Guaranteeing global optimality of divide-and-conquer phase retrieval

Develop an optimization technique for the divide-and-conquer binary phase-retrieval algorithm that synergizes with its blockwise decomposition while guaranteeing convergence to a globally optimal solution of the discrete Hessian-norm minimization problem.

Background

The proposed binary phase-retrieval algorithm decomposes a cosine-transform magnitude into blocks, solves the resulting QUBO subproblems independently, and then resolves sign discrepancies between blocks. Because the subproblems are coupled through the block-border parameter and inter-block curvature, this divide-and-conquer procedure is not guaranteed to recover a global minimizer of the discrete Hessian-norm objective.

The authors identify the development of an optimization method that preserves the computational advantages of the blockwise decomposition while guaranteeing optimality as a direction for future work. They mention the alternating direction method of multipliers (ADMM) as a possible approach, but note that its iterative nature may increase solve times.

References

However, there is no guarantee that Algorithm \ref{alg:phaseRecovery} converges to a global optimum of eq:disHessMin. Another direction for future work would involve employing an optimization technique that synergizes well with the divide-and-conquer nature of Algorithm \ref{alg:phaseRecovery} while guaranteeing optimality, such as the alternating direction method of multipliers (ADMM).

eq:disHessMin:

$\min_{s\in\{-1,1\}^{m\times n}} \sum_{i,j} \Vert #1{H(s \odotg)}{i}{j} \Vert_F^2, $

— Binary Phase Retrieval of Cosine Transforms via Local Curvature Minimization  (2610.02112 - Ogden et al., 1 Oct 2026) in Conclusion