Archimedean levels, semispaces, and majorization of convex cones (Q689825)

As the term is used here, a cone is a nonempty subset \(C\) of a real vector space such that \(0\not\in C\) and \(C=C+C= ]0,\infty[C\). In other words, \(C\) is a convex cone that has the origin 0 as an apex but does not contain the origin. The linear hull of \(C\) is denoted by \(L_ C\). A cone \(C\) is majorized by a cone \(K\) (written \(C<_ m K\)) if there exists a linear transformation \(T: L_ C\to L_ k\) such that \(T(C)\subset K\); and \(C\) and \(K\) are \(m\)-equivalent if each majorizes the other. Each cone \(C\) generates two partial orderings of \(L_ C\), and these play an essential role. They are defined as follows: \[ x<_ C y\Longleftrightarrow y-x\in C; \quad x\ll_ C y\Longleftrightarrow \lambda x<_ C y \text{ for all } \lambda\in \mathbb{R}. \] By means of the following theorem and a corollary, the paper provides a complete solution of the problems of recognizing when one given finite-dimensional cone is majorized by another, and of classifying \(m\)-equivalence classes of such cones. There are also some remarks of infinite-dimensional cones, but they do not lead to any definitive conclusion. Theorem. For each positive integer \(n\) and infinite-dimensional cone \(C\), the following five conditions are equivalent: (a) the archimedean level of \(C\) is at most \(n\); (b) \(C\) does not contain any \((n+1)\)-semispace at the origin; (c) \(C\) is majorized by the lexicopositive cone \(\mathbb{L}^ n_ +\) of \(\mathbb{R}^ n\); (d) the poset \((C,\ll_ C)\) is the union of \(n\) or fewer antichains; (e) \((-C)\cap C_{(n)}=\emptyset\). (A \(k\)-semispace at a point \(p\) is a set which, for some \(k\)-dimensional flat \(F\) through \(p\), is maximal with respect to being a convex subset of \(F\setminus\{p\}\). For each set \(X\), \(X_{(k)}\) denotes the set of all points \(p\in L_ X\) such that \(p\in X\) or \(X\) contains a \(k\)-semispace at \(p\). When there is a finite upper bound for the cardinality of a chain in the poset \((C,\ll_ C)\), the least such bound is called the archimedean level of \(C\)).