Some applications of viscosity solutions to optimal control and differential games (Q4357415)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1066229
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Some applications of viscosity solutions to optimal control and differential games |
scientific article; zbMATH DE number 1066229 |
Statements
18 March 1998
0 references
optimal control
0 references
viscosity solution
0 references
value function
0 references
optimal multivalued feedbacks
0 references
almost optimal feedbacks
0 references
differential games
0 references
0.861128032207489
0 references
0.844417154788971
0 references
Some applications of viscosity solutions to optimal control and differential games (English)
0 references
Following an exposition of Dynamic Programming methods for a variety of optimal control problems, the paper provides extensions of results and treatment of such problems using the notion of viscosity solutions. Let \(E\subseteq\mathbb{R}^m\). Let \(F: E\times \mathbb{R}\times\mathbb{R}^m\to\mathbb{R}\) be continuous. A function \(u\) defined on \(E\), which is upper semicontinuous (respectively lower semicontinuous), is a viscosity subsolution (resp. supersolution) of \(F(x,u, Du)=0\) in \(E\) (or a solution of \(F(x,u,Du)\leq 0\) (resp. \(\geq 0\)) in \(E\)) if, for any \(C^1\) function \(\geq\phi\) defined on \(E\) and any local maximum (resp. minimum) point \(x\) of \(u-\phi\), \(F(x,u(x), D\phi(x))\leq 0\) (resp. \(\geq 0\)). A viscosity solution of \(F(x,u,Du)= 0\) is a subsolution which is also a supersolution. Let \(y'(t)= f(y(t),a(t))\) denote the transition function of an optimal control problem, where \(y\) and \(a\), respectively, denote the state and control, and let \(V\) denote the value function. The control \(a\) takes values in a compact metric space \(A\), while \(y\in\mathbb{R}^N\). Define, for \(z\) in \(\mathbb{R}^N\), \(S(z)=\text{argmin}_{a\in A} f(z,a)\cdot DV(z)\). A map \(\phi:\mathbb{R}^N\to A\) is called an admissible feedback if, for all \(x\in\mathbb{R}^n\), there is a unique solution of \(y'= f(y,\phi(y))\), \(t>0\), and \(y(0)= x\). Say that \(\phi\) is optimal for \(x\) if \(\phi(y(\cdot))\) is an optimal control for \(x\). In many problems one faces the difficulty that the value function \(V\) may not be smooth, \(S(z)\) may not be a singleton or that there are no admissible feedbacks such that \(\phi(z)\in S(z)\) for all \(z\). The paper seeks to show how viscosity solutions help to overcome some of these difficulties of the classical Dynamic Programming method, including extensions of the Hamilton-Jacobi-Bellman equation when the value function \(V\) is only continuous, characterization of the value function as a viscosity solution when \(V\) is also bounded, extension of verification theorems to allow nonsmooth functions, synthesis of optimal multivalued feedbacks and simultaneous approximations of the value function and almost optimal feedbacks. Attention is also devoted to differential games as well as problems in which the value function can be discontinuous. For the entire collection see [Zbl 0868.00050].
0 references