Iterative Algorithms for Ptychographic Phase Retrieval

Ptychography promises diffraction limited resolution without the need for high resolution lenses. To achieve high resolution one has to solve the phase problem for many partially overlapping frames. Here we review some of the existing methods for solving ptychographic phase retrieval problem from a numerical analysis point of view, and propose alternative methods based on numerical optimization.

💡 Research Summary

Ptychography has emerged as a powerful lens‑free imaging technique capable of reaching diffraction‑limited resolution by exploiting the redundancy inherent in overlapping illumination frames. The central computational challenge lies in retrieving the lost phase information from intensity‑only diffraction measurements across many partially overlapping scans. This paper provides a comprehensive review of the major iterative phase‑retrieval algorithms from a numerical‑analysis perspective and introduces a new class of optimization‑driven methods that aim to improve convergence speed, robustness to noise, and computational efficiency.
Traditional algorithms such as the extended Ptychographic Iterative Engine (ePIE) and its variants rely on alternating updates of the object and probe (detector) functions. Each iteration consists of forward Fourier transforms of the current estimate, replacement of the magnitude with the measured intensity, and inverse transforms to update the estimates locally. While these methods are simple, memory‑light, and well‑suited for streaming data, they suffer from slow convergence, sensitivity to the initial guess, and limited tolerance to measurement noise. Global optimization approaches—e.g., the Difference Map (DM), Ptychographic Constrained Gradient (PCG), and variational Bayesian schemes—formulate the reconstruction as a large‑scale nonlinear least‑squares problem that simultaneously considers all frames. By defining a global cost function (often a sum of intensity residuals plus regularization terms) and applying gradient‑based or quasi‑Newton solvers, these methods achieve faster convergence and higher fidelity but at the expense of substantially higher computational and memory demands.
The authors recast the ptychographic phase‑retrieval problem as a constrained optimization over two complex vectors: the object (x) and the probe (p). For each frame (k), the measured intensity (I_k) satisfies (I_k = |F_k(x\odot p)|^2), where (F_k) denotes the frame‑specific Fourier operator and (\odot) the element‑wise product. The shared‑object constraint—that all frames are generated by the same (x) and (p)—is incorporated via Lagrange multipliers or penalty terms, yielding the objective
\