1

$$\sum_{m \ge 1} \frac{(xy)^m}{(2m)!(1-y^m)}, \quad\text{where }x,y \in \mathbb N$$

I have, to start, J.Jacquelin's answer.

MarianD
  • 2,963
  • 1
    If you really want "partial sums" you can always do: take the infinite sum as f(x), then make f(xz) then the expansion of f(xz)/(1-z) in terms of z as an OGF gives the individual infinite sums as coefficients. Then you can extract a particular sum by differentiation. Please look at my proof in https://math.stackexchange.com/questions/400792/general-formula-for-partial-sum-of-series . It's quite common as a "formal" process; i.e. a symbolic process that generates wanted coefficients without things like convergence. Finding your f(x) above doesn't seem impossible, but is harder. – rrogers Feb 20 '19 at 12:42
  • @rrogers Can you elaborate a bit more? I would greatly appreciate it. – user3108815 Mar 01 '19 at 18:19
  • It's better to refer to "The Method of Coefficients Donatella Merlini, Renzo Sprugnoli, Maria Cecilia Verri" for instance at: https://pdfs.semanticscholar.org/bdb4/5bb13c824fb9d7497fb35982c12192adb623.pdf , It is also on JSTOR. Work your way down to Theorem 2.1 . The whole paper is easy to read and I found it educational. Since you are still interested I will continue to try to find a closed form (or not) for your raw sum; which is required by 1/(1-x) . BTW: above "individual infinite sums" should have been "individual partial sums" (: – rrogers Mar 02 '19 at 21:00
  • If you want I will order one of the used books or you can do it. (Bibliomania (especially mathematical) is a sad disease and should have a NIMH psychiatric classification). https://smile.amazon.com/gp/offer-listing/0486661652 – rrogers Mar 03 '19 at 14:31
  • @rrogers is it wrong of me to take advantage of your disease for my own personal gain? Because I am still interested in a solution, but I don't want to pay for that book. – user3108815 Mar 04 '19 at 00:48
  • Done.. Wednesday delivery. From my reading, I am guessing this is a combinatorial type of problem? – rrogers Mar 04 '19 at 19:24
  • @rrogers Correct, but take a look at the Taylor expansion of $cosh(x)$ . – user3108815 Mar 05 '19 at 03:02
  • Repost deleted/corrected: I think I have found an alternate formulation that looks simpler, but just as hard to sum. Are you interested in that? The idea is from "Companion to Concrete Mathematics" Melzak, the end of chapter three. Going on https://www.ias.ac.in/article/fulltext/pmsc/103/03/0269-0293 section 2 (iii) which refers to " Theory and Application of Infinite Series" Knopp; which is on Amazon from 1(!) seller for $8.xx . Which presumably has a proof of the conversion. Sorry I had the wrong link – rrogers Mar 05 '19 at 22:50
  • How about a slightly different problem? I don't know the source so I can't say. I do have other results but they are complicated; i.e. I haven't resolved some problems.
    Generally, these are called Lambert series. Working in the Laplace transform arena we get something like ${\displaystyle \sum_{i=1}^{\infty}x^{i}\cdot\frac{y^{i}}{1-y^{i}}={\displaystyle \sum_{k=1}^{\infty}\frac{x\cdot y^{i}}{1-x\cdot y^{i}}}}$ The sums are the same but the terms are different. The right-hand form seems easy to partially sum.

    Remark 1. BTW: this formula/conversion is presently approved for -1<x,y<1

    – rrogers Mar 11 '19 at 15:52
  • @rrogers you see that the factorial in the denominator introduces all sorts of complication though. – user3108815 Mar 11 '19 at 16:47
  • @rrogers btw, the link you posted can't be reached (the ias.ac.in article) – user3108815 Mar 11 '19 at 16:53
  • Not really, should I post the solution with my suggestion of finding alternate terms? You see x in my above equation is really (1/p) Laplace variable p and the inverse transform creates the factorial. There are two solutions (really the same) one is n multiple integrations the other is inverse Laplace transform of the series int(1/p^n ...) which is the same thing. The integrations are trivial though. – rrogers Mar 11 '19 at 18:23
  • Reaching the article? I can still reach it; would you like a copy? I can put it in dropbox for pickup. – rrogers Mar 11 '19 at 18:25
  • @rrogers yes please (about the Dropbox copy), and the the two solutions you mentioned, I presume you're still working on them before a post, yeah? – user3108815 Mar 11 '19 at 18:30
  • https://www.dropbox.com/s/mp6zccvkbaahe3n/0269-0293.pdf?dl=0 I do have them clear in my mind but I have an appointment. About 4 hours. – rrogers Mar 11 '19 at 18:57

2 Answers2

1

This is a partial answer giving a double path-integral over the function defined in (1).

Let us write for simplicity $$H(x,y) := \sum_{m=1}^\infty \frac{(xy)^m}{(2m)!(1-y^m)}.$$ This sum consits of all diagonal terms of the double series $$G(x,y):=\sum_{m=1}^\infty \frac{x^m}{(2m)!} \sum_{n=1}^\infty \frac{y^n}{1-y^n}.$$ For $x \in \mathbb{C}_{-} := \mathbb{C} \setminus \{ x \in \mathbb{R} : x \le 0 \}$ we have $$\sum_{m=1}^\infty \frac{x^m}{(2m)!} = \cosh(\sqrt{x}) -1$$ where we take the the main branch for the square root. We also have for $y$ with $ 0< \mathrm{Re}(y) < 1$ $$ \sum_{n=1}^\infty \frac{y^n}{1-y^n} = \frac{\psi_y(1) + \log(1-y)}{\log(y)},$$ where $\psi_q$ denotes the q-Polygamma function. Thus $$\tag{1} G(x,y) = \{ \cosh(\sqrt{x}) -1 \} \frac{\psi_y(1) + \log(1-y)}{\log(y)}.$$ Now, we can apply the Cauchy integral formula to get \begin{align} H(x,y) &= \frac{1}{2\pi i} \sum_{n=1}^\infty \int_{ \partial B_\varepsilon(x)} \int_{\partial B_\varepsilon(y)} \frac{G(z,w)}{(z-x)^n (w-y)^n} \, dw dz \\ &= \frac{1}{2\pi i} \int_{\partial B_\varepsilon(x)} \int_{\partial B_\varepsilon(y)} \frac{G(z,w)}{(z- x)(w-y)-1} \, dw dz. \end{align} At this point, one may use the residue theorem.

p4sch
  • 7,863
  • Looks nice but I don't see how the "diagonal" elements are extracted from G(x,y). I also don't see the last conversion. Could we have a reference for it? It almost makes sense but seems a little wrong (to me). – rrogers Mar 20 '19 at 12:40
  • Furthermore, some of us don't know what the residue theorem is, if you have a bit of time, I'd appreciate a full answer! – user3108815 Mar 20 '19 at 21:30
  • @p4sch it's also not clear where this $w$ is coming from? what is that? – user3108815 Mar 21 '19 at 04:33
  • In the last step, I have used the Cauchy integral formula twice times in order to get all diagonal terms. (The Integral gives exactly the $n$-th diagonal term.) See Wikipedia for Cauchy's integral formula. The residue theorem is an extension of Cauchy's theorem. I am not quite sure, if we can real find a nice identity for the last double path-integral. – p4sch Mar 21 '19 at 12:30
  • @p4sch how did you compute the sum as you did in the last step? Aren't the denominator's terms real numbers, and the numerator a constant? Wouldn't then that just be the geometric sum formula? Are there things I'm missing here? – user3108815 Mar 21 '19 at 17:09
  • Also, the article for Cauchy's Integral formula doesn't mention diagonal sums. The closest I could find was this link here which makes explicit the Cauchy product and the diagonalization sums, but not the one we're interested in: one of the two bisecting diagonals, from top left to bottom right. How indeed does the integral grant me that? – user3108815 Mar 21 '19 at 17:56
  • @p4sch I can see how one would model this as a trace of a matrix! Can you obtain the trace from the Cauchy Product of a matrix? – user3108815 Mar 21 '19 at 18:34
  • I am letting the "bounty" expire but will reinstate (and award it) if the summation is explained and a reference is given that I can understand for the diagonalization. The summation looks like a geometric series conversion but the terms are upside down. If I go from the last equation to the preceding one I get inverted terms. Was this intended? I did look at Wikipedia but couldn't see diagonalization. – rrogers Mar 23 '19 at 12:12
1

To try to answer my own question, here's something I started. Be warned, it ends in a tautology, but it may help spark some thought. All variables below are assumed $\in \mathbb N$ unless stated otherwise.

We start with something similar: $$\sum_{m=1}^\infty \frac{x^m}{m!} = e^{x}$$ $$\sum_{m=1}^\infty \frac{x^m(1-y^m)}{m!(1-y^m)} = e^{x}$$ $$\sum_{m=1}^\infty \frac{x^m}{m!(1-y^m)} - \sum_{m=1}^\infty \frac{x^my^m}{m!(1-y^m)}= e^{x}$$ Now we focus on the first term of the LHS. From the CDF of the Poisson distribution (see chart), we have (if we assume some variable $k \in \mathbb N$ and ommit results for $k=0$) : $$e^{-x}\sum_{m=1}^k \frac{x^m}{m!} = \frac{\Gamma(k+1,\ x)}{k!}$$ $$e^{-x}\sum_{m=1}^{k-1} \frac{x^m}{m!} = \frac{\Gamma(k,\ x)}{(k-1)!}$$

The numerators of the RHS's of the above two lines are the upper incomplete gamma function. See relevant information about the upper incomplete gamma function, we'll use it again later

It follows then that: $$\frac{1}{1-y^k} \sum_{m=1}^k \frac{x^m}{m!} = \frac{1}{1-y^k} \frac{e^x\ \Gamma(k+1,\ x)}{k!}$$ $$\frac{1}{1-y^k} \sum_{m=1}^k \frac{x^m}{m!} - \frac{1}{1-y^k} \sum_{m=1}^{k-1} \frac{x^m}{m!} = \frac{1}{1-y^k} \frac{e^x\ \Gamma(k+1,\ x)}{k!} - \frac{1}{1-y^k} \frac{e^x\ \Gamma(k,\ x)}{(k-1)!} $$ $$\sum_{k=1}^\infty \frac{x^k}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1,\ x)}{k!(1-y^k)} - \sum_{k=1}^\infty \frac{e^x\ \Gamma(k,\ x)}{(k-1)!(1-y^k)}$$

The LHS of the above equality should look familiar. Start with the first term of the RHS of the last equality:

$$\sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1,\ x)}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x\ (\Gamma(k+1) - \gamma(k+1,\ x))}{k!(1-y^k)}$$ $$= \sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1)}{k!(1-y^k)} - \sum_{k=1}^\infty e^x \frac{\gamma(k+1,\ x)}{k!(1-y^k)} $$

The lower case gamma is the "lower incomplete gamma function", from the article above on the upper incomplete gamma function. The substitution in the last equality comes from a holomorphic extension cited in the article. The ".gov" site (NIST Digital Library of Mathematical Functions) hosting the quoted result may not be accessible, which is why I "cite" the Wikipedia articles.

$$\sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1)}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x}{1-y^k} = e^x \sum_{k=1}^\infty \frac{1}{1-y^k} = e^x (\frac{\psi_\frac{1}{y}(1)-\ln(y-1)+\ln\frac{1}{y}}{\ln y})$$

$\psi_q(a)$ where $(a,q) \in \mathbb R$ is known as the the q-digamma function (see Wolfram Alpha, Wikipedia, the other answers, etc). A closed form of the RHS is provided above, you can use Wolfram Alpha (Mathematica really) to verify. Onward. There are a few ways to tackle the next part. I don't know which of them will work, so maybe we try the simplest ones first.

$$\sum_{k=1}^\infty e^x\frac{\gamma(k+1,\ x)}{k!(1-y^k)} = \sum_{k=1}^\infty e^x\frac{\sum_{i=0}^\infty \frac{(-1)^i\ (x)^{k+1+i}}{i!(k+1+i)}}{k!(1-y^k)} =\ ...$$

The series representation substitution comes from the Wikipedia article on the Incomplete Gamma Function, specifically the Lower Incomplete Gamma Function section. Here we will cheat; manually input values of $k$ (e.g. increment value by one) and examine the expanded forms of the outputs. These forms are then summed for each $k$. Do this until you can convince yourself that you can and should prove through induction the pattern you observe. We may need Mathematica or some other computational tool in addition to some induction to deduce:

$$...\ = \sum_{k=1}^\infty \frac{e^x}{1-y^k} - \sum_{k=1}^\infty \frac{1}{1-y^k}-\sum_{k=1}^\infty (\frac{x^{k}}{k!} \sum_{i=k}^\infty \frac{1}{1-y^i})$$

The last term's sum looks particularly nasty, and in truth it is. We can make a few observations and simplifications, including that the last term of the RHS is just:

$$-\sum_{k=1}^\infty (\frac{x^{k}}{k!} \sum_{i=k}^\infty \frac{1}{1-y^i}) = -\sum_{k=1}^\infty (\frac{x^{k}}{k!} (\sum_{i=1}^\infty \frac{1}{1-y^i} - \sum_{i=1}^{k-1} \frac{1}{1-y^i}))$$ $$= -\sum_{k=1}^\infty \frac{x^{k}}{k!}(\sum_{i=1}^\infty \frac{1}{1-y^i}+ \frac{\psi_y(k)}{\ln y}-(k-1)-\frac{\psi_y(1)}{\ln y})$$ After doing the distribution of the $\frac{x^k}{k!}$ across the terms and noting that $$\sum_{k=1}^\infty (\frac{x^{k}}{k!} \sum_{i=1}^\infty \frac{1}{1-y^i}) =e^x \sum_{i=1}^\infty \frac{1}{1-y^i} = e^x \sum_{k=1}^\infty \frac{1}{1-y^k} = e^x (\frac{\psi_\frac{1}{y}(1)-\ln(y-1)+\ln\frac{1}{y}}{\ln y})$$ you'll find the new sums equal to the following:$$-\sum_{k=1}^\infty (\frac{x^{k}}{k!} \sum_{i=k}^\infty \frac{1}{1-y^i}) = - \sum_{k=1}^\infty \frac{e^x}{1-y^k}- \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k)\biggl ) +e^x(x-1)+1+\frac{e^x\ \psi_y(1)}{\ln y}$$

I can immediately see a large problem with the sum using $\psi_y(k)$. Press on though, by taking a step back to perform the substitutions to simplify the work a bit:

$$\sum_{k=1}^\infty e^x\frac{\gamma(k+1,\ x)}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x}{1-y^k} - \sum_{k=1}^\infty \frac{1}{1-y^k}-\sum_{k=1}^\infty (\frac{x^{k}}{k!} \sum_{i=k}^\infty \frac{1}{1-y^i})$$ $$ = - \sum_{k=1}^\infty \frac{1}{1-y^k}- \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k)\biggl ) +e^x(x-1)+1+\frac{e^x\ \psi_y(1)}{\ln y}$$

Continue by pausing this particular expansion and "simplify" a previous equality's terms. From: $$\sum_{k=1}^\infty \frac{x^k}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1,\ x)}{k!(1-y^k)} - \sum_{k=1}^\infty \frac{e^x\ \Gamma(k,\ x)}{(k-1)!(1-y^k)}$$

We now focus on the expanded second term of the RHS, and use the same tricks as before: $$- \sum_{k=1}^\infty \frac{e^x\ \Gamma(k,\ x)}{(k-1)!(1-y^k)} = -\sum_{k=1}^\infty \frac{e^x\ \Gamma(k)}{(k-1)!(1-y^k)} + \sum_{k=1}^\infty e^x \frac{\gamma(k,\ x)}{(k-1)!(1-y^k)}$$ $$= -\sum_{k=1}^\infty \frac{e^x}{1-y^k} + \sum_{k=1}^\infty e^x\frac{\sum_{i=0}^\infty \frac{(-1)^i\ (x)^{k+i}}{i!(k+i)}}{(k-1)!(1-y^k)}$$ $$\ =-\sum_{k=1}^\infty \frac{e^x}{1-y^k} +\sum_{k=1}^\infty \frac{e^x}{1-y^k} - \sum_{k=1}^\infty \frac{1}{1-y^k}-\sum_{k=2}^\infty \biggr (\frac{x^{k-1}}{(k-1)!} \sum_{i=k}^\infty \frac{1}{1-y^i}\biggl )$$ $$\ =- \sum_{k=1}^\infty \frac{1}{1-y^k}- \sum_{k=1}^\infty \frac{e^x-1}{1-y^k}- \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) +xe^x+\frac{(e^x-1)\ \psi_y(1)}{\ln y}$$ $$\ =- \sum_{k=1}^\infty \frac{e^x}{1-y^k}- \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) +xe^x+\frac{(e^x-1)\ \psi_y(1)}{\ln y}$$

Now we can start combining our results! First, we have to note a few things: we pulled a trick in the 2nd to last line, on the last term of the RHS. Instead of writing: $$\sum_{k=1}^\infty \biggr (\frac{x^{k}}{k!} \sum_{i=k+1}^\infty \frac{1}{1-y^i}\biggl )$$ We wrote it as: $$\sum_{k=2}^\infty \biggr (\frac{x^{k-1}}{(k-1)!} \sum_{i=k}^\infty \frac{1}{1-y^i}\biggl )$$

I entreat you to prove to me they are not equivalent, given our stated variable constraints. Seriously, please do. Next, we never actually calculated $$\sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1,\ x)}{k!(1-y^k)}$$ by combining our previous results of the expansions of the two terms in its own expansion (the one we got via holomorphic extension from the article, remember?).

$$\sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1,\ x)}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1)}{k!(1-y^k)} - \sum_{k=1}^\infty e^x \frac{\gamma(k+1,\ x)}{k!(1-y^k)}$$ $$ = \sum_{k=1}^\infty \frac{e^x}{1-y^k}+ \sum_{k=1}^\infty \frac{1}{1-y^k}+ \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k)\biggl ) -e^x(x-1)-1-\frac{e^x\ \psi_y(1)}{\ln y}$$ Combine the above result with one of our prior results: $$- \sum_{k=1}^\infty \frac{e^x\ \Gamma(k,\ x)}{(k-1)!(1-y^k)}=- \sum_{k=1}^\infty \frac{e^x}{1-y^k}- \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) +xe^x+\frac{(e^x-1)\ \psi_y(1)}{\ln y}$$ To obtain: $$\sum_{k=1}^\infty \frac{x^k}{k!(1-y^k)} = \sum_{k=1}^\infty \frac{e^x\ \Gamma(k+1,\ x)}{k!(1-y^k)} - \sum_{k=1}^\infty \frac{e^x\ \Gamma(k,\ x)}{(k-1)!(1-y^k)}$$ $$= \sum_{k=1}^\infty \frac{1}{1-y^k}+ \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k)\biggl ) -e^x(x-1)-1-\frac{e^x\ \psi_y(1)}{\ln y}- \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) +xe^x+\frac{(e^x-1)\ \psi_y(1)}{\ln y}$$ $$=\sum_{k=1}^\infty \frac{1}{1-y^k}+ \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k)\biggl ) - \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) +e^x-1 -\frac{\psi_y(1)}{\ln y}$$

The last "trick" is to take care of those pesky q-digamma function sums. You may use Wolfram Alpha to verify. Observe: $$- \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) = - \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k+1)\biggl ) $$ $$\frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}\psi_y(k)\biggl ) - \frac{1}{\ln y} \biggr (\sum_{k=2}^\infty \frac{x^{k-1}}{(k-1)!}\psi_y(k)\biggl ) = \frac{1}{\ln y} \biggr (\sum_{k=1}^\infty \frac{x^{k}}{k!}(\psi_y(k)-\psi_y(k+1))\biggl )$$ $$= -\sum_{k=1}^\infty \frac{x^ky^k}{k!(y^k-1)}$$ Reinsert this new "simplification" into the last main result:

$$\sum_{k=1}^\infty \frac{x^k}{k!(1-y^k)} =\sum_{k=1}^\infty \frac{1}{1-y^k}- \sum_{k=1}^\infty \frac{x^ky^k}{k!(y^k-1)} +e^x-1 -\frac{\psi_y(1)}{\ln y}$$

Finally, we return to one of our earliest results (employ a "change of variables" and use $k$ instead of $m$ like we did above): $$\sum_{k=1}^\infty \frac{x^k}{k!(1-y^k)} - \sum_{k=1}^\infty \frac{x^ky^k}{k!(1-y^k)}= e^{x}$$ Make the appropriate substitution: $$\sum_{k=1}^\infty \frac{1}{1-y^k}- \sum_{k=1}^\infty \frac{x^ky^k}{k!(y^k-1)} +e^x-1 -\frac{\psi_y(1)}{\ln y} - \sum_{k=1}^\infty \frac{x^ky^k}{k!(1-y^k)}= e^{x}$$

Observe another "simplification": $$ \sum_{k=1}^\infty \frac{x^ky^k}{k!(y^k-1)} = - \sum_{k=1}^\infty \frac{x^ky^k}{k!(1-y^k)}$$

Make the substitution: $$\sum_{k=1}^\infty \frac{1}{1-y^k}+ \sum_{k=1}^\infty \frac{x^ky^k}{k!(1-y^k)} +e^x-1 -\frac{\psi_y(1)}{\ln y} - \sum_{k=1}^\infty \frac{x^ky^k}{k!(1-y^k)}= e^{x}$$

$$\sum_{k=1}^\infty \frac{1}{1-y^k} -1 -\frac{\psi_y(1)}{\ln y} = 0$$ Which is not helpful at all! Somewhere we got caught in a tautology, will contemplate how to fix...