A chain rule involving vector functions of bounded variation (Q1095346)

By \(f\in lbv(I,X)\) the authors mean that f is a function of real interval I to a Banach space X with bounded variation on very compact subinterval of I. With every such function f is associated an X-valued measure on \(I- df\) which is commonly called differential (or Stieltjes) measure. In the special case when X equals a real Hilbert space H with norm denoted by \(\| \cdot \|\), the first author has established that for every \(f\in lbv(I,H)\) \[ (*)\quad d(\| f\|^ 2)=(f^++f^-).df. \] Here the dot on the right hand side refers to the scalar product in H and the functions \(f^+\), \(f^-\) are the right and the left-limit of f at every point. The following inequalities hold in the sense of the ordering of real measures \[ 2f^-.df\leq d(\| f\|^ 2)\leq 2f^+.df. \] This result is generalized as the norm in H is replaced by some convex and continuous function \(\gamma\) from an open and convex subset \(\Omega\) of X to R. As for the second type of results in this paper, \(\gamma\) : \(\Omega\to R\) is supposed to be Lipschitz one and Clarke's generalized gradient of \(\gamma\) is used. In both situations if \(\gamma\) happens to be Gateaux- differentiable, and f-continuous, a chain rule of the familiar form is found to hold. Finally the authors have constructed an expression in order to generalize the equality (*) when the function \(\gamma\) is Frechet-differentiable.