AIQM3: Targeting Coupled-Cluster Accuracy with Semi-Empirical Speed across Seven Main-Group Elements.

Area of Science:

• Computational Chemistry
• Materials Science
• Drug Discovery

Background:

• The AIQM series offers neural network models for accurate chemical simulations.
• Previous models (AIQM1, AIQM2) were limited to H, C, N, O elements.
• Broader elemental coverage is crucial for atomistic simulations.

Purpose of the Study:

• Introduce AIQM3, an extension of AIQM methods.
• Expand elemental coverage to include S, F, and Cl.
• Achieve coupled-cluster accuracy at semiempirical speeds.

Main Methods:

• Leveraging Δ-learning for accuracy and robustness.
• Developing a neural network model for molecular simulations.
• Extending the AIQM framework to new chemical elements.

Main Results:

• AIQM3 achieves coupled-cluster level accuracy with high efficiency.
• Surpasses Density Functional Theory (DFT) in molecular interactions.
• Demonstrates accuracy in reaction simulations, drug design, and predicting radical/ion energies without specific training.
• Enables low-cost infrared (IR) spectra calculations.

Conclusions:

• AIQM3 significantly advances atomistic simulations by expanding elemental scope and maintaining high accuracy.
• Offers a competitive and efficient alternative to existing methods like DFT.
• Accessible via web services to facilitate broader scientific use.