1: \begin{abstract}

2: 

3: A number of statistical estimation problems can be addressed by semidefinite programs (SDP). While SDPs are solvable in polynomial time using interior point methods, 

4: in practice generic SDP solvers do not scale well to high-dimensional problems. In order to cope with this problem, Burer and Monteiro proposed a non-convex rank-constrained

5: formulation, which has good performance in practice but is still poorly understood theoretically.  

6: 

7: In this paper we study the rank-constrained version of SDPs  arising in MaxCut and in synchronization problems. We establish a

8: Grothendieck-type inequality that proves that all the local maxima and  dangerous saddle points are within a small multiplicative 

9: gap from the global maximum. We use this structural information to prove that SDPs can be solved within a known accuracy, by applying the Riemannian trust-region 

10: method to this non-convex problem, while constraining the rank to be  of order one. 

11: For the MaxCut problem, our inequality implies that any local maximizer of the rank-constrained SDP provides a $ (1 - 1/(k-1)) \times 0.878$ approximation 

12: of the MaxCut, when the rank is fixed to  $k$. 

13: 

14: We then apply our results to data matrices generated according to the Gaussian $\integers_2$ synchronization problem,

15: and the two-groups stochastic block model with large bounded degree.

16: We prove that the error achieved by local maximizers undergoes a phase transition at the same threshold as for information-theoretically optimal methods.

17: \end{abstract}

18: