1
$\begingroup$

I'm reading notes on the closeability of differential operators where the author says let $\Omega\subset \mathbb{R}^n$, and then defines the operator $A:C_0^\infty(\Omega)\to L^p(\Omega)$ by $$ Au := \sum_{|\alpha|\le m}a_\alpha \partial^\alpha u, $$ and its formal adjoint $B:C_0^\infty(\Omega)\to L^q(\Omega)$ by $$ Bv := \sum_{|\alpha|\le m}(-1)^{|\alpha|}\partial^\alpha (a_\alpha v), $$ He then says that integration by parts shows that $$ \int_\Omega v(Au) = \int_\Omega (Bv) u, \quad \quad \text{for all} \ u,v \in C_0^\infty(\mathbb{R}^n). $$ Now let $u_k\in C_0^\infty(\Omega)$ be a sequence of smooth functions with compact support and let $v\in L^p(\Omega)$ such that $$ (*) \quad \lim_{k\to\infty} ||u_k||_{L^p} = 0, \quad \lim_{k\to\infty} ||v - Au_k||_{L^p} = 0. $$ Then, for every test function $\phi \in C_0^\infty(\Omega)$, we have $$ (**) \quad \int_\Omega \phi v = \lim_{k\to \infty} \int_\Omega \phi (Au_k) = \lim_{k\to \infty} \int_\Omega (B\phi) u_k = 0. $$ Since $C_0^\infty(\Omega)$ is dense in $L^q(\Omega)$, this implies that $\int_\Omega \phi v = 0$ for all $\phi \in L^q(\Omega)$.

  1. I don't see how the fact that $u_k$ converges to zero, and $v$ converges to $Au_k$ in the $L^p$ norm allows us to derive $(**)$ from $(*)$? The integrals in $(**)$ do not even use the $L^p$ norm, so how is statement $(**)$ valid?
  2. The author says that since $C_0^\infty(\Omega)$ is dense in $L^q(\Omega)$, this implies that $\int_\Omega \phi v = 0$ for all $\phi \in L^q(\Omega)$. How can I show this fact?
$\endgroup$
1
$\begingroup$

Strong convergence in a Banach space implies weak convergence. It means that if ${f}_{k} \rightarrow f$ in ${L}^{p} \left({\Omega}\right)$ and if $g \in {L}^{q} \left({\Omega}\right)$, then $\left\langle g , {f}_{k}\right\rangle \rightarrow \left\langle g , f\right\rangle $ where $\left\langle \ \right\rangle $ is the duality bracket. The reason is that

$$\left|\left\langle g , f\right\rangle -\left\langle g , {f}_{k}\right\rangle \right| = \left|\left\langle g , f-{f}_{k}\right\rangle \right| \leqslant {\left\|g\right\|}_{L ^q(\Omega)} {\left\|f-{f}_{k}\right\|}_{L ^p(\Omega)}$$

It turns out that the duality bracket between ${L}^{p} \left({\Omega}\right)$ and ${L}^{q} \left({\Omega}\right)$ is given by the integral

$$\left\langle g , f\right\rangle = \int_{{\Omega}}^{}g f d x$$

The (**) formula can be rewritten

$$\left\langle {\phi} , {\nu}\right\rangle = {\lim }_{k \rightarrow \infty } \left\langle {\phi} , A {u}_{k}\right\rangle = {\lim }_{k \rightarrow \infty } \left\langle B {\phi} , {u}_{k}\right\rangle = 0$$

Now the duality bracket is continuous, that is to say ${\phi} \rightarrow \left\langle {\phi} , {\nu}\right\rangle $ is continuous in ${L}^{q} \left({\Omega}\right)$. If it is zero on the dense subspace ${\mathscr{C}}_{0}^{\infty } {(\Omega)} \subset {L}^{q} \left({\Omega}\right)$, then it must be zero on all of ${L}^{q} \left({\Omega}\right)$.

$\endgroup$
  • $\begingroup$ Ok, that clears some things up thanks! Just to be sure I have the last part correct, is this a correct proof that if $\langle \phi, \nu \rangle = 0$ for $\phi$ in the dense subspace $C_0^\infty(\Omega)$, then it must be zero on all of $L^q(\Omega)$: Since $C_0^\infty(\Omega)$ is dense in $L^q(\Omega)$ we can take a sequence $\phi_k \in C_0^\infty(\Omega)$ s.t. $\phi_k \to \phi \in L^q(\Omega)$. Then we have $$\langle \phi, \nu \rangle = \langle \lim_k \phi_k, \nu \rangle = \lim_k \langle \phi_k, \nu \rangle = 0.$$ Is that right or have I missed some details? $\endgroup$ – ManUtdBloke Aug 30 '17 at 5:10
  • 1
    $\begingroup$ You didn't miss anything, it is correct. Note that the result is very weak: in the sense of distributions, it is true that if $u_n \to 0$ in ${\scr D}^\prime(\Omega)$, then $A u_n\to 0$ in ${\scr D}^\prime(\Omega)$. Now $u_n \to 0$ in $L^p(\Omega)$ implies $u_n \to 0$ in ${\scr D}^\prime(\Omega)$, and similarly $A u_n \to \nu$ in $L^p(\Omega)$ implies $A u_n \to \nu$ in ${\scr D}^\prime(\Omega)$, so that $\nu = 0$. $\endgroup$ – Gribouillis Aug 30 '17 at 6:42
1
$\begingroup$
  1. By Holder's inequality : $$\int_\Omega |\phi( v-Au_k)|=\|\phi( v-Au_k)\|_{L^1} \leq \| \phi \|_{L^{q}}\|v-Au_k \|_{L^{p}} \underset{k \to \infty}\to 0.$$
  2. Let $\psi$ any function of $L^q(\Omega)$ and $(\psi_k)$ a sequence of $C_0^\infty(\Omega)$ s.t. $\|\psi-\psi_k\|_{L^q}\underset{k \to \infty}\to 0$. $$\int_{\Omega}\psi v\overset{\delta}=\lim_{k\to \infty} \int_{\Omega}\psi_kv=0.$$ The equality $\delta$ comes from the same reasonning as before : $$\int_{\Omega}|(\psi-\psi_k) v|=\|(\psi-\psi_k) v\|_{L^1} \leq\|\psi-\psi_k\|_{L^q} \|v\|_{L^p} \underset{k \to \infty}\to 0.$$
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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