PYSEQM 2.0: Accelerated Semiempirical Excited-State Calculations on Graphical Processing Units

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Photochemistry

Background:

• Semiempirical quantum chemical methods are essential for studying molecular properties.
• Calculating electronic excited states is computationally demanding, especially for large systems.
• Existing methods lack efficiency for large-scale simulations.

Purpose of the Study:

• To develop and implement electronic excited-state capabilities for semiempirical quantum chemical methods.
• To enhance the PYSEQM 2.0 software package for efficient simulations.
• To enable accurate excited-state property calculations for large molecular systems.

Main Methods:

• Configuration Interaction Singles (CIS) and Time-Dependent Hartree-Fock (TDHF) levels of theory were implemented.
• The PYSEQM 2.0 software package, utilizing PyTorch for GPU acceleration and automatic differentiation, was employed.
• Benchmarking was performed on systems up to a thousand atoms.

Main Results:

• Excited-state computations for large systems were achieved in under a minute on modern GPUs.
• PYSEQM 2.0 demonstrated substantial performance gains in molecular property evaluations.
• A machine learning interface for Hamiltonian parameter reoptimization and neural network training was integrated.

Conclusions:

• The developed methods facilitate access to excited-state quantum chemistry for large systems.
• The approach is suitable for high-throughput screening, real-time feedback, and large-scale dynamical studies.
• This work lays the foundation for hybrid quantum-machine-learning approaches in photochemistry, photophysics, and materials discovery.