search

ghorn / dynobud

your dynamic optimization buddy
GNU Lesser General Public License v3.0
27 stars 5 forks source link

snopt doesn't like ipopt solutions #11

Closed ghorn closed 10 years ago

ghorn commented 10 years ago

On a test problem, I let ipopt run to convergence, then pass the solution and lagrange multipliers to snopt. Snopt then takes a few major iterations, while @slavakung says that it should accept the solution immediately.

I should try decreasing ipopt's tolerances to make sure this is a real problem

One possible problem is ipopt/snopt have different sign conventions for multipliers. Or maybe one is scaled and the other unscaled.

ghorn commented 10 years ago

also, see https://github.com/casadi/casadi/issues/667

ghorn commented 10 years ago

When IPOPT solves to tighter tolerances, SNOPT will not take any major iterations. It still takes minor iterations, probably because IPOPT doesn't report lagrange multipliers of equality constrained variables. But a casadi segfault is preventing me exploring that for now...