 1) ( 1 1−x 0 if x > 1, if x ≤ 1. 2 Let β := tn , δ := β −1 (log log t)n , and g(x), ψ(x) ∈ Cc∞ ([1, 2]) be test functions. Define Z ∞ 2 Z A dA det z ∗ ˆ I := |L(1 + 2int, f × f )| · EA (z, 1 + it)g ψ d× z β A δ P2n−1,1 \η 2n 0 (2n − 1) × (2n − 1) ∗ where P2n−1,1 is the maximal parabolic and η 2n = GL(2n, R)/K· 0 ∗ R× . Our main theorem is Theorem 1. 1 1 I f δ − 2 β 2 +n (log t)2 as t → ∞. This bound is sharp modulo some log terms. Next time we will get a lower bound. Recall that if φ is a Maass form, then the integral Z |φ|2 EP2n−1,1 (z, s)d× z SL(2n,Z)\η 2n is roughly equal to the Rankin–Selberg L-function Λ(s, φ × φ). 2 The maximal parabolic Eisenstein series is X EP2n−1,1 (z, w) := (det z)w |γ . γ∈P2n−1,1 \ SL2n Unfolding gives I = |L(1 + 2int, f × f )| Z ∞ 2 Z Z dA 1 A −w ˆ∗ (z, 1 + it)g EP2n−1,1 (z, w)d× zdw. · ψ̃(−w)δ · E A 2πi (2) β A SL2n (Z)\η 2n 0 Shifting the w-line to 21 , we pick up the pole of EP2n−1,1 (z, w) at w = 1 and write I = I1 + I2 , where I1 = c|L(1 + 2int, f × f )|ψ̃(−1)δ −1 Z SL(2n,Z)\η Z 2n ∞ 0 2 A ˆ∗ dA × g EA (z, 1 + it) d z β A for some constant c, and 1 I2 = |L(1 + 2int, f × f )| · 2πi Z ψ̃(−w)δ −w Z ( 21 ) SL2n (Z)\η Z 2n ∞ 0 2 · · · EP2n−1,1 (z, w)d× zdw. We want to bound these two integrals. Proposition 2. Suppose F is an automorphic function for SL(k, Z), k ≥ 4. Then F has a Fourier expansion 1 0 · · · 0 u1k Z 1 Z 1 1 · · · 0 u2k .. . . .. z d× u F (z) = ··· F . . . 0 0 1 u k−1,k 1 + X X 1≤l≤k−2 mk−1 6=0 ··· X Z X Z 1 ··· mk−l 6=0 γ∈P̃k−1,l \ SLk−1 1 0 ··· 1 · · · F 1 0 u1,k−l · · · 0 u2,k−l · · · .. . 0 0 u1,k−1 u2,k−1 .. . u1k u2k .. . 1 uk−1,k 1 γk−1 z 1 · e(−mk−1 uk−1,k − · · · − mk−l uk−l,k−l+1 )d× u Z 1 Z 1 X X X + ··· ··· mk−1 6=0 m1 6=0 γ∈Uk−1 \ SLk−1 0 0 3 1 u12 · · · u1,k−1 u1k 1 · · · u2,k−1 u2k γ . . .. . . F z . . . 1 1 uk−1,k 1 · e(−mk−1 uk−1,k − · · · − m1 u12 )d× u. Here P̃k−1,l SLk−1−l ∗ ∗ · · · 1 ∗ ··· 1 ··· = .. . ∗ ∗ ∗ . ∗ 1 The last term is called the nondegenerate term ND(F ). For details, see Goldfeld and Ichino–Yamana. Proposition 3 (Langlands–Shahidi). E(z, s) has constant term ! det m1 n(1−s) det m1 ns f (m1 )f (m2 ) + cs f (m1 )f (m2 ) cP := det m2 det m2 with Λ(2ns − n, f × f ) Λ(1 + 2ns − n, f × f ) along Pn,n . Along other parabolics, the constant terms are 0. cs := Corollary 4. E(z, s) = ! det m1 ns det m1 n(1−s) ∗ + c f (m )f (m ) s 1 2 det m2 det m2 X γ∈P̃2n−1,n−1 \ SL2n−1 + ND(E). ( γ 1) Here X B(m2n−1 , · · · , mn+1 ) WJ (M m2 ), Qn−1 k(n−k) 2 m2n−1 6=0 mn+1 6=0 k=1 m2n−k where B are the Fourier coefficients of f and WJ is the Jacquet–Whittaker function. f ∗ (m2 ) = X ··· Proposition 5. Let X A(m1 , · · · , mk−1 ) F (z) := ··· WJ (M z) . Qk−1 l(k−l) γ 2 mk 6=0 γ∈Uk−1 \ SLk−1 m1 ≥1 l=1 |ml | ( 1) X k(k−1) X √ If maxi yi ≥ (t1+ ) 2 , mini yi ≥ 23 , and the Langlands parameters (α1 , . · · · , αk−1 ) satisfy | Im αi | t, then F (z) (maxi yi )−N for N very big. On a compact Riemannian surface, F (z) |λ| for the spectral eigenvalue λ. On a non-compact surface, this is not true. 4 The Jacquet–Whittaker function has rapid decay Wk,ν (z) (max yi )−N i under the above conditions. Lemma 6 (Bump–Friedberg–Hoffstein). Iν (g · z) = k−2 Y ! Pk−1 kek−i gz ∧ · · · ∧ ek−1 gzk−kνk−i−1 | det z| i=0 iνk−i i=0 for g ∈ SL(k, R). Here Iν (z) = k−1 Y k−1 Y b νj yi ij i=1 j=1 ( ij if i + j ≤ k, with bij = (k − i)(k − j) if i + j ≥ k, and ei = (0, · · · , 0, 1, 0, · · · , 0). Note that kek−i M ∧ · · · ∧ ek M k2 is the sum of squares of all the (i + 1) × (i + 1) minors of the last i + 1 rows of M . This bound I1 . To bound I2 we need to use the maximal parabolic Eisenstein series EP2n−1,1 (z, w) which is essentially the Epstein zeta function. The completed Eisenstein series EP∗ 2n−1 (z, w) = π −nw Γ(nw)ζ(2nw)EP2n−1,1 (z, w) has integral representation Z ∞ Z ∞ 1 1 1 n(1−w) du nw du (θw0 t z−1 w0 (u) − 1)u + − + . (θz (u) − 1)u u u n 1−w w 1 1 Here w0 is the long element in the Weyl group, and the theta function is X 2 2 2 e−π(b1 +b2 +···+b2n )u θz (u) = (a1 ,··· ,a2n )∈Z2n where b 1 = a1 Y 1 , b2 = (a1 x12 + a2 )Y2 , .. . b2n = (a1 x1,2n + a2 x2,2n + · · · + a2n )Y2n , and 1 Yk = y1 · · · y2n−k (y12n−1 · · · y2n−1 )− 2n . √ Proposition 7. For yi ≥ 23 , 1 ≤ i ≤ 2n − 1, Re w = 12 , we have X 1 2n−k 12 k−1 EP2n−1,1 (z, w) (y1 y22 · · · y2n−k ) (y2n−k+1 · · · y2n−1 ) 2 2≤k≤2n 2n−k +(y1 y22 · · · y2n−k ) k−1 2n 5 k−1 (y2n−k+1 · · · y2n−1 ) 2n−k+1 2n ln y1 · · · y2n−1 . Recall the Maass–Selberg relation r+s−1 r−s s−r 1−r−s A A A A 2 + cs + cr + cr cs . hEˆA (·, r), EˆA (·, s)i = |hf, f i| r+s−1 r−s s−r 1−r−s We can apply this to the smooth truncation above, which is just a linear combination of Arthur’s truncation. We need an upper bound for the Rankin–Selberg L-function. Lemma 8. L(1 + 2int, f × f ) n,f log(|t| + 1) and L0 (1 + 2int, f × f ) n,f log2 (|t| + 1). These are all the ingredients of the proof. 6
© Copyright 2025 Paperzz