Consider the following evolution equation

$$u_t=\Delta u+u$$ in a bounded and regular open subset $\Omega$ of $\mathbb{R}^N$, with smooth initial conditions $u_0\geq 0$ and homogeneous Dirichlet boundary conditions.

It is known that this equation has a smooth global solution $u$. My goal is to prove that the solution remains nonnegative. So I consider $w=\min(0,u)$ and its energy $E(t):=\int_\Omega w^2 dx$. We know that \begin{align} E(0) &= \int_\Omega w(0,x)^2 dx \\ &=\int_\Omega \min(0,u(0,x))^2 dx \\ &=\int_\Omega \min(0,u_0(x))^2 dx \\ &=0 \end{align}

By differentiating $E(t)$ and using integration parts we get

\begin{align} E'(t) &= 2\int_\Omega ww_t \\ &= 2\int_\Omega wu_t \\ &= 2\int_\Omega w\Delta u+2\int_\Omega wu \\ &= -2\int_\Omega \nabla w \cdot \nabla u+2\int_\Omega w^2 \\ &= -2\int_\Omega |\nabla w|^2+2E(t) \\ &\leq-\frac{2}{c^2}\int_\Omega w^2 dx+2E(t)\\ &\leq\left(2-\frac{2}{c^2}\right)E(t),\quad \text{for almost every} \ t \end{align} where $c$ is the Poincaré constant. Thus $E(t)\leq e^{\left(2-\frac{2}{c^2}\right)t}E(0)=0$ for almost every $t$ which implies that for a.e $t\geq 0$ $w(t,x)=0$ for a.e $x\in \Omega$. But since $w=\min(0,u)$ is continuous then $w(t,x)=0$ for all $t\geq 0$ and for all $x\in \Omega.$ Therefore $u(t,x)\geq 0$ for all $t\geq 0$ and for all $x\in \Omega.$

My concerns are:

1) How can I justify the derivation under integral sign $E'(t)=2\int_\Omega ww_t$ because unlike $u$ which is smooth, $w=\min(0,u)$ has only weak time derivative $w_t=u_t \mathbb{1}_{\{u\leq0\}}.$

2) In the end I proved that for a.e $t\geq 0$, $\int w(t,x)^2dx=0$, thus for a.e $t\geq 0$: $w(t,x)=0$ for a.e $x\in \Omega$. I then concluded by continuity of $w$ that this holds for all $t\geq 0$ and for all $x\in \Omega.$ I am not use but the space negligable sets of $\Omega$ might depend on time $t$. Does this make my argument still valid?

  • 2
    $\begingroup$ Can't we apply a comparison principle to show non negativity of the solution? $\endgroup$
    – Jonas Lenz
    Jul 31 '19 at 19:29

I think it is possible to write your procedure rigurously in the following way. First of all notice that we can re-write $w(t,x)$ as $$ w(t,x)=\dfrac{u-\vert u\vert}{2}. $$ Thus, differentiating we have $$ w_t=\dfrac{1}{2}\big(u_t-\hbox{sgn}(u)u_t\big)=\dfrac{1}{2}(1-\hbox{sgn}(u))u_t. $$ Now, replacing this in $E'(t)$ we obtain $$ E'(t)=2\int w w_t=\int (\Delta u+u)(1-\hbox{sgn}(u))w $$ Finally, integrating by parts we obtain $$ E'(t)=-\int (1-\hbox{sgn}(u))\nabla u\cdot\nabla w+\int (1-\hbox{sgn}(u))uw, $$ where the Dirac's delta dissapears because it is multiply by $w$, which is zero on the set $\{u=0\}$. Finally, noticing that $$ (1-\hbox{sgn}(u))u=2w \quad \hbox{and} \quad (1-\hbox{sgn}(u))\nabla u=2\nabla w, $$ you obtain the result proceeding exactly as you did on your question.

  • $\begingroup$ Hi, you still used the same derivation under integral sign $E'(t)=2\int w w_t$ using weak derivatives. How can we justify that? $\endgroup$ Aug 1 '19 at 13:49
  • $\begingroup$ There is no problem there, as the proof shows. The important part is that you are inside the integral and that $w$ has a weak-derivative. $\endgroup$
    – Sharik
    Aug 1 '19 at 15:33
  • $\begingroup$ But it's a time weak derivative inside a space integral... $\endgroup$ Aug 1 '19 at 20:08

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.