Representing individual electronic states for machine learning GW band structures of 2D materials

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: Feb. 3, 2022

Choosing optimal representation methods of atomic and electronic structures is essential when machine learning properties materials. We address the problem representing quantum states electrons in a solid for purpose leaning state-specific properties. Specifically, we construct fingerprint based on energy decomposed operator matrix elements (ENDOME) radially projected density (RAD-PDOS), which are both obtainable from standard functional theory (DFT) calculation. Using such fingerprints train gradient boosting model set 46k G

Language: Английский

---

CEGANN: Crystal Edge Graph Attention Neural Network for multiscale classification of materials environment

npj Computational Materials, Journal Year: 2023, Volume and Issue: 9(1)

Published: Feb. 16, 2023

Abstract We introduce Crystal Edge Graph Attention Neural Network (CEGANN) workflow that uses graph attention-based architecture to learn unique feature representations and perform classification of materials across multiple scales (from atomic mesoscale) diverse classes ranging from metals, oxides, non-metals hierarchical such as zeolites semi-ordered mesophases. CEGANN can classify based on a global, structure-level representation space group dimensionality (e.g., bulk, 2D, clusters, etc.). Using representative polycrystals zeolites, we demonstrate its transferability in performing local atom-level tasks, grain boundary identification other heterointerfaces. classifies (thermal) noisy dynamical environments demonstrated for zeolite nucleation growth an amorphous mixture. Finally, use multicomponent systems with thermal noise compositional diversity. Overall, our approach is material agnostic allows multiscale atomic-scale crystals heterointerfaces microscale boundaries.

Language: Английский

---

Machine Learning Paves the Way for High Entropy Compounds Exploration: Challenges, Progress, and Outlook

Advanced Materials, Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 9, 2023

Abstract Machine learning (ML) has emerged as a powerful tool in the research field of high entropy compounds (HECs), which have gained worldwide attention due to their vast compositional space and abundant regulatability. However, complex structure HEC poses challenges traditional experimental computational approaches, necessitating adoption machine learning. Microscopically, can model Hamiltonian system, enabling atomic‐level property investigations, while macroscopically, it analyze macroscopic material characteristics such hardness, melting point, ductility. Various algorithms, both methods deep neural networks, be employed research. Comprehensive accurate data collection, feature engineering, training selection through cross‐validation are crucial for establishing excellent ML models. also holds promise analyzing phase structures stability, constructing potentials simulations, facilitating design functional materials. Although some domains, magnetic device materials, still require further exploration, learning's potential is substantial. Consequently, become an indispensable understanding exploiting capabilities HEC, serving foundation new paradigm Artificial‐intelligence‐assisted exploration.

Language: Английский

---

MatGPT: A Vane of Materials Informatics from Past, Present, to Future

Advanced Materials, Journal Year: 2023, Volume and Issue: 36(6)

Published: Oct. 10, 2023

Abstract Combining materials science, artificial intelligence (AI), physical chemistry, and other disciplines, informatics is continuously accelerating the vigorous development of new materials. The emergence “GPT (Generative Pre‐trained Transformer) AI” shows that scientific research field has entered era intelligent civilization with “data” as basic factor “algorithm + computing power” core productivity. continuous innovation AI will impact cognitive laws methods, reconstruct knowledge wisdom system. This leads to think more about informatics. Here, a comprehensive discussion models infrastructures provided, advances in discovery design are reviewed. With rise paradigms triggered by “AI for Science”, vane informatics: “MatGPT”, proposed technical path planning from aspects data, descriptors, generative models, pretraining directed collaborative training, experimental robots, well efforts preparations needed develop generation informatics, carried out. Finally, challenges constraints faced discussed, order achieve digital, intelligent, automated construction joint interdisciplinary scientists.

Language: Английский

---

ET-AL: Entropy-targeted active learning for bias mitigation in materials data

Applied Physics Reviews, Journal Year: 2023, Volume and Issue: 10(2)

Published: April 10, 2023

Growing materials data and data-driven informatics drastically promote the discovery design of materials. While there are significant advancements in models, quality resources is less studied despite its huge impact on model performance. In this work, we focus bias arising from uneven coverage families existing knowledge. Observing different diversities among crystal systems common databases, propose an information entropy-based metric for measuring bias. To mitigate bias, develop entropy-targeted active learning (ET-AL) framework, which guides acquisition new to improve diversity underrepresented systems. We demonstrate capability ET-AL mitigation resulting improvement downstream machine models. This approach broadly applicable discovery, including autonomous dataset trimming reduce as well other scientific domains.

Language: Английский

---

Rise of machine learning potentials in heterogeneous catalysis: Developments, applications, and prospects

Chemical Engineering Journal, Journal Year: 2024, Volume and Issue: 494, P. 152757 - 152757

Published: June 2, 2024

Language: Английский

---

Accurate and Transferable Machine Learning Potential for Molecular Dynamics Simulation of Sodium Silicate Glasses

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(3), P. 1358 - 1370

Published: Jan. 13, 2024

An accurate and transferable machine learning (ML) potential for the simulation of binary sodium silicate glasses over a wide range compositions (from 0 to 50% Na2O) was developed. The energy surface is approximated by sum atomic contributions mapped neural network algorithm from local geometry comprising information on distances angles with neighboring atoms using DeePMD code [Wang, H. Comput. Phys. Commun. 2018, 228, 178–184]. Our model trained large data set total energies forces computed at density functional theory level structures extracted classical molecular dynamics (MD) simulations performed several temperatures 300 3000 K. This allows generation robust ML applicable full compositional glass formability different that outperforms empirical potentials available in literature reproducing properties such as bond angle distribution, distribution functions, vibrational state. generality approach enables future training other or more elements allowing structures, properties, behavior ternary multicomponent oxide nearly ab initio accuracy fraction computational cost.

Language: Английский

---

An overview about neural networks potentials in molecular dynamics simulation

International Journal of Quantum Chemistry, Journal Year: 2024, Volume and Issue: 124(11)

Published: May 21, 2024

Abstract Ab‐initio molecular dynamics (AIMD) is a key method for realistic simulation of complex atomistic systems and processes in nanoscale. In AIMD, finite‐temperature dynamical trajectories are generated by using forces computed from electronic structure calculations. with high numbers components typical AIMD run computationally demanding. On the other hand, machine learning (ML) subfield artificial intelligence that consist set algorithms show experience use input output data where capable analysing predicting future. At present, main application ML techniques atomic simulations development new interatomic potentials to correctly describe potential energy surfaces (PES). This technique constant progress since its inception around 30 years ago. The combine advantages classical methods, is, efficiency simple functional form accuracy first principles this article we review evolution four generations some their most notable applications. focuses on MLPs based neural networks. Also, present state art topic future trends. Finally, report results scientometric study (covering period 1995–2023) about impact applied simulations, distribution publications geographical regions hot topics investigated literature.

Language: Английский

---

Machine Learning Interatomic Potentials for Reactive Hydrogen Dynamics at Metal Surfaces Based on Iterative Refinement of Reaction Probabilities

The Journal of Physical Chemistry C, Journal Year: 2023, Volume and Issue: 127(50), P. 24168 - 24182

Published: Dec. 4, 2023

The reactive chemistry of molecular hydrogen at surfaces, notably dissociative sticking and evolution, plays a crucial role in energy storage fuel cells. Theoretical studies can help to decipher underlying mechanisms reaction design, but studying dynamics surfaces is computationally challenging due the complex electronic structure interfaces high sensitivity barriers. In addition, ab initio dynamics, based on density functional theory, too demanding accurately predict or desorption probabilities, as it requires averaging over tens thousands initial conditions. High-dimensional machine learning-based interatomic potentials are starting be more commonly used gas-surface yet robust approaches generate reliable training data assess how model uncertainty affects prediction dynamic observables not well established. Here, we employ ensemble learning adaptively while assessing performance with full quantification (UQ) for probabilities scattering different copper facets. We use this approach investigate two message-passing neural networks, SchNet PaiNN. Ensemble-based UQ iterative refinement allow us expose shortcomings invariant pairwise-distance-based feature representation dynamics.

Language: Английский

---

Mixtures Recomposition by Neural Nets: A Multidisciplinary Overview

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(3), P. 597 - 620

Published: Jan. 29, 2024

Artificial Neural Networks (ANNs) are transforming how we understand chemical mixtures, providing an expressive view of the space and multiscale processes. Their hybridization with physical knowledge can bridge gap between predictivity understanding underlying This overview explores recent progress in ANNs, particularly their potential 'recomposition' mixtures. Graph-based representations reveal patterns among mixture components, deep learning models excel capturing complexity symmetries when compared to traditional Quantitative Structure–Property Relationship models. Key such as Hamiltonian networks convolution operations, play a central role representing The integration ANNs Chemical Reaction Physics-Informed for inverse kinetic problems is also examined. combination sensors shows promise optical biomimetic applications. A common ground identified context statistical physics, where ANN-based methods iteratively adapt by blending initial states training data. concept recomposition unveils reciprocal inspiration reactive highlighting behaviors influenced environment.

Language: Английский

---