Enhanced Low-Frequency Dosimetry Simulations with the Energy-Variation Maximum Entropy Snapshot Sampling

Projection-based model order reduction enables efficient low-frequency dosimetry simulations by leveraging precomputed solution vectors from the underlying large-scale system. Since these solution vectors primarily contain localized information, the maximum entropy snapshot sampling method can identify the most informative ones in a preprocessing step. From this subset, a compact reduced basis can be efficiently constructed. While the method is memory-efficient, it relies on state-recurrence quantification of snapshot information to construct the required recurrence matrix, which introduces a computational bottleneck. This work addresses this limitation by proposing an energy-variation-based alternative approach to enhance computational performance.