FBBT emits RuntimeWarning for NumPy scalar bounds

[bernalde]

Summary

pyomo.contrib.fbbt.fbbt() emits a NumPy RuntimeWarning when a variable has NumPy scalar bounds and FBBT reaches interval multiplication patterns that include expected 0 * inf products.

With equivalent native Python float bounds, the same model is quiet.

This appears to be warning hygiene in FBBT interval arithmetic, not an infeasibility in the downstream model: interval.mul() already checks for nan and returns (-inf, inf), but the eager tuple construction evaluates products such as 0 * inf first. If one operand is a NumPy scalar, NumPy emits RuntimeWarning: invalid value encountered in scalar multiply before Pyomo's nan handling runs.

Minimal working example

import warnings

import numpy as np
import pyomo.environ as pyo
from pyomo.contrib.fbbt.fbbt import fbbt


def run_case(bound_type):
    m = pyo.ConcreteModel()
    m.x = pyo.Var(bounds=(bound_type(0), bound_type(1)))
    m.y = pyo.Var(bounds=(bound_type(0), bound_type(1)))
    m.c = pyo.Constraint(expr=m.x * m.y == 0)

    with warnings.catch_warnings(record=True) as records:
        warnings.simplefilter("always", RuntimeWarning)
        fbbt(m.c)

    print(bound_type, len(records), [str(r.message) for r in records])


run_case(float)
run_case(np.float64)

Observed output:

<class 'float'> 0 []
<class 'numpy.float64'> 2 ['invalid value encountered in scalar multiply', 'invalid value encountered in scalar multiply']

Expected behavior

FBBT should not emit a RuntimeWarning for equivalent NumPy scalar bounds. It should either handle the nan case quietly, as it already intends to do, or coerce interval endpoints to native numeric types before multiplication.

Downstream context

This surfaced while working on the GDPLib HDA benchmark refactor:

• GDPLib issue: Refactor hda to fix benchmark issues SECQUOIA/gdplib#67
• GDPLib PR where the warning was observed: Fix hda hull and GLOA benchmark paths SECQUOIA/gdplib#130
• Related GDPLib follow-up considered during this work: Plan PyPI release workflow and Pixi platform support SECQUOIA/gdplib#104

The warning appeared during GDPopt preprocessing/FBBT on HDA flash constraints. A local diagnostic showed that coercing HDA variable bounds and initial values from pandas/NumPy scalars to native Python floats dropped the FBBT runtime warnings from 9 to 0, without changing the algebra.

The promoted-warning traceback reaches:

pyomo/contrib/gdpopt/solve_subproblem.py -> fbbt(...)
pyomo/contrib/fbbt/fbbt.py -> _prop_bnds_root_to_leaf_ProductExpression
pyomo/contrib/fbbt/interval.py -> div(...)
pyomo/contrib/fbbt/interval.py -> mul(...)
RuntimeWarning: invalid value encountered in scalar multiply

Environment

Python 3.12.13
Pyomo 6.10.0
NumPy 2.4.3

I also checked Pyomo/pyomo@main; pyomo/contrib/fbbt/interval.py::mul still appears to use the same eager product tuple:

for i in (xl * yl, xu * yu, xu * yl, xl * yu):
    ...