Population-Balance Tilt Sampling for Gradient-based Free-Energy Reconstruction
Gradient-based free-energy surface (FES) reconstruction typically faces a trade-off between the complexity of evaluating instantaneous forces and the systematic bias introduced by local asymmetry in position-based estimators. To overcome these limitations, we introduce population-balance tilt sampling (PBTS), an adaptive method that applies a linear tilt to a symmetric restraining potential and determines the central free-energy derivative by finding the tilt root that equalizes sample populations across the window center. Because the underlying molecular and symmetric restraining potentials cancel across tilt states within each window, cumulative multistate Bennett acceptance ratio (MBAR) reweighting depends exclusively on collective-variable displacements and tilt coefficients—enabling full trajectory reuse without instantaneous-force evaluations or potential-energy recalculations. Formulating a unified theory of odd-observable estimators to compare sign-based population balance against distance-based mean balance, we prove via moment inequalities that population balance guarantees a lower leading-order finite-window bias under cubic asymmetric perturbations. Furthermore, it bounds the influence of tail outliers on slope estimation while incurring only a modest increase in asymptotic variance. Applications to the Wolfe–Quapp model and gas-phase alanine dipeptide demonstrate accurate gradient extraction and robust FES reconstruction in one and two dimensions. Beyond free-energy estimation, the window-wise cumulative root histories native to PBTS act as parameter-free local convergence probes, uncovering hidden sampling bottlenecks and metastable branching along slow orthogonal degrees of freedom that are otherwise masked by global integration smoothing.
Authors
- Ye‐Guang Fang (ORCID: https://orcid.org/0009-0002-4271-1954)
- Chongqin Zhu
- Wei-Hai Fang
- Yue Liu
Institutions
- Beijing Normal University (CN)
Publication Details
- Journal
- ChemRxiv
- Published
- 2026-11-20
- DOI
- https://doi.org/10.26434/chemrxiv.15012511/v1
- Primary Topic
- Advanced Physical and Chemical Molecular Interactions
- Type
- preprint