Rabinowitz proves (using the Mountain Pass Theorem) that for a bounded smooth domain $\Omega \in \mathbb{R}^n$, and $f(x,\xi)\in C(\bar{\Omega}\times \mathbb{R},\mathbb{R})$ satisyfing the growth condition $$f(x,\xi) \leq A + B|\xi|^s,\ \ \text{where}\ \ s+1<\frac{2n}{n-2}$$ then there is a weak solution $u \in H_0^1(\Omega)$ to the problem $$ -\Delta u = f(x,u), \ x\in \Omega$$ $$ u = 0, \ \ x \in \partial \Omega$$ He then makes the remark that if in addition $f$ is locally Lipschitz, the solution is classical. The proof is made by referencing a paper by Agman, here. Unfortunately, this was made before $\LaTeX$ so it is quite difficult to read, as in it actually hurts my eyes. So, I thought I would try to use some arguments presented in Gilbarg-Trudinger to solve a particular case that came up in Evans presentation for the mountain pass theorem. Let $f = |u|^{s-1}u$, where $s$ as is above. Here, maybe we need this paper, maybe we don't.

My argument goes like this : Since $u \in H_0^1$, by the Sobolev embedding theorem, we have $u \in L^{2^{*}} = L^{\frac{2n}{n-2}}$, and so $f \in L^{\frac{2^{*}}{s}}$. Using the global $L^p$ estimates (GT thm 9.13) $u \in W^{2,\frac{2^{*}}{s}}$. Now use the Sobloev theorem again to get a better estimate for $u$. Continue this process a finite number of times until we hit the critical value for the Sobolev theorem, in which case we get $u \in C^{0,\alpha_1}$ for some $\alpha_1$ (and thus $f \in C^{0,\alpha_2}$ for some $\alpha_2$). Now use the Dirichlet theory for elliptic operators (GT 6.11) to conclude that the solution is $C^{2,\alpha_2}$, by uniqueness.

Problems :

I know this argument can't be correct for at least two reasons :

  1. The theory for the Dirichlet problem says the solution is unique, and it can be proved that there are at least 2 solutions if $f$ is locally Lipschitz continuous (Rabinowitz : Minimax Methods p 11)
  2. The $L^p$ theory used in the "proof" above requires we know apriori that $u$ is a strong solution, ie $u\in W^{2,p}$. The normal regularity result that I would use to get this, GT-theorem 8.12, requires that $f \in L^2$. Our $f$ seems to always be in $L^r$, $r<2$.

Any help you could provide would be appreciated!

Edit : I have found the result in a book "Qualitative Analysis of Nonlinear Elliptic Partial Differential Equations", but I am still having trouble understanding the proof. At least it has some of the general ideas that I had. For convenience, the argument from the book is given below, noting they use different letters for exponents (they consider the problem $-\Delta u = u^p$)

We know until now that $u\in H_0^1(\Omega)\subset L^{2^{*}}(\Omega)$.In a general framework, assuming that $u \in L^q$, it follows that $u^p \in L^{\frac{q}{p}}$, that is, by Schauder regularity and Sobolev embeddings, $u \in W^{2,\frac{q}{p}} \subset L^s$, where $\frac{1}{s} = \frac{p}{q} − \frac{2}{N}$. So, assuming that $q_1 > \frac{(p−1)N}{2}$, we have $u \in L^{q_2}$, where $\frac{1}{q_2} = \frac{p}{q_1} − \frac{2}{N}$. In particular, $q_2 > q_1$. Let $(q_n)$ be the increasing sequence we may construct in this manner and set $q_{\infty} = \lim_{n\rightarrow \infty}{q_n}$. Assuming, by contradiction, that $q_n < \frac{Np}{2}$, we obtain, passing at the limit as $n \rightarrow \infty$, that $q_{\infty} = \frac{N(p − 1)}{2} < q_1$, contradiction. This shows that there exists $r > \frac{N}{2}$ such that $u \in L^r(\Omega)$ which implies $u \in W^{2,r}(\Omega) \subset L^{\infty}(\Omega)$. Therefore, $u \in W^{2,r}(\Omega) \subset C^k(\Omega)$, where $k$ denotes the integer part of $2 − \frac{N}{r}$. Now, by Holder continuity, $u \in C^2(\Omega)$.

I still don't understand how we get $u \in W^{2,\frac{q}{p}}$. Also, the last statement by Holder continuity, $u \in C^2(\Omega)$. Why is this?

  • $\begingroup$ Just a comment on the title (since unfortunately I cannot help with the PDEs): You probably want to add something about PDEs and the problem at hand to the title. I think you'd be more likely to get hits that way. $\endgroup$ – mck Feb 2 '13 at 2:40
  • $\begingroup$ Comment has been noted. Thanks. $\endgroup$ – Euler....IS_ALIVE Feb 2 '13 at 6:59

The theory for the Dirichlet problem says the solution is unique, and it can be proved that there are at least 2 solutions if $f$ is locally Lipschitz continuous.

No contradiction here. The theory to which you refer (GT Chapter 6) is for the Poisson equation $\Delta u=g$ with given $g$. The equation $\Delta u=f(u)$ may well have multiple solutions, since the right-hand side is different for different $u$.

The normal regularity result that I would use to get this, GT-theorem 8.12, requires that $f\in L^2$.

Indeed, GT do not prove a version of Theorem 8.12 for $1<p<\infty$. But it is true: if $u\in H_0^1(\Omega)$ and $\Delta u=g\in L^p(\Omega)$, then $u\in W^{2,p}(\Omega)$. The proof goes like this:

  1. extend $u$ and $g$ by zero to $\mathbb R^n$.
  2. Let $v=\mathcal{N}g$, the Newtonian potential of $g$. GT consider it in Chapter 4, but give only Hölder estimates. Sobolev estimates are based on the theory of singular integrals: $D_{ij}v$ is obtained from $g$ via Riesz transforms which are bounded on $L^p$ for $1<p<\infty$. See Theorem 3.2 here, which gives $v\in W^{2,p}(\mathbb R^n)$.
  3. Both $u$ and $v$ vanish at infinity, and $u-v$ is weakly harmonic, hence harmonic, hence identically zero.

Also, the last statement by Holder continuity, $u\in C^2(\Omega)$. Why is this?

That book is strangely written. Taking "the integer part of $2-N/r$" is not helpful because this part may well be equal to zero. The point is that the Morrey-Sobolev embedding gives $u\in C^{\alpha}(\overline \Omega)$ for some $\alpha>0$, which implies $g:=u^p$ is in $C^\alpha$. By Corollary 4.14 of GT, $\Delta u=g$ has a solution in $C^{2,\alpha}(\overline \Omega)$. Since the classical solution is also a weak solution, and the weak solution is unique, we conclude that $u \in C^{2,\alpha}(\overline \Omega)$.

  • $\begingroup$ Thank you for your response. Indeed, result 2 is proved in GT. Can you expand a bit more please? I'm very surprised that this theorem isn't in GT, since you need it for all the estimates. How can you conclude that $u-v$ is weakly harmonic? We only know that $u$ is in $W^{1,p}$. So, $\int{(u-v)\Delta \phi}$ = $\int{D(u-v)\cdot D\phi}$. To get this integral is $0$, I presume we need to be able to integrate by parts again. Also, how do you know the weak solution is unique? My first thought was Lax-Milgram, but this requires $L^2$ again. $\endgroup$ – Euler....IS_ALIVE Feb 4 '13 at 6:53
  • $\begingroup$ @Euler... I was thinking of $\Delta u$ and $\Delta v$ as distributional laplacian (the distribution happens to be the same $L^p$ function). In the distributional sense $\Delta(u-v)=0$, which by Weyl's lemma means that $u-v$ is harmonic. $\endgroup$ – user53153 Feb 4 '13 at 15:40
  • $\begingroup$ @Euler... There is at most one $u\in H^1_0(\Omega)$ that satisfies $\Delta u=g$ in the weak sense; this follows from Corollary 8.2 in GT. $\endgroup$ – user53153 Feb 4 '13 at 15:54
  • $\begingroup$ This is a good enough explanation for the bounty... although I wonder; Should we always think of the Laplacian as the distributional Laplacian? $\endgroup$ – Euler....IS_ALIVE Feb 5 '13 at 4:47
  • $\begingroup$ @Euler... You don't have to, although I can't think of an instance when it can hurt (as long as Laplacian is a linear operator). Distributions are a convenient common denominator for function spaces. In this situation there is $u\in H_0^1$ which satisfies $\Delta u=g$ in the weak sense, and $v\in W^{2,p}$ which satisfies $\Delta v=g$ in the strong sense. To save myself the headache of thinking of function spaces, I say: whatever, they both have Laplacian $g$ in the sense $\int (u\text{ or }v)\Delta \phi=\int g \phi$ for Schwartz functions $\phi$. This is all Weyl's lemma needs, anyway. $\endgroup$ – user53153 Feb 5 '13 at 4:57

You can see the book Nonlinear Analysis and Semilinear Elliptic Problem by Antonio Ambrosetti and Andrea Malchiodi, page 11, Theorem 1.16.


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.