5
$\begingroup$

Let $f_n(x)\geq0$ be any sequence of nonnegative $L^1(\mathbb{R}^n)$ functions such that $\int_{\mathbb{R}^{n}}f_n(x)dx=1$ where $dx$ is the Lebesgue measure on $\mathbb{R}^n$. For any $a>1,\epsilon>0$, does there exist a sequence $x_n\in\mathbb{R}^n$ such that $$\lim_{n\to+\infty}a^n\frac{\int_{\|x-x_n\|^2\leq n^{1-\epsilon}}f_n(x)dx}{\int_{\|x-x_n\|^2\leq n}f_n(x)dx}=0$$?

If it does not hold, is there any counter example of the sequence $f_n$? What additional conditions do we need on the sequence $f_n$? A similar question is to show that $$\lim_{n\to+\infty}a^n\frac{\int_{\|x-x_n\|^2\leq n^{1-\epsilon}}\exp(-\|x_n-x\|^2)f_n(x)dx}{\int_{\|x-x_n\|^2\leq n}\exp(-\|x_n-x\|^2)f_n(x)dx}=0$$ ?This can be implied by the first display. Therefore, this is a weaker statement.

$\endgroup$
2
$\begingroup$

Such a sequence $(x_n)$ always exists. Indeed, the displayed ratio expression under the limit sign equals $R_n(x_n)$, where \begin{equation*} R_n(y):=\frac{g_{n,r_n}(y)}{g_{n,s_n}(y)},\quad r_n:=n^{(1-\epsilon)/2}, \quad s_n:=n^{1/2}, \end{equation*} \begin{equation*} g_{n,r}(y):=\int f_n(x)1_{|x-y|_n<r}\,dx, \end{equation*} and $|\cdot|_n$ is the Euclidean norm in $\mathbb R^n$. Note that for all real $r>0$ \begin{equation*} \int g_{n,r}(y)\, dy=\int dx\, f_n(x)\int dy\,1_{|x-y|_n<r}=\int dx\, f_n(x)|B_n(r)|=|B_n(r)|, \end{equation*} the volume of any ball in $\mathbb R^n$ of radius $r$, whence \begin{equation*} \inf_y \frac{g_{n,r_n}(y)}{g_{n,s_n}(y)} \le\frac{\int g_{n,r_n}(y)\, dy}{\int g_{n,s_n}(y)\, dy}=\Big(\frac{r_n}{s_n}\Big)^n=n^{-\epsilon n/2}. \tag{1} \end{equation*} So, for some sequence $(x_n)$ (depending only on $\epsilon$ but not on $a$) with $x_n\in\mathbb R^n$ we have \begin{equation*} R_n(x_n)=\frac{g_{n,r_n}(x_n)}{g_{n,s_n}(x_n)}<\inf_y \frac{g_{n,r_n}(y)}{g_{n,s_n}(y)}+n^{-n} \le n^{-\epsilon n/2}+n^{-n} =o(1/a^n) \end{equation*} for all real $a>0$, so that $a^nR_n(x_n)\to0$, as desired.

Details on the inequality in (1): Let \begin{equation*} c:=\inf_y \frac{g_{n,r_n}(y)}{g_{n,s_n}(y)}. \end{equation*} Then $g_{n,r_n}(y)\ge c g_{n,s_n}(y)$ for all $y$, whence $\int g_{n,r_n}(y)\, dy\ge c\int g_{n,s_n}(y)\,dy$ and \begin{equation*} \frac{\int g_{n,r_n}(y)\, dy}{\int g_{n,s_n}(y)\,dy}\ge c=\inf_y \frac{g_{n,r_n}(y)}{g_{n,s_n}(y)}. \end{equation*}

$\endgroup$
  • $\begingroup$ Thank you for your answer! Could you please explain a little bit about the first inequality in the second to last display? The infimum of the ratio is smaller than the ratio of integral? $\endgroup$ – neverevernever Jun 8 at 0:52
  • 1
    $\begingroup$ @neverevernever : I have provided the details you requested. $\endgroup$ – Iosif Pinelis Jun 9 at 2:15

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.