6

Let $\varphi(n):=(-1)^{n+1}(n+1)2^{2n}$.

I am able to prove the following identity (${\color{red}{\mathbf{LHS}}}$=infinite series, ${\color{blue}{\mathbf{RHS}}}$=finite sum) \begin{align*} {\color{red}{\frac{1}{\varphi(n)}\sum_{k\geq1}\frac{(-1)^k}{\binom{3n+k+2}{2n+2}}}} &={\color{blue}{\sum_{j=1}^n\frac{(-1)^j(28j^2+24j+4)}{\binom{3j}j j (3j+2)(3j+1)\,2^{2j}}+3-4\ln 2}}. \tag1 \end{align*} Replacing $n=0$ in equation (1) leads to $\displaystyle\sum_{j\geq1}\frac{(-1)^j}{\binom{j+2}2}=4\ln 2-3$.

Letting $n\rightarrow\infty$ in (1) unearths the less trivial result $$\sum_{j=1}^{\infty}\frac{(-1)^j(28j^2+24j+4)}{\binom{3j}j j (3j+2)(3j+1)\,2^{2j}}=4\ln 2-3.$$

QUESTION. How can one exercise an analytic continuation of the ${\color{blue}{\mathbf{RHS}}}$ in (1), ${\color{purple}{\text{in the variable $n$}}}$, so that we would be able to compute at $n=-\frac23$ and derive (from it) a value for the infinite series on the ${\color{red}{\mathbf{LHS}}}$, i.e. $$\frac1{\varphi(-\frac23)}\sum_{k\geq1}(-1)^k\binom{k}{\frac23}^{-1}\,?$$

T. Amdeberhan
  • 41,802
  • 1
    I wonder why you're interested in an analytic continuation for this function, and in what way the particular value $n = - \frac{2}{3}$ is relevant to your research – Max Muller Jul 01 '23 at 22:27
  • 1
    There are many such continuations. In particular, one can give it any value at $z=-2/3$. The general theorem is: If ${ z_n}\subseteq\mathbb C$ is any discrete set and $a_n\in\mathbb C$ are arbitrary numbers, then there are entire functions $f$ with $f(z_n)=a_n$ for all $n$. – Christian Remling Jul 01 '23 at 23:42
  • True, thank you, but can one give concrete and constructive solution so that we may compute with it here? I mean, instead of existential claims. – T. Amdeberhan Jul 02 '23 at 00:16
  • 2
    A natural way to extend to real variables a partial sum $s(n):=\sum_{k=1}^n f(k)$ is via $$s(x):=\sum_{k=1}^\infty\big{ f(k)- f(k+x)\big}$$ (provided the latter converges); it then satisfies $s(x+1)=s(x)+f(x+1)$ (this is how the digamma function is obtained). In your case it should give an analytic function although maybe too complicated. – Pietro Majer Jul 02 '23 at 10:18
  • 1
    @T.Amdeberhan I would like to know how you proved the identity in the first place. I haven't seen such an identity before. – piepie Jul 04 '23 at 04:49

3 Answers3

5

As for the sum $$\sum_{k\geq1}(-1)^k\binom{k}{\frac23}^{-1}$$ one can evaluate it by means of the Beta function integral, like in this recent computation. $$\sum_{k\geq1}(-1)^k\binom{k}{\frac23}^{-1}=\frac23\sum_{k\geq1}(-1)^k \frac{\Gamma\big(k+\frac13\Big)\Gamma\big(\frac23\big)}{\Gamma(k+1)}=\frac23\sum_{k\geq1}(-1)^kB\Big(k+\frac13,\frac23\Big)=$$ $$=\frac23\sum_{k\geq1} \int_0^1(-1)^k {x^{k-\frac23}}{(1-x)^{-\frac13}}dx=-\frac23 \int_0^1\frac{1}{1+x} \Big(\frac{x }{1-x}\Big)^{1/3}dx=$$

The latter integral is easily rationalised by changing variable; we get $$=-\int_0^\infty\frac{2t^3}{(t^3+1)(2t^3+1)}dt=\int_0^\infty\frac{2dt}{2t^3+1 } -\int_0^\infty\frac{2dt}{ t^3+1}=$$$$=(\sqrt[3]4-2)\int_0^\infty\frac{dt}{ t^3+1}$$ whence $$\sum_{k\geq1}(-1)^k\binom{k}{\frac23}^{-1}= \frac{2\pi}{3\sqrt3} (\sqrt[3]4-2)$$ which evaluates to $-0.4989..$.

Pietro Majer
  • 56,550
  • 4
  • 116
  • 260
4

\begin{align*} \sum_{k\geq 1} (-1)^k \frac{1}{\binom{k}{x}} &=x!(-x)!\sum_{k\geq1}(-1)^k\frac{(-x+k)!}{(-x)!\,k!} =x!(-x)!\sum_{k\geq1}\binom{x-1}{k} \\ &=x!(-x)!(2^{x-1}-1)=\frac{\pi x}{\sin(\pi x)} (2^{x-1}-1):=f(x) \end{align*} for $0< x < 2$, straightforwardly from $(-1)^k \binom{\alpha-1+k}{k}=\binom{-\alpha}{k}$ with $\alpha-1=-x$.

Hence $f(2/3)\simeq -.498914$


Remarks:

(I'm way behind in organizing and formatting many sets of notes into semi-coherent pdfs, so I'm too pressed for time to pursue this to any further depth, but I'm tempted to test (extend my understanding of) some umbral Sheffer operational calculus on this though.)

An infinite sum resolved to a finite sum is typically what you get with a generalized Dobinski-type formula or Euler binomial transform of an e.g.f. where one side has coefficients polynomial in $k$ (see, e.g., this MO-Q on the Dobinski formula).

With $B_n(x)$ the Bell / Touchard / Stirling polynomials of the second kind,

$$g(B.(x)) = e^{-x} \sum_{k \geq 0} g(k) \frac{x^k}{k!} =e^{-x}e^{g.x} = e^{-(1-g.)x}$$

$$ = \sum_{k \geq 0} (-1)^k(1-g.)^k \frac{x^k}{k!} = \sum_{k \geq 0} (-1)^k\triangledown_{j=0}^k g(j) \frac{x^k}{k!}$$

and the last is a finite sum when $g(j)$ is a polynomial. Multiplying by $(1+x)$ sums up pairs of coefficients.

Umbrally substitute the falling factorial polynomial, a.k.a. the Stirling polynomials of the first kind, $(x)_m = m!\binom{x}{m}$ for $x$ in $g(B.(x))$ to retrieve $g(x)$ as the analytic continuation of $g(n)$.

I wonder if you can approach the problem this way to relate the LHS to the RHS, eventually evaluating at $x=1$.

Tom Copeland
  • 9,937
3

I liked the two contributions by Pietro Majer. One is the evaluation of the series in his answer box, the other is a pragmatic direction towards extending the sum for real values of $n$. I would like to expand on the latter.

Denoting the function $$S(N):={\color{blue}{\sum_{j=1}^N\frac{(-1)^j(28j^2+24j+4)}{\binom{3j}j j (3j+2)(3j+1)\,2^{2j}}}}$$ one can perform the analytic continuation by the following procedure \begin{align*} S(N)&=\sum_{j=1}^{\infty}-\sum_{j=N+1}^{\infty} \\ &=4\ln 2-3- \sum_{j=1}^{\infty}\frac{(-1)^j(28(j+N)^2+24(j+N)+4)}{\binom{3j+3N}{j+N}(j+N) (3j+3N+2)(3j+3N+1)\,2^{2j+2N}}. \end{align*} Therefore, \begin{align*} S(-2/3)&=4\ln 2 -3-\sum_{j=1}^{\infty}\frac{(-1)^j(252j^2-120j+4)}{9\binom{3j-2}{j-\frac23}\,j\,(3j-2) (3j-1)\,2^{2j-\frac43}} \end{align*} and hence \begin{align*} \sum_{k\geq1}(-1)^k\binom{k}{\frac23}^{-1} &=\sum_{j=1}^{\infty}(-1)^j\frac{252j^2-120j+4}{27j\,(3j-2) (3j-1)\,2^{2j}}\binom{3j-2}{j-\frac23}^{-1}. \end{align*} Ripping the benefit: we have found a "closed form" for a fairly complicated infinite series $$\sum_{j=1}^{\infty}(-1)^j\frac{252j^2-120j+4}{27j\,(3j-2) (3j-1)\,2^{2j}}\binom{3j-2}{j-\frac23}^{-1} =\frac{2\pi}{3\sqrt3} (\sqrt[3]4-2).$$

Another possible approach for an analytic continuation is: let $f\left(x\right)\in C^{1}\left[a,b\right]$, then \begin{align*} \sum_{a<n\leq b}f\left(n\right)&=\int_{a}^{b}f\left(x\right)dx+\int_{a}^{b}\left(x-\left\lfloor x\right\rfloor-\frac{1}{2}\right)f'\left(x\right)dx \\ & \qquad \qquad \qquad +\left(a-\left\lfloor a\right\rfloor-\frac{1}{2}\right)f\left(a\right)-\left(b-\left\lfloor b\right\rfloor-\frac{1}{2}\right)f\left(b\right) \end{align*} for suitable choices of $a$ and $b$. Here, $\lfloor\cdot\rfloor$ stands for the floor function.

T. Amdeberhan
  • 41,802
  • Another expression, via beta integral again, gives $$\sum_{k\geq1}\frac{(-1)^k}{\binom{3n+k+2}{2n+2}}=-(2n+2)\int_0^1\frac{x^{n+1}(1-x)^{2n+1}}{1+x}dx,$$ that holds for real $n>-1$ (defining the binomial in the LHS via Gamma function). Maple finds a complicated expression for the RHS in terms of hypergeometric functions again. – Pietro Majer Jul 02 '23 at 16:32
  • 1
    @PietroMajer, nice. Also, change $2n +1$ to $s-1$ and divide by $(s-1)!$. Then you have the fractional derivative (and a modified Mellin transform for the HGF) on the RHS of the eqn. in your comment: $D_{x=1}^{-s} (1+x)^{-1} x^{-s/2}.$ Use the binomial expansion or generalized Leibniz formula of Osler to evaluate. $D_x^{-s} H(x) x^{\alpha} / \alpha! =H(x) x^{\alpha+s}/(\alpha+s)!$ with the Heaviside step. – Tom Copeland Jul 02 '23 at 19:22
  • Pietro, so on the LHS you could probably add a factor of $y^k$ and change the limit from $1$ to $y$ in the integral and $(1-x)$ to $(y-x)$ to generalize to $D_{x=y}^{-s}$. Lots of ways to generalize. – Tom Copeland Jul 02 '23 at 19:35
  • I was thinking the same, so you get the convolution of $x_+^{2n+1}$ and $\frac{x_+^{n+1}}{1+x}$... – Pietro Majer Jul 02 '23 at 20:37