Processing math: 100%
Research article

Mean value of the Hardy sums over short intervals

  • Received: 06 April 2020 Accepted: 22 June 2020 Published: 29 June 2020
  • MSC : 11F20, 11L40

  • The main purpose of this paper is using the mean value of Dirichlet L-functions and character sums to study a kind of mean value of the Hardy sums over short intervals, and give some interesting formulae.

    Citation: Lei Liu, Zhefeng Xu. Mean value of the Hardy sums over short intervals[J]. AIMS Mathematics, 2020, 5(6): 5551-5563. doi: 10.3934/math.2020356

    Related Papers:

    [1] Wenjia Guo, Yuankui Ma, Tianping Zhang . New identities involving Hardy sums $S_3(h, k)$ and general Kloosterman sums. AIMS Mathematics, 2021, 6(2): 1596-1606. doi: 10.3934/math.2021095
    [2] Jiankang Wang, Zhefeng Xu, Minmin Jia . Distribution of values of Hardy sums over Chebyshev polynomials. AIMS Mathematics, 2024, 9(2): 3788-3797. doi: 10.3934/math.2024186
    [3] Jiankang Wang, Zhefeng Xu, Minmin Jia . On the generalized Cochrane sum with Dirichlet characters. AIMS Mathematics, 2023, 8(12): 30182-30193. doi: 10.3934/math.20231542
    [4] Zhuoyu Chen, Wenpeng Zhang . A new reciprocity formula of Dedekind sums and its applications. AIMS Mathematics, 2024, 9(5): 12814-12824. doi: 10.3934/math.2024626
    [5] Xuan Wang, Li Wang, Guohui Chen . The fourth power mean of the generalized quadratic Gauss sums associated with some Dirichlet characters. AIMS Mathematics, 2024, 9(7): 17774-17783. doi: 10.3934/math.2024864
    [6] Yanbo Song . On two sums related to the Lehmer problem over short intervals. AIMS Mathematics, 2021, 6(11): 11723-11732. doi: 10.3934/math.2021681
    [7] Huimin Wang, Liqun Hu . Sums of the higher divisor function of diagonal homogeneous forms in short intervals. AIMS Mathematics, 2023, 8(10): 22577-22592. doi: 10.3934/math.20231150
    [8] Jinmin Yu, Renjie Yuan, Tingting Wang . The fourth power mean value of one kind two-term exponential sums. AIMS Mathematics, 2022, 7(9): 17045-17060. doi: 10.3934/math.2022937
    [9] Hu Jiayuan, Chen Zhuoyu . On distribution properties of cubic residues. AIMS Mathematics, 2020, 5(6): 6051-6060. doi: 10.3934/math.2020388
    [10] Xue Han, Tingting Wang . The hybrid power mean of the generalized Gauss sums and the generalized two-term exponential sums. AIMS Mathematics, 2024, 9(2): 3722-3739. doi: 10.3934/math.2024183
  • The main purpose of this paper is using the mean value of Dirichlet L-functions and character sums to study a kind of mean value of the Hardy sums over short intervals, and give some interesting formulae.


    For a positive integer q and an arbitrary integer h, the Dedekind sum S(h,q) is defined by

    S(h,q)=qa=1((aq))((ahq)),

    where

    ((x))={x[x]12,if x not be an integer;0,if x be an integer.

    It plays an important role in the transformation theory of the Dedekind η function. The various properties of S(h,q) were investigated by many authors, related works can be founded in [2,3,4] and [9,10]. Berndt [1] studied the following Hardy sums:

    H(h,q)=q1j=1(1)j+1+[hjq],s1(h,q)=q1j=1(1)[hjq]((jq)),s2(h,q)=q1j=1(1)j((jq))((hjq)),s3(h,q)=q1j=1(1)j((hjq)),s4(h,q)=q1j=1(1)[hjq],s5(h,q)=q1j=1(1)j+[hjq]((jq)).

    In [6], R. Sitaramachandra Rao expressed Hardy sums in terms of Dedekind sum S(h,q) as following:

    H(h,q)=8S(h+q,2q)+4S(h,q), if h+q is odd ;s1(h,q)=2S(h,q)4S(h,2q), if h is even ;s2(h,q)=S(h,q)+2S(2h,q), if q is even ;s3(h,q)=2S(h,q)4S(2h,q), if q is odd ;s4(h,q)=4S(h,q)+8S(h,2q), if h is odd ;s5(h,q)=10S(h,q)+4S(2h,q)+4S(h,2q), if h+q is even .

    Each one of H(h,q) (h+q even), s1(h,q) (h odd), s2(h,q) (q odd), s3(h,q) (q even), s4(h,q) (h even), s5(h,q) (h+q odd) is zero. Z. Xu and W. Zhang [8] proved that if p is a prime and ˉb denotes the multiplicative inverse of b modp, then

    a<p4b<p4H(2aˉb,p)=316p2+O(p1+ϵ).

    W. Liu [5] proved that if p5 is a prime and ˉb denotes the multiplicative inverse of b modp, then

    a<p3b<p3H(2aˉb,p)=15p2+O(p1+ϵ),
    a<p3b<p4H(2aˉb,p)=27320p2+O(p1+ϵ).

    Let 1<N<p, is there a mean value distribution formula of the Hardy sum in the short interval [1,N]? In fact, it is very difficult to get a asymptotic formula for the mean value in the case of one variable as

    a<NH(a,p).

    In this paper, we use the mean value of L-functions and estimate of character sums to study the mean value

    aNbNalbkH(2a¯b,p),

    where l,k be two non-negative integers and ¯b denote the multiplicative inverse of b modulo p, and obtain the following main conclusion:

    Theorem. Let p>2 be a prime and N a positive integer with 1<N<p, l and k be two non-negative integers. Then we have

    aNbNalbkH(2a¯b,p)=C(l,k)72l2ζ(3)(l+k+1)pNl+k+1+{O(pNl+klog2N+Nl+k+2po(1)),if l=k=0l=0,k1 or l1,k=0;O(pNl+klogN+Nl+k+2po(1)),if l1,k1.

    here

    C(l,k)=7ζ(3)8a=1b=a+1(a,2)=1,(b,2)=1al1bl+2b=12ba=b+1(a,2)=1,(b,2)=1al1bl+22l+k+1b=1a=2b+1(a,2)=1,(b,2)=1bk1ak+2

    is a constant depending on l and k.

    It is clear that the asymptotic formulas in the Theorem are non-trivial in the range pϵ<N<p1ϵ.

    From the theorem we may immediately deduce the following corollaries:

    Corollary 1. Let p>2 be a prime and N an integer with 1<N<p. Then we have

    aNbNaH(2a¯b,p)=C(1,0)7ζ(3)pN2+O(pNlog2N+N3po(1)),

    where

    C(1,0)=716ζ(3)π224a=1b=11(a+2b)392a=1b=11(a+2b)2b+2a=1b=11(a+4b)2b

    is a constant.

    Corollary 2. Let p>2 be a prime and N an integer with 1<N<p. Then we have

    aNbNabH(2a¯b,p)=2C(1,1)21ζ(3)pN3+O(pN2logN+N4po(1)),

    where

    C(1,1)=π282516ζ(3)10a=1b=11(a+2b)3+8a=1b=11(a+4b)3

    is a constant.

    To complete the proof of the theorem, we need several lemmas.

    Lemma 1. Let q3 and h be two integers with (h,q)=1. Then we have

    S(h,q)=1π2qd|qd2ϕ(d)χmoddχ(1)=1χ(h)|L(1,χ)|2,

    where ϕ(d) is the Euler function.

    Proof. See Lemma 1 of reference [9].

    Lemma 2. Let q>1 be an odd integer and h an integer with (h,q)=1. Then we have

    H(h,q)={16π2qd|qd2ϕ(d)χmoddχ(1)=1χ(h)|L(1,χχ02)|2,2h;0,2h.

    where χ02 be a principal Dirichlet character modulo 2.

    Proof. From [1], we have

    H(h,q)=8S(h+q,2q)+4S(h,q)=4π2qd|2qd2ϕ(d)χmoddχ(1)=1χ(h+q)|L(1,χ)|2+4π2qd|qd2ϕ(d)χmoddχ(1)=1χ(h)|L(1,χ)|2.

    Note that q be an odd integer, we can get

    d|2qf(d)=d|qf(d)+d|qf(2d)

    Hence we can write

    H(h,q)=4π2q(d|qd2ϕ(d)χmoddχ(1)=1χ(h+q)|L(1,χ)|2+d|q(2d)2ϕ(2d)χmod2dχ(1)=1χ(h+q)|L(1,χ)|2)+4π2qd|qd2ϕ(d)χmoddχ(1)=1χ(h)|L(1,χ)|2.=16π2qd|qd2ϕ(d)χmoddχ(1)=1χ(h+q)χ02(h+q)|L(1,χχ02)|2={16π2qd|qd2ϕ(d)χmoddχ(1)=1χ(h)|L(1,χχ02)|2,2h;0,2h.

    This proves Lemma 2.

    Lemma 3. Let p>2 be a prime, a be a positive integer with (a,p)=1. Then we have the asymptotic formula

    χmodpχ(1)=1χ(2aˉb)|L(1,χχ02)|2=χmodpχ(1)=1χ(2aˉb)|1up2χχ02(u)u|2+O(log2pp).

    Proof. For convenience, we let

    A(χ,y)=p2<uyχ(u).

    Then applying Abel's identity we have

    L(1,χχ02)=1up2χχ02(u)u+p2A(χχ02,y)y2dy.

    Hence

    χmodpχ(1)=1χ(2aˉb)|L(1,χχ02)|2=χmodpχ(1)=1χ(2aˉb)|1up2χχ02(u)u+p2A(χχ02,y)y2dy|2=χmodpχ(1)=1χ(2aˉb)|1up2χχ02(u)u|2+χmodpχ(1)=1χ(2aˉb)(1up2χχ02(u)u)(p2A(¯χχ02,y)y2dy)+χmodpχ(1)=1χ(2aˉb)(1up2¯χχ02(u)u)(p2A(χχ02,y)y2dy)+χmodpχ(1)=1χ(2aˉb)|p2A(χχ02,y)y2dy|2.

    Using the Cauchy inequality and the estimate |mnMχ(n)|plogp, we have the estimates

    |χmodpχ(1)=1χ(2aˉb)(1up2χχ02(u)u)(p2A(¯χχ02,y)y2dy)|[χmodpχ(1)=1|1up2χχ02(u)u|2]12[χmodpχ(1)=1|p2A(¯χχ02,y)y2dy|2]12logpp,
    |χmodpχ(1)=1χ(2aˉb)(1up2¯χχ02(u)u)(p2A(χχ02,y)y2dy)|logpp,

    and

    |χmodpχ(1)=1χ(2aˉb)|p2A(χχ02,y)y2dy|2|log2pp2.

    So we may immediately get the asymptotic formula

    χmodpχ(1)=1χ(2aˉb)|L(1,χχ02)|2=χmodpχ(1)=1χ(2aˉb)|1up2χχ02(u)u|2+O(log2pp).

    This proves Lemma 3.

    Lemma 4. Let p>2 be a prime and N an integer with 1<N<p. Then

    aNbNχmodpχ(2a¯b)|1up2χχ02(u)u|2=O(N2po(1)).

    Proof. From the orthogonality relation for character sums modulo p we have

    aNbNχmodpχ(2a¯b)|1up2χχ02(u)u|2=aNbNχmodpχ(2a¯b)1up2χχ02(u)u1vp2¯χχ02(v)v=ϕ(p)aNbN1up21vp22aubv(modp),(u,2p)=1,(v,2p)=11uv,

    Let J=2logp. By using the similar method used in [7] we can write

    ϕ(p)aNbN1up21vp22aubv(modp),(u,2p)=1,(v,2p)=11uvϕ(p)Ji,j=0eiu<ei+11uejv<ej+11vNa,b=12aubv(modp)12ϕ(p)0ijJeiu<ei+11uejv<ej+11vNa,b=12aubv(modp)12ϕ(p)0ijJe(i+j)eiu<ei+1ejv<ej+1Na,b=12aubv(modp)1

    Thus,

    aNbNχmodpχ(2a¯b)|1up2χχ02(u)u|22ϕ(p)0ijJe(i+j)Ki,j,

    where Ki,j is the number of solutions (a,b,u,v) to the congruence 2aubv(modp) with 1a,bN,eiu<ei+1 and ejv<ej+1.

    For a solution (a,b,u,v), write bv=|k|p2au. We have

    p|k|pmax{2au+bv}N(2ei+1+ej+1)3ej+1N.

    So the product bv can take at most

    3ej+1Npei+1N=3ei+j+2N2p

    values. It is clear that if a,u,k are fixed, then b and v can take at most po(1) possible values. Hence,

    Ki,j3ei+j+2N2ppo(1).

    Noting that

    0ijJe(i+j)=O(1).

    So we have

    aNbNχmodpχ(2a¯b)|1up2χχ02(u)u|2=O(N2po(1)).

    This proves Lemma 4.

    In this section we complete the proof of the Theorem.

    From Lemma 2 and Lemma 3 we have

    aNbNalbkH(2a¯b,p)=16pπ2ϕ(p)aNbNalbkχmodpχ(1)=1χ(2a¯b)|L(1,χχ02)|2=16pπ2ϕ(p)aNbNalbk(χmodpχ(1)=1χ(2aˉb)|1up2χχ02(u)u|2+O(log2pp))=8pπ2ϕ(p)aNbNalbkχmodp(1χ(1))χ(2a¯b)|1up2χχ02(u)u|2+O(Nl+k+2plog2pϕ(p))=8pπ2ϕ(p)aNbNalbkχmodpχ(2a¯b)|1up2χχ02(u)u|2+O(Nl+k+2plog2pϕ(p))+8pπ2ϕ(p)aNbNalbkχmodpχ(2a¯b)|1up2χχ02(u)u|2:=M1+M2+O(Nl+k+2plog2pϕ(p)) (1)

    From the orthogonality relations for characters modulo p, we have

    M1=8pπ2ϕ(p)aNbNalbkχmodpχ(2a¯b)|1up2χχ02(u)u|2=8pπ2aNbN1up21vp22aubv(modp),(u,2)=1,(v,2)=1albkuv=8pπ2aNbN1up21vp22au=bv,(u,2)=1,(v,2)=1albkuv8pπ2aNbN1up21vp22aubv,2aubv,(u,2)=1,(v,2)=1albkuv:=M11+M12

    here 1up2 denotes the summation over u from 1 to p2 with (u,p)=1.

    Now, we calculate M11. First, we write

    M11=8pπ2aNbN1up21vp22au=bv,(u,2)=1,(v,2)=1albkuv=2k+3pπ2dNaNdbN2d1vmin{p2a,p2b}(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v2=2k+3pπ2aNbNdmin{Na,N2b}1vmin{p2a,p2b}(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v2=2k+3pπ21a<Na<bNdN2b1vp2b(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v22k+3pπ21b<N22b<aNdNa1vp2a(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v22k+3pπ21bN2b<a<2bdN2b1vp2a(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v22k+3pπ2dN21vp2(v,2)=1dl+kv2.

    Note that

    1vp2(v,2)=1v2=π28+O(p2)

    and l+k0, we can get

    2k+3pπ2dN21vp2(v,2)=1dl+kv2=pNl+k+12l+1(l+k+1)+O(pNl+k).

    So, we can write

    M11:=A+B+CpNl+k+12l+1(l+k+1)+O(pNl+k). (2)

    We shall calculate the first three terms in the expression (2). First we calculate A. Write

    A=2k+3pπ21a<Na<bNdN2b1vp2b(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v2=pNl+k+12l+1(l+k+1)1a<Na<bN(a,b)=1,(a,2)=1,(b,2)=1al1bl2+O(pNl+k1a<Na<bNal1bl1),

    note that

    1a<Na<bN(a,b)=1,(a,2)=1,(b,2)=1al1bl2=1a<Na<bN(a,2)=1,(b,2)=1al1bl2d|(a,b)μ(d)=1a<Na<bN(a,2)=1,(b,2)=1al1bl2dNb(d,2)=1μ(d)d3,

    we have

    A=pNl+k+12l+1(l+k+1)1a<Na<bN(a,2)=1,(b,2)=1al1bl2(d=1(d,2)=1μ(d)d3d>Nb(d,2)=1μ(d)d3)+O(pNl+k1a<Na<bNal1bl1).

    By using the identity

    d=1(d,2)=1μ(d)ds=2s(2s1)ζ(s),

    we have

    A=pNl+k+172l2ζ(3)(l+k+1)1a<Na<bN(a,2)=1,(b,2)=1al1bl2+O(pNl+k1a<Na<bNal1bl1). (3)

    For the case l=0, we have

    A=4pNk+17ζ(3)(k+1)1a<Na<bN(a,2)=1,(b,2)=11ab2+O(pNk1a<Na<bN1ab)=4pNk+17ζ(3)(k+1)(a=1b=a+1(a,2)=1,(b,2)=11ab2a<Nb>N(a,2)=1,(b,2)=11ab2a>Nb>a(a,2)=1,(b,2)=11ab2)+O(pNklog2N)=4pNk+17ζ(3)(k+1)a=1b=a+1(a,2)=1,(b,2)=11ab2+O(pNklog2N). (4)

    For the case l1, we have

    A=pNl+k+172l2ζ(3)(l+k+1)1a<Na<bN(a,2)=1,(b,2)=1al1bl2+O(pNl+k1a<Na<bNal1bl1)=pNl+k+172l2ζ(3)(l+k+1)(a=1b=a+1(a,2)=1,(b,2)=1al1bl+2a<Nb>N(a,2)=1,(b,2)=1al1bl+2a>Nb>a(a,2)=1,(b,2)=1al1bl+2)+O(pNl+k1a<Na<bNal1bl1)=pNl+k+172l2ζ(3)(l+k+1)a=1b=a+1(a,2)=1,(b,2)=1al1bl+2+O(pNl+klogN). (5)

    By using the same method, we can calculate B:

    B=2k+3pπ21b<N22b<aNdNa1vp2a(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v2=2k+3pNl+k+17ζ(3)(l+k+1)1b<N22b<aN(a,2)=1,(b,2)=1ak2bk1+O(pNl+k1b<N22b<aNak1bk1).

    For the case k=0, we can also get

    B=23pNl+17ζ(3)(l+1)b=1a=2b+1(a,2)=1,(b,2)=11a2b+O(pNllog2N). (6)

    For the case k1, we can also get

    B=2k+3pNl+k+17ζ(3)(l+k+1)b=1a=2b+1(a,2)=1,(b,2)=1bk1ak+2+O(pNl+klogN). (7)

    By using the same method, we can also calculate C:

    C=2k+3pπ21bN2b<a<2bdN2b1vp2a(a,b)=1,(a,2)=1,(v,2)=1,(b,2)=1dl+kal1bk1v2=pNl+k+172l2ζ(3)(l+k+1)1bN2b<a<2b(a,2)=1,(b,2)=1al1bl+2+O(pNl+k1bN2b<a<2bal1bl+1).

    For the case l=0, we can also get

    C=4pNk+17ζ(3)(k+1)b=12ba=b+1(a,2)=1,(b,2)=11ab2+O(pNklog2N). (8)

    For the case l1, we can also get

    C=pNl+k+172l2ζ(3)(l+k+1)b=12ba=b+1(a,2)=1,(b,2)=1al1bl+2+O(pNl+klogN). (9)

    Then from (2)–(9), we can get

    M11=C(l,k)pNl+k+172l2ζ(3)(l+k+1)+{O(pNl+klog2N),if l=k=0l=0,k1 or l1,k=0;O(pNl+klogN),if l1,k1. (10)

    here

    C(l,k)=7ζ(3)8a=1b=a+1(a,2)=1,(b,2)=1al1bl+2b=12ba=b+1(a,2)=1,(b,2)=1al1bl+22l+k+1b=1a=2b+1(a,2)=1,(b,2)=1bk1ak+2

    is a constant related to l,k.

    Now, we calculate M12. Similarly to the proof of Lemma 4 we have

    M12=8pπ2aNbN1up21vp22aubv,2aubv,(u,2)=1,(v,2)=1albkuvpJi,j=0eiu<ei+11uejv<ej+11vNa,b=12aubv(modp)2aubvalbkpNl+k0ijJeijeiu<ei+1ejv<ej+1Na,b=12aubv(modp)2aubv1Nl+k+2po(1). (11)

    In addition, from Lemma 4 we have

    M2=8pπ2ϕ(p)aNbNalbkχmodpχ(2a¯b)|1up2χχ02(u)u|2Nl+kaNbNχmodpχ(2a¯b)|1up2χχ02(u)u|2Nl+k+2po(1). (12)

    Then combining (1) and (10)–(12), we obtain the asymptotic formula

    aNbNalbkH(2a¯b,p)=C(l,k)72l2ζ(3)(l+k+1)pNl+k+1+{O(pNl+klog2N+Nl+k+2po(1)),if l=k=0l=0,k1 or l1,k=0;O(pNl+klogN+Nl+k+2po(1)),if l1,k1.

    This completes the proof of the theorem.

    Taking l=1,k=0 and l=k=1 in the Theorem, we may immediately get the corollaries.

    The authors would like to thank the anonymous referee for very careful reading of the manuscript and helpful comments. This work is supported by the Basic Research Program for Nature Science of Shaanxi Province (2014JM1001, 2015KJXX-27) and N.S.F.(11971381, 11471258, 11701447) of P. R. China.

    We declare that we have no conflict of interest.



    [1] B. C. Berndt, Analytic Eisenstein series, theta-functions, and series relations in the spirit of Ramanujan, J. Reine Angew. Math., 1978 (1978), 332-365. doi: 10.1515/crll.1978.303-304.332
    [2] L. Carlitz, The reciprocity theorem of Dedekind sums, Pac. J. Math., 3 (1953), 513-522. doi: 10.2140/pjm.1953.3.513
    [3] J. B. Conrey, E. Fransen, R. Klein, et al. Mean values of Dedekind sums, J. Number Theory, 41 (1996), 214-226.
    [4] C. Jia, On the mean values of Dedekind sums, J. Number Theory, 87 (2001), 173-188. doi: 10.1006/jnth.2000.2580
    [5] W. Liu, Mean value of Hardy sums over short intervals, Acta Math. Acad. Paedagogicae Nyiregyhaziensis, 28 (2012), 1-11.
    [6] R. Sitaramachandra Rao, Dedekind and Hardy sums, Acta Arith., 48 (1987), 325-340. doi: 10.4064/aa-48-4-325-340
    [7] I. E. Shparlinski, on some weighted average value of L-functions, Bull. Aust. Math. Soc., 79 (2009), 183-186. doi: 10.1017/S0004972708001020
    [8] Z. Xu, W. Zhang, The mean value of Hardy sums over short intervals, P. Roy. Soc. Edinburgh, 137 (2007), 885-894. doi: 10.1017/S0308210505000648
    [9] W. Zhang, On the mean values of Dedekind sums, J. Theor. Nombr. Bordx., 8 (1996), 429-442. doi: 10.5802/jtnb.179
    [10] W. Zhang, A note on the mean square value of the Dedekind sums, Acta Math. Hung., 86 (2000), 275-289. doi: 10.1023/A:1006724724840
  • Reader Comments
  • © 2020 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(4158) PDF downloads(259) Cited by(0)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog