The solution to the equation $- \Delta u = (\lambda - \log u)u$

$\Omega$ is a bounded domain in $\mathbb R^n$ with smooth boundary. Consider the Dirichlet problem:$$-\Delta u = (\lambda - \log u)u ~~~{\rm on}~~~ \Omega~~~ {\rm and}~~~u=0 ~~~{\rm on}~~\partial {\Omega}$$ where the solution $u\in {C^\infty }(\bar \Omega )$ and $u>0$ on $\Omega$ and $\lambda$ is a positive constant.

My question is: what conditions do we need for the existence and uniqueness of the solution $u$ ?

• Why do yout think that a solution of this problem is $C^\infty(\overline{\Omega})$? – Tomás May 14 '13 at 12:25

Define $$f(t) = \left\{ \begin{array}{rl} -t\log{t} &\mbox{ if t\in[0,1]} \\ 0 &\mbox{ otherwise} \end{array} \right.$$

Note that $f$ is continuous. COnsider the problem

$$\tag{1} \left\{ \begin{array}{rl} -\Delta u-\lambda u=f(u) &\mbox{ in \Omega} \\ u=0 &\mbox{ in \partial\Omega} \end{array} \right.$$

1 - Existence

I - There is some methods avaliable to find a solution to $(1)$. For example, we can use the fact that $f$ is continuous with compact support, to prove that the energy functional $I:H_0^1\to\mathbb{R}$ associated with $(1)$ is (under some additional restriction) weakly lower semi continuous and coercive. Remember that if $F(t)=\int_0^t f(s)ds$, then $$Iu=\frac{1}{2}\int_\Omega |\nabla u|^2-\frac{\lambda}{2}\int_\Omega u^2-\int_\Omega F(u)$$

The additional restriction necessary here is $\lambda<\lambda_1$ (to guarantee coercivity).

II - (This is just a guess - I dont have verified the calculations): Also, you can prove existence, by noting that if $\lambda$ is not a eigenvalue of $-\Delta$, then the operator $-\Delta-\lambda I$ is invertible, and you can use arguments similar to this one.

Remark 1: The original function $t\log{t}$ is not monotone, so I dont know if this problem has uniqueness.

Remark 2: The function $f$ being continuous, we can conclude that $u\in C^1(\overline{\Omega})$.

Now, let's find conditions such that the solution $u$ of $(1)$ satisfies $0<u\leq 1$. Because $f\geq 0$ and there is a open set where $f>0$, you can conclude by the maximum principle that $u>0$. On the other hand, by regularity theory, you have that the solution $u$ of $(1)$ (See Breziz chapter 9) satisfies $(1)$ almost everywhere, i.e. $u$ is a Strong solution. If we suppose that the set where $u>1$ is non0empty, then we would have $-\Delta u-\lambda u=0$ almost everywhere in this set. If $\lambda$ is not a eingevalue, then this is a absurd, which implies that $u\geq 1$.

From the last paragraph, we can conclude that in fact $u$ is a solution to the original problem and satisfies $0<u\leq 1$. Hence, by using the first method proposed, we can conclude that if $\lambda<\lambda_1$, then you problem has a solution $u\in C^{2,\alpha}(\overline{\Omega})$ satisfying $0<u\leq 1$. I dont know if we can get more regularity, because the functions $-t\log{t}$ is only Holder continuous. A argument of bootstrap can be applied in the interior of $\Omega$, but in the boudary I dont know, because $-t\log{t}$ is not even Lipschitz for $t=0$.

• What is the $\lambda_1$ ? – lanse7pty Feb 23 '16 at 12:13
• It is the first eigenvalue of the problem $-\Delta u-\lambda u=0$. – Tomás Feb 23 '16 at 12:22
• I still don't understand your answer , I just a beginner of PDE, if I want understand yours , I should read which chapters of Evans PDE ? – lanse7pty Feb 24 '16 at 8:00