2as if ρ t ⋆ , t = 0, . . . , T and Π t ⋆ , t = 0, . . . , T − 1 are optimal for (3), then ρ t ⋆ , t = 0, . . . , T and Π t ⋆ diag(ρ t ⋆ ), t = 0, . . . , T − 1 are feasible for problem (7). Next we need to show that, given the optimal solution ρ t ⋆ , t = 0, . . . , T and M t ⋆ , t = 0, . . . , T − 1 of (7), we can always reconstruct a feasible solution for (3). To do so, consider the matrices Π t ⋆