I can neither confirm nor deny that only one solution (namely $u\equiv 0$) satisfies the following boundary value problem in $\Omega\times [0,T]\subset\mathbb{R}^{n+1}$.

\begin{cases} \displaystyle \frac{\partial u}{\partial t}-\Delta u = -u^3, &\text{in }\Omega \\ u(x,0)=0, &x\in\Omega \\ u(x,t)=0, &x\in\partial\Omega, t>0. \end{cases}

Any suggestions? My thoughts: there is a change of variables that changes the equation to one where the maximum principle holds; suppose $u$ and $v$ are solutions and consider some variation of their difference (e.g., $u-v$, $(u-v)^2$, $e^{\frac{1}{4}(u-v)^4}$, etc.); consider some energy functional. I can't get any of the "standard" techniques to work.


A natural energy functional is $$ E(u) = \int_\Omega |u|^2 \, . $$ Then $E(u(\cdot,0)) = 0$ and $$ \frac{d}{dt} E(u(\cdot,t)) = - 2 \int_\Omega (|\nabla u|^2 + u^4) \le 0 $$ by integration by parts. Therefore $E(u(\cdot,t)) = 0$ for all $t$ and $u = 0$ is the only solution.


Just to be on the safe side, let me affirm at the outset that I assume we are dealing with classical solutions, i.e., sufficiently differentiable functions $u$ satisfying (1) below.

From the given equation

$u_t - \nabla^2u = -u^3, \tag{1}$

we obtain, upon multiplication by $u$,

$uu_t - u\nabla^2 u = -u^4; \tag{2}$

we have

$uu_t = \dfrac{1}{2}(u^2)_t; \tag{3}$

we also have

$\nabla \cdot (u \nabla u) = \langle \nabla u, \nabla u \rangle + u \nabla^2 u, \tag{4}$

from which

$u\nabla^2 u = \nabla \cdot (u\nabla u) - \Vert \nabla u \Vert^2, \tag{5}$


$\langle \nabla u, \nabla u \rangle = \Vert \nabla u \Vert^2. \tag{6}$

Inserting (3) and (5) into (2) yields

$\dfrac{1}{2}(u^2)_t - \nabla \cdot (u \nabla u) + \Vert \nabla u \Vert^2 = -u^4; \tag{7}$

for fixed $s \in [0, T]$, we integrate (7) over $\Omega$ and find

$\dfrac{1}{2}\int_\Omega (u^2(s))_t dV - \int_\Omega \nabla \cdot \ (u(s) \nabla u(s)) dV + \int_\Omega \Vert \nabla u(s) \Vert^2 dV = -\int_\Omega (u(s))^4 dV; \tag{8}$


$\int_\Omega \nabla \cdot (u(s) \nabla u(s)) dV = \int_{\partial \Omega} u(s) \nabla u(s) \cdot dS = 0, \tag{9}$

by the divergence theorem and the boundary conditions. Note that in writing (8), (9), and following, I have adopted the convention that spatial variables are suppressed; thus $u(s)$ is shorthand for $u(\vec x, s)$ with $\vec x \in \Omega$ and $s \in [0, T]$; this notation is especially useful in cases such as the present in which the spatial variables are "integrated out".

We further have

$\int_\Omega (u^2(s))_t dV = \dfrac{d}{dt}\int_\Omega u^2(s) dV, \tag{10}$

and performing a few little algebraic maneuvers we find that (8) becomes

$\dfrac{1}{2} \dfrac{d}{dt}\int_\Omega u^2(s) dV = -\int_\Omega u^4(s) dV - \int_\Omega \Vert \nabla u(s) \Vert^2 dV. \tag{11}$

Integrating (11) 'twixt $t_1, t_2 \in [0, T]$, we obtain

$\dfrac{1}{2} (\int_\Omega u^2(t_2) dV - \int_\Omega u^2(t_1) dV) = \dfrac{1}{2} \int_{t_1}^{t_2} (\dfrac{d}{dt}\int_\Omega u^2 (s) dV) ds$ $ = -\int_{t_1}^{t_2} (\int_\Omega (u^4(s) + \Vert \nabla u(s) \Vert^2 dV)ds. \tag{12}$

We use formula (12) to show that $u = 0$ is the unique solution to (1) satisfying the stated boundary conditions; choising $t_1 = 0$, we find

$\dfrac{1}{2}\int_\Omega u^2(t_2) dV = -\int_0^{t_2} (\int_\Omega (u^4(s) + \Vert \nabla u(s) \Vert^2)dV) ds. \tag{13}$

If now $u(p, t') \ne 0$ for some $(p, t') \in \Omega \times [0, T]$, then since $u$ is continuous, we must have $u \ne 0$ in some neighborhood $W \subset \Omega \times [0, T]$ of $(p, t')$.  Thus $u^2, u^4 > 0$ on $V$, while $\Vert \nabla u \Vert^2 \ge 0$ on $V$.  It follows that both

$\int_\Omega u^2(t') dV > 0 \tag{14}$


$\int_\Omega (u^4(t') + \Vert \nabla u(t') \Vert^2) dV > 0. \tag{15}$

Taken in concert, (13)-(15) form a rather unharmonious trio, bring as they are mutually contradictory.  Thus we must have $u(p, t') = 0$ for $(p, t') \in \Omega \times [0, T]$; the $0$ solution of (1) is indeed unique under the hypothesized boundary conditions.  

Note Added Thursday 11 June 2015 3:34 PM PST: I wanted to leave a few remarks concerning possible generalizations of this result. First of all, for $0 < \alpha \in \Bbb R$ and sufficiently differentiable bounded $\beta(x) > 0$, $x \in \Omega$, the equation

$u_t - \alpha \nabla^2 u = -\beta(x) u^3 \tag{16}$

has precisely one solution $u \equiv 0$ satisfying the given boundary conditions; indeed, it is easy to see that such $u$ satisfy the analogs of (11) and (12):

$\dfrac{1}{2} \dfrac{d}{dt}\int_\Omega u^2(s) dV = -\int_\Omega \beta u^4(s) dV - \alpha\int_\Omega \Vert \nabla u(s) \Vert^2 dV, \tag{17}$

$\dfrac{1}{2} (\int_\Omega u^2(t_2) dV - \int_\Omega u^2(t_1) dV) = \dfrac{1}{2} \int_{t_1}^{t_2} (\dfrac{d}{dt}\int_\Omega u^2 (s) dV) ds$ $ = -\int_{t_1}^{t_2} (\int_\Omega (\beta u^4(s) + \alpha \Vert \nabla u(s) \Vert^2 dV)ds; \tag{18}$

the step-by-step derivations of (17), (18) from (16) are simple extensions of those taking us from (1) to (11)-(12), and our choice of positivity for $\alpha$ and $\beta(x)$ force $u \equiv 0$ just as above. So the result is true for much more general coefficients than as originally stated. Second, the equation

$u_t - \alpha \nabla^2 u = -\beta(x) u^p, \tag{19}$

where $p$ is a positive odd integer leads to the same conclusion, and for essentially the same reasons; in this case (17) becomes

$\dfrac{1}{2} \dfrac{d}{dt}\int_\Omega u^2(s) dV = -\int_\Omega \beta u^{p + 1}(s) dV - \alpha\int_\Omega \Vert \nabla u(s) \Vert^2 dV, \tag{20}$

and since $p + 1$ is even, we have

$\int_\Omega \beta u^{p + 1}(s) dV > 0 \tag{21}$

unless $u = 0$; so the same result binds. This in contrast to the case of $p \ge 0$ even, whence $p + 1$ is odd and we cannot conclude that (21) holds, invalidating the presen line of argument.

It would be engaging to explore the cases $p < 0$ an integer, or even $p$ non-integer, but I think I've gone far enough for the present. End of Note.


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.