Newton's method for computing the nearest correlation matrix with a simple upper bound (Q620444)

A semismooth Newton method is extended to the nearest correlation matrix problem with a simple upper bound. Unlike the original semismooth Newton method proposed by \textit{H. Qi} and \textit{D. Sun} [SIAM J. Matrix Anal. Appl. 28, No. 2, 360--385 (2006; Zbl 1120.65049)] for the unbounded nearest correlation matrix problem, constraint nondegeneracy does not always hold for the optimal solution and thereby the method may lose its quadratic convergence property. The primal nondegeneracy is shown to be equivalent to the positive definiteness of every generalized Hessian (in the sense of Clarke) of the dual objective function. The authors show a quadratic convergence result for a modified semismooth Newton algorithm by adding a perturbation term to the generalized Hessian. Extensive numerical experiments are included.