1
$\begingroup$

I have a big problem to solve this system $\Delta f-hf^2=0$

$|\nabla f|^2+hf^3=0$

where $h$ is a constant, $f$ is a 2-dimensional smooth function, $\Delta f$ is Laplacian of $f$ (i.e. $\Delta f=f_{xx}+f_{yy}$) and $\nabla f$ is the gradient of $f$.

ADD In first case $f$ is defined on $R^2$ and in second case $f$ is defined on surface $S$ ($f:S \rightarrow (0, \infty)$). is there a solution? Thank you for help

MODIFICATION after Igor Khavkine answer: and if the system is

$\Delta f-hf^2+cf=0$

$|\nabla f|^2+hf^3=0$

(c is another constant)

$\endgroup$
10
  • 1
    $\begingroup$ Please give some context as to why you are interested in this problem. $\endgroup$ Commented Nov 14, 2017 at 16:39
  • 1
    $\begingroup$ @Chris Ramsey, I found a particular manifold with Einstein condition in isotropic space... $f$ is defined on surface and it is a positive scalar function $\endgroup$
    – MathDG
    Commented Nov 14, 2017 at 16:46
  • 1
    $\begingroup$ I don't understand your question, you have a system of two PDEs about a scalar function $f$ (which means surely your system is overdetermined). But then you say that simultaneously $f$ is defined on $\mathbb{R}^2$ and a surface $S$: which is it? $\endgroup$ Commented Nov 14, 2017 at 18:23
  • 1
    $\begingroup$ @Willie Wong, No, are two separate case.. I said that badly, sorry! $\endgroup$
    – MathDG
    Commented Nov 14, 2017 at 18:42
  • 1
    $\begingroup$ If you have two different questions, why not ask them separately? It will be easier to understand the questions and the answers if you ask the questions separately. $\endgroup$
    – Ben McKay
    Commented Nov 14, 2017 at 19:19

2 Answers 2

12
$\begingroup$

I assume that, in the surface case, the OP wants to interpret $S$ as a surface endowed with a Riemannian metric and wants to understand the solutions to the equations $\Delta f - hf^2 = 0$ and $|\nabla f|^2 + hf^3 = 0$ for a given constant $h$.

Clearly, if $h=0$, the only solutions are to have $f$ be constant, so one can assume that $h\not=0$. Then, setting $u = -hf$, the above equations are equivalent to $\Delta u + u^2 = 0$ and $|\nabla u|^2 - u^3 = 0$, so it suffices to solve these latter equations.

Let $g$ be the metric on $S$ and assume that $u$ is a nonzero (and, hence, necessarily positive) solution to the above equations on a simply-connected open subset $S'\subset S$. The second equation implies that $\omega_1 = u^{-3/2}\,\mathrm{d}u$ is a $1$-form with $g$-norm $1$ on $S'$, and hence $g$ can be written in the form $g = {\omega_1}^2 + {\omega_2}^2$ on $S'$ for some $\omega_2$, which is also a unit 1-form.

Fix an orientation by requiring that $\omega_1\wedge\omega_2$ be the $g$-area form on $S'$. Then $\star \mathrm{d}u = u^{3/2}\,\omega_2$, and since $\mathrm{d}(\star \mathrm{d}u) = \Delta u\, \omega_1\wedge\omega_2$, it follows that $$ \tfrac32\,u^{1/2}\mathrm{d}u\wedge\omega_2 + u^{3/2}\,\mathrm{d}\omega_2 =\mathrm{d}(u^{3/2}\,\omega_2) = -u^2\,\omega_1\wedge\omega_2 = -u^{1/2}\,\mathrm{d}u\wedge\omega_2\,, $$ or $$ \mathrm{d}\omega_2 = -\tfrac52\,u^{-1}\,\mathrm{d}u\wedge\omega_2\,, $$ which can be written in the form $$ \mathrm{d}\bigl(u^{5/2}\,\omega_2\bigr) = 0. $$ Since $S'$ is simply connected, it follows that there exists a function $v$ on $S'$ such that $u^{5/2}\,\omega_2 = \mathrm{d}v$. Consequently, the metric $g$ has the form $$ g = {\omega_1}^2 + {\omega_2}^2 = u^{-3}\,\mathrm{d}u^2 + u^{-5}\,\mathrm{d}v^2. $$ This metric can be placed in 'polar form' by setting $u = 4r^{-2}$ and $v = 32\theta$, in which case, it becomes $$ g = \mathrm{d}r^2 + r^{10}\,\mathrm{d}\theta^2, $$ This is a singular, incomplete metric at $r=0$ (where $u$ goes to $\infty$), though it is complete at $r=\infty$. Its Gauss curvature is $K = -(r^5)''/r^5 = -20 r^{-2} = -5u < 0$.

The original $f$ then takes the form $f = -u/h = -4/(hr^2)$.

Thus, up to isometry, there is essentially only one solution to the OP's original problem. (In particular, as Igor showed in his answer, there is no nonconstant solution when the background metric $g$ is flat.) The modified problem (with the additional constant $c$) can be solved using essentially the same techniques.

$\endgroup$
15
  • $\begingroup$ thank you very much for your answer, then if I understand for the case with $c$ not zero, the answer is the same, i.e. $f$ constant and $h=c=0$ ? $\endgroup$
    – MathDG
    Commented Nov 15, 2017 at 12:00
  • $\begingroup$ you say there is only one solution (ie $ f $ constant and $ h = 0 $) because if $ h $ is not zero, $ f $ is a not smooth function..is this so? instead in the case of Igor Khavkine (which is the case where the metric is flat) the solution for $h$ not zero is $f=0$. $\endgroup$
    – MathDG
    Commented Nov 15, 2017 at 13:11
  • $\begingroup$ @exxxit8: You have not read my analysis carefully. I showed that, when $h$ is nonzero, then, on the open set where $f$ is nonzero, it has a particular form (as described above). Obviously, $f=0$ is a solution, but that can be discarded as trivial. $\endgroup$ Commented Nov 15, 2017 at 13:18
  • $\begingroup$ you are right! I read fast. And for $c$ not zero, Do you expect something like that? $\endgroup$
    – MathDG
    Commented Nov 15, 2017 at 13:36
  • 1
    $\begingroup$ @AlexanderPigazzini: I'm not sure exactly what you are asking. A non-zero solution $u$ cannot go to zero in finite distance, even if $c$ is not zero. The solution as I've described it would be valid on any simply-connected domain on which $u$ is nonzero, so I'm not sure what the boundary, if one exists, would have to do with anything (assuming that the metric and solution are sufficiently differentiable on the boundary). Does that answer your question? $\endgroup$ Commented Nov 15, 2017 at 21:14
3
$\begingroup$

I will assume that your system is simply defined on $\mathbb{R}^2$. I don't know what you mean for this system to be defined on a surface $S$, since you haven't told us anything about $S$. Note that your equations are in the form $E_2 := f_{xx} + f_{yy} -hf^2 = 0$ and $E_1 := f_x^2+f_y^2 + hf^3=0$. I will assume also that $h$ is a non-zero constant.

If you differentiate $E_2$ you can algebraically solve $(E_2,(E_1)_x, (E_1)_y) = 0$ for the highest derivatives and get \begin{align*} F_1 &:= f_{xy} - 4 f_x f_y/f = 0, \\ F_2 &:= f_{xx} + 4 f_y^2/f + 3hf^2/2 = 0, \\ F_3 &:= f_{yy} + 4 f_x^2/f + 3hf^2/2 = 0. \end{align*}

If you cross-differentiate the $F$-equations you can algebraically eliminate all second derivatives from $((F_1)_x - (F_2)_y, (F_1)_y - (F_3)_x) = 0$ and end up with two first order equations. Using $E_1$ to simplify them, unless I made some calculation mistake, the result is equivalent to $f_x = 0$ and $f_y = 0$. Finally, eliminating all derivatives from $E_1$, we end up with the condition $f=0$. Thus $f(x,y) \equiv 0$ is the only solution to your system of equations.

If $h=0$, then after the first step, you end up with the equations \begin{align*} F_1 &:= f_{xy} = 0, \\ F_2 &:= f_{xx} = 0, \\ F_3 &:= f_{yy} = 0. \end{align*} The only solutions are $f(x,y) = Ax+By+C$ with constants $A$, $B$, $C$. Equation $E_1=0$ forces the constraint $A^2+B^2=0$. So, if you only want positive solutions, the only possibility is $f(x,y) \equiv C$, for some positive constant $C>0$.

$\endgroup$
0

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .