The Reverse-order law for the Moore--Penrose inverse is false in general. TakeThenA sufficient condition isso that has full column rank and has full row rank. Indeed and ; these identities make satisfy all four Penrose equations for .
The Hessian of is . Thus is strongly convex exactly when has full row rank. In that case one may useIn all cases, the gradient has Lipschitz constant bounded by
The identity does not hold for arbitrary matrices. A sufficient condition is that have full column rank and have full row rank.