Serena Della Corte,Richard C. Kraaij
Abstract:We give a new perspective on the existence of viscosity solutions for a stationary and a time-dependent first order Hamilton-Jacobi equation. Following recent comparison principles, we work in a framework in which we consider a subsolution and a supersolution for two equations in terms of two Hamiltonians that can be seen as an upper and lower bound of our original Hamiltonian respectively. The upper and lower bound are formulated in terms of a relaxed Lyapunov function which allows us to restrict part of the analysis to compact sets and to work with almost optimizers of the considered control problems. For this reason, we can relax assumptions on the control problem: most notably, we do not need completeness of set of controlled paths. Moreover, our strategy avoids a-priori analysis on the regularity of the candidate solutions. We then explore the context of Hamilton-Jacobi-Bellman and Hamilton-Jacobi-Isaacs equations to verify our assumptions in a convex and non-convex setting respectively.
What problem does this paper attempt to address?
This paper aims to solve the problem of the existence of viscosity solutions of first - order Hamilton - Jacobi equations on Riemann manifolds. Specifically, the author focuses on the following two types of equations:
1. **Static equation**:
\[
u(x)-\lambda H(x, du(x)) = h(x),
\]
where \(\lambda>0\) and \(h\) is a bounded continuous function.
2. **Time - dependent equation**:
\[
\begin{cases}
\partial_t u(x, t)+\lambda u(x, t)-H(x, du(·, t)(x)) = 0, &\text{if }t > 0,\\
u(x, 0)=u_0(x), &\text{if }t = 0,
\end{cases}
\]
where \(\lambda\geq0\).
To prove the existence of viscosity solutions of these equations, the author introduces a new perspective by using the Lyapunov control method to establish the upper and lower bounds of the Hamiltonian. The specific steps are as follows:
1. **Define candidate solutions**:
- For the static equation, the candidate solution \(R_{\lambda, h}(x)\) is defined as:
\[
R_{\lambda, h}(x)=\sup_{\gamma\in\text{Adm},\gamma(0) = x}J_{\lambda}(\gamma),
\]
where
\[
J_{\lambda}(\gamma)=\int_0^{\infty}\lambda^{-1}e^{-\lambda^{-1}t}\left(h(\gamma(t))-\int_0^t L(\gamma(s),\dot{\gamma}(s))ds\right)dt.
\]
- For the time - dependent equation, the candidate solution \(v_{\lambda}(x, t)\) is defined as:
\[
v_{\lambda}(x, t)=\sup_{\gamma\in\text{Adm},\gamma(0) = x}W_{\lambda}(\gamma, t),
\]
where
\[
W_{\lambda}(\gamma, t)=\int_0^t -e^{-\lambda s}L(\gamma(s),\dot{\gamma}(s))ds+e^{-\lambda t}u_0(\gamma(t)).
\]
2. **Introduce Lyapunov functions**:
- Define two new operators \(H^{\dagger}\) and \(H^{\ddagger}\) through the Lyapunov function \(\Upsilon\), which are used to construct the upper and lower bounds of the Hamiltonian respectively.
3. **Prove the main results**:
- By using tools such as the Dynamic Programming Principle and Young's inequality, it is proved that the upper semi - continuous regularization \((R_{\lambda, h})^*\) of the candidate solution is a viscosity subsolution of the static equation, and its lower semi - continuous regularization \((R_{\lambda, h})_*\) is a viscosity supersolution.
- Similarly, for the time - dependent equation, it is proved that \((v_{\lambda})^*\) is a viscosity subsolution and \((v_{\lambda})_*\) is a viscosity supersolution.
By this method, the author not only avoids the a priori regularity of the candidate solution.