对于一般的非凸函数,其方向导数下具备任何凸性,可以利用一般正齐次函数的回收函数来给出它的一个上凸近似。该上凸近似可用于研究一类非光滑优化——方向可微优化的最优性条件。
For a general convex function, directional derivative is not convex. A recession function of a positively homogenous function can be presented as an upper convex approximation, which can be used to study the optimality conditions of a class of nonsmooth optimization-directionally differentiable optimization.