dopingflow
ML-Driven High-Throughput Doping Workflow for Oxide Materials
The workflow integrates:
Structure generation and symmetry-aware enumeration
ML-based screening and relaxation using configurable backends (M3GNet, UMA, MACE, GRACE)
Sequential gradual doping with stepwise relaxation and energy recomputation
Formation energy calculations using configurable thermodynamic reference schemes
Restricted one-dimensional alloy convex hulls and full multicomponent phase diagrams
Bandgap prediction using ALIGNN
Automated database collection
Fully reproducible, stage-isolated execution
All stages are controlled through a single input.toml file.
The workflow is modular: each stage can be executed independently or combined
into a full pipeline using the run-all command.
User Guide
Workflow Stages
The workflow is organized into modular stages. Each stage can be executed independently and uses its own configuration block.
- 0. Reference Energy Construction
- 1. Structure Generation
- Sequential Doping Workflow
- 2. Symmetry-Reduced Energy Pre-screening (Scanning)
- 3. Candidate Relaxation
- 4. Relaxed-Candidate Filtering
- 5. Bandgap Prediction
- 6. Formation Energy Evaluation
- 7. Database Collection
- Restricted One-Dimensional Alloy Hull
- Phase Diagram and Energy Above Hull
- Surface Generation and Relaxation
Examples
- Explicit Example — Single Target Composition
- Explicit Example — Single Target Composition using Oxide References
- Multiple Oxide References and Relative Energies
- Explicit Example — Multiple Target Compositions
- Enumerate Example — Systematic Dopant Screening
- Smoke Test — Minimal Fast Run
- Sequential Example — Gradual Sb Doping
API Reference
===== END =====