I am having a problem with proving convergence in distribution (or by law).

Consider that the sequence $X_n$ of random variables are IID and that $E[X_n]=0$ and $V[X_n]=1$.

Now define the variable $U_N$ as:

$$U_N= \frac{1}{\sqrt{N}}\sum_{n=1}^N X_n\cdot \sin\left(\frac{n\pi}{N}\right).$$

For $N\rightarrow \infty$, I want to show that $U_N$ converges by distribution, furthermore, I also want to determine the asymptotic distribution of $U_N$.

For the first part, I have tried to show convergence in probability, because this implies conv. in distribution (by law), but this was not possible since I dont have the asymptotic distribution of the $X_n$.

For the second part, I tried with the delta-method, but this did not work because of the summation of in the expression for $U_N$.

Does someone have an idea to this?

  • $\begingroup$ $E[U_N] = 0$. Do you know what $Var(U_N)$ is or at least what it converges to? $\endgroup$ – Henry Aug 13 '18 at 9:37
  • $\begingroup$ Var(U_N) should be 1/Nsum(sin(npi)/N), which also converges to zero for N->\infty $\endgroup$ – Jonathan Kiersch Aug 13 '18 at 9:45
  • $\begingroup$ I suspect $Var(U_N) \to \int\limits_0^1 \sin^2(\pi x)\, dx = \frac12$ $\endgroup$ – Henry Aug 13 '18 at 11:10
  • $\begingroup$ How did you find that? $\endgroup$ – Jonathan Kiersch Aug 13 '18 at 11:57
  • $\begingroup$ A combination of thinking that without the $\sin(\frac{n\pi}{N})$ term the variance would be $1$ as in the central limit theorem, plus simulation and basic calculus. $\endgroup$ – Henry Aug 13 '18 at 12:01

One can try to check Lindeberg's condition with $X_{N,i}:= X_i\sin\left(i\pi/N\right)$, using the following three facts:

  • the limits $\lim_{N\to +\infty}N^{-1}\sum_{n=1}^N\sin^2\left(n\pi/N\right)$ exists and is computable, as a limit of Riemann sums.
  • $s_N=\sum_{i=1}^N\operatorname{Var}\left(X_{N,i}\right)\sim c\sqrt N$.
  • $$\mathbb E\left[X_{N,i}^2\mathbf 1\left\{\left\lvert X_{N,i}\right\vert\gt\varepsilon s_N \right\}\right]=\mathbb E\left[X_1^2\sin^2\left(i\pi/N\right)\mathbf 1\left\{\left\lvert \sin\left(i\pi/N\right) X_{1}\right\vert\gt\varepsilon s_N\right\}\right]\leqslant\mathbb E\left[X_1^2\sin^2\left(i\pi/N\right)\mathbf 1\left\{\left\lvert X_{1}\right\vert\gt\varepsilon s_N \right\}\right].$$
  • $\begingroup$ I see, I was not familiar with the Lindeberg condition for CLT. So if I have understood you correct, I use the Lindeberg condition on X_n,N to find that U_N will converge to a standard normal distribution for N->infinity $\endgroup$ – Jonathan Kiersch Aug 13 '18 at 12:44
  • $\begingroup$ Yes, the limit will be a normal distribution but I have not looked whether the variance is one or not. $\endgroup$ – Davide Giraudo Aug 13 '18 at 13:20
  • $\begingroup$ @DavideGiraudo - the limiting variance is $\frac12$ (it would have been $1$ without the sine term) $\endgroup$ – Henry Aug 13 '18 at 14:04

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.