4
$\begingroup$

I found an interesting property in some lecture notes on weak convergence: lemma 3.2 (2) on page 10: https://www.uio.no/studier/emner/matnat/math/MAT4380/v06/Weakconvergence.pdf . The idea is as follows:

Let $(X,\mu)$ be a $\sigma$ finite measure space and $(f_n)_{n\in\mathbb{N}^*}, (g_n)_{n\in\mathbb{N}^*}, f,g :X\to \overline{\mathbb{R}}$ are $\mu$ measurable functions. Prove that if :

$\bullet$ There is a function $f:X\to\overline{\mathbb{R}}$ such that $\lim\limits_{n\to\infty} f_n(x)= f(x),$ for $\mu$ - almost all $x\in X$. (pointwise convergence);

$\bullet$ There is some positive constant $M>0$ such that: $||f_n||_{\infty}:=\mathrm{ess sup}\ |f_n|<M, \forall n\in\mathbb{N}^*$;

$\bullet$ $g_n\rightharpoonup g$ in $L^1(X,\mu)$,

then $f_n g_n\rightharpoonup fg$ in $L^1(X,\mu)$ i.e. $\lim\limits_{n\to\infty} \displaystyle\int_{X} f_n g_n h\ d\mu=\int_{X} fgh\ d\mu, \forall h\in L^{\infty}(X,\mu)$

I tried to prove this by definition but I cannot make good estimates on $\int_{X} |g_n||f_n-f|\ d\mu$. In the lecture notes sais that this is a simple property and the proof is omitted...Any help is very welcomed.

$\endgroup$
1

1 Answer 1

3
$\begingroup$

The missing argument is a combination of Egorov's theorem and the Dunford-Pettis theorem (for the precise versions of both that we are going to use, see [Brezis, Haim, Functional analysis, Sobolev spaces and partial differential equations, Springer (2011), theorems 4.29 and 4.30 page 115 ]). Roughly speaking, the former tells us that $f_n\to f$ uniformly, up to removing arbitrarily small sets from $X$. And the latter guarantees that $|g_n|$ gives small mass to such small sets, uniformly in $n$ and only depending on the measure of the small set.

Disclaimer: as correctly pointed out by Nik Weaver in his comment, Egorov's theorem crucially requires a finite measure $\mu(X)<+\infty$. Step 1 below gives the key argument in this finite situation. Then Johannes Schürz's comment settles in step 2 the general case, based on the uniform integrability.

Step 1: assume first that $\mu(X)<+\infty$, and pick any $\epsilon>0$. By Egorov's theorem there exists a subset $X_\epsilon\subset X$ with small complement $\mu(X\setminus X_\epsilon)\leq \epsilon$ such that $f_n\to f$ uniformly in $X_{\epsilon}$, in particular also $f_nh\to fh$. Thus by standard $L^\infty-L^1$ strong-weak convergence the term $$ \int_{X_\epsilon} f_ng_nhd\mu\to \int _{X_\epsilon}fg hd\mu. $$ For the remaining term (the integral on $X\setminus X_\epsilon$), the Dunford-Pettis theorem guarantees that $\{g_n\}_n$ is uniformly integrable. Given $\delta>0$, this means that $\int_{X\setminus X_\epsilon} |g_n|\leq \delta$ uniformly in $n$, as soon as $\mu(X\setminus X_\epsilon)\leq \epsilon$ is sufficiently small. Since $|f_n|_\infty\leq M$ and $h\in L^\infty$ this immediately gives $$ \left| \int_{X\setminus X_\epsilon} f_ng_nhd\mu \right|\leq M\|h\|_\infty \delta $$ Putting everything together and playing a bit with $\epsilon,\delta,n\geq n_0$, and quantifiers gives the desired result (also noticinng that $\int _{X_\epsilon}fg hd\mu\to \int _{X}fg hd\mu$ if $\epsilon\to 0$).

Step 2: assume now that $X$ has infinite measure. Since the sequence $g_n$ is $L^1$-weakly converging, it is uniformly integrable (by the Dunford-Pettis theorem) and therefore for any small $\eta>0$ there exists $X_\eta\subset X$ with $\mu(X_\eta)<+\infty$ and $$ \int_{X\setminus X_\eta} |g_n|d\mu \leq \eta \qquad \forall\, n\geq 0 $$ As a consequence $$ \left| \int_{X\setminus X_\eta} f_n g_n hd\mu \right| \leq \|f_n\|_{L^\infty(X)} \|h\|_{L^\infty(X)} \|g_n\|_{L^1(X\setminus X_\eta)}\leq M\|h\|_{L^\infty(X)}\eta $$ can be made arbitrarily small, uniformly in $n$. The result follows next by applying step 1 on the finite measure set $X_\eta$.

PS: I was not aware of this specific statement, and it may turn out to be quite handy at some point so thank you Maxim Diana!

$\endgroup$
6
  • $\begingroup$ Doesn't Egorov only hold for finite measure spaces? $\endgroup$
    – Nik Weaver
    Apr 16, 2020 at 10:56
  • $\begingroup$ Fait point, thanks for pointing this out! At least the argument works for $\mu(X)<+\infty$. Perhaps considering comapctly supported test-functions $h$, and then explointing the $\sigma$-finiteness as usual? $\endgroup$ Apr 16, 2020 at 11:49
  • $\begingroup$ Maybe a change-of-density argument can make one pretend that the measure is finite? $\endgroup$ Apr 16, 2020 at 20:33
  • 2
    $\begingroup$ Use that uniformly integrable also gives you that $\forall \varepsilon >0 \,\, \exists A \subset X \colon \mu(A) < +\infty \land \forall n \in \mathbb{N} \,\, \Vert g_n \cdot \mathbb{1}_A\Vert_{L^1(X,\mu)} < \varepsilon$. Now you can use Egorov for $L^1(A, \mu)$. $\endgroup$ Apr 16, 2020 at 22:22
  • 1
    $\begingroup$ Thank you Johannes, this works indeed! $\endgroup$ Apr 17, 2020 at 15:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.