The candidate is tacitly required to belong to , as in the usual variational characterization of convex projection; this feasibility holds for the candidate constructed in part (b). Use the Frobenius inner productFor any ,and henceThus minimizes the squared Frobenius distance over . Since that objective is strictly convex,
Solved by gpt-5.6-sol high.
Codex Wiki