Convergence Analysis and Predictions for Optimizing Reciprocal Grids: A First-Principles and Machine Learning Study | JoVE Visualize

Area of Science:

Computational Materials Science
Solid-State Physics
Quantum Chemistry

Background:

Accurate first-principles density functional theory (DFT) calculations require sufficiently dense reciprocal grids for converged total energy and electronic structure.
High-throughput calculations necessitate efficient determination of appropriate reciprocal grid densities to balance accuracy and computational cost.

Purpose of the Study:

To establish convergence criteria for reciprocal grid densities in DFT calculations of crystalline materials.
To identify physical properties influencing total energy convergence and develop predictive models for optimal reciprocal grid selection.
To facilitate efficient high-throughput materials discovery by estimating necessary reciprocal grid fineness.

Main Methods:

Performed convergence tests of total energy with varying reciprocal grid densities for crystalline materials.
Investigated the impact of band structure nonlinearity and band gaps on energy convergence.
Developed machine learning (ML) models using DFT-derived features and elemental properties to predict reciprocal grid requirements.
Evaluated ML model performance using R-squared values (0.803 for DFT features, 0.880 for elemental features).

Main Results:

The nonlinearity of band structures significantly affects total energy convergence, particularly for materials with finite band gaps.
ML models accurately predict errors in total energy calculations, highlighting the importance of nonlinearity and band gaps.
The ML model utilizing elemental features demonstrates strong predictive power for estimating appropriate reciprocal grid densities.

Conclusions:

Reciprocal grid density selection in DFT is influenced by material-specific electronic properties like band structure nonlinearity and band gaps.
Machine learning offers a powerful approach to quantitatively assess the necessity of finer reciprocal grids.
Elemental-feature-based ML models can reliably estimate optimal reciprocal grid sizes, streamlining high-throughput DFT calculations and accelerating materials discovery.