Benchmark Data Set of Crystalline Organic Semiconductors

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Accurate prediction of crystalline organic semiconductor properties is crucial for materials design.
• Existing computational methods require rigorous benchmarking against experimental data.
• A comprehensive dataset is needed to assess the performance of various theoretical approaches.

Purpose of the Study:

• To establish a benchmark dataset of crystalline organic semiconductors for evaluating computational methods.
• To test the accuracy of density functional theory (DFT) functionals and approximate DFT methods for predicting structural and electronic properties.
• To compare the performance of different computational approaches regarding geometry, unit cell volume, and computational cost.

Main Methods:

• Compilation of a dataset of 67 crystalline organic semiconductors with available experimental crystal structures.
• Benchmarking of r2SCAN-D3 and PBE-D3 density functionals using a subset of 28 crystals with reliable zero-temperature unit cell volume estimations.
• Evaluation of approximate DFT methods (GFN1-xTB, DFTB3) against r2SCAN-D3 across the entire dataset, including analysis of dispersion corrections.

Main Results:

• r2SCAN-D3 demonstrates high accuracy for geometries (within a few percent) but systematically underestimates unit cell volume by approximately 2%.
• PBE-D3, while faster, provides unbiased volume estimates except for highly polar bonds, where it overestimates volume.
• Approximate DFT methods yield qualitatively correct but overcompressed crystal structures without fitted dispersion corrections.

Conclusions:

• The developed benchmark dataset provides a valuable resource for validating computational methods for organic semiconductors.
• r2SCAN-D3 offers accurate geometries, while PBE-D3 presents a faster alternative with specific limitations.
• Approximate DFT methods require careful consideration of dispersion corrections for accurate unit cell predictions.