Mean First-Passage-Time-Directed Bias Optimization with Gaussian-Kernel Markov Models
- 1Department of Physics and Astronomy, University College London, Gower Street, London, United Kingdom, WC1E 6BT
- 2School of Medicine, Pharmacy and Biomedical Sciences, Faculty of Science and Health
- 3University of Portsmouth
Abstract
Rare transitions limit the direct simulation of molecular kinetics and complicate the design of enhanced-sampling biases that accelerate specific events without obscuring their mechanism. We develop a data-driven framework in which biased trajectory data are combined using dynamic histogram analysis and represented by either conventional grids or normalized kernel-anchor memberships to construct finite row-stochastic kinetic models. These models are used to adapt a bounded bias by directly minimizing an absorbing mean first-passage time (MFPT) objective, while reactive-current diagnostics independently assess whether accelerated trajectories preserve the reference transition mechanism. We first test the kernel-induced representation in equilibrium free-energy reconstruction and then apply the kinetic optimization framework to progressively more challenging rareevent problems, including membrane permeation, multidimensional barrier crossing, and a metastable molecular rearrangement. The benchmark calculations show that MFPT-directed biasing can efficiently promote target arrival and that soft kernel states provide a flexible alternative to rigid spatial discretization. At the same time, pathway analysis shows that faster arrival does not imply preservation of the underlying transition mechanism. A confounded high-dimensional stress test further exposes the limitations of projected kinetic models when the relevant dynamical coordinates are unresolved.
Related
- Correlation-consistent Gaussian basis sets for copper solids from material-constrained atomic optimization
- State-selective molecular orientation by vibrational Autler-Townes adiabatic passage
- Extending the mean first-passage-time formalism to the reconstruction of high free-energy barriers: A top-start approach
- Serial‐Passage Assessment Shows No Confirmed Resistance Development to Norway Spruce (Picea Abies) Resin in Bacterial Species Relevant to Wound Infection
- Radial Kernel Attention for Long-Range Modeling in Polymer Machine Learning Force Fields
BibTeX
@misc{wang2026mean,
title = {{Mean First-Passage-Time-Directed Bias Optimization with Gaussian-Kernel Markov Models}},
author = {Qingying Wang and Tingyue Zhou and Baowen Li and Christian Jorgensen and Edina Rosta},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15009623/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009623/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Mean First-Passage-Time-Directed Bias Optimization with Gaussian-Kernel Markov Models
AU - Qingying Wang
AU - Tingyue Zhou
AU - Baowen Li
AU - Christian Jorgensen
AU - Edina Rosta
PY - 2026
DA - 2026/09/29
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009623/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009623/v1
AB - Rare transitions limit the direct simulation of molecular kinetics and complicate the design of enhanced-sampling biases that accelerate specific events without obscuring their mechanism. We develop a data-driven framework in which biased trajectory data are combined using dynamic histogram analysis and represented by either conventional grids or normalized kernel-anchor memberships to construct finite row-stochastic kinetic models. These models are used to adapt a bounded bias by directly minimizing an absorbing mean first-passage time (MFPT) objective, while reactive-current diagnostics independently assess whether accelerated trajectories preserve the reference transition mechanism. We first test the kernel-induced representation in equilibrium free-energy reconstruction and then apply the kinetic optimization framework to progressively more challenging rareevent problems, including membrane permeation, multidimensional barrier crossing, and a metastable molecular rearrangement. The benchmark calculations show that MFPT-directed biasing can efficiently promote target arrival and that soft kernel states provide a flexible alternative to rigid spatial discretization. At the same time, pathway analysis shows that faster arrival does not imply preservation of the underlying transition mechanism. A confounded high-dimensional stress test further exposes the limitations of projected kinetic models when the relevant dynamical coordinates are unresolved.
ER -