Let $M$ be a closed Riemannian manifold of dimension $6$. We have a function $f\geq 0$ on $M$ satisfying \begin{align*} \Delta f \leq gf-\frac{3}{4}f^2 \end{align*} Where $g$ is another smooth function on $M$ and our Laplacian is $d^*d$, i.e. locally it looks like $-\sum_i \frac{\partial^2}{\partial x_i^2}$. Using maximum principle we can get a $C^0$ bound of $f$ using $C^0$ bound of $g$, my question is: is it possible to get an $L^\infty$ bound of $f$ if we assume some $L^p$ bound of $g$ using some Moser iteration type argument?
-
1$\begingroup$ I don't think that it is possible. Moser's iteration scheme gives you an $L^\infty$ bound for $f^+$ (the positive part of $f$) whenever $f$ is a subsolution of an elliptic problem. In the form your PDE is written $f$ is a supersolution with respect to the elliptic $(-\Delta f)$, so it is going in the opposite direction of the one needed to apply the iteration scheme. $\endgroup$– Michele CaselliCommented Oct 11, 2023 at 7:39
-
1$\begingroup$ You have to consider $-\Delta f$ to put it in a form to apply the result, since it needs the operator in divergence form to be elliptic and positive-definite. Writing $-\Delta f \ge -c(g^2+f^2) $ then $f$ is a supersolution and is in the wrong direction. $\endgroup$– Michele CaselliCommented Oct 11, 2023 at 12:17
-
2$\begingroup$ @MicheleCaselli: the OP uses $\Delta$ for $d*d$, which is locally $- \sum \partial_i^2$, which I think makes the inequality in the question indeed one about a subsolution? $\endgroup$– Willie WongCommented Oct 11, 2023 at 13:08
-
1$\begingroup$ @DeaneYang Yes, thanks. I can see that I can get a bound in terms of $L^p$ norms of both $f$ and $g$, I am asking if it's possible to get a bound only using a bound of $g$? For example a $C^0$ bound of $g$ will give me a $C^0$ bound of $f$, can I improve on that? $\endgroup$– ParthaCommented Oct 11, 2023 at 13:40
-
1$\begingroup$ Moser iteration per se can't do this. For your question, the specific form of the right side is crucial, specifically the presence of the negative quadratic term. So you need an ad hoc argument. If you can't find one yourself, you can try to dig through the PDE literature. It's not unlikely that someone else has had to study a PDE of this form, probably on a domain in Euclidean space. If so, it's possible that their argument can be adapted to a closed Riemannian manifold. $\endgroup$– Deane YangCommented Oct 11, 2023 at 14:10
1 Answer
\begin{align*} &\Delta f\leq gf-\frac{3}{4}f^2\\ &\Rightarrow\int f^{p+1}\Delta f\leq \int(gf^{p+2}-\frac{3}{4}f^{p+3})\\ &\Rightarrow\frac{4(p+1)}{(p+2)^2}\int |\nabla(f^{\frac{p+2}{2}})|^2\leq\int |g|f^{p+2}-\frac{3}{4}\int f^{p+3}\\ &\Rightarrow\frac{4(p+1)}{(p+2)^2}\int |\nabla(f^{\frac{p+2}{2}})|^2\leq \int \frac{(c|g|)^{p+3}}{p+3}+\int \frac{f^{p+3}}{c^{\frac{p+3}{p+2}}\frac{p+3}{p+2}}-\frac{3}{4}\int f^{p+3}\hspace{1 ex}[c \text{ any positive constant}] \end{align*} Notice that for any given $p\geq -1,\exists \hspace{0.4 ex}c,$ such that \begin{align*} &c^{\frac{p+3}{p+2}}\geq \frac{4}{3},\text{ and for that $c$ we get,}\\ &c^{\frac{p+3}{p+2}}>\frac{4}{3}\times\frac{p+2}{p+3}\Rightarrow\bigg(\frac{1}{c^{\frac{p+3}{p+2}}\frac{p+3}{p+2}}-\frac{3}{4}\bigg)< 0 \end{align*} Hence, we can write, \begin{align*} \int f^{p+3}\leq C(p)\int |g|^{p+3} \end{align*} Where the constant $C(p)$ depends on $p$ but not on $f$ or $g$. Hence, we get an $L^{p+3}$ bound of $f$ from an $L^{p+3}$ bound of $g$ for any $p\geq -1$.
We can actually do better! We start with the inequality: \begin{align*} &\frac{4(p+1)}{(p+2)^2}\int |\nabla(f^{\frac{p+2}{2}})|^2\leq\int |g|f^{p+2}-\frac{3}{4}\int f^{p+3}\\ &\Rightarrow\frac{4(p+1)}{(p+2)^2}\big(\int |\nabla(f^{\frac{p+2}{2}})|^2+\int (f^{\frac{p+2}{2}})^2\big)\leq \frac{4(p+1)}{(p+2)^2}\int |\nabla(f^{\frac{p+2}{2}})|^2+4\int f^{p+2}\leq\int (|g|+4)f^{p+2}-\frac{3}{4}\int f^{p+3} \end{align*} We do the same trick on the right hand side as before but we do it with $(|g|+4)$ and $f$ instead of $|g|$ and $f$ (the trick is nothing but Young's inequality) and we end up with \begin{align*} \int |\nabla(f^{\frac{p+2}{2}})|^2+\int \big(f^{\frac{p+2}{2}}\big)^2\leq \tilde{C}(p)\int (|g|+4)^{p+3} \end{align*} Where the constant $\tilde{C}(p)$ depends on $p$ but not on $f$ or $g$. Now we use Sobolev embedding $L^2_1\hookrightarrow L^3$ in dimension $6$ and this gives us: \begin{align*} &\big(\int f^{\frac{3(p+2)}{2}}\big)^{\frac{1}{3}}\leq C_S\big(\int |\nabla(f^{\frac{p+2}{2}})|^2+\int (f^{\frac{p+2}{2}})^2\big)^{\frac{1}{2}} \end{align*} So, we can control the $L^{\frac{3(p+2)}{2}}$ norm of $f$ using the $L^{p+3}$ norm of $g.$ Using elliptic regularity, we get \begin{align*} ||f||_{L^q_2}&\leq C\big(||\Delta f||_{L^q}+||f||_{L^q}\big)\\ &\leq C\big(||gf||_{L^q}+||f^2||_{L^q}\big)\\ &\leq C\big(||\frac{g^m}{m}||_{L^q}+||\frac{f^n}{n}||_{L^q}+||f^2||_{L^q}\big) \hspace{1 ex}[\frac{1}{m}+\frac{1}{n}=1,m,n>1] \end{align*} For $q>3,L^q_2\hookrightarrow C^0$ (using Sobolev embedding). If we take $m=n=2,$ then an $L^s$ bound of $g$ for $s>6$ gives us an $L^{\frac{s}{2}}$ bound for $f^2$ and $g^2$ and hence an $L^q_2$ bound of $f$ for $q>3$ and hence a $C^0$ bound of $f.$
-
$\begingroup$ This looks correct, but it makes me wonder if maybe the only possible solutions are the constant solutions $f=0$ and, if $g$ is constant, $f = \frac{4}{3}g$. Have you checked this? Also, I think it might be simpler to observe that $\int f^{p+1} \le \frac{4}{3}\int gf^p$ and use Cauchy-Schwarz or Holder to bound $\int f^{p+1}$ by a constant times $\int g^{p+1}$. If you choose any $p > 0$, then you can use Moser iteration to get a bound for $\|f\|_\infty$ in terms of $\int |g|^{p+1}$. $\endgroup$ Commented Oct 12, 2023 at 16:40
-
1$\begingroup$ @DeaneYang, when $g$ is constant, this type of inequality appears from the Seiberg-Witten equations in dimension $4$ and $3$ and there are non-trivial solutions of SW equations, I think there are non-trivial solutions definitely. $\endgroup$– ParthaCommented Oct 13, 2023 at 12:04
-
$\begingroup$ But I wonder if it can be proved that there are no solutions on $\mathbb{R}^6$ except the trivial constant solutions. $\endgroup$– ParthaCommented Oct 13, 2023 at 13:57