I posted this question first in Math.StackExchange one week ago here, but I didn't get an answer or a helpful comment so I repost it here:

Let $d=3$ and $\Omega\subset \mathbb R^d$ is a bounded Lipschitz domain and $u$ is a measurable function. A sufficient condition for the integral $\int\limits_{\Omega}{uvdx}<\infty,\forall v\in H_0^1(\Omega)$ is that $u\in L^{6/5}(\Omega)$ which follows from Holder's inequality and the (continuous) embedding $H^1(\Omega)\hookrightarrow L^6(\Omega)$.

Question: Is the opposite true, i.e is it true that $$\int\limits_{\Omega}{uvdx}<\infty,\forall v\in H_0^1(\Omega)$$ implies $u\in L^{6/5}(\Omega)$ or at least $u\in L^1(\Omega)$ ?

My thoughts: It is easy to see that $u\in L^1_{loc}(\Omega)$ by taking $v$ to be smooth cut-off functions equal to $1$ in compact subsets of $\Omega$ and $0$ in a neighborhood of the boundary $\partial \Omega$.

The motivation for this question is the "correct" weak formulation of a nonlinear problem - whether to formulate it as $(1)$ or as $(2)$:

$(1)$ Find $u\in H_0^1(\Omega)$ such that $f(u)\in L^{6/5}(\Omega)$ and $$a(u,v)+\int\limits_{\Omega}{f(u)vdx}=0,\forall v\in H_0^1(\Omega)$$


$(2)$ Find $u\in H_0^1(\Omega)$ such that $\int\limits_{\Omega}{f(u)vdx}<\infty,\forall v\in H_0^1(\Omega)$ and $$a(u,v)+\int\limits_{\Omega}{f(u)vdx}=0,\forall v\in H_0^1(\Omega)$$

where $a(.,.)$ is a bilinear form and $f(.)$ is in general a nonlinear function. If the answer to my question is affirmative then both formulations are equivalent.

Note that $(2)$ is less restrictive, because the set in which we search for a solution is bigger, so it might be easier to find such.

  • 7
    $\begingroup$ You are asking if $H^{-1} \subset L^{6/5}$, which is false $\endgroup$ – Piero D'Ancona Jan 8 '16 at 0:24
  • $\begingroup$ @Piero D'Ancona Can you give me some argument or a reference, so that I can see why is this? $\endgroup$ – Svetoslav Jan 11 '16 at 14:11
  • 2
    $\begingroup$ Take $u=\partial_1(\chi(x)v(x_1))$ where $\chi$ is a test function on $R^3$ equal to 1 near the origin and $v(s)$ is the sign function. From $\chi v\in L^2$ it follows $u\in H^{-1}$, and of course $u$ is not even a function since it is a distribution (a delta in the direction $x_1$). $\endgroup$ – Piero D'Ancona Jan 11 '16 at 14:23
  • $\begingroup$ @Piero D'Ancona Yes, this is fine, but my functional is not just any functional (distribution) but it is a regular distribution, i.e it is represented by an integral of the measurable function $u(x)$ times a function $v(x)\in H_0^1$ $\endgroup$ – Svetoslav Jan 11 '16 at 15:15

That doesn't work because $H_0^1$ functions are small near the boundary, so testing against them won't detect bad behavior of $u$ near $\partial\Omega$.

For a concrete example, take $\Omega$ as the unit ball and $u(x)=1/(1-|x|)\notin L^1$. Then $$ \int |uv|\, dx \le \left( \int \frac{v^2\, dx}{(1-|x|)^{3/2}} \int \frac{dx}{(1-|x|)^{1/2}} \right)^{1/2} . $$ If $v\in H_0^1$ is also smooth, then we can estimate the first integral in the same way as in this related question (by just integrating the gradient, starting from the boundary, to bound $v$). This gives $\int v^2/(1-|x|)^{3/2}\lesssim \|v\|^2_{H^1}$, so $\int |uv| \lesssim \|v\|_{H^1}$ for all such $v$, and by density of the smooth functions, this also holds for arbitrary $v\in H_0^1$.

| cite | improve this answer | |
  • $\begingroup$ sorry for the late response. I like your answer, but I cann't get the crucial inequality $|v(x)|\leq (1-|x|)^{1/2}\|v\|_{H_0^1}$. Could you elaborate a bit on it ? $\endgroup$ – Svetoslav Jan 9 '16 at 19:18
  • $\begingroup$ If I understand correctly the issue, I think that this inequality can not be true (for $C_0^\infty$ functions), because (I think) there are positive functions $f\in C_0^\infty(B_0(1))$ with arbitrary small $H^1$ norm and arbitrary "high": $\forall M>0\forall \delta>0\exists f\in H_0^1(B_0(1)): \|f\|_{H^1(B_0(1))}<\delta$ and $\exists \epsilon>0: f(x)\ge M$ a.e in $B_0(\epsilon)\subset B_0(1)$. For example, if we take $g=K \log{\frac{1}{|x|/\epsilon}}$ and construct the function $\tilde g=g$ on $B_0(\epsilon)$ and $\tilde g=0$ on $B_0(1)\setminus B_0(\epsilon)$. Here $K$ controls the $H^1$-norm $\endgroup$ – Svetoslav Jan 10 '16 at 20:58
  • $\begingroup$ Then, if $h(x)=\tilde g(x)$ where $x: \tilde g(x)\leq M$ and $h(x)=M$ where $x: \tilde g(x)\ge M$. Now taking $f$ to be a mollified version of $h$ should give us $C_0^\infty(B_0(1))$ function, for which the inequality does not hold. Maybe I am missing something, but it will be nice if you check my construction. $\endgroup$ – Svetoslav Jan 10 '16 at 21:06
  • 1
    $\begingroup$ Notice that the $t$ integration is essentially the same thing as integrating with respect to the radial variable in spherical coordinates, so integrating over just the angular part of $dx$ gives you a quantity that is estimated by $\|v\|^2$, and then the extra $d|x|$ integration just affects the constant. $\endgroup$ – Christian Remling Jan 12 '16 at 21:46
  • 1
    $\begingroup$ @Svetoslav: I think Fatou's Lemma works best here: we also have that $v_n\to v$ pointwise a.e. (pass to a subsequence if necessary), so $\int |uv|=\int\liminf |uv_n|\le\liminf\int|uv_n|\lesssim \liminf\|v_n\|=\|v\|$. $\endgroup$ – Christian Remling Jan 13 '16 at 18:20

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.