Let $X_1,\cdots,X_n$ be i.i.d. random variables with density $f$ and let $\hat{f}$ be an estimator of $f$. Is the following inequality direct from the standard properties of expectation and sup-norm?

$$\mathbb{E}\left[ \lVert \hat{f}- \mathbb{E}(\hat{f})\rVert_\infty + \lVert \mathbb{E}(\hat{f})-f\rVert_\infty \right]^2 \leq 2\mathbb{E}\lVert \hat{f}- \mathbb{E}(\hat{f})\rVert_\infty^2 + 2\lVert \mathbb{E}(\hat{f})-f\rVert_\infty^2.$$

This inequality is found in page 44 of Tsybakov's book.

  • $\begingroup$ You are right. I missed to apply the expectation. $\endgroup$ – Celine Harumi Mar 24 at 12:36

Since$$(a+b)^2=a^2+2ab+b^2\le a^2+\overbrace{(a^2+b^2)}^{\ge 2ab}+b^2=2a^2+2b^2$$ $$\mathbb{E}\left[ \lVert \hat{f}- \mathbb{E}(\hat{f})\rVert_\infty + \lVert \mathbb{E}(\hat{f})-f\rVert_\infty \right]^2 \leq 2\Bbb E[\lVert \hat{f}- \mathbb{E}(\hat{f})\rVert_\infty]^2 + 2\lVert \mathbb{E}(\hat{f})-f\rVert_\infty^2.$$

  • $\begingroup$ Can you show me why 2ab need to be less than or equal the sum of squares? This is exactly what I was struggled with. $\endgroup$ – Celine Harumi Mar 24 at 12:48
  • 1
    $\begingroup$ Hint:$$a^2+b^2\ge 2ab$$is equivalent to$$a^2+b^2-2ab\ge 0$$and then... $\endgroup$ – Xi'an Mar 24 at 13:53

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.