feat: Add fix(), unfix(), and fixed to Variable and Variables by FBumann · Pull Request #625 · PyPSA/linopy

FBumann · 2026-03-18T08:42:23Z

Changes proposed in this Pull Request

Adds fix(), unfix(), and fixed to both Variable and the Variables container for fixing variables to values via equality constraints.

Features

fix(value=None, decimals=8, relax=False) — Fixes a variable by adding a __fix__{name} equality constraint. Uses the current solution if no value is given.
unfix() — Removes the fix constraint and restores integrality if it was relaxed.
fixed — Property indicating whether the variable is currently fixed.
Precision handling — Rounds integers/binaries to 0 decimal places, continuous variables to decimals. Clips to variable bounds to prevent infeasibility from floating-point noise.
Integrality relaxation — fix(relax=True) temporarily converts integer/binary variables to continuous (tracked in Model._relaxed_registry), enabling dual extraction after re-solve. unfix() restores the original type.
Cleanup — Model.remove_variables() automatically cleans up fix constraints and relaxed registry entries.

Use cases

MILP dual extraction: Solve MILP, fix binaries with relax=True, re-solve as LP, read duals.
Fixing decision variables after an initial (aggregated) solve.

Includes

Tests covering rounding, clipping, relaxation, restore, container delegation, and cleanup.
Example added to the manipulating-models notebook.

Checklist

Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in doc.
Unit tests for new features were added (if applicable).
A note for the release notes doc/release_notes.rst of the upcoming release is included.
I consent to the release of this PR's code under the MIT license.

Add methods to fix variables to values via equality constraints, with: - Automatic rounding (0 decimals for int/binary, configurable for continuous) - Clipping to variable bounds to prevent infeasibility - Optional integrality relaxation (relax=True) for MILP dual extraction - Relaxed state tracked in Model._relaxed_registry and restored by unfix() - Cleanup on variable removal Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Serialize the relaxed registry as a JSON string in netCDF attrs so that unfix() can restore integrality after a save/load roundtrip. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

FabianHofmann

convenient feature which does not hurt. this could be extended to fix the variables on the solver side as well. I remember the LP files allow to define fixed variables (without an explicit constraint). but this should wait after #630

FabianHofmann · 2026-03-24T08:46:29Z

linopy/variables.py

+        if value is None:
+            value = self.solution
+
+        value = DataArray(value).broadcast_like(self.labels)


should probably use as_dataarray here

FabianHofmann · 2026-03-24T08:47:01Z

linopy/variables.py

+
+        value = DataArray(value).broadcast_like(self.labels)
+
+        # Round: integers/binaries to 0 decimals, continuous to `decimals`


remove comments like these which explain the code

FabianHofmann · 2026-03-24T08:48:01Z

linopy/variables.py


 FILL_VALUE = {"labels": -1, "lower": np.nan, "upper": np.nan}

+FIX_CONSTRAINT_PREFIX = "__fix__"


move to constants.py

FabianHofmann · 2026-03-24T08:49:03Z

linopy/variables.py

+            value = value.round(decimals)
+
+        # Clip to bounds
+        value = value.clip(min=self.lower, max=self.upper)


I would fail-fast raise here in case values are incompatible with bounds

The clipping is meant to prevent infeasibilities due to floating point precision. Ill check if it can be done differntly while not hiding real infeasible values.

FabianHofmann · 2026-03-24T08:50:20Z

linopy/variables.py

+        constraint_name = f"{FIX_CONSTRAINT_PREFIX}{self.name}"
+
+        # Remove existing fix constraint if present
+        if constraint_name in self.model.constraints:


add an additional argument overwrite that allows to drop existing fix constraints

I did so. I argue that this defaults to True, as the semantics of "fixing" are pretty straight forward: fix this variable to this value. But if you disagree we can default it to false, which might catch unexpected refixing.

FabianHofmann · 2026-03-24T08:52:40Z

linopy/variables.py

+        # Handle integrality relaxation
+        if relax and (self.attrs.get("integer") or self.attrs.get("binary")):
+            original_type = "binary" if self.attrs.get("binary") else "integer"
+            self.model._relaxed_registry[self.name] = original_type


that is interesting, are we relaxing fixed binary and integer variable to save computation in the solving?

This makes the problem pure LP if all binaries are fixed. My intention was to allow computing duals etc, but it might also speed up the solve, although i imagine the MILP solver to be as fast with fixed binaries.
This is the main reason for this PR, as this cannot be implemented outside of linopy easily. The equality constraints themselves could be done by a user with reasonable effort.

Makes perfect sense!

- Use as_dataarray() instead of DataArray() for value conversion - Remove explanatory inline comments - Move FIX_CONSTRAINT_PREFIX to constants.py - Raise ValueError for out-of-bounds fix values instead of clipping - Add overwrite parameter to fix() (default True) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

FabianHofmann

feel free to merge when you want

…nfix

FBumann and others added 4 commits March 18, 2026 08:46

Add example to notebook covering fixing and dual extraction

ef734ac

Add example to notebook covering fixing and dual extraction

464743a

fix: Add type annotations to test_fix.py for mypy

cc0edda

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

FBumann changed the title ~~feat: variable fix unfix relax~~ feat: Add fix(), unfix(), and fixed to Variable and Variables Mar 18, 2026

FBumann and others added 2 commits March 18, 2026 09:53

docs: Add release note for fix/unfix/fixed feature

dd2b797

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

feat: Persist _relaxed_registry through netCDF IO

724e3f1

Serialize the relaxed registry as a JSON string in netCDF attrs so that unfix() can restore integrality after a save/load roundtrip. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

FBumann requested a review from FabianHofmann March 18, 2026 09:17

FBumann marked this pull request as ready for review March 18, 2026 09:17

FBumann added the enhancement New feature or request label Mar 18, 2026

FabianHofmann requested changes Mar 24, 2026

View reviewed changes

FBumann and others added 2 commits March 24, 2026 10:14

test: Parametrize fix tests across scalar and array data types

c13fe69

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

FabianHofmann approved these changes Mar 24, 2026

View reviewed changes

FBumann added 2 commits March 24, 2026 11:43

docs: Remove clipping from changelog for new feature

cb625d6

Merge remote-tracking branch 'origin/master' into feat/variable-fix-u…

66a2a56

…nfix

FBumann merged commit f8dcfea into master Mar 24, 2026
21 checks passed

FBumann deleted the feat/variable-fix-unfix branch March 24, 2026 10:54


		value = DataArray(value).broadcast_like(self.labels)

		# Round: integers/binaries to 0 decimals, continuous to `decimals`


		FILL_VALUE = {"labels": -1, "lower": np.nan, "upper": np.nan}

		FIX_CONSTRAINT_PREFIX = "__fix__"

Conversation

FBumann commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes proposed in this Pull Request

Features

Use cases

Includes

Checklist

Uh oh!

FabianHofmann left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

FBumann Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

FabianHofmann left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

FBumann commented Mar 18, 2026 •

edited

Loading

FBumann Mar 24, 2026 •

edited

Loading