Skip to main content
2 of 4
added 6 characters in body
Iosif Pinelis
  • 127.9k
  • 8
  • 107
  • 229

$\newcommand{\R}{\mathbb R}\newcommand{\EE}{\mathcal E}\newcommand{\de}{\delta}\newcommand\ep\varepsilon $The answer is yes, and the key here is the Dunford--Pettis theorem (Theorem 3.1). According to this theorem, the condition that the sequence $(f_n)$ converges weakly in $L^1(\R)$ implies that this sequence is equi-bounded and equi-integrable. So, (i) \begin{equation*} M:=\sup_n\|f_n\|_1<\infty \tag{1}\label{1} \end{equation*} and (ii) for each real $\ep>0$ there exist a measurable set $A_\ep\subseteq\R$ with Lebesgue measure $|A_\ep|<\infty$ and a real number $\de_\ep>0$ such that \begin{equation*} \sup_n\int_{A_\ep^c}|f_n|\le\ep \tag{2}\label{2} \end{equation*} and \begin{equation*} \sup_n\sup_{E\in\EE_\de}\int_E |f_n|\le\ep, \tag{3}\label{3} \end{equation*} where $A_\ep^c:=\R\setminus A_\ep$ and $\EE_\de$ is the set of all measurable subsets $E$ of $\R$ with Lebesgue measure $|E|\le\de$.

Without loss of generality (wlog), the weak limit $f$ of the sequence $(f_n)$ is $0$, so that \begin{equation*} I_n(h):=\int_\R f_n h\to0 \end{equation*} for all $h\in L^\infty(\R)$ (as $n\to\infty$). We have to show that \begin{equation*} J_n(h):=\int_\R g_n h\to0 \end{equation*} for all $h\in L^\infty(\R)$.

For any $h\in L^\infty(\R)$, we have $$J_n(h)=\int_\R g_n(x)h(x)\,dx \\ =\int_\R f_n(x-a_n)h(x)\,dx=\int_\R f_n(y)h(y+a_n)\,dy$$ and hence \begin{equation*} J_n(h)-I_n(h)=D_n(\R,h), \quad\text{where}\quad D_n(A,h):=\int_A f_n(x)[h(x+a_n)-h(x)]\,dx. \end{equation*} It remains to show that $D_n(\R,h)\to0$ for all $h\in L^\infty(\R)$.

Take any real $\ep>0$, and let $A_\ep$ and $\de_\ep$ be as in \eqref{2} and \eqref{3}. By \eqref{2},
\begin{equation*} |D_n(\R,h)-D_n(A_\ep,h)|\le2\ep\|h\|_\infty. \tag{4}\label{4} \end{equation*} So, it is enough to show that $D_n(A_\ep,h)\to0$ for all $h\in L^\infty(\R)$.

The measurable set $A_\ep$ of finite measure can be approximated in measure by a finite union of finite intervals. So, by \eqref{3}, it is enough to show that \begin{equation*} D_n(l,h)\overset{\text{(?)}}\to0 \tag{5}\label{5} \end{equation*} for any finite interval $l$ and any $h\in L^\infty(\R)$. Let $L$ be a finite interval containing $l\cup\bigcup_n(a_n+l)$.

The function $h\in L^\infty(\R)$ can be uniformly approximated on the interval $L$ by a linear combination of indicators of measurable sets. So, in view of \eqref{1}, wlog $h=1_B$ for some measurable subset $B$ of the interval $L$.

The measurable set $B$ can be approximated in measure by a finite union of finite intervals. So, again by \eqref{3}, wlog $B=[u,v]$ for some real $u$ and $v$, and then $h=1_{[u,v]}$. Therefore and, again, in view of \eqref{3}, \begin{equation*} |D_n(l,h)| \le\int_{[u,v]+[u-a_n,v-a_n]} |f_n|\to0, \end{equation*} where $[u,v]+[u-a_n,v-a_n]$ denotes the symmetric difference of $[u,v]$ and $[u-a_n,v-a_n]$.

Thus, \eqref{5} follows, as desired.

Iosif Pinelis
  • 127.9k
  • 8
  • 107
  • 229