Evaluate DFT energies
Compute single-point DFT energies on stored reactant, TS and product geometries.
Stored source calculations, a published DFT Program and a CPU or CUDA worker.
Compute single-point DFT energies across stored reactant, transition-state, and product geometries on CPU (PySCF) or GPU (GPU4PySCF) workers.
Important: This workflow evaluates single-point energies at fixed geometries. It does not re-optimize structures, compute a DFT Hessian, or run a DFT IRC.
1. Check available DFT Programs
List the Programs published in your project to find the DFT Program reference (for example, dft.single-point@1):
2. Dry-run your DFT evaluation
Preview which geometries and settings will be evaluated for a stored calculation before submitting jobs:
Key options:
--calculation-id CALCULATION_UUID— Evaluates the geometries of one stored source calculation. (Always pass--calculation-idwhen targeting a single reaction; omitting it selects all eligible calculations in the project.)--device cpu | cuda— Selects CPUPySCFor CUDA 13GPU4PySCF. CUDA jobs never silently fall back to CPU.--settings dft.yaml(optional) — Custom YAML settings for functional, basis set, and SCF convergence policy.
3. Submit the calculation batch
Remove --dry-run to schedule the single-point jobs:
The command freezes the source geometries and method settings, fans out one job per geometry via a Flow, and prints both the BATCH_UUID and FLOW_UUID.
4. Check status and retry failed points
Monitor the evaluation batch or individual geometry jobs:
If one geometry point fails (for example, due to a worker preemption), retry only the incomplete points while keeping the completed sibling energies:
5. Compare barriers safely
When inspecting published DFT energies in your project:
- Spin multiplicity is explicit:
1= singlet,2= doublet,3= triplet (the calculator converts to PySCF's2S = multiplicity - 1internally). - Match charge and level of theory: Only compare electronic energies and barriers that share the same total charge, functional, basis set, and reference state (such as optimized IRC endpoints).
See Optimize a structure for geometry relaxation, Search and verify a transition state for IRC generation, or Connect a worker to add CPU/GPU capacity.