Support for JuMP's new Nonlinear interface

[ccoffrin]

trafficstars

It would be great if this supported the new NL interface in JuMP. Right now it seems to ignore all constraints provided in that way.

CC @odow

[abelsiqueira]

You're right, I though this was added in #171, except for quadratic constraints (#173), but apparently it is failing for any constraint. This also means that https://github.com/JuliaSmoothOptimizers/Percival.jl/pull/152 is wrong.

cc. @blegat @tmigot

[blegat]

What is failing ? It's working for me: With quadratic

julia> using JuMP, NLPModelsJuMP, Percival

julia> model = Model(NLPModelsJuMP.Optimizer)

A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: NLPModels

julia> set_attribute(model, "solver", Percival.PercivalSolver)

julia> @variable(model, x[i=1:2], start = [-1.2; 1.0][i])
2-element Vector{VariableRef}:
 x[1]
 x[2]

julia> @objective(model, Min, (x[1] - 1)^2 + 100 * (x[2] - x[1]^2)^2)
(x[1]² - 2 x[1] + 1) + (100.0 * ((-x[1]² + x[2]) ^ 2.0))

julia> @constraint(model, x[1]^2 + x[2]^2 == 1)
x[1]² + x[2]² = 1

julia> optimize!(model)
[ Info:   iter        fx    normgp    normcx         μ     normy    sumc     inner_status        iter_type  
[ Info:      0   2.4e+01   2.3e+02   1.4e+00   1.0e+01   0.0e+00       3
[ Info:      1   2.4e+01   2.3e+02   1.4e+00   1.0e+02   0.0e+00       7      first_order         update_μ
[ Info:      2   3.2e+00   2.1e+00   7.9e-04   1.0e+02   7.9e-02      20      first_order         update_y
[ Info:      3   3.2e+00   7.3e-07   9.4e-03   1.0e+02   1.0e+00      30      first_order         update_y
[ Info:      4   3.2e+00   9.8e-07   5.8e-05   1.0e+02   1.0e+00      40      first_order         update_y
[ Info:      5   3.2e+00   4.5e-14   3.6e-07   1.0e+02   1.0e+00      52      first_order         update_y
[ Info:      6   3.2e+00   2.2e-11   2.2e-09   1.0e+02   1.0e+00      61      first_order         update_y

julia> solution_summary(model)
* Solver : NLPModels

* Status
  Result count       : 1
  Termination status : LOCALLY_SOLVED
  Message from the solver:
  "first-order stationary"

* Candidate solution (result #1)
  Primal status      : FEASIBLE_POINT
  Dual status        : NO_SOLUTION
  Objective value    : 3.18638e+00

* Work counters
  Solve time (sec)   : 5.94959e-01


julia> value.(x)
2-element Vector{Float64}:
 -0.7839301868019912
  0.6208489868346048

With NL

julia> model = Model(NLPModelsJuMP.Optimizer)
A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: NLPModels

julia> set_attribute(model, "solver", Percival.PercivalSolver)

julia> @variable(model, x[i=1:2], start = [-1.2; 1.0][i])
2-element Vector{VariableRef}:
 x[1]
 x[2]

julia> @objective(model, Min, sqrt((x[1] - 1)^2 + 100 * (x[2] - x[1]^2)^2))
sqrt((x[1]² - 2 x[1] + 1) + (100.0 * ((-x[1]² + x[2]) ^ 2.0)))

julia> @constraint(model, sqrt(x[1]^2 + x[2]^2) == 1)
sqrt(x[1]² + x[2]²) - 1.0 = 0

julia> optimize!(model)
[ Info:   iter        fx    normgp    normcx         μ     normy    sumc     inner_status        iter_type  
[ Info:      0   4.9e+00   2.4e+01   5.6e-01   1.0e+01   0.0e+00       3
[ Info:      1   4.9e+00   2.4e+01   5.6e-01   1.0e+02   0.0e+00       7      first_order         update_μ
[ Info:      2   1.8e+00   1.5e-01   5.2e-03   1.0e+02   5.2e-01      25      first_order         update_y
[ Info:      3   1.8e+00   7.2e-09   5.2e-04   1.0e+02   5.7e-01      35      first_order         update_y
[ Info:      4   1.8e+00   6.3e-14   1.5e-06   1.0e+02   5.7e-01      47      first_order         update_y
[ Info:      5   1.8e+00   4.1e-11   4.6e-09   1.0e+02   5.7e-01      54      first_order         update_y

julia> solution_summary(model)
* Solver : NLPModels

* Status
  Result count       : 1
  Termination status : LOCALLY_SOLVED
  Message from the solver:
  "first-order stationary"

* Candidate solution (result #1)
  Primal status      : FEASIBLE_POINT
  Dual status        : NO_SOLUTION
  Objective value    : 1.78504e+00

* Work counters
  Solve time (sec)   : 2.96116e-04

julia> value.(x)
2-element Vector{Float64}:
 -0.78393018356842
  0.6208489817564649

The versions are

(Percival) pkg> st
Project Percival v0.7.0
Status `~/.julia/dev/Percival/Project.toml`
  [10dff2fc] JSOSolvers v0.11.0
  [4076af6c] JuMP v1.18.1
  [ba0b0d4f] Krylov v0.9.5
  [5c8ed15e] LinearOperators v2.6.0
  [a4795742] NLPModels v0.20.0
  [792afdf1] NLPModelsJuMP v0.12.5
  [e01155f1] NLPModelsModifiers v0.6.6
  [ff4d7338] SolverCore v0.3.7
  [b5612192] SolverTools v0.8.7
  [37e2e46d] LinearAlgebra
  [56ddb016] Logging

[abelsiqueira]

The conversion to an NLPModel doesn't work.

using JuMP, NLPModelsJuMP, Percival
model = Model()
@variable(model, x[1:2])
@objective(model, Min, (x[1] - 1)^2 + 4 * (x[2] - x[1]^2)^2)
@constraint(model, x[1]^2 + x[2]^2 == 1)
@constraint(model, x[1]^4 + x[2]^4 == 1)
@constraint(model, x[1] * exp(x[2]) == 1)
nlp = MathOptNLPModel(model) # Warns Function MathOptInterface.ScalarQuadraticFunction{Float64} is not supported.
                             # And shows no constraints
output = percival(nlp) # Warns Problem does not have general constraints; calling tron
output.solution # Shows close to [1;1]

[blegat]

This is https://github.com/JuliaSmoothOptimizers/NLPModelsJuMP.jl/issues/173 but aren't constraints other than quadratic working ?

[abelsiqueira]

Apparently not