The Bourgain-Brézis-Mironescu formula in arbitrary bounded domains

Irene Drelichman and Ricardo G. Durán

arXiv: 2012.14505v2 · PDF · 6 January 2021
Categories: math.AP
Licence: CC BY 4.0 The site shows the text.

Abstract

We obtain a Bourgain-Brézis-Mironescu formula on the limit behaviour of a modified fractional Sobolev seminorm when s\nearrow 1, which is valid in arbitrary bounded domains. In the case of extension domains, we recover the classical result.

Items 10

Theorem

Theorem 1.1

POUC

Let \Omega \subset \mathbb{R}^n be a bounded domain, d(x)=\mbox{dist}(x, \partial\Omega), and \tau \in (0,1). If 1<p<\infty and f\in L^p(\Omega),

\lim_{s\nearrow 1} (1-s) \int_\Omega \int_{|x-y|<\tau d(x)} \frac{|f(y)-f(x)|^p}{|x-y|^{n+sp}} \, dy \, dx = K_{n,p} \int_\Omega |\nabla f (x)|^p \, dx,

where K_{n,p}= \frac{1}{p} \int_{\mathbb{S}^{n-1}} |\sigma_n|^p \, d\sigma, with the convention that \|\nabla f\|_p =\infty if f\not\in W^{1,p}(\Omega).

Used by Remark 1.1

Remark

Remark 1.1

1EWL

When |x-y|<\tau d(x) for \tau \in (0,1), it is clear that the geometric distance equals the euclidean distance, that is, \delta(x,y)=|x-y|. Therefore, Theorem 1.1 and the obvious inequality

\begin{aligned} \lim\inf_{s\nearrow 1} (1-s)&\int_\Omega \int_{|x-y|<\tau d(x)} \frac{|f(x)-f(y)|^p}{\delta(x,y)^{n+sp}} \, dx \, dy \\ &\le \lim\sup_{s\nearrow 1} (1-s)\int_\Omega \int_\Omega \frac{|f(x)-f(y)|^p}{\delta(x,y)^{n+sp}} \, dx \, dy \end{aligned}

show that the answer to the first part of [B, Open problem 1] is positive.

Also, in view of Theorem 1.1, the second part of the problem is true if and only if

\lim_{s\nearrow 1} (1-s) \int_\Omega \int_{|x-y|>\tau d(x)} \frac{|f(y)-f(x)|^p}{\delta(x,y)^{n+sp}} \, dy \, dx =0.

Uses Theorem 1.1

Lemma

Lemma 2.1

H0U8

For 1<p<\infty, h\in L^p(\mathbb{R}^n), h\ge 0, and \nu a unit vector, let

h_1(x,\nu) = \sup_{t>0} \frac{1}{t} \int_0^t h(x+\nu s) \, ds

and

h^*(x) = \left(\int_{\mathbb{S}^{n-1}} h_1(x,\nu)^p \, d\sigma_\nu \right)^\frac{1}{p}.

Then, h^* \in L^p and \|h^*\|_p\le C \|h\|_p.

Used by Proof

Lemma

Lemma 2.2

UHVV

Let \Omega \subset \mathbb{R}^n be a bounded domain,

F_s(x)= \int_{|h|<\tau d(x)} \frac{|f(x+h)-f(x)|^p}{|h|^{n+sp} } \, dh,

and

|f|_{\widetilde W^{s,p}(\Omega)} = \left( \int_\Omega F_s(x) \, dx\right)^\frac{1}{p}.

Then, for 1<p<\infty, we have the pointwise bound

(1-s) F_s(x) \le C |\nabla f|^*(x)^p.

In particular,

(1-s) |f|_{\widetilde W^{s,p}(\Omega)}^p \le C \| \nabla f\|_{p}^p,

which means W^{1,p}(\Omega) \subset \widetilde{W}^{s,p}(\Omega).

Used by Proof, Proof

Proof of Lemma 2.2

By the Meyers-Serrin theorem [MeSe], we may assume that f is smooth. Then, if \sigma=\frac{h}{|h|} and r=|h|,

f(x+h)-f(x) =\int_0^1 \nabla f(x+th) \cdot h \, dt = \int_0^r \nabla f(x+s\sigma) \cdot \sigma \, ds

which implies that, extending |\nabla f| by zero,

\begin{aligned} (1-s) F_s(x) &= (1-s) \int_{|h|<\tau d(x)} \left| \int_0^r \nabla f (x+s\sigma) \cdot \sigma \, ds \right|^p \frac{1}{|h|^{n+sp}} \, dh\\ & \le (1-s) \int_0^{\tau d(x)} \int_{\mathbb{S}^{n-1}} \left( \int_0^r |\nabla f (x+s\sigma)| \, ds \right)^p \, d\sigma \, \frac{1}{r^{sp+1}} \, dr\\ & = (1-s) \int_0^{\tau d(x)} \int_{\mathbb{S}^{n-1}} \left( \frac{1}{r} \int_0^r |\nabla f (x+s\sigma)| \, ds \right)^p \, d\sigma \, r^{p-sp-1} \, dr\\ & \le (1-s) \int_0^{\tau d(x)} |\nabla f|^* (x)^p \, r^{p-sp-1} \, dr\\ &= \frac{(\tau d(x))^{p(1-s)}}{p} |\nabla f|^*(x)^p. \end{aligned}

Finally, by Lemma 2.1, we obtain

(1-s) |f|_{\widetilde W^{s,p}(\Omega)}^p \le C \|\nabla f\|_p^p.

Uses Lemma 2.1

Proof

V0VV

Since x is fixed, we may assume, by rotation, that the vector \nabla f(x) has only its n-th coordinate different from zero. Then,

\begin{aligned} \int_{|x-y|< \tau d(x)} \left(\left| \nabla f(x) \cdot \frac{x-y}{|x-y|^s}\right|\right)^p \frac{1}{|x-y|^n}\, dy &= \int_0^{\tau d(x)} \int_{\mathbb{S}^{n-1}} |\nabla f(x) \cdot \sigma|^p r^{(1-s)p} \, d\sigma \frac{dr}{r}\\ &= \int_0^{\tau d(x)} \int_{\mathbb{S}^{n-1}} |\nabla f(x)|^p |\sigma_n|^p r^{(1-s)p} \, d\sigma \frac{dr}{r}\\ &= K_{n,p} \, |\nabla f(x)|^p \, \frac{(\tau d(x))^{(1-s)p}}{(1-s)} \end{aligned}

and, therefore

\lim_{s \nearrow 1} (1-s) \int_{|x-y|< \tau d(x)} \left(\left| \nabla f(x) \cdot \frac{x-y}{|x-y|^s}\right|\right)^p \frac{1}{|x-y|^n} \, dy = K_{n,p} |\nabla f(x)|^p

Proof

6P23

By the previous step, it suffices to show that

\begin{aligned} \lim_{s \nearrow 1} (1-s) & \int_{|x-y|< \tau d(x)} \frac{|f(x)-f(y)|^p}{|x-y|^{sp}} \frac{1}{|x-y|^n} \, dy \\ &= \lim_{s \nearrow 1} (1-s) \int_{|x-y|< \tau d(x)} \left(\left| \nabla f(x) \cdot \frac{x-y}{|x-y|^s}\right|\right)^p \frac{1}{|x-y|^n} \, dy. \end{aligned}

Now, since the limit of the right-hand side exists, it suffices to show that

\lim_{s \nearrow 1} (1-s) \int_{|x-y|< \tau d(x)} \left| \left(\frac{|f(x)-f(y)|}{|x-y|^s}\right)^p - \left( \left| \nabla f(x) \cdot \frac{x-y}{|x-y|^s}\right| \right)^p \right| \frac{1}{|x-y|^n} \, dy =0.

For fixed x and all y such that |x-y|<\tau d(x), since taking p powers is locally Lipschitz, we have that

\begin{aligned} \left| \left(\frac{|f(x)-f(y)|}{|x-y|^s}\right)^p - \left( \left| \nabla f(x) \cdot \frac{x-y}{|x-y|^s}\right| \right)^p \right| & \le C \frac{|f(x)-f(y)-\nabla f(x)\cdot (x-y)|}{|x-y|^s}\\ &\le C |x-y|^{2-s} \end{aligned}

(where the constant depends on x).

Combining this bound with

\int_{|x-y|<\tau d(x)} |x-y|^{2-s-n} \, dy = \omega_n \int_0^{\tau d(x)} r^{1-s} \, dr = \frac{\omega_n}{2-s} (\tau d(x))^{2-s}

and

\lim_{s\nearrow 1} (1-s) \int_{|x-y|<\tau d(x)} |x-y|^{2-s-n} \, dy = 0,

we obtain the desired equality.

Proof

9533

By dominated convergence, it suffices to show that there exists an integrable g such that (1-s)|F_s(x)|\le g(x) with

F_s(x) =\int_{|x-y|< \tau d(x)} \frac{|f(x)-f(y)|^p}{|x-y|^{n+sp}} \, dy,

but this follows immediately by Lemma 2.2, taking g=(|\nabla f|^*)^p.

Uses Lemma 2.2

Proof

947D

This follows by density. Namely, we consider (f_k)_{k\in \mathbb{N}} \subset C^2(\Omega)\cap W^{1,p}(\Omega) such that f_k\to f in W^{1,p}(\Omega). Then,

\begin{aligned} (1-s) |f|_{\widetilde W^{s,p}}^p - K_{n,p} \|\nabla f\|_p^p &\le (1-s) \Big| |f|_{\widetilde W^{s,p}}^p - |f_k|_{\widetilde W^{s,p}}^p \Big| \\ &+ \Big| (1-s) |f_k|_{\widetilde W^{s,p}}^p - K_{n,p} \|\nabla f_k\|_p^p \Big| \\ &+ K_{n,p} \Big| \|\nabla f_k\|_p^p - \|\nabla f\|_p^p \Big|. \end{aligned}

But, since f_k \to f in W^{1,p}(\Omega), by Lemma 2.2 the first and last terms in the right hand-side of the previous inequality converge to zero. Finally, by Step 3,

\lim_{s\nearrow 1} (1-s) |f_k|_{\widetilde W^{s,p}}^p = K_{n,p} \|\nabla f_k\|_p^p,

and we obtain (?).

Uses Lemma 2.2

Proof

QWXQ

Consider an exhaustion \{\Omega_j\}_{j \in \mathbb{N}} of smooth bounded domains such that

\overline\Omega_j\subset \Omega_{j+1} \, \mbox{ and } \, \bigcup_{j=1}^\infty \Omega_j =\Omega.

Then, there exist a decreasing sequence of positive numbers \{\alpha_j\}_{j\in \mathbb{N}} such that, for all x \in \Omega_j, d(x)=dist(x, \partial \Omega)\ge \alpha_j >0.

We want to apply the Bourgain-Brézis-Mironescu formula (?) to the sets \Omega_j and prove that the limit is a finite quantity independent of j. To this end, we write

\begin{aligned} \int_{\Omega_j} \int_{\Omega_j} \frac{|f(y)-f(x)|^p}{|x-y|^{n+sp} } \, dy \, dx &= \int_{\Omega_j} \int_{\{y\in \Omega_j : |x-y|<\tau d(x)\}} \frac{|f(y)-f(x)|^p}{|x-y|^{n+sp} } \, dy \, dx\\ & + \int_{\Omega_j} \int_{\{y\in \Omega_j : |x-y|\ge\tau d(x)\}} \frac{|f(y)-f(x)|^p}{|x-y|^{n+sp} } \, dy \, dx\\ &= {\Large{\textcircled{\small I}}} +{ \Large{\textcircled{\small II}} } \end{aligned}

Clearly,

{\Large{\textcircled{\small I}}} \le \int_{\Omega} \int_{|x-y|<\tau d(x)} \frac{|f(y)-f(x)|^p}{|x-y|^{n+sp} } \, dy \, dx

so that, by hypothesis,

\liminf_{s \nearrow 1} (1-s) \, {\Large{\textcircled{\small I}}} \le \widetilde{A}.

To bound II, we use that d(x)\ge \alpha_j>0 for all x\in \Omega_j and obtain

\begin{aligned} {\Large{\textcircled{\small II}}} &\le \int_{\Omega_j} \int_{\{y\in \Omega_j : |x-y|\ge \tau \alpha_j\}} \frac{|f(y)-f(x)|^p}{|x-y|^{n+sp} } \, dy \, dx\\ &\le \frac{1}{(\tau \alpha_j)^{n+sp}} \int_{\Omega_j} \int_{\Omega_j} |f(y)-f(x)|^p \, dy \, dx\\ &\le \frac{2^{p+1} |\Omega_j|}{(\tau \alpha_j)^{n+sp}} \|f\|_p^p \end{aligned}

so that,

\liminf_{s \nearrow 1} (1-s) \, {\Large{\textcircled{\small II}}} =0.

Therefore, since \liminf_{s \nearrow 1} (1-s) |f|_{W^{s,p}(\Omega_j)}\le \widetilde A, by (?) we have that f \in W^{1,p}(\Omega_j) for each j and that the seminorms are uniformly bounded. Hence, f \in W^{1,p}(\Omega), as we wanted to see.

Source and licence

The text on this page is from arXiv:2012.14505v2 by Irene Drelichman and Ricardo G. Durán, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2012/2012.14505.jsonl.