Richard Hoekstra Papers · Atlas · Data

Introduction

Let V=S2(Γ0(163)).V=S_2(\Gamma_0(163)). Since 1637(mod12)163\equiv 7\pmod{12}, the modular curve X0(163)X_0(163) has genus g=dimV=13.g=\dim V=13. The prime 163163 is the largest class-number-one Heegner prime . At level 163163 the relevant Heegner operator set is S={3,7,11,19,43,67,163},S=\{3,7,11,19,43,67,163\}, where for d=163d=163 we set T163:=U163T_{163}:=U_{163}, the Atkin–Lehner operator at the level. At prime level NN, the new subspace equals the full cuspidal space, and UNU_N acts as wN-w_N where wNw_N is the Atkin–Lehner involution . This convention is used consistently throughout.

For d,dSd,d'\in S define G(d,d)=Tr(TdTdV).G(d,d')=\operatorname{Tr}\!\bigl(T_dT_{d'}\mid V\bigr). The matrix G=(G(d,d))d,dSG=(G(d,d'))_{d,d'\in S} is symmetric (since the Hecke operators are self-adjoint with respect to the Petersson inner product and trace is cyclic) and integer-valued (since the operators preserve the integral modular-symbols lattice). Writing f1,,f13f_1,\ldots,f_{13} for the normalized Hecke eigenforms, G(d,d)=i=113ad(fi)ad(fi),G(d,d')=\sum_{i=1}^{13}a_d(f_i)\,a_{d'}(f_i), so GG is a Gram matrix and in particular positive semidefinite; positive definiteness follows from the computed eigenvalues below.

Our main result is the following.

For all d,dSd,d'\in S one has |Tr(TdTdV)|2min(Tr(Td2V),Tr(Td2V)).\bigl|\operatorname{Tr}(T_dT_{d'}\mid V)\bigr| \le 2\min\!\bigl(\operatorname{Tr}(T_d^2\mid V),\operatorname{Tr}(T_{d'}^2\mid V)\bigr).

The proof is computationally exact but not uniform. It splits the 2121 unordered pairs in SS into four categories:

  1. 99 pairs by Cauchy–Schwarz, because the larger diagonal is at most four times the smaller;

  2. 55 vacuum-row pairs (163,d)(163,d) by the prime-level Eichler–Selberg formula;

  3. 44 pairs by orbitwise Cauchy–Schwarz on the 1+5+71+5+7 newform decomposition;

  4. 33 residual pairs by explicit finite Hurwitz-class-number sums.

The proof is therefore exact but case-by-case. We do not know a uniform conceptual proof of Theorem [thm:main].

The Hecke Trace Gram Matrix

With rows and columns ordered as (3,7,11,19,43,67,163)(3,7,11,19,43,67,163), the trace matrix is G=(461812204470618921044460412101363888120620443828210266444881025831641370601206164640146466131413).G= \begin{pmatrix} 46 & -18 & -12 & 20 & -44 & -70 & 6\\ -18 & 92 & -10 & -44 & -4 & -60 & 4\\ -12 & -10 & 136 & -38 & -88 & 120 & 6\\ 20 & -44 & -38 & 282 & -102 & 6 & -6\\ -44 & -4 & -88 & -102 & 583 & -164 & -13\\ -70 & -60 & 120 & 6 & -164 & 640 & -14\\ 6 & 4 & 6 & -6 & -13 & -14 & 13 \end{pmatrix}. Its eigenvalues are approximately 10.37,19.44,76.78,98.09,274.14,491.62,821.55.10.37,\ 19.44,\ 76.78,\ 98.09,\ 274.14,\ 491.62,\ 821.55. In particular, GG is positive definite.

Proof. The entries of GG are computed in SageMath using the integral modular-symbols realization of VV. Symmetry follows from trace cyclicity: Tr(TdTd)=Tr(TdTd)\operatorname{Tr}(T_dT_{d'})=\operatorname{Tr}(T_{d'}T_d). The eigenvalues are computed numerically and are all positive, confirming positive definiteness. ◻

One has G(163,163)=13=g.G(163,163)=13=g.

Proof. At prime level NN, the operator UNU_N acts on the new subspace by UN=wNU_N=-w_N, where wNw_N is the Atkin–Lehner involution . Hence U1632=w1632=IdV,U_{163}^2=w_{163}^2=\mathrm{Id}_V, so G(163,163)=Tr(U1632V)=dimV=13.G(163,163)=\operatorname{Tr}(U_{163}^2\mid V)=\dim V=13. ◻

Proof of the Trace Inequality

We write B(d,d)=2min(G(d,d),G(d,d)).B(d,d')=2\min(G(d,d),G(d',d')). Theorem [thm:main] is the statement that |G(d,d)|B(d,d)|G(d,d')|\le B(d,d') for all d,dSd,d'\in S.

Nine pairs by Cauchy–Schwarz

The inequality of Theorem [thm:main] holds for the nine pairs (3,7),(3,11),(3,163),(7,11),(7,19),(11,19),(19,43),(19,67),(43,67).(3,7),\ (3,11),\ (3,163),\ (7,11),\ (7,19),\ (11,19),\ (19,43),\ (19,67),\ (43,67).

Proof. Since GG is a Gram matrix, Cauchy–Schwarz gives |G(d,d)|G(d,d)G(d,d).|G(d,d')|\le \sqrt{G(d,d)G(d',d')}. If max(G(d,d),G(d,d))4min(G(d,d),G(d,d))\max(G(d,d),G(d',d'))\le 4\min(G(d,d),G(d',d')), then G(d,d)G(d,d)2min(G(d,d),G(d,d)).\sqrt{G(d,d)G(d',d')} \le 2\min(G(d,d),G(d',d')). The nine pairs above satisfy this ratio condition. Their diagonal ratios are listed in Table 1. ◻

The nine pairs covered by Cauchy–Schwarz.
pair diagonals larger/smaller bound
(3,7)(3,7) (46,92)(46,92) 2.0002.000 9292
(3,11)(3,11) (46,136)(46,136) 2.9572.957 9292
(3,163)(3,163) (46,13)(46,13) 3.5383.538 2626
(7,11)(7,11) (92,136)(92,136) 1.4781.478 184184
(7,19)(7,19) (92,282)(92,282) 3.0653.065 184184
(11,19)(11,19) (136,282)(136,282) 2.0742.074 272272
(19,43)(19,43) (282,583)(282,583) 2.0672.067 564564
(19,67)(19,67) (282,640)(282,640) 2.2702.270 564564
(43,67)(43,67) (583,640)(583,640) 1.0981.098 11661166

Five vacuum-row pairs by Eichler–Selberg

For the vacuum row we use the prime-level Eichler–Selberg trace formula specialized to weight 22 and level 163163 . For nonsquare nn prime to 163163 it takes the form Tr(TnV)=12t2<4n(1+(Δ163))H(Δ)an(a,163)=1a,Δ=4nt2,\label{eq:es} \operatorname{Tr}(T_n\mid V) = \frac12\sum_{t^2<4n}\bigl(1+\bigl(\tfrac{\Delta}{163}\bigr)\bigr)H(\Delta) -\!\!\sum_{\substack{a\mid n\\(a,163)=1}}\! a, \qquad \Delta=4n-t^2, where H(Δ)H(\Delta) denotes the Hurwitz class number.

The inequality of Theorem [thm:main] holds for the five pairs (7,163),(11,163),(19,163),(43,163),(67,163).(7,163),\ (11,163),\ (19,163),\ (43,163),\ (67,163).

Proof. Write n=163pn=163p with p{7,11,19,43,67}p\in\{7,11,19,43,67\}. Then Δ=4163pt2t2(mod163).\Delta=4\cdot 163\,p-t^2\equiv -t^2\pmod{163}. Since 1633(mod4)163\equiv 3\pmod4, one has (1163)=1\bigl(\frac{-1}{163}\bigr)=-1. Hence if 163t163\nmid t, then $$\Bigl(\frac{\Delta}{163}\Bigr)=\Bigl(\frac{-t^2}{163}\Bigr)=-1,$$ so the factor 1+(Δ163)1+\bigl(\frac{\Delta}{163}\bigr) vanishes. Because t2<4163p<(2163)2,t^2<4\cdot 163\,p<(2\cdot 163)^2, the only possible surviving values are t=0t=0 and, for p{43,67}p\in\{43,67\}, also t=163t=163.

The five exact evaluations are: Tr(T1637)=12(24)8=4,Tr(T16311)=12(36)12=6,Tr(T16319)=12(28)20=6,Tr(T16343)=12(52+10)44=13,Tr(T16367)=12(44+64)68=14.\begin{aligned} \operatorname{Tr}(T_{163\cdot 7}) &= \tfrac12(24)-8 = 4,\\ \operatorname{Tr}(T_{163\cdot 11}) &= \tfrac12(36)-12 = 6,\\ \operatorname{Tr}(T_{163\cdot 19}) &= \tfrac12(28)-20 = -6,\\ \operatorname{Tr}(T_{163\cdot 43}) &= \tfrac12(52+10)-44 = -13,\\ \operatorname{Tr}(T_{163\cdot 67}) &= \tfrac12(44+64)-68 = -14. \end{aligned} Since G(163,163)=13G(163,163)=13, the required bound is always 213=262\cdot 13=26, and each of the five absolute values is at most 2626. ◻

Four pairs by the 1+5+7 orbit decomposition

The space VV decomposes over 𝐐\mathbf Q into three Hecke-stable subspaces corresponding to the three Galois conjugacy classes of normalized newforms:

This gives 1+5+7=13=dimV1+5+7=13=\dim V. Accordingly, the trace Gram matrix splits as G=C1+C5+C7,G=C_1+C_5+C_7, where Ck(d,d)=Tr(TdTdVk)=fVkad(f)ad(f).C_k(d,d')=\operatorname{Tr}(T_dT_{d'}\mid V_k)=\sum_{f\in V_k}a_d(f)\,a_{d'}(f). At level 163163 the Atkin–Lehner signs are w163=(+1,+1,1)w_{163}=(+1,+1,-1) on (V1,V5,V7)(V_1,V_5,V_7), equivalently U163=(1,1,+1).U_{163}=(-1,-1,+1).

The inequality of Theorem [thm:main] holds for the four pairs (3,19),(7,43),(11,43),(11,67).(3,19),\ (7,43),\ (11,43),\ (11,67).

Proof. For each orbit one has orbitwise Cauchy–Schwarz |Ck(d,d)|Ck(d,d)Ck(d,d).|C_k(d,d')|\le \sqrt{C_k(d,d)\,C_k(d',d')}. Using the exact orbit decompositions from the modular-symbol computation gives the bounds in Table 2. In each case the resulting orbitwise bound is below the required bound B(d,d)B(d,d'). ◻

The four pairs proved from the 1+5+71+5+7 orbit decomposition.
pair exact C1+C5+C7C_1+C_5+C_7 orbitwise bound required bound B(d,d)B(d,d')
(3,19)(3,19) 0+0+200+0+20 50.0750.07 9292
(7,43)(7,43) 14+102814+10-28 46.4046.40 184184
(11,43)(11,43) 42+147-42+1-47 257.85257.85 272272
(11,67)(11,67) 12+96+1212+96+12 253.80253.80 272272

Three residual pairs by exact finite sums

The inequality of Theorem [thm:main] holds for the three residual pairs (3,43),(3,67),(7,67).(3,43),\ (3,67),\ (7,67).

Proof. These are the indices n=129=343n=129=3\cdot 43, n=201=367n=201=3\cdot 67, and n=469=767n=469=7\cdot 67. The divisor sums are a129(a,163)=1a=176,a201(a,163)=1a=272,a469(a,163)=1a=544.\textstyle\sum_{\substack{a\mid 129\\(a,163)=1}} a=176,\qquad \sum_{\substack{a\mid 201\\(a,163)=1}} a=272,\qquad \sum_{\substack{a\mid 469\\(a,163)=1}} a=544. For each nn, the Eichler–Selberg class-number sum runs over those integers tt with t2<4nt^2<4n for which (4nt2163)1\bigl(\frac{4n-t^2}{163}\bigr)\ne -1. The individual terms (1+(Δ163))H(Δ)\bigl(1+\bigl(\frac{\Delta}{163}\bigr)\bigr)H(\Delta) for the surviving tt-values are listed in Table 3. The exact evaluations are: Tr(T129)=12(24+32+72+32+16+20+32+20+16)176=44,Tr(T201)=12(24+40+16+40+64+16+40+72+32+24+16+20)272=70,Tr(T469)=12(32+20+56+80+48+128+40+32+48+96+16+96+24+24+40+72+32+20+32+32)544=60.\begin{aligned} \operatorname{Tr}(T_{129}) &= \tfrac12(24+32+72+32+16+20+32+20+16)-176=-44,\\ \operatorname{Tr}(T_{201}) &= \tfrac12(24+40+16+40+64+16+40+72+32+24+16+20)-272=-70,\\ \operatorname{Tr}(T_{469}) &= \tfrac12(32+20+56+80+48+128+40+32+48+96\\ &\hspace{4em}{}+16+96+24+24+40+72+32+20+32+32)-544=-60. \end{aligned} These satisfy |44|<92,|70|<92,|60|<184,|-44|<92,\qquad |-70|<92,\qquad |-60|<184, which are exactly the required bounds B(3,43)=92B(3,43)=92, B(3,67)=92B(3,67)=92, B(7,67)=184B(7,67)=184. ◻

Surviving Eichler–Selberg terms for the three residual indices. For each nn, the surviving tt-values are those with t2<4nt^2<4n and (4nt2163)1\bigl(\frac{4n-t^2}{163}\bigr)\ne -1. Each |t|>0|t|>0 contributes twice (from ±t\pm t). The column headers show |t||t| only; the half-sum accounts for multiplicities.
nn surviving |t||t|-values 12(1+(Δ163))H(Δ)\frac12\sum(1{+}(\frac{\Delta}{163}))H(\Delta)
129129 1,8,10,11,12,13,16,17,191,8,10,11,12,13,16,17,19 132132
201201 0,1,3,4,8,9,13,14,15,17,24,250,1,3,4,8,9,13,14,15,17,24,25 202202
469469 0,3,5,6,7,10,13,15,18,20,0,3,5,6,7,10,13,15,18,20, 544544
21,22,23,27,29,30,31,33,36,4021,22,23,27,29,30,31,33,36,40

Proof of Theorem [thm:main]. Combine Lemmas [lem:cs], [lem:vacuum], [lem:block], and [lem:residual]. ◻

Enlargement Data

The trace inequality is specific to the Heegner set SS and does not persist under arbitrary enlargement.

At level 163163:

  1. adjoining the individual primes 17,53,71,89,9717,53,71,89,97 to SS already produces violations of the trace inequality;

  2. adjoining all primes 97\le 97 together with 163163 produces eight violations, the worst being |G(53,163)|=43>26.|G(53,163)|=43>26.

Proof. Direct computation gives the following individual failures: |G(17,163)|=34>26,|G(53,163)|=43>26,|G(71,3)|=100>92,|G(71,163)|=29>26,|G(89,163)|=32>26,|G(97,3)|=106>92.\begin{aligned} |G(17,163)|&=34>26,\\ |G(53,163)|&=43>26,\\ |G(71,3)|&=100>92,\qquad |G(71,163)|=29>26,\\ |G(89,163)|&=32>26,\\ |G(97,3)|&=106>92. \end{aligned} For the full enlarged set of primes 97\le 97 together with 163163, the computational scan records eight violations, with worst pair (53,163)(53,163) of ratio 43/261.65443/26\approx 1.654. ◻

The saved scan data do not support the stronger statement that SS is inclusion-maximal among all subsets of primes 97\le 97. For example, the one-prime extensions by 2,5,13,23,29,31,37,41,47,59,61,73,79,832,\ 5,\ 13,\ 23,\ 29,\ 31,\ 37,\ 41,\ 47,\ 59,\ 61,\ 73,\ 79,\ 83 still satisfy the trace inequality.

Remarks

By Proposition [prop:vacuumdiag], 1632G(163,163)=16326=137.163-2G(163,163)=163-26=137. We record this as an arithmetic coincidence: 1632g=137.163-2g=137.

The inequality of Theorem [thm:main] is not generic for Hecke operators. For example, at level 163163 one has |G(53,163)|=43>26=2min(G(53,53),G(163,163)),|G(53,163)|=43>26=2\min(G(53,53),G(163,163)), so the Heegner-set phenomenon fails even at the same level once the operator set is enlarged.

The proof given here is exact, but it does not explain the inequality conceptually. The vacuum row is controlled by a local factor in the prime-level trace formula, whereas the off-vacuum residual cases still require explicit whole-sum cancellation. A uniform proof remains open.

Acknowledgements

Computations were carried out in SageMath, NumPy, and SciPy.

9

A. Baker, Linear forms in the logarithms of algebraic numbers, Mathematika 13 (1966), 204–216.

F. Diamond and J. Shurman, A First Course in Modular Forms, Springer, 2005.

K. Heegner, Diophantische Analysis und Modulfunktionen, Math. Z. 56 (1952), 227–253.

T. Miyake, Modular Forms, Springer, 1989.

The Sage Developers, SageMath, the Sage Mathematics Software System, https://www.sagemath.org.

H. M. Stark, A complete determination of the complex quadratic fields of class-number one, Michigan Math. J. 14 (1967), 1–27.