Redirecting to original paper in 30 seconds...

Click below to go immediately or wait for automatic redirect

arxiv_ml 80% Match Research Paper Computational Physicists,Computational Chemists,Machine Learning Researchers,Materials Scientists,Biophysicists 1 week ago

FEAT: Free energy Estimators with Adaptive Transport

graph-neural-networks › molecular-modeling
📄 Abstract

Abstract: We present Free energy Estimators with Adaptive Transport (FEAT), a novel framework for free energy estimation -- a critical challenge across scientific domains. FEAT leverages learned transports implemented via stochastic interpolants and provides consistent, minimum-variance estimators based on escorted Jarzynski equality and controlled Crooks theorem, alongside variational upper and lower bounds on free energy differences. Unifying equilibrium and non-equilibrium methods under a single theoretical framework, FEAT establishes a principled foundation for neural free energy calculations. Experimental validation on toy examples, molecular simulations, and quantum field theory demonstrates improvements over existing learning-based methods. Our PyTorch implementation is available at https://github.com/jiajunhe98/FEAT.
Authors (7)
Jiajun He
Yuanqi Du
Francisco Vargas
Yuanqing Wang
Carla P. Gomes
José Miguel Hernández-Lobato
+1 more
Submitted
April 15, 2025
arXiv Category
stat.ML
arXiv PDF Code

Key Contributions

FEAT is a novel framework for free energy estimation using learned transports via stochastic interpolants. It provides consistent, minimum-variance estimators based on established theorems and unifies equilibrium and non-equilibrium methods under a single theoretical foundation for neural calculations.

Business Value

Accelerates scientific discovery in fields like drug design and materials science by enabling more accurate and efficient computation of crucial thermodynamic properties.

Paper Metadata

Innovation Type

Framework/Methodology

Deployment Feasibility

Feasible, as it provides a theoretical framework and an open-source implementation, applicable to various simulation domains.

Limitations Addressed

Addresses the critical challenge of accurate and efficient free energy estimation, unifying disparate theoretical approaches and establishing a principled foundation for neural methods in this domain.

Performance Gains

Improvements over existing learning-based methods.

View Code on GitHub

Technical Tags

free energy estimationadaptive transportstochastic interpolantsJarzynski equalityCrooks theoremvariational boundsequilibrium methodsnon-equilibrium methodsneural free energy calculationPyTorch

Research Topics

Statistical MechanicsComputational PhysicsMachine LearningMolecular DynamicsThermodynamics

Methods & Architectures

Learned transportsStochastic interpolantsEscorted Jarzynski equalityControlled Crooks theoremVariational upper and lower boundsUnified framework for equilibrium/non-equilibrium methods Neural Networks (implied)Stochastic processes

Applications & Tasks

Molecular Dynamics Simulations Computational Chemistry Quantum Field Theory Materials Science Drug Discovery Free Energy CalculationThermodynamic Property EstimationSimulation Analysis Estimating free energy differencesUnifying equilibrium and non-equilibrium simulation methodsDeveloping principled neural free energy calculations

Datasets & Benchmarks

Benchmarks
Toy examples • Molecular simulations • Quantum field theory

Related Fields

Computational PhysicsStatistical MechanicsMachine LearningComputational ChemistryScientific Computing

Keywords

free energyestimationadaptive transportstochastic interpolantsJarzynskiCrooksneural networksmolecular dynamicsquantum field theorythermodynamicssimulationPyTorch

Academic Context

#Statistical Mechanics#Computational Physics#Machine Learning#Molecular Dynamics#Thermodynamics

Technology Stack

Frameworks & Libraries
PyTorch
Programming Languages
Python

Commercial Potential

Potential Products
Software libraries for enhanced molecular simulation analysisTools for predicting drug efficacy and material propertiesAccelerated scientific computing platforms
Target Industries
PharmaceuticalsBiotechnologyMaterials ScienceChemical IndustryAcademia/Research
Use Case Examples
Predicting binding affinities of drug candidates to target proteins.Designing novel materials with desired thermodynamic properties.Studying phase transitions in physical systems.
Competitive Edge
Offers a unified and principled framework for neural free energy calculations, improving upon existing learning-based methods by leveraging adaptive transport and unifying different theoretical approaches.
Market Opportunity
Significant market for computational tools in drug discovery, materials science, and fundamental research.
Revenue Models
Licensing of softwareconsulting servicesdevelopment of specialized simulation tools.

Resource Requirements

Compute Needs
High, requires significant computational resources for running simulations and training neural networks.
Data Requirements
Requires simulation data (e.g., trajectories from molecular dynamics) and potentially labeled datasets for training.
Deployment Constraints
Integration with existing simulation software, computational cost for large-scale applications.
Scalability
Scalability depends on the efficiency of the learned transport methods and the underlying simulation techniques.

Production Readiness

Maturity Level

Research/Development

Time to Market

2-5 years for widespread adoption in specific scientific domains.

Licensing

MIT License (implied by GitHub availability)

Patent Potential

Moderate, potential for patents on novel learned transport mechanisms or specific applications in drug discovery/materials science.

View Full Paper Back to Papers