Make eigenvalue PSD/NSD tolerance configurable in hull exact quadratic reformulation

[Copilot]

The numerical tolerance used for eigenvalue-based convexity checks in the exact hull quadratic reformulation (exact_hull_quadratic=True) was hardcoded at 1e-10, making it impossible to tune for ill-conditioned Q matrices without modifying source code.

Changes

• New config parameter eigenvalue_tolerance (default 1e-10, PositiveFloat) added to Hull.CONFIG, following the same pattern as EPS
• Replaces hardcoded value in the PSD/NSD eigenvalue check with self._config.eigenvalue_tolerance
• Updated class docstring to document the new parameter

Usage

from pyomo.gdp import TransformationFactory

hull = TransformationFactory('gdp.hull')
hull.apply_to(
    model,
    exact_hull_quadratic=True,
    eigenvalue_tolerance=1e-8,  # looser tolerance for ill-conditioned Q
)

Larger values make the convexity check more conservative; smaller values more permissive.

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.

[Copilot]

Pull request overview

Adds a configurable numerical tolerance for eigenvalue-based PSD/NSD checks used by the GDP hull transformation’s exact quadratic hull reformulation, replacing a previously hardcoded constant so users can tune behavior for ill-conditioned quadratic forms.

Changes:

• Document eigenvalue_tolerance in Hull_Reformulation docstring.
• Declare eigenvalue_tolerance in Hull_Reformulation.CONFIG (default 1e-10).
• Use self._config.eigenvalue_tolerance in the eigenvalue convexity check instead of a hardcoded 1e-10.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This change introduces a new public config option (eigenvalue_tolerance) that affects whether the conic vs general exact-hull formulation is selected, but there do not appear to be any existing tests covering exact_hull_quadratic (and thus none validating that this tolerance is honored). Please add a focused unit test that sets exact_hull_quadratic=True and varies eigenvalue_tolerance to flip the PSD/NSD classification for a nearly-singular Q, asserting the chosen reformulation differs.