atrfu

A-TRFu, an advanced trust-region funnel method for grey-box optimisation with Pyomo. A grey-box problem couples an equation-based Pyomo model with black boxes, such as simulations or legacy code, that can only be evaluated.

Documentation: https://atrfu.readthedocs.io

Install

pip install atrfu
conda install -c conda-forge ipopt     # the subproblem solver (not pip-installable)
atrfu check                            # confirms IPOPT and gjh are found

AMPL's gjh is downloaded automatically on first use, or in advance with atrfu install-gjh.

Example

from pyomo.environ import ConcreteModel, Constraint, ExternalFunction, Objective, Var
from atrfu import TrustRegionSolver

def distance(a, b):                    # a black box: only its values are used
    return a**2 + b**2

m = ConcreteModel()
m.x1 = Var(bounds=(-3, 3), initialize=0.5)
m.x2 = Var(bounds=(-3, 3), initialize=0.5)
m.y = Var(bounds=(0, 18))
m.bb = ExternalFunction(distance)
m.link = Constraint(expr=m.y == m.bb(m.x1, m.x2))
m.disc = Constraint(expr=m.y <= 1)
m.obj = Objective(expr=(m.x1 - 2)**2 + (m.x2 - 1)**2)

result = TrustRegionSolver(verbose=1).solve(m, {"distance": [m.bb]})
print(result["status"], result["final_obj"], m.x1.value, m.x2.value)

See the user guide and the tuning guide.

Citing

Please cite G. Hameed, Rigorous Trust-Region Algorithms for Surrogate-based Grey-box Optimisation, PhD thesis, University of Surrey, 2026, and the software release (CITATION.cff, Zenodo DOI per release).

License

MIT. Files derived from Pyomo keep their BSD-3-Clause notice.