2
$\begingroup$

I've been struggling a bit with a double sum that arose as the trace of an operator:

$$\sum_{(j,k)\in Z^2 \setminus (0,0)} \frac{(j+k)^n}{(j^2+k^2)^n},$$

where $n$ is an even natural number. Is there a closed form for this sum?

Even in the case of $n=2$, I'm a bit miffed. If there were no numerator, I would use the Mellin transform and rewrite the sum in terms of theta functions, which is a standard approach in lattice sums (cf. MR3135109). However, there is a numerator, and I can't seem to be clever enough to make some change of variables work well.

I am also interested in the corresponding $d$-dimensional case.

Thanks for any thoughts/suggestions.

$\endgroup$
4
  • 3
    $\begingroup$ For $n=2$ it diverges, does not it? $\endgroup$ Commented May 8, 2023 at 16:15
  • $\begingroup$ @FedorPetrov, yes that is possible that it diverges for $n=2$ (it's the boundary case), which argument do you have in mind? I'm also, interested in the cases $n=4,6,8,...$ which (I believe) converge. $\endgroup$ Commented May 8, 2023 at 22:18
  • $\begingroup$ For large $X$ consider $j, k$ between $X$ and $2 X$, they are of order $1/X^2$, so their sum is bounded from below $\endgroup$ Commented May 9, 2023 at 5:15
  • $\begingroup$ Righto. Any thoughts on the n=4 case? $\endgroup$ Commented May 11, 2023 at 16:25

1 Answer 1

4
$\begingroup$

The $n=4$ sum is accessible as follows. Expand $(j+k)^4$; by symmetry the odd terms $4j^3k$ and $4jk^3$ cancel out so we need only sum $(j^4 + 6j^2k^2 + k^4) / (j^2+k^2)^4$. You say you already know the sum of $1 / (j^2+k^2)^2$, which is $(j^4 + 2j^2k^2 + k^4) / (j^2+k^2)^4$ [see also the "P.S." paragraph below]; so we only need one more sum $\sum^\prime (j^4 + c j^2 k^2 + k^4) / (j^2+k^2)^4$. Well, we know the sum of $1 / (j+ik)^4$ from the theory of the Weierstrass $\wp$-function; if I did this right it comes to $\Gamma(1/4)^8 / (960 \pi^2) = 3.1512120+$. But $$ \frac1{(j+ik)^4} = \frac{(j-ik)^4}{j^2+k^2)^4} = \frac{(j^4 - 4i j^3 k - 6 j^2 k^2 + 4i j k^3 + k^4}{(j^2+k^2)^4}, $$ and the odd terms again cancel out, so this gives us the sum of $(j^4 - 6j^2k^2 + k^4) / (j^2+k^2)^4$ — and we're done because $$ 2 (j^4 + 6j^2k^2 + k^4) = 3 (j^2+k^2)^2 - (j^4 - 6j^2k^2 + k^4). $$

P.S. In general the sum of $1/(j^2+k^2)^n$ is $4 \zeta(n) L(n,\chi_4)$ [because it's 4 times the value at $n$ of the zeta function for ${\bf Q}(i)$]. Thus when $n=2$ that's $2\pi^2 {\bf G} / 3$ where ${\bf G}$ is Catalan's constant $$ L(2,\chi_4) = 1 - \frac1{3^2} + \frac1{5^2} - \frac1{7^2} + - \cdots = .915965594\ldots $$ so $2\pi^2 {\bf G} / 3 = 6.026812\ldots$; no further simplification is known or expected, though there are various ways known to compute the numerical value of ${\bf G}$ efficiently to any desired accuracy.

$\endgroup$
1
  • $\begingroup$ Nice, thank you. $\endgroup$ Commented May 12, 2023 at 13:59

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.