0
$\begingroup$

Suppose we have a coin with $0 < p < 1$ probability of giving heads when tossed. Let $X_n$ be the random variable denoting the number of heads we obtained from flipping $n$ of the described coins independently.

For any positive integer $k$, does the following holds? $$\Pr[X_n = m] \le \Pr[X_{kn} = km]$$

$\endgroup$
  • $\begingroup$ Inequality appears to depend on values of $n,m,k,p$.. $\endgroup$ – Arpan Mar 26 '15 at 12:26
  • $\begingroup$ @ArpanBanerjee Hm, so it doesn't only depends on $p$? For my research, I have that $m = E[X_n]$, $2 \le k < n^{0.5}$ and $p = 1/k$. Is there any related strategy to solve this one? $\endgroup$ – Irvan Mar 26 '15 at 12:48
2
$\begingroup$

Hint/idea: you can check whether this is true of false explicitly (even get a hunch on the dependence on $p$) by using the explicit formulae for Binomial distributions. What you are trying to compare are the two quantities $$ \binom{n}{m} p^m(1-p)^{n-m}, \qquad \binom{kn}{km} p^{km}(1-p)^{kn-km} $$ Your inequality is equivalent to checking whether $$ 1 \leq A(k, n,m, p) \stackrel{\rm def}{=} \frac{\binom{kn}{km}}{\binom{n}{m}} \left(p^{m}(1-p)^{n-m}\right)^{k-1} $$ If you are interested in the regime where $n\to\infty$ and $m=\mathbb{E}[X_n] = np$, you can approximate the RHS by Striling's approximation to get something easier to analyse as a function of $n,p,k$ only.

$\endgroup$
  • $\begingroup$ Stirling approximation is an excellent suggestion! To clarify we don't really need $m = E[X_n]$ to do that, right? (i mean, as long as $m$ is known). $\endgroup$ – Irvan Mar 27 '15 at 4:55
  • $\begingroup$ @Irvan: indeed, you just need to know something about $m$ (to take the asymptotics: i.e., how it compares to $n$.) $\endgroup$ – Clement C. Mar 27 '15 at 12:30

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.