Integrated workflows and interfaces for data-driven semi-empirical electronic structure calculations

Area of Science:

• Computational Chemistry
• Materials Science
• Software Engineering

Background:

• Traditional electronic structure codes often use monolithic workflows.
• Software modularity offers greater flexibility for computational tasks.
• Integrating electronic structure calculations with data-driven analysis presents opportunities.

Purpose of the Study:

• To discuss approaches for creating modular interfaces between electronic structure codes and big-data workflows.
• To explore diverse use cases enabled by these interfaces.
• To present specific interface implementations for the DFTB+ package.

Main Methods:

• Developed two distinct interface approaches for the DFTB+ package.
• One approach uses DFTB+ as a library providing data to external workflows.
• The other approach uses external bindings for DFTB+ to receive and process data internally.
• A general framework for data exchange workflows was established.

Main Results:

• Demonstrated modular interfaces connecting DFTB+ with external workflows.
• Enabled DFTB+ to act as both a data provider and a data consumer.
• Facilitated the embedding of machine-learning-based Hamiltonians within DFTB+.
• Enabled deep integration of DFTB+ into multiscale embedding workflows.

Conclusions:

• Modular interfaces enhance the flexibility and applicability of electronic structure codes.
• These interfaces accelerate scientific discovery by enabling novel software and data workflows.
• The presented framework supports the integration of advanced computational methods like machine learning.