A physics-driven statistical mechanics framework for many-body thermodynamic and transport property computation

Abstract

Many-body interactions play an important role in accurately modeling high-pressure thermodynamic and transport properties of real fluids. However, analytically resolving many-body interactions is computationally prohibitive. This work introduces a physics-driven statistical mechanics framework, which establishes a many-body virial (MB-Virial) equation of state (EoS) to compute real-fluid thermodynamic and transport properties. For predictions of compressibility, enthalpy, heat capacity, and viscosity, MB-Virial EoS consistently outperforms previous pairwise models. For polar species like H₂O and NH₃, it accurately captures the non-linear behaviors of the properties above within 15% errors between 1-600 atm, while pairwise models reach 270% errors. For long-chain alkanes such as n-C₅H₁₂, MB-Virial EoS maintains errors below 10% for heat capacity and viscosity predictions, whereas pairwise models often diverge beyond 50 atm. By accelerating computations $10^4$ times faster than quantum calculations while maintaining an accuracy of $<0.1$ kJ/mol, the MB-Virial approach offers a universal framework for many-body thermodynamic and transport predictions with various deep potential (DP) models, enabling next-generation simulations of supercritical fluids.

Novelty and significance statement

This work introduces a physics-driven statistical mechanics framework, the MB-Virial EoS, to compute real-fluid thermodynamic and transport properties. In this framework, various deep potential models are used to compute virial coefficients of the MB-Virial EoS directly from statistical mechanics, enabling systematic corrections to thermodynamic and transport properties, including enthalpy, heat capacity, viscosity, and thermal conductivity, over a wide range of temperatures and pressures. By explicitly linking microscopic many-body interactions with macroscopic fluid behavior, the MB-Virial EoS provides a general physics-driven statistical mechanics framework for formulating real-fluid properties directly from underlying potential models.