5

In the book of PDE, evans, the proof of the characterization of $W^{1,\infty}$ uses the following argument:

Suppose $u$ has compact support and Lipchitz continuous. Then $$ \Vert D_i^{-h}u\Vert_{L^\infty(\mathbb{R}^n)}\leq Lip(u), $$ where $D_i^{-h}u$ the difference quotient defined by $$ D_i^hu=\frac{u(x+he_i)-u(x)}{h}. $$ Then there exists a function $v_i\in L^\infty(\mathbb{R}^n)$ and a subsequence $h_k\rightarrow0$ such that $$ D_i^{-h_k}u\rightharpoonup v_i \quad\text{weakly in }L^2_{loc}(\mathbb{R}^n). $$

Question: I don't know why such a subsequence exists. Since $u$ has some compact support $\Omega$, then $u\in L^2(\Omega)$. Then there exists a function $v_i\in L^2(\Omega)$ such that $$ D_i^{-h_k}u\rightharpoonup v_i \quad\text{weakly in }L^2(\Omega). $$ But I don't know why this $v_i\in L^\infty$.

Q-Y
  • 1,589
  • Good question and good answer. It is uncharacteristic of Evans's books to brush such a key observation under the rug. They have the same proof in measure theory and fine properties of functions book and there too this is not addressed. – Behnam Esmayli Jul 12 '22 at 19:52

1 Answers1

7

Note that the $D_i^{-h_k}$ are all bounded in $L^\infty(\Omega)$ by the Lipschitz constant of $u$; call this constant $M$.

Claim. Suppose $f_k \in L^2(\Omega)$, $f_k \rightharpoonup f$ in $L^2(\Omega)$, and $\|f_k\|_{L^\infty} \le M$ for all $k$. Then $\|f\|_{L^\infty} \le M$ as well.

Proof. Fix $\epsilon > 0$ and let $A = \{ f \ge M+\epsilon \}$. Since $1_A$ is bounded, it is therefore in $L^2$, so by weak convergence we have $$\int_A f_k = \int_\Omega 1_A f_k \to \int_\Omega 1_A f = \int_A f.$$ However, since $\|f_k\|_{L^\infty} \le M$, we have $\int_A f_k \le M \mu(A)$. And since $f \ge M+\epsilon$ on $A$, we have $\int_A f \ge (M+\epsilon) \mu(A)$. The only way to reconcile these statements is to have $\mu(A) = 0$, which is to say that $f < M+\epsilon$ almost everywhere. Letting $\epsilon \downarrow 0$ along a sequence, we have $f \le M$ almost everywhere. A similar argument will show that $f \ge -M$ almost everywhere.

Alternate proof. Since $L^1 \cap L^2$ is dense in $L^1$, by $L^p$ duality we have $$\|f\|_{L^\infty} = \sup\left\{\int_\Omega fg : g \in L^1 \cap L^2, \|g\|_{L^1} \le 1\right\}.$$ So suppose $g \in L^1 \cap L^2$ with $\|g\|_{L^1} \le 1$. We have $\int_\Omega f_k g \le M$ for every $k$, so by weak convergence, $\int_\Omega fg \le M$ also. Taking the supremum over $g$, we have $\|f\|_{L^\infty} \le M$.

Yet another proof. Let $B = \{h \in L^2 : \|h\|_{L^\infty} \le M\}$. I claim first $B$ is strongly closed in $L^2$. For suppose $\|h_k\|_{L^\infty} \le M$ and $h_k \to h$ in $L^2$. There is a subsequence $h_{k_j}$ which converges to $h$ almost everywhere. Since $|h_{k_j}| \le M$ almost everywhere, the same is therefore true of $h$. Now since $M$ is strongly closed and convex, by Mazur's lemma it is also weakly closed, which is the desired statement.

Nate Eldredge
  • 97,710