Problems with COUENNE when constraints are swapped

[svigerske]

Issue created by migration from Trac.

Original creator: linh

Original creation time: 2011-10-27 02:34:39

Assignee: somebody

Version:

CC: [email protected]

I am using AMPL and COUENNE with a very small example

var x1;
var x2;

minimize error: (x1 - 1)^2;

subject to condition1: 0.0002612 + 1.43476/(1 + 100*(x2*0.3)^2) - 0.21*x1 = 0;

subject to condition2: 0.0001974 + 0.81608/(1 + 100*x1^2) - 0.2*x2 = 0;

COUENNE can solve and return the results well but when I swap two conditions (see below), COUENNE returns the message "The LP relaxation is infeasible or too expensive" and then terminates without returning any solution.

var x1;
var x2;

minimize error: (x1 - 1)^2;

subject to condition2: 0.0001974 + 0.81608/(1 + 100*x1^2) - 0.2*x2 = 0;

subject to condition1: 0.0002612 + 1.43476/(1 + 100*(x2*0.3)^2) - 0.21*x1 = 0;

I write their .nl files and realize that AMPL writes the dual initial guess and primal initial guess for the first case into the .nl file

d2
0 0
1 0
x2
0 6.833221401328508
1 0.00186069279294463

while AMPL does not do that for the second case.

I also tried these examples with MINOS and they still can solve both cases.

Could you please explain to me what does the message "The LP relaxation is infeasible or too expensive" mean and why does it appear in the case we swap constraints?

Thank you very much,

Linh