7

Can you think of a simple way of proving that $$ \sum_{n=k+1}^\infty \frac{1}{n \choose k} $$ is rational for any $k \geq 2$?

Here's the background. Consider a series: $$ \sum_{n=1}^\infty \frac{1}{n(n+1)} $$

Elementary algebra gives us that: $$ \sum_{n=1}^\infty \frac{1}{n(n+1)} = \sum_{n=1}^\infty \left(\frac{1}{n} - \frac{1}{n+1}\right) = \lim_{k\to\infty} (1 - \frac{1}{k+1}) = 1 $$

so the sum turns out to be rational.

Next, consider $$ \sum_{n=1}^\infty \frac{1}{n(n+1)(n+2)(n+3)(n+4)} $$

With the same method, but much more effort we can show that: $$ \sum_{n=1}^\infty \frac{1}{n(n+1)(n+2)(n+3)(n+4)} = \sum_{n=1}^\infty \left(\frac{1}{24 n}-\frac{1}{6(n+1)}+\frac{1}{4(n+2)}-\frac{1}{6(n+3)}+\frac{1}{24(n+4)}\right) $$ and again we'll see that stuff cancels out, and the sum is again rational.

So the obvious conjecture is that this method will work for arbitrary (but $\geq 2$) fixed number of factors in denominator, and the sum will always be rational. Indeed that's the case. I provide a solution as an answer, but I'm not fully satisfied with it (it seems for me to be too brute force), so I'm looking for alternative solutions.

xyzzyz
  • 7,884

3 Answers3

4

Consider the generating function of the terms of our series (after simplifying renumbering): $$ F_k(x) = \sum_{n=k+1}^\infty \frac{x^n}{n(n-1)\cdots(n-k)} $$ By explicit calculation, it's easy to see that $k-th$ derivative of $F_k$ is equal to $-\log(1-x)$, so indeed we can calculate $F_k$ by taking antiderivative of $-\log(1-x)$ $k$ times. Surprisingly, it actually leads us somewhere in our case: Inductive argument using integration by parts show that the $k$-th antiderivative of $-\log(1-x)$ is $C (1-x)^k \log(1-x) + W(x)$, where $W$ is some polynomial of degree $k$ and $C$ is some rational constant (the exact value of C and the polynomial $W$ are not really hard to obtain, but they're not important here).

So, by general argument involving power series, we see that for $n > k$: $$ \frac{1}{n(n-1)\cdots(n-k)} = \frac{F^{(n)}_k(0)}{n!} $$

But then $$ \frac{F^{(n)}_k(x)}{n!} = C \frac{((1-x)^k log(1-x))^{(n)}}{n!} = C \sum_{i=0}^k (-1)^i {k \choose i} \frac{(x^i \log(1-x))^{(n)}}{n!} $$ The polynomial $W$ disappears because $n$ is larger than degree of $W$. Now, $\frac{(x^i \log(1-x))^{(n)}}{n!}$ is precisely the $n$-th coefficient in the power series expansion of $x^i \log(1-x) = x^i(-x - \frac{1}{2} x^2 - \ldots - \frac{1}{n} x^n + \ldots) = -x^{i+1} - \frac{1}{2} x^{i+2} - \ldots - \frac{1}{n} x^{i+n} + \ldots$, so it's exactly $\frac{1}{n-i}$. In the end, we obtain: $$ \sum_{n=k+1}^\infty \frac{1}{n(n-1)\cdots(n-k)} = \sum_{n=k+1}^\infty \frac{F^{(n)}_k(0)}{n!} = \sum_{n=k+1}^\infty C \sum_{i=0}^k (-1)^i {k \choose i} \frac{1}{n-i} $$ Using the fact that $\sum_{i=0}^k (-1)^i {k \choose i} = 0$ and appropriately reordering terms, we again see that most terms cancel out, and just like we saw for $k = 1$ and $k = 4$, sum is rational (remember that $C$ was rational). Bonus point is that since $C$ is easily computable (left as an exercise for reader), we can actually calculate the sum.

xyzzyz
  • 7,884
2

We show the result stated by @IgorRivin is valid.

\begin{align*} \sum_{n=k+1}^{\infty}\frac{1}{\binom{n}{k}}=\frac{1}{k-1}\qquad\qquad k\geq 2\tag{1} \end{align*}

We start proving a related identity from which we can easily derive (1).

The following is valid for $N\geq 1,k\geq 1$

\begin{align*} \sum_{n=1}^{N}\frac{1}{\binom{n+k+1}{k+1}}=\frac{1}{k}-\frac{\left(1+\frac{1}{k}\right)}{\binom{N+k+1}{k}}\tag{2} \end{align*}

We obtain

\begin{align*} \sum_{n=1}^{N}\frac{1}{\binom{n+k+1}{k+1}}&=\sum_{n=1}^{N}\frac{n!(k+1)!}{(n+k+1)!}\\ &=(k+1)!\sum_{n=1}^N\frac{1}{(n+1)\cdots(n+k+1)}\\ &=\frac{(k+1)!}{k}\sum_{n=1}^N\frac{(n+k+1)-(k+1)}{(n+1)\cdots(n+k+1)}\\ &=\frac{(k+1)!}{k}\sum_{n=1}^N\left(\frac{1}{(n+1)\cdots(n+k)}-\frac{1}{(n+2)\cdots(n+k+1)}\right)\tag{3}\\ &=\frac{(k+1)!}{k}\left(\frac{1}{(k+1)!}-\frac{1}{(N+2)\cdots(N+k+1)}\right)\\ &=\frac{1}{k}-\frac{\left(1+\frac{1}{k}\right)k!}{(N+2)\cdots(N+k+1)}\tag{4}\\ &=\frac{1}{k}-\frac{\left(1+\frac{1}{k}\right)}{\binom{N+k+1}{k}}\\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\Box \end{align*}

In (3) we get a telescoping sum leaving only the first and last term.

We will now apply identity (2) to answer OPs question. We use the representation from (4) with $k-1$ instead of $k$ and the claim follows for $k\geq 2$:

\begin{align*} \sum_{n=k+1}^{\infty}\frac{1}{\binom{n}{k}}&=\sum_{n=1}^{\infty}\frac{1}{\binom{n+k}{k}} =\lim_{N\rightarrow\infty}\sum_{n=1}^{N}\frac{1}{\binom{n+k}{n}}\\ &=\lim_{N\rightarrow\infty}\left(\frac{1}{k-1}-\frac{\left(1+\frac{1}{k-1}\right)(k-1)!}{(N+2)\cdots(N+k)}\right)\\ &=\frac{1}{k-1}\\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\Box \end{align*}

Note: These formulae can be found in H.W.Goulds Combinatorial Identities Vol. 3, (3.26 - 3.29).

Markus Scheuer
  • 112,413
2

Mathematica gives the sum in closed form: $$ \frac{\Gamma (k-1) \Gamma (k+2)}{(k+1) \Gamma (k) \Gamma (k+1)}. $$ If you ask it to FullSimplify[] it, it comes up with

$$\frac{1}{k-1}.$$

Here is the relevant command: Assuming[Element[k, Integers], FullSimplify[Sum[1/Binomial[n, k], {n, k + 1, Infinity}]]]

Igor Rivin
  • 26,372