0

I'm trying to solve below exercise in Brezis' Functional Analysis, i.e.,

Exercise 4.8 Let $(\Omega, \mathcal F, \mu)$ be a $\sigma$-finite measure space. Let $X$ be a closed vector subspace of $L^1 (\Omega)$ such that $$ X \subset \bigcup_{q \in (1, \infty]} L^q (\Omega). $$

  1. Prove that there is $p>1$ such that $X \subset L^p (\Omega)$.
  2. Prove that there is $C \in [0, \infty)$ such that $$ \|f\|_p \le C \|f\|_1 \quad \forall f \in X. $$

Could you confirm if my below attempt is correct?


Proof 1. It follows from $X$ is closed in $L^1 (\Omega)$ that $X$ is complete w.r.t. $\| \cdot \|_1$. For each $n \in \mathbb N^*$, $$ \begin{align} Y_n &:= \{f \in L^1 (\Omega) : \|f\|_{1+(1/n)} \le n\} \\ X_n &:= X \cap Y_n . \end{align} $$

We have $Y_n$ is closed in $L^1 (\Omega)$. Then $X_n$ is closed in $(X, \|\cdot\|_1)$. Let $f \in X \cap L^q (\Omega)$ for some $q>1$. By interpolation inequality, $$ \|f\|_r \le \|f\|_1^\alpha \|f\|_q^{1-\alpha} \quad \forall r \quad \text{s.t.} \quad \frac{1}{r} = \frac{\alpha}{1} + \frac{1-\alpha}{q}. $$

So $X = \bigcup_n X_n$. By Baire category theorem, there is $p$ such that $$ \operatorname{int}_{(X, \|\cdot\|_1)} (X_p) \neq \emptyset, $$ where $\operatorname{int}_{(X, \|\cdot\|_1)} (X_p)$ is the interior of $X_p$ in $(X, \|\cdot\|_1)$. Fix $f \in \operatorname{int}_{(X, \|\cdot\|_1)} (X_p)$. There is $r>0$ such that $$ B_{(X, \|\cdot\|_1)} (f, r) \subset X_p \subset L^p(\Omega). $$

Here $B_{(X, \|\cdot\|_1)} (f, r)$ is the open ball centered at $f$ with radius $r$ in $(X, \|\cdot\|_1)$. Let $g \in X$ be arbitrary. Because $X$ is a vector space, there is $\lambda>0$ such that $$ \lambda f + (1-\lambda)g \in B_{(X, \|\cdot\|_1)} (f, r). $$

So $\lambda f + (1-\lambda)g \in L^p(\Omega)$. We already have $f \in L^p(\Omega)$. Thus $g \in L^p(\Omega)$. Hence $X \subset X_p$.

  1. Consider the embedding $$ T:X \to L^p (\Omega), f \mapsto f. $$

Then $T$ is linear. It suffices to prove that $T$ is continuous. Notice that $X$ and $L^p (\Omega)$ are both Banach spaces. By closed graph theorem, it suffices to prove that the graph of $T$ is closed.

Let $u,u_n \in X$ and $f \in L^p (\Omega)$ such that $(u_n, u_n) \to (u, f)$ in $X \times L^p$. There is a sub-sequence $\varphi$ of $\mathbb N$ such that $u_{\varphi (n)} \to u$ $\mu$-a.e. Clearly, $(u_{\varphi(n)})$ is a Cauchy sequence in $L^p (\Omega)$. So $u_{\varphi (n)} \to u$ in $L^p$. Hence $u=f$ $\mu$-a.e. This completes the proof.

Mittens
  • 39,145
Akira
  • 17,367
  • Please ask one question at a time. – Shaun Apr 04 '23 at 15:07
  • 1
    Your solution is hard to read because you refer to exercises that you have solved before and make it hard to go back and forth from link to link. It is probably more helpful to the reader that as you explain your arguments, you provide a general justification to it and then the link to the corresponding posting. For example,
    1. The fact that $Y_n$ is closed in $L_1$ follows by an application of Fatou's Lemma and Hölder's inequality (see [your_link1]).
    – Mittens Apr 04 '23 at 16:04
  • @OliverDíaz You're right. I should have done that... – Akira Apr 04 '23 at 16:06
  • 1
    Another observation. In part (1), the $p$ (which in the context of your argument is an integer) fir which $\operatorname{int}(X_p)\neq\emptyset$ is not the $p$ for which $X\subset L_p$. It is $P=1+\tfrac1p$ the index for which $X\subset L_P$. – Mittens Apr 04 '23 at 16:18
  • @OliverDíaz Yess, it should be $B_{(X, |\cdot|_1)} (f, r) \subset X_p \subset L^{1 + (1/p)}(\Omega)$. – Akira Apr 04 '23 at 16:19

1 Answers1

1

The proof that $X=\bigcup_nX_n$ is not very clear to me. Here is a short proof:

If $f\in X$, then $f\in L_1\cap L_p$ for some $1<p\leq\infty$ and so, $f\in L_r$ for all $1<r<p$. Without loss of generality, we may assume that $1<p<\infty$. By choosing $n\in\mathbb{N}$ large enough so that $\|f\|_1\vee\|f\|_p<n$ and $1+\tfrac1n<p$ we obtain $f\in L_{1+1/n}$.

Here we use the fact that if $0<p<q\leq \infty$, then $L_p\cap L_q\subset L_r$ for all $p<r<q$, and $\|f\|_r\leq \|f\|_p\vee\|f\|_r$, which follows from Hölder's inequality.


Some other observations:

  1. In part (1), the $p$ (which in the context of your argument is an integer) fir which $\operatorname{int}(X_p)\neq\emptyset$ is not the $p$ for which $X\subset L_p$. It is $q:=1+\tfrac1p$ the index for which $X\subset L_q$.
  2. in part (b), the fat that $u_{\phi(n)}$ is Cauchy in $L_q$ ($q=1+\tfrac1p$) is not relevant. You have that $u_{\phi(n)}\xrightarrow{n\rightarrow\infty} u$ $\mu-a.s.$ and that $u_{\phi(n)}\xrightarrow{n\rightarrow\infty}f$ in $L_p$ (since $u_n$ converges to $f$ in $L_p$). Then, along a subsequence $psi(n)$ of $\phi$ $u_{psi(n)}\xrightarrow{n\rightarrow\infty}f$ $\mu$-a.s. It then follows that $u=f$ $\mu$-a.s.

Mittens
  • 39,145
  • In "$u_{\phi(n)}\xrightarrow{n\rightarrow\infty}f$ in $L_p$ (since $u_n$ converges to $f$ in $L_p$)", did you mean $q$ rather than $p$? – Akira Apr 04 '23 at 16:59
  • I usually use Cauchy sequence for $L^p$ space because of the definition of Bochner integral for vector-valued functions... – Akira Apr 04 '23 at 17:02
  • 1
    @Akira: Yes, $q$ instead of $p$. The Bochner integral is an overkill here. Once we start to consider vector valued functions then we have to care about strongly-measurability issues and things become more subtle. The exercises are about (abstract) Lebesgue integration, – Mittens Apr 04 '23 at 17:55
  • 1
    @Akira: here is a problem that may be of interest you based on your most recent questions. – Mittens Apr 07 '23 at 16:23