JuMP SCIP error

Specific Domains Optimization (Mathematical)
1

I get the error below when I try to solve a model in JuMP using SCIP. There is no error if I instead use HiGHS.

$ julia Oscar.jl
ERROR: LoadError: MathOptInterface.DeleteNotAllowed{MathOptInterface.VariableIndex}: Deleting the index MathOptInterface.VariableIndex(368) cannot be performed: Can not delete variable while model contains constraints! You may want to use a `CachingOptimizer` in `AUTOMATIC` mode or you may need to call `reset_optimizer` before doing this operation if the `CachingOptimizer` is in `MANUAL` mode.
Stacktrace:
  [1] delete(o::SCIP.Optimizer, vi::MathOptInterface.VariableIndex)
    @ SCIP ~/.julia/packages/SCIP/BUSgq/src/MOI_wrapper/variable.jl:45
  [2] delete(b::MathOptInterface.Bridges.LazyBridgeOptimizer{SCIP.Optimizer}, vi::MathOptInterface.VariableIndex)
    @ MathOptInterface.Bridges ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/bridge_optimizer.jl:601
  [3] delete(model::MathOptInterface.Bridges.LazyBridgeOptimizer{SCIP.Optimizer}, bridge::MathOptInterface.Bridges.Constraint.CountBelongsToMILPBridge{Float64, MathOptInterface.VectorOfVariables})
    @ MathOptInterface.Bridges.Constraint ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/Constraint/bridges/count_belongs.jl:139
  [4] final_touch(bridge::MathOptInterface.Bridges.Constraint.CountBelongsToMILPBridge{Float64, MathOptInterface.VectorOfVariables}, model::MathOptInterface.Bridges.LazyBridgeOptimizer{SCIP.Optimizer})
    @ MathOptInterface.Bridges.Constraint ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/Constraint/bridges/count_belongs.jl:280
  [5] _final_touch(bridges::OrderedCollections.OrderedSet{MathOptInterface.Bridges.Constraint.CountBelongsToMILPBridge{Float64, MathOptInterface.VectorOfVariables}}, model::MathOptInterface.Bridges.LazyBridgeOptimizer{SCIP.Optimizer})
    @ MathOptInterface.Bridges.Constraint ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/Constraint/map.jl:324
  [6] final_touch(map::MathOptInterface.Bridges.Constraint.Map, model::MathOptInterface.Bridges.LazyBridgeOptimizer{SCIP.Optimizer})
    @ MathOptInterface.Bridges.Constraint ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/Constraint/map.jl:331
  [7] final_touch
    @ ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/bridge_optimizer.jl:459 [inlined]
  [8] optimize!
    @ ~/.julia/packages/MathOptInterface/57fsF/src/Bridges/bridge_optimizer.jl:375 [inlined]
  [9] optimize!
    @ ~/.julia/packages/MathOptInterface/57fsF/src/MathOptInterface.jl:83 [inlined]
 [10] optimize!(m::MathOptInterface.Utilities.CachingOptimizer{MathOptInterface.Bridges.LazyBridgeOptimizer{SCIP.Optimizer}, MathOptInterface.Utilities.UniversalFallback{MathOptInterface.Utilities.Model{Float64}}})
    @ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/57fsF/src/Utilities/cachingoptimizer.jl:316
 [11] optimize!(model::Model; ignore_optimize_hook::Bool, _differentiation_backend::MathOptInterface.Nonlinear.SparseReverseMode, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ JuMP ~/.julia/packages/JuMP/vuP7I/src/optimizer_interface.jl:185
 [12] optimize!
    @ ~/.julia/packages/JuMP/vuP7I/src/optimizer_interface.jl:163 [inlined]
 [13] top-level scope
    @ ./timing.jl:220
2

Do you have a reproducible example? I cannot reproduce:

julia> using JuMP

julia> import SCIP

julia> model = Model(SCIP.Optimizer);

julia> @variable(model, 0 <= x[i = 1:4] <= i, Int)
4-element Vector{VariableRef}:
 x[1]
 x[2]
 x[3]
 x[4]

julia> @variable(model, n, Int)
n

julia> @objective(model, Max, sum(x))
x[1] + x[2] + x[3] + x[4]

julia> @constraint(model, [n; x] in MOI.CountBelongs(1 + length(x), Set([2, 3])))
[n, x[1], x[2], x[3], x[4]] ∈ MathOptInterface.CountBelongs(5, Set([2, 3]))

julia> optimize!(model)
presolving:
(round 1, fast)       6 del vars, 3 del conss, 0 add conss, 2 chg bounds, 0 chg sides, 0 chg coeffs, 0 upgd conss, 3 impls, 8 clqs
(round 2, fast)       7 del vars, 3 del conss, 0 add conss, 2 chg bounds, 4 chg sides, 1 chg coeffs, 0 upgd conss, 3 impls, 8 clqs
   (0.0s) running MILP presolver
   (0.0s) MILP presolver (2 rounds): 1 aggregations, 2 fixings, 0 bound changes
(round 3, medium)     10 del vars, 9 del conss, 4 add conss, 2 chg bounds, 4 chg sides, 1 chg coeffs, 0 upgd conss, 3 impls, 6 clqs
(round 4, exhaustive) 10 del vars, 9 del conss, 4 add conss, 2 chg bounds, 4 chg sides, 1 chg coeffs, 4 upgd conss, 3 impls, 6 clqs
(round 5, fast)       10 del vars, 9 del conss, 4 add conss, 2 chg bounds, 4 chg sides, 3 chg coeffs, 4 upgd conss, 3 impls, 6 clqs
(round 6, medium)     12 del vars, 11 del conss, 6 add conss, 2 chg bounds, 7 chg sides, 9 chg coeffs, 4 upgd conss, 3 impls, 6 clqs
(round 7, fast)       12 del vars, 13 del conss, 6 add conss, 2 chg bounds, 7 chg sides, 9 chg coeffs, 4 upgd conss, 3 impls, 6 clqs
presolving (8 rounds: 8 fast, 4 medium, 2 exhaustive):
 19 deleted vars, 15 deleted constraints, 6 added constraints, 2 tightened bounds, 0 added holes, 7 changed sides, 9 changed coefficients
 3 implications, 0 cliques
transformed 1/1 original solutions to the transformed problem space
Presolving Time: 0.00

SCIP Status        : problem is solved [optimal solution found]
Solving Time (sec) : 0.00
Solving Nodes      : 0
Primal Bound       : +1.00000000000000e+01 (1 solutions)
Dual Bound         : +1.00000000000000e+01
Gap                : 0.00 %
3

Reproducible example is:

using JuMP
import SCIP
model = Model(SCIP.Optimizer)
@variable(model, 0 <= s <= 1, Bin)
@variable(model, n)
@constraint(model, [n, s] in MOI.CountBelongs(2, Set([1])))
@constraint(model, [n, s] in MOI.CountBelongs(2, Set([1])))
optimize!(model)

I’ll take a look.

4

This seems like a bug somewhere in MOI: Bridges and final_touch · Issue #2088 · jump-dev/MathOptInterface.jl · GitHub

A work-around for the meantime is to use:

model = Model() do
    return MOI.Utilities.CachingOptimizer(
        MOI.Utilities.UniversalFallback(MOI.Utilities.Model{Float64}()),
        SCIP.Optimizer(),    
    )
end
5

In your example, instead of repeating the same constraint, did you instead mean the following?
@constraint(model, [n, s] in MOI.CountBelongs(2, Set([0])))
@constraint(model, [n, s] in MOI.CountBelongs(2, Set([1])))

6

Is there a work-around if we set SCIP as the JuMP solver as follows?

using JuMP, SCIP, HiGHS
use_SCIP = true
model = Model()
if(use_SCIP)
opt = SCIP.Optimizer
else
opt = HiGHS.Optimizer
end
set_optimizer(model, opt)

7

In your example, instead of repeating the same constraint, did you instead mean the following?

No. I meant what I typed, but it’s somewhat arbitrary. It just needs to be another constraint of any type.

Is there a work-around if we set SCIP as the JuMP solver as follows?

opt = () -> MOI.Utilities.CachingOptimizer(
      MOI.Utilities.UniversalFallback(MOI.Utilities.Model{Float64}()),
      SCIP.Optimizer(),    
  )
set_optimizer(model, opt)
8

Was this SCIP error fixed? If so, what packages must be updated to capture the fix.

9

This is mostly fixed in the latest MathOptInterface v1.12.0.

But you can still trigger it in some situations, for example, if you try to delete a variable.