5

Inspired by these two questions asking about the case n = 3 and n=4, I was wondering what is $$S =\sum_{k=1}^\infty\frac{(-1)^{k-1}}{k^n 2^k {2k \choose k}}$$ for positive integer $n \ge 3$.

For $n = 3$, the sum is $\frac{1}{4}\zeta (3)-\frac{1}{6}\ln^3(2) = \frac{1}{2}\operatorname{Li}_3\left(\frac{1}{2}\right)+\frac{1}{2}\ln(2)\operatorname{Li}_2\left(\frac{1}{2}\right)-\frac{3}{16}\zeta(3)$.

For $n = 4$, the sum is $4\operatorname{Li}_4\left(\frac12\right)-\frac72\zeta(4)+\frac{13}4\ln2\zeta(3)-\ln^22\zeta(2)+\frac5{24}\ln^42$.

Using similar work as from the two linked questions, the sum can be re-expressed as the integral $$\frac{2\cdot(-1)^{n-1}}{(n-3)!}\int_0^1\text{arcsinh}^2\left(\sqrt{\frac{x}{8}}\right)\frac{\ln^{n-3}(x)}{x}dx$$

Setting $u = \text{arcsinh}\left(\sqrt{\frac{x}{8}}\right)$, we get $$S = \frac{4\cdot(-1)^{n-1}}{(n-3)!}\underbrace{\int_0^{\frac{\ln(2)}{2}} u^2\ln^{n-3}(8\sinh^2(u))\coth(u) du}_{\large {I}}$$

$$I = \int_0^{\frac{\ln(2)}{2}}u^2\coth(u)\sum_{k=0}^{n-3}\left({n-3\choose k}\ln^{n-3-k}(8)(2\ln(\sinh(u)))^{k}\right) du$$ $$I = \sum_{k=0}^{n-3}{n-3\choose k}\ln^{n-3-k}(8)2^{k}\underbrace{\int_0^{\frac{\ln(2)}{2}}u^2\coth(u)\ln^{k}(\sinh(u))du}_{\large J}$$

Making the substitution $v = \sinh(u)$ and simplifying, we get $$J = \int_0^{\frac{1}{2\sqrt{2}}}\frac{\text{arcsinh}^2(v)\ln^k(v)}{v}dv$$

Although this may or may not help, making the substitution $w = \ln(v)$, we get $$J = \int_{-\infty}^{-\ln(2\sqrt{2})}w^k\text{arcsinh}^2(e^w)dw$$

From here, I don't know what to do to find $J$.

How can I, either through this process or a completely different one, find:

$1.$ The value of $S$ for integer $n \ge 3$?

$2.$ The value of $J$ for integer $k \ge 0$?

  • I did case n=5 and I'll post it today but there is no closed form for some higher weights as we will come across alternating sums with no known closed form. – Ali Shadhar Jul 29 '19 at 17:33
  • @AliShather why is there no closed form for higher $n$? Is there no closed form for $J$ for $k \ge 3$? – Varun Vejalla Jul 29 '19 at 17:51
  • If we work with higher weights we will come across $\sum_{n=1}^\infty \frac{(-1)^n H_n}{n^a}$ and there is no closed form for this sum for odd $ a>3$. As for even a, we have a generalization. – Ali Shadhar Jul 29 '19 at 17:58
  • @AliShather I made some further work on this problem and edited to include my attempts. That may help you and others. – Varun Vejalla Jul 30 '19 at 02:41
  • @AliShather Could you post the closed form for $n = 5$? It might help in finding a pattern that could then be proven. – Varun Vejalla Jul 30 '19 at 22:06
  • 1
    sure . check here https://math.stackexchange.com/questions/3308137/challenging-sum-calculate-sum-k-1-infty-frac-1k-1k52k2k-choose/3308153#3308153 – Ali Shadhar Jul 30 '19 at 22:09
  • @Editprofileandsettings Thanks for your comment and the article. I am reading the article now. – Varun Vejalla May 01 '20 at 20:46

1 Answers1

3

Nothing about the closed-form, but a generalization that might be helpful. (Note: The closed-form for $n = 5$ can be found in this other post.)

Consider the following function:

$$f(a,s)=\int_0^1\text{arcsinh}^2\left(\sqrt{ax}\right) x^{s-1} dx$$

It is known that:

$$\text{arcsinh} \sqrt{ax}=\text{arctanh} \frac{\sqrt{ax}}{\sqrt{1+ax}}$$

Using a generating function from this answer we can write:

$$\text{arctanh}^2 \frac{\sqrt{ax}}{\sqrt{1+ax}}= \frac{1}{2} \frac{ax}{1+ax} \sum_{m=0}^\infty \frac{H_{m+1/2}+\log 4}{m+1} \frac{a^mx^m}{(1+ax)^m}$$

Now consider the integral:

$$g_m(a,s)=\int_0^1 \frac{x^{m+s}}{(1+a x)^{m+1}} dx$$

Technically, this is a hypergeometric function, but we can express it as a more simple series, by repeatedly differentiating $g_0(a,x)$ w.r.t. $a$:

$$g_m(a,s)=\frac{(-1)^m}{m!} \sum_{k=m}^\infty \frac{(-1)^k k! a^{k-m}}{(k+s+1) (k-m)!}$$

Combining the two results, we have:

$$f(a,s)=\frac{1}{2}\sum_{m=0}^\infty (-1)^m \frac{H_{m+1/2}+\log 4}{(m+1)!} \sum_{k=m}^\infty \frac{(-1)^k k!~ a^{k+1}}{(k+s+1) (k-m)!}$$

Now consider:

$$F_n(a)=\frac{2\cdot (-1)^n}{n!}\int_0^1\text{arcsinh}^2\left(\sqrt{a x}\right)\frac{\ln^n(x)}{x}dx$$

It's clear that:

$$F_n(a)= \frac{2\cdot(-1)^n}{n!} \frac{\partial^n f(a,s)}{\partial s^n} \bigg| _{s=0}$$

Which makes for a neat double series:

$$F_n(a)=\sum_{m=0}^\infty (-1)^m \frac{H_{m+1/2}+\log 4}{m+1} \sum_{k=m}^\infty \binom{k}{m} \frac{(-1)^k a^{k+1}}{(k+1)^{n+1}}$$

$$ S=\sum_{k=1}^\infty\frac{(-1)^{k-1}}{k^n 2^k {2k \choose k}}=F_{n-3} \left( \frac{1}{8} \right)$$

Note that $H_{m+1/2}+\log 4$ are rational numbers.

The original series is definitely less complicated, but it might be that the double series may offer some insight.


As an example to judge the convergence rate of the new series, for $n=5$ we get $20$ correct digits by using the following number of terms:

$$\sum_{m=0}^{15} (-1)^m \frac{H_{m+1/2}+\log 4}{m+1} \sum_{k=m}^{15} \binom{k}{m} \frac{(-1)^k }{(k+1)^3 8^{k+1}}=0.24872280253516023269 \ldots$$

$$\sum_{k=1}^{16} \frac{(-1)^{k-1}}{k^5 2^k {2k \choose k}}=0.24872280253516023269 \ldots$$

Which means that the original sum converges faster.

But again, the generalization might be useful:

$$\sum_{k=1}^\infty \frac{(-1)^{k-1} (4a)^k}{k^n {2k \choose k}}=F_{n-3}(a)$$


By the uniqueness of power series, we can now collect the coefficients:

$$\sum_{m=0}^\infty \frac{H_{m+1/2}+\log 4}{m+1} \sum_{k=0}^\infty \binom{m+k}{m} \frac{(-1)^k a^{m+k+1}}{(m+k+1)^{n-2}}=\sum_{l=0}^\infty \frac{(-1)^l (4a)^{l+1}}{(l+1)^n {2l+2 \choose l+1}}$$

$$k=l-m$$

$$\sum_{m=0}^\infty (-1)^m \frac{H_{m+1/2}+\log 4}{m+1} \binom{l}{m} =\frac{4^{l+1}}{(l+1)^2 {2l+2 \choose l+1}}$$

Simplifying, we obtain:

$$\sum_{m=0}^\infty (-1)^m \binom{l}{m} \frac{H_{m+1/2}+\log 4}{m+1} =\frac{2^{2l+1}}{(2l+1) (l+1) {2l \choose l}}$$

Multiplying by $x^l$ and summing from $0$ to $\infty$, we obtain:

$$\sum_{m=0}^\infty (-1)^m \frac{H_{m+1/2}+\log 4}{m+1} \frac{x^m}{(1-x)^{m+1}}= \sum_{l=0}^\infty \frac{2^{2l+1} x^l}{(2l+1) (l+1) {2l \choose l}}$$

After some simplifications, the right hand side gives us the Taylor series for $\frac{2}{x} \arcsin^2 \sqrt{x}$, or:

$$\frac{1}{x} \arcsin^2 \sqrt{x}=\sum_{l=0}^\infty \frac{(4 x)^l}{(2l+1) (l+1) {2l \choose l}}$$

Yuriy S
  • 31,474