Fault-tolerant quantum chemical calculations with improved machine-learning models.

Area of Science:

• Computational chemistry
• Quantum chemistry
• Machine learning

Background:

• Efficient use of computational resources is vital for scientific calculations.
• Previous work introduced machine-learning (ML) assisted scheduling optimization.
• Load balancing and fault tolerance are key challenges in distributed computing.

Purpose of the Study:

• To improve ML models for more accurate computational load predictions.
• To integrate coded computation for fault tolerance in distributed calculations.
• To apply these advancements to the re-normalized exciton model with time-dependent density functional theory (REM-TDDFT) for excited-state calculations.

Main Methods:

• Development of enhanced ML models for load prediction.
• Implementation of coded computation (gradient coding) for fault tolerance.
• Integration of ML-assisted coded computation with REM-TDDFT.

Main Results:

• Improved load-balancing and cluster utilization were achieved.
• Significant gains in fault tolerance were demonstrated.
• Successful application to benchmark calculations including P38 protein and solvent models.

Conclusions:

• The combined approach of ML-assisted coded computation offers enhanced performance for distributed quantum chemical calculations.
• This methodology facilitates automated calculations for both ground and excited states.
• The findings pave the way for more robust and efficient scientific computing workflows.