2
$\begingroup$

In a paper I am reading, the following result is used. If $(\mathcal{X},\mathcal{F}) $ is a measurable space, $\mathcal{G}$ is a class of functions with elements $f : \mathcal{X} \to \mathbb{R}$. We assume $\mathcal{G}$ is a bounded metric space. Then I wish to show that if

$$ \int_{0}^R\sqrt{H_B(u,\mathcal{G},L_2(P))}du < \infty $$

where $H_B(u,\mathcal{G},L_2(P))$ is the bracketing entropy, then

$$ \sup_{f \in \mathcal{G}} \Big| \frac{1}{n}\sum_{i=1}^nf(X_i)-\mathbb{E}(f(X)) \Big| \to 0 $$ in probability. Here, $R$ is a bound on the distance between functions in $\mathcal{G}$, i.e.

$$ \sup_{f_1,f_2 \in \mathcal{G}} d(f_1,f_2) < R. $$

I have looked various resources on empirical process theory, for example page 50 of http://www.stat.columbia.edu/~bodhi/Talks/Emp-Proc-Lecture-Notes.pdf#page50, but to no avail. The paper where it has been used is page 3 of https://projecteuclid.org/journals/supplementalcontent/10.1214/14-AOS1260/suppdf_1.pdf

$\endgroup$

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.