Let p be a prime. In this paper, we study the sum
∑m≥1∑n≥1anλg(m)λf(m+pn)U(mX)V(nH)
for any newforms g∈Bk(1) (or B∗λ(1)) and f∈Bk(p) (or B∗λ(p)), with the aim of determining the explicit dependence on the level, where a={an∈C} is an arbitrary complex sequence. As a result, we prove a uniform bound with respect to the level parameter p, and present that this type of sum is non-trivial for any given H,X≥2.
Citation: Fei Hou, Bin Chen. On triple correlation sums of Fourier coefficients of cusp forms[J]. AIMS Mathematics, 2022, 7(10): 19359-19371. doi: 10.3934/math.20221063
[1] | Guangwei Hu, Huixue Lao, Huimin Pan . High power sums of Fourier coefficients of holomorphic cusp forms and their applications. AIMS Mathematics, 2024, 9(9): 25166-25183. doi: 10.3934/math.20241227 |
[2] | Fei Hou, Bin Chen . Triple correlation sums of coefficients of $ \theta $-series. AIMS Mathematics, 2023, 8(10): 25275-25287. doi: 10.3934/math.20231289 |
[3] | Fei Hou . On the exponential sums estimates related to Fourier coefficients of $ GL_3 $ Hecke-Maaß forms. AIMS Mathematics, 2023, 8(4): 7806-7816. doi: 10.3934/math.2023392 |
[4] | Xiuhai Fei, Zhonghua Wang, Cuixian Lu, Haifang Zhang . Higher Jordan triple derivations on $ * $-type trivial extension algebras. AIMS Mathematics, 2024, 9(3): 6933-6950. doi: 10.3934/math.2024338 |
[5] | Huafeng Liu, Rui Liu . The sum of a hybrid arithmetic function over a sparse sequence. AIMS Mathematics, 2024, 9(2): 4830-4843. doi: 10.3934/math.2024234 |
[6] | Wen Teng, Fengshan Long, Yu Zhang . Cohomologies of modified $ \lambda $-differential Lie triple systems and applications. AIMS Mathematics, 2023, 8(10): 25079-25096. doi: 10.3934/math.20231280 |
[7] | Li Zhou, Liqun Hu . Sum of the triple divisor function of mixed powers. AIMS Mathematics, 2022, 7(7): 12885-12896. doi: 10.3934/math.2022713 |
[8] | Stefano Bonaccorsi, Bernard Hanzon, Giulia Lombardi . A generalized Budan-Fourier approach to generalized Gaussian and exponential mixtures. AIMS Mathematics, 2024, 9(10): 26499-26537. doi: 10.3934/math.20241290 |
[9] | Tingting Jiang, Jiantao Jiang, Jing An . An efficient Fourier spectral method and error analysis for the fourth order problem with periodic boundary conditions and variable coefficients. AIMS Mathematics, 2023, 8(4): 9585-9601. doi: 10.3934/math.2023484 |
[10] | S. Akansha, Aditya Subramaniam . Exploring Chebyshev polynomial approximations: Error estimates for functions of bounded variation. AIMS Mathematics, 2025, 10(4): 8688-8706. doi: 10.3934/math.2025398 |
Let p be a prime. In this paper, we study the sum
∑m≥1∑n≥1anλg(m)λf(m+pn)U(mX)V(nH)
for any newforms g∈Bk(1) (or B∗λ(1)) and f∈Bk(p) (or B∗λ(p)), with the aim of determining the explicit dependence on the level, where a={an∈C} is an arbitrary complex sequence. As a result, we prove a uniform bound with respect to the level parameter p, and present that this type of sum is non-trivial for any given H,X≥2.
A basic but important problem in number theory is the triple correlation sums problem, which concerns the non-trivial bounds for
∑h≤H∑n≤Xa(n)b(l1n+l2h)c(h)or∑h≤H∑n≤Xa(n)b(n+l1h)c(n+l2h). |
Here, a(n),b(n) and c(n) are three arithmetic functions, H,X≥2 and l1,l2∈Z. These type of sums play the vital roles of their own in many topics, such as the moments of L-functions (or zeta-functions), subconvexity, the Gauss circle problem and the Quantum Unique Ergodicity (QUE) conjecture, etc (see for instance [1,5,7,8,9,11,12,13,15,19] and the references therein). In the case of all the arithmetic functions being the divisor functions, Browning [4] found that
∑h≤H,n≤Xa(n)b(n+h)c(n+2h)=118Υ(h)∏p(1−1p)2(1+2p)HXlog3X+o(HXlog3X) |
for certain function Υ(h), provided that H≥X3/4+ε. It is remarkable that Blomer [2] used the spectral decomposition for partially smoothed triple correlation sums to prove that
∑h≥1∑n≤XW(hH)τ(n)a(n+h)τ(n+2h)=HˆW(1)∑n≤Xa(n)∑d≥1S(2n,0;d)d2×(logn+2γ−2logd)2+O((H2√X+HX14+√XH+X√H)‖a‖2). |
Here, the gamma constant γ≈0.57721, W is a bump function supported on [1/2,5/2], satisfying that xjW(j)(x)≪1 for any j∈N+; while, ˆW is the Mellin transform of W and
‖a‖2=√∑n≤X|a2(n)| |
is the ℓ2-norm. Notice that, the parameter H is reduced to H≥X1/3+ε. Let k,k′∈2N+. Let f1∈B∗k(1) and f2∈B∗k′(1) be two Hecke newforms on GL2, with λf1 and λf2 being their n-th Hecke eigenvalues, respectively (see §2 for definitions). Subsequently, Lin [20] claimed that
∑h≥1∑n≤XW(hH)λf1(n)a(n+h)λf2(n+2h)≪Xε(√XH+X√H)‖a‖2, |
which is non-trivial, provided that H≥X2/3+ε. Recently, Singh [28], however, were able to attain that, for any f1,f2,f3∈B∗k(1) (or B∗λ(1)) and some constant η>0,
∑h≥1∑n≥1W1(hH)W2(nX)λf1(n)λf2(n+h)λf3(n+2h)≪X1−η+εH, |
where W1,W2 are any bump functions compactly supported on [1/2,5/2] with bounded derivatives. Here, for any N∈N+, B∗λ(N) denotes the collection of the primitive newforms of Laplacian eigenvalue λ on Γ0(N) (see §2 for backgrounds). Until now, the best result is due to Lü-Xi [21,22] who achieved that
∑h≥1∑n≤XW(hH)a(n)b(n+h)λf1(n+2h)≪XεΔ1(X,H)‖a‖2‖b‖2, |
which allows one to take H≥X2/5+ε. Here, the definition of Δ1(X,H) can be referred to [22,Theorem 3.1]. More recently, Hulse et al. [14] successfully attained
∑h≥1∑n≥1λg1(n)λg2(h)λg3(2n−h)exp(−hH−nX)≪Xκ−1+ϑ+12+εHκ−12−ϑ+12+ε, | (1.1) |
where ϑ<7/64 denotes the currently best record for the Generalized Ramanujan Conjecture. Here, λg1(n) (resp. λg1(n) and λg1(n)) denote the n-th non-normalized coefficients of holomorphic cusp forms g1 (resp. g2 and g3), each of weight κ≥2 and level M≥2.
In the present paper, we aim to consider the level aspect for the triple correlation sums. It is noticeable that, just lately, Munshi [26] obtained that, whenever X1/3+ε≤p≤X, one has the inequality
∑n≥1λf(n)λf(n+pm)≪p14X34+ε | (1.2) |
for any f∈B∗k(p) and fixed integer m such that |m|≤X/p. In this paper, we would like to go further obtaining the following quantitive estimate.
Theorem 1.1. Let X,H≥2, and p be a prime such that p≤X. Let U,V be two smooth weight functions supported [1/2,5/2] with bounded derivatives. Then, for any sequence a={an∈C} and any newforms g∈B∗k(1) (or B∗λ(1)) and f∈B∗k(p) (or B∗λ(p)), we have
∑m≥1∑n≥1anλg(m)λf(m+pn)U(mX)V(nH)≪Xεmax(√XHp,X)‖a‖2, | (1.3) |
where the implied constant depends only on the weight k (or the spectral parameter λ) and ε.
Remark 1.2. Our main result (1.3) is non-trivial for any given parameters X and H; particularly, for any automorphic cusp form π of any rank N, N≥2, with λπ(n) being its n-th normoailzed Fourier coefficient, we find
∑m≥1∑n≥1λπ(n)λg(m)λf(m+pn)U(mX)V(nH)≪Xεmax(H√Xp,X√H) |
by the Rankin-Selberg's bound, which says that ∑n≤X|λπ(n)|2≪π,εX1+ε (see for instance [6,Remark 12.1.8]).
Remark 1.3. The merits that comes from (1.3) is that the implied constant does not depend on the level parameter anymore. One may verify that our result (1.3), however, is exhibited to be a strengthened upper-bound whenever H≥√X/p, compared with an application of Munshi's estimate (1.2). Indeed, Munshi's estimate implies the upper-bound ≪p1/4X3/4+ε√H‖a‖2 for the triple sum above. Moreover, one might save roughly a magnitude of √H in the interesting case of pH≍X; in the average sense, the main result improves upon the estimate due to Munshi. One, on the other hand, wanders whether or not the non-trivial bounds for the scenarios where the cusp forms f,g being of higher rank could be achieved; we shall plan to investigates this problem in the future work.
Notations. Throughout the paper, ε always denotes an arbitrarily small positive constant. n∼X means that X<n≤2X. For any integers m,n, (m,n) means the great common divisor of m,n. Finally, μ(n) denotes the Möbius function of n.
For any k∈2N+ and N∈N+, let us denote by Sk(N) the vector space of the normalized holomorphic cusp forms on Γ0(N) of weight k and trivial nebentypus. Whenever f∈Sk(N), one has
f(z)=∑n≥1λf(n)nk−12e(nz) |
for Im(z)>0. We also denote by Sλ(N) the vector space of the normalized Maaß forms on Γ0(N) of weight 0, Laplacian eigenvalue λ=1/4+r2 (r∈R) and trivial nebentypus. For any f∈Sλ(N), there exists the following Fourier expansion
f(z)=2√|y|∑n≠0λf(n)Kir(2π|ny|)e(nx), |
where z=x+iy. The set of the primitive forms B∗k(N) (or B∗λ(N)) consists of common eigenfunctions of all the Hecke operators Tn for any n≥1. Regarding the individual bounds for λf(n), we have
λf(n)≪(nN)ε, | (2.1) |
whenever f∈B∗k(N) (or B∗λ(N)).
We will need the Voronoi summation formula in the analysis; see [18,Theorem A.4].
Lemma 2.1. Let k,N and the form f be as before. For any a,q∈N+ such that (a,q)=1, we set N2:=N/(N,q). Let h be a bump function of bounded derivatives. Then, there exists a constant ϱ of modulus one and a newform f⋆∈B∗k(N) (or B∗λ(N)) such that
∑n≥1λf(n)e(anq)h(nX)=2πϱq√N2∑n≥1λf⋆(n)e(−¯aN2nq)H♭(nXq2N2;h)+2πϱq√N2∑n≥1λf⋆(n)e(¯aN2nq)H♮(nXq2N2;h), |
where
H♭(x;h)=∫∞0h(ξ)Jf(4π√xξ)dξ,andH♮(x;h)=∫∞0h(ξ)Kf(4π√xξ)dξ. |
Here, if f is holomorphic
Jf(x)=2πikJk−1(x),Kf(x)=0; |
while, if f is a Maaß form
Jf(x)=−πsin(πir)(J2ir(x)−J−2ir(x)),Kf(x)=4cosh(πr)K2ir(x). |
For any s∈R, one may write
Jk−1(s)=s−12(F+k(s)e(x)+F−k(s)e(−s)) | (2.2) |
for some smooth functions F± satisfying that
sjF±k(j)(s)≪k,js(1+s)32 |
for any j≥0; the resource might be referred to [29,Section 6.5] if s<1 and [29,Section 3.4] if s≥1. One thus sees that, for H♭, the n-variable is essentially truncated at n≪q2N2/X1−ε, by repeated integration by parts. Furthermore, notice that, by Appendix of [3],
K2ir(s)≪r,ε{s−12exp(−s),s>1+π|r|,(1+|r|)ε,0<s≤1+π|r|; | (2.3) |
one will find the n-variable enjoys the analogous truncation range for H♮ with that for H♭.
Now, let us recall the following Wilton-type bound; the resource, however, may be referred to [17], together with [10,27].
Lemma 2.2. Let X≥2 and W be a smooth function, compactly supported on [1/2,5/2] such that xjW(j)(x)≪1 for any j∈N+. For any newform f∈B∗k(N) (or B∗λ(N)) and α∈R, we thus have
∑n≥1λf(n)e(nα)W(nX)≪√XN13+ε, | (2.4) |
where the implied ≪-constant depends merely on k (or λ) and ε.
As a variant of the circle method, the δ-symbol method plays a focal role in number theory. We will now briefly recall a version of the circle method; see for instance [16,Chapter 20].
Lemma 2.3. Fix X,Q≥1. For any n≤X, one has
δ(n)=1Q∑q≤Q1q∑amodq(a,q)=1e(anq)∫Rg(q,τ)e(nτqQ)dτ, |
where
g(q,τ)=1+h(q,τ)withh(q,τ)=O(1qQ(|τ|+qQ))A,τj∂j∂τjg(q,τ)≪logQmin(Qq,1|τ|), |
and g(q,τ)≪|τ|−A for any sufficiently large A. In particular, the effective range ofthe τ-integral is [−Xε,Xε].
We will have a need of the following lemmas which will be applied in §3.
Lemma 2.4. Let Q≥2. Let F(x,y) be a smooth bump functionsupported on [1/2,5/2]×[1/2,5/2], which satisfies that
XiYj∂i∂xi∂j∂yjF(xX,yY)≪i,j1 |
for any integers i,j∈N+ and any X,Y≥1. Then, for any c∈Z, sequence a={an∈C} and newform f∈B∗k(1) (or B∗λ(1)), there holds that
∑q≥1∑n≥1anS(n,c;q)F(nX,qQ)≪Xε(√XQ+1Q2>XQ2)‖a‖2, | (2.5) |
where the symbol 1P equals 1 if the assertion P is true, and 0 otherwise.
Proof. First, via the Cauchy-Schwarz inequality, we might evaluate the double sum as
≤(∑q1,q2≥1∑n≥1S(n,c;q1)¯S(n,c;q1)F(nX,q1Q)F(nX,q2Q))12‖a‖2. |
It thus follows from the Weil bound that the non-generic terms q1=q2 shall contribute a upper-bound ≪X1+εQ2 to the parentheses above, which gives the term √XQ‖a‖2 on the RHS of (2.5); while, for the generic terms q1≠q2, if one writes q1=ˆq1δ, q2=ˆq2δ with δ=(q1,q2) satisfying that (δ,ˆq1)=1, Poisson summation formula with the modulus ˆq1ˆq2δ thus might produce the following bound for the triple sum that
≪∑δ≤Q∑ˆq1,ˆq2≤Q/δsup0<|l|≪ˆq1ˆq2δ/X1−ε|∑αmodˆq1ˆq2δS(α,c;ˆq1δ)¯S(α,c;ˆq2δ)e(αlˆq1ˆq2δ)|. |
Notice, here, the inner-most sum vanishes, if l=0, and it is necessary that Q2>X as well. At this point, on applying Chinese remainder theorem, the sum over α turns out to be
ˆq1ˆq2δe(−c¯δl⋅ˆq2ˆq1)∑∑∗smodδe(¯aˆq1sδ−ˆq1c⋅¯(ˆq2s+l)ˆq2δ) |
with ¯δδ≡1modˆq1, ¯ll≡1modˆq1, ¯ˆq1ˆq1≡1modδ and ¯ˆq2s+l(ˆq2s+l)≡1modˆq2δ; trivially evaluating everything thus exactly leads to the term Q2‖a‖2 in (2.5).
Lemma 2.5. Let the parameters X,Q,c, the form f and the sequence a={an∈C} be as in Lemma 2.4. Let W(x,y,z) be a smooth bump functionsupported on [1/2,5/2]×[1/2,5/2]×[1/2,5/2], with the partial derivatives satisfying
XiYjZk∂i∂xi∂j∂yj∂k∂zkW(xX,yY,zZ)≪i,j,k1 |
for ever integers i,j,k∈N+ and any X,Y,Z≥1. There thus holds that
∑q≥1∑n≥1an∑m≥1λf(m)√mS(m−np,c;q)W(mX,nH,qQ)≪Xε(√HQ+Q2)‖a‖2. | (2.6) |
Proof. To show the lemma, the initial procedure is to invoke the Cauchy-Schwarz inequality; we are thus led to evaluating
∑q1,q2≥1∑n≥1∑m1,m2≥1λf(m1)¯λf(m2)√m1m2S(m1−np,c;q1)¯S(m2−np,c;q2)W(m1X,nH,q1Q)W(m2X,nH,q2Q). | (2.7) |
(1) First, let us begin with considering the generic terms q1=q2=q, say. In this moment, Poisson is applicable, which yields an alternative form for (2.7):
H∑l1∈Z∑q∼Q1q∑m1,m2≥1λf(m1)¯λf(m2)√m1m2Y0(l1,m1,m2,p,c;q)I0(l1,m1,m2), |
where
Y0(l,m1,m2,p,c;q)=∑αmodqS(m1−αp,c;q)¯S(m2−αp,c;q)e(−αlq), |
and
I0(l,m1,m2)=∫RW(m1X,ξ,qQ)W(m2X,ξ,qQ)e(lHξq)dξ. |
Notice that the exponential sum modulo q asymptotically equals
qe(−m1¯plq)∑∑∗αmodqe((m1−m2)α+n⋅¯α−¯pl−n¯αq), |
where we have employed the relation involving Ramanujan sum that
S(n,0;q)=∑ab=qμ(a)∑βmodqe(βnb). | (2.8) |
Upon combining with Lemma 2.2, one thus sees that the zero-frequency shall contribute a bound by ≪HQXε for any ε>0. While, on the other hand, if Q>H, one may find the non-zero frequencies will be indispensable to contribute a magnitude to (2.7). It can be demonstrated that, in this situation, the contribution, however, is estimated as ≪Q2Xε; this gives totally a quantity by
≪HQXε+1Q>HQ2Xε. | (2.9) |
(2) Now, we are left with the non-generic case where q1≠q2. One writes q1=q′1h, q2=q′2h, with (q1,q2)=h and (q′1,q′2)=1. Notice that h is co-prime with one of factors q′1,q′2; without loss of generality, one assumes that (h,q′1)=1. The expression in (2.7) thus becomes
∑h≪Q∑q1,q2≤Q/h∑n≥1∑m1,m2≥1λf(m1)¯λf(m2)√m1m2S(m1−np,c;q′1h)ׯS(m2−np,c;q′2h)W(m1X,nH,q′1hQ)W(m2X,nH,q′2hQ). | (2.10) |
Upon exploiting the Poisson twice, we thus arrive at
H∑l2∈Z∑h≪Q∑q′1,q′2≤Q/h1q′1q′2h∑m1,m2≥1λf(m1)¯λf(m2)√m1m2Y†(l2,m1,m2,p,c;q′1,q′2,h)I†(l2,m1,m2,q′1,q′2h), |
where the exponential sum Y† and the resulting integral I† are defined as
Y†(l,m1,m2,p,c;q′1,q′2,h)=∑αmodq′1q′2hS(m1−αp,c;q′1h)¯S(m2−αp,c;q′2h)e(−αlq′1q′2h), |
and
I†(l,m1,m2,q′1,q′2,h)=∫RW(m1X,ξ,q′1hQ)W(m2X,ξ,q′2hQ)e(lHξq′1q′2h)dξ. |
Here, one finds that the zero-frequency l2=0 does exist anymore. Indeed, upon recalling that (h,q′1)=1, one writes α=q′1¯q′1x+q′2h¯q′2y, with xmodq′2h and ymodq′1 such that (x,q′2h) and (y,q′1)=1; applying Chinese remainder theorem and (2.8), the sum over α thus essentially turns out to be
q′1q′2he(m1¯q′2pl+nq′2p¯hlq′1+m2¯q′1plq′2h)∑∑∗αmodhe((m1−m2)¯q′1q′2α−nq′1⋅¯q′2α−¯pl+nq′2¯q′1¯αq′2h). |
Via Lemma 2.2, it thus follows that the display (2.10) is dominated by ≪εXεQ4, upon opening the Kloosterman sum above. This, together with (2.9), shows the desired estimates in the parentheses of (2.6).
In this part, let us focus on the proof of Theorem 1.1. We shall first manage to separate the variables n,m by applying Lemma 2.3; in this paper, we shall employ a vital trick, that is, the 'conductor lowering mechanism' (see [23,24] or the survey [25]). One may see that actually there holds the following
δ(n)=1pQ∑q≤Q1q∑amodqp(a,q)=1e(anqp)∫Rg(q,τ)e(nτqQp)dτ; | (3.1) |
while, the parameter Q shall be taken as Q=√X/p. Now, for three smooth functions U,V,R, supported [1/2,5/2] with bounded derivatives, we shall detect the shift l=m+pn via (3.1), which yields an alternative form for the triple sum in (1.3) as follows
S(X,p,H)=X√HpQ∫R∑q≤Qg(q,τ)q∑γmodpq(γ,q)=1∑l≥1al√le(−γlq)Vτ(lH)∑m≥1λg(m)√m×e(mγpq)Uτ(mX)∑n≥1λf(n)√ne(−nγpq)Rτ(nX)dτ, | (3.2) |
where
Uτ(m)=U(m)e(mXτpqQ),Rτ(n)=R(n)e(−nXτpqQ),Vτ(u)=V(u)e(−uHτqQ). |
We shall proceed to distinguish whether (γ,p)=1 or not in the analysis, so that we are led to three parts, i.e., the non-degenerate term SNon−de., the degenerate term SDeg. and the error term SErr., which are respectively given by
SNon−de.(X,p,H)=X√HpQ∫R∑q≤Q(q,p)=1g(q,τ)q∑∑∗γmodpq∑l≥1al√le(−lγpq)Vτ(lH)×∑m≥1λg(m)√me(mγpq)Uτ(mX)∑n≥1λf(n)√ne(−nγq)Rτ(nX)dτ, | (3.3) |
SDeg.(X,p,H)=X√HpQ∫R∑q≤Q(q,p)=1g(q,τ)q∑∑∗γmodq∑l≥1al√ue(−lpγq)Vτ(lH)×∑m≥1λπ(1,m)√me(mγq)Uτ(mX)∑n≥1λf(n)√ne(−nγq)Rτ(nX)dτ, | (3.4) |
and
SErr.(X,p,H)=X√HpQ∫R∑q≤Qp|qg(q,τ)q∑∑∗γmodpq∑l≥1al√le(−lγpq)Vτ(lH)×∑m≥1λg(m)√me(mγpq)Uτ(mX)∑n≥1λf(n)√ne(−nγq)Rτ(nX)dτ. | (3.5) |
One might see that, here, it suffices to consider SNon−de.; the same argument works for SErr. which serves as a noisy term and for which we save more. We shall now now begin with SNon−de.; the analysis of the term SDeg. will be postponed to the end of this paper.
In this part, let us concentrate on the analysis of SNon−de.. For any ι,ν,υ,ρ∈R, write
Wτ(ι,ν,υ,ρ)=Uτ(ι)Rτ(ν)Vτ(υ)ηQ(ρQ), |
where ηQ is a smooth bump function supported on [1Q/2,5Q/2], satisfying that η(j)Q(x)≪1 for any j∈N+. One might find that the quantity we are focusing on is the following
X√HpQsupτ≪XεsupQ≤Q∑q≥1(q,p)=1g(q,τ)q∑∑∗γmodpq∑u≥1al√le(−lγq)∑m≥1λg(m)√m×e(mγpq)∑n≥1λf(n)√ne(−nγpq)Wτ(mX,nX,lH,qQ). | (3.6) |
We intend to invoking the Voronoi formula, Lemma 2.1; we thus arrive at
X√HpQsupτ≪XεsupQ≤Q∑q≥1(q,p)=1g(q,τ)q∑l≥1al√l∑m≥1λg(m)√m∑n≥1λf(n)√n×S(n−m,pl;pq){^W♭τ(mXpQ2,nXpQ2,lH,qQ)+^W♮τ(mXpQ2,nXpQ2,lH,qQ)}, | (3.7) |
where, for any ⋆,∗∈{♭,♮}, each integral ^W⋆,∗τ is defined as
^W⋆,∗τ(ι,ν,υ,ρ)=ηQ(ρQ)Vτ(υ)H⋆(Q2ιq2;Rτ)H∗(Q2νq2;Uτ). |
It is remarkable that, here, from (2.2) and (2.3), we have the identical crude estimate that ^W⋆,∗τ≪Xε for any ε>0; in this sense, one sees that it suffices to deal simply with ^W♭,♭τ, upon noticing that the argument of the other terms (i.e., ^W♭,♮τ, ^W♮,♭τ and ^W♮,♮τ) can follow similarly with it. One, however, on the other hand, sees that the inner-most sum modulo pq can be converted into
pS(¯p(n−m),l;q) |
with n≡mmodp. Now, if one writes n=m+pk with k≪Xε, we find that (3.6) is no more than
≪X√HQsupτ≪XεsupQ≤Q∑m≪pXε∑k≪Xελf(m)λf(pk+m)√m(pk+m)×∑q≥1(q,p)=1g(q,τ)q∑l≥1al√lS(l,k;q)^W♭,♭τ(mXpQ2,nXpQ2,lH,qQ). |
At this point, an application of Lemma 2.3 shows that the RHS of the expression above is bounded by
≪X1+εQsupτ≪XεsupQ≤Q(√H+1Q2>HQ)‖a‖2≪Xεmax(√XHp,X)‖a‖2, | (3.8) |
upon recalling the value of Q.
Now, let us have a look at the multiple-sum SDeg.. One might verify that SDeg. is of the form
X√HpQsupτ≪XεsupQ≤Q∑q≥1(q,p)=1g(q,τ)q∑∑∗γmodq∑l≥1al√ue(−lpγq)∑m≥1λg(m)√m×e(mγq)∑n≥1λf(n)√ne(−nγq)Wτ(nX,mX,uH,qQ) | (3.9) |
with Wτ being as before. We will now proceed by appealing to the Voronoi formula, Lemma 2.1, again to transform the sums over n into the dualized form, so that we infer that the expression above should be controlled by
X√HpQsupτ≪XεsupQ≤Qsupn≪pQ2X−1+ε|∑q≥1(q,p)=1g(q,τ)q∑l≥1al√l∑m≥1λg(m)√m=×S(m−lp,n;q){~W♭τ(mX,nXpQ2,lH,qQ)+~W♮τ(mX,nXpQ2,lH,qQ)}|, |
where, for ⋆∈{♭,♮}, each integral transform ~W⋆τ is given by
~W⋆τ(ι,ν,υ,ρ)=ηQ(ρQ)Vτ(υ)Uτ(ι)H⋆(Q2νq2;Rτ). |
Via Lemma 2.5, one thus deduces
SDeg.(X,p,H)≪X1+εpQsupτ≪XεsupQ≤Q(√H+Q)‖a‖2≪Xε(√XHp+Xp)‖a‖2. |
This leads to the estimates we would like to prove in Theorem 1.1, upon combining with (3.8).
In this paper, we investigate the triple correlations sums of Fourier coefficient of newforms on GL2, with the levels aspects being explicitly determined; our method is flexible enough to deal with the Maaß new forms. It is also remarkable that more recently, the authors are able to establish a sharp bound in the scenario where one of the froms f,g in (1.3) is a Maaß cuspidal form on GL3 (not necessarily self-dual) with the trivial level.
This research was funded by National Natural Science Foundation of China (grant numbers 413618022 and 61402335).
The authors declare that they have no conflicts of interest.
[1] |
V. Blomer, Shifted convolution sums and subconvexity bounds for automorphic L-functions, IMRN, 2004 (2004), 3905–3926. https://doi.org/10.1155/S1073792804142505 doi: 10.1155/S1073792804142505
![]() |
[2] |
V. Blomer, On triple correlations of divisor functions, Bull. Lond. Math. Soc., 49 (2017), 10–22. https://doi.org/10.1112/blms.12004 doi: 10.1112/blms.12004
![]() |
[3] |
V. Blomer, G. Harcos, P. Michel, A Burgess-like subconvex bound for twisted L-functions (with Appendix 2 by Z. Mao), Forum Math., 19 (2007), 61–105. https://doi.org/10.1515/FORUM.2007.003 doi: 10.1515/FORUM.2007.003
![]() |
[4] | T. D. Browning, The divisor problem for binary cubic form, J. Théor. Nombr. Bordx., 23 (2011), 579–602. |
[5] |
W. Duke, J. B. Friedlander, H. Iwaniec, Bounds for automorphic L-functions, Invent. Math., 112 (1993), 1–8. https://doi.org/10.1007/BF01232422 doi: 10.1007/BF01232422
![]() |
[6] | D. Goldfeld, Automorphic forms and L-functions for the group GL(n,R), Cambridge Studies in Advanced Mathematics, Vol. 99, Cambridge University Press, 2009. https://doi.org/10.1017/CBO9780511542923 |
[7] |
A. Good, Beitrage zur theorie der Dirichletreihen, die spitzenformen zugeordnet sind, J. Number Theory, 13 (1981), 18–65. https://doi.org/10.1016/0022-314X(81)90028-7 doi: 10.1016/0022-314X(81)90028-7
![]() |
[8] |
A. Good, Cusp forms and eigenfunctions of the Laplacian, Math. Ann., 255 (1981), 523–548. https://doi.org/10.1007/BF01451932 doi: 10.1007/BF01451932
![]() |
[9] |
G. Harcos, P. Michel, The subconvexity problem for Rankin-Selberg L-functions and equidistribution of Heegner points, Ⅱ, Invent. Math., 163 (2006), 581–655. https://doi.org/10.1007/s00222-005-0468-6 doi: 10.1007/s00222-005-0468-6
![]() |
[10] |
G. Harcos, N. Templier, On the sup-norm of Maaß cusp forms of large level, Ⅲ, Math. Ann., 356 (2013), 209–216. https://doi.org/10.1007/s00208-012-0844-7 doi: 10.1007/s00208-012-0844-7
![]() |
[11] |
D. R. Heath-Brown, The fourth power moment of the Riemann zeta function, Proc. Lond. Math. Soc., S3-38 (1979), 385–422. https://doi.org/10.1112/plms/s3-38.3.385 doi: 10.1112/plms/s3-38.3.385
![]() |
[12] |
R. Holowinsky, K. Soundararajan, Mass equidistribution for Hecke eigenforms, Ann. Math., 172 (2010), 1517–1528. http://doi.org/10.4007/annals.2010.172.1517 doi: 10.4007/annals.2010.172.1517
![]() |
[13] |
T. A. Hulse, C. I. Kuan, D. Lowry-Duda, A. Walker, Second moments in the generalized Gauss circle problem, Forum Math. Sigma, 6 (2018), 1–49. https://doi.org/10.1017/fms.2018.26 doi: 10.1017/fms.2018.26
![]() |
[14] |
T. A. Hulse, C. I. Kuan, D. Lowry-Duda, A. Walker, Triple correlation sums of coefficients of cusp forms, J. Number Theory, 220 (2021), 1–18. https://doi.org/10.1016/j.jnt.2020.08.007 doi: 10.1016/j.jnt.2020.08.007
![]() |
[15] | A. Ivić, A note on the Laplace transform of the square in the circle problem, Studia Sci. Math. Hung., 37 (2001), 391–399. |
[16] | H. Iwaniec, E. Kowalski, Analytic number theory, Vol. 53, American Mathematical Society Colloquium Publications, 2004. |
[17] |
M. Jutila, Sums of the additive divisor problem type and the inner product method, J. Math. Sci., 137 (2006), 4755–4761. https://doi.org/10.1007/s10958-006-0271-y doi: 10.1007/s10958-006-0271-y
![]() |
[18] |
E. Kowalski, P. Michel, J. VanderKam, Rankin-Selberg L-functions in the level aspect, Duke Math. J., 114 (2002), 123–191. https://doi.org/10.1215/S0012-7094-02-11416-1 doi: 10.1215/S0012-7094-02-11416-1
![]() |
[19] | Y. K. Lau, J. Liu, Y. Ye, Shifted convolution sums of Fourier coefficients of cusp forms, In: Number theory: Sailing on the sea of number theory, 2007,108–135. https://doi.org/10.1142/9789812770134_0005 |
[20] |
Y. Lin, Triple correlations of Fourier coefficients of cusp forms, Ramanujan J., 45 (2018), 841–858. https://doi.org/10.1007/s11139-016-9874-1 doi: 10.1007/s11139-016-9874-1
![]() |
[21] |
G. Lü, P. Xi, On triple correlations of Fourier coefficients of cusp forms, J. Number Theory, 183 (2018), 485–492. https://doi.org/10.1016/j.jnt.2017.08.028 doi: 10.1016/j.jnt.2017.08.028
![]() |
[22] |
G. Lü, P. Xi, On triple correlations of Fourier coefficients of cusp forms. Ⅱ, Int. J. Number Theory, 15 (2019), 713–722. https://doi.org/10.1142/S1793042119500374 doi: 10.1142/S1793042119500374
![]() |
[23] |
R. Munshi, The circle method and bounds for L-functions-Ⅱ. Subconvexity for twists of GL(3) L-functions, Amer. J. Math., 137 (2012), 791–812. https://doi.org/10.1353/AJM.2015.0018 doi: 10.1353/AJM.2015.0018
![]() |
[24] | R. Munshi, The circle method and bounds for L-functions-Ⅲ. t-aspect subconvexity for GL(3) L-functions, J. Amer. Math. Soc., 28 (2015), 913–938. |
[25] | R. Munshi, On some recent applications of circle method, Math. Students, 84 (2015), 23–38. |
[26] |
R. Munshi, On a shifted convolution sum problem, J. Number Theory, 230 (2022), 225–232. https://doi.org/10.1016/j.jnt.2020.12.011 doi: 10.1016/j.jnt.2020.12.011
![]() |
[27] |
A. Saha, Sup-norms of eigenfunctions in the level aspect for compact arithmetic surfaces, Math. Ann., 376 (2020), 609–644. https://doi.org/10.1007/s00208-019-01923-3 doi: 10.1007/s00208-019-01923-3
![]() |
[28] |
S. K. Singh, On double shifted convolution sum of SL(2,Z) Hecke eigenforms, J. Number Theory, 191 (2018), 258–272. https://doi.org/10.1016/j.jnt.2018.03.008 doi: 10.1016/j.jnt.2018.03.008
![]() |
[29] | G. N. Watson, A treatise on the theory of Bessel functions, Cambridge Mathematical Library, Cambridge University Press, Cambridge, 1995. |