1: \begin{abstract}

2:   This paper concerns with a noisy structured low-rank matrix recovery problem which can be modeled

3:   as a structured rank minimization problem. We reformulate this problem as

4:   a mathematical program with a generalized complementarity constraint \!(MPGCC),

5:   and show that its penalty version, yielded by moving the generalized complementarity constraint

6:   to the objective, has the same global optimal solution set as the MPGCC does

7:   whenever the penalty parameter is over a threshold. Then, by solving the exact penalty

8:   problem in an alternating way, we obtain a multi-stage convex relaxation approach.

9:   We provide   theoretical guarantees for our approach under a mild restricted eigenvalue condition,

10:   by quantifying the reduction of the error and approximate rank bounds of the first stage convex

11:   relaxation (which is exactly the nuclear norm relaxation) in the subsequent stages

12:   and establishing the geometric convergence of the error sequence in a statistical sense.

13:   Numerical experiments are conducted for some structured low-rank matrix recovery examples  to

14:   confirm our theoretical findings.

15:  \end{abstract}