Let $\Omega$ be an open bounded subset of $\mathbb{R}^N$, working in the space $H_0^1(\Omega)$ with the inner product $$(u,v)_{H_0^1} = \int_\Omega \nabla u \cdot \nabla v$$ for $u\in H_0^1$ and $\mu(\{|\nabla u|>1\}) >0$, is there a way to define a function $v$ such that $$(u,v)_{H_0^1} = \int_{\{|\nabla u|>1\}} |\nabla u|^2. $$ So we want $\nabla v = \nabla u$ on $\{|\nabla u|>1\}$, and $\nabla v= 0$ on the set $\{|\nabla u|\leq 1\}$, but naively defining $v$ in this way will not guarantee that $v\in H_0^1(\Omega)$.

Edit: Here is the main problem I am trying to solve. Define $$E(u) = \sup_{v\in H_0^1} \Bigg\{\int_\Omega \nabla u\cdot \nabla v - \int_\Omega |\nabla v| \Bigg\}$$ I claim this is equal to $$E'(u) = \begin{cases} 0 & \text{if } |\nabla u|\leq 1 \text{ a.e.} \\ +\infty & \text{else } \end{cases}$$

By substituting $v = Cu$ for some $C \geq 0$, this is when $\int\nabla u\cdot\nabla v \leq \int |\nabla u||\nabla v|$ becomes an equality, the sup becomes $$E(u) = \sup_{C\geq 0} \Bigg\{C\bigg(\int_\Omega |\nabla u|^2- |\nabla u|\bigg) \Bigg\}$$

Now when taking $\sup_{v\in H_0^1}$, if we could somehow restrict the support of $v = Cu$ onto a subset $K \subset \{|\nabla u|>1\}$ (whenever $\mu(\{|\nabla u|>1\}) >0$, we require $\mu(K) >0$), we will get $$E(u) = \sup_{C\geq 0} \Bigg\{C\bigg(\int_K |\nabla u|^2- |\nabla u|\bigg) \Bigg\} = E'(u)$$ which is what is claimed.

Thank you very much!


There is generally no $v\in H^1_0$ with $\nabla v=\chi_{\{|\nabla u|>1\}}\nabla u$. Consider for example $u:(-1/2,2)\to\mathbb R$, $u(x)=\min\{1-x/2,1+2x\}$. Now $u\in H^1_0((-1/2,2))$ but there is no $v\in H^1_0$ with $v'=2\chi_{(-1/2,0)}$ by the fundamental theorem of calculus.

If $u\in C^1_0$ and $u$ is constant on $\{|\nabla u|=1\}$ then a truncation of $u$ works as $v$. You need to assume something like this before you can hope for the existence of $v$.

Another issue is that it is not clear if the vector field $\chi_{\{|\nabla u|>1\}}\nabla u$ has a potential in the first place. In dimension three a local requirement is that the curl of the vector field vanish. Writing $\phi=\chi_{\{|\nabla u|>1\}}$, we have formally $$ \nabla\times(\phi\nabla u)=\phi\nabla\times\nabla u+(\nabla\phi)\times\nabla u=(\nabla\phi)\times\nabla u. $$ Assuming the set $V=\{|\nabla u|>1\}$ and the function $u$ are sufficiently smooth, the condition $(\nabla\phi)\times\nabla u=0$ means that $\nabla u$ should be normal to $\partial V$. (The vector valued distribution $\nabla\phi$ is supported on $\partial V$ and is normal to it.) Thus $u$ should be constant on connected components of $\partial V$. This condition is very similar to the sufficient condition I gave in the previous paragraph.

You need to be more specific about your second question. I understand it like this: "Let $u\in H^1_0$ and $K\subset\{|\nabla u|>1\}$ with positive measure. Does there exist $v\in H^1_0$ with $(u,v)_{H^1_0}=\int_K|\nabla u|^2$?" The answer is yes: for some $\lambda\in[0,1]$ the choice $v=\lambda u$ works.

Edit: This is a remark regarding your edit. If $\phi$ is a cut-off function supported in $K$, a natural attempt is $v=\phi(u-c)$ for some constant $c$. One quickly ends up with the estimate $$ \int_\Omega\nabla u\cdot\nabla v-|\nabla v| \geq \int_\Omega\phi(|\nabla u|^2-|\nabla u|)+(u-c)\nabla u\cdot\nabla\phi-|u-c||\nabla\phi|. $$ But you can't in general choose $\phi$ so that the RHS would be positive since you can't increase $\int\phi$ arbitrarily but keep $\int|\nabla\phi|$ bounded. Something more clever is needed.

  • $\begingroup$ Thank you for the very helpful answer! For the last part, I knew the problem will be at the boundary of $\{|\nabla u| > 1\}$. So by relaxing the condition, I thought maybe there is some hope we could define $v$ on a smaller subset. I actually need $\nabla v$ to be supported on the set $\{|\nabla u| > 1\}$ and I will explain this in detail in my edit. $\endgroup$ – Xiao Nov 17 '14 at 7:52

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.