Reviews: Dynamic matrix recovery from incomplete observations under an exact low-rank constraint

My primary concerns are listed below: A. Proof 1. a) In line 379 (supplementary material) Theorem 2.3 from 5 cannot be directly used to establish RIP for \sum_t \sqrt{wt} At(\Delta) as \sum_t wt At(\Delta) _2 2 NOT \sum_t \sqrt{wt} At(\Delta) _2 2. However, as theorem 3.1 requires bound on only the former, the proof in [5] can be extended. The proof needs some work though. B. Theory 2. The results are derived for exact global solution to (2) which is a non-convex optimization. The paper is incomplete without the suggested future work on analyzing alternating minimization. I believe the analysis will follow through with a bit of linear algebra machinery, but the exact expression of the joint error term arising from Thm 3.4 and 3.8 and the weighted power iteration is more useful towards understanding the tradeoff of sample complexity and accuracy using the dynamic estimation in practical implementations.