Abstract:We present eMAGPIE (extended Multilevel-Adaptive-Guided Ptychographic Iterative Engine), a stochastic multigrid method for blind ptychographic phase retrieval that jointly recovers the object and the probe. We recast the task as the iterative minimization of a quadratic surrogate that majorizes the exit-wave misfit. From this surrogate, we derive closed-form updates, combined in a geometric-mean, phase-aligned joint step, yielding a simultaneous update of the object and probe with guaranteed descent of the sampled surrogate. This formulation naturally admits a multigrid acceleration that speeds up convergence. In experiments, eMAGPIE attains lower data misfit and phas…
Abstract:We present eMAGPIE (extended Multilevel-Adaptive-Guided Ptychographic Iterative Engine), a stochastic multigrid method for blind ptychographic phase retrieval that jointly recovers the object and the probe. We recast the task as the iterative minimization of a quadratic surrogate that majorizes the exit-wave misfit. From this surrogate, we derive closed-form updates, combined in a geometric-mean, phase-aligned joint step, yielding a simultaneous update of the object and probe with guaranteed descent of the sampled surrogate. This formulation naturally admits a multigrid acceleration that speeds up convergence. In experiments, eMAGPIE attains lower data misfit and phase error at comparable compute budgets and produces smoother, artifact-reduced phase reconstructions.
| Subjects: | Numerical Analysis (math.NA); Optimization and Control (math.OC) |
| Cite as: | arXiv:2511.01793 [math.NA] |
| (or arXiv:2511.01793v1 [math.NA] for this version) | |
| https://doi.org/10.48550/arXiv.2511.01793 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Borong Zhang [view email] [v1] Mon, 3 Nov 2025 17:47:34 UTC (18,475 KB)