Repository navigation
Support exact Hessians of Lagrangian NonLinMPC and MovingHorizonEstimator #181
Description
@baggepinnen I think for Hessians, it would make sense to always default to sparse computations (when activated, I would keep the LBFGS approximation as the default)? Should it be pure ForwardDiff, or mixed mode?
Activity
franckgaga commentedon Mar 22, 2025
Also maybe @gdalle has tips on that subject ? Thanks!
gdalle commentedon Mar 28, 2025
If your problems are reaasonably small, pure-ForwardDiff sparse Hessians are not a bad idea.
Otherwise you'd have to combine backends, and that's where it gets tricky:
Pure-Enzyme sparse Hessians would also be an option but at the moment they're not so optimized, and there are still bugs I don't understand with respect to
DI.Cachein second-order Enzymefranckgaga commentedon Apr 5, 2025
Alright let's always sparse forwarddiff by default. The cost might be prohibitive on large systems, but the default setting will still be the LBFGS approximation anyway.
Any thoughts @baggepinnen 💭?
baggepinnen commentedon Apr 5, 2025
sparse ForwardDiff is probably the only "safe" default in the sense that it's most likely to work, all others have quite spotty support in my experience.
franckgaga commentedon Apr 5, 2025
Btw, how's your trip to Jordan ? 🇯🇴
Will visit the Iceland for the first time in July, really looking forward for this trip 🇮🇸!
franckgaga commentedon Apr 17, 2025
Come to think of it, as mentionned by Robbybp here, it would not be efficient at all to compute individual Hessians for each element of the nonlinear constraint equality and inequality vectors. Maybe I should wait that Ippot.jl and JuMP officially support user-defined vector functions, at least for the nonlinear constraints.
A good compromise here would be to add the support for the objective function Hessian with a keyword argument
hessianin the short term (since this part does not require user-defined vector functions). The default would be sparseAutoForwardDiff.I would add a new keyword argument
hessian_lagrangianlater, when these functionalities will be ready and stable.[-]Support exact Hessians of nonlinear objective and constraints in `NonLinMPC` and `MovingHorizonEstimator`[/-][+]Support exact Hessians of nonlinear objective in `NonLinMPC` and `MovingHorizonEstimator`[/+]JinNonLinMPCandMovingHorizonEstimator#193[-]Support exact Hessians of nonlinear objective in `NonLinMPC` and `MovingHorizonEstimator`[/-][+]Support exact Hessians of Lagrangian `NonLinMPC` and `MovingHorizonEstimator`[/+]franckgaga commentedon Nov 4, 2025
Implemented in #275 and #277.