Some existence theorems for functional equations arising in dynamic programming (Q790054): Difference between revisions

The authors prove some existence theorems for functional equations which arose in a certain type of continuous multistage decision processes which are described below. Let \(S\subset X\) be the state space and \(D\subset Y\) be the decision space. The state vectors are denoted by x, the decision vectors by y. Let T:\(S\times D\to S\), g:\(S\times D\to R\) and G:\(S\times D\times R\to R\), R being the field of real numbers. The return function f:\(S\to R\) of the continuous decision process is defined by the functional equation \[ (*)\quad f(x)=\sup_{y\in D}[g(x,y)+G(x,y,f(T(x,y)))]\quad(x\in S). \] The authors prove the existence of a solution to the functional equation (*) under various conditions; in some cases they also prove uniqueness. Among the results are the following two typical ones: Theorem A. (i) Let g and G be bounded. (ii) Let \(| G(x,y,z_ 1)- G(x,y,z_ 2)| \leq \phi(| z_ 1-z_ 2|)\) for all \((x,y,z_ 1)\), \((x,y,z_ 2)\) in \(S\times D\times R\), where \(\phi\) :[0,\(\infty)\to [0,\infty)\) is nondecreasing, continuous and \(\phi(r)<r\) for every \(r>0\). Then the functional equation (*) possesses a unique bounded solution on S. Theorem B. (i) Let \(| g(x,y)| \leq M\| x\|\) for (x,y)\(\in S\times D\), where M is a positive constant. (ii) Let \(\| T(x,y)\| \leq \phi(\| x\|)\) for (x,y)\(\in S\times D\), where \(\phi\) :[0,\(\infty)\to [0,\infty)\) is nondecreasing and \(\sum \phi^ n(r)\) converges for every \(r>0\). Then the functional equation \[ f(x)=\sup_{y\in D}[g(x,y)+f(T(x,y))]\quad(x\in S) \] possesses a unique solution \(f_ 0\) satisfying the condition (iii): (iii) If \(x\in S\), \(\{Y_ n\}^{\infty}\!_{n=1}\subset D\) and \(x_ n=T(x_{n-1},Y_ n)\), \((x_ 0=x\), \(n=1,2,3,...)\), then \(f_ 0(x_ n)\to 0\) as \(n\to \infty\).