mmapy source package

mmapy is a small, stateful implementation of the Method of Moving Asymptotes (MMA). It solves a separable convex approximation at each call and retains the design and asymptote history for the next outer iteration.

The package is independent of finite-element and topology-optimization libraries. Supply objective and inequality-constraint values and gradients at the current point. Constraints use the convention g(x) <= 0.

Install

pip install mmahelpy

Example

import numpy as np
from mmapy import MMA, MMAOptions

optimizer = MMA(MMAOptions(move_limit=0.2))
x = np.array([0.8, 0.8])
lower = np.zeros(2)
upper = np.ones(2)

for _ in range(30):
    # Minimize x[0]**2 + x[1]**2 subject to x[0] + x[1] >= 0.9.
    objective = float(x @ x)
    objective_gradient = 2.0 * x
    constraints = np.array([0.9 - x.sum()])
    constraint_jacobian = np.array([[-1.0, -1.0]])
    result = optimizer.step(
        x,
        objective,
        objective_gradient,
        constraints,
        constraint_jacobian,
        lower,
        upper,
    )
    x = result.x

print(x)

Create one MMA object per optimization run. Its step method returns an MMAResult with the next design, dual multipliers, approximate constraint values, and dual-solver status. Call reset() to reuse the object for a new problem. Multiple inequality constraints are supported.

Development

PYTHONPATH=src pytest

The distribution is named mmahelpy because the mmapy name is already used on PyPI. The installed import remains mmapy. Scikit-Topt depends on mmahelpy and adapts its public MMA_Config to the generic optimizer for topology optimization. The solver focuses on continuous, differentiable problems with box bounds and inequality constraints.