PAC Studio Machine Learning: Human-in-the-Loop Analysis of TDPAC Spectra

Abstract

Time-differential perturbed angular correlation (TDPAC or PAC) analysis is an ill-conditioned inverse problem in which site count, interaction type, correlated hyperfine parameters, damping, and initialization choices can produce competing numerical solutions. This software paper presents PAC Studio ML, a human-in-the-loop Python desktop environment for physics-informed inverse analysis of PAC spectra. The software integrates a Hamiltonian-based forward PAC model, user-defined synthetic training libraries, feature extraction, one-, two-, and three-site machine-learning predictors, direct parameter prediction, Auto sites model-family screening, ML-seeded nonlinear least-squares refinement, visualization, benchmarking, diagnostics, model-card reporting, and export tools. The ML component is designed to support, not replace, conventional fitting and expert interpretation by accelerating parameter exploration, suggesting plausible initialization regions, comparing site-count hypotheses, and improving reproducibility. Held-out synthetic tests demonstrate proof of operation and illustrate the unequal recoverability of PAC parameters in difficult inverse problems. Selected BiFeO3 examples demonstrate conventional, direct-ML, ML-seeded, and Auto sites workflows as software case studies, not as a complete experimental validation corpus. PAC Studio ML is therefore positioned as a supporting tool for expert PAC analysis: it improves workflow speed and diagnostic transparency while final model choice, physical constraints, and materials interpretation remain the responsibility of the researcher.