Skip to content

Solve with JuMP ​

A JuMP problem is a JuMP.Model, already discretized: the scheme, the grid and the order of the variables are chosen when the problem is built. The model reproduces the NLP that OptimalControl builds for the same scheme; see JuMP schemes.

julia
using JuMP, Ipopt, OptimalControlProblems

Build and solve ​

julia
nlp = beam(JuMPBackend(); grid_size=grid_size_data(:beam, JuMPBackend()), scheme=:trapeze)
set_optimizer(nlp, Ipopt.Optimizer)
set_silent(nlp)
optimize!(nlp)
termination_status(nlp), objective_value(nlp)

The keywords ​

KeywordValuesDefault
grid_sizenumber of stepsthe default grid of the problem
scheme:euler, :euler_implicit, :trapeze, :midpoint, as in OptimalControl.Collocation:trapeze
ordering:interleaved (as CTDirect) or :blocks (grouped by kind):interleaved
parametersthe parameters to changethe default parameters
initanother initial guess (see Initial guess)the shared one

Other arguments are passed to JuMP.Model.

Read the solution ​

The JuMP models store the names of the components and the time grid, so the solution reads as an OptimalControl one:

julia
x = state(nlp)      # a function of the time
u = control(nlp)
t = time_grid(nlp)
x(t[end]), u(t[1])
([-7.130607050246146e-39, -1.0], -6.644794477620481)

With OptimalControl loaded, solution builds a full OptimalControl solution from the solved model and the Problem of the same problem, to display or plot it:

julia
using OptimalControl
pb = beam(OptimalControlBackend())
sol = solution(nlp, pb)
objective(sol), iterations(sol)
(8.889282231873594, 22)