JuMP interface with register: AssertionError: d.want_hess error

[franckgaga]

Related to #115

The JuMP.jl interface does not work with user-registered nonlinear operators. Here is a simple example running JuMP v1.21.0 and MadNLP v0.8.0:

using JuMP, MadNLP
optim = Model(MadNLP.Optimizer)
nvar = 2
function Jfunc(ΔŨtup...)
    ΔŨ = collect(ΔŨtup)
    return sum(ΔŨ.^2)
end
@variable(optim, ΔŨvar[1:nvar])
register(optim, :Jfunc, nvar, Jfunc, autodiff=true)
@NLobjective(optim, Min, Jfunc(ΔŨvar...))
optimize!(optim)
value.(ΔŨvar)

giving:

ERROR: AssertionError: d.want_hess
Stacktrace:
  [1] hessian_lagrangian_structure
    @ ~/.julia/packages/MathOptInterface/mz9FK/src/Nonlinear/ReverseAD/mathoptinterface_api.jl:296 [inlined]
  [2] hessian_lagrangian_structure(evaluator::MathOptInterface.Nonlinear.Evaluator{…})
    @ MathOptInterface.Nonlinear ~/.julia/packages/MathOptInterface/mz9FK/src/Nonlinear/evaluator.jl:179
  [3] hessian_lagrangian_structure(model::MadNLPMOI.Optimizer)
    @ MadNLPMOI ~/.julia/packages/MadNLP/waaOd/ext/MadNLPMOI/MadNLPMOI.jl:733
  [4] MadNLPMOI.MOIModel(model::MadNLPMOI.Optimizer)
    @ MadNLPMOI ~/.julia/packages/MadNLP/waaOd/ext/MadNLPMOI/MadNLPMOI.jl:839
  [5] optimize!(model::MadNLPMOI.Optimizer)
    @ MadNLPMOI ~/.julia/packages/MadNLP/waaOd/ext/MadNLPMOI/MadNLPMOI.jl:893
  [6] optimize!
    @ ~/.julia/packages/MathOptInterface/mz9FK/src/Bridges/bridge_optimizer.jl:380 [inlined]
  [7] optimize!
    @ ~/.julia/packages/MathOptInterface/mz9FK/src/MathOptInterface.jl:85 [inlined]
  [8] optimize!(m::MathOptInterface.Utilities.CachingOptimizer{…})
    @ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/mz9FK/src/Utilities/cachingoptimizer.jl:316
  [9] optimize!(model::Model; ignore_optimize_hook::Bool, _differentiation_backend::MathOptInterface.Nonlinear.SparseReverseMode, kwargs::@Kwargs{})
    @ JuMP ~/.julia/packages/JuMP/glJ0u/src/optimizer_interface.jl:457
 [10] optimize!(model::Model)
    @ JuMP ~/.julia/packages/JuMP/glJ0u/src/optimizer_interface.jl:409
 [11] top-level scope
    @ ~/.julia/dev/ModelPredictiveControl/src/controller/test_yo.jl:11
Some type information was truncated. Use `show(err)` to see complete types.

[frapac]

Thank you for providing a MWE for your issue. I confirm that MadNLP is not supporting user-defined operator in MadNLP. A solution is to fallback to the LBFGS algorithm implemented inside MadNLP if no Hessian is detected (as it is done in Ipopt). This is addressed in this PR, which should solve your issue:
#322

[frapac]

Solved by #322

[franckgaga]

@frapac Thanks for your work!

The current MadNLP.jl dev version does solve this issue indeed on this MWE. It may or may not be related, but on the specific case of ModelPredictiveControl.jl (that internally uses user-registered nonlinear operator):

using ModelPredictiveControl, JuMP, MadNLP
f(x,u,_) = 0.1x
h(x,_) = x
model = NonLinModel(f, h, 1, 1, 1, 1, solver=nothing)
nmpc = NonLinMPC(model, Hp=10, optim=Model(MadNLP.Optimizer))
nmpc([1])

there is no longer the want_hess error, but a new one:

ERROR: UndefVarError: `∇f` not defined
Stacktrace:
  [1] eval_constraint_jacobian_transpose_product(block::MadNLPMOI.QPBlockData{…}, Jtv::Vector{…}, x::Vector{…}, v::Vector{…})
    @ MadNLPMOI ~/.julia/dev/MadNLP/ext/MadNLPMOI/utils.jl:528
  [2] eval_constraint_jacobian_transpose_product(model::MadNLPMOI.Optimizer, Jtv::Vector{…}, x::Vector{…}, v::Vector{…})
    @ MadNLPMOI ~/.julia/dev/MadNLP/ext/MadNLPMOI/MadNLPMOI.jl:768
  [3] jtprod!
    @ ~/.julia/dev/MadNLP/ext/MadNLPMOI/MadNLPMOI.jl:810 [inlined]
  [4] _eval_jtprod_wrapper!(cb::MadNLP.SparseCallback{…}, x::Vector{…}, v::Vector{…}, jvt::Vector{…})
    @ MadNLP ~/.julia/dev/MadNLP/src/nlpmodels.jl:620
  [5] eval_lag_hess_wrapper!(solver::MadNLPSolver{…}, kkt::MadNLP.SparseKKTSystem{…}, x::MadNLP.PrimalVector{…}, l::Vector{…}; is_resto::Bool)
    @ MadNLP ~/.julia/dev/MadNLP/src/IPM/callbacks.jl:169
  [6] eval_lag_hess_wrapper!
    @ ~/.julia/dev/MadNLP/src/IPM/callbacks.jl:142 [inlined]
  [7] regular!(solver::MadNLPSolver{…})
    @ MadNLP ~/.julia/dev/MadNLP/src/IPM/solver.jl:262
  [8] solve!(nlp::MadNLPMOI.MOIModel{…}, solver::MadNLPSolver{…}, stats::MadNLP.MadNLPExecutionStats{…}; x::Nothing, y::Nothing, zl::Nothing, zu::Nothing, kwargs::@Kwargs{})
    @ MadNLP ~/.julia/dev/MadNLP/src/IPM/solver.jl:165
  [9] solve!
    @ ~/.julia/dev/MadNLP/src/IPM/solver.jl:128 [inlined]
 [10] solve!
    @ ~/.julia/dev/MadNLP/src/IPM/solver.jl:14 [inlined]
 [11] solve!(solver::MadNLPSolver{…})
    @ MadNLP ~/.julia/dev/MadNLP/src/IPM/solver.jl:17
 [12] optimize!(model::MadNLPMOI.Optimizer)
    @ MadNLPMOI ~/.julia/dev/MadNLP/ext/MadNLPMOI/MadNLPMOI.jl:956
 [13] optimize!
    @ ~/.julia/packages/MathOptInterface/2rAFb/src/Bridges/bridge_optimizer.jl:380 [inlined]
 [14] optimize!
    @ ~/.julia/packages/MathOptInterface/2rAFb/src/MathOptInterface.jl:85 [inlined]
 [15] optimize!(m::MathOptInterface.Utilities.CachingOptimizer{…})
    @ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/2rAFb/src/Utilities/cachingoptimizer.jl:316
 [16] optimize!(model::Model; ignore_optimize_hook::Bool, _differentiation_backend::MathOptInterface.Nonlinear.SparseReverseMode, kwargs::@Kwargs{})
    @ JuMP ~/.julia/packages/JuMP/as6Ji/src/optimizer_interface.jl:457
 [17] optimize!
    @ ~/.julia/packages/JuMP/as6Ji/src/optimizer_interface.jl:409 [inlined]
 [18] optim_objective!(mpc::NonLinMPC{Float64, UnscentedKalmanFilter{…}, Model, ModelPredictiveControl.var"#98#100"})
    @ ModelPredictiveControl ~/.julia/dev/ModelPredictiveControl/src/controller/execute.jl:476
 [19] moveinput!(mpc::NonLinMPC{…}, ry::Vector{…}, d::Vector{…}; Dhat::Vector{…}, Rhaty::Vector{…}, Rhatu::Vector{…}, ym::Nothing, D̂::Vector{…}, R̂y::Vector{…}, R̂u::Vector{…})
    @ ModelPredictiveControl ~/.julia/dev/ModelPredictiveControl/src/controller/execute.jl:67
 [20] moveinput!(mpc::NonLinMPC{…}, ry::Vector{…}, d::Vector{…})
    @ ModelPredictiveControl ~/.julia/dev/ModelPredictiveControl/src/controller/execute.jl:52
 [21] #_#123
    @ ~/.julia/dev/ModelPredictiveControl/src/predictive_control.jl:49 [inlined]
 [22] PredictiveController
    @ ~/.julia/dev/ModelPredictiveControl/src/predictive_control.jl:44 [inlined]
 [23] (::NonLinMPC{Float64, UnscentedKalmanFilter{…}, Model, ModelPredictiveControl.var"#98#100"})(ry::Vector{Int64})
    @ ModelPredictiveControl ~/.julia/dev/ModelPredictiveControl/src/predictive_control.jl:44
 [24] top-level scope
    @ ~/Dropbox/Programmation/Julia/TestMPC/src/test_yo.jl:67
Some type information was truncated. Use `show(err)` to see complete types.