A note on the sup norm of Eisenstein series

A note on the sup norm of Eisenstein series Abstract We prove a sup norm bound on the real-analytic Eisenstein series, of the form E(z,1/2+iT)≪T3/8+ε, uniformly for z in a fixed compact subset of H. 1. Introduction Let E(z,s) be the usual real-analytic Eisenstein series for the group Γ=PSL2(Z). Theorem 1.1. Let Kbe a fixed compact subset of H. Then for T≥1, maxz∈K∣E(z,1/2+iT)∣≪K,εT3/8+ε. (1.1) For comparison, Iwaniec and Sarnak showed, for uj a Hecke–Maass cusp form of spectral parameter tj, that maxz∈K∣uj(z)∣≪Ktj5/12+ε. (1.2) The sup norm problem has been an active area and now there exist non-trivial estimates for cusp forms of large level, on higher rank groups, and for half-integral weight forms [2–4, 7, 8, 10, 13, 15, 19, 20]. Nevertheless, the basic estimate for Γ=PSL2(Z) in the eigenvalue aspect has not been improved. The case of Eisenstein series seems to have been largely neglected up to now, at least for the sup norm problem, but not for some other norms: Luo and Sarnak proved QUE for Eisenstein series [14], Spinu estimated the L4 norm [18], and the author has investigated QUE for geodesic restrictions [21]. The Eisenstein series case is similar in some ways to the cuspidal case, but has some technical problems because of the constant term in the Fourier expansion. Actually, the main impetus for this note was the realization that one can choose an efficient amplifier for the Eisenstein series, which leads to the improved exponent compared to the cusp form case (see [12, Remark 1.6]). 2. Notation and summary of the results of Iwaniec and Sarnak For Re(s)>1, let E(z,s)=∑γ∈Γ∞⧹Γ(Im(γz))s. (2.1) As shorthand, we sometimes write Et(z)=E(z,1/2+it). The Fourier expansion states ζ*(2s)E(z,s)=ζ*(2s)ys+ζ*(2(1−s))y1−s+2y∑n≠0τs−12(n)e(nx)Ks−12(2π∣n∣y), (2.2) where τw(n)=∑ab=∣n∣(a/b)w, and ζ*(2s)=π−sΓ(s)ζ(2s). (2.3) The Fourier expansion implies the functional equation ζ*(2s)E(z,s)=ζ*(2(1−s))E(z,1−s). Specializing to s=1/2+it, and setting φ(s)=ζ*(2(1−s))/ζ*(2s), one obtains Et(z)=y1/2+it+φ(1/2+it)y1/2−it+2yζ*(1+2it)∑n≠0τit(n)e(nx)Kit(2π∣n∣y). (2.4) Let αn be an arbitrary sequence of complex numbers. The main technical result proved by Iwaniec and Sarnak [12, (A.12)] is ∑T≤tj≤T+1∣∑n≤Nαnλj(n)∣2∣uj(z)∣2+∫TT+1∣∑n≤Nαnτit(n)∣2∣Et(z)∣2dt≪Tε(T∑n≤N∣αn∣2+T1/2(N+N1/2y)(∑n≤N∣αn∣)2). (2.5) Here λj(n) are the Hecke eigenvalues of the Hecke–Maass cusp forms uj, scaled so the Ramanujan–Petersson conjecture is ∣λj(n)∣≤τ0(n). The implied constant is uniform in y for y≫1. We have two main problems to overcome to obtain a bound on Et. The first problem is to relate a pointwise bound on Et to an integral bound of the type occurring in (2.5). We are able to accomplish this by modifying a method of Heath-Brown [9, Lemma 3]. This shows, roughly, that ∣ET(z)∣2⪅∫T−1T+1∣Et(z)∣2dt (see Corollary 4.2 below for the true result). Normally one constructs an amplifier to be large at a specified point. Because of the above relationship between the integral of ∣Et∣2 and the pointwise bound, we cannot simply choose αn to be large at a single value of t. Rather, we need the amplifier to be large on an interval of t’s of length ≫T−ε. This we accomplish with Lemma 5.1 below. 3. Preliminary estimates For purposes of comparison, it is helpful to record the effects of trivially bounding the Eisenstein series using the Fourier expansion. Let F(z,s)=E(z,s)−ys−φ(s)y1−s. Lemma 3.1. For t≥1, and y≫1, we have F(z,1/2+it)≪(t/y)1/2log2t+y1/2t−1/3+ε, (3.1)and, therefore, E(z,1/2+it)≪y1/2+(t/y)1/2log2t. (3.2) This is analogous to [20, Proposition 6.2]. Proof Suppose that t≥1. By Stirling’s formula, F(z,1/2+it)≪y∣ζ(1+2it)∣∑n=1∞τ0(n)∣Kit(2πny)∣cosh(πt/2). (3.3) Next we need uniform bounds on the K-Bessel function which we extract from the uniform asymptotic expansions due to Balogh [1]: cosh(πt/2)Kit(u)≪{t−1/4(t−u)−1/4,if0<u<t−Ct1/3t−1/3,if∣u−t∣≤Ct1/3,u−1/4(u−t)−1/4·exp(−c(ut)3/2(u−tt1/3)3/2),ifu>t+Ct1/3. (3.4) We break up the sum over n according to the different pieces. For instance, the range 2πny≤t2 gives a bound y∣ζ(1+2it)∣∑n≪t/yτ0(n)t1/2≪(t/y)1/2log2t, (3.5) using ∣ζ(1+2it)∣−1≪logt. Similarly, the range ∣2πny−t∣≍Δ with t1/3≪Δ≪t gives y∣ζ(1+2it)∣(tΔ)−1/4∑n=t2πy+O(Δy)τ0(n)≪ylogt(tΔ)1/4(Δylogt+tε), (3.6) using Shiu’s bound on the divisor function in a short interval [17]. These terms then give the same bound as (3.5), plus an additional term that is ≪y1/2t−1/3+ε. (3.7) It is easily checked that the transition range also leads to (3.7). The range with u>t+Ct1/3 is even easier to bound because there is an additional exponential factor aiding the estimation. In all, this shows (3.1). The bound (3.2) is immediate from (3.1).□ We also record some bounds on F(z,s) and E(z,s) valid not just for Re(s)=1/2. In principle, a similar approach to the proof of Lemma 3.1 should apply; however, the author has been unable to find sharp estimates analogous to (3.4) for Kσ+it with σ>0, and so we present some weaker bounds that suffice for our later application. Lemma 3.2. If Re(s)>1, then F(z,s)≪σy1−σ, (3.8)uniformly for y≫1, t∈R. If Re(s)≥1/2−clog(2+∣t∣), for a sufficiently small c>0, then F(z,σ+it)≪σ,ε(1+∣t∣)1+εy−σ. (3.9)Moreover, if y≫t1+ε, and tis sufficiently large, then F(z,σ+it)≪σ,ε(yt)−100. (3.10) Proof We begin with (3.8). Using the original definition (2.1), we have F(z,s)=−φ(s)y1−s+∑γ∈Γ∞⧹Γγ≠1(Im(γz))s, (3.11) and so by a trivial bound, ∣F(z,s)∣≤∣φ(s)∣y1−σ+∑γ∈Γ∞⧹Γγ≠1(Im(γz))σ=∣φ(s)∣y1−σ+(E(z,σ)−yσ). (3.12) By the Fourier expansion, and using the standard bound Kα(x)≪αx−1/2e−x, (3.13) for x≫1, we have E(z,σ)−yσ≪σy1−σ+∑n=1∞τσ−12(n)n−1/2exp(−2πny)≪σy1−σ, (3.14) since the infinite sum in (3.14) is ≪σexp(−2πy). Furthermore, note that for σ>1, we have φ(σ+it)=ζ*(2−2σ−2it)ζ*(2σ+2it)=ζ*(−1+2σ+2it)ζ*(2σ+2it)≪σ∣Γ(σ−12+it)∣∣Γ(σ+it)∣≪σ(1+∣t∣)−1/2. (3.15) Combining the above estimates, we derive (3.8). Next we show (3.9) and (3.10). By the Fourier expansion (2.2), we have ∣F(z,s)∣=∣2y1/2π−sΓ(s)ζ(2s)∑n≠0τs−12(n)e(nx)Ks−12(2π∣n∣y)∣. (3.16) We will momentarily show that if t≫1, then Kσ+it(y)Γ(12+σ+it)≪σ,λy−σ(t/y)λ, (3.17) where λ>0 may be chosen at will. If t≪1, then (3.13) gives an even stronger bound than (3.17) (one should note that the implied constant in (3.13) depends continuously on α). Now we insert this bound into (3.16), giving ∣F(z,σ+it)∣≪σ,λy1−σ−λtλ∣ζ(2σ+2it)∣∑n=1∞nσ−12τ0(n)n12−σ−λ≪σ,λy1−σ(t/y)λ∣ζ(2σ+2it)∣, (3.18) where we hereby assume λ>1 to ensure convergence. If y≫t1+ε we may take λ very large to give (3.10). In any case, we may take λ=1+ε with ε>0 small, giving (3.9). Finally, we show (3.17). By [6, 17.43.18], we have 4Kν(y)Γ(12+ν)=12πi∫(δ)(y/2)−wΓ(w+ν2)Γ(w−ν2)Γ(12+ν)dw, (3.19) where δ>∣Re(ν)∣. By Stirling, we have with w=δ+iv and ν=σ+it, that Γ(w+ν2)Γ(w−ν2)Γ(12+ν)≪σ,δt−σ(1+∣v+t∣)δ+σ−12(1+∣v−t∣)δ−σ−12exp(−π2q(v,t)), (3.20) where q(v,t)=0 for ∣v∣≤∣t∣, and q(v,t)=∣v−t∣ for ∣v∣≥∣t∣. Therefore, Kσ+it(y)Γ(12+σ+it)≪y−δt−σ∫−∞∞(1+∣v+t∣)δ+σ−12(1+∣v−t∣)δ−σ−12exp(−π2q(v,t))dv. (3.21) By symmetry, suppose v,t≥0. The part of the integral with 0≤v≤t gives a bound y−δt−σ+δ+σ−12∫0t(1+v)δ−σ−12dv≪y−δt−σ+δ+σ−12+δ−σ+12=y−δtδ−σ. (3.22) We shall choose δ=σ+λ with λ>0 arbitrary, which gives a bound consistent with (3.17). The part of the integral with v≥t is easier to bound, aided by the exponential decay that is present in this range. First, one may note that the range with t≤v≤2t gives precisely the same bound as (3.22), even if we only use the very crude bound q(v,t)≥0. For the range with v≥2t, then we have the bound y−δt−σ∫v≥2tvδ−1exp(−π2(v−t))dv≪δy−δexp(−πt/2). (3.23) □ 4. A pointwise bound via an integral bound Lemma 4.1. Suppose y,T≫1and y≪T100. Then ∣F(z,1/2+iT)∣2≪log5Ty+log5T∫∣r∣≤4logT∣F(z,1/2+iT+ir)∣2dr. (4.1) The proof is analogous to that of Heath-Brown [9], but some modifications are necessary since the Eisenstein series is not a Dirichlet series, and z is an additional parameter to control. The key idea is to work with F(z,s), and to deduce analogous results for E(z,s) only at the end (see Corollary 4.2 below). The logarithmic powers could be reduced with a little more work, if it were necessary. Proof Note that F(z,s) satisfies the functional equation ζ*(2s)F(z,s)=ζ*(2(1−s))F(z,1−s), just as the Eisenstein series does. Furthermore, ζ*(2s)F(z,s) is entire, and so F(z,s) is analytic for Re(s)≥1/2. Suppose that Re(s)∈[1/4,3/4], and let I=12πi∫(2)F(z,s+w)2exp(w2)wdw. (4.2) By (3.8), we have that I=O(y−1), uniformly in the stated range of s, and for y≫1. Suppose that 0<δ<1/8, and s=1/2+δ+iT. Then by shifting the contour of integration to Re(w)=−δ, we obtain F(z,1/2+δ+iT)2+12πi∫(−δ)F(z,1/2+δ+iT+w)2exp(w2)wdw=I=O(y−1). (4.3) By bounding the integral trivially with absolute values, we have ∣F(z,1/2+δ+iT)∣2≪y−1+∫−∞∞exp(−v2)∣−δ+iv∣∣F(z,1/2+iT+iv)∣2dv. (4.4) Alternatively, we have by Cauchy’s theorem that F(z,1/2+iT)2=12πi∳F(z,1/2+iT+u)2exp(u2)udu, (4.5) where the contour of integration is a small loop around u=0. We choose the contour to be a rectangle with corners ±δ±2ilogT, where δ=clogT with c>0 small enough to ensure that ζ(1+2iT+2u)≫(logT)−1 inside the contour (by the standard zero-free region of ζ). Using (3.9), it follows that the top and bottom sides of this integral are bounded by y−1exp(−log2T), (4.6) which additionally uses the fact that y−1/2−Re(u)≍y−1/2 since logy≪logT. For the side of the rectangle with Re(u)=−δ, we change variables u→−u and apply the functional equation of F, which gives F(z,1/2−δ+iT−iv)=ζ*(1+2δ−2iT+2iv)ζ*(1−2δ+2iT−2iv)F(z,1/2+δ−iT+iv). (4.7) Using standard bounds on the zeta function, we derive ζ*(1+2δ−2iT+2iv)ζ*(1−2δ+2iT−2iv)≪log2T. (4.8) Therefore, we conclude ∣F(z,1/2+iT)∣2≪y−1exp(−log2T)+(logT)4∫∣v∣≤2logTexp(−v2)∣δ+iv∣∣F(z,1/2+δ+iT+iv)∣2dv. (4.9) Inserting (4.4) into (4.9), we derive ∣F(z,1/2+iT)∣2≪y−1exp(−log2T)+(logT)4∫∣v∣≤2logTexp(−v2)∣δ+iv∣·[y−1+∫−∞∞exp(−r2)∣−δ+ir∣∣F(z,1/2+iT+iv+ir)∣2dr]dv. (4.10) The inner r-integral can be safely truncated at ∣r∣≤2logT without introducing a new error term. Changing variables r→r−v, and extending the r-integral to ∣r∣≤4logT by positivity, we derive ∣F(z,1/2+iT)∣2≪y−1log5T+(logT)4∫∣r∣≤4logT∣F(z,1/2+iT+ir)∣2·(∫∣v∣≤2logTexp(−v2)∣δ+iv∣exp(−(r−v)2)∣−δ+i(r−v)∣dv)dr. (4.11) Using Cauchy–Schwarz, and the obvious estimate ∫−∞∞1∣δ+iv∣2dv=πδ≪logT, (4.12) we may bound the inner v-integral by O(logT). Putting everything together gives the desired bound (4.1).□ Using E(z,s)=ys+φ(s)y1−s+F(z,s), and Cauchy–Schwarz, we derive a corresponding result for the Eisenstein series itself. Corollary 4.2. Suppose y,T≫1. Then ∣E(z,1/2+iT)∣2≪ylog6T+log5T∫∣r∣≤4logT∣E(z,1/2+iT+ir)∣2dr. (4.13) 5. A lower bound for the amplifier Let w be a fixed, compactly supported function on the positive reals, with ∫−∞∞w(t)dt≠0. Define AN(t,r)=∑n=1∞w(n/N)τit(n)τir(n). (5.1) Lemma 5.1. Suppose that logN≫(logT)2/3+δ, and t,r=T+O((logN)−1−δ), for some fixed δ>0. Then AN(t,r)=w˜(1)NlogNζ(2)∣ζ(1+2iT)∣2(1+o(1)). (5.2) Fouvry et al. [5, Lemma 2.4] prove a result with a similar conclusion, but their method requires N≫T3, while here we eventually will want N=T1/4. Proof Taking a Mellin transform, and using a well-known identity of Ramanujan [16, (15)], we derive AN(t,r)=12πi∫(2)Nsw˜(s)ζ−1(2s)ζ(s+it+ir)ζ(s+it−ir)ζ(s−it+ir)ζ(s−it−ir)ds. (5.3) Next we move the contour to the left, to one along the straight line segments L1,L2,L3 defined by L1={1−c(logT)2/3+it:∣t∣≤100T}, L2={it:∣t∣≥100T}, and the short horizontal segments L3={σ±100iT:1−c(logT)2/3≤σ≤1}. The integrals along the line segments L2 and L3 are trivially bounded by O(T−100) by the rapid decay of w˜. The new line L1 gives an amount that is ≪Nlog8/3Texp(−clogN(logT)2/3)≪N(logT)100, (5.4) using the Vinogradov–Korobov bound ∣ζ(σ+it)∣≪(log∣t∣)2/3 for t≫1 and 1−c(log∣t∣)2/3≤σ≤1 (see [11, Corollary 8.28]). Here what is important is that σ can be taken to be 1−O((log∣t∣)−1+δ), for some small δ>0. We need to analyze the residues of the poles. Temporarily assume that t≠±r. The poles at s=1+it+ir and s=1−it−ir have very small residues from the rapid decay of w˜(s). The residue at s=1+ir−it contributes R1=N1+ir−itw˜(1+ir−it)ζ(2(1+ir−it))ζ(1+2ir)ζ(1+2ir−2it)ζ(1−2it). (5.5) By symmetry the residue at s=1−ir+it, say R2, is the same as R1 but with r and t interchanged. Let us write r=t+η (by assumption, η=O((logN)−1−δ), so R1=N1+iηw˜(1+iη)ζ(2(1+iη))ζ(1+2iη)ζ(1+2it+2iη)ζ(1−2it). (5.6) By simple Taylor approximations, we have N1+iηw˜(1+iη)ζ(2(1+iη))ζ(1+2iη)=w˜(1)NlogN2ζ(2)(1iηlogN+1+O(∣η∣logN)), (5.7) and by the Vinogradov–Korobov bound ζ′ζ(1+2it)≪(log∣t∣)2/3+ε (see [11, Theorem 8.29]), we have ζ(1+2it+2iη)ζ(1−2it)=∣ζ(1+2it)∣2(1+O(∣η∣(logT)2/3+ε)). (5.8) Combining (5.7) and (5.8), we derive R1=Nw˜(1)∣ζ(1+2it)∣2logN2ζ(2)·[1+1iηlogN+O(∣η∣logN+(logT)2/3+εlogN+∣η∣(logT)2/3+ε)]. (5.9) The conditions η≪(logN)−1−δ, (logN)≫(logT)2/3+δ are enough to imply that the error term in (5.9) is o(1). Similarly, R2=Nw˜(1)∣ζ(1+2it)∣2logN2ζ(2)[1+1−iηlogN+o(1)], (5.10) and, therefore, R1+R2=Nw˜(1)∣ζ(1+2it)∣2logNζ(2)(1+o(1)). (5.11) □ Remark. If we had moved the contour to the line σ=1/2, then instead of (5.4) we would have obtained an error term of size O(N1/2T1/3+ε) using Weyl’s subconvexity bound. This is o(N) for N≫T2/3+ε, which is far from our desired choice of N=T1/4. 6. Completion of the proof By Corollary 4.2, we have that ∣E(z,1/2+iT)∣2≪ylog6T+log5T∫∣r∣≤4logT∣E(z,1/2+iT+ir)∣2dr. (6.1) On the right-hand side above, we dissect the integral into subintervals, each of length ≍(logT)−2, say. Let U be one of these intervals, and choose a point tU∈U. Then by Lemma 5.1, we have ∫r∈U∣E(z,1/2+iT+ir)∣2dr≪N−2Tε∫r∈U∣AN(T+r,T+tU)∣2∣E(z,1/2+iT+ir)∣2dr. (6.2) By (2.5), with αn=w(n/N)τi(T+tU)(n), this is in turn ≪(NT)ε(TN+T1/2(N+N1/2y)). (6.3) If 1≪y≪T1/8, we choose N as T1/4, which in all gives ∣E(z,1/2+iT)∣≪T3/8+ε. (6.4) If T1/8≪y≪T1/6, we set N=y−2/3T1/3, giving ∣E(z,1/2+iT)∣≪y1/3T1/3+ε. (6.5) The bound (3.2) is superior to (6.5) for y≫T1/6. Funding This material is based upon work supported by the National Science Foundation under Agreement no. DMS-1401008. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. References 1 C. Balogh , Asymptotic expansions of the modified Bessel function of the third kind of imaginary order , SIAM J. Appl. Math. 15 ( 1967 ), 1315 – 1323 . Google Scholar CrossRef Search ADS 2 V. Blomer and R. Holowinsky , Bounding sup-norms of cusp forms of large level , Invent. Math. 179 ( 2010 ), 645 – 681 . Google Scholar CrossRef Search ADS 3 V. Blomer and P. Maga , Subconvexity for sup-norms of cusp forms on PGL(n) , Selecta Math . 22 ( 2016 ), 1269 – 1287 . 4 V. Blomer and A. Pohl , The sup-norm problem on the Siegel modular space of rank two, Amer. J. Math . 136 ( 2016 ), 999 – 1027 . 5 É. Fouvry , E. Kowalski and P. Michel , Algebraic trace functions over the primes , Duke Math. J. 163 ( 2014 ), 1683 – 1736 . Google Scholar CrossRef Search ADS 6 I. S. Gradshteyn and I. M. Ryzhik , Table of Integrals, Series, and Products , 6th edn, Academic Press Inc , San Diego, CA, 2000 , xlvii, 1163p. Translated from the Russian, Translation edited and with a preface by Alan Jeffrey and Daniel Zwillinger. 7 G. Harcos and N. Templier , On the sup-norm of Maass cusp forms of large level: II , Int. Math. Res. Not. IMRN 2012 ( 2012 ), 4764 – 4774 . Google Scholar CrossRef Search ADS 8 G. Harcos and N. Templier , On the sup-norm of Maass cusp forms of large level: III , Math. Ann. 356 ( 2013 ), 209 – 216 . Google Scholar CrossRef Search ADS 9 D. R. Heath-Brown , The twelfth power moment of the Riemann-function, Quart , J. Math. Oxford Ser. (2) 29 ( 1978 ), 443 – 462 . Google Scholar CrossRef Search ADS 10 R. Holowinsky , G. Ricotta and E. Royer , On the sup-norm of SL3 Hecke-Maass cusp form, preprint, 2014 , http://arxiv.org/abs/1404.3622. 11 H. Iwaniec and E. Kowalski , Analytic Number Theory. American Mathematical Society Colloquium Publications, 53 , American Mathematical Society , Providence, RI , 2004 . 12 H. Iwaniec and P. Sarnak , L∞ norms of eigenfunctions of arithmetic surfaces , Ann. of Math. (2) 141 ( 1995 ), 301 – 320 . Google Scholar CrossRef Search ADS 13 E. Kiral , Bounds on sup-norms of half-integral weight modular forms , Acta Arith. 165 ( 2014 ), 385 – 400 . Google Scholar CrossRef Search ADS 14 W. Luo and P. Sarnak , Quantum ergodicity of eigenfunctions on PSL2(Z)\H2 , Inst. Hautes Études Sci. Publ. Math. 81 ( 1995 ), 207 – 237 . Google Scholar CrossRef Search ADS 15 S. Marshall , Sup norms of Maass forms on semisimple groups, preprint, 2014 http://arxiv.org/abs/1405.7033. 16 S. Ramanujan , Some formulæ in the analytic theory of numbers Collected papers of Srinivasa Ramanujan , AMS Chelsea Publ , Providence, RI , 2000 , 133 – 135 . 17 P. Shiu , A Brun-Titchmarsh theorem for multiplicative functions , J. Reine Angew. Math. 313 ( 1980 ), 161 – 170 . 18 F. Spinu , The L4 norm of the Eisenstein series, Ph. D. Thesis, Princeton University. 2003 . 19 N. Templier , On the sup-norm of Maass cusp forms of large level , Selecta Math. (N.S.) 16 ( 2010 ), 501 – 531 . Google Scholar CrossRef Search ADS 20 N. Templier , Hybrid sup-norm bounds for Hecke-Maass cusp forms , J. Eur. Math. Soc. (JEMS) 17 ( 2015 ), 2069 – 2082 . Google Scholar CrossRef Search ADS 21 M. Young , The quantum unique ergodicity conjecture for thin sets , Adv. Math. 286 ( 2016 ), 9581016 . Google Scholar CrossRef Search ADS © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Quarterly Journal of Mathematics Oxford University Press

A note on the sup norm of Eisenstein series

Loading next page...
 
/lp/ou_press/a-note-on-the-sup-norm-of-eisenstein-series-Kf82E0HH55
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0033-5606
eISSN
1464-3847
D.O.I.
10.1093/qmath/hay019
Publisher site
See Article on Publisher Site

Abstract

Abstract We prove a sup norm bound on the real-analytic Eisenstein series, of the form E(z,1/2+iT)≪T3/8+ε, uniformly for z in a fixed compact subset of H. 1. Introduction Let E(z,s) be the usual real-analytic Eisenstein series for the group Γ=PSL2(Z). Theorem 1.1. Let Kbe a fixed compact subset of H. Then for T≥1, maxz∈K∣E(z,1/2+iT)∣≪K,εT3/8+ε. (1.1) For comparison, Iwaniec and Sarnak showed, for uj a Hecke–Maass cusp form of spectral parameter tj, that maxz∈K∣uj(z)∣≪Ktj5/12+ε. (1.2) The sup norm problem has been an active area and now there exist non-trivial estimates for cusp forms of large level, on higher rank groups, and for half-integral weight forms [2–4, 7, 8, 10, 13, 15, 19, 20]. Nevertheless, the basic estimate for Γ=PSL2(Z) in the eigenvalue aspect has not been improved. The case of Eisenstein series seems to have been largely neglected up to now, at least for the sup norm problem, but not for some other norms: Luo and Sarnak proved QUE for Eisenstein series [14], Spinu estimated the L4 norm [18], and the author has investigated QUE for geodesic restrictions [21]. The Eisenstein series case is similar in some ways to the cuspidal case, but has some technical problems because of the constant term in the Fourier expansion. Actually, the main impetus for this note was the realization that one can choose an efficient amplifier for the Eisenstein series, which leads to the improved exponent compared to the cusp form case (see [12, Remark 1.6]). 2. Notation and summary of the results of Iwaniec and Sarnak For Re(s)>1, let E(z,s)=∑γ∈Γ∞⧹Γ(Im(γz))s. (2.1) As shorthand, we sometimes write Et(z)=E(z,1/2+it). The Fourier expansion states ζ*(2s)E(z,s)=ζ*(2s)ys+ζ*(2(1−s))y1−s+2y∑n≠0τs−12(n)e(nx)Ks−12(2π∣n∣y), (2.2) where τw(n)=∑ab=∣n∣(a/b)w, and ζ*(2s)=π−sΓ(s)ζ(2s). (2.3) The Fourier expansion implies the functional equation ζ*(2s)E(z,s)=ζ*(2(1−s))E(z,1−s). Specializing to s=1/2+it, and setting φ(s)=ζ*(2(1−s))/ζ*(2s), one obtains Et(z)=y1/2+it+φ(1/2+it)y1/2−it+2yζ*(1+2it)∑n≠0τit(n)e(nx)Kit(2π∣n∣y). (2.4) Let αn be an arbitrary sequence of complex numbers. The main technical result proved by Iwaniec and Sarnak [12, (A.12)] is ∑T≤tj≤T+1∣∑n≤Nαnλj(n)∣2∣uj(z)∣2+∫TT+1∣∑n≤Nαnτit(n)∣2∣Et(z)∣2dt≪Tε(T∑n≤N∣αn∣2+T1/2(N+N1/2y)(∑n≤N∣αn∣)2). (2.5) Here λj(n) are the Hecke eigenvalues of the Hecke–Maass cusp forms uj, scaled so the Ramanujan–Petersson conjecture is ∣λj(n)∣≤τ0(n). The implied constant is uniform in y for y≫1. We have two main problems to overcome to obtain a bound on Et. The first problem is to relate a pointwise bound on Et to an integral bound of the type occurring in (2.5). We are able to accomplish this by modifying a method of Heath-Brown [9, Lemma 3]. This shows, roughly, that ∣ET(z)∣2⪅∫T−1T+1∣Et(z)∣2dt (see Corollary 4.2 below for the true result). Normally one constructs an amplifier to be large at a specified point. Because of the above relationship between the integral of ∣Et∣2 and the pointwise bound, we cannot simply choose αn to be large at a single value of t. Rather, we need the amplifier to be large on an interval of t’s of length ≫T−ε. This we accomplish with Lemma 5.1 below. 3. Preliminary estimates For purposes of comparison, it is helpful to record the effects of trivially bounding the Eisenstein series using the Fourier expansion. Let F(z,s)=E(z,s)−ys−φ(s)y1−s. Lemma 3.1. For t≥1, and y≫1, we have F(z,1/2+it)≪(t/y)1/2log2t+y1/2t−1/3+ε, (3.1)and, therefore, E(z,1/2+it)≪y1/2+(t/y)1/2log2t. (3.2) This is analogous to [20, Proposition 6.2]. Proof Suppose that t≥1. By Stirling’s formula, F(z,1/2+it)≪y∣ζ(1+2it)∣∑n=1∞τ0(n)∣Kit(2πny)∣cosh(πt/2). (3.3) Next we need uniform bounds on the K-Bessel function which we extract from the uniform asymptotic expansions due to Balogh [1]: cosh(πt/2)Kit(u)≪{t−1/4(t−u)−1/4,if0<u<t−Ct1/3t−1/3,if∣u−t∣≤Ct1/3,u−1/4(u−t)−1/4·exp(−c(ut)3/2(u−tt1/3)3/2),ifu>t+Ct1/3. (3.4) We break up the sum over n according to the different pieces. For instance, the range 2πny≤t2 gives a bound y∣ζ(1+2it)∣∑n≪t/yτ0(n)t1/2≪(t/y)1/2log2t, (3.5) using ∣ζ(1+2it)∣−1≪logt. Similarly, the range ∣2πny−t∣≍Δ with t1/3≪Δ≪t gives y∣ζ(1+2it)∣(tΔ)−1/4∑n=t2πy+O(Δy)τ0(n)≪ylogt(tΔ)1/4(Δylogt+tε), (3.6) using Shiu’s bound on the divisor function in a short interval [17]. These terms then give the same bound as (3.5), plus an additional term that is ≪y1/2t−1/3+ε. (3.7) It is easily checked that the transition range also leads to (3.7). The range with u>t+Ct1/3 is even easier to bound because there is an additional exponential factor aiding the estimation. In all, this shows (3.1). The bound (3.2) is immediate from (3.1).□ We also record some bounds on F(z,s) and E(z,s) valid not just for Re(s)=1/2. In principle, a similar approach to the proof of Lemma 3.1 should apply; however, the author has been unable to find sharp estimates analogous to (3.4) for Kσ+it with σ>0, and so we present some weaker bounds that suffice for our later application. Lemma 3.2. If Re(s)>1, then F(z,s)≪σy1−σ, (3.8)uniformly for y≫1, t∈R. If Re(s)≥1/2−clog(2+∣t∣), for a sufficiently small c>0, then F(z,σ+it)≪σ,ε(1+∣t∣)1+εy−σ. (3.9)Moreover, if y≫t1+ε, and tis sufficiently large, then F(z,σ+it)≪σ,ε(yt)−100. (3.10) Proof We begin with (3.8). Using the original definition (2.1), we have F(z,s)=−φ(s)y1−s+∑γ∈Γ∞⧹Γγ≠1(Im(γz))s, (3.11) and so by a trivial bound, ∣F(z,s)∣≤∣φ(s)∣y1−σ+∑γ∈Γ∞⧹Γγ≠1(Im(γz))σ=∣φ(s)∣y1−σ+(E(z,σ)−yσ). (3.12) By the Fourier expansion, and using the standard bound Kα(x)≪αx−1/2e−x, (3.13) for x≫1, we have E(z,σ)−yσ≪σy1−σ+∑n=1∞τσ−12(n)n−1/2exp(−2πny)≪σy1−σ, (3.14) since the infinite sum in (3.14) is ≪σexp(−2πy). Furthermore, note that for σ>1, we have φ(σ+it)=ζ*(2−2σ−2it)ζ*(2σ+2it)=ζ*(−1+2σ+2it)ζ*(2σ+2it)≪σ∣Γ(σ−12+it)∣∣Γ(σ+it)∣≪σ(1+∣t∣)−1/2. (3.15) Combining the above estimates, we derive (3.8). Next we show (3.9) and (3.10). By the Fourier expansion (2.2), we have ∣F(z,s)∣=∣2y1/2π−sΓ(s)ζ(2s)∑n≠0τs−12(n)e(nx)Ks−12(2π∣n∣y)∣. (3.16) We will momentarily show that if t≫1, then Kσ+it(y)Γ(12+σ+it)≪σ,λy−σ(t/y)λ, (3.17) where λ>0 may be chosen at will. If t≪1, then (3.13) gives an even stronger bound than (3.17) (one should note that the implied constant in (3.13) depends continuously on α). Now we insert this bound into (3.16), giving ∣F(z,σ+it)∣≪σ,λy1−σ−λtλ∣ζ(2σ+2it)∣∑n=1∞nσ−12τ0(n)n12−σ−λ≪σ,λy1−σ(t/y)λ∣ζ(2σ+2it)∣, (3.18) where we hereby assume λ>1 to ensure convergence. If y≫t1+ε we may take λ very large to give (3.10). In any case, we may take λ=1+ε with ε>0 small, giving (3.9). Finally, we show (3.17). By [6, 17.43.18], we have 4Kν(y)Γ(12+ν)=12πi∫(δ)(y/2)−wΓ(w+ν2)Γ(w−ν2)Γ(12+ν)dw, (3.19) where δ>∣Re(ν)∣. By Stirling, we have with w=δ+iv and ν=σ+it, that Γ(w+ν2)Γ(w−ν2)Γ(12+ν)≪σ,δt−σ(1+∣v+t∣)δ+σ−12(1+∣v−t∣)δ−σ−12exp(−π2q(v,t)), (3.20) where q(v,t)=0 for ∣v∣≤∣t∣, and q(v,t)=∣v−t∣ for ∣v∣≥∣t∣. Therefore, Kσ+it(y)Γ(12+σ+it)≪y−δt−σ∫−∞∞(1+∣v+t∣)δ+σ−12(1+∣v−t∣)δ−σ−12exp(−π2q(v,t))dv. (3.21) By symmetry, suppose v,t≥0. The part of the integral with 0≤v≤t gives a bound y−δt−σ+δ+σ−12∫0t(1+v)δ−σ−12dv≪y−δt−σ+δ+σ−12+δ−σ+12=y−δtδ−σ. (3.22) We shall choose δ=σ+λ with λ>0 arbitrary, which gives a bound consistent with (3.17). The part of the integral with v≥t is easier to bound, aided by the exponential decay that is present in this range. First, one may note that the range with t≤v≤2t gives precisely the same bound as (3.22), even if we only use the very crude bound q(v,t)≥0. For the range with v≥2t, then we have the bound y−δt−σ∫v≥2tvδ−1exp(−π2(v−t))dv≪δy−δexp(−πt/2). (3.23) □ 4. A pointwise bound via an integral bound Lemma 4.1. Suppose y,T≫1and y≪T100. Then ∣F(z,1/2+iT)∣2≪log5Ty+log5T∫∣r∣≤4logT∣F(z,1/2+iT+ir)∣2dr. (4.1) The proof is analogous to that of Heath-Brown [9], but some modifications are necessary since the Eisenstein series is not a Dirichlet series, and z is an additional parameter to control. The key idea is to work with F(z,s), and to deduce analogous results for E(z,s) only at the end (see Corollary 4.2 below). The logarithmic powers could be reduced with a little more work, if it were necessary. Proof Note that F(z,s) satisfies the functional equation ζ*(2s)F(z,s)=ζ*(2(1−s))F(z,1−s), just as the Eisenstein series does. Furthermore, ζ*(2s)F(z,s) is entire, and so F(z,s) is analytic for Re(s)≥1/2. Suppose that Re(s)∈[1/4,3/4], and let I=12πi∫(2)F(z,s+w)2exp(w2)wdw. (4.2) By (3.8), we have that I=O(y−1), uniformly in the stated range of s, and for y≫1. Suppose that 0<δ<1/8, and s=1/2+δ+iT. Then by shifting the contour of integration to Re(w)=−δ, we obtain F(z,1/2+δ+iT)2+12πi∫(−δ)F(z,1/2+δ+iT+w)2exp(w2)wdw=I=O(y−1). (4.3) By bounding the integral trivially with absolute values, we have ∣F(z,1/2+δ+iT)∣2≪y−1+∫−∞∞exp(−v2)∣−δ+iv∣∣F(z,1/2+iT+iv)∣2dv. (4.4) Alternatively, we have by Cauchy’s theorem that F(z,1/2+iT)2=12πi∳F(z,1/2+iT+u)2exp(u2)udu, (4.5) where the contour of integration is a small loop around u=0. We choose the contour to be a rectangle with corners ±δ±2ilogT, where δ=clogT with c>0 small enough to ensure that ζ(1+2iT+2u)≫(logT)−1 inside the contour (by the standard zero-free region of ζ). Using (3.9), it follows that the top and bottom sides of this integral are bounded by y−1exp(−log2T), (4.6) which additionally uses the fact that y−1/2−Re(u)≍y−1/2 since logy≪logT. For the side of the rectangle with Re(u)=−δ, we change variables u→−u and apply the functional equation of F, which gives F(z,1/2−δ+iT−iv)=ζ*(1+2δ−2iT+2iv)ζ*(1−2δ+2iT−2iv)F(z,1/2+δ−iT+iv). (4.7) Using standard bounds on the zeta function, we derive ζ*(1+2δ−2iT+2iv)ζ*(1−2δ+2iT−2iv)≪log2T. (4.8) Therefore, we conclude ∣F(z,1/2+iT)∣2≪y−1exp(−log2T)+(logT)4∫∣v∣≤2logTexp(−v2)∣δ+iv∣∣F(z,1/2+δ+iT+iv)∣2dv. (4.9) Inserting (4.4) into (4.9), we derive ∣F(z,1/2+iT)∣2≪y−1exp(−log2T)+(logT)4∫∣v∣≤2logTexp(−v2)∣δ+iv∣·[y−1+∫−∞∞exp(−r2)∣−δ+ir∣∣F(z,1/2+iT+iv+ir)∣2dr]dv. (4.10) The inner r-integral can be safely truncated at ∣r∣≤2logT without introducing a new error term. Changing variables r→r−v, and extending the r-integral to ∣r∣≤4logT by positivity, we derive ∣F(z,1/2+iT)∣2≪y−1log5T+(logT)4∫∣r∣≤4logT∣F(z,1/2+iT+ir)∣2·(∫∣v∣≤2logTexp(−v2)∣δ+iv∣exp(−(r−v)2)∣−δ+i(r−v)∣dv)dr. (4.11) Using Cauchy–Schwarz, and the obvious estimate ∫−∞∞1∣δ+iv∣2dv=πδ≪logT, (4.12) we may bound the inner v-integral by O(logT). Putting everything together gives the desired bound (4.1).□ Using E(z,s)=ys+φ(s)y1−s+F(z,s), and Cauchy–Schwarz, we derive a corresponding result for the Eisenstein series itself. Corollary 4.2. Suppose y,T≫1. Then ∣E(z,1/2+iT)∣2≪ylog6T+log5T∫∣r∣≤4logT∣E(z,1/2+iT+ir)∣2dr. (4.13) 5. A lower bound for the amplifier Let w be a fixed, compactly supported function on the positive reals, with ∫−∞∞w(t)dt≠0. Define AN(t,r)=∑n=1∞w(n/N)τit(n)τir(n). (5.1) Lemma 5.1. Suppose that logN≫(logT)2/3+δ, and t,r=T+O((logN)−1−δ), for some fixed δ>0. Then AN(t,r)=w˜(1)NlogNζ(2)∣ζ(1+2iT)∣2(1+o(1)). (5.2) Fouvry et al. [5, Lemma 2.4] prove a result with a similar conclusion, but their method requires N≫T3, while here we eventually will want N=T1/4. Proof Taking a Mellin transform, and using a well-known identity of Ramanujan [16, (15)], we derive AN(t,r)=12πi∫(2)Nsw˜(s)ζ−1(2s)ζ(s+it+ir)ζ(s+it−ir)ζ(s−it+ir)ζ(s−it−ir)ds. (5.3) Next we move the contour to the left, to one along the straight line segments L1,L2,L3 defined by L1={1−c(logT)2/3+it:∣t∣≤100T}, L2={it:∣t∣≥100T}, and the short horizontal segments L3={σ±100iT:1−c(logT)2/3≤σ≤1}. The integrals along the line segments L2 and L3 are trivially bounded by O(T−100) by the rapid decay of w˜. The new line L1 gives an amount that is ≪Nlog8/3Texp(−clogN(logT)2/3)≪N(logT)100, (5.4) using the Vinogradov–Korobov bound ∣ζ(σ+it)∣≪(log∣t∣)2/3 for t≫1 and 1−c(log∣t∣)2/3≤σ≤1 (see [11, Corollary 8.28]). Here what is important is that σ can be taken to be 1−O((log∣t∣)−1+δ), for some small δ>0. We need to analyze the residues of the poles. Temporarily assume that t≠±r. The poles at s=1+it+ir and s=1−it−ir have very small residues from the rapid decay of w˜(s). The residue at s=1+ir−it contributes R1=N1+ir−itw˜(1+ir−it)ζ(2(1+ir−it))ζ(1+2ir)ζ(1+2ir−2it)ζ(1−2it). (5.5) By symmetry the residue at s=1−ir+it, say R2, is the same as R1 but with r and t interchanged. Let us write r=t+η (by assumption, η=O((logN)−1−δ), so R1=N1+iηw˜(1+iη)ζ(2(1+iη))ζ(1+2iη)ζ(1+2it+2iη)ζ(1−2it). (5.6) By simple Taylor approximations, we have N1+iηw˜(1+iη)ζ(2(1+iη))ζ(1+2iη)=w˜(1)NlogN2ζ(2)(1iηlogN+1+O(∣η∣logN)), (5.7) and by the Vinogradov–Korobov bound ζ′ζ(1+2it)≪(log∣t∣)2/3+ε (see [11, Theorem 8.29]), we have ζ(1+2it+2iη)ζ(1−2it)=∣ζ(1+2it)∣2(1+O(∣η∣(logT)2/3+ε)). (5.8) Combining (5.7) and (5.8), we derive R1=Nw˜(1)∣ζ(1+2it)∣2logN2ζ(2)·[1+1iηlogN+O(∣η∣logN+(logT)2/3+εlogN+∣η∣(logT)2/3+ε)]. (5.9) The conditions η≪(logN)−1−δ, (logN)≫(logT)2/3+δ are enough to imply that the error term in (5.9) is o(1). Similarly, R2=Nw˜(1)∣ζ(1+2it)∣2logN2ζ(2)[1+1−iηlogN+o(1)], (5.10) and, therefore, R1+R2=Nw˜(1)∣ζ(1+2it)∣2logNζ(2)(1+o(1)). (5.11) □ Remark. If we had moved the contour to the line σ=1/2, then instead of (5.4) we would have obtained an error term of size O(N1/2T1/3+ε) using Weyl’s subconvexity bound. This is o(N) for N≫T2/3+ε, which is far from our desired choice of N=T1/4. 6. Completion of the proof By Corollary 4.2, we have that ∣E(z,1/2+iT)∣2≪ylog6T+log5T∫∣r∣≤4logT∣E(z,1/2+iT+ir)∣2dr. (6.1) On the right-hand side above, we dissect the integral into subintervals, each of length ≍(logT)−2, say. Let U be one of these intervals, and choose a point tU∈U. Then by Lemma 5.1, we have ∫r∈U∣E(z,1/2+iT+ir)∣2dr≪N−2Tε∫r∈U∣AN(T+r,T+tU)∣2∣E(z,1/2+iT+ir)∣2dr. (6.2) By (2.5), with αn=w(n/N)τi(T+tU)(n), this is in turn ≪(NT)ε(TN+T1/2(N+N1/2y)). (6.3) If 1≪y≪T1/8, we choose N as T1/4, which in all gives ∣E(z,1/2+iT)∣≪T3/8+ε. (6.4) If T1/8≪y≪T1/6, we set N=y−2/3T1/3, giving ∣E(z,1/2+iT)∣≪y1/3T1/3+ε. (6.5) The bound (3.2) is superior to (6.5) for y≫T1/6. Funding This material is based upon work supported by the National Science Foundation under Agreement no. DMS-1401008. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. References 1 C. Balogh , Asymptotic expansions of the modified Bessel function of the third kind of imaginary order , SIAM J. Appl. Math. 15 ( 1967 ), 1315 – 1323 . Google Scholar CrossRef Search ADS 2 V. Blomer and R. Holowinsky , Bounding sup-norms of cusp forms of large level , Invent. Math. 179 ( 2010 ), 645 – 681 . Google Scholar CrossRef Search ADS 3 V. Blomer and P. Maga , Subconvexity for sup-norms of cusp forms on PGL(n) , Selecta Math . 22 ( 2016 ), 1269 – 1287 . 4 V. Blomer and A. Pohl , The sup-norm problem on the Siegel modular space of rank two, Amer. J. Math . 136 ( 2016 ), 999 – 1027 . 5 É. Fouvry , E. Kowalski and P. Michel , Algebraic trace functions over the primes , Duke Math. J. 163 ( 2014 ), 1683 – 1736 . Google Scholar CrossRef Search ADS 6 I. S. Gradshteyn and I. M. Ryzhik , Table of Integrals, Series, and Products , 6th edn, Academic Press Inc , San Diego, CA, 2000 , xlvii, 1163p. Translated from the Russian, Translation edited and with a preface by Alan Jeffrey and Daniel Zwillinger. 7 G. Harcos and N. Templier , On the sup-norm of Maass cusp forms of large level: II , Int. Math. Res. Not. IMRN 2012 ( 2012 ), 4764 – 4774 . Google Scholar CrossRef Search ADS 8 G. Harcos and N. Templier , On the sup-norm of Maass cusp forms of large level: III , Math. Ann. 356 ( 2013 ), 209 – 216 . Google Scholar CrossRef Search ADS 9 D. R. Heath-Brown , The twelfth power moment of the Riemann-function, Quart , J. Math. Oxford Ser. (2) 29 ( 1978 ), 443 – 462 . Google Scholar CrossRef Search ADS 10 R. Holowinsky , G. Ricotta and E. Royer , On the sup-norm of SL3 Hecke-Maass cusp form, preprint, 2014 , http://arxiv.org/abs/1404.3622. 11 H. Iwaniec and E. Kowalski , Analytic Number Theory. American Mathematical Society Colloquium Publications, 53 , American Mathematical Society , Providence, RI , 2004 . 12 H. Iwaniec and P. Sarnak , L∞ norms of eigenfunctions of arithmetic surfaces , Ann. of Math. (2) 141 ( 1995 ), 301 – 320 . Google Scholar CrossRef Search ADS 13 E. Kiral , Bounds on sup-norms of half-integral weight modular forms , Acta Arith. 165 ( 2014 ), 385 – 400 . Google Scholar CrossRef Search ADS 14 W. Luo and P. Sarnak , Quantum ergodicity of eigenfunctions on PSL2(Z)\H2 , Inst. Hautes Études Sci. Publ. Math. 81 ( 1995 ), 207 – 237 . Google Scholar CrossRef Search ADS 15 S. Marshall , Sup norms of Maass forms on semisimple groups, preprint, 2014 http://arxiv.org/abs/1405.7033. 16 S. Ramanujan , Some formulæ in the analytic theory of numbers Collected papers of Srinivasa Ramanujan , AMS Chelsea Publ , Providence, RI , 2000 , 133 – 135 . 17 P. Shiu , A Brun-Titchmarsh theorem for multiplicative functions , J. Reine Angew. Math. 313 ( 1980 ), 161 – 170 . 18 F. Spinu , The L4 norm of the Eisenstein series, Ph. D. Thesis, Princeton University. 2003 . 19 N. Templier , On the sup-norm of Maass cusp forms of large level , Selecta Math. (N.S.) 16 ( 2010 ), 501 – 531 . Google Scholar CrossRef Search ADS 20 N. Templier , Hybrid sup-norm bounds for Hecke-Maass cusp forms , J. Eur. Math. Soc. (JEMS) 17 ( 2015 ), 2069 – 2082 . Google Scholar CrossRef Search ADS 21 M. Young , The quantum unique ergodicity conjecture for thin sets , Adv. Math. 286 ( 2016 ), 9581016 . Google Scholar CrossRef Search ADS © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

The Quarterly Journal of MathematicsOxford University Press

Published: Apr 11, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off