0
$\begingroup$

I am considering the following section of Peter Grindrods „Pattern and Waves“.

We begin by considering a scalar equation $$ u_t=\Delta u+f(u,x,t),\quad x\in\Omega\subseteq\mathbb{R}^n, t>0. $$ Here $f$ is smooth (say, continously differentiable).

Suppose $\overline{u}\colon\Omega\times [0,T]\to B$, some bounded subset of $\mathbb{R}$, and $$ \overline{u}_t\geq\Delta\overline{u}+f(\overline{u},x,t); $$ then we say that $\overline{u}$ is a super-solution. If $\underline{u}\colon\Omega\times [0,T]\to B$, and $$ \underline{u}_t\leq\Delta\underline{u}+f(\underline{u},x,t), $$ then we say that $\underline{u}$ is a sub-solution.

Now suppose that there exist constants $\alpha,\beta (\alpha^2+\beta^2\neq 0)$ such that $$ \alpha\overline{u}-\beta\nabla\overline{u}.n\geq \alpha\underline{u}-\beta\nabla\underline{u}.n,\quad x\in\partial\Omega, t>0, $$ (n is the outer normal to $\partial\Omega$); and that $$ \overline{u}(x,0)\geq\underline{u}(x,0), x\in\Omega. $$ Then we claim that $$ \overline{u}(x,t)\geq \underline{u}(x,t),\quad x\in\Omega. $$


The author also gives a short "proof".

To see this, set $$ w=\overline{u}-\underline{u}. $$ Then the mean value theorem implies $$ w_t-\Delta w\geq f_u(\underline{u}+\theta (\overline{u}-\underline{u}))w $$ for some mapping $\theta\colon\Omega\times [0,T]\to [0,1]$. The result follows from the strong maximum principle for linear parabolic equations [15] (The point is that $w$ is initially nonnegative and the boundary conditions ensure that if it becomes negative it must do so at an interior point of $\Omega$. We then can construct a contradiction.)


Unfortunately, I have no opportunity to have a look at the mentioned maximum principle cited from book [15] (which is a book by P.C. Fife).

So we set $w=\overline{u}-\underline{u}$ and note that $$ w_t-\Delta w\geq f(\overline{u},x,t)-f(\underline{u},x,t)=f_u(\underline{u}+\theta(\overline{u}-\underline{u})w,x,t)w $$ where the last identity comes from applying the mean value theorem.

Up to here, I can follow the proof.

Now, i write this inequality by using a parabolic linear operator in order to bring it in the usual textbook form for maximum principles:

$$ Lw = \sum_{i=1}^n\frac{\partial^2 w}{\partial^2 x_i}-f_u(\underline{u}+\theta(\overline{u}-\underline{u},x,t)w-(-w_t)\geq 0. $$ Now, if $$ f_u(\underline{u}+\theta(\overline{u}-\underline{u},x,t)\leq 0, $$ I could use the strong maximum principle for the function $-w$, i.e. making some statement about non-negative maxima of the function $-w$ (resp. about negative minima of the function $w$).

However, I can neither find the assumption $$ f_u(\underline{u}+\theta(\overline{u}-\underline{u},x,t)\leq 0 $$ nor deduce it from the text.


But let's suppose for a moment that $$ f_u(\underline{u}+\theta(\overline{u}-\underline{u},x,t)\leq 0. $$

The proof by contradiction which the author mentions maybe works like this:

Assume we have at least one point $(x_0,t_0), t_0>0$ such that $$ \overline{u}(x,t)<\underline{u}(x,t). $$ Then $-w(x_0,t_0)>0$.

By initial- and boundary conditions, the point $(x_0,t_0)$ had to be an inner point of $\Omega\times [0,T]$ since $w<0$ can only happen at inner points.

From this, can we deduce that in the interior of $\Omega\times [0,T]$ we have a negative minimum $M$ (say at some point $(x',t')$ which equals $(x_0,t_0)$ in case this is the only point such that $\overline{u}<\underline{u}$) of $w$ and that - by the strong maximum principle - $$ w(x,t)=M $$ for all $(x,t)$ which are on the horizontal line on which $(x',t')$ lies? (in particular for boundary points on this line, which gives a contradiction since there $w$ cannot be negative)?

Thats how I did understand the proof.

$\endgroup$
  • $\begingroup$ Is $f_u\geq 0$? $\endgroup$ – Jeff Mar 2 '18 at 22:25
  • $\begingroup$ @Jeff I do not know. At least the text does not say that. $\endgroup$ – Salamo Mar 2 '18 at 22:37
  • $\begingroup$ Is the PDE $u_t = \Delta u + f(u,x,t)$? (there are some typos in your post). Generally the maximum principle only holds when $f_u \leq 0$ (sorry I wrote wrong way in previous comment). Are there any other assumptions on $f$ in your book (there must be something)? $\endgroup$ – Jeff Mar 2 '18 at 22:45
  • $\begingroup$ @Jeff Sorry for the typos. Yes, the Pde is $u_t=\Delta u+f(u,x,t)$. - Considering $w_t-\Delta w\geq f_u(...)w$, why does the maximal principle only hold for $f_u\leq 0$? I am a bit confused since the maximum principles are always formulated in the form $Lu\leq g$ or $Lu\geq g$ and I am not sure what are $L$ and $g$ here and whether we need $\leq$ or $\geq 0$. - - I will have a closer look at the book. (Unfortunately, I only have the first 100 pages.) $\endgroup$ – Salamo Mar 2 '18 at 22:50
  • $\begingroup$ For a linear equation $w_t-\Delta w + cw = 0$, maximum principle requires $c\geq 0$. Here $c=-f_u$. A good reference for the maximum principle for parabolic equations is Evans PDE book Chapter 6. $\endgroup$ – Jeff Mar 2 '18 at 22:58
3
$\begingroup$

There is a trick you can use to adjust the sign of the zeroth order term so that the maximum principle applies. You have a solution $w$ of

$$w_t - Lw \geq cw$$

in a parabolic cylinder, with $w\geq 0$ on sides and base, and you want to show $w\geq 0$ in the interior. Define $v = e^{-\lambda t}w$. Then $v$ satisfies

$$v_t + (\lambda-c)v - Lv \geq 0.$$

Provided $c$ is bounded, you can choose $\lambda$ large enough so that $\lambda -c\geq 0$. Now you can apply the standard strong (or weak) maximum principle to conclude that $v \geq 0$ everywhere, and so $w\geq 0$.

$\endgroup$
  • $\begingroup$ Am I right that your operator L is $Lw=\Delta w$ and your c is $c=f_u(...)$? And the parabolic cylinder is $\Omega\times [0,T]$? $\endgroup$ – Salamo Mar 3 '18 at 19:47
  • $\begingroup$ How can we conclude that $v\geq 0$ everywhere? $\endgroup$ – Salamo Mar 3 '18 at 20:02
  • $\begingroup$ Yes. You use the minimum principle. $v$ attains its minimum on the sides and base, where $v\geq 0$ (since $w\geq 0$ on sides and base). There are similar versions of the maximum principle for your boundary conditions. The whole point is the trick to change variables and introduce a zeroth order term $(\lambda-c)v$ with the right sign. $\endgroup$ – Jeff Mar 3 '18 at 20:04
  • $\begingroup$ Sorry I did not see yet why Minimum principle. You mean if we assume that w is negative, then w (and v) has a negative minimum M in the interior and hence v also has value M on the sides and the base which is a contradiction since there v is nonnegative? $\endgroup$ – Salamo Mar 3 '18 at 20:10
  • $\begingroup$ But for minimum principle I need $...\leq 0$ and not $\geq 0$, right? $\endgroup$ – Salamo Mar 3 '18 at 20:13

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.