The problem interface projected-gradient descent consumes: a backend, an operator A and linear term q defining the quadratic 0.5 x^T A x + q^T x, a convex space to project onto, and a mutable operator workspace. More...

#include <pgd.hpp>

Concept definition

template<class Problem>
concept ProjectedProblem = requires(const Problem& p) {
  p.backend();
  p.A();
  p.q();
  p.space();
  p.workspace();
}

Detailed Description

A CQPP satisfies this. An LCP does NOT (and must not): its constraint space appears only under reduction to a CQPP, so an LCP is solved by PGD via to_cqpp rather than natively.