Suppose $\{X_n\}_{n \geq 1}$ is a sequence of random variables which converges weakly to some random variable: $$ X_n \overset{w}{\longrightarrow} X $$ Question: what happens to the Cesaro sums of $X_n$ (assuming all moments exist)? $$ A_n := \frac{1}{n} \sum_{i=1}^n X_n \overset{w}{\longrightarrow} \; ? $$ At first I thought maybe $A_n \to \mathbb{E}X$ by writing $X_n = (X_n-X)+X$, but then something seemed fishy. I can't seem to grasp the intuition for whether this should converge to a constant (as opposed to random) limit.

  • $\begingroup$ Well why it could not converge to a constant r.v.? Think at what the law of large numbers says (although you need to check the assumptions required by that theorem as opppsed to the assumptions in your problem). $\endgroup$ – Kolmo Nov 9 '15 at 15:27
  • $\begingroup$ @Kolmo: well, it definitely can in certain cases, but I guess I'm wondering if there are cases where it doesn't. $\endgroup$ – gogurt Nov 9 '15 at 16:28
  • The sequence $(A_n)_{n\geqslant 1}$ may even be not tight. For example, define $X_n:=2^n\mathbf 1(E_n)$, where the sequence $(E_n)_{n\geqslant 1}$ is a sequence of independent events such that $\mu(E_n)=1/n$. Then for each $R$, $$\limsup_{n\to +\infty}\mu\{|A_n|\gt R\}\geqslant \limsup_{n\to +\infty}\mu\{2^{2n}/(2n)\bigcup_{i=n+1}^{2n}E_i \gt R\} \geqslant\limsup_{n\to +\infty}\mu\left(\bigcup_{i=n+1}^{2n}E_i\right)\geqslant 1/2$$ while $X_n\to 0$ in probability.

  • If $X$ is a random variable with symmetric distribution, then for each sequence $(\varepsilon_n)_{n\geqslant 1}$ of numbers among $-1$ and $1$, the sequence $\left(\varepsilon_n X=:X_n\right)_{n\geqslant 1}$ converges to $X$ in distribution. With some good choices of $\varepsilon_n$, we may have $A_n\to \alpha X$ in distribution, where $\alpha$ is a rational number in $[-1,1]$.


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.