WorksMathematics
Failure of the Li–Liu spectral hypothesis R under Dirichlet GRH
Abstract
Under Dirichlet GRH, the Li–Liu spectral hypothesis R fails. For every fixed and , each sufficiently large interval contains at least even integers for which . We construct these integers unconditionally and use GRH only at the modulus to obtain the norm bound. Each has at least representations as a sum of two distinct primes. The proof isolates a matrix entry in which the continuous model cancels and the two arithmetic terms have different Buchstab densities.
Introduction
Li and Liu [5] attach a compact self-adjoint operator to each even satisfying , the Riemann hypothesis for the primitive Dirichlet -functions of conductor dividing , including the Riemann zeta function. Their hypothesis R places every zero of an auxiliary spectral product on the line for every sufficiently large such . By their Theorem 6.1, this is equivalent at to
Theorem 1.1. Fix and . There exists such that, for every sufficiently large , at least even integers have both of the following properties.
has at least distinct representations by primes with .
If holds, then .
Since , taking contradicts (1) under global Dirichlet GRH. The first conclusion holds unconditionally on the same set of moduli.
Maier’s matrix method detects fluctuations of primes in short intervals [6]. Granville and Soundararajan developed the connection between intervals and progressions for general arithmetic sequences [3], and Pandey and Woo obtained related results for systems of prime linear forms [7]. Here the shifts and occur in the same matrix entry. Their continuous contributions agree, while their progression averages sample Buchstab’s function at two consecutive arguments.
Two opposite shifts
Write for the von Mangoldt function. For a smooth real weight supported in , set
Lemma 2.1. Assume GRH(). Suppose and is supported in and in an interval of length . If , then
Proof. Put , . For write and . On the Hardy strip , let represent the functional
With arc length on the two boundary lines, the norm formula of [5 Prop. 7.4] is
Indeed, writing , the functional pairs with . Parseval on the two boundary lines gives (2) for these supports.
Take and fix equal to one on . We claim that
To evaluate the entry, apply the two functionals to the reproducing kernel of [5 Prop. 5.4]. The factors cancel its denominators. In the character term, move each contour to real part one and expand the logarithmic derivatives. The odd characters are nonprincipal, and GRH() makes the shift pole-free. Mellin inversion gives . Orthogonality on units gives the equal residue class with a plus sign and the opposite class with a minus sign. The supports select and , respectively, and exclude the diagonal.
For the continuous term extend by zero and put
The three beta integrals in the Li–Liu kernel give this expression under Mellin inversion [5 Lemma 7.2]. For example, the first integral of the powers is . Interchanging gives the second, and integration over $0<u<y$ gives the third. Also , since the first two terms partition the double integral of , which the third subtracts. The Mellin transform of therefore cancels the pole at of . Moving the contour to $\Re w=3/2 and expanding gives . The tests are smooth with compact support, so the Mellin integrals converge absolutely on the shifted lines. Only can contribute, and on the support of . Thus , proving (3).
Finally (2) gives . The support of has length at most , and there and . Hence . Apply Cauchy–Schwarz to (3). ▫
The progression average
Put , , , and . We use an even divisor of with two properties. First, for an absolute ,
@page 2/4 starts_mid=0 ends_mid=1 Put , , , and . We use an even divisor of with two properties. First, for an absolute ,
Second, for fixed , , and ,
where is Buchstab’s function.
To choose , use the exceptional prime of Ford, Green, Konyagin, Maynard and Tao [2 Cor. 6 and (7.2)]. It is a number equal to one or a prime . Take . Since , their estimate (7.2) applies to each conductor dividing . Summing its primitive character bounds over costs at most , by . Orthogonality and the ordinary prime number theorem give (4), absorbing this factor in the exponential error. Passing from a primitive character to its induced character removes only prime powers above primes dividing , of total weight .
Mertens’ theorem gives the product estimate in (6). For , the weighted estimate is partial summation from Buchstab’s asymptotic [1 (1.3)]. If a prime is removed, write . For , the extra integers up to number at most
For use Buchstab’s bound for , and sum the remaining geometric tail trivially. This proves (6).
Proof of Theorem 1.1. Choose a fixed so large that
Such occur arbitrarily far out. If eventually, the equation makes nonincreasing on a tail. It must then be constant there. The same equation propagates a constant tail backwards to , where .
Set and . Fix nonnegative of integral one. For each integer put
Summing the first term over separates the -sum from the sum over . The prime number theorem and (6) at give . For the second term put . Uniformly on its support ,
Indeed . Huxley’s prime number theorem in short intervals [4] gives uniformly for , : use his asymptotic for and the trivial bound below . Partial summation against the smooth weight gives the formula, since and the powers of two cost . There are admissible by (6). Summing over and applying (6) at gives the second term . Thus
Only values of occur, and odd contribute zero because . Fix an even with .
Let , so . For odd in the support of , (4) gives
In a nonunit class every nonzero term is a power of a single prime dividing . Terms removed by are powers of . Each costs in total. Since by the short-interval estimate, we obtain
Here . Pointwise , so for members of , with a fixed positive lower-bound constant.
Discard any such within of a proper prime power in . This removes integers. Since and , at least moduli remain. For each of them the sum with alone has weight and every positive term has prime. Proper prime powers contribute by (7). Each remaining pair contributes , so there are at least distinct prime pairs with smaller prime at most .
Finally uniformly in . At each of these moduli satisfying GRH, Lemma 2.1 and (7) give
Acknowledgment. The author used OpenAI’s GPT-6.2 Astra to formulate the results, develop and review the proofs, and draft the manuscript.
References
- [1]K. Fan, Numerically explicit estimates for the distribution of rough numbers, J. Number Theory 260 (2024), 120–150.arxiv.org/abs/2306.03347
- [2]K. Ford, B. Green, S. Konyagin, J. Maynard and T. Tao, Long gaps between primes, J. Amer. Math. Soc. 31 (2018), 65–105.arxiv.org/abs/1412.5029
- [3]A. Granville and K. Soundararajan, An uncertainty principle for arithmetic sequences, Ann. of Math. (2) 165 (2007), 593–635.DOI
- [4]M. N. Huxley, On the difference between consecutive primes, Invent. Math. 15 (1972), 164–170.DOI
- [5]J. Li and J. Liu, A conditional proof of the Goldbach conjecture, preprint, 2026.
- [6]H. Maier, Primes in short intervals, Michigan Math. J. 32 (1985), 221–225.DOI
- [7]M. Pandey and K. Woo, Small-scale distribution of linear patterns of primes, J. Lond. Math. Soc. (2) 110 (2024), e13001.arxiv.org/abs/2304.14267