Generalized constraint identities for odd divisor functions

From papers

Let aa be a positive odd integer, let k0k\geqslant0 be an integer, let g(s)=πs/2(s1)Γ(s/2+1)g(s)=\pi^{-s/2}(s-1)\Gamma(s/2+1), and let σa(n)\sigma_a(n) be the generalized divisor function. Let the contour Re(s)=σ\operatorname{Re}(s)=\sigma lie to the right of all poles.

General constraint conjecture.

n=1σa(n)Re(s)=σ[g(s)g(sa)ns(s)k+g(s)g(s+a)ns+a(s1)k]ds=0.\sum_{n=1}^{\infty}\sigma_a(n)\int_{\operatorname{Re}(s)=\sigma}\left[-\frac{g(s)g(s-a)}{n^s}(-s)^k+\frac{g(s)g(s+a)}{n^{s+a}}(s-1)^k\right]ds=0.

The corresponding identities are established for the cases treated in the paper, while this formulation is presented as a conjectural generalization beyond those proven cases.

Progress summary

Open

The full identity remains conjectural, although matching identities are proved in the cases a=1a=1, a=3a=3, and a=5a=5.

The conjecture asserts a contour-integral identity for every positive odd integer aa and every integer k0k\geqslant0. It appears as Conjecture 2 in the 2023 paper The Riemann zeta function and exact exponential sum identities of divisor functions, which presents the general statement beyond its proved cases.

Known results

  • Theorem 3 proves the corresponding identities for a=1a=1, a=3a=3, and a=5a=5, for every integer k0k\geqslant0.
  • The paper notes that analogous relations can be derived for other odd values of aa, but for odd a>5a>5 the associated exponential-sum functions become rational rather than polynomial.

Current status (as of August 2026): The cases a=1a=1, a=3a=3, and a=5a=5 are settled, while the stated identity for general positive odd aa remains an open conjecture with no independently reported proof or counterexample.

Sources
Sources & referencesView supporting material

Primary source

Maria Nastasescu, Nicolas Robles, Bogdan Stoica and Alexandru Zaharescu, “The Riemann zeta function and exact exponential sum identities of divisor functions”, arXiv:2311.07657 (2023).

Solutions 1

Proof

In fact, the constraint follows directly from the functional equation of the completed Riemann zeta function; the restriction that aa be odd is unnecessary.

Put

g(s)=πs/2(s1)Γ ⁣(s2+1),ξ(s)=g(s)ζ(s).g(s)=\pi^{-s/2}(s-1)\Gamma\!\left(\frac{s}{2}+1\right), \qquad \xi(s)=g(s)\zeta(s).

Then

ξ(s)=12s(s1)πs/2Γ ⁣(s2)ζ(s)\xi(s)=\frac12s(s-1)\pi^{-s/2} \Gamma\!\left(\frac s2\right)\zeta(s)

is entire and satisfies

ξ(1s)=ξ(s).\xi(1-s)=\xi(s).

Initially choose σ>1+max{0,Rea}\sigma>1+\max\{0,\operatorname{Re}a\}. The absolutely convergent Dirichlet series

n1σa(n)ns=ζ(s)ζ(sa),n1σa(n)ns+a=ζ(s+a)ζ(s)\sum_{n\ge1}\frac{\sigma_a(n)}{n^s} =\zeta(s)\zeta(s-a), \qquad \sum_{n\ge1}\frac{\sigma_a(n)}{n^{s+a}} =\zeta(s+a)\zeta(s)

and the exponential decay of the gamma factors justify interchanging summation and integration. The proposed left-hand side therefore becomes

Res=σξ(s)ξ(sa)(s)kds+Res=σξ(s)ξ(s+a)(s1)kds.-\int_{\operatorname{Re}s=\sigma} \xi(s)\xi(s-a)(-s)^k\,ds + \int_{\operatorname{Re}s=\sigma} \xi(s)\xi(s+a)(s-1)^k\,ds.

Define the entire function

F(s)=ξ(s)ξ(s+a)(s1)k.F(s)=\xi(s)\xi(s+a)(s-1)^k.

By the functional equation,

F(1s)=ξ(s)ξ(sa)(s)k.F(1-s)=\xi(s)\xi(s-a)(-s)^k.

The substitution w=1sw=1-s, with the induced reversal of the vertical path canceled by ds=dwds=-dw, gives

Res=σF(1s)ds=Rew=1σF(w)dw.\int_{\operatorname{Re}s=\sigma}F(1-s)\,ds = \int_{\operatorname{Re}w=1-\sigma}F(w)\,dw.

Since FF is entire and decays exponentially along horizontal edges in every fixed vertical strip, Cauchy's theorem yields

Rew=1σF(w)dw=Res=σF(s)ds.\int_{\operatorname{Re}w=1-\sigma}F(w)\,dw = \int_{\operatorname{Re}s=\sigma}F(s)\,ds.

The two terms consequently cancel, proving the asserted identity for every integer k0k\ge0.

For any other contour lying to the right of the poles in the original statement, shift each individual summand's gamma-product contour to a line satisfying σ>1+max{0,Rea}\sigma>1+\max\{0,\operatorname{Re}a\}. No pole is crossed, and the horizontal integrals vanish by Stirling's estimate. The individual integrals are therefore unchanged, so the same identity holds on every admissible contour. In particular, it holds for every positive odd integer aa as conjectured.

0 endorsements
Shivam Patel ·