7

I'm studying section 3.3 of Analysis by Lieb and Loss, about symmetric-decreasing rearrangement of functions.

Let $A\subset \mathbb{R}^n$ a Borel set of finite Lebesgue measure. They define $A^*$ to be the ball centered at 0 with the same measure that $A$.

The symmetric-decreasing rearrangement of a measurable function $f:\mathbb{R}^n \to \mathbb{R}$ is then defined by

$$f^*(x):=\int_0^{\infty} \chi_{\{|f|>t\}^*}(x)dt,$$

by comparison to the "layercake" representation of $f$, namely $$f(x)=\int_0^{\infty} \chi_{\{f>t\}}(x)dt.$$

They say that it is then an obvious property that

$$\{x: f^*(x)>t\}=\{x: |f(x)|>t\}^* .$$

But I can't see why/how...

Guy Fsone
  • 25,237
Klaus
  • 4,265

3 Answers3

8

Fix $t>0$ et $y\in \left\{x \in \mathbb{R}^n: |f(x)| > t \right\}^{*}$. This is equivalent to say that $$c_d|y|^d<|\{|f|>t\}|,\qquad \qquad c_d=|B(0,1)|. $$ Assume that $$\int_{t}^{+\infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds=0.$$ Then $y\not \in \{|f| > s\}^{*}$ for all $s\geq t$. That is, for all $s\geq t$ we have $$c_d|y|^d\geq |\{|f|>s\}| $$ The Fatou lemma implies $$c_d|y|^d\geq |\{|f|\geq t\}|\geq |\{|f|>t\}|.$$ This is in contradiction with the fact that $c_d|y|^d<|\{|f|>t\}|$. Therefore we deduce $$\int_{t}^{\infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds>0.$$ On the other hand, one can check that for every, $0<s< t$ one has $$\left\{x \in \mathbb{R}^n: |f(x)| > t \right\}\subset \left\{x \in \mathbb{R}^n: |f(x)| > s \right\}$$ this entails that, \begin{equation}\label{eq-inclu t-s}\tag{I} \left\{x \in \mathbb{R}^n: |f(x)| > t \right\}^{*}\subset \left\{x \in \mathbb{R}^n: |f(x)| > s \right\}^{*}~~~\textrm{for all $s\in (0,t)$}. \end{equation} this implies that,$$ \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) =1 ~~~s\in (0,t)$$

From definition of $f^{*}$, if $y\in \{|f|>t\}^*$ then we have $$\begin{align*} f^{*}(y) &:= \int_{0}^{+ \infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds\\ &= \int_{0}^{t} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds+ \int_{t}^{+\infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds\\ & = \int_{0}^{t} ds+\int_{t}^{+\infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds >t. \end{align*}$$

Whence, $$\left\{x \in \mathbb{R}^n: |f(x)| > t \right\}^{*}\subset \left\{ x \in \mathbb{R}^n:f^{*}(x)> t \right\}.$$ Conversely, suppose, $y\notin \left\{x \in \mathbb{R}^n: |f(x)| > t \right\}^{*}$. If $t\leq s$ then analogously to relation \eqref{eq-inclu t-s} we get $$y\notin \left\{x \in \mathbb{R}^n: |f(x)| > s \right\}^{*}.$$ That is for all $s\geq t$ we get $$\mathbf{1}_{\{ | f| > s\}^*}(y)=0$$ Hence necessarily for all $s>0$ such that $ y\in \left\{x \in \mathbb{R}^n: |f(x)| > s \right\}^{*}$ one has $0<s\leq t$. This means that, $$\sup\left\{s>0 : y\in \left\{x \in \mathbb{R}^n: |f(x)| > s \right\}^{*}\right\}\leq t. $$ We then deduce that, $$\begin{align*} f^{*}(y) &:= \int_{0}^{+ \infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds\\ &= \int_{0}^{t} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds+ \underbrace{\int_{t}^{+\infty} \mathbf{1}_{\left\{ | f| > s \right\}^*}(y) ds}_{=0}\leq t \end{align*}$$ that is $f^*(y)\leq t$ or that $y\notin \left\{x \in \mathbb{R}^n: f^*(x) > t \right\}$. We've just prove that,

\begin{equation}\label{eq}\tag{II} \Bbb R^n\setminus \left\{x \in \mathbb{R}^n: |f(x)| > t \right\}^{*}\subset \Bbb R^n\setminus\left\{x \in \mathbb{R}^n: f^*(x) > t \right\}. \end{equation} Which ends the prove by taking the complementary.

Guy Fsone
  • 25,237
  • 2
    Nice answer. I am amused that the authors see this as "obvious" (of course it is if you look at a picture...). – lasik43 Nov 19 '19 at 10:35
  • 1
    I am delighted that someone finally finds this answer useful – Guy Fsone Nov 20 '19 at 11:36
  • It looks like that you can only get $f^(y)\ge t$ in your first part, and you need more explanation to show $f^(y)\neq t$? – IMOS Mar 27 '23 at 03:11
  • https://math.stackexchange.com/questions/1997959/on-rearrangement-of-level-set-ft-ft-text?rq=1 – Guy Fsone Jan 08 '24 at 11:59
  • Thank you @GuyFsone for the very clear solution. I didn't understand the final step. How to conclude that $\int_0^t \chi_{{|f| > s}^\ast} (y) ds \leq t$ ? Is it true just because $\chi_{{|f| > s}^\ast} (y) \leq 1 ?$ Or do we have to use the supremum ? – ThiagoGM Feb 21 '24 at 00:03
  • @GuyFsone do you know where can I find a proof for the question in the link https://math.stackexchange.com/questions/4866925/inequality-involving-the-l2-norm-of-the-gradient-of-the-symmetric-decreasing ? – ThiagoGM Feb 21 '24 at 12:33
4

Hint 1: show that if $t_1 > t_2$, then $\{ |f| > t_1\}^* \subseteq \{ |f| > t_2\}^*$.

Hint 2: use this to show that if $y\in \{ |f| > t \}^*$, then $y \in \{|f| > s\}^*$ for every $0 \leq s \leq t$. Notice that this implies that $f^*(y) \geq t$ by the definition.

Hint 3: use hint 1 again to show that if $y\not\in \{|f| > t\}^*$, $$ \sup \left\{ s \geq 0 ~~|~~ y \in \{|f| > s\}^*\right\} \leq t $$ this implies in particular $f^*(y) \leq t$ (why?).

Willie Wong
  • 75,276
  • hint 3 is unnecessary. You already proved that the level set of $f^{*}$ is subset of the star set. – Thomas Kojar Feb 10 '14 at 23:48
  • It seems that you only showed ${ f^>t} \subset { |f|>t}^$. But how to show the another side. – IMOS Mar 27 '23 at 10:28
  • @IMOS: Reading literally what I wrote 11 years ago: Hint 2 leads to "${ |f| > t }^* \subseteq { f^* \geq t }$" // Hint 3 is meant to lead to the other direction (the one you mentioned in your comment) // I do not understand the comment by TKM. – Willie Wong Mar 27 '23 at 13:01
  • @WillieWong Yes, I found that you showed ${|f|>t}^\subset { f^\ge t}$. And we need more explanation to show that ${|f|>t}^\subset { f^\ge t} \subset {f^>t}$. I just found this can be done by the continuous of Lebesgue measure. (By showing ${f=t} is of zero measure$. Thanks for your reply. – IMOS Mar 28 '23 at 03:11
0

I have been working on this subject for some months now and your question was one of the main obstacles I encountered. I think what makes this particular problem hard to attack is that it seems so obvious and geometrically correct that most would prefer not to bother with a formal proof. Anyway, what worked for me was proving first that $$ r(t) = C_n\sqrt[n]{\mathcal{L}^n\left(\{ |f| > t \}\right)} $$ where $C_n = \left( \frac{n\Gamma(n/2)}{2\pi^{n/2}} \right)^{\frac{1}{n}}$, is a lower semicontinuous function. Then you can prove that the parameter $t$ in the integral which defines $f^*$ will define a set that contains $x$ just in an OPEN interval. Showing that it is an interval is pretty easy, but the open part is where you need the lower semicontinuity of the function $r$.

Best regards!