2
$\begingroup$

Consider the first-order Hamilton-Jacobi equation (HJ):

$$H(x,u,\nabla u) = 0 \quad \text{ on } \ \Omega,$$ where $\Omega$ is an open set of $\mathbb{R}^n$, $u:\Omega \to \mathbb{R}$, and $H:\Omega \times \mathbb{R} \times \mathbb{R}^n \to \mathbb{R}$ (Hamiltonian) is continuous.

Definition 1 (Crandall-Lions-Evans): We say that $u \in C(\Omega)$ is a viscosity solution of (HJ) iff, $\forall \phi \in C^1(\Omega)$,

$\forall x_0$ point of local maximum of $u-\phi$, $\ H(x_0, u(x_0), \nabla\phi(x_0)) \le 0$;

$\forall x_0$ point of local minimum of $u-\phi$, $\ H(x_0, u(x_0), \nabla\phi(x_0)) \ge 0$.


Q1: Where can I find a detailed proof that we can replace

  • "local maximum" by "strict local maximum" or "global maximum" or "strict global maximum";
  • "local minimum" by "strict local minimum" or "global minimum" or "strict global minimum"

and obtain an equivalent definition? (Or anyway how would that proof go?)

Q2: Is it true that we can replace $C^1$ by $C^k$ or $C^k_{\text{comp}}$, with $1 < k \le \infty$, and obtain an equivalent definition?

$\endgroup$
2
  • $\begingroup$ Just get the ball rolling, surely we can replace the test function $\phi$ by compactly supported ones because all conditions here are local. $\endgroup$
    – Fan Zheng
    Commented Oct 3, 2016 at 17:07
  • $\begingroup$ @FanZheng: That's right. Thanks for your remark. $\endgroup$
    – user99249
    Commented Oct 3, 2016 at 18:53

1 Answer 1

2
$\begingroup$

Only the subsolution case is proven, as the supersolution case is identical.

Q1

Suppose $u$ is a subsolution under the definition with strict extremum. Let $\phi$ be a test function such that $u-\phi$ has a possibly nonstrict maximum at $x_0$. Let $$\psi(x)=\phi(x)+|x-x_0|^2$$ so that $x_0$ is a strict maximum of $u-\psi$. Moreover, $$H(x_0,u,\nabla \phi(x_0)) = H(x_0,u,\nabla \psi(x_0)) \leq 0.$$

Q2

Suppose $u$ is a subsolution under the definition with compact test functions and $1 < k < \infty$. Let $\phi$ be a noncompact test function such that $u-\phi$ has a local maximum at $x_0$. Let $B_r$ denote the Euclidean ball of radius $r$ around $x_0$. Without loss of generality, we assume that $B_3 \subset \Omega$ (otherwise, perform some scaling to fix things). Now, let \begin{align*} \psi(x)&=\phi(x) 1_{B_1}(x) + \zeta(x) \phi(\hat{x}) 1_{B_2 \setminus B_1}(x),\\ \zeta(x)&=\exp(1-1/(1-|x-\hat{x}|^{2k})), \end{align*} and $\hat{x}$ is the unique closest point to $x$ in $\operatorname{cl}B_1$. $\psi$ inherits all the local properties of $\phi$ at $x_0$ and hence $$H(x_0,u,\nabla \phi(x_0)) = H(x_0,u,\nabla \psi(x_0)) \leq 0.$$ The supersolution case is identical.

$\endgroup$
6
  • $\begingroup$ Thank you. So this answer leaves open only the cases "global maximum" (or "minimum") and replacing $C^1$ with $C^k$, $1 < k \le \infty$. $\endgroup$
    – user99249
    Commented Oct 4, 2016 at 12:03
  • $\begingroup$ Also, what is the motivation behind defining $\zeta$ that way? $\endgroup$
    – user99249
    Commented Oct 4, 2016 at 12:05
  • $\begingroup$ @Kei: The global case is already handled by my answer in Q1. To answer your second question, intuitively, $\zeta$ is used to mollify the value of $\psi$ on $\partial B_{1}$ (where it is equal to $\varphi$) with its value on $\partial B_{2}$ (where it is equal to zero). The $C^k$ replacement can be obtained by a density argument. $\endgroup$
    – parsiad
    Commented Oct 4, 2016 at 16:08
  • $\begingroup$ Thanks for your reply. 1. Why is the global case already handled? 2. What kind of density argument are you thinking of? $\endgroup$
    – user99249
    Commented Oct 5, 2016 at 19:17
  • $\begingroup$ Oh sorry. I misread your question. The answer handles "global max" $\iff$ "strict global max". I believe the idea is similar for "local max" $\iff$ "global max". Start with a test function in which a local max is admitted and try to turn that local max into a global max while maintaining all of the local properties. As for the density argument, start with a test function in $C^1$ and approximate it by $C^k$ functions via mollification. $\endgroup$
    – parsiad
    Commented Oct 5, 2016 at 19:27

You must log in to answer this question.