Efficiently searching extreme mechanical properties via boundless objective-free exploration and minimal first-principles calculations

Abstract

Despite the machine learning (ML) methods have been largely used recently, the predicted materials properties usually cannot exceed the range of original training data. We deployed a boundless objective-free exploration approach to combine traditional ML and density functional theory (DFT) in searching extreme material properties. This combination not only improves the efficiency for screening large-scale materials with minimal DFT inquiry, but also yields properties beyond original training range. We use Stein novelty to recommend outliers and then verify using DFT. Validated data are then added into the training dataset for next round iteration. We test the loop of training-recommendation-validation in mechanical property space. By screening 85,707 crystal structures, we identify 21 ultrahigh hardness structures and 11 negative Poisson’s ratio structures. The algorithm is very promising for future materials discovery that can push materials properties to the limit with minimal DFT calculations on only ~1% of the structures in the screening pool.

High-throughput computational discovery of 3218 ultralow thermal conductivity and dynamically stable materials by dual machine learning models

“…Similarly, the efficiency of ALIGNN model has also been demonstrated on training and prediction of phonon density of states, 70 electronic density of states, 71 superconducting properties, 72 etc. It should be emphasized that, the performance of all ML models on LTC is slightly lower than that on other properties such as mechanical properties, 25 heat capacity, 28 sound speed, group velocity, etc. Although from domain knowledge the heat capacity is one of the dominant factors in obtaining LTC, it is easier to train because it is harmonic property (only depending on harmonic frequencies and temperature).…”

“…Alternatively, other empirical models to evaluate LTC have been applied, including the Debye-Callaway model, 11,12 Slack model, 13 but these methods are less accurate. 14 The classical molecular dynamics (MD) simulation [15][16][17][18][19] can also be used to study the thermal transport processes and to predict the LTC of materials, however, the accuracy of this approach relies signicantly on that of the underlying interatomic potentials which are challenging to obtain for a large number of materials, 20,21 even with the help of the recently developed machine learning (ML) interatomic potential techniques. 22,23 ML has been successfully applied for solving complex problems and improving decision making at both the academic and industrial levels.…”

“…For instance, multi-delity ML models offer the possibility of bridging the gap between accurate DFTlevel and rough classical-level (such as MD and empirical models) results. 24,25 With given data, ML algorithms can determine the underlying associations, even if the relationship is highly nonlinear. 26 This allows reducing the number of DFT calculations needed to discover new materials, because ML models are based on statistical prediction and hence they are computationally less expensive.…”

Machine Learning Accelerated Discovery of Promising Thermal Energy Storage Materials with High Heat Capacity

“…For solid materials, the specific heat per mole of a substance has an upper limit of about 3 NR (the so-called Dulong–Petit law), where R = 8.31441 J/mol-K being the molar gas constant and N the molar number of atoms in the material. Thus, the molar thermal energy q mol stored in solids can be approximated by In the past decade, material scientists have been using high-throughput screening (HTS) coupled with density functional theory (DFT) for structure–property prediction with high accuracy in search of novel materials. , DFT is highly accurate but less efficient, and hence it is computationally expensive and time-consuming. , Machine learning (ML) methods offer the possibility of reducing the number of DFT calculations needed to discover new materials because ML models are based on statistical predictions rather than physical-based calculations, hence they are computationally less expensive …”

“…In the past decade, material scientists have been using highthroughput screening (HTS) coupled with density functional theory (DFT) for structure−property prediction with high accuracy in search of novel materials. 7,8 DFT is highly accurate but less efficient, and hence it is computationally expensive and time-consuming. 7,9 Machine learning (ML) methods offer the possibility of reducing the number of DFT calculations needed to discover new materials because ML models are based on statistical predictions rather than physical-based calculations, hence they are computationally less expensive.…”