ML-wPBE

ML-wPBE predicts molecule-specific range-separation parameters (ω) using a stacked ensemble machine learning model to accelerate optimally tuned range-separated hybrid (OT-RSH) density functional theory calculations for organic semiconducting molecules.

Key Features:

• Stacked ensemble machine learning: Employs a stacked ensemble ML model to predict optimal range-separation parameter ω for OT-RSH functionals.
• System-dependent descriptors: Leverages system-dependent structural and electronic configurations as inputs to predict ω.
• Accuracy: Achieves a mean absolute error of 0.00504 a_0^{-1} for predicted optimal ω values while retaining comparable predictive power for optical properties relative to traditional OT-RSH functionals.
• Computational cost reduction: Reduces computational cost by approximately 2.66 orders of magnitude compared to nonempirical OT-ωPBE methods.
• Training and validation sets: Trained on a database of 1,970 molecules and validated on an additional set of 1,956 molecules.
• Addresses self-interaction errors: Enables OT-RSH behavior that transitions from short-range (semi)local functionals to long-range Hartree-Fock exchange to mitigate self-interaction errors in DFT for organic semiconductors.

Scientific Applications:

• High-throughput materials discovery: Enables rapid OT-RSH DFT screening of organic semiconducting molecules for materials design.
• Drug discovery: Provides accelerated DFT predictions relevant to molecular property evaluation in drug discovery pipelines.
• Electronics and photonics: Facilitates exploration and optimization of optoelectronic and optical properties of molecular systems for electronic and photonic applications.

Methodology:

Uses a stacked ensemble machine learning model trained on structural and electronic descriptors from 1,970 molecules and validated on 1,956 molecules to predict optimal ω values for use in optimally tuned range-separated hybrid (OT-RSH) density functional calculations.