Avoiding local minima in FROG retrievals with a convergence evaluation metric

Accurate retrieval of ultrashort laser pulses using frequency-resolved optical gating (FROG) is often hindered by local convergence in existing algorithms. We introduce the Sigma Check, a novel supervised evaluation step, to our knowledge, designed to detect and mitigate erroneous convergence to local minima. This method evaluates the difference between measured and retrieved spectrograms, applying targeted perturbation when significant discrepancies are detected. Numerical simulations and experimental measurements demonstrate that the Sigma Check enhances retrieval accuracy across multiple algorithms, including line-search FROG, extended ptychographic iterative engine, and principal components generalized projections algorithm, providing a systematic approach to improving FROG pulse reconstruction.