Suppose that for some integer $p\geq 2$, we have

$$(\mathbb{E}[\lVert \mathbf{X} \rVert^p])^{\frac{1}{p}}\leq C$$ where $C>0$ is a constant as upper bound.

Now, my question is how to convert this into a high probability result? Specifically, I want to have the result like

with probability at least $1-\delta$, $$\lVert \mathbf{X} \rVert \leq D$$ Here, $D = f(C,\delta)$. Besides, hope the order of $\delta$ is $log$-order.

A simple method is use Markov inequality, then $D=C/\delta^{1/p}$, so is it possible to be tighter for a high probability bound?


Let $X$ be a real random variable such that $\mathbb{P}(X=0)=1-\delta$ and $\mathbb{P}(X=D)=\delta$ where $D=C/\delta^{1/p}$. Then we have $(\mathbb{E}(X^p))^{1/p}=C$ so one cannot hope to get a better bound than $C/\delta^{1/p}$ without any other assumptions.


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.