Difference Answer in solver CONOPT GAMSversion 23.8.2 and 22.7.2

GAMS
1

User GAMS,

I have GAMS version 22.7.2 and version 23.8.2. I run my code in both
version and I have different answers(objective Value). I am using the
same solver in both cases CONOPT. I was wondering why I got different
answer If I am using the same solver?. I got better values (more
logical) using version 22.7.2. So what is the main difference between
these versions CONOPT version 3.14S and version 3.15D?

I’ll appreciate any feedback.

Omar Alminagorta
Ph.D Candidate
Utah State University

/////////// Version 22.7.2

S O L V E S U M M A R Y

MODEL Nov30 OBJECTIVE Obj
TYPE RMINLP DIRECTION MAXIMIZE
SOLVER CONOPT FROM LINE 1027

**** SOLVER STATUS 1 NORMAL COMPLETION
**** MODEL STATUS 2 LOCALLY OPTIMAL
**** OBJECTIVE VALUE 379902341.3100

RESOURCE USAGE, LIMIT 269.258 1000.000
ITERATION COUNT, LIMIT 5341 10000
EVALUATION ERRORS 0 0


C O N O P T 3 version 3.14S
Copyright (C) ARKI Consulting and Development A/S
Bagsvaerdvej 246 A
DK-2880 Bagsvaerd, Denmark

Using default options.


The model has 8149 variables and 11235 constraints
with 24189 Jacobian elements, 4750 of which are nonlinear.
The Hessian of the Lagrangian has 1500 elements on the diagonal,
1750 elements below the diagonal, and 2400 nonlinear variables.

** Optimal solution. Reduced gradient less than tolerance.

//////////////////////// version 23.8.2

S O L V E S U M M A R Y

MODEL Nov30 OBJECTIVE Obj
TYPE RMINLP DIRECTION MAXIMIZE
SOLVER CONOPT FROM LINE 1027

**** SOLVER STATUS 1 Normal Completion
**** MODEL STATUS 2 Locally Optimal
**** OBJECTIVE VALUE 422962605.3883

RESOURCE USAGE, LIMIT 547.548 1000.000
ITERATION COUNT, LIMIT 5442 10000
EVALUATION ERRORS 0 0
CONOPT 3 Mar 17, 2012 23.8.2 WEX 31442.32372 WEI x86_64/MS
Windows


C O N O P T 3 version 3.15D
Copyright (C) ARKI Consulting and Development A/S
Bagsvaerdvej 246 A
DK-2880 Bagsvaerd, Denmark


The model has 8149 variables and 11235 constraints
with 24189 Jacobian elements, 4750 of which are nonlinear.
The Hessian of the Lagrangian has 1500 elements on the diagonal,
1750 elements below the diagonal, and 2400 nonlinear variables.

\

2

Omar:

There are a number of ‘small’ changes between the two CONOPT versions,
mainly fixing some errors that very rarely happen. But we have also had to
re-scale more often to avoid stopping in non-optimal points. In addition, we
have also changed compilers in between.

These changes should not matter much for small and well-behaved models, but
for large and especially non-convex models they may result in different
turns of the solution path at some stage. After a difference has occurred
the solution path should converge again to the same point if the solution is
unique (at least within reasonable a tolerance). If there are multiple local
optima then there is no guarantee where the solution path will end.

You seem to like the 22.7.2 solution, but the objective for 23.8.2 is
better. If the solution with 23.8 is not ‘logical’ then you have a problem
with the formulation of the model. You need something to get rid of the
‘non-logical’ properties. Of course, you are solving an RMINLP model and
with binary variables at fractional values it can be difficult to say what
is logical.

Regards

Arne

Arne Stolbjerg Drud
ARKI Consulting & Development A/S
Bagsvaerdvej 246A, DK-2880 Bagsvaerd, Denmark
Phone: (+45) 44 49 03 23, Fax: (+45) 44 49 03 33, email: adrud@arki.dk

-----Original Message-----
From: gamsworld@googlegroups.com [mailto:gamsworld@googlegroups.com] On
Behalf Of Omar Alminagorta
Sent: Monday, June 04, 2012 10:06 PM
To: gamsworld
Subject: Difference Answer in solver CONOPT GAMSversion 23.8.2 and 22.7.2

User GAMS,

I have GAMS version 22.7.2 and version 23.8.2. I run my code in both version
and I have different answers(objective Value). I am using the same solver in
both cases CONOPT. I was wondering why I got different answer If I am using
the same solver?. I got better values (more
logical) using version 22.7.2. So what is the main difference between these
versions CONOPT version 3.14S and version 3.15D?

I’ll appreciate any feedback.

Omar Alminagorta
Ph.D Candidate
Utah State University

/////////// Version 22.7.2

S O L V E S U M M A R Y

MODEL Nov30 OBJECTIVE Obj
TYPE RMINLP DIRECTION MAXIMIZE
SOLVER CONOPT FROM LINE 1027

**** SOLVER STATUS 1 NORMAL COMPLETION
**** MODEL STATUS 2 LOCALLY OPTIMAL
**** OBJECTIVE VALUE 379902341.3100

RESOURCE USAGE, LIMIT 269.258 1000.000
ITERATION COUNT, LIMIT 5341 10000
EVALUATION ERRORS 0 0


C O N O P T 3 version 3.14S
Copyright (C) ARKI Consulting and Development A/S
Bagsvaerdvej 246 A
DK-2880 Bagsvaerd, Denmark

Using default options.


The model has 8149 variables and 11235 constraints
with 24189 Jacobian elements, 4750 of which are nonlinear.
The Hessian of the Lagrangian has 1500 elements on the diagonal,
1750 elements below the diagonal, and 2400 nonlinear variables.

** Optimal solution. Reduced gradient less than tolerance.

//////////////////////// version 23.8.2

S O L V E S U M M A R Y

MODEL Nov30 OBJECTIVE Obj
TYPE RMINLP DIRECTION MAXIMIZE
SOLVER CONOPT FROM LINE 1027

**** SOLVER STATUS 1 Normal Completion
**** MODEL STATUS 2 Locally Optimal
**** OBJECTIVE VALUE 422962605.3883

RESOURCE USAGE, LIMIT 547.548 1000.000
ITERATION COUNT, LIMIT 5442 10000
EVALUATION ERRORS 0 0
CONOPT 3 Mar 17, 2012 23.8.2 WEX 31442.32372 WEI x86_64/MS
Windows


C O N O P T 3 version 3.15D
Copyright (C) ARKI Consulting and Development A/S
Bagsvaerdvej 246 A
DK-2880 Bagsvaerd, Denmark


The model has 8149 variables and 11235 constraints
with 24189 Jacobian elements, 4750 of which are nonlinear.
The Hessian of the Lagrangian has 1500 elements on the diagonal,
1750 elements below the diagonal, and 2400 nonlinear variables.

–
“gamsworld” group.
To post to this group, send email to gamsworld@googlegroups.com.
To unsubscribe from this group, send email to
gamsworld+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/gamsworld?hl=en.

\