1
$\begingroup$

I'm trying to understand $(W_0^{1,p} (Ω))^*=W_0^{-1,p^*} (Ω)$, and what a proper representation of its elements is. I understand the basics such as: if $f∈L^{p^*} (Ω)⇒f∈W_0^{-1,p^*}(Ω)$ and the duality is given by $(f,u)=∫_Ωfudx$ $∀ u∈W_0^{1,p^*}(Ω). $

Now my question is the following: assume that $F∈ (W_0^{1,p} (Ω))^*$ but $F∉L^{p^*}(Ω)$ what is a proper representation of $F$ ?

here is my attempt:

$∀ F∈(W_0^{1,p} (Ω))^* ∃ f_0...f_n ∈L^{p^*} (Ω)$ such that $∀ u∈W_0^{1,p}(Ω)$

$F(u)=lim_{n→∞} ⟨f_0-∑_{i=1}^n∂_i f_i,ϕ_n ⟩_{D^*×D}$ (1)

Where:

$∂_i f_i$ is the distribution of derivative of $f_i$ with respect to the $i$ variable.

$⟨*,*⟩_{D^*×D}$ is the dual bracket of $D^* (Ω)×D(Ω)$

for this to be correct I have to justify the expression itself, and then prove that the particular choice of sequence is irrelevant

Here is what I think proves my claim:

Step 1 we can find an isometry between $W_0^{1,p}(Ω)$ and $(L^p (Ω))^{n+1}$. Without going too much into detail using this isometry we conclude that $∀F∈(W_0^{1,p}(Ω))^*∃ f_0...f_n ∈L^{p^*}(Ω)$ such that $∀ u∈W_0^{1,p}(Ω):$

$F(u)=∫_Ω uf_0dx+∑_{i=1}^N∫_Ωf_i ∂_i udx$ (2)

Step 2. this is in my opinion the tricky part. Firstly since we are working in $W_0^{1,p}(Ω)$ we can always choose a sequence of test functions $ϕ_n$ that converges to $u$ in $W_0^{1,p}(Ω)$.Now by continuity of F we have:

$F(u)=lim_{n→∞}⁡ ∫_Ω ϕ_n f_0dx+∑_{i=1}^N∫_Ω f_i ∂_i ϕ_n dx.$

However since these are test functions, we can consider the integration to be the duality bracket in $D^* (Ω)×D(Ω)$, thus we get.

$F(u)=lim_{n→∞} ⟨f_0 ,ϕ_n⟩_{D^*×D}+∑_{i=1}^N⟨f_i,∂_i ϕ_n ⟩_{D^*×D}$ (3)

by using the derivative of distribution we get (1) directly from (3). As far as I can tell the important part is that: convergence in$W_0^{1,p}(Ω)$ does not imply convergence in $D(Ω)$ even so the limit in (3) exists by the calculations in step 1 and step 2; namely we know that the limit it is equal to some $F$ acting on some $u$.

Step 3. we prove that the choice of sequence is irrelevant. Let $ψ_n$ be another sequence converging to $u$ in $W_0^{1,p}(Ω)$ by using (3) we get:

$lim_{n→∞} ⟨f_0,ϕ_n-ψ_n ⟩_{D^*×D}+∑_{i=1}^N ⟨f_i,∂_i ϕ_n-ψ_n ⟩_{D^*×D} =0$

Is my approach correct ? And if not can you please correct some mistakes.

$\endgroup$
5
  • $\begingroup$ Embed $W_0^{1,p}(\Omega)$ isometrically onto a subspace of $L^p(\Omega)^{n+1}$ via $f\mapsto (f,\partial_1f,\dots,\partial_nf)$, and use Hahn-Banach. This yields to a (not unique) representation for every linear form by $L^q(\Omega)^{n+1}$. $\endgroup$ Commented Dec 4 at 19:16
  • $\begingroup$ @PietroMajer sorry can you clarify ? the problem is that the representation si not unique ? but is (1) a correct representation ? $\endgroup$ Commented Dec 4 at 20:07
  • $\begingroup$ It is well understood how to characterize elements of dual spaces of the Sobolev spaces $W^{1,p}_0(\Omega)$, see for examples Proposition 9.20 in the book of Brezis. Essentially, one can represent it by $f - \operatorname{div} F$ with $(f,F) \in L^q(\Omega)^{1+n}$ and $q$ the conjugate exponent of $p$, as hinted at by Pietro. $\endgroup$
    – Hannes
    Commented Dec 4 at 20:41
  • $\begingroup$ @Hannes I know that, this is exactly Step 1. But $f-divF$ makes sense only when acting on test functions,not on $W^{1,p}$ function , so then we must apply $f-divF$ to a sequence of test functions conversion in $W^{1,p}$ right ? $\endgroup$ Commented Dec 4 at 21:12
  • $\begingroup$ Well, yes, if you want to do this extremely rigorously coming from a distributional point of view; the point being that $\langle \operatorname{div} F,\varphi\rangle$ can be estimated by $\lVert F\rVert_{L^q(\Omega)^n} \lVert \nabla\varphi\rVert_{L^p(\Omega)^n}$, so you extend from test functions to $W^{1,p}_0(\Omega)$ by density. (Alternatively just define the $\operatorname{div}$ directly on $W^{1,p}_0(\Omega)$ as in the second part of (2) in OP and convince yourself that it gives you a bounded linear operator..) $\endgroup$
    – Hannes
    Commented Dec 5 at 13:43

0

You must log in to answer this question.