Overview

mma4py is a parallel gradient-based optimizer that implements the method of moving asymptotes (MMA). The algorithm is implemented in C++, and can be accessed from Python. This project is based on TopOpt_in_PETSc, from which the source code of the implementation of MMA algorithm is obtained.

The original code can be found here: https://github.com/topopt/TopOpt_in_PETSc. The corresponding publication is Aage, N., Andreassen, E., & Lazarov, B. S. (2015). Topology optimization using PETSc: An easy-to-use, fully parallel, open source topology optimization framework. Structural and Multidisciplinary Optimization, 51(3), 565–572. https://doi.org/10.1007/s00158-014-1157-0.

How to use

To use mma4py, simply define a custom class that implements the evaluation of objective, constraints and gradients, and feed the object to the optimizer. An illustrative code snippet is shown below:

from mma4py import Problem, Optimizer

class MyProb(Problem):
    def __init__(self, comm, nvars, nvars_l):
        ...

    def getVarsAndBounds(self, x, lb, ub):
        ...

    def evalObjCon(self, x, cons):
        ...

    def evalObjConGrad(self, x, g, gcon):
        ...

# Create problem instance
prob = MyProb(comm, nvars, nvars_l)

# Create optimization instance
opt = Optimizer(prob)

# Validate input gradients using finite difference
opt.checkGradients()

# Run optimization
opt.optimize(niter=100, verbose=True)

# Get the (distributed) optimized solution
xopt = opt.getOptimizedDesign()

See examples/quadratic/quad-min.py for more details.

Installation

• First, install python dependencies by pip install -r requirements.txt
• Then, install PETSc:
  • Obtain source code from https://gitlab.com/petsc/petsc
  • Configure PETSc by ./configure --with-mpi=1 --with-debugging=0 --prefix=dir/to/install/petsc, then follow the steps in the printout
• Install mma4py:
  • Specify location of PETSc by export MMA4PY_PETSC_PREFIX=dir/to/install/petsc
  • In root directory of mma4py, do pip install .