Answers are grounded in this paper and cite their sources.
For primes \(d≡1 4\) let \(θχ_d(u)=∑n≥1χ_d(n)e-π n^2u/d\) and \(L_d(t)=logθχ_d(e^t)\). We study the set \(\) of those \(d\) for which \(θχ_d\) is log-concave on \((0,∞)\), equivalently \(L_d''≤0\). We prove two elementary universal facts. First, \(L_d'(0)=-14\) for every \(d\). Second, \(L_d''(log d)=-π+3π(3π-1)χ_d(2)e-3π+O(10⁻⁶)\), and consequently \(L_d''(t)≤-3.13\) for every \(t≥log d\). Together with the functional equation these reduce the verification of log-concavity to the finite window \(t∈[0,log d)\). We also prove that \(L_d'''<0\) on the whole half-axis \(t≥log d\), so that the functional \(F=tL_d''-L_d'\) is strictly decreasing there; for \(d=5\) this combines with an interval certificate on the finite window to give \(F<14\) for all \(t>0\). Numerically, among the \(39\,175\) primes \(d≡14\) with \(d≤10^6\), our computation finds \(777\) members of \(\), and the counting function grows like \(X0.616\) over three decades; we conjecture \(#(X) Xα\) with \(α≈0.62\), in particular that \(\) is infinite. The largest concavity margin occurs at \(d=5\), where \(max_tL_5''=-1.2225\), that is \(3.385\) times the second largest value (at \(d=17\)). Failures split into three numerically stable classes. Since certified inclusions are presently out of reach, the counts are upper bounds; we state precisely which assertions are proved, which are certified by interval arithmetic, and which are numerical.
No takes yet. Share an insight, caveat, or question.
Xia et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: