Bourgain-Brezis-Mironescu formula for Riesz Potentials
Extracted source text ยท arXiv 2604.06827
\documentclass[reqno, 11pt, oneside]{amsart}
\usepackage{latexsym, amsmath,amssymb}
%\usepackage{latexsym,amsmath,amssymb}
%\usepackage[abbrev,backrefs]{amsrefs}
\usepackage{hyperref}
\hypersetup{colorlinks=true, citecolor=cyan, linkcolor=blue}
\usepackage[initials, alphabetic,msc-links]{amsrefs}
\usepackage{graphicx}
%\usepackage{esint}
\usepackage{xcolor}
\usepackage{enumitem}
%\usepackage{showkeys}
\DeclareMathOperator*{\argmin}{arg\,min}
\usepackage[latin1]{inputenc}
%\usepackage[utf8]{inputenc}
\newcommand{\cal}{\mathcal}
%\setcounter{secnumdepth}{1}
%\setcounter{secnumdepth}{3}
\newcommand{\subscript}[2]{$#1 _ #2$}
\numberwithin{equation}{section}
% \newcommand{\half}{{\textstyle \frac 12}}
\newcommand{\acal}{\mathcal{A}}
\newcommand{\bcal}{\mathcal{B}}
\newcommand{\ccal}{\mathcal{C}}
\newcommand{\dcal}{\mathcal{D}}
\newcommand{\ecal}{\mathcal{E}}
\newcommand{\fcal}{\mathcal{F}}
\newcommand{\gcal}{\mathcal{G}}
\newcommand{\hcal}{\mathcal{H}}
\newcommand{\ical}{\mathcal{I}}
\newcommand{\jcal}{\mathcal{J}}
%\newcommand{\cal J}{\jcal}
\newcommand{\kcal}{\mathcal{K}}
\newcommand{\lcal}{\mathcal{L}}
\newcommand{\mcal}{\mathcal{M}}
\newcommand{\ncal}{\mathcal{N}}
\newcommand{\ocal}{\mathcal{O}}
\newcommand{\pcal}{\mathcal{P}}
\newcommand{\qcal}{\mathcal{Q}}
\newcommand{\rcal}{\mathcal{R}}
\newcommand{\scal}{\mathcal{S}}
\newcommand{\tcal}{\mathcal{T}}
\newcommand{\ucal}{\mathcal{U}}
\newcommand{\vcal}{\mathcal{V}}
\newcommand{\Q}{{\mathbb Q}}
\newtheorem{theo}{{\sc Theorem}}[section]
\newcommand{\wcal}{\mathcal{W}}
\newcommand{\zcal}{\mathcal{Z}}
\newtheorem{defin}{{\sc Definition}}
\newtheorem{mainprop}{{\sc Proposition}}
%\newtheorem{cor}[theo]{{\sc Corollary}}
%\newtheorem{maincor}[maintheo]{{\sc Corollary}}
\newtheorem{lem}[theo]{{\sc Lemma}}
%\newtheorem{prop}[theo]{{\sc Proposition}}
\newtheorem{conj}{\sc Conjecture}
\newtheorem{mainconj}{\sc Conjecture}
\newenvironment{example}{\medskip\noindent{\it Example:\/} }{\medskip}
\newenvironment{rem}{\medskip\noindent{\it Remark:\/} }{\medskip}
\usepackage{amsmath}
\newcommand{\wedgevee}{\mathrel{\substack{\wedge\\\vee}}}
\newcommand{\veewedge}{\mathrel{\substack{\vee\\\wedge}}}
\newcommand{\ave}[2]{{#2}_{#1}}
\newcommand{\avg}[2]{\frac{1}{|#1|}\int_{#1} #2}
\newtheorem{defn}[theo]{{\sc Definition}}
\newenvironment{claim}{\medskip\noindent{\it Claim:\/} }{\medskip}
%we define average integral
\def\Xint#1{\mathchoice
{\XXint\displaystyle\textstyle{#1}}%
{\XXint\textstyle\scriptstyle{#1}}%
{\XXint\scriptstyle\scriptscriptstyle{#1}}%
{\XXint\scriptscriptstyle%
\scriptscriptstyle{#1}}%
\!\int}
\def\XXint#1#2#3{{\setbox0=\hbox{$#1{#2#3}{%
\int}$ }
\vcenter{\hbox{$#2#3$ }}\kern-.6\wd0}}
\def\ddashint{\Xint=}
\def\dashint{\Xint-}
\setlength\evensidemargin{.5in}
%\setlength\textheight{8.1in}
\setlength\textheight{44cc} \setlength\textwidth{30cc}
\setlength\topmargin{0in} \setlength\parskip{5pt}
\renewcommand{\epsilon}{\varepsilon}
%\newcommand{\newsection}[1]
%{\subsection{#1}\setcounter{theorem}{0} \setcounter{equation}{0}
%\par\noindent}
%\renewcommand{\theequation}{\arabic{section}.\arabic{equation}}
%\renewcommand{\thesection}{\arabic{subsection}}
\newtheorem{theorem}{Theorem}
\renewcommand{\thetheorem}{\arabic{section}.\arabic{theorem}}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{prop}[theorem]{Prop}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{remark}[theorem]{Remark}
\newcommand{\bth}{\begin{theorem}}
\newcommand{\ble}{\begin{lemma}}
\newcommand{\bcor}{\begin{corr}}
\newcommand{\ltrt}{{L^2({\Bbb R}^3)}}
\newcommand{\ltoo}{{L^2({\mathbb{R}}^3\backslash\mathcal{K})}}
\newcommand{\bdeff}{\begin{deff}}
\newcommand{\lirt}{{L^\infty({\Bbb R}^3)}}
\newcommand{\lioo}{{L^{\infty}({\mathbb{R}}^3 \backslash \mathcal{K})}}
\newcommand{\bprop}{\begin{proposition}}
\newcommand{\ele}{\end{lemma}}
\newcommand{\ecor}{\end{corr}}
\newcommand{\edeff}{\end{deff}}
\numberwithin{theorem}{section}
\newcommand{\ii}{i}
\newcommand{\eprop}{\end{proposition}}
\newcommand{\rlnu}{{R_{\lambda}^{\nu}}}
\newcommand{\trlnu}{{\tilde{R}_{\lambda}^{\nu}}}
\newcommand{\tlnu}{{T_{\lambda}^{\nu}}}
\newcommand{\ttlnu}{{\tilde{T}_{\lambda}^{\nu}}}
\newcommand{\slnu}{{S_{\lambda}^{\nu}}}
\newcommand{\slnut}{{}^t\!\slnu}
\newcommand{\cd}{\, \cdot\, }
\newcommand{\mlnu}{{m_{\lambda}^{\nu}}}
\newcommand{\psilnu}{\psi_{\lambda}^{\nu}}
\newcommand{\xilnu}{\xi_{\lambda}^{\nu}}
\newcommand{\nlnu}{N_{\lambda}^{\nu}}
\newcommand{\nl}{N_{\lambda}}
\newcommand{\Rn}{{\mathbb R}^n}
\newcommand{\jump}{{}}
\newcommand{\la}{\lambda}
\newcommand{\st}{{\Bbb R}^{1+3}_+}
\newcommand{\eps}{\varepsilon}
\newcommand{\e}{\varepsilon}
\renewcommand{\l}{\lambda}
%\newcommand{\loc}{{\text{\rm loc}}}
\newcommand{\comp}{{\text{\rm comp}}}
\newcommand{\Coi}{C^\infty_0}
\newcommand{\supp}{\text{supp }}
\renewcommand{\Pi}{\varPi}
\renewcommand{\Re}{\rm{Re} \,}
\renewcommand{\Im}{\rm{Im} \,}
\renewcommand{\epsilon}{\varepsilon}
\newcommand{\sgn}{{\text {sgn}}}
\newcommand{\Gmid}{\Gamma_{\text{mid}}}
\newcommand{\Rt}{{\Bbb R}^3}
\newcommand{\Mdel}{{{\cal M}}^\alpha}
%\newcommand{\dist}{{{\rm dist}}}
\newcommand{\Adel}{{{\cal A}}_\delta}
\newcommand{\Kob}{{\cal K}}
\newcommand{\Dia}{\overline{\Bbb E}^{1+3}}
\newcommand{\Diap}{\overline{\Bbb E}^{1+3}_+}
\newcommand{\Cyl}{{\Bbb E}^{1+3}}
\newcommand{\Cylp}{{\Bbb E}^{1+3}_+}
\newcommand{\Penrose}{{\cal P}}
\newcommand{\Rplus}{{\Bbb R}_+}
\newcommand{\parital}{\partial}
\newcommand{\tidle}{\tilde}
\newcommand{\grad}{\text{grad}\,}
\newcommand{\ob}{{\mathcal K}}
\newcommand{\R}{{\mathbb R}}
\newcommand{\Z}{{\mathbb Z}}
\DeclareMathOperator*{\esssup}{ess\,sup}
\DeclareMathOperator*{\essinf}{ess\,inf}
\newcommand{\pa}[1]{\left(#1\right)}
\newcommand{\norm}[1]{\left\|#1\right\|}
\newcommand{\abs}[1]{\left\vert#1\right\vert}
\newcommand{\ip}[2]{\langle#1 ,#2\rangle}
\newcommand{\set}[2]{\left\{#1 :#2\right\}}
%\newcommand{\square}{\Box}
\newcommand{\subheading}[1]{{\bf #1}}
\newcommand{\1}{{\rm 1\hspace*{-0.4ex}%
\rule{0.1ex}{1.52ex}\hspace*{0.2ex}}}
%I add this because in my pc esint does not work
%\def\fint{\mathop{\,\rlap{-}\!\!\int}\nolimits}
\usepackage{esint}
\usepackage{comment}
\newcommand{\ac}{\color{magenta}}
\newcommand{\cpm}{\color{red}}
\newcommand{\ed}{\color{black}}
\allowdisplaybreaks
%
% \vint barred integral, needs one index (use {} if none)
% Definition submitted by Tero Kilpel\"ainen and Pekka Koskela
% with articles for Arkiv f\"or matematik 37:2 (1999).
% Modified to give the same spacing to the left of the sign as the
% usual integral by Anders Bj\"orn, 4 January 2000.
%
% Modified to have distance in em and ex rather than pt to scale
% properly in 11pt and 12pt by Anders Bj\"orn, 16 February 2005
%
% \vintslides is provided for use with the slides style,
% uncomment the last line in this case. Anders Bj\"orn, 23 February 2005
%
\def\vint_#1{\mathchoice%
{\mathop{\kern 0.2em\vrule width 0.6em height 0.69678ex depth -0.58065ex
\kern -0.8em \intop}\nolimits_{\kern -0.4em#1}}%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.69678ex depth -0.60387ex
\kern -0.6em \intop}\nolimits_{#1}}%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.69678ex
depth -0.60387ex
\kern -0.6em \intop}\nolimits_{#1}}%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.69678ex depth -0.60387ex
\kern -0.6em \intop}\nolimits_{#1}}}
\def\vintslides_#1{\mathchoice%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.697ex depth -0.581ex
\kern -0.6em \intop}\nolimits_{\kern -0.4em#1}}%
{\mathop{\kern 0.1em\vrule width 0.3em height 0.697ex depth -0.604ex
\kern -0.4em \intop}\nolimits_{#1}}%
{\mathop{\kern 0.1em\vrule width 0.3em height 0.697ex depth -0.604ex
\kern -0.4em \intop}\nolimits_{#1}}%
{\mathop{\kern 0.1em\vrule width 0.3em height 0.697ex depth -0.604ex
\kern -0.4em \intop}\nolimits_{#1}}}
\newcommand{\kint}{\vint}
\newcommand{\intav}{\vint}
\newcommand{\aveint}[2]{\mathchoice%
{\mathop{\kern 0.2em\vrule width 0.6em height 0.69678ex depth -0.58065ex
\kern -0.8em \intop}\nolimits_{\kern -0.45em#1}^{#2}}%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.69678ex depth -0.60387ex
\kern -0.6em \intop}\nolimits_{#1}^{#2}}%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.69678ex depth -0.60387ex
\kern -0.6em \intop}\nolimits_{#1}^{#2}}%
{\mathop{\kern 0.1em\vrule width 0.5em height 0.69678ex depth -0.60387ex
\kern -0.6em \intop}\nolimits_{#1}^{#2}}}
\newcommand{\ud}{\, d}
\newcommand{\half}{{\frac{1}{2}}}
\newcommand{\parts}[2]{\frac{\partial {#1}}{\partial {#2}}}
\newcommand{\ol}{\overline}
\newcommand{\Om}{\Omega}
\newcommand{\I}{\textrm{I}}
\newcommand{\II}{\textrm{II}}
\newcommand{\dist}{\operatorname{dist}}
\renewcommand{\P}{\mathbb{P\,}}
\newcommand{\F}{\mathcal{F}g}
\newcommand{\om}{\omega}
\newcommand{\trm}{\textrm}
\newcommand{\spt}{\operatorname{spt}}
\newcommand{\loc}{\mathrm{loc}}
\newcommand{\M}{\operatorname{M}}
\def\Xint#1{\mathchoice
{\XXint\displaystyle\textstyle{#1}}%
{\XXint\textstyle\scriptstyle{#1}}%
{\XXint\scriptstyle\scriptscriptstyle{#1}}%
{\XXint\scriptscriptstyle\scriptscriptstyle{#1}}%
\!\int}
\def\XXint#1#2#3{{\setbox0=\hbox{$#1{#2#3}{\int}$}
\vcenter{\hbox{$#2#3$}}\kern-.5\wd0}}
\def\ddashint{\Xint=}
\def\avgint{\Xint-}
\newcommand{\vertiii}[1]{{\left\vert\kern-0.25ex\left\vert\kern-0.25ex\left\vert #1
\right\vert\kern-0.25ex\right\vert\kern-0.25ex\right\vert}}
\newcommand{\vertii}[1]{{\left\vert\kern-0.25ex\left\vert\kern-0.25ex #1
\kern-0.25ex\right\vert\kern-0.25ex\right\vert}}
\makeatletter
\@namedef{subjclassname@2020}{\textup{2020} Mathematics Subject Classification}
\makeatother
\let\udot\relax
\usepackage{mathabx}
\usepackage{geometry}
\begin{document}
\title[Bourgain--Brezis--Mironescu formula for Riesz Potentials]
{Bourgain--Brezis--Mironescu formula for Riesz Potentials}
\author[A. Claros]{Alejandro Claros}
\address[A. Claros]{BCAM -- Basque Center for Applied Mathematics, Bilbao, Spain\newline
Universidad del Pa\'is Vasco / Euskal Herriko Unibertsitatea (UPV/EHU), Bilbao, Spain}
\email{aclaros@bcamath.org, aclaros003@ikasle.ehu.eus}
\author[C. P\'erez]{Carlos P\'erez}
\address[C. P\'erez]{BCAM -- Basque Center for Applied Mathematics, Bilbao, Spain\newline
Universidad del Pa\'is Vasco / Euskal Herriko Unibertsitatea (UPV/EHU), Bilbao, Spain\newline
Ikerbasque, Bilbao, Spain}
\email{cperez@bcamath.org}
\thanks{A. Claros is supported by the Basque Government through the BERC 2022-2025 program, by the Ministry of Science and Innovation through Grant PRE2021-099091 funded by BCAM Severo Ochoa accreditation CEX2021-001142-S/MICIN/AEI/10.13039/501100011033 and by ESF+, and by the project PID2023-146646NB-I00 funded by MICIU/AEI/10.13039/501100011033 and by ESF+.
\\ \hspace*{1.5em}C. P\'erez is supported by the Spanish government through the grant PID2023-146646NB-I00 and by Severo Ochoa accreditation CEX2021-001142-S, both at BCAM, and also by the Basque Government through grant IT1615-22 at the University of the Basque Country and by the BERC programme 2022-2025 at BCAM}
\subjclass[2020]{Primary 46E35; Secondary 42B35}
%\date{\today}
%\dedicatory{}
\keywords{Riesz potential, nonlinear fractional derivative, Sobolev spaces}
\begin{abstract}
We identify the Bourgain--Brezis--Mironescu pointwise limit of the nonlocal potential operator $(1-\alpha)\, I_\alpha(\mathcal D^\alpha f)$, $0<\alpha<1$, where $I_\alpha$ denotes the Riesz potential and $\mathcal D^\alpha$ a nonlinear fractional differential operator. Specifically, for every $f\in C_c^\infty(\R^n)$ and every $x\in\R^n$, we show that
\begin{equation*}
\lim_{\alpha\to 1^-} (1-\alpha)\, I_\alpha(\mathcal D^\alpha f)(x) = K_n\, I_1(|\nabla f|)(x),
\end{equation*}
where $K_n$ is the geometric constant appearing in the well-known Bourgain--Brezis--Mironescu formula \cite{BBM2}. By a density argument, we further extend this result to every $f\in W^{1,1}(\R^n)$, obtaining almost everywhere convergence along subsequences.
\end{abstract}
\maketitle
\section{Introduction and statement of the main result}
A classical starting point in Sobolev theory is the subrepresentation formula
\begin{equation}\label{eq:subrep-classical}
|f(x)| \lesssim_n I_1(|\nabla f|)(x),
\end{equation}
valid for $f\in C_c^\infty(\R^n)$, with $n\ge2$. Here, for $\alpha\in (0,n)$, the Riesz potential is defined by
\begin{equation*}
I_\alpha f(x):=\gamma_{n,\alpha}\int_{\R^n}\frac{f(y)}{|x-y|^{n-\alpha}}\,dy,
\end{equation*}
where $\gamma_{n,\alpha}$ is a normalization constant chosen so that $\mathcal F(I_\alpha f)(\xi)=|\xi|^{-\alpha}\mathcal F(f)(\xi)$, where $\mathcal F(f)(\xi)=\int_{\R^n} f(x) \, e^{-ix\cdot \xi}\,dx$ denotes the Fourier transform. Explicitly, this constant is given by
\begin{equation}\label{Defgamma}
\gamma_{n,\alpha}=\frac{\Gamma\bigl(\tfrac{n-\alpha}{2}\bigr)}{2^\alpha \pi^{\frac{n}{2}}\Gamma\bigl(\tfrac{\alpha}{2}\bigr)}.
\end{equation}
%
The estimate \eqref{eq:subrep-classical} can be derived from the classical local $(1,1)$-Poincar\'e inequality on balls (or cubes) via a standard chain argument, which converts local mean oscillation control into a pointwise potential estimate. Recently, several extensions of \eqref{eq:subrep-classical} have been obtained. On the one hand, the authors in \cite{HMP2} refine the left-hand side through the substitution $f \mapsto Tf$, with $T$ ranging from various maximal operators to singular integrals with rough kernels. On the other hand, \cite{HMP} improves the right-hand side by replacing it with a smaller operator (see \eqref{eq:HMP} below). This improvement of \eqref{eq:subrep-classical} is closely related to the very interesting nonlinear fractional differential operator introduced by D. Spector in \cite{SpectorJFA}:
\begin{equation*}
\mathcal{D}^\alpha f(x):=\int_{\R^n}\frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}},dy,
\qquad 0<\alpha<1,
\end{equation*}
which serves as a nonlocal ``fractional gradient" of order $\alpha$. This operator is of interest for several reasons, one of them is the recent fractional subrepresentation formula established in \cite{HMP} (see \eqref{eq:frac-subrep} below). On the other hand, it preserves some of the structural properties of first-order differential operators. For instance, it satisfies the following Leibniz-type inequality (see \cite{NahasPonce}),
\begin{equation*}
\norm{\mathcal{D}^\alpha (fg)}_{L^1(\R^n)} \le \norm{f\, \mathcal{D}^\alpha g}_{L^1(\R^n)} +\norm{g\, \mathcal{D}^\alpha f}_{L^1(\R^n)}.
\end{equation*}
The link between the fractional scale and the gradient scale is given by the Bourgain--Brezis--Mironescu (BBM) formula \cite{BBM2}\footnote{The original proof of the BBM formula was stated and proved in \cite{BBM2} for smooth bounded domains $\Omega \subset \R^n$. The result can be extended to the whole $\R^n$, see \cite[Appendix A]{BSY} and \cite{Kaushik}.}. For $f\in C^\infty_c(\R^n)$, it identifies the limit of the fractional Gagliardo seminorm
\begin{equation}\label{eq:BBM-limit}
\lim_{\alpha\to1^-}(1-\alpha)\int_{\R^n}\int_{\R^n}\frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}\,dx\,dy =K_n\int_{\R^n}|\nabla f(x)|\,dx,
\end{equation}
where
\begin{equation}\label{eq:Kn}
K_n:=\int_{\mathbb{S}^{n-1}}|\omega\cdot e|\,d\sigma(\omega)
\end{equation}
and $e\in \mathbb{S}^{n-1}$ is arbitrary. In particular, the factor $(1-\alpha)$ is crucial, otherwise the limit is not even finite for nonconstant functions (see \cite{BrezisConstant}). BBM-type limits have also been investigated in extension domains \cite{KKP}, in arbitrary bounded domains, where boundary effects require modifications of the inner integral in the seminorm \cite{DrelichmanDuran}, and in fully arbitrary domains with $W^{\alpha,p}_q$ seminorms \cite{Kaushik}, which are related to Triebel--Lizorkin spaces. Further extensions have been obtained in fractional Orlicz--Sobolev spaces \cite{Orlicz,Orlicz2,Orlicz3}, in the setting of interpolation spaces \cite{DominguezMilman}, in ball Banach function spaces \cite{DachunYang, ZhuYangYuan2023}, in connection with Triebel--Lizorkin spaces \cite{BSY}, and for anisotropic fractional energies \cite{FernandezSalort}.
In \cite{BBM1}, the following fractional Poincar\'e inequality on cubes is proved with the BBM extra term $(1-\alpha)$. For every cube $Q\subset\R^n$ and every $f\in C^\infty(Q)$,
\begin{equation}\label{eq:BBM-Poincare}
\dashint_Q |f(x)-f_Q|\,dx \lesssim (1-\alpha)\,\ell(Q)^\alpha\, \dashint_Q\int_Q \frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}\,dx\,dy, \qquad 0<\alpha<1,
\end{equation}
where $f_Q=\dashint_Q f$. Once \eqref{eq:BBM-Poincare} is available, a standard telescoping argument (see for instance \cite{FLW, HMPV}) yields the fractional subrepresentation formula
\begin{equation}\label{eq:frac-subrep}
|f(x)| \lesssim_n (1-\alpha)\, I_\alpha(\mathcal{D}^\alpha f)(x),
\qquad x\in\R^n,\quad f\in C^\infty_c(\R^n).
\end{equation}
as shown in \cite{HMP}. In particular, $(1-\alpha)I_\alpha(\mathcal{D}^\alpha f)$ plays the role of a fractional version of the first-order Riesz potential of the gradient, in the spirit of \eqref{eq:subrep-classical}. Interesting related subrepresentation formulae were established for $s$-John domains in \cite{DD2}, although without the corresponding BBM-type factor $(1-\alpha)$.
An alternative motivation for studying $(1-\alpha)I_\alpha(\mathcal{D}^\alpha f)$ stems from the theory of rough singular integrals. In the recent work \cite{HMP}, it is established that, for every $\alpha \in (0,1)$, if $\Omega$ is homogeneous of degree zero, has vanishing average on $\mathbb{S}^{n-1}$, and belongs to the Marcinkiewicz space $L^{\frac{n}{\alpha},\infty}(\mathbb{S}^{n-1})$, then the maximal truncated rough operator
\begin{equation*}
T^\star_{\Omega} f(x):=\sup_{\varepsilon>0}\Bigl|\int_{|x-y|>\varepsilon}\frac{\Omega(x-y)}{|x-y|^n}\,f(y)\,dy\Bigr|
\end{equation*}
satisfies the pointwise estimate
\begin{equation}\label{HMPsubrep}
T^\star_{\Omega} f(x)\leq c_n\, \|\Omega\|_{L^{\frac{n}{\alpha},\infty}(\mathbb S^{n-1})}\, (1-\alpha) \,I_{\alpha}(\mathcal{D}^{\alpha}f)(x), \qquad f\in C^\infty_c(\R^n).
\end{equation}
%
The same work also establishes that for every $f\in C_c^\infty(\R^n)$,
\begin{equation}\label{eq:HMP}
(1-\alpha)\, I_\alpha(\mathcal{D}^\alpha f)(x)\le C_{n,\alpha}\, I_1(|\nabla f|)(x),
\end{equation}
%
where the constant $C_{n,\alpha}$ bounded as $\alpha\to1^-$.
Combining this with \eqref{eq:frac-subrep}, we obtain
\begin{equation*}
|f(x)| \lesssim_n (1-\alpha)\, I_\alpha(\mathcal{D}^\alpha f)(x) \le C_{n,\alpha}\, I_1(|\nabla f|)(x)
\end{equation*}
for every $x\in \R^n$. The boundedness of $C_{n,\alpha}$ when $\alpha\to1^-$ in \eqref{eq:HMP} suggests that the operator $(1-\alpha)I_\alpha(\mathcal{D}^\alpha f)$ should have a meaningful limit as $\alpha\to 1^-$, in the spirit of \eqref{eq:BBM-limit}.
Our main result confirms this intuition by identifying the pointwise BBM limit of the operator appearing on the left-hand side of \eqref{eq:HMP}.
\begin{theorem}\label{thm:1}
Let $f\in C_c^\infty(\R^n)$ with $n\ge 2$. Then for every $x\in\R^n$,
\begin{equation}\label{eq:main-limit}
\lim_{\alpha\to 1^-} (1-\alpha)\, I_\alpha(\mathcal{D}^\alpha f)(x) =K_n\, I_1(|\nabla f|)(x),
\end{equation}
where $K_n$ is given by \eqref{eq:Kn}.
\end{theorem}
\begin{remark}\label{rem:Kn}
The constant $K_n$ coincides with the geometric constant appearing in the BBM limit \eqref{eq:BBM-limit}, and it is independent of the choice of $e\in \mathbb{S}^{n-1}$ by rotation invariance.
\end{remark}
In particular, Theorem \ref{thm:1} shows that the inequality \eqref{eq:HMP}, proved in \cite{HMP}, is asymptotically sharp as $\alpha\to 1^-$.
\subsection*{Outline of the paper}
The following section is devoted to several auxiliary lemmas. Section 3 is devoted to the proof of Theorem \ref{thm:1}. In Section 4, we extend the pointwise convergence in \eqref{eq:main-limit} from $C_c^\infty(\R^n)$ to $W^{1,1}(\R^n)$ by a density argument. Finally, we discuss further extensions, including an $L^p$-variant of the operator and its corresponding pointwise BBM limit.
\section{Auxiliary lemmas}
As a first step toward the proof of the main result, we analyze the pointwise limit of the fractional gradient $\mathcal{D}^\alpha f$. The following lemma describes this behavior for smooth functions.
\begin{lemma}\label{lem:1}
Let $f\in C_c^2(\R^n)$, then for every $x\in \R^n$ we have
\begin{equation*}
\lim_{\alpha\to 1^-} (1-\alpha) \mathcal{D} ^\alpha f(x) = K_n\, |\nabla f(x)|.
\end{equation*}
\end{lemma}
The previous result is implicit in \cite{BBM1}; we include a proof for the convenience of the reader. We refer to \cite{Kaushik} for a more general version involving general domains and mixed Sobolev seminorms $W_q^{\alpha, p}$.
\begin{proof}
Fix $x\in \R^n$. We split the fractional derivative into two terms, the local and nonlocal parts,
\begin{align*}
(1-\alpha) \mathcal{D} ^\alpha f(x) = & (1-\alpha) \int_{\R^n} \frac{|f(x+h)-f(x)|}{|h|^{n+\alpha}}dh\\
= & (1-\alpha) \int_{|h|<1} \frac{|f(x+h)-f(x)|}{|h|^{n+\alpha}}dh\\
& +(1-\alpha) \int_{|h|\ge 1} \frac{|f(x+h)-f(x)|}{|h|^{n+\alpha}}dh\\
= & I_1 + I_2.
\end{align*}
First, we observe that the nonlocal part does not contribute to the limit,
\begin{align*}
I_2 \le & 2\|f\|_{L^\infty} (1-\alpha)\int_{|h|\ge 1} \frac{1}{|h|^{n+\alpha}}dh \\
= & C_n \|f\|_{L^\infty} (1-\alpha)\int_{1}^\infty \frac{1}{r^{1+\alpha}}dr \\
= & C_n \|f\|_{L^\infty}\frac{1-\alpha}{\alpha} \longrightarrow 0,
\end{align*}
when $\alpha\to 1^-$. To study the local part, we consider the Taylor expansion of $f$ of order $2$,
\begin{equation*}
f(x+h)= f(x)+ \nabla f(x) \cdot h + R(h),
\end{equation*}
where the remainder term satisfies $|R(h)|\le \tfrac{1}{2}\|D^2 f\|_{L^\infty}\,|h|^2$ for $|h|<1$. Using the elementary inequality $\bigl||a+b|-|a|\bigr|\le |b|$ for $a,b\in\R$, with $a=\nabla f(x)\cdot h$ and $b=R(h)$, we obtain
\begin{equation*}
\bigl||f(x+h)-f(x)|-|\nabla f(x)\cdot h|\bigr|\le |R(h)|,\qquad |h|<1.
\end{equation*}
Therefore,
\begin{equation}\label{eq:sandwich}
|I_1-I_{11}|\le I_{12},
\end{equation}
where
\begin{align*}
I_{11}:=(1-\alpha)\int_{|h|<1}\frac{|\nabla f(x)\cdot h|}{|h|^{n+\alpha}}\,dh,
\quad \text{ and } \quad
I_{12}:=(1-\alpha)\int_{|h|<1}\frac{|R(h)|}{|h|^{n+\alpha}}\,dh.
\end{align*}
The remainder term satisfies
\begin{align*}
I_{12} &\le C_f(1-\alpha)\int_{|h|<1}\frac{|h|^2}{|h|^{n+\alpha}}\,dh\\
&= C_f C_n (1-\alpha)\int_0^1 r^{1-\alpha}\,dr \\
&= C_f C_n \frac{1-\alpha}{2-\alpha}\longrightarrow 0,
\end{align*}
when $\alpha\to 1^-$. Next, we compute $I_{11}$. If $\nabla f(x)=0$, then $I_{11}=0$ and \eqref{eq:sandwich} gives $0\le I_1\le I_{12}\longrightarrow 0$, and then $\lim_{\alpha\to 1^-}(1-\alpha)\mathcal{D}^\alpha f(x)=0=K_n|\nabla f(x)|$. Assume now that $\nabla f(x)\neq 0$. Using polar coordinates $h=r\omega$, we get
\begin{align*}
I_{11}
&=(1-\alpha)\int_0^1\int_{\mathbb{S}^{n-1}}
\frac{|\nabla f(x)\cdot (r\omega)|}{r^{n+\alpha}}\,r^{n-1}\,d\sigma(\omega)\,dr\\
&=(1-\alpha)\left(\int_{\mathbb{S}^{n-1}}|\nabla f(x)\cdot \omega|\,d\sigma(\omega)\right)\int_0^1 r^{-\alpha}\,dr\\
&=(1-\alpha)\left(\int_{\mathbb{S}^{n-1}}|\nabla f(x)\cdot \omega|\,d\sigma(\omega)\right)\frac{1}{1-\alpha}\\
&=\int_{\mathbb{S}^{n-1}}|\nabla f(x)\cdot \omega|\,d\sigma(\omega).
\end{align*}
By homogeneity and rotational invariance,
\begin{equation*}
\int_{\mathbb{S}^{n-1}}|\nabla f(x)\cdot \omega|\,d\sigma(\omega)
=|\nabla f(x)|\int_{\mathbb{S}^{n-1}}\left|\frac{\nabla f(x)}{|\nabla f(x)|}\cdot \omega\right|\,d\sigma(\omega)
=K_n|\nabla f(x)|.
\end{equation*}
Therefore $I_{11}=K_n|\nabla f(x)|$ and, since $I_{12}\to 0$, \eqref{eq:sandwich} yields $\lim_{\alpha\to 1^-} I_1 = K_n|\nabla f(x)|$. Together with $I_2\to 0$ we conclude that
\begin{equation*}
\lim_{\alpha\to 1^-}(1-\alpha)\mathcal{D}^\alpha f(x)
=\lim_{\alpha\to 1^-}(I_1+I_2)=K_n|\nabla f(x)|.
\end{equation*}
This concludes the proof.
\end{proof}
\begin{remark}
As demonstrated in the proof, the nonlocal term vanishes in the limit. Actually, we have established the following result:
\begin{equation*}
\lim_{\alpha\to 1^-} (1-\alpha) \int_{B(x,r)} \frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}dy = K_n\, |\nabla f(x)|,
\end{equation*}
for every $x\in \R^n$ and every $0<r\le \infty$. In particular, if $\Omega$ is a domain and $\tau \in (0,1)$ we have proved
\begin{equation}\label{eq:domains}
\lim_{\alpha\to 1^-} (1-\alpha) \int_{B(x,\tau \dist (x, \partial \Omega))} \frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}dy = K_n\, |\nabla f(x)|,
\end{equation}
for every $x\in \Omega$.
\end{remark}
To justify the exchange of limits and integrals later in the proof of Theorem \ref{thm:1}, we need a uniform pointwise bound for $(1-\alpha) \mathcal{D}^\alpha f$ that holds uniformly for $\alpha$ close to $1$.
\begin{lemma}\label{lem:2}
Let $f\in C^2_c(\R^n)$. Then for each $\alpha \in (\tfrac{1}{2}, 1)$ and $x\in \R^n,$ we have
\begin{equation*}
(1-\alpha) \mathcal{D} ^\alpha f(x) \le C_n \pa{\|f\|_{L^\infty} + \|\nabla f\|_{L^\infty}}.
\end{equation*}
\end{lemma}
\begin{proof}
We again split the fractional derivative into two terms: the local and nonlocal parts. We have
\begin{align*}
(1-\alpha) \mathcal{D} ^\alpha f(x) = & (1-\alpha) \int_{|x-y|<1} \frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}dy \\
&+ (1-\alpha) \int_{|x-y|\ge 1} \frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}dy\\
= & I_1 + I_2.
\end{align*}
To estimate the local term, by the mean value theorem, we have
\begin{align*}
I_1 \le & C \| \nabla f\|_{L^\infty} (1-\alpha) \int_{|x-y|<1} \frac{|x-y|}{|x-y|^{n+\alpha}}dy \\
= & C_n \| \nabla f\|_{L^\infty} (1-\alpha) \int_0^1 r^{-\alpha} dr\\
= & C_n \| \nabla f\|_{L^\infty}.
\end{align*}
On the other hand, we use the fact that $f$ is bounded,
\begin{align*}
I_2 \le & 2\|f\|_{L^\infty} (1-\alpha) \int_{|x-y|\ge 1} \frac{1}{|x-y|^{n+\alpha}}dy\\
= & C_n \|f\|_{L^\infty} (1-\alpha) \int_1^\infty r^{-1-\alpha} dr \\
= & C_n \|f\|_{L^\infty} \frac{1-\alpha}{\alpha} \\
\le & C_n \|f\|_{L^\infty}
\end{align*}
where in the last inequality we use $\alpha \in (\tfrac{1}{2}, 1)$.
\end{proof}
\begin{remark}
Let $\Omega\subset\R^n$ be an arbitrary bounded domain and fix $\tau\in(0,1)$. Combining \eqref{eq:domains} with the uniform bound from Lemma \ref{lem:2}, one may apply the dominated convergence theorem to obtain the well-known BBM limit in $\Omega$ with the truncated seminorm
\begin{equation*}
\lim_{\alpha \to 1^-} (1-\alpha)\int_{\Omega}\int_{B(x,\tau\,\dist(x,\partial\Omega))}\frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}\,dy\,d\mu(x) =K_n\int_{\Omega}|\nabla f(x)|\,d\mu(x),
\end{equation*}
where $\mu$ is a locally finite measure. In particular, this recovers the main result of \cite{DrelichmanDuran} for $p=1$ and $f\in C_c^2(\R^n)$.
\end{remark}
\section{Proof of Theorem \texorpdfstring{\ref{thm:1}}{1.1}}
\begin{proof}[Proof of Theorem \ref{thm:1}]
Fix $x\in\R^n$ and set $f_x(u):=f(u+x)$. Since both $\mathcal{D}^\alpha$ and the Riesz potential $I_\alpha$ are translation invariant, we have, for every $\alpha\in(0,1)$,
\begin{equation*}
(1-\alpha)\,I_\alpha(\mathcal{D}^\alpha f)(x)=(1-\alpha)\,I_\alpha(\mathcal{D}^\alpha f_x)(0).
\end{equation*}
Moreover, $|\nabla f_x(u)|=|\nabla f(u+x)|$, and therefore
\begin{equation*}
I_1(|\nabla f|)(x)=I_1(|\nabla f_x|)(0).
\end{equation*}
Hence, it is enough to prove \eqref{eq:main-limit} in the case $x=0$, applied to the translated function $f_x$. In what follows, we assume $x=0$ and write $f$ in place of $f_x$ for simplicity.
For each $\alpha \in (0,1)$, define
\begin{equation*}
F_\alpha(y):=\gamma_{n,\alpha}(1-\alpha)\frac{\mathcal{D}^\alpha f(y)}{|y|^{n-\alpha}},
\end{equation*}
%
and
\begin{equation*}
F(y):=K_n\gamma_{n,1}\frac{|\nabla f(y)|}{|y|^{n-1}},
\end{equation*}
%
for $y\in\R^n\setminus\{0\}$, where $\gamma_{n,\alpha}$ is defined in \eqref{Defgamma}. By Lemma \ref{lem:1}, together with the continuity of $\gamma_{n,\alpha}$ as a function of $\alpha$, we have
\begin{equation*}
\lim_{\alpha\to1^-}F_\alpha(y)=F(y)
\end{equation*}
for every $y\in\R^n\setminus\{0\}$. Thus, it remains to show that
\begin{equation}\label{eq:limint}
\lim_{\alpha\to1^-}\int_{\R^n}F_\alpha(y)\,dy=\int_{\R^n}F(y)\,dy.
\end{equation}
To justify the interchange of limit and integral, we shall apply the dominated convergence theorem. Let $R_0>0$ be such that $\supp(f)\subset B(0,R_0)$, and decompose
\begin{equation*}
\R^n=A_1\cup A_2\cup A_3,
\end{equation*}
where
\begin{equation*}
A_1=B(0,1), \qquad A_2=\R^n\setminus B(0,2R_0+1), \qquad A_3=B(0,2R_0+1)\setminus B(0,1).
\end{equation*}
We begin with $y\in A_1$. If $\alpha\in(\tfrac12,1)$, then Lemma \ref{lem:2} gives
\begin{align*}
F_\alpha(y)
&=\gamma_{n,\alpha}(1-\alpha)\frac{\mathcal{D}^\alpha f(y)}{|y|^{n-\alpha}}\\
&\le C_n\frac{\|f\|_{L^\infty}+\|\nabla f\|_{L^\infty}}{|y|^{n-\alpha}}\\
&\le C_{n,f}\frac{1}{|y|^{n-\frac12}}\\
& =: h(y).
\end{align*}
Moreover, $h\in L^1(A_1)$, since
\begin{equation*}
\int_{A_1}h(y)\,dy=\int_{B(0,1)}\frac{1}{|y|^{n-\frac12}}\,dy
=C_n\int_0^1r^{-\frac12}\,dr<\infty.
\end{equation*}
Next, let $y\in A_2$. Then $f(y)=0$, and therefore
\begin{align*}
\mathcal{D}^\alpha f(y)
&=\int_{\R^n}\frac{|f(y)-f(z)|}{|y-z|^{n+\alpha}}\,dz\\
&=\int_{\supp(f)}\frac{|f(z)|}{|y-z|^{n+\alpha}}\,dz\\
&\le \int_{\supp(f)}|f(z)|\frac{2^{n+\alpha}}{|y|^{n+\alpha}}\,dz\\
&=2^{n+\alpha}\|f\|_{L^1}\frac{1}{|y|^{n+\alpha}},
\end{align*}
because $z\in \supp(f)\subset B(0,R_0)$ and $|y|\ge 2R_0+1$, so
\begin{equation*}
|y-z|\ge |y|-|z|\ge |y|-R_0\ge \frac{|y|}{2}.
\end{equation*}
%
Hence, for $y\in A_2$
%
\begin{align*}
F_\alpha(y)
&=\gamma_{n,\alpha}(1-\alpha)\frac{\mathcal{D}^\alpha f(y)}{|y|^{n-\alpha}}\\
&\le \gamma_{n,\alpha}(1-\alpha)2^{n+\alpha}\|f\|_{L^1}\frac{1}{|y|^{n+\alpha}|y|^{n-\alpha}}\\
&\le C_n\|f\|_{L^1}\frac{1}{|y|^{2n}}\\
& =:g(y),
\end{align*}
and $g\in L^1(A_2)$, since
\begin{equation*}
\int_{A_2}g(y)\,dy
=C_{n,f}\int_{\R^n\setminus B(0,2R_0+1)}\frac{1}{|y|^{2n}}\,dy
=C_{n,f}\int_{2R_0+1}^\infty r^{-n-1}\,dr<\infty.
\end{equation*}
Finally, let $y\in A_3$. Since $|y|\ge 1$, Lemma \ref{lem:2} yields
\begin{equation*}
F_\alpha(y)=\gamma_{n,\alpha}(1-\alpha)\frac{\mathcal{D}^\alpha f(y)}{|y|^{n-\alpha}}\le C_{n,f},
\end{equation*}
for every $\alpha\in(\tfrac12,1)$. Since $|A_3|<\infty$, this gives an integrable dominating function on $A_3$.
Combining the estimates on $A_1$, $A_2$, and $A_3$, we obtain an $L^1$-dominating function on $\R^n$. Therefore, the dominated convergence theorem yields \eqref{eq:limint} which completes the proof.
\end{proof}
\section{Extension to the Sobolev space \texorpdfstring{$W^{1,1}(\R^n)$}{W11}}
In this section, we extend Theorem \ref{thm:1} to the Sobolev space $W^{1,1}(\R^n)$ by a density argument. First, we prove the following inequality for $p=1$ and the global Gagliardo seminorm.
\begin{proposition}\label{prop:inverse}
Let $f\in W^{1,1}(\R^n)$ and let $0<\alpha<1$. Then
\begin{equation*}
\alpha(1-\alpha)\int_{\R^n}\int_{\R^n} \frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}\,dy\,dx \le C_n \|f\|_{W^{1,1}(\R^n)}.
\end{equation*}
\end{proposition}
This proposition is a consequence of the Gagliardo-Nirenberg interpolation inequality proved in \cite{BrezisMironescu} (see also \cite[(1.10)]{HLYY}). For completeness, we include a proof.
\begin{proof}
By the change of variables $y=x+h$,
\begin{align*}
\int_{\R^n}\int_{\R^n}\frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}\,dy\,dx = \int_{\R^n}\int_{\R^n}\frac{|f(x+h)-f(x)|}{|h|^{n+\alpha}}\,dx\,dh =: I_1+I_2,
\end{align*}
where in $I_1$ we integrate over $|h|<1$, while in $I_2$ we integrate over $|h|\ge 1$. For $I_1$, we use the following difference quotients estimate,
\begin{equation*}
\int_{\R^n}|f(x+h)-f(x)|\,dx \le |h|\int_{\R^n}|\nabla f(x)|\,dx,
\end{equation*}
for every $h\in\R^n$ (see for instance \cite[Proposition 9.3]{BrezisBook}). Indeed, this is immediate for $f\in C_c^\infty(\R^n)$ from
\begin{equation*}
f(x+h)-f(x)=\int_0^1 \nabla f(x+th)\cdot h\,dt,
\end{equation*}
and the general case follows by density in $W^{1,1}(\R^n)$. Hence
\begin{align*}
I_1 &\le \int_{|h|<1}\frac{1}{|h|^{n+\alpha-1}} \left(\int_{\R^n}|\nabla f(x)|\,dx\right)dh \le \frac{C_n}{1-\alpha}\int_{\R^n}|\nabla f(x)|\,dx.
\end{align*}
For $I_2$, the triangle inequality gives
\begin{align*}
I_2 &\le 2\int_{|h|\ge 1}\frac{dh}{|h|^{n+\alpha}}\int_{\R^n}|f(x)|\,dx \le \frac{C_n}{\alpha}\int_{\R^n}|f(x)|\,dx.
\end{align*}
Combining both estimates, we obtain
\begin{equation*}
\int_{\R^n}\int_{\R^n}\frac{|f(x)-f(y)|}{|x-y|^{n+\alpha}}\,dy\,dx \le \frac{C_n}{1-\alpha}\int_{\R^n}|\nabla f(x)|\,dx + \frac{C_n}{\alpha}\int_{\R^n}|f(x)|\,dx.
\end{equation*}
We conclude the proof by multiplying both sides by $\alpha(1-\alpha)$ and using that $\alpha < 1$ and $1-\alpha < 1$.
\end{proof}
We can now state the main result of this section.
\begin{theorem}
Let $f\in W^{1,1}(\R^n)$ with $n\ge 2$. Then, for every sequence $\{\alpha_k\}_{k} \subset (0,1)$ with $\alpha_k \to 1^-$, there exists a subsequence $\{\alpha_{k_j}\}_{j}$ such that
\begin{equation*}
\lim_{j\to \infty} (1-\alpha_{k_j}) I_{\alpha_{k_j}} (\mathcal{D} ^{\alpha_{k_j}} f)(x) = K_n\, I_1(|\nabla f|)(x),
\end{equation*}
for almost every $x\in \R^n$.
\end{theorem}
\begin{proof}
Let $f\in W^{1,1}(\R^n)$. By the density of smooth compactly supported functions, there exists a sequence $\{\varphi_k\}_{k=1}^\infty\subset C_c^\infty(\R^n)$ such that $\varphi_k\to f$ in $W^{1,1}(\R^n)$ (see for instance \cite{AdamsFournier, BrezisBook}). We first show that
\begin{equation*}
(1-\alpha) I_\alpha(\mathcal{D}^\alpha f)\to K_n I_1(|\nabla f|)
\end{equation*}
in measure on every compact set $E \subset \R^n$ as $\alpha\to 1^-$.
For every $k\in \mathbb{N}$ and every $\alpha\in(0,1)$, we use the triangle inequality to write
\begin{align*}
\left|(1-\alpha) I_\alpha(\mathcal{D}^\alpha f)(x)-K_n I_1(|\nabla f|)(x)\right|
\le &\ (1-\alpha)\left|I_\alpha(\mathcal{D}^\alpha f)(x)-I_\alpha(\mathcal{D}^\alpha \varphi_k)(x)\right|\\
&+\left|(1-\alpha)I_\alpha(\mathcal{D}^\alpha \varphi_k)(x)-K_n I_1(|\nabla \varphi_k|)(x)\right|\\
&+K_n\left|I_1(|\nabla \varphi_k|)(x)-I_1(|\nabla f|)(x)\right|\\
=:&\ A_{1,\alpha,k}(x)+A_{2,\alpha,k}(x)+A_{3,k}(x).
\end{align*}
We begin by bounding the first term. Combining the weak-type estimate for the Riesz potential with Proposition \ref{prop:inverse} for $\alpha \in (\tfrac12, 1)$, we obtain
\begin{align*}
\|A_{1,\alpha,k}\|_{L^{\frac{n}{n-\alpha},\infty}(\R^n)}
&\le (1-\alpha)\|I_\alpha(\mathcal{D}^\alpha(f-\varphi_k))\|_{L^{\frac{n}{n-\alpha},\infty}(\R^n)}\\
&\le C_n (1-\alpha)\int_{\R^n}\mathcal{D}^\alpha(f-\varphi_k)(x)\,dx\\
&\le C_n \|f-\varphi_k\|_{W^{1,1}(\R^n)}.
\end{align*}
Similarly, for the third term,
\begin{align*}
\|A_{3,k}\|_{L^{\frac{n}{n-1},\infty}(\R^n)} &\le K_n \|I_1(|\nabla \varphi_k-\nabla f|)\|_{L^{\frac{n}{n-1},\infty}(\R^n)}\\
&\le C_n \|\nabla \varphi_k-\nabla f\|_{L^1(\R^n)}\\
&= C_n \|\nabla (\varphi_k- f)\|_{L^1(\R^n)}\\
&\le C_n \|f-\varphi_k\|_{W^{1,1}(\R^n)}.
\end{align*}
Given any $\varepsilon>0$ and $\eta>0$, since $\varphi_k \to f$ in $W^{1,1}(\R^n)$, we can choose $k$ sufficiently large such that
\begin{equation*}
\frac{C_n\|f-\varphi_k\|_{W^{1,1}(\R^n)}}{\varepsilon} < \min\left\{ 1,\, \frac{\eta}{3} ,\,
\left(\frac{\eta}{3}\right)^{\frac{n-1}{n}}\right\}.
\end{equation*}
Fix a compact set $E\subset \R^n$. Since $\frac{n}{n-\alpha}\ge 1$ for every $\alpha\in(0,1)$, applying Chebyshev's inequality to $A_{1,\alpha,k}$ we have
\begin{align*}
\left|\left\{x\in E:\ A_{1,\alpha,k}(x)>\varepsilon\right\}\right|
&\le \left|\left\{x\in \R^n:\ A_{1,\alpha,k}(x)>\varepsilon\right\}\right|\\
&\le \left(\frac{\|A_{1,\alpha,k}\|_{L^{\frac{n}{n-\alpha},\infty}(\R^n)}}{\varepsilon}\right)^{\frac{n}{n-\alpha}} \\
&\le \left(\frac{C_n\|f-\varphi_k\|_{W^{1,1}(\R^n)}}{\varepsilon}\right)^{\frac{n}{n-\alpha}}\\
& < \frac{\eta}{3}.
\end{align*}
Applying Chebyshev's inequality to $A_{3,k}$, we also obtain
\begin{align*}
\left|\left\{x\in E:\ A_{3,k}(x)>\varepsilon\right\}\right| \le & \left(\frac{\|A_{3,k}\|_{L^{\frac{n}{n-1},\infty}(\R^n)}}{\varepsilon}\right)^{\frac{n}{n-1}} \\
\le & \left(\frac{C_n \|f-\varphi_k\|_{W^{1,1}(\R^n)}}{\varepsilon}\right)^{\frac{n}{n-1}} \\
<& \frac{\eta}{3}.
\end{align*}
%
For this fixed choice of $k$, since $\varphi_k \in C_c^\infty(\mathbb{R}^n)$, we may apply Theorem~\ref{thm:1} to conclude that $A_{2,\alpha,k}(x) \to 0$ for every $x \in \mathbb{R}^n$ as $\alpha \to 1^-$. Since $|E|<\infty$, pointwise convergence implies convergence in measure on $E$. Thus, there exists $\alpha_0\in(\tfrac12,1)$ (depending on $\varepsilon$ and $\eta$) such that for every $\alpha\in(\alpha_0,1)$,
\begin{equation*}
\left|\left\{x\in E:\ A_{2,\alpha,k}(x)>\varepsilon\right\}\right|<\frac{\eta}{3}.
\end{equation*}
Combining the estimates for $A_{1,\alpha,k}$, $A_{2,\alpha,k}$, and $A_{3,k}$ yields
\begin{equation*}
\left|\left\{x\in E:\ \left|(1-\alpha) I_\alpha(\mathcal{D}^\alpha f)(x)-K_n I_1(|\nabla f|)(x)\right|>3\varepsilon\right\}\right|<\eta
\end{equation*}
for every $\alpha\in(\alpha_0,1)$. Since $\varepsilon, \eta>0$ were arbitrary, this proves that $(1-\alpha) I_\alpha(\mathcal{D}^\alpha f)\to K_n I_1(|\nabla f|)$ in measure on every compact set $E\subset \R^n$ as $\alpha\to1^-$.
Finally, let $\{\alpha_k\}_{k}\subset(0,1)$ be any sequence such that $\alpha_k\to1^-$. By the previous step, the sequence $\{(1-\alpha_k) I_{\alpha_k}(\mathcal{D}^{\alpha_k}f)\}_{k}$ converges to $K_n I_1(|\nabla f|)$ in measure on every compact subset of $\R^n$. Therefore, there exists a subsequence $\{\alpha_{k_j}\}_{j}$ such that
\begin{equation*}
(1-\alpha_{k_j}) I_{\alpha_{k_j}}(\mathcal{D}^{\alpha_{k_j}}f)(x) \to K_n I_1(|\nabla f|)(x)
\end{equation*}
for almost every $x\in\R^n$. This completes the proof.
\end{proof}
\section{Further extensions}
In this final section, we record a natural $L^p$-variant of the nonlinear fractional differential operator. For $0<\alpha<1$ and $p\ge1$, we define
\begin{equation*}
\mathcal{D}^\alpha_p f(x):=\left(\int_{\R^n}\frac{|f(x)-f(y)|^p}{|x-y|^{n+\alpha p}}\,dy\right)^{\frac{1}{p}}.
\end{equation*}
This operator (in particular for $p=2$) appears, for instance, in \cite{NahasPonce}. This operator also behaves as a differential operator; it satisfies the following Leibniz-type inequality \begin{equation*}
\norm{\mathcal{D}^\alpha_p (fg)}_{L^p(\R^n)}\le \norm{f\, \mathcal{D}^\alpha_p g}_{L^p(\R^n)}+\norm{g\, \mathcal{D}^\alpha_p f}_{L^p(\R^n)}.
\end{equation*}
We next state the corresponding pointwise BBM limit for the Riesz potential of $\mathcal{D}^\alpha_p$.
\begin{theorem}
Let $p\ge1$ and let $f\in C_c^\infty(\R^n)$. Then for every $x\in\R^n$,
\begin{equation*}
\lim_{\alpha\to 1^-} (1-\alpha)^{\frac{1}{p}}\, I_\alpha(\mathcal{D}^\alpha_p f)(x) =K_{n,p}\, I_1(|\nabla f|)(x),
\end{equation*}
where
\begin{equation*}
K_{n,p}:=\left( \frac{1}{p} \int_{\mathbb{S}^{n-1}}|\omega\cdot e|^p\,d\sigma(\omega) \right)^\frac{1}{p} .
\end{equation*}
\end{theorem}
\begin{proof}[Sketch of the proof]
The proof follows the strategy of Theorem \ref{thm:1}. By translation invariance, it suffices to consider $x=0$. Setting
\begin{equation*}
F_{\alpha,p}(y):=\gamma_{n,\alpha}\,(1-\alpha)^{\frac1p}\,\frac{\mathcal{D}^\alpha_p f(y)}{|y|^{n-\alpha}},
\qquad
F(y):=K_{n,p}\,\gamma_{n,1}\,\frac{|\nabla f(y)|}{|y|^{n-1}},
\end{equation*}
it is enough to justify that $\lim_{\alpha\to 1^-}\int_{\R^n}F_{\alpha,p}=\int_{\R^n}F$. The pointwise convergence $F_{\alpha,p}(y)\to F(y)$ is provided by \cite[Lemma 13]{Kaushik} (in place of Lemma~\ref{lem:1}) together with the continuity of $\gamma_{n,\alpha}$ in $\alpha$. The required domination to apply the dominated convergence theorem is obtained by the same splitting estimates as in the proof of Theorem \ref{thm:1}, adapted to $\mathcal{D}^\alpha_p f$.
\end{proof}
%\section*{Acknowledgment}
%\section*{Conflicts of Interest}
%The author has no conflicts of interest to declare.
%\section*{Data Availability Statement}
%Data sharing is not applicable to this article, as no datasets were generated or analyzed during the current study.
\begin{bibdiv}
\begin{biblist}
\bib{AdamsFournier}{book}{
author={Adams, Robert A.},
author={Fournier, John J. F.},
title={Sobolev spaces},
series={Pure and Applied Mathematics (Amsterdam)},
volume={140},
edition={2},
publisher={Elsevier/Academic Press, Amsterdam},
date={2003},
pages={xiv+305},
isbn={0-12-044143-8},
review={\MR{2424078}},
}
\bib{Orlicz}{article}{
author={Alberico, Angela},
author={Cianchi, Andrea},
author={Pick, Lubo\v s},
author={Slav\'ikov\'a, Lenka},
title={On the limit as $s \to 1^-$ of possibly non-separable fractional
Orlicz-Sobolev spaces},
journal={Atti Accad. Naz. Lincei Rend. Lincei Mat. Appl.},
volume={31},
date={2020},
number={4},
pages={879--899},
issn={1120-6330},
review={\MR{4215683}},
doi={10.4171/rlm/918},
}
\bib{Orlicz2}{article}{
author={Alberico, Angela},
author={Cianchi, Andrea},
author={Pick, Lubo\v s},
author={Slav\'ikov\'a, Lenka},
title={Fractional Orlicz-Sobolev embeddings},
language={English, with English and French summaries},
journal={J. Math. Pures Appl. (9)},
volume={149},
date={2021},
pages={216--253},
issn={0021-7824},
review={\MR{4239001}},
doi={10.1016/j.matpur.2020.12.007},
}
\bib{Orlicz3}{article}{
author={Alberico, Angela},
author={Cianchi, Andrea},
author={Pick, Lubo\v s},
author={Slav\'ikov\'a, Lenka},
title={On fractional Orlicz-Sobolev spaces},
journal={Anal. Math. Phys.},
volume={11},
date={2021},
number={2},
pages={Paper No. 84, 21},
issn={1664-2368},
review={\MR{4237489}},
doi={10.1007/s13324-021-00511-6},
}
\bib{KKP}{article}{
author={Bal, Kaushik},
author={Mohanta, Kaushik},
author={Roy, Prosenjit},
title={Bourgain-Brezis-Mironescu domains},
journal={Nonlinear Anal.},
volume={199},
date={2020},
pages={111928, 10},
issn={0362-546X},
review={\MR{4093822}},
doi={10.1016/j.na.2020.111928},
}
\bib{BBM1}{article}{
author={Bourgain, Jean},
author={Brezis, Haim},
author={Mironescu, Petru},
title={Another look at Sobolev spaces},
conference={
title={Optimal control and partial differential equations},
},
book={
publisher={IOS, Amsterdam},
},
isbn={1-58603-096-5},
date={2001},
pages={439--455},
review={\MR{3586796}},
}
\bib{BBM2}{article}{
author={Bourgain, Jean},
author={Brezis, Ha\"im},
author={Mironescu, Petru},
title={Limiting embedding theorems for $W^{s,p}$ when $s\uparrow1$ and
applications},
note={Dedicated to the memory of Thomas H.\ Wolff},
journal={J. Anal. Math.},
volume={87},
date={2002},
pages={77--101},
issn={0021-7670},
review={\MR{1945278}},
doi={10.1007/BF02868470},
}
\bib{BSY}{article}{
author={Brazke, Denis},
author={Schikorra, Armin},
author={Yung, Po-Lam},
title={Bourgain-Brezis-Mironescu convergence via Triebel-Lizorkin spaces},
journal={Calc. Var. Partial Differential Equations},
volume={62},
date={2023},
number={2},
pages={Paper No. 41, 33},
issn={0944-2669},
review={\MR{4525722}},
doi={10.1007/s00526-022-02382-6},
}
\bib{BrezisConstant}{article}{
author={Brezis, Haim},
title={How to recognize constant functions. A connection with Sobolev
spaces},
language={Russian, with Russian summary},
journal={Uspekhi Mat. Nauk},
volume={57},
date={2002},
number={4(346)},
pages={59--74},
issn={0042-1316},
translation={
journal={Russian Math. Surveys},
volume={57},
date={2002},
number={4},
pages={693--708},
issn={0036-0279},
},
review={\MR{1942116}},
doi={10.1070/RM2002v057n04ABEH000533},
}
\bib{BrezisBook}{book}{
author={Brezis, Haim},
title={Functional analysis, Sobolev spaces and partial differential
equations},
series={Universitext},
publisher={Springer, New York},
date={2011},
pages={xiv+599},
isbn={978-0-387-70913-0},
review={\MR{2759829}},
}
\bib{BrezisMironescu}{article}{
author={Brezis, Ha\"im},
author={Mironescu, Petru},
title={Gagliardo-Nirenberg inequalities and non-inequalities: the full
story},
journal={Ann. Inst. H. Poincar\'e{} C Anal. Non Lin\'eaire},
volume={35},
date={2018},
number={5},
pages={1355--1376},
issn={0294-1449},
review={\MR{3813967}},
doi={10.1016/j.anihpc.2017.11.007},
}
\bib{DachunYang}{article}{
author={Dai, Feng},
author={Grafakos, Loukas},
author={Pan, Zhulei},
author={Yang, Dachun},
author={Yuan, Wen},
author={Zhang, Yangyang},
title={The Bourgain-Brezis-Mironescu formula on ball Banach function
spaces},
journal={Math. Ann.},
volume={388},
date={2024},
number={2},
pages={1691--1768},
issn={0025-5831},
review={\MR{4700381}},
doi={10.1007/s00208-023-02562-5},
}
\bib{DominguezMilman}{article}{
author={Dom\'inguez, Oscar},
author={Milman, Mario},
title={Bourgain-Brezis-Mironescu-Maz\cprime ya-Shaposhnikova limit
formulae for fractional Sobolev spaces via interpolation and
extrapolation},
journal={Calc. Var. Partial Differential Equations},
volume={62},
date={2023},
number={2},
pages={Paper No. 43, 37},
issn={0944-2669},
review={\MR{4525724}},
doi={10.1007/s00526-022-02383-5},
}
\bib{DD2}{article}{
author={Drelichman, Irene},
author={Dur\'an, Ricardo G.},
title={Improved Poincar\'e{} inequalities in fractional Sobolev spaces},
journal={Ann. Acad. Sci. Fenn. Math.},
volume={43},
date={2018},
number={2},
pages={885--903},
issn={1239-629X},
review={\MR{3839841}},
doi={10.5186/aasfm.2018.4352},
}
\bib{DrelichmanDuran}{article}{
author={Drelichman, Irene},
author={Dur\'an, Ricardo G.},
title={The Bourgain-Br\'ezis-Mironescu formula in arbitrary bounded
domains},
journal={Proc. Amer. Math. Soc.},
volume={150},
date={2022},
number={2},
pages={701--708},
issn={0002-9939},
review={\MR{4356180}},
doi={10.1090/proc/15665},
}
\bib{FernandezSalort}{article}{
author={Fern{\'a}ndez Bonder, Julian},
author={Salort, Ariel},
title={Asymptotic behavior for anisotropic fractional energies},
journal={Israel Journal of Mathematics},
date={2025},
issn={1565-8511},
doi={10.1007/s11856-025-2892-1},
url={https://doi.org/10.1007/s11856-025-2892-1}
}
\bib{FLW}{article}{
author={Franchi, Bruno},
author={Lu, Guozhen},
author={Wheeden, Richard L.},
title={A relationship between Poincar\'e-type inequalities and
representation formulas in spaces of homogeneous type},
journal={Internat. Math. Res. Notices},
date={1996},
number={1},
pages={1--14},
issn={1073-7928},
review={\MR{1383947}},
doi={10.1155/S1073792896000013},
}
\bib{HMP2}{article}{
author={Hoang, Cong},
author={Moen, Kabe},
author={P\'erez, Carlos},
title={Pointwise estimates for rough operators with applications to
Sobolev inequalities},
journal={J. Anal. Math.},
volume={155},
date={2025},
number={1},
pages={43--74},
issn={0021-7670},
review={\MR{4905307}},
doi={10.1007/s11854-024-0345-0},
}
\bib{HMP}{article}{
author={Hoang, Cong},
author={Moen, Kabe},
author={P\'erez, Carlos},
title={A new look at the subrepresentation formulas},
journal={Matematica},
volume={4},
date={2025},
number={2},
pages={293--309},
review={\MR{4930303}},
doi={10.1007/s44007-025-00151-3},
}
\bib{HLYY}{article}{
author={Hu, Pingxu},
author={Li, Yinqin},
author={Yang, Dachun},
author={Yuan, Wen},
title={A sharp localized weighted inequality related to Gagliardo and
Sobolev seminorms and its applications},
journal={Adv. Math.},
volume={481},
date={2025},
pages={Paper No. 110537, 66},
issn={0001-8708},
review={\MR{4961027}},
doi={10.1016/j.aim.2025.110537},
}
\bib{HMPV}{article}{
author={Hurri-Syrj\"anen, Ritva},
author={Mart\'inez-Perales, Javier C.},
author={P\'erez, Carlos},
author={V\"ah\"akangas, Antti V.},
title={On the BBM-phenomenon in fractional Poincar\'e-Sobolev
inequalities with weights},
journal={Int. Math. Res. Not. IMRN},
date={2023},
number={20},
pages={17205--17244},
issn={1073-7928},
review={\MR{4659850}},
doi={10.1093/imrn/rnac246},
}
\bib{Kaushik}{article}{
author={Mohanta, Kaushik},
title={Bourgain-Brezis-Mironescu formula for $W^{s,p}_q$-spaces in
arbitrary domains},
journal={Calc. Var. Partial Differential Equations},
volume={63},
date={2024},
number={2},
pages={Paper No. 31, 17},
issn={0944-2669},
review={\MR{4686398}},
doi={10.1007/s00526-023-02637-w},
}
\bib{NahasPonce}{article}{
author={Nahas, J.},
author={Ponce, G.},
title={On the persistent properties of solutions to semi-linear
Schr\"odinger equation},
journal={Comm. Partial Differential Equations},
volume={34},
date={2009},
number={10-12},
pages={1208--1227},
issn={0360-5302},
review={\MR{2581970}},
doi={10.1080/03605300903129044},
}
\bib{SpectorJFA}{article}{
author={Spector, Daniel},
title={An optimal Sobolev embedding for $L^1$},
journal={J. Funct. Anal.},
volume={279},
date={2020},
number={3},
pages={108559, 26},
issn={0022-1236},
review={\MR{4093790}},
doi={10.1016/j.jfa.2020.108559},
}
\bib{ZhuYangYuan2023}{misc}{
author={Zhu, Chenfeng},
author={Yang, Dachun},
author={Yuan, Wen},
title={Extension theorem and {B}ourgain--{B}rezis--{M}ironescu-type characterization of ball {B}anach {S}obolev spaces on domains},
eprint={2307.11392},
archiveprefix={arXiv},
primaryclass={math.FA},
date={2023},
note={Available at \url{https://arxiv.org/abs/2307.11392}},
}
\end{biblist}
\end{bibdiv}
\end{document}