This is a very interesting conjecture of large scale property of hypergeometric distribution.

Let $a>1$ be a integer constant, $N\in\mathbb{N_+}$, for any $x<N-1$, consider $N+(a-1)x$ balls in a bag, in which $ax$ of them are colored white and the other $N-x$ are colored black. Now randomly select $N$ balls without replacement from the bag. Let random variable $X$ denote the number of white balls selected. For another $y=x+1$, we also select $N$ balls from a bag with $N+(a-1)y$ balls, in which $ay$ are colored white, then we get a similar random variable $Y$.

In short, $X \sim H(N+(a-1)x,ax,N)$ and $Y \sim H(N+(a-1)y,ay,N)$, with $y=x+1$.

We want to prove that $\mathbb{E}(\frac{xy}{X}-\frac{xy}{Y})\rightarrow \frac{1}{a}$, when $N\rightarrow\infty$. The convergence is uniform w.r.t to $x$, namely, the convergence rate is independent of $x$. (For large enough $N$, the expectation can be arbitrarily close to $1/a$, regardless of which $x<N-1$ we choose. See the comment of the first answer.)

It is likely that the convergence must use a suitable coupling of $X$ and $Y$.


In your setting, when $N\to\infty$ we have $\mathbb{P}(X=ax)\to 1$ and so $\mathbb{E}(\frac{xy}{X})\to \frac{y}{a}$. Similarly, $\mathbb{E}(\frac{xy}{Y})\to \frac{x}{a}$, which together with $y=x+1$ yield the desired result.

The convergence rate cannot be independent of $x$ as the model is only defined when $N>x$.

  • $\begingroup$ Thank you for your answer. Here independence means, $\forall \epsilon>0$, there is an number $N(\epsilon)$, such that $\forall N>N(\epsilon)$, we have $|\mathbb{E}(xy/X-xy/Y)-1/a|<\epsilon$. Note that $\mathbb{E}(\frac{xy}{X})$ may tend to infinity, because $x$ can tend to infinity when $N$ grows, so this answer is not correct :-) $\endgroup$ – user2173168 Feb 3 '15 at 13:09

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.