Primal-dual active-set methods for large-scale optimization (Q493264): Difference between revisions

The author considers the nearest low-rank correlation matrix problem: \(\underset{M \i S^n} \min\frac12 \| M -C \|{}^2_F\) subject to \(\text{diag}(M)=e\), \(M\geq 0\), \(\text{rank}(M)\leq p\). This problem plays an important role in mathematical finance. The author presents a new easily computed constraint-preserving update scheme, which can be viewed as a generalization of the Cayley transform, a classical retraction on the Stiefel manifold.