Let $\mathscr F$ be the collection of smooth functions $f \colon \mathbb R \to \mathbb R$ such that

  1. $f \in C^\infty_c(\mathbb R)$, with $\text{supp } f \subset [-1,1]$;
  2. $\int_0^1 x f(x) dx = \frac{1}{2\pi}$.

I would like to compute $$ \inf_{\mathscr F} \int_0^1 \vert f^\prime (x) \vert x^2 \, dx. $$

My (last) hope is that this infimum is $0$ but I do not manage neither to prove it nor to disprove it. The point is that the functional I am trying to minimize is invariant under rescaling, in the sense that given $f \in \mathscr F$ the functional computed on $f$ and on $f_\varepsilon(x) = \varepsilon^{-3} f(\frac{x}{\varepsilon})$ (for every $\varepsilon>0$) has the same value. What is this telling?


$\newcommand{\al}{\alpha} \newcommand{\de}{\delta} \newcommand{\De}{\Delta} \newcommand{\ep}{\varepsilon} \newcommand{\ga}{\gamma} \newcommand{\Ga}{\Gamma} \newcommand{\la}{\lambda} \newcommand{\Si}{\Sigma} \newcommand{\thh}{\theta} \newcommand{\om}{\omega} \newcommand{\R}{\mathbb{R}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\F}{\mathcal{F}} \newcommand{\E}{\operatorname{\mathsf E}} \renewcommand{\P}{\operatorname{\mathsf P}} \newcommand{\ii}[1]{\operatorname{\mathsf I}\{#1\}} \newcommand{\tf}{\widetilde{f}}$

This infimum is $\frac1\pi$. Indeed, integrating by parts, for any $f\in\F$ we have \begin{equation} \frac1{2\pi}=I(f):=\int_0^1 x f(x) dx =-J(f)/2,\quad J(f):=\int_0^1 x^2 f'(x) dx, \end{equation} whence \begin{equation} H(f):=\int_0^1 x^2 |f'(x)| dx\ge|J(f)|=\frac1\pi, \end{equation} so that the infimum in question is $\ge\frac1\pi$.

On the other hand, for any $\ep\in(0,1)$ let \begin{equation} f_\ep:=c_\ep g_\ep,\quad g_\ep(x):=\frac x\ep\,1_{(0,\ep]}(x)+\frac{1-x}{1-\ep}\,1_{(\ep,1)}(x), \end{equation} where $c_\ep$ is such that $I(f_\ep)=\frac1{2\pi}$, so that $c_\ep\to3/\pi$; the convergence everywhere here is for $\ep\downarrow0$. Then $H(f_\ep)\to\frac1\pi$. It remains to approximate in $L^1[0,1]$, however closely, both $f_\ep$ and $f'_\ep$ by $f$ and $f'$ for some $f\in\F$; this can be done by using, say, the convolution of $f_\ep$ (or, rather, of a close version of $f_\ep$ with support on $(0,1)$) with a smooth kernel and the multiplication by a constant factor close to $1$. So, the infimum in question is indeed $\frac1\pi$.

  • $\begingroup$ Dear Prof. Iosif Pinelis, I do thank you for your answer. Your help in these days has been extremely valuable to me and I therefore thank you. +1, of course $\endgroup$ – Y.B. May 30 '18 at 20:33
  • $\begingroup$ I am glad this was helpful. $\endgroup$ – Iosif Pinelis May 30 '18 at 21:59

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.