This Title All WIREs
How to cite this WIREs title:
WIREs Comput Mol Sci
Impact Factor: 16.778

Simulation and design of energy materials accelerated by machine learning

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Abstract In the light of mature mathematical algorithms and material database construction, a basic research framework of machine learning (ML) method integrated with computational chemistry toolkits exhibits great potentials and advantages in the field of material researches. In this review, we introduce a work flow of ML in energy materials and demonstrate its recent applications in accelerating the material exploration, especially significant progresses in designing novel catalysts, organic and inorganic battery materials and metal–organic framework materials. As a rising research direction, we also identify the prospects and challenges of ML. More automated and intelligent workflows will be widely used in energy material design with the development of ML. Our review provides a guideline to study and design energy materials in the framework of ML. This article is categorized under: Structure and Mechanism > Computational Materials Science
Workflow about the application of machine learning in energy materials research
[ Normal View | Magnified View ]
(a) Flowchart of approach to ionic conductivity model building and structure screening. This approach consists of two main efforts: Ionic conductivity model building and structure screening. First screen for the prerequisite requirements of bandgap, electrochemical stability, energy above the convex hull (metastability), and materials cost. Then build a structure‐based ionic conductivity predictor by defining a feature space and learning from experimentally reported lithium conductors in the literature. This model enables screening for materials with high likelihood of superionic character based on atomic structural characteristics. (Reprinted with permission from Reference . Copyright 2017 The Royal Society of Chemistry). (b) Artificial neural network with 10 input variables and two hidden layers. (c) Comparison of the redox potentials predicted from machine learning with those predicted from DFT. (Reprinted with permission from Reference . Copyright 2018 The Royal Society of Chemistry)
[ Normal View | Magnified View ]
(a) Scheme of polymer design by combining the RF screening and manual screening/modification. The picked‐up molecule or polymer in each stage is shown. (Reprinted with permission from Reference . Copyright 2018 American Chemical Society). (b) Theoretically predicted versus experimental PCE for the testing set (30 molecules) (a) and all data points using the leave‐one‐out cross‐validation technique (b) for the GB model. Inset shows probability density of prediction errors. The descriptor importance for the GB (c) and RF (d) model are depicted. Descriptors are in the following order: (1) number of unsaturated atoms in the main conjugation path of donor molecules, , (2) polarizability of donor molecules, (3) the energetic differences of LUMO and LUMO + 1 of donor molecules ΔL, (4) the energetic differences of HOMO and HOMO − 1 of donor molecules, ΔH, (5) IP(ν), vertical ionization potential of donor molecules (6) reorganization energy for holes in donor molecules, λh, (7) hole–electron binding energy in donor molecules, Ebind, (8) the energetic difference of LUMO of donor and LUMO of acceptor, , (9) the energetic difference of HOMO of donor and LUMO of acceptor, , (10) energy of the electronic transition to a singlet excited state with the largest oscillator strength,Eg, (11) change in dipole moment in going from the ground state to the first excited state for donor molecules, Δge, (12) energy of the electronic transition to the lowest‐lying triplet state, ET1 and (13) the energetic difference of LUMO and LUMO + 1 of acceptors . (Reprinted with permission from Reference . Copyright 2018 John Wiley & Sons)
[ Normal View | Magnified View ]
(a) Workflow for the prediction of new ABO3 cubic perovskites. (Reprinted with permission from Reference . Copyright 2018 American Physical Society). (b) General schematic of the artificial neural network (ANN). The ANN comprises an input layer of descriptors (the Pauling electronegativity and ionic radii on each site), followed by a number of hidden layers, and finally an output layer (Ef). The large circle in the center shows how the output of the ith neuron in lth layer,, is related to the received inputs from (l − 1)th layer . and denote the weight and bias between the jth neuron in (l − 1)th layer and ith neuron in lth layer. σ is the activation function (rectified linear unit in this work). (Reprinted with permission from Reference . Copyright 2018 Springer Nature). (c) A comparison between ML‐predicted (GBR) and DFT‐calculated (PBE) results of six selected HOIPs. Excellent agreement (no larger than 0.1 eV) is found between the ML predicted and DFT calculated bandgap values, verifying the great superiority of the current ML technology. (Reprinted with permission from Reference . Copyright 2018 Springer Nature)
[ Normal View | Magnified View ]
(a) Combinatorial challenge of identifying active sites and surfaces for bimetallic catalysts. (A) Four Ni/Ga intermetallics made experimentally and identified as the lower hull by the Materials Project. (B) 40 identified facets/terminations, up to Miller index (3,3,3). Facets often expose two asymmetric terminations so much is considered separately. (C) 583 adsorption configurations identified with unique average coordination of bonding metal atoms. (D) High‐throughput methodology developed to catalog and rapidly evaluate necessary thermodynamic quantities for this combinatorial problem. (Reprinted with permission from Reference . Copyright 2017 American Chemical Society). (b) DFT‐calculated CO adsorption energies on a set of idealized bimetallic surfaces versus prediction from the two‐level interaction model and the machine‐learning model. Insets show schematics of the two‐level interaction diagram and the ANN. Both reported RMSEs of the model prediction are the average over 16 repeated data randomization to avoid sampling bias. (Reprinted with permission from Reference . Copyright 2015 American Chemical Society) True and predicted activation energy using (A) random forest regression and (B) linear regression. Note that the data is split into 80% trained data and 20% test data. (Reprinted with permission from Reference . Copyright 2018 John Wiley & Sons)
[ Normal View | Magnified View ]

Browse by Topic

Structure and Mechanism > Computational Materials Science

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts