feasible problem declared infeasible

[mlubin]

The problem is a bit messy, but if I simplify it any further then the issue goes away. The problem is trivially feasible if you set all variables (except aux3) to zero. Used ECOS 1.0.5 here.

using JuMP
import ECOS
m = Model(solver=ECOS.ECOSSolver())

@defVar(m, a1)
@defVar(m, a2)
@defVar(m, c1)
@defVar(m, c2)
@defVar(m, aux1)
@defVar(m, aux2)
@defVar(m, aux3 ≥ 0)

@addConstraint(m, aux3 == 1)
@addConstraint(m, a2 == 0)
@addConstraint(m, c1 == 0)
@addConstraint(m, a1 == 0)
@addConstraint(m, c2 == 0)

@defVar(m, t1 >= 0)

@defVar(m, aux1_l)
@defVar(m, aux1_u)
@defVar(m, term_a1)
@defVar(m, term_a2)
@addConstraint(m, term_a1 == a1)
@addConstraint(m, term_a2 == a2)

@addConstraint(m, aux1_l + aux1 == 0)
@addConstraint(m, aux1_u - aux1 == 0)
@addConstraint(m, aux1_l + 1.6448536269514715*t1 ≤ 0)
@addConstraint(m, aux1_u - 1.6448536269514715*t1 ≥ 0)
@addConstraint(m, aux1_u - aux1_l - 3.919927969080107*t1 ≥ 0)

@defVar(m, t2 >= 0)

@defVar(m, aux2_l)
@defVar(m, aux2_u)
@defVar(m, term_c1)
@defVar(m, term_c2)
@addConstraint(m, term_c1 == c1)
@addConstraint(m, term_c2 == c2)

@addConstraint(m, aux2_l + aux2 == 0)
@addConstraint(m, aux2_u - aux2 == 0)
@addConstraint(m, aux2_l + 1.6448536269514715*t2 ≤ 0)
@addConstraint(m, aux2_u - 1.6448536269514715*t2 ≥ 0)
@addConstraint(m, aux2_u - aux2_l - 3.919927969080107*t2 ≥ 0)

@addConstraint(m, aux1^2 + aux2^2 ≤ aux3^2)

@addConstraint(m, term_a1^2 + term_a2^2 ≤ t1^2)
@addConstraint(m, term_c1^2 + term_c2^2 ≤ t2^2)

solve(m)

CC @adomahidi @echu

[IainNZ]

Quite possibly fixed by now: https://github.com/embotech/ecos/issues/109 was fixed for example. We are quite far behind in terms of bug fixes.

[tkelman]

https://github.com/JuliaOpt/ECOS.jl/issues/16#issuecomment-72537761 and https://github.com/embotech/ecos/issues/112 are not yet fixed afaict (which is what's been holding up an upgrade to something more recent)

[mlubin]

Seems to be fine now

[tkelman]

Worth adding a test?