Segfault on Mac

[kibaekkim]

When solving the following problem on Mac, PIPS-NLP returns Segfault.

using StructJuMP

m = StructuredModel(num_scenarios=3)
@variable(m, w[i=1:3], start = 0.0)
@NLobjective(m, Min, 1.5 * sum(w[i]^2 for i=1:3))

for j in 1:3
    bl = StructuredModel(parent=m,id=j)
    if j == 1
        @variable(bl, z[i=1:2] >= 0, start = 0.0)
        @NLconstraint(bl, z[1]+z[2]==1)
        @NLobjective(bl, Min,
            - 183*w[1]*z[1] - 67*w[2]*z[1] - 250*w[3]*z[1] + 50239*z[1]^2
            - 67*w[1]*z[2] - 183*w[2]*z[2] - 250*w[3]*z[2] + 87022*z[1]*z[2] + 50239*z[2]^2
            - 56436.5961078333202749*z[1]
            - 55000.9027938333310885*z[2])
    elseif j == 2
        @variable(bl, z[i=1] >= 0, start = 0.0)
        @NLconstraint(bl, z[1]==1)
        @NLobjective(bl, Min,
            - 120*w[1]*z[1] - 80*w[2]*z[1] - 300*w[3]*z[1] + 55400*z[1]^2
            - 39798.0662713333222200*z[1])
    else
        @variable(bl, z[i=1:2] >= 0, start = 0.0)
        @NLconstraint(bl, z[1]+z[2]==1)
        @NLobjective(bl, Min,
            - 100*w[1]*z[1] - 25*w[2]*z[1] - 375*w[3]*z[1] + 75625*z[1]^2
            - 400*w[1]*z[2] - 100*w[2]*z[2] + 42500*z[1]*z[2] + 85000*z[2]^2
            - 17699.9170663333425182*z[1]
            - 8199.6667933333355904*z[2])
    end
end

status = solve(m; solver="PipsNlpSerial")
@show status
@show getObjectiveVal(m)

The PIPS results are

  -----------------------------------------------
  NLP Solver
  Argonne National Laboratory
  Lawrence Livermore National Laboratory
  2010-2018
  -----------------------------------------------
  Variables ================ 8
  Equality Constraints ===== 3
  Inequality Constraints === 0
  Variable Lower Bounds ==== 5
  Variable Upper Bounds ==== 0

  Linear system solver ------    Ma57.

 Iter     Objective     Inf_pr    Inf_du     Comp    lg(mu)   ||d||   lg(PrRg)   alpha_du   alpha_pr    LS   lg(DuRg)   AugSysErr KKTSysErr    dWd0      dWd       thd0       thd      #RegFact   #KryIt
 0  -1.7267473e+03  9.90e-01  5.46e+04  9.00e-02  -1.0
Segmentation fault: 11

But, it works fine on linux.

[cnpetra]

Can you raise PIPS' print level to 3 ('prtLvl 3' in pips.parameter file) and rerun? I don't have a mac so I'll need some assistance from you in debugging this.

[kibaekkim]

This is the output with the print level 3:

load option file: pipsnlp.parameter
OPTION: Set DoIR_Aug to 1
OPTION: Set DoIR_Full to 0
OPTION: Set printing level to 3
OPTION: Set Iteration Limit to 500
OPTION: Set Convergence Tolerance to 1.00e-06
OPTION: Max Line search step:   50
OPTION: Set Sym linear solver as MA57.
OPTION: MA57 pivot level = 1.00e-04
OPTION: MA57 ordering method = 5, (see doc)
OPTION: Do IR on Aug sys.
OPTION: Set MaxIR to 10
OPTION: Set IRtol to 1.00e-12
OPTION: Require small constraint violation in SWC

  -----------------------------------------------
  NLP Solver
  Argonne National Laboratory
  Lawrence Livermore National Laboratory
  2010-2018
  -----------------------------------------------
  Variables ================ 8
  Equality Constraints ===== 3
  Inequality Constraints === 0
  Variable Lower Bounds ==== 5
  Variable Upper Bounds ==== 0

  Linear system solver ------    Ma57.

 Iter     Objective     Inf_pr    Inf_du     Comp    lg(mu)   ||d||   lg(PrRg)   alpha_du   alpha_pr    LS   lg(DuRg)   AugSysErr KKTSysErr    dWd0      dWd       thd0       thd      #RegFact   #KryIt
 0  -1.7267473e+03  9.90e-01  5.46e+04  9.00e-02  -1.0
Segmentation fault: 11