Originally posted by: yanzhi

I use two methods to solve a simple MILP problem, but the results of x and cpxsol.x are different, I am wondering what's the reason. can anyone give me a clue? thanks a lot.

x is obtained from method 1:

Aeq=[];beq=[];

options = cplexoptimset('cplex');cplexoptimset('Display','on');

[x,fval,exitflag,output] = cplexmilp(f,Aineq,bineq,[],[],[],[],[],lb,ub,ctype,[],options );

cpxsol.mat is obtained from method 2:

 cplex=Cplex('');
    cplex.Model.sense = 'minimize';
    cplex.addCols(f,[],lb,ub,ctype);
    cplex.addRows([-inf*ones(size(bineq,1),1);beq], [Aineq;Aeq], [bineq;beq]);

cplex.solve();
cpxsol=cplex.Solution;

Originally posted by: John Cui

You can export both model to LP files, then compare with these 2 LP files to find out the difference.

The first way, you need to set a parameter(exportmodel) in cplexoptimset function.

{code}

options = cplexoptimset('cplex'); options.exportmodel = 'debug1.lp'

{code}

The second way, you just use cplex.exportModel("debug2.lp");

John Cui

Originally posted by: yanzhi

thank you for your reply.

actually, I did it and compared them.

the myModel.lp is generated through the first way, and the enr.lp is from the 2nd way.

but I didn't find any difference between them.

however, in the solution , there are differences , for example, x(22) compared with cpxsol.x(22) or x(24) with cpxsol.x(24).

and I check the constraints, and found that the solution from the first way did meet the requirements of the problem, but the 2nd way didn't.

so I'm still confused.

I cannot reproduce this problem here. What version of CPLEX and matlab do you use? Which operating system?

EDIT: Could you please also provide the log output for the two solves? To get the log output for the cplexmilp run replace this line
options = cplexoptimset('cplex');cplexoptimset('Display','on');
by
options = cplexoptimset('Diagnostics', 'on');

Originally posted by: yanzhi

the cplex version is 12.4, matlab is 7.11.0(R2010b).

operating system is windows 7.

the log file for  cplexmilp method is

{Tried aggregator 1 time.
MIP Presolve eliminated 123 rows and 111 columns.
MIP Presolve modified 6 coefficients.
Reduced MIP has 60 rows, 63 columns, and 181 nonzeros.
Reduced MIP has 0 binaries, 0 generals, 0 SOSs, and 0 indicators.
Probing time =    0.00 sec.
Tried aggregator 1 time.
Reduced MIP has 60 rows, 63 columns, and 181 nonzeros.
Reduced MIP has 0 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time =    0.09 sec.
Probing time =    0.00 sec.
MIP emphasis: balance optimality and feasibility.
MIP search method: dynamic search.
Parallel mode: deterministic, using up to 2 threads.
Root relaxation solution time =    0.08 sec.

        Nodes                                         Cuts/
   Node  Left     Objective  IInf  Best Integer    Best Bound    ItCnt     Gap

*     0     0      integral     0     5321.4324     5321.4324       31    0.00%
Elapsed real time =   0.20 sec. (tree size =  0.00 MB, solutions = 1)

Root node processing (before b&c):
  Real time             =    0.11
Parallel b&c, 2 threads:
  Real time             =    0.00
  Sync time (average)   =    0.00
  Wait time (average)   =    0.00
                          -------
Total (root+branch&cut) =    0.11 sec.}

the log file for cplex.solve(); method is

{Default variable names x1, x2 ... being created.
Default row names c1, c2 ... being created.
Tried aggregator 1 time.
MIP Presolve eliminated 123 rows and 111 columns.
MIP Presolve modified 6 coefficients.
Reduced MIP has 60 rows, 63 columns, and 181 nonzeros.
Reduced MIP has 0 binaries, 0 generals, 0 SOSs, and 0 indicators.
Probing time =    0.00 sec.
Tried aggregator 1 time.
Reduced MIP has 60 rows, 63 columns, and 181 nonzeros.
Reduced MIP has 0 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time =    0.00 sec.
Probing time =    0.00 sec.
MIP emphasis: balance optimality and feasibility.
MIP search method: dynamic search.
Parallel mode: deterministic, using up to 2 threads.
Root relaxation solution time =    0.00 sec.

        Nodes                                         Cuts/
   Node  Left     Objective  IInf  Best Integer    Best Bound    ItCnt     Gap

*     0     0      integral     0     5321.4324     5321.4324       31    0.00%
Elapsed real time =   0.00 sec. (tree size =  0.00 MB, solutions = 1)

Root node processing (before b&c):
  Real time             =    0.00
Parallel b&c, 2 threads:
  Real time             =    0.00
  Sync time (average)   =    0.00
  Wait time (average)   =    0.00
                          -------
Total (root+branch&cut) =    0.00 sec.}

Sorry, I could not reproduce the problem here. Neither with 12.4 nor with 12.5.0.1 (the latest version). I also looked at the code directly but could not find anything that would explain this weird behavior.

The matlab script I used was this:

function problem(varargin)
try
   load('Aineq.mat');
   load('bineq.mat');
   load('ctype.mat');
   load('f.mat');
   load('lb.mat');
   load('ub.mat');
   Aeq=[];beq=[];

   fprintf('------------------------------ Solve 1 --------------------------\n');
   options = cplexoptimset('Diagnostics', 'on');
   [x,fval,exitflag,output] = cplexmilp(f,Aineq,bineq,[],[],[],[],[],lb,ub,ctype,[],options );

   fprintf('------------------------------ Solve 2 --------------------------\n');
   cplex=Cplex('');
   cplex.Model.sense = 'minimize';
   cplex.addCols(f,[],lb,ub,ctype);
   cplex.addRows([-inf*ones(size(bineq,1),1);beq], [Aineq;Aeq], [bineq;beq]);
   cplex.solve();
   cpxsol=cplex.Solution;

   x - cpxsol.x
catch m
   display m;
   throw (m);
   clear all;
   clear classes;
end

The vector that is printed by that in the end is all-zero.
Can you try to run this script at your place? Does it produce a non-zero vector in the end?

Could you try to upgrade to CPLEX 12.5.0.1 and see if the problem still exists with that new version?