Construct series of independent random variables all with different distribution with convergence to standard normal.

85 Views Asked by At

I'm asked to construct a series of independent random variables $X_k, k \in \mathbb{N}$ all with different distributions, such that

$$N^{-\frac{1}{2}}\sum_{k=1}^NX_k \rightarrow_d X \sim \mathcal{N}(0,1)$$

as $N \rightarrow \infty$. I thought about having them all normal-distributed with expectation 0 and different variance. That the sum of every $2^k, k\in \mathbb{N}$ random variables is again standard normal distributed, but I'm unsure on how to chose the variance. Since they all distributions have to be different.

1

There are 1 best solutions below

0
On BEST ANSWER

The variance of the sum of independent random variables is the sum of their variances. So choosing the variances of the $\ X_k\ $ to converge to 1 sufficiently fast as $\ N\rightarrow\infty\ $should work. If you set $\ \mathrm{Var}\left(X_1\right)=\frac{3}{2}\ $ and $\ \mathrm{Var}\left(X_k\right)=1 + \left(2^{-k}-2^{-(k-1)}\right)\ $ for $\ k\ge 2\ $, for instance, then $\ \mathrm{Var}\left(N^{-\frac{1}{2}}\sum_\limits{k=1}^NX_1\right)=1 + \frac{1}{N2^N}\ $.