7 0 0 2 ON THE FOURTH MOMENT IN n a THE RANKIN-SELBERG PROBLEM J 1 3 ] T Aleksandar Ivic´ N . h Abstract. If t a ∆(x) := cn−Cx m X n≤x [ denotes the error term in the classical Rankin-Selberg problem, then it is proved 1 that 2v Z X∆4(x)dx≪ε X3+ε, Z X∆41(x)dx≪ε X11/2+ε, 0 0 1 9 where ∆1(x)= 0x∆(u)du. The latter bound is, up to ‘ε’, best possible. 1 R 0 7 0 / h 1. Introduction and statement of results t a m The classical Rankin-Selberg problem consists of the estimation of the error : term function v i X (1.1) ∆(x) := c −Cx, r n a n≤x X where the notation is as follows. Let ϕ(z) be a holomorphic cusp form of weight κ with respect to the full modular group SL(2,Z), and denote by a(n) the n-th Fourier coefficient of ϕ(z). We suppose that ϕ(z) is a normalized eigenfunction for the Hecke operators T(n), that is, a(1) = 1 and T(n)ϕ = a(n)ϕ for every n ∈ N. The classical example is a(n) = τ(n) (κ = 12), the Ramanujan function defined by ∞ τ(n)xn = x (1−x)(1−x2)(1−x3)··· 24 (|x| < 1). n=1 X (cid:8) (cid:9) 1991 Mathematics Subject Classification. 11 N 37, 11 M 06, 44 A 15, 26 A 12. Key words and phrases. The Rankin-Selberg problem, Vorono¨ı type formula, Selberg class, mean square estimates. Typeset by AMS-TEX 1 2 Aleksandar Ivi´c The constant C(> 0) in (1) may be written down explicitly (see e.g., [7]), and c n is the convolution function defined by n 2 c = n1−κ m2(κ−1) a . n m2 mX2|n (cid:12) (cid:16) (cid:17)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) The classical Rankin-Selberg bound of 1939 is (1.2) ∆(x) = O(x3/5), hitherto unimproved. In theirworks, done independently, R.A.Rankin [10]derives (1.2) from a general result of E. Landau [9], while A. Selberg [12] states the result with no proof. Although it seems very difficult at present to improve the bound in (1.2), recently there have been some results on mean square estimates in the Rankin-Selberg problem (see the author’s works [5], [6]). Namely, let as usual µ(σ) denote the Lindel¨of function log|ζ(σ +it)| µ(σ) := limsup (σ ∈ R). logt t→∞ Then we have X 2 (1.3) ∆2(x)dx ≪ X1+2β+ε, β = . ε 5−2µ(1) Z0 2 Here and later ε denotes positiveconstants which may be arbitrarily small, but are not necessarily the same at each occurrence, while ≪ means that the ≪–constant ε depends on ε. Note that with the sharpest known result (see M.N. Huxley [2]) µ(1) ≤ 32/205 we obtain β = 410/961 = 0.426638917... . The limit of (1.3) is 2 the value β = 2/5 if the Lindel¨of hypothesis (that µ(1) = 0) is true. 2 We propose to contribute here to the subject of mean value results for ∆(x) by proving (unconditionally) the following results. THEOREM 1. For any given ε > 0 we have X (1.4) ∆4(x)dx ≪ X3+ε. ε Z0 Note that (1.4) follows from (1.3) only if the Lindel¨of hypothesis µ(1) = 0 is true. 2 Corollary. For any given ε > 0 we have (1.5) ∆(x) ≪ x3/5+ε. ε On the fourth moment in the Rankin-Selberg problem 3 The bound in (1.5)is only by an ‘ε’–factorweaker than the strongest known bound (1.2). To obtain (1.5) from (1.4) note that we have (see Lemma 1 below) 1 X+H (1.6) ∆(X) = ∆(x)dx+O(H) (Xε ≤ H ≤ 1X). 2H 2 ZX−H It follows from (1.6) by H¨older’s inequality for integrals that X+H ∆4(X) ≪ H−1 ∆4(x)dx+H4 ≪ H−1X3+ε +H4 ε ZX−H by (1.4), and (1.2) follows with H = X3/5. Note that if (1.4) holds with the exponent θ on the right-hand side of (1.4), then the above argument gives ∆(x) ≪ xθ/5, and the best possible exponent θ must satisfy θ ≥ 15/8 since ∆(x) = Ω (x3/8) ± (see the author’s work [4]). x THEOREM 2. If ∆ (x) = ∆(u)du, then for any given ε > 0 we have 1 0 R X (1.6) ∆4(x)dx ≪ X11/2+ε. 1 ε Z0 Note that it was proved in [7] that X 2 ∞ (1.7) ∆2(x)dx = (2π)−4 c2n−7/4 X13/4 +O (X3+ε), 1 13 n ε Z0 n=1 ! X so that from (1.7) and the Cauchy-Schwarz inequality for integrals we obtain that X ∆4(x)dx ≫ X11/2. 1 Z0 This shows that, up to ‘ε’, the bound in (1.6) is best possible. 4 Aleksandar Ivi´c 2. The necessary lemmas Inthissectionweshall statethelemmasnecessary fortheproofofourtheorems. LEMMA 1. For Xε ≤ H ≤ 1X we have 2 1 X+H (2.1) ∆(X) = ∆(x)dx+O(H). 2H ZX−H Proof. 1 X+H 1 X+H ∆(X)− ∆(x)dx = (∆(X)−∆(x))dx 2H 2H ZX−H ZX−H 1 X+H = c +C(X −x) dx n 2H   ZX−H x<n≤X X   ≪ c +H ≪ H, n X−H≤n≤X+H X where in the last step a well-known result of P. Shiu on multiplicative functions [14] in short intervals was used (see also Lemma 4 of [7]). The next two lemmas are the explicit, truncated formula of the Vorono¨ı type for for ∆(x) and ∆ (x), respectively. 1 LEMMA 2. For 1 ≪ K ≪ x a parameter we have 0 x3/8 (2.2) ∆(x) = c k−5/8sin 8π(kx)1/4 + 3π +O x3/4+εK−1/4 , 2π k 4 ε 0 kX≤K0 (cid:16) (cid:17) (cid:16) (cid:17) Choosing K = x3/5 and estimating trivially the sum in (2.2) we obtain again 0 the bound ∆(x) ≪ x3/5+ε. The formula (2.2) was proved (see [7, Lemma 2] ε with ρ = 0) by Ivi´c, Matsumoto and Tanigawa, where the general (Riesz) sum (x −n)ρc for fixed ρ ≥ 0 is investigated, and a proof of (1.2) is given. A n≤x n similar formula holds in general for Dirichlet series of degree four in the Selberg P class (see e.g., the survey article [8] of Kaczorowski-Perelli on functions from the Selberg class S, and Selberg’s original paper [13]). Thus the generalization of our results will hold for error terms associated to suitable Dirichlet series in S. However, in some cases the special structure of the problem at hand allows for sharper results. For example, consider the estimation of ∆ (x), the error term 4 in the asymptotic formula for the summatory function of the divisor function On the fourth moment in the Rankin-Selberg problem 5 d (n) = 1. The generating function in this case is ζ4(s), and we 4 abcd=n;a,b,c,d∈N have (see e.g., [3, Chapter 13]) P X (2.3) ∆2(x)dx ≪ X7/4+ε, 4 ε Z0 which is (up to ‘ε’) best possible in view of ∆ (x) = Ω (x3/8) (see [4]). Since one 4 ± has ∆ (x) ≪ x1/2+ε, it trivially follows from (2.3) that 4 ε X (2.4) ∆4(x)dx ≪ X11/4+ε, 4 ε Z0 and the exponent in (2.4) is better than the exponent in (1.4). LEMMA 3. For 1 ≪ K ≪ x2 a parameter we have 0 (2.5) x9/8 ∆ (x) = c k−7/8sin 8π(kx)1/4 + π +O x1+ε +x3/2+εK−1/2 . 1 (2π)2 k 4 ε 0 kX≤K0 (cid:16) (cid:17) (cid:16) (cid:17) Lemma 3 is the case ρ = 1 of [7, Lemma 2]. The chief ingredient in the proof of Theorem 2 is the new result of O. Robert–P. Sargos [11], which is the following LEMMA 4. Let k ≥ 2 be a fixed integer and δ > 0 be given. Then the number of integers n ,n ,n ,n such that N < n ,n ,n ,n ≤ 2N and 1 2 3 4 1 2 3 4 |n1/k +n1/k −n1/k −n1/k| < δN1/k 1 2 3 4 is, for any given ε > 0, (2.6) ≪ Nε(N4δ +N2). ε 3. Proof of Theorem 1 To prove (1.4), it is sufficient to prove the bound in question for the integral over the interval [X, 2X]. Let, for (3.1) X1/2 ≤ V ≪ X3/5, 2X (3.2) I(V,X) := ∆4(x)dx, ZX,V≤|∆(x)|<2V 6 Aleksandar Ivi´c where the upper bound in (3.1) holds because of (1.2). Clearly we have 2X (3.3) ∆4(x)dx ≪ X3 +logX max I(V,X). ZX X1/2≤V≪X3/5 Hence the problem is reduced to the estimation of I(V,X) in (3.2). Thus we fix a value V = C2−jX3/5(j = 0,1,2,... , C > 0) and split the interval [X, 2X] into subintervals of length H (X1/2 ≤ H ≤ X1−ε), where the last of these intervals may be shorter. Suppose there are R = R(V) of these subintervals which contain a point x for which V ≤ |∆(x)| < 2V holds. Further suppose that x is the point r in the r-th of these intervals where the largest value of |∆(x)| is attained. To obtain the spacing condition (3.4) |x −x | ≥ H (r 6= s; r,s = 1,... ,R) r s we consider separately the points with even and odd indices and, with a slight abuse of notation, each of these two systems of points is again denoted by {x }R . r r=1 Therefore we have |∆(x )| ≥ V (r = 1,2...), and observe that r x3/8 c k−5/8sin 8π(kx)1/4 + 3π ≪ δ3/8X1/2 2π k 4 k≤XδX1/3 (cid:16) (cid:17) for x = x ∈ [X, 2X], δ > 0. Thus for δ small enough it follows from (2.1) and r (2.2)(changingK toX3H−4 andrecallingthate(z) = e2πiz),thatforr = 1,... ,R 0 X3/8 xr+H/3 (3.5) V ≪ c k−5/8e(4π(kx)1/4) dx+HXε, k H Zxr−H/3 (cid:12)δX1/3≤Xk≤X3H−4 (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) where all the intervals [x −H/3, x +H/3] are disjoint in view of (3.4). We take r r in (3.5) H = VX−2ε, square, use the Cauchy-Schwarz inequality for integrals and sum the resulting expressions. We obtain 5X/2 2 R ≪ max X3/4+2εV−3 c k−5/8e(4π(kx)1/4) dx. ε k δX1/3≤K≤X3+8εV−4 ZX/2 (cid:12)K<Xk≤2K (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) To evaluate the integral on the right-hand side we square out the sum and use the first derivative test (i.e., Lemma 3.1 of [1] or Lemma 2.1 of [3]). It follows that, On the fourth moment in the Rankin-Selberg problem 7 since c ≪ nε, n ε 5X/2 2 c k−5/8e(4π(kx)1/4) dx k ZX/2 (cid:12)K<Xk≤2K (cid:12) (cid:12) (cid:12) (cid:12) 5X/(cid:12)2 = c c (k k )−5/8 e 4πx1/4(k1/4 −k1/4) dx k1 k2 1 2 1 2 K<kX1,k2≤2K ZX/2 (cid:16) (cid:17) X3/4+ε ≪ X1+εK−1/4 +K−5/4 ε 1/4 1/4 |k −k | K<k1X6=k2≤2K 1 2 1 ≪ X1+εK−1/4 +K−1/2X3/4+ε ε |k −k | 2 1 K<Xk1≤2KK<k2≤X2K,k26=k1 ≪ X1+εK−1/4 +X3/4+εK1/2 ≪ X3/4+εK1/2 ε ε for K ≫ X1/3, which is the case in (3.5). Consequently (3.6) R ≪ max X3/4+2εV−3X3/4+εK1/2 ≪ X3+εV−5. ε ε δX1/3≤K≤X3+8εV−4 From (3.6) we obtain that I(V,X) ≪ XεV ·V4R ≪ X3+ε, ε ε and (3.3) gives 2X ∆4(x)dx ≪ X3+ε. ε ZX Theorem 1 follows if we replace X by X2−j in the above bound and sum over j = 1,2,... . An alternative proof of (1.4) may be obtained by going through the appropriate modification of the proof of Theorem 13.8 in [3], taking k = 4 and R = R , T = T in (13.65). The choice (κ,λ) = (1, 1) will provide again the 0 0 2 2 bound in (3.6) for the relevant range. 4. Proof of Theorem 2 ToproveTheorem2weuse(2.5)ofLemma3withK = X forX/2 ≤ x ≤ 5X/2. 0 This gives (4.1) 2X 5X/2 4 ∆4(x)dx ≪ X9/2logX max ϕ(x) c k−7/8e(4(xk)1/4) dx 1 ε k K≪X ZX ZX/2 (cid:12)K<Xk≤2K (cid:12) (cid:12) (cid:12) +X5+ε, (cid:12) (cid:12) 8 Aleksandar Ivi´c where ϕ(x) (≥ 0) is a smooth function supported in [X/2, 5X/2] such that ϕ(x) = 1 for X ≤ x ≤ 2X and ϕ(r)(x) ≪ X−r (r = 0,1,...). If we set r ∆ := k1/4 +ℓ1/4 −m1/4 −n1/4 (k,ℓ,m,n ∈ N), then 5X/2 5X/2 ϕ(x)··· dx = c c c c (kℓmn)−7/8 ϕ(x)e(4π∆x1/4)dx. k ℓ m n Z K<k,ℓ,m,n≤2K Z X/2 X X/2 But integration by parts shows that 5X/2 5X/2 e(4π∆x1/4) ′ ϕ(x)e(4π∆x1/4)dx = − x3/4ϕ(x) dx. 2πi∆ XZ/2 XZ/2 (cid:16) (cid:17) Thus the exponential factor remained the same, but the order of the integrand has decreased by ∆−1X−1/4, provided that ∆ 6= 0. Since this will be repeated after every integration by parts, then it follows that the contribution of quadruples (k,ℓ,m,n) will be negligible if ∆ ≥ Xε−1/4 for any given ε > 0. The contribution of the quadruples satisfying ∆ ≤ Xε−1/4 is estimated by Lemma 4, where in (2.6) we take k = 4,δ ≍ ∆K−1/4. The ensuing integral is estimated trivially (using c ≪ nε), and we obtain n ε 2X ∆4(x)dx ≪ X9/2+εX max K−7/2 K4(KX)−1/4 +K2 +X5+ε 1 ε K≪X ZX (cid:16) (cid:17) ≪ X21/4+ε max K1/4 +X11/2+ε ≪ X11/2+ε. ε ε K≪X Theorem 2 follows if we replace X by X2−j in the above bound and sum over j = 1,2,.... On the fourth moment in the Rankin-Selberg problem 9 References [1] S.W. Graham and G. Kolesnik, Van der Corput’s method of exponential sums, LMS Lecture Note Series 126, Cambridge University Press, Cambridge, 1991. [2] M.N. Huxley, Exponential sums and the Riemann zeta-function V, Proc. London Math. Soc. (4) 90(2005), 1-41. [3] A. Ivi´c, The Riemann Zeta-Function, John Wiley & Sons, New York, 1985 (2nd ed. Dover, 2003). [4] A. Ivi´c, Large values of certain number-theoreticerror terms, Acta Arith. 56(1990), 135-159. [5] A. Ivi´c, Estimates of convolutions of certain number-theoretic error terms, Intern. J. Math. and Math. Sciences, Vol. 2004, No. 1(2004), 1-23. [6] A. Ivi´c, On some mean square estimates in the Rankin-Selberg problem, Applicable Analysis and Discrete Mathematics 1(2007), 1-11. [7] A. Ivi´c, K. Matsumoto and Y. Tanigawa, On Riesz mean of the coefficients of the Rankin– Selberg series, Math. Proc. Camb. Phil. Soc. 127(1999), 117-131. [8] A. Kaczorowski and A. Perelli, The Selberg class: a survey, in “Number Theory in Progress, Proc. Conf. in honour of A. Schinzel (K. Gyo¨ry et al. eds)”, de Gruyter, Berlin, 1999, pp. 953-992. [9] E. Landau, U¨ber die Anzahl der Gitterpunkte in gewissen Bereichen II, Nachr. Ges. Wiss. Go¨ttingen 1915, 209-243. [10] R. A. Rankin, Contributions to the theory of Ramanujan’s function τ(n) and similar arith- metical functions II. The order of the Fourier coefficients of integral modular forms, Proc. Cambridge Phil. Soc. 35(1939), 357-372. [11] O. Robert and P. Sargos, Three-dimensional exponential sums with monomials, J. reine angew. Math. 591(2006), 1-20. [12] A.Selberg,Bemerkungenu¨bereineDirichletscheReihe,diemitderTheoriederModulformen nahe verbunden ist, Arch. Math. Naturvid. 43(1940), 47-50. [13] A. Selberg, Old and new conjectures and results about a class of Dirichlet series, in “Proc. Amalfi Conf. Analytic NumberTheory 1989 (E. Bombieriet al. eds.)”, Universityof Salerno, Salerno, 1992, pp. 367–385. [14] P. Shiu, A Brun–Titchmarsh theorem for multiplicative functions, J. reine angew. Math. 313(1980), 161-170. Katedra Matematike RGF-a, Universitet u Beogradu, D- uˇsina 7, 11000 Beograd, Serbia E-mail address: [email protected], [email protected]