Macroscopic material properties, such as grain boundary energy, are typically difficult to obtain directly through first-principles calculations. However, some fundamental microscopic properties of materials, like elastic constants and lattice constants, can be readily computed using first-principles methods. If correlations can be found between these fundamental microscopic properties and macroscopic properties, it becomes possible to predict macroscopic properties based on microscopic properties, thereby accelerating the discovery and design of new materials.
Methods
Molecular dynamics simulations are conducted using interatomic potentials (IPs) to generate extensive data for "synthetic materials."
Canonical properties (fundamental microscopic properties computable from first principles) and quantities of interest (macroscopic properties to be predicted) are extracted from this data, and regression models are established.
Density functional theory (DFT) calculations of grain boundary energies are used to validate the accuracy of the regression models.
Fig. 1. The relationship between grain boundary energy and tilt angle for different symmetric tilt boundaries obtained from interatomic potential simulations of aluminum, extracted from the OpenKIM database.
Grain boundary energy can be effectively predicted using regression models.
Certain canonical properties, such as vacancy formation energy, vacancy migration energy, and the elastic constant C44, exhibit strong correlations with grain boundary energy.
The regression model built from interatomic potential data is consistent with DFT data, suggesting that interatomic potentials and DFT belong to the same statistical pool and can be used to predict DFT results.
This study provides a new method for predicting macroscopic properties using microscopic properties.
It enables the training of interatomic potentials to predict more complex material properties.
It contributes to the acceleration of new material discovery and design.
Fig. 3. Regression model prediction versus actual grain boundary energy scaling coefficients. The root mean squared error (RMSE) as a function of species is given in the legend. Error bars represent the estimate for prediction uncertainty (1.96x RMSE) for each species.
Acta Materialia | Grain Boundary Database: 327 GBs 58 Elements Acta Materialia | Predicting Grain Boundary Sliding of Metals Four Open Source Materials Gene Databases PSU Acta Materialia | Ab-initio Calc. of O diffusion Acta Materialia | A New Hall-Petch for Yield Strength of Metals ASME | Rules of High Temperature Reactors
