3
$\begingroup$

Let $p\geq 1$ and consider the space $W^{1,p}(B)$ where $B\subset \mathbb{R}^{n}$ is the standard unit ball. Moreover, let $f_{k} \in C^{\infty}(B)$ be a Cauchy sequence in $W^{1,p}(B)$ of smooth function. How can one deduce that also $f_{k}^{+}$ is a Cauchy sequence in $W^{1,p}(B)$, where $f_{k}^{+}$ are defined by $$ f_{k}^{+}(x) = \text{max}\{f_{k}(x),0\}? $$

Greetings.

$\endgroup$
6
  • 9
    $\begingroup$ In other words, $F_:f\mapsto |f|$ is a continuous self-map of $W^{1,p}(B)$. One way to show it is approximating with $F_\epsilon:f\mapsto |f^2+\epsilon^2|^{1/2}$ and letting $\epsilon\to0$ $\endgroup$ Commented Dec 4, 2019 at 8:05
  • 3
    $\begingroup$ I think Pietro meant "positive part $f^+=\max\{f,0\}$", and not "absolute value $|f|$". Anyway, this is true because the composition $F(f)$ of $1$-Lipschitz maps $F$ satisfying $F(0)=0$ and Sobolev maps $f\in W^{1,p}$ (including $p=1$) is again Sobolev, with $\|F(f)\|_{W^{1,p}}\leq \|f\|_{W^{1,p}}$ $\endgroup$ Commented May 18, 2020 at 13:19
  • $\begingroup$ And as Pietro mentioned, the proof goes by approximation (of $F=\lim F_\epsilon$) $\endgroup$ Commented May 18, 2020 at 13:21
  • 3
    $\begingroup$ I have fixed a misprint in the title. Sobolev was as much a real person as Cauchy. $\endgroup$
    – markvs
    Commented Jan 3, 2021 at 19:13
  • 2
    $\begingroup$ More generally, the following is true: if $\phi:{\mathbb R}\rightarrow{\mathbb R}$ is globally Lipschitz, then $f\mapsto\phi\circ f$ maps $W^{1,p}(B)$ into itself continuously. In particular, if $(f_k)_k$ is Cauchy, then so is $(\phi\circ f_k)_k$. $\endgroup$ Commented Feb 2, 2021 at 21:01

1 Answer 1

1
$\begingroup$

You have $ \bigl\vert\vert f_{k+l}(x)\vert-\vert f_{k}(x)\vert\bigr\vert\le \vert f_{k+l}(x)-f_{k}(x)\vert $ and thus $$ \Vert\vert f_{k+l}\vert-\vert f_{k}\vert\Vert_{W^{1,1}} \le \Vert f_{k+l}- f_{k}\Vert_{L^{1}}+\Vert \nabla{\vert f_{k+l}\vert} -\nabla{\vert f_{k}\vert}\Vert_{L^{1}}. $$ Now, let us calculate $\nabla \vert f\vert$, say for $f\in C^1$. We claim that we have $$ \nabla \vert f\vert=S(f) \nabla f, \quad S(f)=\mathbf 1(f>0)-\mathbf 1(f<0). \label{1}\tag{1}$$ Note that the function $S(f)$ is bounded measurable and that the product makes sense, even if you have only $\nabla f\in L^1$. Let us prove our claim: with brackets of duality and $\phi\in C^\infty_c$, we have \begin{multline} \langle \nabla \vert f\vert, \phi\rangle=-\int\vert f\vert \nabla \phi dx=-\lim_{\epsilon \rightarrow 0_+}\int(f^2+\epsilon^2)^{1/2} \nabla \phi dx= \lim_{\epsilon \rightarrow 0_+}\int(f^2+\epsilon^2)^{-1/2} f(\nabla f) \phi dx \\ =\lim_{\epsilon \rightarrow 0_+}\int(f^2+\epsilon^2)^{-1/2} f(\nabla f) \phi dx =\int S(f) (\nabla f) \phi dx, \end{multline} proving \eqref{1}, using Lebesgue's Dominated Convergence Theorem. I feel a bit uncomfortable to extend this formula to $f\in W^{1,1}$, but it might be a first step.

$\endgroup$
2
  • $\begingroup$ This is about $\lvert f\rvert$, but I guess one can use it to handle $f^+ = \frac1 2(f + \lvert f\rvert)$. $\endgroup$
    – LSpice
    Commented Jan 3, 2021 at 19:46
  • 1
    $\begingroup$ If I am not mistaken, identity (1) for a general $f \in W^{1,1}$ is a straightforward application of the ACL characterisation of weakly differentiable functions, is it not? $\endgroup$ Commented Jan 3, 2021 at 22:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .