From electrons to phase diagrams with classical and machine learning potentials - automated workflows for materials science with pyiron

Aug 6, 2026

A comprehensive, user-friendly framework for machine learning interatomic potentials (MLPs) is presented using the pyiron integrated development environment. The workflow covers the full life cycle: creating systematic DFT reference databases, fitting DFT data to empirical potentials or MLPs, and validating the resulting parametrization largely automatically. Demonstrations span three distinct potential classes, including EAM, HDNNP, and ACE. As an application and validation case, the framework computes a binary composition-temperature phase diagram for Al-Li, a lightweight aerospace-relevant alloy system.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

Introduction

Machine learning interatomic potentials and motivation

Life cycle of MLP development

Reference database generation

Parameter fitting and validation

Validation importance and phase-diagram applications

arXiv :2403 .05724v1 [ cond-mat .mtrl-sci ] 8 Mar 2024 From electrons to phase diagrams with classical and machine learning potentials: automated workflows for materials science with pyiron Sarath Menon , 1, ∗ Yury Lysogorskiy,2 Alexander L. M. Knoll,3, 4 Niklas Leimeroth ,5 Marvin Poul , 1 Minaam Qamar ,2 Jan Janssen , 1 Matous Mrovec ,2 Jochen Rohrer ,5 Karsten Albe ,5 J¨org Behler ,3, 4 Ralf Drautz ,2 and J¨org Neugebauer 1,† 1 Max-Planck-Institut f¨ur Eisenforschung GmbH, 40237 D¨usseldorf, Germany 2 ICAMS, Ruhr-Universit¨at Bochum, 44801 Bochum, Germany 3 Lehrstuhl f¨ur Theoretische Chemie II, Ruhr-Universit¨at Bochum, 44780 Bochum, Germany 4 Research Center Chemical Sciences and Sustainability, Research Alliance Ruhr, 44780 Bochum, Germany 5 Technische Universit¨at Darmstadt, Fachbereich Material und Geowissenschaften, Fachgebiet Materialmodellierung, 64287 Darmstadt, Germany (Dated: March 12, 2024) We present a comprehensive and user-friendly framework built upon the pyiron integrated development environment (IDE), enabling researchers to perform the entire Machine Learning Potential (MLP) development cycle consisting of (i) creating systematic DFT databases,(ii) fitting the Density Functional Theory (DFT) data to empirical potentials or MLPs, and (iii) validating the potentials ina largely automatic approach. The power and performance of this framework are demonstrated for three conceptually very different classes of interatomic potentials: an empirical potential (embedded atom method-EAM), neural networks (high-dimensional neural network potentials-HDNNP) and expansions in basis sets (atomic cluster expansion-ACE) . As an advanced example for validation and application, we show the computation of a binary composition-temperature phase diagram for Al-Li, a technologically important lightweight alloy system with applications in the aerospace industry. I. INTRODUCTION The advent of machine learning interatomic potentials (MLPs) is revolutionising the field of computational materials science, enabling simulations of large systems and complex material properties with ab initio accuracy [1– 5] . However, the development of these data-driven interatomic potentials is a computationally intensive task that needs automated and reliable workflows. The life cycle of MLP development can be broadly divided into the following tasks: (i) generating a database containing reference data, (ii) fitting the model parameters to the reference data, and (iii) validating the resulting parametrization for a specified range of properties. Furthermore, it is often necessary to provide a feedback loop between the tasks via an active learning approach to ascertain transferability [6, 7] . The initial task of setting up the reference database usually requires to perform many thousands of density functional theory (DFT) calculations for a broad range of atomic environments that span the configuration space of interest as completely as possible. Such computations can nowadays be facilitated using either general workflow frameworks [8–11] or tools designed specifically fora particular MLP class [12–17] . Nevertheless, there is still lack of standardized workflow setups, computational metaparameters and structural databases. Therefore, ∗ [s.menon@mpie.de](s.menon@mpie.de)[ ](s.menon@mpie.de)† [neugebauer@mpie.de](neugebauer@mpie.de) each research group relies mostly on their own expertise and experience. This may not only lead to inconsistencies in the generated data (for instance, due to variations in DFT settings, such as the exchange-correlation functional, Brillouin zone sampling or plane wave cutoff) [18, 19], but it can also strongly limit exchange of data from different sources and their collection into greater databases. The situation remains similar when it comes to the second stage of MLP development, namely, fitting the model parameters. Many optimization algorithms and software tools are tailored to a p

FAQs

What capabilities does pyiron-based automation provide for MLP development?

It supports creating DFT reference databases, fitting model parameters to DFT data, and performing largely automated validation to check performance beyond simple correlations.

Why is validation considered more than correlating energies and forces?

Simple energy/force correlations can be misleading, so the framework emphasizes physical properties and finite-temperature dynamical simulations to detect spurious model behavior.

What potential classes are used to demonstrate the framework?

The examples cover an empirical potential (EAM), neural network potentials (HDNNP), and basis-set expansions (ACE).

How is the framework applied in an advanced example?

It computes a binary composition-temperature phase diagram for the Al-Li alloy system to validate and demonstrate practical use for a technologically important lightweight material.

Aug 6, 2026

Upload by Lucas Martin