betaFeedback

WorksMathematics

Extreme central LL-values of almost prime quadratic twists of elliptic curves II

Abstract

We prove lower bounds for central LL-values of quadratic twists of an elliptic curve by discriminants with at most five prime factors. Under the rank-zero Birch–Swinnerton-Dyer formula, these bounds give large Tate–Shafarevich groups.

Introduction

The resonance method finds extreme values of LL-functions by weighting them with suitable Dirichlet polynomials. Soundararajan [12] used it to study large values of the Riemann zeta-function and extreme central values in families of LL-functions. We use the method to find large values in a sparse family of quadratic twists.

For quadratic twists of elliptic curves, the resonance method gives large central LL-values [6]. Hua and Huang [7] proved such bounds for discriminants with at most WW prime factors, for every fixed W≥20W \ge20. Here we obtain bounds for W≥5W \ge5.

Let E/QE/\mathbb{Q} be an elliptic curve of conductor NN, and let EdE_d be its quadratic twist by dd. We normalize the LL-function so that its central point is s=1/2s = 1/2. Put N0=lcm⁡(8,N)N_0 = \operatorname{lcm}(8,N). A class aa (mod N0N_0) is admissible if (a,N0)=1(a,N_0)=1 and a≡1a \equiv1 or 55 (mod 8). For such a class, let Da−\mathcal{D}^{-}_a be the set of negative fundamental discriminants d≡ad \equiv a (mod N0N_0). Write waw_a for the common root number of these twists. We fix an admissible class with wa=+1w_a = +1. For an integer W≥1W \ge1, let Fa,W(X)\mathcal{F}_{a,W}(X) be the set of d∈Da−d \in\mathcal{D}^{-}_a with X≤∣d∣≤2XX \le|d| \le2X and ω(∣d∣)≤W\omega(|d|) \le W, where ω(n)\omega(n) counts the distinct prime divisors of nn. Put HX=log⁡X/log⁡log⁡XH_X = \sqrt{\log X/\log\log X}.

Theorem 1. Let EE and aa (mod N0N_0) be as above. For every fixed integer W≥5W \ge5 and every fixed 0<c<(7W−34)/(7W+1)0 < c < \sqrt{(7W-34)/(7W+1)}, the set Fa,W(X)\mathcal{F}_{a,W}(X) is nonempty for all sufficiently large XX, and

max⁡d∈Fa,W(X)L(1/2,Ed)≥exp⁡(cHX).\max_{d\in\mathcal{F}_{a,W}(X)} L(1/2,E_d) \ge\exp(cH_X).

For W=5W = 5, one may take any 0<c<1/60 < c < 1/6. The proof combines the resonance method with a weighted sieve.

Large central values also give large Tate–Shafarevich groups under the Birch–Swinnerton-Dyer formula. Goldfeld and Szpiro [4] studied bounds for their orders, and Radziwiłł and Soundararajan [11] studied their distribution in quadratic twist families. When L(1/2,Ed)≠0L(1/2,E_d) \ne0, the group \Sha(E_d/\mathbb{Q}) is finite [5, 8]. The rank-zero formula relates its order to the central value, the real period, the rational torsion and the Tamagawa product. For the twists here, the period is proportional to ∣d∣−1/2|d|^{-1/2}, and the Tamagawa product is bounded when ω(∣d∣)≤W\omega(|d|) \le W. Hua and Huang [7] used these facts to find large Tate–Shafarevich groups in almost-prime twist families. Theorem 1 gives the following result with at most five prime factors.

Corollary 1. Let EE, aa, WW and cc be as in the Theorem 1. Assume the rank-zero Birch–Swinnerton-Dyer formula for the twists EdE_d with d∈Da−d \in\mathcal{D}^{-}_{a} and L(1/2,Ed)≠0L(1/2,E_d) \ne0. For every sufficiently large XX, there is d∈Fa,W(X)d \in\mathcal{F}_{a,W}(X) such that

#III(Ed/Q)≥Xexp⁡(cHX).(1)\#\mathrm{III}(E_d/\mathbb{Q}) \ge\sqrt{X}\exp(cH_X). \tag*{(1)}

For elliptic curves with complex multiplication the conclusion holds unconditionally.

The first moment and the resonator

Write L(s,E)=∑n≥1a(n)n−sL(s,E)=\sum_{n\ge1}a(n)n^{-s} and χd(n)=(dn)\chi_d(n)=\left(\frac{d}{n}\right). The central values Ld=L(1/2,Ed)L_d=L(1/2,E_d) are nonnegative [11]. Fix a nonzero nonnegative Φ∈Cc∞((1,2))\Phi\in C^\infty_c((1,2)).

For p∤N0p\nmid N_0, write Qp=1−(a(p)2−2)/p+p−2Q_p=1-(a(p)^2-2)/p+p^{-2} and Tp=1+p−1+Qp/pT_p=1+p^{-1}+Q_p/p. Define gg multiplicatively by g(1)=1g(1)=1, g(p2r−1)=Tp−1g(p^{2r-1})=T_p^{-1} and g(p2r)=(1+p−1)Tp−1g(p^{2r})=(1+p^{-1})T_p^{-1} for r≥1r\ge1. Put δ(p)=1−g(p2)=Qp/(pTp)\delta(p)=1-g(p^2)=Q_p/(pT_p), and extend δ\delta multiplicatively to squarefree integers coprime to N0N_0. By [6 Lemma 2.1, r = 0], for (t,N0)=1(t,N_0)=1 and t=t1t22t=t_1t_2^2 with t1t_1 squarefree,

∑d∈Da−Ldχd(t)Φ(∣d∣/X)=C1Xa(t1)t1g(t)+OE,Φ,ε(X1/2+εt11/2+ε).\sum_{d\in\mathcal{D}^{-}_{a}}L_d\chi_d(t)\Phi(|d|/X)=C_1X\frac{a(t_1)}{\sqrt{t_1}}g(t)+O_{E,\Phi,\varepsilon}\left(X^{1/2+\varepsilon}t_1^{1/2+\varepsilon}\right).

where C1>0C_1>0 depends only on the fixed data. Write

Sa(0;X,u,ℓ)=∑d∈Da−ℓ∣dLdχd(u)Φ(∣d∣/X).S_a(0;X,u,\ell)=\sum_{\substack{d\in\mathcal{D}^{-}_{a}\\ \ell\mid d}}L_d\chi_d(u)\Phi(|d|/X).

Fix θ>0\theta>0, 0<ρ<10<\rho<1, and put M=XθM=X^\theta, L=ρlog⁡Mlog⁡log⁡M\mathcal{L}=\rho\sqrt{\log M\log\log M}, P−=L2P_-=\mathcal{L}^2, and P+=exp⁡((log⁡L)2)P_+=\exp((\log\mathcal{L})^2). Define a real multiplicative function bb supported on squarefree integers by b(1)=1b(1)=1. For primes P−≤p≤P+P_-\le p\le P_+ with p∤N0p\nmid N_0, set b(p)=a(p)L/(plog⁡p)b(p)=a(p)\mathcal{L}/(\sqrt{p}\log p); set b(p)=0b(p)=0 for all other primes. Let R(d)=∑m≤Mb(m)χd(m)R(d)=\sum_{m\le M}b(m)\chi_d(m), A0(p)=1+b(p)2p/(p+1)A_0(p)=1+b(p)^2p/(p+1) and A1(p)=1+b(p)2g(p2)+2a(p)b(p)g(p)/pA_1(p)=1+b(p)^2g(p^2)+2a(p)b(p)g(p)/\sqrt{p}. Let Bi=∏p∤N0Ai(p)B_i=\prod_{p\nmid N_0}A_i(p) for i=0,1i=0,1. These products are finite. Rankin–Selberg theory and partial summation give

∑P−≤p≤P+a(p)2plog⁡p=1+o(1)2log⁡L,∑P−≤p≤P+a(p)2p=O(log⁡log⁡L);(2)\sum_{P_-\le p\le P_+}\frac{a(p)^2}{p\log p}=\frac{1+o(1)}{2\log\mathcal{L}},\qquad \sum_{P_-\le p\le P_+}\frac{a(p)^2}{p}=O(\log\log\mathcal{L}); \tag*{(2)}

see [11 Lemma 3]. Since P−→∞P_-\to\infty, the primes in this interval do not divide N0N_0 for large XX.

Lemma 1. For the coefficients above,

∑m≤M∣b(m)∣m≤M1+o(1).(3)\sum_{m\le M}|b(m)|\sqrt{m}\le M^{1+o(1)}. \tag*{(3)}

There is cρ>0c_\rho>0 such that, with τM=exp⁡(−cρlog⁡M/(log⁡log⁡M)3)\tau_M=\exp(-c_\rho\log M/(\log\log M)^3), uniformly for squarefree ℓ\ell coprime to N0N_0,

∑m1,m2≤M(m1m2,ℓ)=1b(m1)b(m2)g(m1m2)a((m1m2)1)(m1m2)1=∏p∤ℓN0A1(p)(1+O(τM)).(4)\sum_{\substack{m_1,m_2\le M\\(m_1m_2,\ell)=1}}b(m_1)b(m_2)g(m_1m_2)\frac{a((m_1m_2)_1)}{\sqrt{(m_1m_2)_1}}=\prod_{p\nmid\ell N_0}A_1(p)(1+O(\tau_M)). \tag*{(4)}

Here (m1m2)1(m_1m_2)_1 is the squarefree part of $m_1m_2.

Proof. Hasse’s bound ∣a(p)∣≤2|a(p)| \le2 and the elementary harmonic estimate give

∑m≤M∣b(m)∣m≤M∏P−≤p≤P+(1+∣a(p)∣Lplog⁡p)≤Mexp⁡(4L(1+(log⁡L)2))=M1+o(1).\sum_{m \le M}|b(m)|\sqrt{m} \le M \prod_{P_{-} \le p \le P_{+}}\left(1+\frac{|a(p)|\mathcal{L}}{p\log p}\right) \le M\exp\left(4\mathcal{L}\left(1+(\log\mathcal{L})^{2}\right)\right)=M^{1+o(1)}.

Here log⁡p≥log⁡2\log p \ge\log2, and the harmonic sum bounds the exponent sum by 4(1+log⁡P+)4(1+\log P_{+}) for large MM. Also L(1+(log⁡L)2)=o(log⁡M)\mathcal{L}(1+(\log\mathcal{L})^{2})=o(\log M). For the second assertion write m1=qrm_{1}=qr, m2=qsm_{2}=qs, with q,r,sq,r,s pairwise coprime and squarefree. The summand becomes

b(q)2g(q2)a(r)b(r)g(r)ra(s)b(s)g(s)s,b(q)^{2}g(q^{2})\frac{a(r)b(r)g(r)}{\sqrt{r}}\frac{a(s)b(s)g(s)}{\sqrt{s}},

which is nonnegative. The unrestricted sum is the product of A1(p)A_{1}(p) over p∤ℓN0p \nmid\ell N_{0}. Put ν=(log⁡L)−3\nu=(\log\mathcal{L})^{-3}. Rankin’s trick bounds the relative contribution of qr>Mqr>M or qs>Mqs>M by

2exp⁡(−νlog⁡M+∑p∤ℓN0(b(p)2g(p2)+a(p)b(p)g(p)p)(pν−1)).2\exp\left(-\nu\log M+\sum_{p\nmid\ell N_{0}}\left(b(p)^{2}g(p^{2})+\frac{a(p)b(p)g(p)}{\sqrt{p}}\right)(p^{\nu}-1)\right).

On the support, pν−1=νlog⁡p(1+o(1))p^{\nu}-1=\nu\log p(1+o(1)) uniformly. By (2), the quadratic term with the factor log⁡p\log p sums to (ρ2+o(1))log⁡M(\rho^{2}+o(1))\log M. The linear term sums to O(Llog⁡log⁡L)=o(log⁡M)O(\mathcal{L}\log\log\mathcal{L})=o(\log M). Omitting primes dividing ℓ\ell decreases these sums. The exponent is at most −(1−ρ2+o(1))νlog⁡M-(1-\rho^{2}+o(1))\nu\log M.

Lemma 2. We have log⁡(B1/B0)=(2ρθ+o(1))HX\log(B_{1}/B_{0})=(2\rho\sqrt{\theta}+o(1))H_{X}.

Proof. On the support of bb, put xp=2a(p)b(p)g(p)/px_{p}=2a(p)b(p)g(p)/\sqrt{p} and yp=b(p)2g(p2)y_{p}=b(p)^{2}g(p^{2}). Then

log⁡A1(p)A0(p)=log⁡1+yp1+b(p)2p/(p+1)+log⁡(1+xp1+yp).\log\frac{A_{1}(p)}{A_{0}(p)}=\log\frac{1+y_{p}}{1+b(p)^{2}p/(p+1)}+\log\left(1+\frac{x_{p}}{1+y_{p}}\right).

Since g(p),g(p2)=1+OE(p−1)g(p),g(p^{2})=1+O_{E}(p^{-1}) and b(p)2≪(log⁡L)−2b(p)^{2}\ll(\log\mathcal{L})^{-2}, this is

2a(p)b(p)g(p)p+O(b(p)2p+a(p)2b(p)2p+∣a(p)∣∣b(p)∣3p).\frac{2a(p)b(p)g(p)}{\sqrt{p}}+O\left(\frac{b(p)^{2}}{p}+\frac{a(p)^{2}b(p)^{2}}{p}+\frac{|a(p)||b(p)|^{3}}{\sqrt{p}}\right).

Partial summation gives total error O(L/(log⁡L)4)=o(HX)O(\mathcal{L}/(\log\mathcal{L})^{4})=o(H_{X}). Replacing g(p)g(p) by 1 in the linear term contributes o(1)o(1). By (2),

log⁡B1B0=(1+o(1))Llog⁡L=(2ρθ+o(1))HX.\log\frac{B_{1}}{B_{0}}=(1+o(1))\frac{\mathcal{L}}{\log\mathcal{L}}=(2\rho\sqrt{\theta}+o(1))H_{X}.

Divisor sums

Lemma 3. Let ℓ\ell be squarefree and (u,ℓ)=(uℓ,N0)=1(u,\ell)=(u\ell,N_{0})=1. Writing u=u1u22u=u_{1}u_{2}^{2} with u1u_{1} squarefree, we have

Sa(0;X,u,ℓ)=C1Xa(u1)u1g(u)δ(ℓ)+O(X1/2+εu11/2+εℓ1+ε).S_{a}(0;X,u,\ell)=C_{1}X\frac{a(u_{1})}{\sqrt{u_{1}}}g(u)\delta(\ell)+O\left(X^{1/2+\varepsilon}u_{1}^{1/2+\varepsilon}\ell^{1+\varepsilon}\right).

Proof. For a fundamental discriminant dd, 1ℓ∣d=∑q∣ℓμ(q)χd(q2)1_{\ell\mid d}=\sum_{q\mid\ell}\mu(q)\chi_{d}(q^{2}). Apply [6 Lemma 2.1, r = 0] to each insertion uq2uq^{2}. Since (u,ℓ)=1(u,\ell)=1, all squarefree parts are u1u_{1} and g(uq2)=g(u)g(q2)g(uq^{2})=g(u)g(q^{2}). As ℓ\ell is squarefree, ∑q∣ℓμ(q)g(q2)=δ(ℓ)\sum_{q\mid\ell}\mu(q)g(q^{2})=\delta(\ell). Thus ∑q∣ℓμ(q)g(uq2)=g(u)δ(ℓ)\sum_{q\mid\ell}\mu(q)g(uq^{2})=g(u)\delta(\ell). The errors sum to O(X1/2+εu1/2+ε∑q∣ℓq1+2ε)O(X^{1/2+\varepsilon}u^{1/2+\varepsilon}\sum_{q\mid\ell}q^{1+2\varepsilon}). Use ∑q∣ℓq1+2ε≪εℓ1+3ε\sum_{q\mid\ell}q^{1+2\varepsilon}\ll_{\varepsilon}\ell^{1+3\varepsilon} and choose smaller intermediate exponents. □

Lemma 4. Let M=XθM=X^{\theta}, D=XγD=X^{\gamma} and 4θ+4γ<14\theta+4\gamma<1. Put X=C1XB1\mathcal{X}=C_{1}XB_{1}, G(p)=δ(p)/A1(p)G(p)=\delta(p)/A_{1}(p) for p∤N0p\nmid N_{0}, and G(p)=0G(p)=0 for p∣N0p\mid N_{0}. Extend GG multiplicatively to squarefree integers. For squarefree ℓ\ell coprime to N0N_{0}, write

Aℓ=∑d∈Da−ℓ∣dL(1/2,Ed)R(d)2Φ(∣d∣/X)=XG(ℓ)+rℓ.A_{\ell}=\sum_{\substack{d\in D_{a}^{-}\\ \ell\mid d}}L(1/2,E_{d})R(d)^{2}\Phi(|d|/X)=\mathcal{X}G(\ell)+r_{\ell}.

Set rℓ=0r_{\ell}=0 for all other ℓ\ell. Then, for every fixed A>0A>0,

∑ℓ≤Dμ(ℓ)2=1(ℓ,N0)=1∣rℓ∣≪AX(log⁡X)−A.\sum_{\substack{\ell\leq D\\ \mu(\ell)^{2}=1\\ (\ell,N_{0})=1}}|r_{\ell}|\ll_{A}\mathcal{X}(\log X)^{-A}.

Proof. Expand R(d)2R(d)^{2}. Terms with (m1m2,ℓ)>1(m_{1}m_{2},\ell)>1 vanish. Apply Lemma 3 to the other terms. The truncation formula (4) gives the main term XG(ℓ)\mathcal{X}G(\ell) and a relative error O(τM)O(\tau_{M}). The coefficient bound (3) gives

∣rℓ∣≪XG(ℓ)τM+X1/2+εM2+o(1)ℓ1+ε.|r_{\ell}|\ll\mathcal{X}G(\ell)\tau_{M}+X^{1/2+\varepsilon}M^{2+o(1)}\ell^{1+\varepsilon}.

We have 0≤G(p)<1/p0\leq G(p)<1/p. To see this, set x=1/px=1/p and q=Qpq=Q_{p}. Hasse’s bound gives

(1−x)2≤q≤(1+x)2,Tp−q=(1−x)((1+x)2−q)+x2(1+x)>0.(1-x)^{2}\leq q\leq(1+x)^{2},\qquad T_{p}-q=(1-x)((1+x)^{2}-q)+x^{2}(1+x)>0.

Moreover g(p),g(p2)>0g(p),g(p^{2})>0 and a(p)b(p)≥0a(p)b(p)\geq0, so A1(p)≥1A_{1}(p)\geq1. It follows that 0≤G(p)≤δ(p)<1/p0\leq G(p)\leq\delta(p)<1/p when p∤N0p\nmid N_{0}; for p∣N0p\mid N_{0} this is immediate from G(p)=0G(p)=0. In particular, B1≥1B_{1}\geq1, and hence X≥C1X\mathcal{X}\geq C_{1}X. Thus

∑ℓ≤D∣rℓ∣≪XτMlog⁡D+X1/2+εM2+o(1)D2+ε.\sum_{\ell\leq D}|r_{\ell}|\ll\mathcal{X}\tau_{M}\log D+X^{1/2+\varepsilon}M^{2+o(1)}D^{2+\varepsilon}.

Indeed, for each sufficiently small fixed ε>0\varepsilon>0 the exponent of XX in the second term is at most 1/2+2θ+2γ+ε(1+γ)+o(1)1/2+2\theta+2\gamma+\varepsilon(1+\gamma)+o(1), which is strictly less than 11 by 4θ+4γ<14\theta+4\gamma<1. Since X≥C1X\mathcal{X}\geq C_{1}X, this gives a fixed power saving relative to X\mathcal{X}. The first term is smaller than X(log⁡X)−A\mathcal{X}(\log X)^{-A} for every fixed AA. □

The weighted sieve

Proof of Theorem 1. Choose fixed ρ∈(0,1)\rho\in(0,1) and θ,γ>0\theta,\gamma>0 with c<2ρθc<2\rho\sqrt{\theta}. We also require 35/(4(7W+1))<γ<1/4−θ35/(4(7W+1))<\gamma<1/4-\theta. Take the resonator of § 2, and put D=XγD=X^{\gamma}, z=D1/5z=D^{1/5}, P(z)=∏p<zp∤N0pP(z)=\prod_{\substack{p<z\\p\nmid N_{0}}}p and V(z)=∏p∣P(z)(1−G(p))V(z)=\prod_{p\mid P(z)}(1-G(p)). The inequality G(p)<1/pG(p)<1/p gives the uniform dimension-one condition and V(z)≫E1/log⁡zV(z)\gg_{E}1/\log z by Mertens’ theorem.

Let FF and ff be the upper and lower linear sieve functions [3 Theorem 11.13], and let C=2eγEMC=2e^{\gamma_{\mathrm{EM}}}, where γEM\gamma_{\mathrm{EM}} is Euler’s constant. The sieve formulas give F(t)=C/tF(t)=C/t for 1<t≤31<t\leq3 and f(4)=Clog⁡3/4f(4)=C\log3/4. For 3≤r≤43\leq r\leq4, the delay equation gives

rF(r)=C(1+∫3rlog⁡(t−2)t−1 dt)≤Cr3,rF(r)=C\left(1+\int_{3}^{r}\frac{\log(t-2)}{t-1}\,dt\right)\leq\frac{Cr}{3},

since log⁡(t−2)≤t−3≤(t−1)/3\log(t-2)\leq t-3\leq(t-1)/3 on this interval. It also gives

5f(5)=4f(4)+∫34F(t) dt≥Clog⁡4,f(5)≥Clog⁡45>f(4).5f(5)=4f(4)+\int_{3}^{4}F(t)\,dt\geq C\log4,\qquad f(5)\geq\frac{C\log4}{5}>f(4).

For 1<v≤41<v\le4, put I(v)=∫1v(1−t/v)F(5−t) dt/tI(v)=\int_{1}^{v}(1-t/v)F(5-t)\,\mathrm{d}t/t. Splitting at t=2t=2 and using the preceding bounds gives

I(4)≤C(815log⁡2−112−120log⁡3)<67f(4).I(4)\le C\left(\frac{8}{15}\log2-\frac{1}{12}-\frac{1}{20}\log3\right)<\frac{6}{7}f(4).

Thus I(v)≤I(4)<(6/7)f(5)I(v)\le I(4)<(6/7)f(5) for v≤4v\le4. Fix η=6/7\eta=6/7 and choose 1<v<41<v<4 close enough to 44 that γ>5/(v(W+1/7))\gamma>5/(v(W+1/7)). Put y=zvy=z^{v} and define the logarithmic weight

w(d)=1−1η∑z≤p<yp∣d(1−log⁡plog⁡y).w(d)=1-\frac{1}{\eta}\sum_{\substack{z\le p<y\\p\mid d}}\left(1-\frac{\log p}{\log y}\right).

For (d,P(z))=1(d,P(z))=1, all prime factors of ∣d∣|d| are at least zz. If ω(∣d∣)≥W+1\omega(|d|)\ge W+1, then

∑z≤p<yp∣d(1−log⁡plog⁡y)≥W+1−log⁡(2X)log⁡y>η\sum_{\substack{z\le p<y\\p\mid d}}\left(1-\frac{\log p}{\log y}\right)\ge W+1-\frac{\log(2X)}{\log y}>\eta

for large XX. Thus w(d)≤0w(d)\le0 on these twists, and w(d)≤1w(d)\le1 on all sifted twists. Write ad=L(1/2,Ed)R(d)2Φ(∣d∣/X)≥0a_d=L(1/2,E_d)R(d)^2\Phi(|d|/X)\ge0, and put

S(z)=∑(d,P(z))=1ad,Sp(z)=∑(d,P(z))=1p∣dad.S(z)=\sum_{(d,P(z))=1}a_d,\qquad S_p(z)=\sum_{\substack{(d,P(z))=1\\p\mid d}}a_d.

All sums over dd below are over Dα−\mathcal{D}_{\alpha}^{-}. The lower linear sieve at level DD gives S(z)≥XV(z)(f(5)+o(1))S(z)\ge\mathcal{X}V(z)(f(5)+o(1)). For z≤p<yz\le p<y, apply the upper linear sieve to the sequence {ad1p∣d}\{a_{d1_{p\mid d}}\} at level D/pD/p. Its divisor sums are XG(p)G(ℓ)+rpℓ\mathcal{X}G(p)G(\ell)+r_{p\ell} for ℓ∣P(z)\ell\mid P(z), and its sieve parameter is 5−log⁡p/log⁡z∈[5−v,4]5-\log p/\log z\in[5-v,4]. Consequently,

Sp(z)≤XG(p)V(z)(F(5−log⁡plog⁡z)+o(1))+∑ℓ≤D/pℓ∣P(z)∣rpℓ∣.S_p(z)\le\mathcal{X}G(p)V(z)\left(F\left(5-\frac{\log p}{\log z}\right)+o(1)\right)+\sum_{\substack{\ell\le D/p\\\ell\mid P(z)}}|r_{p\ell}|.

The o(1)o(1) is uniform in pp since 5−v>15-v>1 is fixed. Each pℓp\ell in this sum has exactly one prime factor at least zz, so Lemma 4 gives

∑z≤p<y∑ℓ≤D/pℓ∣P(z)∣rpℓ∣≤∑q≤D∣rq∣=o(XV(z)).\sum_{z\le p<y}\sum_{\substack{\ell\le D/p\\\ell\mid P(z)}}|r_{p\ell}|\le\sum_{q\le D}|r_q|=o(\mathcal{X}V(z)).

Here log⁡L=O(log⁡log⁡X)\log\mathcal{L}=O(\log\log X), so log⁡P+=o(log⁡X)\log P_{+}=o(\log X). Since log⁡z=(γ/5)log⁡X\log z=(\gamma/5)\log X, we have P+<zP_{+}<z for large XX. Also z>N0z>N_{0} eventually. Consequently, for every prime p≥zp\ge z, b(p)=0b(p)=0, p∤N0p\nmid N_{0}, and G(p)=δ(p)G(p)=\delta(p); the formulas above and Hasse’s bound give the uniform estimate ∣G(p)−1/p∣≤6/p2|G(p)-1/p|\le6/p^{2}. The function FF is bounded on the fixed compact interval [5−v,4][5-v,4], and the logarithmic weight lies in [0,1][0,1]. Hence replacing G(p)G(p) by 1/p1/p in the following sum changes it by at most Ov(∑n≥zn−2)=o(1)O_v(\sum_{n\ge z}n^{-2})=o(1). The prime number theorem and partial summation then give

∑z≤p<y(1−log⁡plog⁡y)G(p)F(5−log⁡plog⁡z)=I(v)+o(1).\sum_{z\le p<y}\left(1-\frac{\log p}{\log y}\right)G(p)F\left(5-\frac{\log p}{\log z}\right)=I(v)+o(1).

The sieve bounds give

T:=∑(d,P(z))=1adw(d)≥XV(z)(f(5)−I(v)η+o(1))≫XB1log⁡X.T := \sum_{(d,P(z))=1} a_d w(d) \geq X V(z)\left(f(5)-\frac{I(v)}{\eta}+o(1)\right) \gg\frac{X B_1}{\log X}.

Since T>0T>0, some term has w(d)>0w(d)>0 and ω(∣d∣)≤W\omega(|d|)\leq W. Discard the terms with w(d)≤0w(d)\leq0. For the remaining terms, w(d)≤1w(d)\leq1, so

T≤max⁡d∈Fa,W(X)L(1/2,Ed)∑d∈Da−R(d)2Φ(∣d∣/X).T \leq\max_{d\in\mathcal{F}_{a,W}(X)} L(1/2,E_d) \sum_{d\in\mathcal{D}_a^-} R(d)^2\Phi(|d|/X).

The sum on the right is ≪XB0\ll X B_0 by [6 Proposition 3.1], since θ<1/4\theta<1/4. Therefore

max⁡d∈Fa,W(X)L(1/2,Ed)≫B1B0log⁡X.\max_{d\in\mathcal{F}_{a,W}(X)} L(1/2,E_d) \gg\frac{B_1}{B_0\log X}.

By Lemma 2, log⁡(B1/B0)=(2ρθ+o(1))HX\log(B_1/B_0)=(2\rho\sqrt{\theta}+o(1))H_X. As log⁡log⁡X=o(HX)\log\log X=o(H_X) and c<2ρθc<2\rho\sqrt{\theta}, the asserted bound follows. □

Large Tate–Shafarevich groups

For an elliptic curve A/QA/\mathbb{Q}, our normalization gives L(1/2,A)=LHW(1,A)L(1/2,A)=L_{\mathrm{HW}}(1,A), where LHWL_{\mathrm{HW}} is the classical Hasse–Weil LL-function. Let ωA\omega_A be a minimal Néron differential and put Ω(A)=∫A(R)∣ωA∣\Omega(A)=\int_{A(\mathbb{R})}|\omega_A| and Tam⁡(A)=∏p<∞cp(A)\operatorname{Tam}(A)=\prod_{p<\infty}c_p(A). The integral includes all c∞(A)=#π0(A(R))c_\infty(A)=\#\pi_0(A(\mathbb{R})) connected components. Write H1(A(C),Z)±H^1(A(\mathbb{C}),\mathbb{Z})^\pm for the eigensubgroups of complex conjugation, and define Ω−(A)=∫γAωA\Omega^-(A)=\int_{\gamma_A}\omega_A using a generator γA\gamma_A of the minus subgroup. This period is purely imaginary and determined up to sign. Put

B(A)=L(1/2,A)#A(Q)tors2Ω(A)Tam⁡(A).(5)\mathcal{B}(A)=\frac{L(1/2,A)\#A(\mathbb{Q})_{\mathrm{tors}}^2}{\Omega(A)\operatorname{Tam}(A)}. \tag*{(5)}

The rank-zero Birch–Swinnerton-Dyer formula is

\#\Sha(A/\mathbb{Q})=\mathcal{B}(A) \qquad\text{when } L(1/2,A)\ne0. \tag*{(6)}

Lemma 5. Let d<0d<0 be a fundamental discriminant with (d,2N)=1(d,2N)=1. Then

Ω(Ed)=CE−∣d∣,CE−=c∞(E)∣Ω−(E)∣.(7)\Omega(E_d)=\frac{C_E^-}{\sqrt{|d|}}, \qquad C_E^-=c_\infty(E)|\Omega^-(E)|. \tag*{(7)}

The complementary period satisfies

∣Ω−(Ed)∣=Ω(E)c∞(E)∣d∣,∣d∣Ω(Ed)∣Ω−(Ed)∣=Ω(E)∣Ω−(E)∣.|\Omega^-(E_d)|=\frac{\Omega(E)}{c_\infty(E)\sqrt{|d|}}, \qquad|d|\Omega(E_d)|\Omega^-(E_d)|=\Omega(E)|\Omega^-(E)|.

The Tamagawa product satisfies

1≤Tam⁡(Ed)≪E4ω(∣d∣).(8)1\leq\operatorname{Tam}(E_d)\ll_E 4^{\omega(|d|)}. \tag*{(8)}

For any fixed W≥1W\geq1, we have B(Ed)≫E,W∣d∣L(1/2,Ed)\mathcal{B}(E_d)\gg_{E,W}\sqrt{|d|}L(1/2,E_d) if ω(∣d∣)≤W\omega(|d|)\leq W.

Proof. By [10 Theorem 3.2], Ω(Ed)=u~(d)c∞(Ed)∣Ω−(E)∣/∣d∣\Omega(E_d)=\tilde{u}(d)c_\infty(E_d)|\Omega^-(E)|/\sqrt{|d|}, where u~(d)\tilde{u}(d) is the minimal-model correction. Here dd is odd, squarefree, and congruent to 1 modulo 4. It is coprime to the minimal discriminant of EE, so u~(d)=1\tilde{u}(d)=1 by [10 Proposition 2.5 and Corollary 2.6]. Quadratic twisting preserves the sign of the discriminant, so c∞(Ed)=c∞(E)c_\infty(E_d)=c_\infty(E). This proves (7).

Choose the twisting isomorphism ϕ:Ed→E\phi: E_d \to E over C\mathbb{C} so that ϕ∗ωE=d ωEd\phi^*\omega_E = \sqrt{d}\,\omega_{E_d}; see [10 Lemma 3.1]. Complex conjugation satisfies σE∘ϕ=[−1]∘ϕ∘σEd\sigma_E \circ\phi= [-1] \circ\phi\circ\sigma_{E_d}. Thus ϕ∗\phi_* exchanges the two integral homology eigensubgroups and maps γEd\gamma_{E_d} to a generator of H1(E(C),Z)+H_1(E(\mathbb{C}),\mathbb{Z})^+. The absolute period of this generator is Ω(E)/c∞(E)\Omega(E)/c_\infty(E); see [10 Lemma 5.1]. Hence

∣d∣ ∣Ω−(Ed)∣=∣∫ϕ∗γEdωE∣=Ω(E)c∞(E).\sqrt{|d|}\,|\Omega^-(E_d)| = \left|\int_{\phi_*\gamma_{E_d}}\omega_E\right| = \frac{\Omega(E)}{c_\infty(E)}.

The complementary formula and the product identity follow.

We have cp(Ed)=1c_p(E_d)=1 for p∤dNp\nmid dN. If p∣dp\mid d, then EE has good reduction at pp and EdE_d has type I0∗I_0^*, with cp(Ed)=#E(Fp)[2]∈{1,2,4}c_p(E_d)=\#E(\mathbb{F}_p)[2]\in\{1,2,4\}; see [1 Theorem 4.1 and Table 4]. For p∣2Np\mid2N, the local twist depends only on the squareclass of dd in Qp×\mathbb{Q}_p^\times. There are finitely many such classes, so their Tamagawa numbers are bounded in terms of EE. This gives (8). Since #Ed(Q)tors≥1\#E_d(\mathbb{Q})_{\mathrm{tors}}\ge1, the last bound follows from (7), (8) and (5). □

Proof of Corollary 1. Take c′c' such that c<c′<(7W−34)/(7W+1)c<c'<\sqrt{(7W-34)/(7W+1)}. Theorem 1 gives d∈Fa,W(X)d\in\mathcal{F}_{a,W}(X) with L(1/2,Ed)≥exp⁡(c′HX)L(1/2,E_d)\ge\exp(c'H_X). Lemma 5 gives B(Ed)≫E,WXexp⁡(c′HX)B(E_d)\gg_{E,W}\sqrt{X}\exp(c'H_X). Since c′>cc'>c, we have B(Ed)≥Xexp⁡(cHX)B(E_d)\ge\sqrt{X}\exp(cH_X) for all sufficiently large XX.

The nonzero central value implies that Ed(Q)E_d(\mathbb{Q}) and \Sha(E_d/\mathbb{Q}) are finite by the Gross–Zagier–Kolyvagin theorem [5, 8]. Under (6), this proves (1). If EE has complex multiplication, then so does EdE_d. By [2 Corollary 2 and the paragraph following its proof, p. 362], the full rank-zero BSD formula holds for every CM elliptic curve over Q\mathbb{Q} with nonzero central value. This gives the unconditional assertion for CM curves. □

Remark 1 (The Tamagawa denominator). For d∈Da−d\in D_a^-, the local squareclass of dd is fixed at every p∣2Np\mid2N. The number BE,a=∏p∣2Ncp(Ed)B_{E,a}=\prod_{p\mid2N}c_p(E_d) is therefore constant, and Tam⁡(Ed)=BE,a∏p∣d#E(Fp)[2]\operatorname{Tam}(E_d)=B_{E,a}\prod_{p\mid d}\#E(\mathbb{F}_p)[2]. Frobenius acting on E[2]E[2] determines each factor; see also [11 §5]. We have

max⁡d∈Da−X≤∣d∣≤2Xlog⁡Tam⁡(Ed)=(log⁡4+o(1))log⁡Xlog⁡log⁡X.\max_{\substack{d\in D_a^-\\X\le|d|\le2X}}\log\operatorname{Tam}(E_d)=(\log4+o(1))\frac{\log X}{\log\log X}.

For the upper bound, use Tam⁡(Ed)≤BE,a4ω(∣d∣)\operatorname{Tam}(E_d)\le B_{E,a}4^{\omega(|d|)}. If r=ω(n)r=\omega(n), then n≥(r+1)!n\ge(r+1)!. Stirling’s formula gives ω(n)≤(1+o(1))log⁡X/log⁡log⁡X\omega(n)\le(1+o(1))\log X/\log\log X uniformly for n≤2Xn\le2X.

For the lower bound, let K=Q(E[2])K=\mathbb{Q}(E[2]) and δ=[K:Q]−1\delta=[K:\mathbb{Q}]^{-1}. Let PP be the primes p∤2Np\nmid2N that split completely in KK, and put πP(y)=#{p≤y:p∈P}\pi_P(y)=\#\{p\le y:p\in P\}. Each such prime satisfies #E(Fp)[2]=4\#E(\mathbb{F}_p)[2]=4. Chebotarev gives

πP(y)∼δylog⁡y,∑p≤yp∈Plog⁡p∼δy;\pi_P(y)\sim\frac{\delta y}{\log y},\qquad\sum_{\substack{p\le y\\p\in P}}\log p\sim\delta y;

see [9 Lemma 2.4]. Fix 0<ε<10<\varepsilon<1, and put y=(1−ε)log⁡X/δy=(1-\varepsilon)\log X/\delta and m=∏p≤yp∈Ppm=\prod_{\substack{p\le y\\p\in P}}p. Then log⁡m=(1−ε+o(1))log⁡X\log m=(1-\varepsilon+o(1))\log X and Y=X/m=Xε+o(1)Y=X/m=X^{\varepsilon+o(1)}. The prime number theorem in fixed arithmetic progressions gives a prime q∈[Y,2Y]q\in[Y,2Y] with q≡−am−1(modN0)q\equiv-am^{-1}\pmod{N_0}. For large XX, we have q>yq>y and q∤mq\nmid m. The integer d=−qmd=-qm is squarefree, belongs to Da−D_a^-, and satisfies X≤∣d∣≤2XX\le|d|\le2X. Therefore

log⁡Tam⁡(Ed)≥πP(y)log⁡4=((1−ε)log⁡4+o(1))log⁡Xlog⁡log⁡X.\log\operatorname{Tam}(E_d)\ge\pi_P(y)\log4=((1-\varepsilon)\log4+o(1))\frac{\log X}{\log\log X}.

Let ε→0\varepsilon\to0 to obtain the lower bound.

If E(Q)[2]≠0E(\mathbb{Q})[2] \ne0, the local product gives Tam⁡(Ed)≥BE,a2ω(∣d∣)\operatorname{Tam}(E_d) \ge B_{E,a}2^{\omega(|d|)}. If all 2-torsion points are rational, then Tam⁡(Ed)=BE,a4ω(∣d∣)\operatorname{Tam}(E_d) = B_{E,a}4^{\omega(|d|)}. For these curves, a uniform bound for Tam⁡(Ed)\operatorname{Tam}(E_d) also bounds ω(∣d∣)\omega(|d|). The almost prime condition controls this denominator for every fixed elliptic curve. More generally, the condition log⁡Tam⁡(Ed)=o(HX)\log\operatorname{Tam}(E_d) = o(H_X) on the selected twists preserves the central-value gain.

Acknowledgements

S.H. would like to thank Chung Pang Mok for the constant encouragement. S.H. is supported by the National Natural Science Foundation of China (No. 12601008) and the China Postdoctoral Science Foundation (No. 2026M793348).

AI Use Disclosure

The author introduced sieve weights into the argument, using calculations based on earlier work, and first reduced the number of prime factors to six. ChatGPT introduced Richert’s logarithmic weighted sieve to reduce the bound to five and prepared the draft.

References

Shanghai Institute for Mathematics and Interdisciplinary Sciences (SIMIS), Shanghai 200433, China

Research Institute of Intelligent Complex Systems, Fudan University, Shanghai, 200433, China

Email address: [email protected]

References

  1. [1]Barrios, A. J.; Roy, M.; Sahajpal, N.; Tallana, D.; Tobin, B.; Wiersema, H. Local data of elliptic curves under quadratic twist. Res. Number Theory 11 (2025), no. 3, Paper No. 75, 39 pp.DOI
  2. [2]Burungale, Ashay A.; Flach, Matthias. The conjecture of Birch and Swinnerton-Dyer for certain elliptic curves with complex multiplication. Cambridge J. Math. 12 (2024), no. 2, 357–415.arxiv.org/abs/2206.09874
  3. [3]Friedlander, John B.; Iwaniec, Henryk. Opera de cribro. American Mathematical Society Colloquium Publications 57, American Mathematical Society, 2010.
  4. [4]Goldfeld, Dorian; Szpiro, Lucien. Bounds for the order of the Tate-Shafarevich group. Compositio Math. 97 (1995), no. 1–2, 71–87.
  5. [5]Gross, Benedict H.; Zagier, Don B. Heegner points and derivatives of L-series. Invent. Math. 84 (1986), no. 2, 225–320.
  6. [6]Hamdar, Mohammad H.; Leung, Sun-Kai. Extreme values of central L-derivatives and heights of Heegner points. (Preprint). arXiv:2609.26786v1 (2026).arxiv.org/abs/2609.26786
  7. [7]Hua, Shenghao; Huang, Bingrong. Extreme central L-values of almost prime quadratic twists of elliptic curves. Sci. China Math. 66 (2023), 2755–2766.DOI
  8. [8]Kolyvagin, V. A. Finiteness of E(Q) and Ш(E,Q) for a subclass of Weil curves. Math. USSR-Izv. 32 (1989), no. 3, 523–541.DOI
  9. [9]Loughran, Daniel; Matthiesen, Lilian. Frobenian multiplicative functions and rational points in fibrations. J. Eur. Math. Soc. (JEMS) 26 (2024), no. 12, 4779–4830.arxiv.org/abs/1904.12845
  10. [10]Pal, Vivek. Periods of quadratic twists of elliptic curves. With an appendix by A. Agashe. Proc. Amer. Math. Soc. 140 (2012), no. 5, 1513–1525.arxiv.org/abs/1012.0094
  11. [11]Radziwiłł, Maksym; Soundararajan, Kannan. Moments and distribution of central L-values of quadratic twists of elliptic curves. Invent. Math. 202 (2015), 1029–1068.DOI
  12. [12]Soundararajan, Kannan. Extreme values of zeta and L-functions. Math. Ann. 342 (2008), no. 2, 467–486.arxiv.org/abs/0708.3990

Paper details

vlnc:W_s8tz94wq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT

View statement
Looking for related papers…
Contents