Scaling, initialization and diagnostics
Use this route when two nominally similar solves behave differently, or when constructing a reproducible numerical experiment. The detailed equations and supported scaling domains remain in the OPF reference. This page connects the controls to the evidence you should inspect.
Choose coordinates deliberately
The input network and extracted results use SI units. Solver coordinates are a separate choice. OpfScaling exposes SI, classic per-unit, and custom consistent-base policies; an explicit scaling_policy takes precedence over per_unit and s_base.
| Task | Where to go |
|---|---|
| Learn voltage/current/power bases and objective units | Units, bases and economics. Its feeder examples require the external dataset stated on that page. |
| Choose a typed policy or experiment with zone-specific bases | Units and scaling, including the qualified transformer/converter domains and rejected combinations. |
| Inspect the coordinates at an AC or DC location | opf_coordinate_bases. Use the location's bases rather than assuming a universal power base. |
| Inspect model blocks, scaling and initialization evidence | opf_diagnostic_schema. Check the returned capabilities before relying on semantic-block completeness. |
| Interpret objective weights or compare different norm encodings | Objective weights, norms and pitfalls. |
A scaling change is an experiment in numerical coordinates. Compare the recovered physical quantities and residuals, as well as termination status and solve time. The supported policy does not guarantee better conditioning or an identical local solution from every starting point.
Inspect the start before tuning the solver
The warm-start initialization uses network-wide source, grounding, conductor-map and winding relations to form a zero-current phasor guess. Its transport diagnostics are available through opf_diagnostic_schema(ctx).initialization. They describe the generated start, not feasibility of the loaded network.
Follow that section for the construction and its scope, then Trusting the solver for solution branches and start-dependence. The solver guide connects model coordinates, tolerances and independent validation. Changing backends and changing a formulation are separate experiments.
Keep a numerical experiment interpretable
For repeated coefficient changes, follow Repeated solves before reusing a model. For a reported operating point, capture opf_research_provenance using the experiment-provenance guide. It records scaling and model/parameter fingerprints as well as solver evidence; full-model provenance collection can be expensive, so use study checkpoints.
Finish with independent solution verification. For an additional re-solve comparison, use solution projection. These checks serve different purposes from model-coordinate diagnostics.