On the intersection of continuous local tents (Q1318943): Difference between revisions

Let \(X\) be an Euclidean affine space and \(B\) its unit ball centered at 0. A system of convex cones \(K_ 1,\dots, K_ s\) in \(X\) with common vertex \(x_ 0\) is said to be separable if some one of them can be separated by a hyperplane from the intersection of the others. Further, let \(\Omega\subset X\) and \(x_ 0\in \Omega\). A convex cone \(Q\) in \(X\) with vertex at \(x_ 0\) is called a tent of \(\Omega\) at \(x_ 0\) if there exists a mapping \(\psi: X\to X\) defined in a neighborhood of \(x_ 0\) in \(X\) such that 1) \(\psi\) is differentiable at \(x_ 0\) with \(\psi(x_ 0)= x_ 0\) and \(\psi'(x_ 0)\)= the identity mapping and 2) \(\psi(Q\cap (x_ 0+ \varepsilon B))\subset\Omega\) for some \(\varepsilon>0\). More generally, a convex cone \(K\) in \(X\) with vertex \(x_ 0\) is called a local tent of \(\Omega\) at \(x_ 0\) if, for any \(x\in \text{rint }K\) (= the relative interior of \(K\)), there exists a tent \(Q\subset K\) of \(\Omega\) at \(x_ 0\) such that \(x\in \text{rint } Q\) and \(\text{aff } Q= \text{aff } K\), where aff denotes the affine hull. Also, the notion of continuous or smooth (local) tent is defined by assuming the mapping \(\psi\) above continuous or smooth. These notions are effectively used in deriving systematically necessary conditions for optimality on the basis of Theorem A: Let \(\Omega_ 1,\dots, \Omega_ s\) be subsets of \(X\) having a point \(x_ 0\) in common and \(K_ 1,\dots, K_ s\) a system of continuous local tents of \(\Omega_ 1,\dots, \Omega_ s\) at \(x_ 0\), respectively. Assume that the system is not separable and \(K_ j\not\subset \text{aff } K_ j\) for some \(j\). Then there exists a point \(x_ 1\in \Omega:= \bigcap_{j=1}^ s \Omega_ j\) other than \(x_ 0\). \textit{V. G. Boltjanskij} gave a proof in his monograph: Optimal control of discrete systems (1978; Zbl 0397.49001). But the subtlety of his proof has been preventing the very useful notions from coming into general use. Theorem A is a simple consequence of Theorem B: Let \(\Omega_ 1,\dots, \Omega_ s\) be subsets of \(X\) having a point \(x_ 0\) in common and \(K_ 1,\dots, K_ s\) a system of continuous local tents of \(\Omega_ 1,\dots, \Omega_ s\) at \(x_ 0\), respectively. Assume that the system is not separable. Then \(K:= \bigcap_{j=1}^ s K_ j\) is a local tent of \(\Omega:= \bigcap_{j=1}^ s \Omega_ j\) at \(x_ 0\). In the conclusion of Theorem B, ``local tent'' cannot be replaced by ``continuous local tent'' by virtue of Boltjanskii's counter- example. The author proves Theorem B using a characterization of the non- separability of a system of convex cones and a form of generalized implicit function theorem.