Abstract:Using normalizing flows and reweighting, Boltzmann Generators enable equilibrium sampling from a Boltzmann distribution, defined by an energy function and thermodynamic state. In this work, we introduce Thermodynamic Interpolation (TI), which allows for generating sampling statistics in a temperature-controllable way. We introduce TI flavors that work directly in the ambient configurational space, mapping between different thermodynamic states or through a latent, normally distributed reference state. Our ambient-space approach allows for the specification of arbitrary target temperatures, ensuring generalizability within the temperature range of the training set and demonstrating the potential for extrapolation beyond it. We validate the effectiveness of TI on model systems that exhibit metastability and non-trivial temperature dependencies. Finally, we demonstrate how to combine TI-based sampling to estimate free energy differences through various free energy perturbation methods and provide corresponding approximated kinetic rates estimated through generator extended dynamic mode decomposition (gEDMD).

What problem does this paper attempt to address?

The problem that this paper attempts to solve is efficient sampling and computation in molecular thermodynamics and kinetics. Specifically, the author introduced a new method - Thermodynamic Interpolation (TI), aiming to generate sample statistics at different temperatures through the generative model, thereby achieving the estimation of free - energy differences and kinetic rates in molecular systems. This method can improve the efficiency of state - transition from high - temperature to low - temperature, and can make effective predictions within the temperature range outside the training data. ### Main contributions of Thermodynamic Interpolation (TI): 1. **Generate samples at controllable temperatures**: - The TI method allows the generation of sample statistics at controllable temperatures, which is helpful for studying the properties of molecular systems under different temperature conditions. - Through two different TI methods (environmental - space TI and latent - space TI), samples can be directly transformed in the configuration space or latent space, thereby achieving efficient temperature transformation. 2. **Estimation of free - energy differences**: - Using the samples generated by TI, the free - energy differences between different thermodynamic states can be estimated through the Free Energy Perturbation (FEP) method and other related techniques. - The author showed how to combine the samples generated by TI and Generator - Extended Dynamic Mode Decomposition (gEDMD) to estimate free - energy differences and kinetic rates. 3. **Estimation of kinetic rates**: - Through the gEDMD method, kinetic information can be extracted from the generated samples, thereby estimating the kinetic exchange rates at different temperatures. - This method can directly estimate the kinetic rates of slow processes and rare events from equilibrium - state data without the need for long - term correlated trajectories. ### Specific applications and verification: - **One - dimensional asymmetric double - well model**: - The author verified the effectiveness of the TI method on a one - dimensional asymmetric double - well model system, demonstrating the ability of this method to generate samples at different temperatures. - Through the Effective Sample Size (ESS) index, the high efficiency of the TI method in generating low - temperature samples was proved. - **Molecular systems**: - The author used the TI method to process the molecular dynamics simulation data of two molecules (N - methylformamide and 3 - prop - 2 - ylpent - 1 - yne) from the QM9 data set. - By generating low - temperature conformation ensembles, calculating free - energy differences, and estimating kinetic exchange rates, the application effect of the TI method in actual molecular systems was demonstrated. ### Summary: This paper proposed a new thermodynamic interpolation method. By generating samples at different temperatures through the generative model, efficient estimation of free - energy differences and kinetic rates in molecular systems was achieved. This method not only improves the sampling efficiency but also shows good generalization ability within the temperature range outside the training data.

Thermodynamic Interpolation: A generative approach to molecular thermodynamics and kinetics

Neural Thermodynamic Integration: Free Energies from Energy-based Diffusion Models

Transferable Boltzmann Generators

From data to noise to data for mixing physics across temperatures with generative artificial intelligence

Multireference Generalization of the Weighted Thermodynamic Perturbation Method.

A Boltzmann generator for the isobaric-isothermal ensemble

Self-adaptive Enhanced Sampling in the Energy and Trajectory Spaces: Accelerated Thermodynamics and Kinetic Calculations

Sampling from Boltzmann densities with physics informed low-rank formats

A path integral methodology for obtaining thermodynamic properties of nonadiabatic systems using Gaussian mixture distributions

Boltzmann Generators: Sampling Equilibrium States of Many-Body Systems with Deep Learning

Efficient mapping of phase diagrams with conditional Boltzmann Generators

Enhancing torsional sampling using fully adaptive simulated tempering

Free Energy Landscapes, Diffusion Coefficients, and Kinetic Rates from Transition Paths

Target-aware Bayesian inference via generalized thermodynamic integration

From Thermodynamics to Kinetics: Enhanced Sampling of Rare Events

Molecular free energies, rates, and mechanisms from data-efficient path sampling simulations

An introduction to thermodynamic integration and application to dynamic causal models

Iterated Energy-based Flow Matching for Sampling from Boltzmann Densities

Flow Perturbation to Accelerate Unbiased Sampling of Boltzmann distribution

Generalized Flow Matching for Transition Dynamics Modeling

Normalizing flows as an enhanced sampling method for atomistic supercooled liquids