Let $\displaystyle\pi_{lr}\left(p\right) := {l \choose r}p^{r}\left(1 - p\right)^{l - r}\quad$ ( i.e., the binomial probability with parameters $\displaystyle l$ and $\displaystyle r$ ).

I'm interested in computing the following sum: $ \displaystyle\sum_{l = r}^{\infty}\pi_{lr}\left(p\right) $

I've two questions:

  1. Does this summation converge to a simple function of $\displaystyle p$ and $\displaystyle r$ ?.
  2. If I were to settle on approximating it with $\displaystyle\sum_{l = r}^{N}\pi_{lr}(p)$ for some $\displaystyle N$. How large should I choose $\displaystyle N$ ( as a function of $\displaystyle p$ and $\displaystyle r$ ) ?.
  • $\begingroup$ Could you clarify the second part of the second question ? $\endgroup$ – Claude Leibovici Feb 10 '19 at 3:39
  • $\begingroup$ sure, if the summation didn't admit a clean expression (such as $1/p$), I was planning to study it computationally. In which case, the obvious thing to do is to try to approximate the infinite sum with a finite one. However, one needs to choose a large enough $N$ so that this approximation is a good one. My second question was how to choose $N$. But it is now irrelevant, given the answer below. $\endgroup$ – Ozzy Feb 10 '19 at 3:54
  • $\begingroup$ Thank you for the answer ! It was just not clear to me; now, I understand. Cheers. $\endgroup$ – Claude Leibovici Feb 10 '19 at 3:57

$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$

$\ds{\pi_{\ell r}\pars{p} \equiv {\ell \choose r}p^{r}\pars{1 - p}^{\ell - r}.\qquad \sum_{\ell = r}^{\infty}\pi_{\ell r}\pars{p}:\ {\LARGE ?}}$.

\begin{align} &\bbox[10px,#ffd]{\sum_{\ell = r}^{\infty}\pi_{\ell r}\pars{p}} = \sum_{\ell = r}^{\infty}{\ell \choose r}p^{r}\pars{1 - p}^{\ell - r} = p^{r}\sum_{\ell = r}^{\infty}\overbrace{\ell \choose \ell - r} ^{\ds{=\ {\ell \choose r}}}\ \pars{1 - p}^{\ell - r} \\[5mm] = &\ p^{r}\sum_{\ell = r}^{\infty} \overbrace{{-r - 1 \choose \ell - r}\pars{-1}^{\ell - r}} ^{\ds{=\ {\ell \choose \ell - r}}}\ \pars{1 - p}^{\ell - r} \\[5mm] \stackrel{\ell\ -\ r\ \mapsto\ \ell}{=}\,\,\,& p^{r}\sum_{\ell = 0}^{\infty} {-r - 1 \choose \ell}\pars{-1}^{\ell}\pars{1 - p}^{\ell} \\[5mm] = &\ p^{r}\sum_{\ell = 0}^{\infty}{-r - 1 \choose \ell}\pars{p - 1}^{\ell} = p^{r}\,\bracks{1 + \pars{p - 1}}^{-r - 1} = \bbx{1 \over p} \end{align}

  • $\begingroup$ very nice, thank you! $\endgroup$ – Ozzy Feb 10 '19 at 0:32
  • $\begingroup$ @Ozzy You're welcome. $\endgroup$ – Felix Marin Feb 10 '19 at 0:35

As far as approximations goes, we have $$ \sum_{l=r}^\infty{l\choose r}p^r(1-p)^{l-r}=\left({p\over 1-p}\right)^r\sum_{l=r}^\infty{l\choose r}(1-p)^l $$ Write $a_l = {l\choose r}(1-p)^l.$ Then $${a_{l+1}\over a_l} = {l+1\over l+1-r}(1-p)\to 1-p$$ so that we essentially have a geometric series. Observing that ${l+1\over l+1-r}$ decreases to $1$ as $l\to\infty$ makes it easy to bound the omitted terms.

As to the first part of the question, I don't know, but I doubt it, except perhaps when $r$ is very small. Since the ratio of successive terms is a rational function of $l$ for fixed $p$ and $r$, this is a hypergeometric series. There is a large theory of these, which I unfortunately know nothing about. I've added the hypergeometric-function tag in hopes of attracting someone knowledgeable.

  • $\begingroup$ thank you for your thoughtful answer and also adding the additional tag! $\endgroup$ – Ozzy Feb 9 '19 at 23:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.