Tips for using DuaLip

This page contains some things you could try if DuaLip does not work in the way you expected.

Issue: The solver converges slowly or does not converge.

Potential things to try:

1. Apply Jacobi Preconditioning: This rescales each constraint constraints by the norm of constraint coefficients. This helps in making the parameters roughly of the same magnitude. (If not, a small set of constraints might dominate the algorithm path.)

2. Try re-running the solver with a larger value of the gamma parameter. This helps in smoothing the dual objective function. (For more details, please see the The DuaLip Solver section.) You may also use adaptive smoothing by starting with a large gamma and gradually decreasing it over time. This helps in finding a good solution faster while using a smaller gamma.