Machine learning force fields for molecular liquids - Ethylene Carbonate/Ethyl Methyl Carbonate binary solvent

Aug 5, 2026

Highly accurate ab initio molecular dynamics is the benchmark for molecular mechanisms in condensed phases, but it is too costly to capture slowly converging properties over long time scales. This work develops a robust machine-learning potential for the EC:EMC binary solvent, relevant to liquid electrolytes in rechargeable Li-ion batteries. It addresses scale separation between intra- and inter-molecular interactions, identifies modeling ingredients for the mixture, and evaluates how total-loss fitting affects accuracy and thermodynamic reproduction compared with reference methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Motivation: bridging ab initio accuracy and long time scales

Scale separation challenges in molecular mixtures

ML force-field construction for EC:EMC

Total-interaction fitting without explicit scale separation

Battery-electrolyte relevance and composition (3:7 EC:EMC)

Modeling strategy and interaction modeling

Role of intra- vs inter-molecular accuracy

Electrostatics and short-range model sufficiency

Training-set design via iterative training

[www.nature.com/npjcompumats](www.nature.com/npjcompumats) ARTICLE OPEN Machine learning force ﬁelds for molecular liquids: Ethylene Carbonate/Ethyl Methyl Carbonate binary solvent Ioan-Bogdan Magdău1 ✉ , Daniel J. Arismendi-Arrieta 2, Holly E. Smith 3, Clare P. Grey 3, Kersti Hermansson2 and Gábor Csányi 1 | | Highly accurate ab initio molecular dynamics (MD) methods are the gold standard for studying molecular mechanisms in the condensed phase, however, they are too expensive to capture many key properties that converge slowly with respect to simulation length and time scales. Machine learning (ML) approaches which reach the accuracy of ab initio simulation, and which are, at the same time, sufﬁciently affordable hold the key to bridging this gap. In this work we present a robust ML potential for the EC:EMC binary solvent, a key component of liquid electrolytes in rechargeable Li-ion batteries. We identify the necessary ingredients needed to successfully model this liquid mixture of organic molecules. In particular, we address the challenge posed by the separation of scale between intra- and inter-molecular interactions, which is a general issue in all condensed phase molecular | | | --- | --- | --- | | | | | | | | | | | | | | | | | | | | | | | | | | | systems. | | | | npj Computational Materials (2023)9:146; [https://doi.org/10.1038/s41524-023-01](https://doi.org/10.1038/s41524-023-01)100-w | | | | | | INTRODUCTION The development of ML methods for molecular modelling has opened the possibility to simulate large length scales and longtime scales with ab initio-level accuracy1–3. ML methods have been employed successfully to model isolated molecules4–8, where the directional intra-molecular interactions dominate, as well as inorganic solids and liquids, where the interactions are homogeneous9–22. The molecular condensed phase presents unique challenges owing to a large separation of scales between intra-and inter-molecular interactions23–28. Molecular mixtures, in particular, present further complications owing to the heterogeneity of the inter-molecular environments. Many important properties of the molecular liquids such as: density, viscosity and dielectric constant; depend speciﬁcally, and critically on inter-molecular forces. It is therefore crucial that ML models obtain good accuracies on both intra- and inter-scales. Numerous studies have demonstrated ML potentials for water and aqueous solutions29–33, methane26, ionic liquids28 and, more recently, electrolyte solutions34–36. The standard approach to modelling these molecular systems is to create separate force ﬁelds for the intra-and inter-contributions2,37–39. This approach is inspired by classical force ﬁelds and neatly solves the problem of scale separation. However the approximation breaks down in reactive systems where the molecular identity changes during the course of the simulation. In this work, we intentionally choose to ﬁt ML models on total interactions, without an explicit separation of scales, so our potential are entirely general and could, in principle, capture reactivity40,41. We aim to answer the overarching question: can we develop an accurate and robust ML force ﬁeld for molecular liquids mixtures without making the scale separation explicit? In particular, can these models generate long and stable molecular dynamics trajectories and reproduce the thermodynamic properties of the reference method? As shown in other studies, even reproducing the density of a molecular system is not a trivial task42 and merits a meticulous investigation. Here we show that a good ﬁt on the total loss function results in good intra-, but poor inter-relative accuracies. In practice, this makes it difﬁcult to ﬁt the inter-contribution which underlies the thermodynamic properties of the liquid state. Speciﬁcally, this paper develops an ML force ﬁeld for the binary solvent EC:EMC (3:7 M) of the standard LP57 electrolyte (1M L

FAQs

Why are ab initio molecular dynamics simulations difficult for molecular liquids?

They provide gold-standard accuracy but are prohibitively expensive to run long enough to capture properties that converge slowly with simulation length and time scales.

What is the main goal of the proposed machine-learning force field?

To develop an accurate, robust ML potential for the EC:EMC binary solvent mixture without explicitly separating intra- and inter-molecular interactions, while still reproducing thermodynamic properties.

How does fitting on total interactions affect intra- and inter-molecular accuracy?

A good fit on the total loss improves intra-molecular relative accuracy but can lead to poor inter-molecular relative accuracy, making inter-contribution fitting difficult.

Aug 5, 2026

Upload by Olivia Brown