Bibliography

(eng): Nonconvex optimization problems need a relaxation to handle effectively fast oscillation (and possibly also concentration) effects. This uses Young measures or their generalizations. Approximation of the relaxed problem can then be made by various ways, but computationally the most effective way appears to use adaptively a maximum principle (if it forms also a sufficient optimality condition) with the Hamiltonian guessed approximately from a previous iteration, e.g. from a coarser mesh.