Let $X_1,X_2,\dots,X_n$ be independent random variables such that $\mathbf{E}(X_i)=0,\mathbf{E}(X_i^2)\leq 1$ for $1\le i\le n$ then show that $$\mathbf{E}\left(\max_{1\le i\le > n}(X_1+X_2+\dots+X_i)^2\right)\leq \dfrac{n(n+1)}{2}$$

Now what I did was, the maximum would be some $(X_1+X_2+\dots+X_j)^2$ so we need to compute $$\mathbf{E}((X_1+\dots+X_j)^2)=\sum\mathbf{E}(X_i)^2+\sum\mathbf{E}(X_kX_\ell)=\sum_{i=1}^{j}\mathbf{E}(X_i^2)\le j$$ because $\mathbf{E}(X_kX_\ell)=\mathbf{E}(X_k)\mathbf{E}(X_\ell)=0$ is due to independence. But then $$\mathbf{E}\left(\max_{1\le i\le n}(X_1+X_2+\dots+X_i)^2\right)\le \max_{1\le j\le n}j=n$$ Which reduces the bound considerably. So, I am a bit worried this proof is flawed, can someone check it? Thanks a lot.

  • $\begingroup$ Note $E[\max X_i] \ge \max E[X_i]$ and so on. $\endgroup$ – Macavity Feb 11 '15 at 14:55

Your attempt is complete because you proved that $$\max_{1\leqslant i\leqslant n}\mathbb E\left[\left(\sum_{j=1}^i X_j\right)^2 \right] \leqslant n,$$ but in general, for non-negative random variables $Y_i$, $\max_{1\leqslant i\leqslant n} \mathbb EY_i\lt \mathbb E\left[\max_{1\leqslant i\leqslant n} Y_i \right] $.

In this case, however, we can notice that $$\max_{1\leqslant i\leqslant n}\left(\sum_{j=1}^i X_j\right)^2\leqslant \sum_{i=1}^n\left(\sum_{j=1}^i X_j\right)^2,$$ then take the expectation and use the fact that the $X_j$'s are uncorrelated.


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.