The Compactness of a Policy Space in Dynamic Programming Via an Extension Theorem for Carathéodory Functions (Q4346231)

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 1042555
Language Label Description Also known as
default for all languages
No label defined
    English
    The Compactness of a Policy Space in Dynamic Programming Via an Extension Theorem for Carathéodory Functions
    scientific article; zbMATH DE number 1042555

      Statements

      The Compactness of a Policy Space in Dynamic Programming Via an Extension Theorem for Carathéodory Functions (English)
      0 references
      9 December 1998
      0 references
      compactness theorem
      0 references
      set of measures on a subset of a product space
      0 references
      optimal policies in stochastic dynamic programming
      0 references
      Carathéodory functions
      0 references
      This paper provides a simplified formulation and self-contained proof of a known compactness theorem for a set of measures on a subset of a product space. Consider the product space \(X\times A\), where \(X\) and \(A\) are complete separable metric spaces, and let \(K\) be a Borel subset of \(X\times A\) with the property that every nonempty cross-section below is compact: \[ K(x) = \left\{ a\in A : (x,a)\in K \right\}, \quad x\in X. \] Form the set \(S\) of all finite measures \(\mu\) on \(K\) enjoying the following relationship to an arbitrary but fixed finite measure \(m\) on \(X\): \[ \mu( (B\times A) \cap K) = m(B), \qquad\forall B\in{\mathcal B}(X). \tag{\(*\)} \] (Here \({\mathcal B}(X)\) denotes the \(\sigma\)-algebra of Borel sets in \(X\).) Theorem 1 asserts that if \(\mu_j\) is a sequence of measures in \(S\), then there must be a subsequence \(\mu_{j'}\) and a measure \(\mu\) in \(S\) such that for every Carathéodory function \(f\) defined on \(K\), one has \[ \int_K f d\mu_{j'} \to \int_K f d\mu. \tag{\(\dag\)} \] (This mode of convergence defines the ``weak-strong'' topology on the set of finite measures on \(K\); a Borel-measurable function \(f\) on \(K\) is called ``Carathéodory'' if, for every \(x\) in \(X\), the function \(a\mapsto f(x,a)\) is continuous on \(K(x)\).) This result can be used to demonstrate the existence of optimal policies in stochastic dynamic programming. The proof relies on a version of the classical Urysohn lemma for Carathéodory functions that has independent interest: Theorem 2 asserts that under the hypotheses above, any Carathéodory function \(f\) on \(K\) can be extended to a Carathéodory function \(F\) on all of \(X\times A\) in a way that preserves its norm and, for every \(x\) in \(X\) where \(K(x)\neq\emptyset\), ensures that the supremum and infimum of the continuous function \(a \mapsto F(x,a)\) are attained at points in \(K(x)\).
      0 references

      Identifiers