CVXOPT

To use the CVXOPT solver, first install the ‘cvxopt’ package with pip install cvxopt. You can then follow the same process as for other optimizers except when importing the optimizer.

Warning

CVXOPT can only solve convex optimization problems. Therefore, users should ensure their problems are convex before applying the optimizer. modOpt does not perform any checks to determine if the user-defined problem is indeed a convex optimization problem. Users must ensure, at a minimum, that the equality constraints are linear.

Import the optimizer as shown in the following code:

from modopt import CVXOPT

Solver options could be set by just passing them within the solver_options dictionary when instantiating the CVXOPT optimizer object. For example, we can set the maximum number of iterations maxiters and the convergence tolerance abstol for the algorithm as shown below.

optimizer = CVXOPT(prob, solver_options={'maxiters': 100, 'abstol': 1e-9})

A limited number of options are available for the CVXOPT solver, as given in the following table. More details about the nonlinear convex optimization algorithm and the solver options can be found here.

CVXOPT solver options
Option	Type (default value)	Description
`show_progress`	bool (`True`)	Set to `False` to turn off the console output.
`maxiters`	int (`100`)	Maximum number of iterations.
`abstol`	float (`1e-7`)	Absolute accuracy.
`reltol`	float (`1e-6`)	Relative accuracy.
`feastol`	float (`1e-7`)	Tolerance for the feasibility conditions.
`refinement`	int (`1`)	Number of iterative refinement steps to use when solving the KKT equations.

Note

Like any other gradient-based optimizer, CVXOPT requires users to define the objective gradient and constraint Jacobian for their problem. It additionally requires the Lagrangian or objective Hessian depending on whether the problem has constraints or not. However, for small problems, users can leverage the finite difference approximations for these derivatives implemented in the Problem/ProblemLite classes in modOpt.