3
$\begingroup$

Suppose that $\mathbb H$ is a separable Hilbert space and $X_1,\ldots,X_n$ are independent zero mean $\mathbb H$-valued random elements such that $\|X_i\|\le s$ for each $1\le i\le n$, where $\|\cdot\|$ is the norm of $\mathbb H$. Denote $S_n=X_1+\ldots+X_n$.

Is it true that $$ P(\|S_n\|\ge t)\le Ce^{-c\frac{t^2}{ns^2}} $$ for $t>0$, where $C$ and $c$ are positive constants?

This reduces to Hoeffding's inequality if $\mathbb H=\mathbb R$. So basically what I am asking is if there exists a generalization of Hoeffding's inequality to Hilbert spaces. Unfortunately, I could not find a result of this sort.

Any help is much appreciated.

$\endgroup$
3
$\begingroup$

This is a special case of more general Theorem 3.5, which holds for martingales in $2$-smooth Banach spaces. Note that (i) any Hilbert space is $2$-smooth, with the $2$-smoothness constant $D=1$ and (ii) the consecutive sums of independent zero-mean random vectors constitute a martingale.

The inequality you need can also be easily deduced from the earlier result: Theorem 3 specifically for sums of independent zero-mean random vectors in a Hilbert space.

$\endgroup$

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.