Variance conjecture for divisor sums in arithmetic progressions

At least 9 years old · documented by

Let k≥2k\geq 2 be fixed, and let

dk(n):=#{(a1,…,ak)∈Nk:a1⋯ak=n}d_k(n):=\#\{(a_1,\ldots,a_k)\in\mathbb{N}^k:a_1\cdots a_k=n\}

be the kk-fold divisor function. For integers q≥1q\geq 1 and X≥1X\geq 1, define

vk(q;X):=∑1≤a≤q(a,q)=1∣∑n≡a(modq) ≤Xdk(n)−1ϕ(q)∑(n,q)=1 ≤Xdk(n)∣2.v_k(q;X):=\sum_{\substack{1\leq a\leq q\\(a,q)=1}}\left|\sum_{\substack{n\equiv a\pmod q\ \leq X}}d_k(n)-\frac{1}{\phi(q)}\sum_{\substack{(n,q)=1\ \leq X}}d_k(n)\right|^2.

Also define

ak(q):=lim⁡s→1+(s−1)k2∑n≥1(n,q)=1dk(n)2ns,a_k(q):=\lim_{s\to1^+}(s-1)^{k^2}\sum_{\substack{n\geq1\\(n,q)=1}}\frac{d_k(n)^2}{n^s},

where γk(c)\gamma_k(c) is a piecewise polynomial of degree k2−1k^2-1.

Variance conjecture for divisor sums in arithmetic progressions. If X,q→∞X,q\to\infty in such a way that log⁡Xlog⁡q→c∈(0,k)\frac{\log X}{\log q}\to c\in(0,k), then

vk(q;X)∼ak(q)γk(c)X(log⁡q)k2−1.v_k(q;X)\sim a_k(q)\gamma_k(c)X(\log q)^{k^2-1}.

The function γk(c)\gamma_k(c) is positive for c∈(0,k)c\in(0,k) and is described more fully later in the source. This conjecture predicts the asymptotic variance of kk-fold divisor sums over reduced residue classes when both the modulus and the summation range grow. The range c<1c<1 corresponds to progressions containing at most one term, while values of cc just above 11 concern moduli close to, but smaller than, XX. The paper proves an averaged version in a restricted range; the full pointwise asymptotic stated here remains open.

References

Primary source

Brad Rodgers and Kannan Soundararajan, “The variance of divisor sums in arithmetic progressions”, arXiv:1610.06900 (2017).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.