1
$\begingroup$

Can any one give me an example of random vector $X = (X_1,\ldots,X_n)$ such that $$||X||_{\psi_2} \gg \max_i||X_i||_{\psi_2}.$$ Here $||X||_{\psi_2} = \sup_{x\in S^{n-1}}||\langle X,x\rangle||_{\psi_2}$ and $||X_i||_{\psi_2} = \inf\left\{t>0:E\left(\exp\left(X_i^2/t^2\right)\leq2\right)\right\}$. This is a question from high-dimensional probability and what I know is that if the coordinates of $X$ are independent then $||X||_{\psi_2} \leq C \max_i||X_i||_{\psi_2}$ for some constant $C$.

Thanks in advance!

$\endgroup$
1
$\begingroup$

Let $X_i:=Y$ for all $i\in\left\{1,\dots,n\right\}$. Then \begin{align} \left\lVert X\right\rVert_\psi&=\sup_{x\in S^{n-1}} \left\lVert \left\langle x,X\right\rangle\right\rVert_\psi\\ &=\sup_{x\in S^{n-1}} \left\lVert \sum_{i=1}^n x_iY \right\rVert_\psi\\ &= \sup_{x\in S^{n-1}}\left\lvert \sum_{i=1}^n x_i\right\rvert \left\lVert Y \right\rVert_\psi \end{align} and since $\sup_{x\in S^{n-1}}\left\lvert \sum_{i=1}^n x_i\right\rvert $ is achieved for $x_i=1/\sqrt n$, we get $$ \left\lVert X\right\rVert_\psi\geqslant \sqrt n\max_{1\leqslant i\leqslant n}\left\lVert X_i \right\rVert_\psi. $$

$\endgroup$

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.