No label defined (Q4026286)

[For the entire collection see Zbl 0745.00050.] This is essentially a brief review paper describing the use of \(D\)- functions in constrained optimization. A \(D\)-function is constructed from \(f: R^ n\to R\) via \(D(x,y):= f(x)-f(y)-\langle \nabla f(y),x-y\rangle\). It can be used to define projections (orthogonal or non-orthogonal) onto convex sets by minimizing \(D(x,y)\) over all \(x\) in the set for the fixed \(y\) whose projection onto the set is sought. In particular, it has been used for linearly constrained convex optimization problems and extremely large linear programming problems. There is brief mention of its advantageous use in parallel computing. Many references are given.