How graph neural network interatomic potentials extrapolate: Role of the message-passing algorithm DOI
Sungwoo Kang

The Journal of Chemical Physics, Год журнала: 2024, Номер 161(24)

Опубликована: Дек. 23, 2024

Graph neural network interatomic potentials (GNN-IPs) are gaining significant attention due to their capability of learning from large datasets. Specifically, universal based on GNN, usually trained with crystalline geometries, often exhibit remarkable extrapolative behavior toward untrained domains, such as surfaces and amorphous configurations. However, the origin this extrapolation is not well understood. This work provides a theoretical explanation how GNN-IPs extrapolate geometries. First, we demonstrate that can capture non-local electrostatic interactions through message-passing algorithm, evidenced by tests toy models density-functional theory data. We find GNN-IP models, SevenNet MACE, accurately predict forces in indicating they have learned exact functional form Coulomb interaction. Based these results, suggest ability learn interactions, coupled embedding nature GNN-IPs, explains ability. GNN-IP, SevenNet-0, effectively infers domains but fails arising kinetic term, which supports suggested theory. Finally, address impact hyperparameters performance potentials, SevenNet-0 MACE-MP-0, discuss limitations capabilities.

Язык: Английский

Crystal Structure Prediction of Cs–Te with Supervised Machine Learning DOI Creative Commons
Holger‐Dietrich Saßnick, Caterina Cocchi

Advanced Theory and Simulations, Год журнала: 2025, Номер unknown

Опубликована: Янв. 24, 2025

Abstract Crystal structure prediction methods aim to determine the ground‐state crystal for a given material. The vast combinatorial space associated with this problem makes conventional computationally prohibitive routine use. To overcome these limitations, novel approach combining high‐throughput density functional theory calculations machine learning is proposed. It predicts stable structures within binary and ternary systems by systematically evaluating various structural descriptors algorithms. superiority of models based on atomic coordination environments shown, transfer‐learned graph neural networks emerging as particularly promising technique. By validating proposed method Cs–Te crystals, its ability generate proved, suggesting potential advancing established computational schemes.

Язык: Английский

Процитировано

1
Band Alignment of Oxides by Learnable Structural-Descriptor-Aided Neural Network and Transfer Learning DOI Creative Commons
Shin Kiyohara, Yoyo Hinuma, Fumiyasu Oba

и другие.

Journal of the American Chemical Society, Год журнала: 2024, Номер 146(14), С. 9697 - 9708

Опубликована: Март 28, 2024

The band alignment of semiconductors, insulators, and dielectrics is relevant to diverse material properties device structures utilizing their surfaces interfaces. In particular, the ionization potential electron affinity are fundamental quantities that describe surface-dependent band-edge positions with respect vacuum level. Their accurate systematic determination, however, demands elaborate experiments or simulations for well-characterized surfaces. Here, we report machine learning nonmetallic oxides using a high-throughput first-principles calculation data set containing about 3000 oxide Our neural network accurately predicts relaxed binary simply by information on bulk surface termination planes. Moreover, extend model naturally include multiple-cation effects transfer it ternary oxides. present approach enables vast number solid surfaces, thereby opening way understanding materials screening.

Язык: Английский

Процитировано

9
Designing semiconductor materials and devices in the post-Moore era by tackling computational challenges with data-driven strategies DOI
Jiahao Xie, Yansong Zhou, Muhammad Faizan

и другие.

Nature Computational Science, Год журнала: 2024, Номер 4(5), С. 322 - 333

Опубликована: Май 23, 2024

Язык: Английский

Процитировано

7
Computational design of energy‐related materials: From first‐principles calculations to machine learning DOI
Haibo Xue, Guanjian Cheng, Wan‐Jian Yin

и другие.

Wiley Interdisciplinary Reviews Computational Molecular Science, Год журнала: 2024, Номер 14(5)

Опубликована: Сен. 1, 2024

Abstract Energy‐related materials are crucial for advancing energy technologies, improving efficiency, reducing environmental impacts, and supporting sustainable development. Designing discovering these through computational techniques necessitates a comprehensive understanding of the material space, which is defined by constituent atoms, composition, structure. Depending on search space involved in investigation, design can be categorized into four primary approaches: atomic substitution fixed prototype structures, crystal structure prediction (CSP), variable‐composition CSP, inverse across entire space. This review provides an overview paradigms, detailing concepts, strategies, applications pertinent to energy‐related materials. The progression from first‐principles calculations machine learning emphasized, with aim enhancing elucidating new advancements computationally article under: Structure Mechanism > Computational Materials Science Data Artificial Intelligence/Machine Learning Electronic Theory Density Functional

Язык: Английский

Процитировано

5
Data-Efficient Multifidelity Training for High-Fidelity Machine Learning Interatomic Potentials DOI
Jaesun Kim,

Jisu Kim,

Jaehoon Kim

и другие.

Journal of the American Chemical Society, Год журнала: 2024, Номер unknown

Опубликована: Дек. 17, 2024

Machine learning interatomic potentials (MLIPs) are used to estimate potential energy surfaces (PES) from

Язык: Английский

Процитировано

4
Crystal Structure Prediction Using a Self-Attention Neural Network and Semantic Segmentation DOI

Wuling Zhao,

Mingting Zhou,

Jialin Shao

и другие.

Journal of Chemical Information and Modeling, Год журнала: 2025, Номер unknown

Опубликована: Апрель 14, 2025

The development of new materials is a time-consuming and resource-intensive process. Deep learning has emerged as promising approach to accelerate this However, accurately predicting crystal structures using deep remains significant challenge due the complex, high-dimensional nature atomic interactions scarcity comprehensive training data that captures full diversity possible configurations. This work developed neural network model based on set comprising thousands crystallographic information files from existing structure databases. incorporates self-attention mechanism enhance prediction accuracy by extracting both local global features three-dimensional structures, treating atoms in each point sets. enables effective semantic segmentation accurate unit cell prediction. Experimental results demonstrate for cells containing up 500 atoms, achieves 89.78%.

Язык: Английский

Процитировано

0
Machine Learning-Assisted Exploration and Identification of Aqueous Dispersants in the Vast Diversity of Organic Chemicals DOI
Hirokuni Jintoku, Don N. Futaba

ACS Applied Materials & Interfaces, Год журнала: 2024, Номер 16(9), С. 11800 - 11808

Опубликована: Фев. 23, 2024

Dispersion represents a central processing method in the organization of nanomaterials; however, strong interparticle interaction significant obstacle to fabricating homogeneous and stable dispersions. While dispersants can greatly assist overcoming this obstacle, appropriate type is dependent on such factors as nanomaterial, solvent, experimental conditions, etc., there no general guide selection from vast number possibilities. We report strategy successful demonstration machine-learning-based "Dispersant Explorer", which surveys identifies suitable open databases. Through combined use molecular descriptors derived SMILES databases, model showed exceptional predictive accuracy surveying about ∼1000 chemical compounds identifying those that could be applied dispersants. Furthermore, fabrication transparent conducting films using predicted previously unknown dispersant exhibited highest sheet resistance transmittance compared with other reported undoped films. This result highlights that, addition opening new avenues for novel discovery, machine learning has potential elucidate structures essential optimal dispersion performance advancement complex topic nanomaterial processing.

Язык: Английский

Процитировано

3
Intelligent Structure Searching and Designs for Nanoclusters: Effective Units in Atomic Manufacturing DOI Creative Commons
Junfeng Gao,

Luneng Zhao,

Yuan Chang

и другие.

Advanced Intelligent Systems, Год журнала: 2024, Номер 6(6)

Опубликована: Апрель 26, 2024

Clusters, an aggregation of several to thousands atoms, molecules, or ions, are the building blocks novel functional materials by atomic manufacturing and exhibit excellent applications in catalysis, quantum information, nanomedicine. The evolution cluster structures has been studied for many years. Many effective structural search methods, such as genetic algorithm, basin‐hopping, so on, have developed. However, efficient execution these methods relies on precise energy calculators, density theory (DFT) calculations. Up now, limited computational capabilities, researches mainly focus free‐standing clusters, which different from clusters practical applications. Recently, rapid development big data‐driven machine learning is expected replace DFT high‐precision large‐scale computing. In this review, present challenges currently faced summarized. It proposed that artificial intelligence potential solve some problems including properties complex environment, causing revolutionary developments fields nanomedicine based clusters.

Язык: Английский

Процитировано

3
Machine learning interatomic potentials in engineering perspective for developing cathode materials DOI
Dohyeong Kwon, Duho Kim

Journal of Materials Chemistry A, Год журнала: 2024, Номер 12(35), С. 23837 - 23847

Опубликована: Янв. 1, 2024

Machine learning interatomic potentials (MLIPs) predict thermodynamic phase stability and structural parameters like density functional theory (DFT) but are much faster, making them valuable for engineering applications.

Язык: Английский

Процитировано

3
AI-empowered digital design of zeolites: Progress, challenges, and perspectives DOI Creative Commons
Mengfan Wu, Shiyi Zhang, Jie Ren

и другие.

APL Materials, Год журнала: 2025, Номер 13(2)

Опубликована: Фев. 1, 2025

The rise of artificial intelligence (AI) as a powerful research tool in materials science has been extensively acknowledged. Particularly, exploring zeolites with target properties is vital significance for industrial applications, integrating AI technologies into zeolite design undoubtedly brings immense promise the advancements this field. Here, we provide comprehensive review AI-empowered digital zeolites. It showcases state-of-the-art progress predicting zeolite-related properties, employing machine learning potentials simulations, using generative models inverse design, and aiding experimental synthesis challenges perspectives are also discussed, emphasizing new opportunities at intersection This expected to offer crucial guidance advancing innovations through future.

Язык: Английский

Процитировано

0