Genetic algorithms for computational materials discovery accelerated by machine learning

Paul C. Jennings, Steen Lysgaard, Jens Strabo Hummelshøj, Tejs Vegge*, Thomas Bligaard

*Corresponding author for this work

Research output: Contribution to journalJournal articleResearchpeer-review

217 Downloads (Pure)

Abstract

Materials discovery is increasingly being impelled by machine learning methods that rely on pre-existing datasets. Where datasets are lacking, unbiased data generation can be achieved with genetic algorithms. Here a machine learning model is trained on-the-fly as a computationally inexpensive energy predictor before analyzing how to augment convergence in genetic algorithm-based approaches by using the model as a surrogate. This leads to a machine learning accelerated genetic algorithm combining robust qualities of the genetic algorithm with rapid machine learning. The approach is used to search for stable, compositionally variant, geometrically similar nanoparticle alloys to illustrate its capability for accelerated materials discovery, e.g., nanoalloy catalysts. The machine learning accelerated approach, in this case, yields a 50-fold reduction in the number of required energy calculations compared to a traditional "brute force" genetic algorithm. This makes searching through the space of all homotops and compositions of a binary alloy particle in a given structure feasible, using density functional theory calculations.
Original languageEnglish
Article number46
Journaln p j Computational Materials
Volume5
Number of pages6
ISSN2057-3960
DOIs
Publication statusPublished - 2019

Cite this

@article{c7e593937f644bca942db63b74fa0cfa,
title = "Genetic algorithms for computational materials discovery accelerated by machine learning",
abstract = "Materials discovery is increasingly being impelled by machine learning methods that rely on pre-existing datasets. Where datasets are lacking, unbiased data generation can be achieved with genetic algorithms. Here a machine learning model is trained on-the-fly as a computationally inexpensive energy predictor before analyzing how to augment convergence in genetic algorithm-based approaches by using the model as a surrogate. This leads to a machine learning accelerated genetic algorithm combining robust qualities of the genetic algorithm with rapid machine learning. The approach is used to search for stable, compositionally variant, geometrically similar nanoparticle alloys to illustrate its capability for accelerated materials discovery, e.g., nanoalloy catalysts. The machine learning accelerated approach, in this case, yields a 50-fold reduction in the number of required energy calculations compared to a traditional {"}brute force{"} genetic algorithm. This makes searching through the space of all homotops and compositions of a binary alloy particle in a given structure feasible, using density functional theory calculations.",
author = "{C. Jennings}, Paul and Steen Lysgaard and Hummelsh{\o}j, {Jens Strabo} and Tejs Vegge and Thomas Bligaard",
year = "2019",
doi = "10.1038/s41524-019-0181-4",
language = "English",
volume = "5",
journal = "n p j Computational Materials",
issn = "2057-3960",
publisher = "Nature Publishing Group",

}

Genetic algorithms for computational materials discovery accelerated by machine learning. / C. Jennings, Paul; Lysgaard, Steen; Hummelshøj, Jens Strabo; Vegge, Tejs; Bligaard, Thomas .

In: n p j Computational Materials, Vol. 5, 46, 2019.

Research output: Contribution to journalJournal articleResearchpeer-review

TY - JOUR

T1 - Genetic algorithms for computational materials discovery accelerated by machine learning

AU - C. Jennings, Paul

AU - Lysgaard, Steen

AU - Hummelshøj, Jens Strabo

AU - Vegge, Tejs

AU - Bligaard, Thomas

PY - 2019

Y1 - 2019

N2 - Materials discovery is increasingly being impelled by machine learning methods that rely on pre-existing datasets. Where datasets are lacking, unbiased data generation can be achieved with genetic algorithms. Here a machine learning model is trained on-the-fly as a computationally inexpensive energy predictor before analyzing how to augment convergence in genetic algorithm-based approaches by using the model as a surrogate. This leads to a machine learning accelerated genetic algorithm combining robust qualities of the genetic algorithm with rapid machine learning. The approach is used to search for stable, compositionally variant, geometrically similar nanoparticle alloys to illustrate its capability for accelerated materials discovery, e.g., nanoalloy catalysts. The machine learning accelerated approach, in this case, yields a 50-fold reduction in the number of required energy calculations compared to a traditional "brute force" genetic algorithm. This makes searching through the space of all homotops and compositions of a binary alloy particle in a given structure feasible, using density functional theory calculations.

AB - Materials discovery is increasingly being impelled by machine learning methods that rely on pre-existing datasets. Where datasets are lacking, unbiased data generation can be achieved with genetic algorithms. Here a machine learning model is trained on-the-fly as a computationally inexpensive energy predictor before analyzing how to augment convergence in genetic algorithm-based approaches by using the model as a surrogate. This leads to a machine learning accelerated genetic algorithm combining robust qualities of the genetic algorithm with rapid machine learning. The approach is used to search for stable, compositionally variant, geometrically similar nanoparticle alloys to illustrate its capability for accelerated materials discovery, e.g., nanoalloy catalysts. The machine learning accelerated approach, in this case, yields a 50-fold reduction in the number of required energy calculations compared to a traditional "brute force" genetic algorithm. This makes searching through the space of all homotops and compositions of a binary alloy particle in a given structure feasible, using density functional theory calculations.

U2 - 10.1038/s41524-019-0181-4

DO - 10.1038/s41524-019-0181-4

M3 - Journal article

VL - 5

JO - n p j Computational Materials

JF - n p j Computational Materials

SN - 2057-3960

M1 - 46

ER -