Non-commutative optimization under constraints fails

[araujoms]

The following code gives me a stack overflow error:

using DynamicPolynomials
using SumOfSquares
import SCS

model = Model(SCS.Optimizer)

@ncpolyvar a0 a1 b0 b1
p = a0*b0 + a0*b1 + a1*b0 - a1*b1
@variable(model, λ)
@objective(model, Min, λ)
S = @set a0^2 == 1 && a1^2 == 1 && b0^2 == 1 && b1^2 == 1
con_ref = @constraint(model, λ - p in SOSCone(), domain = S)

optimize!(model)

Here I'm optimizing the CHSH functional, the simplest problem in nonlocality. The documentation is not clear whether this is supposed to work. If I change @ncpolyvar to @polyvar then it works, and the answer is 2*sqrt(2) as expected.

[odow]

A more minimal example is

julia> using DynamicPolynomials

julia> using SumOfSquares

julia> import SCS

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

julia> @ncpolyvar a
(a,)

julia> @constraint(model, a in SOSCone(), domain = @set a^2 == 1)
(1)a is SOS

julia> optimize!(model)
ERROR: StackOverflowError:
Stacktrace:
 [1] mapexponents(f::Function, m1::Monomial{false}, m2::Monomial{false}) (repeats 79984 times)
   @ MultivariatePolynomials ~/.julia/packages/MultivariatePolynomials/sWAOE/src/monomial.jl:128

[blegat]

Thanks for reporting this, the noncommutative part does not have enough tests unfortunately, especially with constraints