We construct an explicit dual-feasible point for the Cohn–Elkies linear program in dimension 36, built from the space of weight-18 modular forms for \Gamma_0(24) following the method of Cohn and Triantafillou. The certificate shows that the two-point linear programming bound on the sphere packing density in dimension 36 exceeds the density of the best packing currently known – the Kschischang–Pasupathy packing, of center density 2^{18}/3^{10} – by a factor of at least 32.91. In particular, no Cohn–Elkies auxiliary function can certify the best known packing in dimension 36 as optimal. To our knowledge this is the first such dual bound in any dimension above 32, extending the table of Cohn–Triantafillou (d=12,16,20,28,32), Li (3\le d\le 13), and de Courcy-Ireland–Dostert–Viazovska (d=6). The certificate is rigorous and machine-checkable, with exact rational data and certified interval bounds: the dual point is a rational vector, coefficient nonnegativity is verified by exact arithmetic up to n=800, and eventual positivity of the two relevant q-expansions is proved via an explicit Deligne-type tail bound whose constant is certified with outward-rounded interval arithmetic. Two methodological points may be of independent interest: a constraint-generation (cutting-plane) formulation of the exact rational LP, which is what makes an exact vertex whose Eisenstein data supports the tail argument reachable; and a sharpened, lift-aware form of the Deligne bookkeeping constant that discounts deep oldform lifts, without which the finite verification in dimension 36 fails (the crossover moves past the verified window).
\delta_{\mathrm{LP}}(36)\;\ge\;B\;=\;32.91044\ldots\times \dfrac{2^{18}}{3^{10}}\;>\;146.1036, where B is the explicit rational number given in (?), with outward decimal enclosures B\in[146.1036734821,\,146.1036734822] and B/(2^{18}/3^{10})\in[32.91044546,\,32.91044547]. In particular the Cohn–Elkies linear programming bound in dimension 36 cannot certify the Kschischang–Pasupathy packing [KP], of center density 2^{18}/3^{10}=4.4394\ldots, as an optimal sphere packing.
The certificate data — the explicit list of the 29 basis forms, the 20 exact rational coordinates of the vertex in the (pivot-ordered) reduced basis, the coordinate-to-form map, the exact rational b_0 and B, the exact Eisenstein coefficients, and all constants below — are in the ancillary file ancillary/certificate_exact_data.txt; the construction and verification receipts are ancillary/receipt_d36_cutgen.txt and ancillary/receipt_d36_cs.txt, and the single-driver verification is ancillary/code/verify_certificate.py (§?).
Let S=\sum_{f,e}\lambda_{f,e}\,f(ez)\in S_k(\Gamma_0(N)) with f normalized newforms (each Galois orbit expanded into its individual complex embeddings, so that every |\lambda_{f,e}| is the modulus of a single complex coordinate). Then for all n\ge1,
By the divisor bound |a_f(m)|\le\sigma_0(m)m^{(k-1)/2}, valid for every m\ge1 and each normalized newform — at the unramified primes p\nmid M, Deligne’s estimate and the Hecke recursion give |a_f(p^r)|\le(r+1)\,p^{r(k-1)/2}; at the ramified primes p\mid M we distinguish two cases. If p\,\|\,M (exactly divides the level), the local newform bound |a_f(p)|\le p^{(k-2)/2} with a_f(p^r)=a_f(p)^r (the local Euler factor at a prime exactly dividing the level, for a newform of trivial character; see e.g. [DiamondShurman, §5.8]) gives |a_f(p^r)|\le p^{r(k-2)/2}\le\sigma_0(p^r)\,p^{r(k-1)/2}; if p^2\mid M, then (trivial character) a_f(p)=0, so a_f(p^r)=0 and the bound holds a fortiori. Both cases occur here, as the newform levels M\mid 24 include M\in\{4,8,12,24\} with 2^2\mid M. The bound for general m follows by multiplicativity. The coefficient of q^n in f(ez) is a_f(n/e) if e\mid n and 0 otherwise, and for e\mid n, \sigma_0(n/e)\le\sigma_0(n) and (n/e)^{(k-1)/2}=n^{(k-1)/2}e^{-(k-1)/2}. Summing over the pairs (f,e) with e\mid n gives the claim.
The text on this page is from arXiv:2607.11319v2 by Rifat Jumagulov, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2607/2607.11319.jsonl.