Speaker
Description
Accurate evaluation of partition functions in molecular and solid-state systems remains computationally challenging due to high dimensionality and expensive energy evaluations. We present two complementary approaches to address this problem.
First, we introduce ZoRRO, a framework based on tailored coordinate transformations into translational, rotational, and vibrational (TRV) degrees of freedom. This representation isolates physically relevant modes, leading to improved energy landscapes and more efficient integration. Within this framework, thermodynamic properties, diffusion coefficients, and molecular uptake in external structures can be computed in a unified manner.
Second, we develop Slice Monte Carlo, a surrogate-guided integration scheme inspired by nested sampling. By leveraging lower-cost surrogate models, such as machine-learned potentials or force fields, to guide exploration, the method enables efficient evaluation of partition functions. The approach is athermal and robust to surrogate misspecification; even partial structural information is sufficient to guide the integration and improve performance, enabling the use of high-level models.
Together, these approaches provide a scalable route to efficient and reliable thermodynamic integration in complex systems.