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

Abstract

Choosing optimal representation methods of atomic and electronic structures is essential when machine learning properties of materials. We address the problem of representing quantum states of electrons in a solid for the purpose of machine leaning state-specific electronic properties. Specifically, we construct a fingerprint based on energy decomposed operator matrix elements (ENDOME) and radially decomposed projected density of states (RAD-PDOS), which are both obtainable from a standard density functional theory (DFT) calculation. Using such fingerprints we train a gradient boosting model on a set of 46k G0W0 quasiparticle energies. The resulting model predicts the self-energy correction of states in materials not seen by the model with a mean absolute error of 0.14 eV. By including the material’s calculated dielectric constant in the fingerprint the error can be further reduced by 30%, which we find is due to an enhanced ability to learn the correlation/screening part of the self-energy. Our work paves the way for accurate estimates of quasiparticle band structures at the cost of a standard DFT calculation.

Machine learning approach to genome of two-dimensional materials with flat electronic bands

“…Features like band crossings, which is both common and small, may require developing automatic graphical pattern search tools 31 . Moreover, in addition to using structural fingerprints, such as lattice coordination patterns as described in the current work, to distinguish different flat band materials, electronic fingerprints, e.g., similarity of electronic states 52 , may also be attempted in the future to offer new perspectives towards discovery of materials.…”

“…Machine learning algorithms can take two different routes to identify flat band materials. In the first approach, materials are grouped based on electronic structures in the reciprocal space, followed by classifying their electronic bands based on topology or other features [26][27][28][29] . The main bottleneck for implementing this method is the need of complete description of wavefunctions in a database, which is not yet available in any open materials database because of the enormous data storage requirements.…”

Discovery of Two-Dimensional Multinary Component Photocatalysts Accelerated by Machine Learning

“…First, we have developed two ML models to precisely predict bandgaps and band edge positions for each structure in C2DB. Note that although the GW method may be more accurate than the Heyd−Scuseria− Ernzerhof (HSE) hybrid functional, 43 the available data based on the GW method is still limited. Consequently, it is hard to accurately predict the band edge positions at the GW level.…”

Design of Thermally Activated Delayed Fluorescence Materials with High Intersystem Crossing Efficiencies by Machine Learning-Assisted Virtual Screening

“…34−36 For example, Thygesen et al introduced novel methods for representing electronic states in solids, which could be used as input in ML models to acurately estimate the quasiparticle band structures of 2D materials. 18 Aspuru-Guzik and co-workers found out efficient new donor molecules for organic photovoltaic materials by highthroughput virtual screening in the Harvard Clean Energy Project. 37 between the material/device structural factors and the external quantum efficiency (EQE) and established the ML models for the prediction of EQE of TADF OLEDs.…”

“…The data-driven study combining density functional theory (DFT), machine learning (ML), and available experimental data is becoming a fast and efficient method to predict the properties of materials and accelerate the discovery of promising materials on a large scale by autonomous learning, which could extract the regularities and correlations from existing knowledge. Material property predictions and designs using such a data-driven method have been successfully applied in two-dimensional (2D) materials, − catalysts, photovoltaic materials, − organic field-effect transistor materials, and OLED materials. − For example, Thygesen et al introduced novel methods for representing electronic states in solids, which could be used as input in ML models to acurately estimate the quasiparticle band structures of 2D materials . Aspuru-Guzik and co-workers found out efficient new donor molecules for organic photovoltaic materials by high-throughput virtual screening in the Harvard Clean Energy Project .…”