Perspective: Atomistic Simulations of Water and Aqueous Systems with Machine Learning Potentials

Aug 5, 2026

Upload by Ivy

Water is central to Earth, climate, geophysics, and technology, yet its complex anomalies and phase behavior arise from simple H2O interactions. Computer simulations complement limited experimental resolution, revealing hydrogen-bond networks, solute and surface effects, and extreme pressure/temperature regimes. High-cost ab initio molecular dynamics has constrained system size and timescales, while modern machine-learning potentials combine first-principles accuracy with force-field efficiency. This perspective reviews progress from gas-phase molecules and clusters to bulk water, electrolyte solutions, and solid-liquid interfaces.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Introduction

Water’s scientific importance and experimental limits

Hydrogen-bond networks and simulation insights

Empirical force fields and their limitations

Ab initio molecular dynamics and motivation for ML potentials

arXiv :2401 . 17875v1 [ cond-mat .soft] 31 Jan 2024 Perspective: Atomistic Simulations of Water and Aqueous Systems with Machine Learning Potentials∗ Amir Omranpour, 1, 2 Pablo Montero De Hijes,3, 4 J¨org Behler, 1, 2,† and Christoph Dellago3,‡ 1 Lehrstuhl f¨ur Theoretische Chemie II, Ruhr-Universit¨at Bochum, 44780 Bochum, Germany 2 Research Center Chemical Sciences and Sustainability, Research Alliance Ruhr, 44780 Bochum, Germany 3 University of Vienna, Faculty of Physics, Boltzmanngasse 5, A-1090 Vienna, Austria 4 University of Vienna, Faculty of Earth Sciences, Geography and Astronomy, Josef-Holaubuek-Platz 2, 1090 Vienna, Austria (Dated: February 1, 2024) As the most important solvent, water has been at the center of interest since the advent of computer simulations. While early molecular dynamics and Monte Carlo simulations had to make use of simple model potentials to describe the atomic interactions, accurate ab initio molecular dynamics simulations relying on the first-principles calculation of the energies and forces have opened the way to predictive simulations of aqueous systems. Still, these simulations are very demanding, which prevents the study of complex systems and their properties. Modern machine learning potentials (MLPs) have now reached a mature state, allowing to overcome these limitations by combining the high accuracy of electronic structure calculations with the efficiency of empirical force fields. In this Perspective we give a concise overview about the progress made in the simulation of water and aqueous systems employing MLPs, starting from early work on free molecules and clusters via bulk liquid water to electrolyte solutions and solid-liquid interfaces. INTRODUCTION A large fraction of the surface of the Earth is covered by water and, still, some ice, giving our planet its distinctive blue color when viewed from space. Water is carried down deep into the Earth’s crust at subduction zones, influencing volcanism and plate tectonics, and in the atmosphere, in form of vapor, liquid or ice, water is a key climate factor from the troposphere up to the stratosphere and mesosphere. Down at the Earth’s surface, water shapes landscapes, provides the basis for life and is central to many technologies that sustain humanity. Given its significance and abundance, it is no surprise that over the centuries much research has been undertaken to understand the properties of water and their physical origin. One of the central scientific questions addressed in water research is how the complex behavior of water, exhibiting many anomalies and a rich phase diagram, arises from the interactions of the chemically rather simple H2 O molecules. Due to the limited temporal and spatial resolution of many experimental probes, much of what we know about water has been learned from computer simulations. Specifically, atomistic simulations have provided detailed insights into the directed network of hydrogen bonds between molecules that governs the structure and dynamics of water and its interaction with solutes and surfaces [1–4] . Moreover, computer simulations have made it possible to investigate water at extreme conditions that are not accessible in experiments. For instance, simulations have been used to study water and ice at pressure and temperature conditions prevailing in the deep Earth [5] and in the interiors of the giant planets Uranus and Neptune [6], as well as in the deeply supercooled state, the so-called “no-man’s land”, where crystallization occurs extremely quickly [7, 8] . Following the pioneering Monte Carlo (MC) simulations of Barker [9] and molecular dynamics (MD) studies of Rahman and Stillinger [1] in the late 1960s and early 1970s, respectively, many computer simulations of water and aqueous systems were carried out. Initially, these simulations were based on empirical potentials [10], but later they relied increasingly on forces and energies obtained from electronic st

FAQs

Why are computer simulations important for studying water properties?

Many experimental probes have limited temporal and spatial resolution, so simulations provide detailed microscopic insights into structure, dynamics, and interactions with solutes and surfaces.

What limits traditional ab initio molecular dynamics simulations of water?

Ab initio simulations rely on first-principles energies and forces, which are computationally demanding and therefore restrict the study of complex systems and longer timescales.

How do machine learning potentials address the limitations of both empirical force fields and ab initio methods?

Machine learning potentials aim to retain the accuracy of electronic structure calculations while achieving near force-field efficiency, enabling more extensive simulations of water and aqueous systems.

Aug 5, 2026

Upload by Ivy