12

I wonder why this is true

$$ \sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{m^2 + 58 n^2} = - \frac{\pi \ln( 27 + 5 \sqrt {29})}{\sqrt {58}} $$

Where the sum omits the case $n = m = 0$ ofcourse.

mick
  • 15,946
  • Not sure If related but is N(sqrt(58)) a ufd ? I think so. – mick Oct 12 '17 at 23:17
  • 1
    For sure, $h(-232)=2$ (see http://mathworld.wolfram.com/ClassNumber.html) hence there are just two binary reduced quadratic forms of discriminant $-232$, namely $m^2+58n^2$ and $2m^2+29n^2$, so the problem boils down to computing the value of a Dirichlet L-function. – Jack D'Aurizio Oct 12 '17 at 23:53
  • Im not into Class Numbers , L-functions etc :( A pure calculus explaination ?? Btw where does - 232 come from ?? – mick Oct 13 '17 at 00:43
  • 1
    The discriminant of $a^2+58b^2$ is $-4\cdot 58=-232$. Similar results do not hold if $58$ is replaced by a random integer, such series encode the arithmetic structure of certain quadratic fields in a very strict way. – Jack D'Aurizio Oct 13 '17 at 00:46
  • I only know the discriminant for univariate polynomials ... – mick Oct 13 '17 at 00:54
  • 1
    But a homogeneous quadratic polynomial is not that different. The discriminant of $Ax^2+Bxy+Cy^2$ is defined as $B^2-4AC$ as expected. Would you mind mentioning why you are digging into the theory of integer quadratic forms (and beyond) and your actual attempts? – Jack D'Aurizio Oct 13 '17 at 00:56
  • 1
    I was Just amazed and inspired. – mick Oct 13 '17 at 01:02
  • @mick: MathJax hint: to get multicharacter things under the square root sign, put them in braces. For example, \sqrt{58} gives $\sqrt {58}$ compared with \sqrt 58 which gives $\sqrt 58$. The Chaz fixed the body but not the title. It works everywhere-exponents, subscripts, fractions, etc. – Ross Millikan Oct 13 '17 at 04:59
  • @JackD'Aurizio: The integer $10$ works as well. The family $d=5,,13,,37$ probably involves a cross term. Kindly see comment/answer below. – Tito Piezas III Oct 13 '17 at 07:25
  • It is an irony that Ramanujan never went after these sums even after being an expert in elliptic function theory. Had he been there algebraic number theory would not have been so unnecessarily difficult topic to handle. The sum can be evaluated easily via Kronecker second limit formula. – Paramanand Singh Oct 13 '17 at 07:39
  • @ParamanandSingh: Can you edit my answer and add the closed-forms for $$\sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{2m^2 + p,n^2}=;\color{red}?$$ for $p=3,5,11,29$? Thanks. – Tito Piezas III Oct 13 '17 at 07:54
  • @TitoPiezasIII: I have not evaluated them in closed form. Perhaps it will require some laborious calculations related to eta functions. – Paramanand Singh Oct 13 '17 at 10:35
  • 1
    Using Kronecker limit formula (see previous comment) we get the general formula $$\sum_{m, n\in \mathbb {Z}}'\frac{(-1)^{m}}{m^{2}+pn^{2}}=-\frac{\pi\log(2g_{p} ^{4})}{\sqrt{p}}$$ where $p>0$ and $g_{p} $ denotes one of Ramanujan's class invariants. If $p$ is rational then $g_{p} $ is algebraic. – Paramanand Singh Oct 13 '17 at 12:14
  • @JackD'Aurizio: there is a closed form available for many values of $p$ as my last comment suggests. Although I must say getting the algebraic number $g_{p} $ in explicit form for any arbitrary positive rational $p$ is difficult. – Paramanand Singh Oct 13 '17 at 12:19
  • @ParamanandSingh: I added an update to express the Ramanujan class invariant $g_p$ into a more familiar eta quotient. – Tito Piezas III Oct 13 '17 at 18:04
  • @TitoPiezasIII Please try to give all the clues you know in your answers (see mine, I took 1 hour to write it) – reuns Oct 14 '17 at 01:33
  • @reuns: the formula I gave in my comment has been used by Tito to give the evaluations of the sums. If one accepts the Kronecker (second) limit formula, then one immediately gets the answer in terms of Ramanujan's class invariants. Kronecker limit formula does not appear to have an easy and accessible proof (at least my online searches have been totally ineffective in finding easy proofs of any result with Kronecker's name in it, say Kronecker Weber Theorem). – Paramanand Singh Oct 14 '17 at 03:37
  • @reuns : Ramanujan defines two functions of a real variable $q$ via infinite products $$G(q) =2^{-1/4}q^{-1/24}\prod_{n=1}^{\infty}(1+q^{2n-1}),g(q)=2^{-1/4}q^{-1/24}\prod_{n=1}^{\infty} (1-q^{2n-1}) $$ and proved that these functions take algebraic values if $q=\exp(-\pi\sqrt{p}) $ if $p$ is a positive rational. The corresponding values of the function are denoted by $G_{p}, g_{p} $ and called class invariant (the term is borrowed from Weber). He also gave a method to evaluate these algebraic numbers and worked out details for small values of $p$. – Paramanand Singh Oct 14 '17 at 03:50
  • @reuns : Ramanujan was not so much of an abstract theorist. When he said that a certain quantity can be evaluated in closed form, he meant that he had performed the calculations and had a general procedure to do these evaluations. So much better than the modular form guys / cft guys who are trying to guess algebraic numbers based on the degree of polynomials satisfied by them. – Paramanand Singh Oct 14 '17 at 03:54
  • @reuns : you may have a look at http://paramanands.blogspot.com/2012/03/ramanujans-class-invariants.html where some class invariants have been evaluated in a manner specified by Ramanujan. – Paramanand Singh Oct 14 '17 at 03:58
  • @ParamanandSingh It doesn't seem very different to evaluating $j(a+b\sqrt{-d})$, which is the same as finding the Hilbert class field of $\mathbb{Q}(\sqrt{-d})$ or finding an elliptic curve with complex multiplication by $\mathbb{Z}[\sqrt{-d}]$. In general, once we know $j(a+b\sqrt{-d})$ and if $\mathbb{Q}(\sqrt{-d},j(a+b\sqrt{-d}))/\mathbb{Q}$ is abelian then we can evaluate $\sum_{n,m} \frac{(-1)^n}{n^2+d m^2}$ in term of Dirichlet L-functions as in my answer. – reuns Oct 14 '17 at 04:03
  • @reuns : that's what Weber did. Unfortunately his work is also lost in translation (https://math.stackexchange.com/q/438643/72031) and modern mathematicians tried best to make the work of guys like Kronecker / Weber inaccessible. – Paramanand Singh Oct 14 '17 at 04:04
  • @reuns: Hi reuns. I believe Paramanand addressed your query. The case $p=10$ was inspired guesswork, but I used his closed-form and Mathematica's integer-relations to find $p=6,22$. – Tito Piezas III Oct 14 '17 at 12:33

3 Answers3

6

Finally I managed to sum this series using Ramanujan's class invariants. We have the definition $$g(q) = 2^{-1/4}q^{-1/24}\prod_{n = 1}^{\infty}(1 - q^{2n - 1}), \, g_{p} = g(e^{-\pi\sqrt{p}})\tag{1}$$ Ramanujan established that if $p$ is a positive rational number then $g_{p}$ is an algebraic real number. Moreover we have the non-trivial identity $$g_{4/p} = 1/g_{p}\tag{2}$$ Another ingredient we need is the formula $$\sin \pi z = \pi z\prod_{n = 1}^{\infty}\left(1 - \frac{z^{2}}{n^{2}}\right)\tag{3}$$ Taking logs and differentiating with respect to $z$ we get $$\pi\cot\pi z = \frac{1}{z} -2z \sum_{n = 1}^{\infty}\frac{1}{n^{2} - z^{2}}$$ Replacing $z$ by $iz$ we get $$\pi\coth\pi z = \frac{1}{z} + 2z\sum_{n = 1}^{\infty}\frac{1}{n^{2} + z^{2}}$$ The above sum can be written as $$\sum_{n = 1}^{\infty}\frac{1}{n^{2} + z^{2}} = \frac{\pi}{2z} - \frac{1}{2z^{2}} + \frac{\pi e^{-2\pi z}}{z(1 - e^{-2\pi z})}\tag{4}$$ Consider the sum in question \begin{align} S(p) &= \sum_{m, n\in\mathbb{Z}}'\frac{(-1)^{m}}{m^{2} + pn^{2}}\notag\\ &= 2\sum_{m = 1}^{\infty}\frac{(-1)^{m}}{m^{2}} + 2\sum_{n = 1}^{\infty}\sum_{m\in\mathbb{Z}}\frac{(-1)^{m}}{m^{2} + pn^{2}}\notag\\ &= 2\sum_{m = 1}^{\infty}\frac{(-1)^{m}}{m^{2}} + \frac{2}{p}\sum_{n = 1}^{\infty}\frac{1}{n^{2}}+ 4\sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}\frac{(-1)^{m}}{m^{2} + pn^{2}}\notag\\ &=\frac{(2- p)\pi^{2}}{6p} + 4\sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}\frac{(-1)^{m}}{m^{2} + pn^{2}}\notag\\ &=\frac{(2- p)\pi^{2}}{6p} + 4\sum_{m = 1}^{\infty}(-1)^{m}\sum_{n = 1}^{\infty}\frac{1}{m^{2} + pn^{2}}\notag\\ &= \frac{(2- p)\pi^{2}}{6p} + \frac{4}{p}\sum_{m = 1}^{\infty}(-1)^{m}\sum_{n = 1}^{\infty}\frac{1}{m^{2}(1/p) + n^{2}}\notag\\ &= \frac{(2- p)\pi^{2}}{6p} + \frac{4}{p}\sum_{m = 1}^{\infty}(-1)^{m}\left(\frac{\pi\sqrt{p}}{2m} - \frac{p}{2m^{2}} + \frac{\pi\sqrt{p} e^{-2\pi m/\sqrt{p}}}{m(1 - e^{-2\pi m/\sqrt{p}})}\right)\text{ (using equation }(4)) \notag\\ &=\frac{\pi^{2}}{3p}-\frac{2\pi\log 2}{\sqrt{p}}+\frac{4\pi}{\sqrt{p}}\sum_{m=1}^{\infty} \frac{(-1)^{m}q^{m}}{m(1-q^{m})},\,q=e^{-2\pi/\sqrt{p}}\notag\\ &=\frac{\pi^{2}}{3p}-\frac{\pi\log 4}{\sqrt{p}}+\frac{4\pi}{\sqrt{p}}(a(q^{2})-a(q))\notag\\ &\, \, \, \, \,\,\,\,\text{where }a(q) =\sum_{n=1}^{\infty} \frac{q^{n}} {n(1-q^{n})}=-\log\prod_{n=1}^{\infty}(1-q^{n})\notag\\ &=\frac{\pi^{2}}{3p}-\frac{\pi\log 4} {\sqrt{p}}+\frac{4\pi}{\sqrt{p}}\log\prod_{n=1}^{\infty}(1-q^{2n-1}) \notag\\ &=\frac{\pi^{2}}{3p}-\frac{\pi}{\sqrt {p}} \log\frac{2}{q^{1/6}g^{4}(q)}\text{ (via equation }(1)) \notag\\ &= -\frac{\pi\log(2/g_{4/p}^{4})}{\sqrt{p}}\notag\\ &= -\frac{\pi\log (2g_{p}^{4})}{\sqrt{p}}\text{ (using equation }(2))\notag \end{align} It is well known that $g_{58} = \sqrt{(5 + \sqrt{29})/2}$ and this gives the desired closed form for $S(58)$. The above technique can also be used (with some more effort) to prove the Kronecker's second limit formula.


The function $a(q) $ used above is related to work of Simon Plouffe. See this answer for details.

  • nice answer (+1)...i was fiddeling around with dedekind etas but gave up. they seem to be quiet closely related to the class invariants you are using.. – tired Oct 21 '17 at 09:40
  • 1
    @tired: oh yes the class invariants can be expressed in terms of eta functions. For the current case just note that $\prod(1-q^{2n-1})=\prod(1-q^{n})/(1-q^{2n})$ so that it is essentially $\eta(q) /\eta(q^{2}) $. – Paramanand Singh Oct 21 '17 at 10:30
5
  • Let $F=\mathbb{Q}(\sqrt{-58}),\mathcal{O}_F=\mathbb{Z}[\sqrt{-58}]$. Its ideal class group is $C_F= \{ (1),(2,\sqrt{-58})\}$

    thus the ideals with their norm are $N((n+\sqrt{-58}m))= n^2+58m^2$, $ N(\frac{2n+\sqrt{-58}m}{2}(2,\sqrt{-58}))= \frac12(4n^2+58m^2)$

    then your series is $\ \ 2 \ L(1,\psi)$

    where $\psi((n+\sqrt{-58}m))=1, \psi(\frac{2n+\sqrt{-58}m}{2}(2,\sqrt{-58}))=-1$ is the Hecke character induced by the non-trivial character of $C_F$.
    $$ L(s,\psi) = \sum_I \psi(I) N^{-s} =\frac{1}{|\mathcal{O}_F^\times|} \sum_{n,m \in \mathbb{Z}^2}' N((n+\sqrt{-58}m))^{-s}- 2^{s} N((2n+\sqrt{-58}m))^{-s}$$

  • Using $|C_F| = 2$ and some class field theory, we find that $H = F(\sqrt{-2})$ is the Hilbert class field of $F$ and $$\zeta_H(s) = \zeta_F(s) L(s,\psi)$$

    Now by chance it happens that $H/\mathbb{Q}$ is itself an abelian extension. Thus we can write $\zeta_H$ as a product of Dirichlet L-functions

    $$\zeta_H(s) = \zeta(s)\ L(s,{\scriptstyle \left(\frac{-58}{.}\right)})\ L(s,{\scriptstyle \left(\frac{-2}{.}\right)})\ L(s,{\scriptstyle \left(\frac{29}{.}\right)})$$ Together with $\displaystyle\zeta_F(s) = \zeta(s)\ L(s,{\scriptstyle \left(\frac{-58}{.}\right)})$ it means $$L(1,\psi) = L(1,{\scriptstyle \left(\frac{-2}{.}\right)})\ L(1,{\scriptstyle \left(\frac{29}{.}\right)})$$ and we use quadratic reciprocity to write $(\frac{-d}{.}) = (\frac{.}{\Delta})$ and conclude.

reuns
  • 77,999
  • 1
    Funny to see a rather innocuous looking sum solved using Hecke Characters and Class Field Theory. +1 –  Oct 13 '17 at 04:45
3

(Too long for a comment.)

We have,

$$ \sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{m^2 + 10n^2} = - \frac{2\pi \ln( \sqrt2\; U_{5})}{\sqrt {10}} $$

$$ \sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{m^2 + 58 n^2} = - \frac{2\pi \ln( \sqrt2\; U_{29})}{\sqrt {58}} $$

with fundamental units $U_5 = \frac{1+\sqrt5}2$ and $U_{29} = \frac{5+\sqrt{29}}2$.

P.S. Presumably there might be a family for $d = 5,\,13,\,37$.


Added:

Courtesy of a comment by Paramanand Singh, we have the closed-form in terms of the Dedekind eta function $\eta(\tau)$ as,

$$\sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{m^2 + pn^2} = - \frac{2\pi \ln(\sqrt2\,g_p^2)}{\sqrt {p}} =- \frac{\pi \ln(2\,g_p^4)}{\sqrt {p}} $$ where,

$$g_p = 2^{-1/4}\frac{\eta(\tfrac12\sqrt{-p})}{\eta(\sqrt{-p})}$$

In particular, $$\sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{m^2 + 6n^2} = - \frac{2\pi \ln\big(\sqrt2\,(1+\sqrt2)^{1/3}\big)}{\sqrt{6}}$$ $$\sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{m^2 + 22n^2} = - \frac{2\pi \ln\big(\sqrt2\,(1+\sqrt2)\big)}{\sqrt{22}}$$ and more complicated ones for $d=5,\,13,\,37$.

  • Glad to see your update. When there are too my comments people may not read all of them. Putting that comment as a part of answer (with additional details) will make it more accessible. I had already delivered my +1 so just thanks for now. – Paramanand Singh Oct 13 '17 at 19:19
  • @ParamanandSingh: I upvoted your comment with the formula so it goes into the top 5. – Tito Piezas III Oct 14 '17 at 12:51
  • @ParamanandSingh: I believe there should be a similar formula for $$\sum_{m,n = - \infty}^{\infty} \frac{(-1)^m}{am^2 +amn+p n^2}$$ especially for $a=1$ and $a=2$. – Tito Piezas III Oct 14 '17 at 15:21
  • I am not sure, perhaps some algebraic number theory expert can help. You may ask it as a separate question. It might be possible to transform $am^{2}+amn+pn^{2} $ into $rM^{2}+sN^{2}$ where $M, N$ are linear combinations of $m, n$ in such a manner that if $m, n$ take all integer values then $M, N$ also take all integer values. This will reduce the problem to current one. – Paramanand Singh Oct 14 '17 at 15:33
  • @ParamanandSingh: I've asked the question in this post. – Tito Piezas III Oct 18 '17 at 16:26