Novel computational routes to materials discovery

Lead Research Organisation: University of Warwick
Department Name: Chemistry

Abstract

Understanding the behaviour of materials on the atomic scale is fundamental to modern science and technology, because most properties and phenomena are ultimately controlled by the details of atomistic processes. During the past decades computer simulations on the atomistic
level became a powerful tool in modern chemistry, augmenting experiments, by making initial predictions, aiding studies under extreme conditions or providing an atomistic insight into mechanisms. For example, predicting the state of matter in planetary interiors or in nuclear reactors where measurements are impossible or dangerous, or pinpointing stable structures and properties efficiently, such as for trial drugs or alloys, reduces the amount of expensive and time-consuming experiments.

One of the major fields where computer simulations became widely used is material science, studying phase transitions and phase diagrams. A phase diagram shows the properties of a given material at specific conditions, for example, tells whether a substance is found as gas, liquid or solid at a particular temperature and pressure, or at a particular composition in case of a multicomponent system. It also shows when these phases transform into each other, corresponding to phase transitions. It is of great technological importance to have a complete picture of the phase diagram, and computational tools are widely employed to enable this. Nonetheless, the main difficulty in using computer simulations is that the number of possible ways atoms can be arranged in space is enormous, and no technique is capable of considering all of them, hence we need importance sampling. A plethora of computational techniques exist, however, these are usually problem specific and rely on prior knowledge of the atomic structure, limiting their predictive power. I have been developing a novel computational technique, nested sampling (NS), which addresses these challenges from a new perspective: it automatically generates all relevant atomic configurations (a small subset of all possible variations), and determines their relative stability, offering complete thermodynamic information without any advance knowledge of the material, except its composition.

I have already shown how NS can be used to calculate the phase diagram of metals and alloys, in an automated way, and my aim is to extend its applicability to a broader range of problems: augment crystal structure prediction studies (highly relevant in developing pharmaceuticals), a novel application in calculating spectroscopic properties (for accurate measurements of composition in climate science and astrochemistry), and develop strategies to determine and improve the reliability of potential models (the mathematical formulation of atomic interactions) benefiting computational research in a wide context.

Planned Impact

Academic impact: This project will achieve cross-disciplinary academic impact spanning through not only the the computational and theoretical chemistry community but through those fields that intensively use data generated by computer modelling on the atomistic level: materials science, geology, spectroscopy and environmental chemistry. The new computational technique and the protocol to generate improved computational models will be widely disseminated. The project is also expected to aid me in consolidating my membership of the computational materials science community as an independent researcher.

Economic impact: As the available computational resources are becoming more powerful, it is increasingly more important that high-throughput computational techniques are developed in order to optimise efficiency both for academic and industrial applications, as human resources are expensive and limited. The proposed methodology fulfils all necessary criteria: it can be fully automated, predictions can be made without advance knowledge, technique is not limited to a single type of material, allowing the straightforward calculation of thermodynamic properties under a wide range of conditions. In the long term, this will have a high impact in areas where a large number of screening computations have to be performed, hence automation is crucial for high efficiency and cost-effective manufacturing, such as in pharmaceuticals (screening large amount of candidate molecules while searching for crystal structures with appropriate properties) or alloy development (screening a large number of different compositions, searching for structures stable under certain conditions). Looking beyond the lifetime of this project, it is envisaged that technology transfer will allow accurate, predictive, high-throughput simulations to be carried out directly by industrial partners.

Knowledge: Many of the issues that will be investigated in this project have great fundamental interest whose importance extends beyond computational modelling. Studying the behaviour and structure of materials under a variety conditions help us better understand the interior of our planet, the chemistry of the atmosphere or the unique properties of certain elements. I expect to deliver significant contributions to the state of knowledge on these topics.

People and societal impact: The high costs of laboratory investigations and conditions where measurements are impossible or dangerous (e.g. extreme high pressure or radiation) mean that theory must come in to support experiments to produce new knowledge. During the fellowship I will use and develop new methodology that will help scientists and engineers to model materials, thus saving time and precious resources, enabling more sustainable production in the long term. I will also contribute to inform the general public of these research advances, through a series of events, including public lectures and school demonstrations.
 
Description Our research in computational materials modelling, focusing on phase transitions and phase stability has led us to uncover unexpected phase properties in various model systems, such as a new crystalline structure or phase transition, highlighting the importance of unbiased sampling for revealing new material characteristics. Our work also shows that our configuration phase sampling technique has a great potential to contribute to the development of state-of-the-art machine learning-based potentials -- by assessing the reliability of models and suggesting configurations for automated data extension, contributing to more accurate simulations in material science.
By expanded our computational tool to work on 2D systems, surfaces, and confined spaces, we are allowing a broader range of systems to be studied with this powerful tool and lead to better understanding how materials behave under different conditions.
Exploitation Route The outcomes of this funding offer a valuable computational methodology, poised to be adopted by other researcher groups as well. Its applicability extends to the development of more reliable atomistic models, aiding in the prediction of material properties and has a potential to contribute to the design of e.g. functional materials or catalysts.
Sectors Chemicals

Energy

Manufacturing

including Industrial Biotechology

 
Title Exploring the configuration space of elemental carbon with empirical and machine learned interatomic potentials 
Description This dataset contains a vertical slice of the data used to generate the results found in the publication "Exploring the configuration space of elemental carbon with empirical and machine learned interatomic potentials" It contains nested sampling input files and trajectory files for each potential studied, as well as the xml files and training data for the new potential, GAP-20U+gr. 
Type Of Material Database/Collection of data 
Year Produced 2022 
Provided To Others? Yes  
Impact Improved version of the carbon potential, made available for the community 
URL https://zenodo.org/record/7463706
 
Title Platinum nanoparticle database 
Description This is a database of Pt nanoparticles generated with a GAP interatomic potential for platinum [1]. The files are provided in ASE's extended XYZ format. The database contains the following entries: NP-DB01.yxz: 8000 Pt nanoparticles with sizes between 10 and 349 atoms, generated with a "cooking" protocol. NP-DB02.yxz: 3400 Pt nanoparticles with sizes between 10 and 349 atoms, generated with a "crystallization" protocol. This is the paper reference for this database: J. Kloppenburg, L. B. Pártay, H. Jónsson, M. A. Caro. "A general-purpose machine learning Pt interatomic potential for an accurate description of bulk, surfaces and nanoparticles". J. Chem. Phys. 158, 134704 (2023) Protocols "Cooking": annealing at 1500 K for 20 ps from fcc + quenching down to 100 K over 20 ps + geometry relaxation to local minimum. "Crystallization": annealing at 1150 K for 20 ps from a random spherical distribution + quenching down to 100 K over 20 ps + geometry relaxation to local minimum. References J. Kloppenburg and M.A. Caro. General-purpose GAP potential for platinum. DOI: 10.5281/zenodo.7415219. 
Type Of Material Database/Collection of data 
Year Produced 2022 
Provided To Others? Yes  
Impact Training dataset for ML potential, to enable the community to assess its reliability and enable future improvements or modifications. 
URL https://zenodo.org/record/7415542
 
Title Surface phase diagrams from nested sampling - raw calculation data 
Description This dataset contains raw input and output of calculations presented in the manuscript "Surface phase diagrams from nested sampling" by Mingrui Yang, Livia B. Pártay & Robert B. Wexler, https://arxiv.org/abs/2308.08509. The input and checkpoint trajectories can be used to reproduce the published results, or resume the calculations. The Python scripts used to analyze the output and to produce the figures are also included. 
Type Of Material Database/Collection of data 
Year Produced 2024 
Provided To Others? Yes  
Impact These data are part of proof of concept calculations of new methodology and thus important to assess the reliability and usage of the methodology. 
URL https://data.library.wustl.edu/record/103650
 
Description Application of NS in soft-core models with Gyorgy Hantal 
Organisation University of Natural Resources and Life Sciences
Country Austria 
Sector Academic/University 
PI Contribution My research group brought expertise of computational methods into this collaboration, performed the computations and performed the majority of analysis of data.
Collaborator Contribution Partner brought in expertise in soft-core potential models and fluid-fluid phase transitions.
Impact Publication: Insight into Liquid Polymorphism from the Complex Phase Behavior of a Simple Model PHYSICAL REVIEW LETTERS 127, 015701 (2021)
Start Year 2020
 
Description Calculating surface phase diagrams 
Organisation Washington University in St Louis
Country United States 
Sector Academic/University 
PI Contribution Our group has expertise on configuration space sampling and exploration methods, using computational techniques to locate and characterise phase transitions. We bring methodological and simulation knowhow to this collaboration.
Collaborator Contribution Dr Wexler's group has experience in simulating adsorption processes on crystalline surfaces and modelling nanomaterials, and thus brings extensive knowledge on surfaces processes and current challenges of modelling surface phenomena.
Impact One publication under review currently: arXiv:2308.08509
Start Year 2022
 
Description Nested sampling of the configuration space of machine learnt potentials 
Organisation Aalto University
Country Finland 
Sector Academic/University 
PI Contribution Provide know-how of nested sampling simulations and perform calculations to understand thermodynamic properties of machine learnt interatomic potentials.
Collaborator Contribution Knowledge of the Gaussian Approximation Potential (GAP) framework as well as neural network potentials to train and fit ML potentials. Provided access to existing models and databases and collaborated on developing new models, using nested sampling in assessing their reliability and performance.
Impact Three publications to date: 10.1103/PhysRevMaterials.7.123804 10.1038/s41524-023-01081-w 10.1063/5.0143891
Start Year 2022
 
Description ACORN Warwick 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Schools
Results and Impact A day long conference (in-person and/or online) intended for A-level students to hear about chemical research.
Year(s) Of Engagement Activity 2020,2021