FunctionalAgent automates on-top functional design in MC-PDFT
The introduction of FunctionalAgent marks a significant advancement in fully automated functional development. This agentic system coordinates various specialized sub-agents to break down the development process into several stages: dataset construction, active-space generation, MCSCF calculation, descriptor generation, loss-function creation, and functional fitting, optimization, and evaluation, all interconnected in a closed-loop workflow. Utilizing FunctionalAgent, the MC26 hybrid meta-GGA on-top functional was created, demonstrating enhanced overall accuracy on the training set when compared to alternative methods using the same benchmark dataset. This research is documented on arXiv (2605.06215) and pertains to multiconfiguration pair-density functional theory (MC-PDFT), which provides a reliable framework for calculating electronic energies in strongly correlated molecular systems. The effectiveness of the on-top functional is crucial for the predictive accuracy of MC-PDFT.
Key facts
- FunctionalAgent is an agentic system for automated functional development.
- It uses specialized sub-agents to manage dataset construction, active-space generation, MCSCF calculation, descriptor generation, loss-function construction, and functional fitting, optimization, and evaluation.
- The system creates a closed-loop automated workflow.
- MC26 is a hybrid meta-GGA on-top functional developed using FunctionalAgent.
- MC26 achieves improved overall accuracy on the training set compared to other methods on the same benchmark dataset.
- The work is published on arXiv with identifier 2605.06215.
- MC-PDFT is used for computing electronic energies in strongly correlated molecular systems.
- The quality of the on-top functional is key to MC-PDFT's accuracy.
Entities
Institutions
- arXiv