Acta Mathematica Sinica, New Series 1997, Oct., Vol.13, No.4, pp. 443–452 On the Third and Fourth Power Moments of Fourier Coefficients of Cusp Forms Cai Yingchun (Department of Mathematics, Shandong Normal University, Jinan 250014, China) Abstract The asymptotic formulae for the third and fourth power moments of Fourier coefficients of cusp forms are proved in this paper. Keywords Cusp form, Fourier coefficients, The estimation of mean value, Asymptotic formula 1991MR Subject Classification 11F Chinese Library Classification O156 1 Introduction Let a(n) be the nth Fourier coefficient of a cusp form of weight ω = 2m(m ≥ 6) for the full modular group. In 1974 Deligne[1] proved the following deep result: a(n) n ω−1 2 d(n) (1) where d(n) denotes the divisor function, and the constant in is absolute. In fact this result is best possible and is one of the crowning achievements of mathematics. Assume that the a(n)’s are real and let a(n). A(x) = n≤x Some mathematicians have made contributions to the estimation of A(x). In 1973 Joris[2] proved: ω 1 (2) A(x) = Ω± (x 2 − 4 log log log x). √ In 1990 Ivic[3] showed that the interval [T, T + C T ] contains two points t1 and t2 such that ω ω −1 −1 (3) A(t1 ) > Bt12 4 , A(t2 ) < −Bt22 4 where B > 0, C > 0. By some similarity between A(x) and (x) (the error term in the Dirichlet divisor problem) it is conjectured that ω−1 1 (4) A(x) = O(x 2 + 4 +ε ) Received February 23, 1995, Revised March 12, 1996, Accepted May 6, 1996 444 Acta Mathematica Sinica, New Series Vol.13 No.4 should hold. Jutila[4] proved a truncated formula for A(x), that is, √ ω 1 ω 1 ω 1 1 π A(x) = √ x 2 − 4 a(n)n− 2 − 4 cos 4π xn − + O(x 2 +ε N − 2 ) 4 2π n≤N (5) where 1 N x. By (1), (5) and a trivial estimation one gets ω 1 A(x) x 2 − 6 +ε . (6) 1990 Ivic[5] proved the following square mean value formula for A(x), X 1 A2 (x)dx = CX ω+ 2 + B(X), (7) 1 where C= ∞ 1 1 a2 (n)n−ω− 2 , (4ω + 2)π 2 n=1 B(X) = O(X ω log5 X), 3 1 (log log log X) . B(X) = Ω X ω− 4 log X Ivic[3] also proved the following upper bound estimation for the eighth power moment of A(x), that is, X A8 (x)dx X 4ω−1+ε . (8) 1 Thus in a moderate sense the conjecture (4) is true, one can easily derive this result from (7) or (8). As a corollary of (7) Ivic[3] proved the following sharpening of (6), that is, ω 1 A(x) x 2 − 6 log2 x. By the large value technique, Ivic[3] proved X ω−1 1 |A(x)|B dx X 1+B( 2 + 4 )+ε , 1 (9) 0 ≤ B ≤ 8. (10) Inspired by [5], in this paper we shall prove the following third and fourth power moments for A(x), that is, Theorem 1 X 3ω 1 3ω 5 A3 (x)dx = c1 X 2 + 4 + O(X 2 + 28 +ε ), 1 √ where c1 = Theorem 2 2 π (6ω + 1) √ 3 √ √ n+ m= k 1 X 1 ω a(m)a(n)a(k)(mnk)− 4 − 2 . 1 A4 (x)dx = c2 X 2ω + O(X 2ω− 23 +ε ), Cai Yingchun On the Third and Fourth Power Moments of Fourier Coefficients of Cusp Forms where c2 = 3 64π 4 ω √ ω √ √ √ m+ n= k+ l 445 1 a(m)a(n)a(k)a(l)(mnkl)− 2 − 4 . As a corollary of Theorem 2 and (7) we get the following sharpening of (10) for 0 ≤ B ≤ 4, that is, X ω−1 1 Corollary |A(x)|B dx X 1+B( 2 + 4 ) , 0 ≤ B ≤ 4. 1 The proofs of Theorems 1 and 2 are based on the Voronoi-type formula (5) for A(x). In the proofs we have not made attempt to get the best results for the O terms. 2 Some Lemmas Lemma 1[1] Lemma 2[4] a(n) n ω−1 2 d(n). ω 1 √ ω 1 ω 1 x 2 −4 π A(x) = √ a(n)n− 2 − 4 cos 4π nx − + O(x 2 +ε N − 2 ), 4 2π n≤N (1 N x). Lemma 3[5] If m, n, k are natural numbers such that √ m+ √ √ n = k then √ √ √ 1 3 | m + n − k| ≥ (max(m, n, k))− 2 . 27 Lemma 4[5] If m, n, k, and l are natural numbers such that √ √ √ √ √ √ √ √ m + n = k + l or m + n + k = l, then √ √ √ √ 2 | m + n − k − l| (max(m, n, k, l))− 7 or √ √ √ √ 2 | m + n + k − l| (max(m, n, k, l))− 7 . Lemma 5[5] . For any real numbers α = 0, β and 0 < δ < 12 , we have 1 1 1 3 Kδ + |α| 3 K 2 + |α|− 2 K 4 , K<k≤2K √ α k+β<δ the implied constant in being absolute. The following Lemma 6 is well known. Lemma 6 If g(x) and h(x) are continuous real-valued functions of x and g(x) is monotonic then v b g(x)h(x)dx max |g(x)| max h(x)dx . a≤x≤b a≤u<v≤b a u 446 Acta Mathematica Sinica, New Series Vol.13 No.4 The Third Power Moment of A(x) 3 Let ω 1 √ ω 1 π x 2 −4 a(n)n− 2 − 4 cos 4π xn − , (x, N ) = √ 4 2π n≤N ω 1 R(x, N ) = O(x 2 +ε N − 2 ), 1 N x. For simplicity, put ω 1 r = r(m, n, k) = a(m)a(n)a(k)(mnk)− 2 − 4 , m, n, k ≤ N and r = 0, otherwise. 9 14 In this section 2 ≤ H ≤ X 2 ,N = H . By the elementary formula cos A cos B cos C = 1 (cos(−A + B + C) + cos(A − B + C) + cos(A + B − C) + cos(A + B + C)), 4 we have 3 where (x, N ) √ √ √ √ 3 3 = √3 3 x 2 ω− 4 r cos 4π x( m + n − k) − π 4 8 2π √ √ √ √ 3ω 3 r cos 4π x( m + n + k) − 34 π + √1 3 x 2 − 4 8 2π = 0+ 1+ 2 0 3ω 3 = √3 3 x 2 − 4 8 2π 1 3ω 3 = √3 3 x 2 − 4 8 2π 2 3ω 3 = √1 3 x 2 − 4 8 2π Thus 2H √ √ √ m+ n= k r, √ √ √ √ π r cos 4π x( m + n − k) − , 4 √ √ √ m+ n= k √ √ √ √ r cos 4π x( m + n + k) − 34 π . 2H 3ω 3 3 dx = √ 3 x 2 − 4 dx. r 0 8 2π H H By Lemma 1 or ω m,n, k>N √ √ √ m+ n= k 1 a(m)a(n)a(k)(mnk)− 2 − 4 √ √ √ m+ n= k m,n, k>N or 3 d(m)d(n)d(k)(mnk)− 4 Cai Yingchun On the Third and Fourth Power Moments of Fourier Coefficients of Cusp Forms 447 m>N √ √ √ m+ n= k + 3 d(m)d(n)d(k)(mnk)− 4 , k>N √ √ √ m+ n= k 3 d(m)d(n)d(k)(mnk)− 4 m>N √ √ √ m+ n= k √ √ √ 1 ( k = m + n (mn) 4 ) 3 (mnk)− 4 +ε m>N √ √ √ m+ n= k 3 m>N 3 (mn)− 4 +ε (mn)− 8 +ε 3 k>N √ √ √ m+ n= k r= 2H 3 (mnk)− 4 +ε k>N √ √ √ m+ n= k 3 1 (mnk)− 4 +ε N − 8 +ε . m> N √ 10 √ √ m+ n= k ω 1 1 a(m)a(n)a(k)(mnk)− 2 − 4 + O(N − 8 +ε ), H 1 m>N d(m)d(n)d(k)(mnk)− 4 Hence 9 m− 8 +ε N − 8 +ε , ω 1 √3 3 a(m)a(n)a(k)(mnk)− 2 − 4 0 dx = 8 2π +O(H 3ω 1 2 +4 2H x 3ω 3 2 −4 dx (3.1) H 1 N − 8 +ε ). By Lemma 6 2H H Hence 2H H 2 dx √ xp cos(A x + B)dx (A = 0) √ A √ cos(A x + B) dx = 2A x 2 x H v √ 1 A p+ 12 −1 √ cos(A x + B)dx H p+ 2 A−1 . A max H H≤u<v≤2H u 2 x 2H −1 p+ 12 H 3ω 1 2 −4 H 3ω 1 2 −4 H 3ω 1 2 −4 (3.2) r √ √ √ m+ n+ k 3 (mnk)− 4 +ε √ √ √ m+ n+ k m,n,k≤N 11 (mnk)− 12 +ε H √ √ √ 1 1 ( m + n + k ≥ (mnk) 6 ) 3 3ω 1 2 −4 1 N 4. m,n,k≤N For simplicity, let = √ √ √ 3 m + n − k. When | | ≥ H − 7 by (3.2) we have 2H x H 3ω 3 2 −4 √ 3ω 1 π cos 4π x − dx H 2 − 4 −1 , 4 (3.3) 448 Acta Mathematica Sinica, New Series Vol.13 No.4 3 and when | | < H − 7 we have the trivial upper bound 2H x 3ω 3 2 −4 H √ 3ω 1 π cos 4π x − dx H 2 + 4 . 4 (3.4) We dedeuce from (3.3) and (3.4) that 2H H H 1 dx 3ω 1 2 −4 r −1 +H 3 ||≥H − 7 ||<H H 3ω 1 2 −4 3 1 =H (M1 + M2 ) + H r 3 7 3ω 1 r −1 +H 2 + 4 + H − 7 <||<m 4 3ω 1 2 −4 3ω 1 2 +4 (3.5) r 3 1 ||<H − 7 ||>m 4 3ω 1 2 +4 M3 . 3 for given m√≤ n ≤ N there is at most one k such that | | < H − 7 , since √ For√M3 and √ k = m + n − and | N | = o(1). Such k, if exists, must be greater than n. By Lemma 3 3 3 2 4, 0 < | | < H − 7 implies n− 2 H − 7 , that is, n H 7 ; hence 3 3 5 1 (mn)− 4 n− 4 H ε n− 4 H − 14 +ε . (3.6) M3 H ε m≤n≤N 2 nH 7 In the sum M1 , for m ≤ n ≤ N, √ √ √ k = ( m + n)2 + O(| | n) > n, so there are √ 1+|| m such k. Hence Hε M1 √ 3 3 (mn)− 4 | |−1 (1 + | m|)m− 4 m≤n≤N Hε H Finally, M2 3 3 3 n− 4 m− 2 H 7 + H ε m≤n≤N 3 ε 7 +ε 1 3 n− 4 m−1 (3.7) m≤n≤N 3 + H N 4 H 7 +ε . r| |−1 H ε 3 1 1 (mnk)− 4 m− 4 H ε N 2 . (3.8) m,n,k≤N 1 ||>m 4 From (3.5), (3.6), (3.7) and (3.8) we deduce that 2H H 1 dx H 3ω 5 2 + 28 +ε . (3.9) Combiming (3.1), (3.2) and (3.9) we get 2H H 3 (x, N )dx = c1 ((2H) 3ω 1 2 +4 −H 3ω 1 2 +4 ) + O(H 3ω 5 2 + 28 +ε ). (3.10) Cai Yingchun On the Third and Fourth Power Moments of Fourier Coefficients of Cusp Forms 449 By the same but simpler arguments it can be proved that 2H H 2 1 (x, N )dx H ω+ 2 . (3.11) Since (a + b)3 = a3 + O(a2 |b| + |b|3 ), it follows from (3.10) and (3.11) that 2H 3 A (x)dx 2H = H H 2H = 3 (x, N )dx + O H ω 9 2 − 28 +ε 2H 2 (x, N )dx + O(H 3ω 1 2 − 28 +ε 3 (x, N )dx + O(H 3ω 5 2 + 28 +ε ). H (3.12) Lastly it is a routine argument to deduce from (3.10) and (3.12) that X 1 4 A3 (x)dx = c1 X 3ω 1 2 +4 + O(x 3ω 5 2 + 28 +ε ). The Fourth Power Moment of A(x) For simplicity, put ω 1 r1 = r1 (m, n, k, l) = a(m)a(n)a(k)a(l)(mnkl)− 2 − 4 , r1 = 0, m, n, k, l ≤ N, otherwise, and all the notations are the same as in Section 3 except 27 N = H 46 . By the elementary formula cos A cos B cos C cos D = 81 (cos(−A + B + C + D) + cos(A − B + C + D) + cos(A + B − C + D) + cos(A + B + C − D) + cos(−A + B + C − D) + cos(A − B + C − D) + cos(A + B − C − D) + cos(A + B + C + D)), we have 4 (x, N ) ) H √ √ √ √ 2ω−1 √ = x 4 r1 cos 4π x( m + n + k − l) − π 2 8π √ √ 2ω−1 √ √ √ r1 cos(4π x( m + n − k − l)) + 3x 4 32π √ √ 2n−1 √ √ √ +x 4 r1 cos(4π x( m + n + k + l) − π) 32π = 3+ 4+ 5+ 6 450 Acta Mathematica Sinica, New Series where 3 4 5 2ω−1 = 3x 4 32π √ √ √ √ m+ n= k+ l 2ω−1 = 3x 4 32π √ √ √ √ m+ n= k+ l r1 , √ √ √ √ √ r1 cos(4π x( m + n − k − l)), √ √ √ √ 2ω−1 √ =x 4 r1 cos 4π x( m + n + k − l) − π 2 , 8π 2ω−1 6 Vol.13 No.4 = −x 4 32π √ √ √ √ √ r1 cos(4π x( m + n + k + l)). The following arguments are similar to that of [5] Section 3.√ √ √ √ For natural numbers m, n, k and l, the condition m+ n = k+ l holds iff (m, n) = (k, l) or m, n, k, l all have the same square-free part, h, say, such that m = α2 h, n = β 2 h, k = θ2 h, l = ϕ2 h and α + β = θ + ϕ. So in the sum ω 1 = a(m)a(n)a(k)a(l)(mnkl)− 2 − 4 √ √ √ √ m+ n= k+ l m,n,k,l≤N if let the variables m, n, k and l run over all natural numbers, then by Lemma 1 the error thus induced is ∞ 3 k=1 n>N N 3 n− 2 k− 2 d2 (n)d2 (k) + − 12 +ε + α − 32 α,β,θ,ϕ≥1 α2h >N d(α2 )h−3 d4 (h) α2 h>N 1 3 (αβθϕ)− 2 h−3 d(α2 )d(β 2 )d(θ2 )d(ϕ2 )d4 (h) 1 1 N − 2 +ε + N − 4 +ε N − 4 +ε . Thus 2H 3 H The treatment of 3 32π 4 dx = √ √ √ √ m+ n= k+ l is similar to that of 6 2H H r1 2 2H 1 x2ω−1 dx + O(H 2ω N − 4 +ε ). (4.1) H in Section 3. Applying (3.2) we have √ √ √ √ √ x2ω−1 cos(4π x( m + n + k + l))dx √ √ √ 1 √ H 2ω− 2 ( m + n + k + l)−1 . Hence by Lemma 1 2H H 6 dx 1 H 2ω− 2 1 H 2ω− 2 +ε √ √ √ √ r1 ( m + n + k + l)−1 m,n,k,l<N √ √ √ 3 √ (mnkl)− 4 ( m + n + k + l)−1 √ √ √ √ 1 ( m + n + k + l ≥ 41 (mnkl) 8 ) 7 1 1 1 (mnkl)− 8 H 2ω− 2 +ε N 2 . H 2ω− 2 +ε m,n,k,l<N (4.2) Cai Yingchun On the Third and Fourth Power Moments of Fourier Coefficients of Cusp Forms 451 √ √ √ √ 21 . Similarly to (3.2), when | | ≥ H −c3 we have m + n + k − l and c3 = 46 1 2H √ √ √ √ √ 1 π dx H 2ω− 2 | 1 |−1 . x2ω−1 cos 4π x( m + n + k − l) − 2 H Let 1 = When | 1 | < H −c3 we have the trivial bound 2H √ √ √ √ √ π x2ω−1 cos 4π x( m + n + k − l) − dx H 2ω . 2 H Thus 2H H 5 dx H 2ω 1 r1 + H 2ω− 2 0<|1 |<H −c3 1 r1 | 1 |−1 = H 2ω M4 + H 2ω− 2 M5 , (4.3) |1 |≥H −c3 say. First consider M4 . Without loss of generality we may assume m ≤ n ≤ k ≤ N. The condition | 1 | ≤ H −c3 implies √ √ √ √ √ l ≤ m + n + k + | 1 | ≤ 4 k, so that √ √ √ √ √ √ √ √ √ | m + n + k + l| | m + n + k − l| ≤ 7 kH −c3 , that is, √ √ √ √ |1 − (m + n + k + 2( mn + nk + mk))| ≤ 7 kH −c3 . Since (4.4) √ √ 7 kH −c3 ≤ 7 N H −c3 = o(1), there will be no natural number l for which 0 < | l | < H −c3 unless √ √ √ √ 0 < ||2( mn + nk + mk|| < 7 kH −c3 . (4.5) In the latter case, there will be at most one eligible l. In view of (4.5), such an l must also satisfy k ≤ l ≤ 10k. Hence by Lemma 1, 3 3 3 (mnkl)− 4 N ε N ε (mn)− 4 k− 2 . M4 m≤n≤k≤N m≤n≤k≤N (4.5) 0<|1 |≤H −c3 √ √ √ √ Applying Lemma 5 with α = 2( m + n), β = 2 mn and δ = 7 2kH −c3 , we have 3 3 3 1 1 1 3 k− 2 K − 2 (K 2 H −c3 + n 6 K 2 + n− 4 K 4 ), (4.7) K<k<2K whence m≤n≤2K K≤k≤2K (4.5) 3 3 (mn)− 4 k− 2 3 1 1 3 (mn)− 4 (H −c3 + n 6 K −1 + n− 4 K − 4 ) m≤n≤2K 1 2 K H −c3 + K (4.8) − 13 +K − 12 1 2 K H −c3 + K 2 − 13 . 3 By Lemma 4, 0 < |1 | < H −c3 implies k− 7 H −c3 , since k ≤ l ≤ 10k. Hence H 7 c3 k N. We deduce from (4.7) and (4.8) that 1 2 1 M4 N ε (N 2 H −c3 + H − 21 c3 ) H ε− 23 . (4.9) 452 Acta Mathematica Sinica, New Series Vol.13 No.4 1 To estimate M5 , we split this sum into two sub-sums according to H −c3 < | 1 | ≤ k 4 1 and | 1 | > k 4 respectively. Following the same arguments for M2 and M3 in Section 3, we 3 find that these sub-sums are bounded by H c3 +ε and N 4 +ε respectively. Hence M5 H c3 +ε . Combining the estimations for M4 and M5 we get 2H 5 H 1 dx H 2ω− 23 +ε . (4.10) By a similar argument, it can be proved that 2H 4 H 1 dx H 2ω− 23 +ε . (4.11) Collecting (4.1), (4.2), (4.10) and (4.11) we obtain 2H 4 H 1 (x, N )dx = c2 ((2H)2ω − H 2ω ) + O(H 2ω− 23 +ε ). (4.12) By the elementary formula: (a + b)4 = a4 + O(|a|3 |b| + |b|4 ), we get 2H A4 (x)dx = H 2H 4 ω 1 (x, N )dx + O(H 2 +ε N − 2 ) H 2H H | (x, N )|3 dx + O(H 2ω+1 N −2 ). (4.13) Then by Holder’s inequality, this yields 2H H | (x, N )|3 dx 2H (x, N ) 4 34 dx H 14 2H dx H 3 1 H 2 ω+ 4 . Finally we obtain 2H H 1 A4 (x)dx = c2 ((2H 2ω ) − H ω ) + O(H 2ω− 23 +ε ). (4.14) Lastly it is a routine argument to deduce from (4.14) that 1 X 1 A4 (x)dx = c2 X 2ω + O(X 2ω− 23 +ε ). References 1 2 3 4 5 Deligne P. La conjecture de weil. Inst Hautes Etudes Sci Publ Math, 1974, 43: 273–307 Joris H. Ω-Satye fur gewisce multiplikative arithmetische Funktionen. Comment Math Helv, 1973, 48: 409–435 Ivic A. Large values of certain number-theoretic error term. Acta Arithmetica, LVI, 1990, 135–159 Jutila M. Rienann’s Zeta-function and the divisor problem. Ark Mat, 1983, 21:75–96 KAI-MAN TSANG. Higher-power moments of (x), E(t) and P (x). Proc London Math Soc, 1992, 61: 65–84
© Copyright 2026 Paperzz