Central limit theorem for varying random variables

202 Views Asked by At

I'm reading through Random Graphs by Janson, Luczak, Rucinski. There is an argument (proposition 1.15) that uses the result $$\Pr(X_n<M_n)\to 1/2$$ as $n\to\infty$, where $X_n\sim \mathrm{Bin}(N_n,M_n/N_n)$, and $N_n,M_n$ vary with $n$ such that $0\le M_n\le N_n$ and $M_n(N_n-M_n)/N_n\to \infty$.

The book appeals to the central limit theorem for this result, but I'm having trouble seeing why it follows. $X_n$ is a sum of $\mathrm{Be}(p_n:=M_n/N_n)$ random variables, but as far as I know the CLT needs $p_n$ to be constant. After some googling, the Berry Esseen theorem seems to give a bound that works, but I don't think it's what the authors intended. Is there a simple argument I'm missing?

(really, it suffices to prove that $\Pr(X_n<M_n)$ doesn't converge to 0 or 1, but I can't even see why that's true.)

1

There are 1 best solutions below

1
On BEST ANSWER

Indeed CLT holds and yields the result, as can be seen by considering $Y_n=(X_n-M_n)/s_n$ where $s_n^2=M_n(N_n-M_n)/N_n$ is the variance of $X_n$.

To see this, note that the hypothesis that $s_n\to\infty$ is exactly what is needed to guarantee, by an expansion of up to second order around zero, that $E[\mathrm e^{\mathrm itY_n}]\to\mathrm e^{-t^2/2}$ for every real $t$. Thus, $Y_n$ converges in distribution to a standard normal random variable $Y$, in particular $$ P[X_n\lt M_n]=P[Y_n\lt0]\to P[Y\lt0]=\tfrac12. $$