Skip to content

Possible numerical instability in problem solving #200

@yuxies

Description

@yuxies

SYMPHONY stable/5.7 report infeasibility for the attached problem, while CPLEX and Cbc report an optimal solution.

Starting Preprocessing...
Preprocessing finished...
coefficients modified: 3603
constraints removed: 9337
variables fixed: 22360
Solving...
Time Done Queued LB UB Gap
27 1 0 20699.00
Problem Infeasible
The problem is infeasible!
No Solution Found

Comments in Slack for reference: "The incorrect result for vpphard comes from the root LP relaxation being declared infeasible. If you solve with Clp, the last couple of lines of the output are

3332 Obj 20699 Primal inf 0.015085897 (275) Dual inf 1273.9215 (874)
3333 Obj 20699 Primal inf 0.015085897 (275) Dual inf 326.00733 (482)
3333 Obj 20699 Primal inf 0.015085897 (275)
Primal infeasible - objective value 20699
Presolved problem not optimal, resolve after postsolve
After Postsolve, objective 20699, infeasibilities - dual 1.9999998 (2), primal 0.03370455 (1013)
Perturbing problem by 1e-08% of 1 - largest nonzero change 8.1782865e-06 ( 0.00081782865%) - largest zero change 8.1781075e-06
0 Obj 20698.922 Dual inf 2.9999903 (3)
0 Obj 20699 Dual inf 2.9999997 (3)
6 Obj 20699
Optimal - objective value 20699
Optimal objective 20699 - 3339 iterations time 34.912, Presolve 0.59

"
vpphard_100_0-ub.txt

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions