Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery

Matter, Journal Year: 2021, Volume and Issue: 4(5), P. 1578 - 1597

Published: April 5, 2021

The modular nature of metal–organic frameworks (MOFs) enables synthetic control over their physical and chemical properties, but it can be difficult to know which MOFs would optimal for a given application. High-throughput computational screening machine learning are promising routes efficiently navigate the vast space have rarely been used prediction properties that need calculated by quantum mechanical methods. Here, we introduce Quantum MOF (QMOF) database, publicly available database computed quantum-chemical more than 14,000 experimentally synthesized MOFs. Throughout this study, demonstrate how models trained on QMOF rapidly discover with targeted electronic structure using theoretically band gaps as representative example. We conclude highlighting several predicted low gaps, challenging task electronically insulating most

Language: Английский

---

Materials science in the artificial intelligence age: high-throughput library generation, machine learning, and a pathway from correlations to the underpinning physics

MRS Communications, Journal Year: 2019, Volume and Issue: 9(3), P. 821 - 838

Published: July 22, 2019

Language: Английский

---

SchNetPack 2.0: A neural network toolbox for atomistic machine learning

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 158(14)

Published: March 21, 2023

SchNetPack is a versatile neural network toolbox that addresses both the requirements of method development and application atomistic machine learning. Version 2.0 comes with an improved data pipeline, modules for equivariant networks, PyTorch implementation molecular dynamics. An optional integration Lightning Hydra configuration framework powers flexible command-line interface. This makes easily extendable custom code ready complex training tasks, such as generation 3D structures.

Language: Английский

---

Database of Two-Dimensional Hybrid Perovskite Materials: Open-Access Collection of Crystal Structures, Band Gaps, and Atomic Partial Charges Predicted by Machine Learning

Chemistry of Materials, Journal Year: 2020, Volume and Issue: 32(17), P. 7383 - 7388

Published: Aug. 11, 2020

We describe a first open-access database of experimentally investigated hybrid organic-inorganic materials with two-dimensional (2D) perovskite-like crystal structure. The includes 515 compounds, containing 180 different organic cations, 10 metals (Pb, Sn, Bi, Cd, Cu, Fe, Ge, Mn, Pd, Sb) and 3 halogens (I, Br, Cl) known so far will be regularly updated. contains geometrical chemical analysis the structures, which are useful to reveal quantitative structure-property relationships for this class compounds. show that penetration depth spacer cation into inorganic layer M-X-M bond angles increase in number layers (n). machine learning model is developed trained on database, prediction band gap accuracy within 0.1 eV. Another atomic partial charges 0.01 e. predicted values gaps decrease an n single-layered perovskites. In general, proposed models shown tools rational design new 2D perovskite materials.

Language: Английский

---

Artificial intelligence-enabled smart mechanical metamaterials: advent and future trends

International Materials Reviews, Journal Year: 2020, Volume and Issue: 66(6), P. 365 - 393

Published: Sept. 8, 2020

Mechanical metamaterials have opened an exciting venue for control and manipulation of architected structures in recent years. Research the area mechanical has covered many their fabrication, mechanism characterisation application aspects. More recently, however, a paradigm shift emerged to research direction towards designing, optimising characterising using artificial intelligence (AI) techniques. This new line aims at addressing difficulties (i.e. design, analysis, fabrication industrial application). review article discusses advent development metamaterials, future trends applying AI obtain smart with programmable response. We explain why materials prominent advantages, what are main challenges metamaterial domain, how surpass limit via finally envision potential avenues emerging AI-enabled innovations.

Language: Английский

---

Molecular representations for machine learning applications in chemistry

International Journal of Quantum Chemistry, Journal Year: 2021, Volume and Issue: 122(7)

Published: Dec. 27, 2021

Abstract Machine learning (ML) methods enable computers to address problems by from existing data. Such applications are becoming commonplace in molecular sciences. Interest applying ML techniques across chemical compound space, predicting properties designing molecules and materials is the surge. Especially, models have started accelerate computational chemistry, often as accurate state‐of‐the‐art electronic/atomistic models. Being an integral part of architecture, representation a entity, uniquely encoded, plays crucial role what extent model would be accurately desired property. This review aims demonstrate hierarchy representations which has been introduced, capture all degrees freedom molecule or atom best, map quantum mechanical properties. We discuss their diverse how they instrumental harnessing growing field accelerated modeling.

Language: Английский

---

Computational Approaches for Organic Semiconductors: From Chemical and Physical Understanding to Predicting New Materials

Chemical Reviews, Journal Year: 2023, Volume and Issue: 123(12), P. 7498 - 7547

Published: May 4, 2023

While a complete understanding of organic semiconductor (OSC) design principles remains elusive, computational methods─ranging from techniques based in classical and quantum mechanics to more recent data-enabled models─can complement experimental observations provide deep physicochemical insights into OSC structure-processing-property relationships, offering new capabilities for silico discovery design. In this Review, we trace the evolution these methods their application OSCs, beginning with early quantum-chemical investigate resonance benzene building machine-learning (ML) ever sophisticated scientific engineering challenges. Along way, highlight limitations how physical mathematical frameworks have been created overcome those limitations. We illustrate applications range specific challenges OSCs derived π-conjugated polymers molecules, including predicting charge-carrier transport, modeling chain conformations bulk morphology, estimating thermomechanical properties, describing phonons thermal name few. Through examples, demonstrate advances accelerate deployment OSCsin wide-ranging technologies, such as photovoltaics (OPVs), light-emitting diodes (OLEDs), thermoelectrics, batteries, (bio)sensors. conclude by providing an outlook future development discover assess properties high-performing greater accuracy.

Language: Английский

---

On the Sufficiency of a Single Hidden Layer in Feed-Forward Neural Networks Used for Machine Learning of Materials Properties

Physchem, Journal Year: 2025, Volume and Issue: 5(1), P. 4 - 4

Published: Jan. 16, 2025

Feed-forward neural networks (NNs) are widely used for the machine learning of properties materials and molecules from descriptors their composition structure (materials informatics) as well in other physics chemistry applications. Often, multilayer (so-called “deep”) NNs used. Considering that universal approximator hold single-hidden-layer NNs, we compare here performance (SLNN) with (MLNN), including those previously reported different We consider three representative cases: prediction band gaps two-dimensional materials, reorganization energies oligomers, formation polyaromatic hydrocarbons. In all cases, results good or better than obtained an MLNN could be SLNN, a much smaller number neurons. As SLNNs offer advantages (including ease construction use, more favorable scaling nonlinear parameters, modulation NN model by choice neuron activation function), hope this work will entice researchers to have closer look at when is genuinely needed SLNN sufficient.

Language: Английский

---

Machine Learning-Based Prediction of Crystal Systems and Space Groups from Inorganic Materials Compositions

ACS Omega, Journal Year: 2020, Volume and Issue: 5(7), P. 3596 - 3606

Published: Feb. 13, 2020

Structural information of materials such as the crystal systems and space groups are highly useful for analyzing their physical properties. However, enormous composition makes experimental X-ray diffraction (XRD) or first-principle-based structure determination methods infeasible large-scale material screening in space. Herein, we propose evaluate machine-learning algorithms determining type materials, given only compositions. We couple random forest (RF) multiple layer perceptron (MLP) neural network models with three types features: Magpie, atom vector, one-hot encoding (atom frequency) system group prediction materials. Four predicting proposed, trained, evaluated including one-versus-all binary classifiers, multiclass polymorphism predictors, multilabel classifiers. The synthetic minority over-sampling technique (SMOTE) is conducted to mitigate effects imbalanced data sets. Our results demonstrate that RF Magpie features generally outperforms other groups, while MLP frequency best one structural prediction. For prediction, relevance respectively. analysis related descriptors identifies a few key contributing structural-type electronegativity, covalent radius, Mendeleev number. work thus paves way fast composition-based inorganic via predicted

Language: Английский

---

Artificial intelligence driven in-silico discovery of novel organic lithium-ion battery cathodes

Energy storage materials, Journal Year: 2021, Volume and Issue: 44, P. 313 - 325

Published: Oct. 25, 2021

Organic electrode materials (OEMs) combine key sustainability and versatility properties with the potential to enable realisation of next generation truly green battery technologies. However, for OEMs become a competitive alternative, challenging issues related energy density, rate capability cycling stability need be overcome. In this work, we have developed applied an alternative yet systematic methodology accelerate discovery suitable cathode-active by interplaying artificial intelligence (AI) quantum mechanics. This AI-kernel has allowed high-throughput screening huge library organic molecules, leading 459 novel promising candidates offering achieve theoretical densities superior 1000 W h kg−1. Moreover, machinery accurately identified common molecular functionalities that lead such higher-voltage electrodes pointed out interesting donor-accepter-like effect may drive future design OEMs.

Language: Английский

---