Choosing the Right Solver

GAMS
1

Dear Friends,

I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.

My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).

Which solver I must choose?


z2 T1 T2 v g Solver Status

LINDOGLOBAL 855434.1617686 0.054794521 0.01 15.14004249 4 optimal

BONMIN 855434.1617756 0.05479452 0.01 15.14004623 4 integer solution

OQNLP 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

SBB 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

KNITRO 855434.1617677 0.054794521 0.01 15.14004631 4 integer solution

DICOPT 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution


Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

2

Hi Zohreh

How precise do you want your solution to be…? This is almost the default precision level of most solvers. This is certainly not a criteria for the choice of solver for this model

Cheers

Renger



From: gamsworld@googlegroups.com [mailto:gamsworld@googlegroups.com] On Behalf Of Zohreh
Sent: Dienstag, 4. März 2014 04:59
To: gamsworld@googlegroups.com
Subject: Choosing the Right Solver



Dear Friends,



I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.



My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).



Which solver I must choose?






z2


T1


T2


v


g


Solver Status


LINDOGLOBAL


855434.1617686


0.054794521


0.01


15.14004249


4


optimal


BONMIN


855434.1617756


0.05479452


0.01


15.14004623


4


integer solution


OQNLP


855434.1617686


0.054794521


0.01


15.14004623


4


integer solution


SBB


855434.1617686


0.054794521


0.01


15.14004623


4


integer solution


KNITRO


855434.1617677


0.054794521


0.01


15.14004631


4


integer solution


DICOPT


855434.1617686


0.054794521


0.01


15.14004623


4


integer solution





Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

3

Dear Renger,

Thank you very much for your kind response.

I am actually looking for the global optimum. Can I choose BONMIN as the solver and mention that it is the global optimum (as it has almost the same answer as LINDOGLOBAL which has found the global optimal)?



On Tuesday, March 4, 2014 11:59:10 AM UTC+8, Zohreh wrote:

Dear Friends,

I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.

My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).

Which solver I must choose?


z2 T1 T2 v g Solver Status

LINDOGLOBAL 855434.1617686 0.054794521 0.01 15.14004249 4 optimal

BONMIN 855434.1617756 0.05479452 0.01 15.14004623 4 integer solution

OQNLP 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

SBB 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

KNITRO 855434.1617677 0.054794521 0.01 15.14004631 4 integer solution

DICOPT 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution


Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

4

Zohreh:

Reformulating Renger’s words:

You are looking at the results at a level of numerical precision which is near the level of precision the solvers handle internally. Most probably, all solvers are converging to the same point.

¿Would it even be realistic for your problem to have a local optimum and a global optimum so close in the variable space (1E-9 in T1 and 1E-6 in v)?

¿Would you even be able to tell apart the decisions influenced by the results of your model at that level of precision (you must have very good controlling and measurement equipment :stuck_out_tongue:)?

Again: all solvers are giving you the same solution.

Hope to help you through !
Regards!
Claudio




On Tue, Mar 4, 2014 at 11:38 AM, Zohreh wrote:

Dear Renger,

Thank you very much for your kind response.

I am actually looking for the global optimum. Can I choose BONMIN as the solver and mention that it is the global optimum (as it has almost the same answer as LINDOGLOBAL which has found the global optimal)?



On Tuesday, March 4, 2014 11:59:10 AM UTC+8, Zohreh wrote:

Dear Friends,

I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.

My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).

Which solver I must choose?


z2 T1 T2 v g Solver Status

LINDOGLOBAL 855434.1617686 0.054794521 0.01 15.14004249 4 optimal

BONMIN 855434.1617756 0.05479452 0.01 15.14004623 4 integer solution

OQNLP 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

SBB 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

KNITRO 855434.1617677 0.054794521 0.01 15.14004631 4 integer solution

DICOPT 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution


Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

5

Dear Claudio,

Thank you very much for your kind response.

Best regards.

On Tuesday, March 4, 2014 11:59:10 AM UTC+8, Zohreh wrote:

Dear Friends,

I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.

My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).

Which solver I must choose?


z2 T1 T2 v g Solver Status

LINDOGLOBAL 855434.1617686 0.054794521 0.01 15.14004249 4 optimal

BONMIN 855434.1617756 0.05479452 0.01 15.14004623 4 integer solution

OQNLP 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

SBB 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

KNITRO 855434.1617677 0.054794521 0.01 15.14004631 4 integer solution

DICOPT 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution


Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

6

What if all the solvers are not giving the same solution?

Best regards,

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

7

Sam: They appear to be, considering the quantitative relation between the variables and objective function different values and solvers internal precision.

What would be evidence for the opposite ?

Regards,
Claudio


On Tue, Mar 11, 2014 at 5:15 PM, Sam wrote:

What if all the solvers are not giving the same solution?

Best regards,

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

\

To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

8

Hi, Claudio: Its not about the precision of the solver, but something else. For the same problem the solver giving one result of the objective function, yet changing one variable to a fixed value within the bound, the same solver is giving an improved solution. What do you think, the problem might be in this case?

I have attached the sample code, and I am doing a comparison between SBB and BARON.

Regards,

Subir

On Wednesday, March 12, 2014 6:40:23 PM UTC+5:30, Claudio Delpino wrote:

Sam: They appear to be, considering the quantitative relation between the variables and objective function different values and solvers internal precision.

What would be evidence for the opposite ?

Regards,
Claudio


On Tue, Mar 11, 2014 at 5:15 PM, Sam wrote:

What if all the solvers are not giving the same solution?

Best regards,

–
Untitled_12.gms (2.1 KB)

9

Hi

I have now set optcr=0. Shouldn’t it reduce the difference between the integer solution and the optimal one to zero?




On Tuesday, March 4, 2014 11:59:10 AM UTC+8, Zohreh wrote:

Dear Friends,

I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.

My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).

Which solver I must choose?


z2 T1 T2 v g Solver Status

LINDOGLOBAL 855434.1617686 0.054794521 0.01 15.14004249 4 optimal

BONMIN 855434.1617756 0.05479452 0.01 15.14004623 4 integer solution

OQNLP 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

SBB 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

KNITRO 855434.1617677 0.054794521 0.01 15.14004631 4 integer solution

DICOPT 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution


Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

10

Actually, when I checked the results now, I found that the best integer is greater than the best estimate (Final integrality gap (abs / rel) =-4.08e-004 / -4.76e-010 (-0.00%)), and that is why the algorithm stops and returns the solver status as integer solution, instead of optimal solution.

Is it normal to have negative value for this gap?

If yes, can I set the absolute value of optcr to zero to get the optimal value?


On Tuesday, March 4, 2014 11:59:10 AM UTC+8, Zohreh wrote:

Dear Friends,

I run my MINLP model by different solvers as follows. I also run it by BARON, but after 9 hours I stopped it as it was still working. Moreover, except LINDOGLOBAL which lasted for around 6 hours, the others took less than 1 minute.

My question is that why the result of LINDOGLOBAL which is a global optimizer and has found the optimal solution is 7.01413955539465E-06 less than BONMIN which has found the integer solution (the problem is maximization).

Which solver I must choose?


z2 T1 T2 v g Solver Status

LINDOGLOBAL 855434.1617686 0.054794521 0.01 15.14004249 4 optimal

BONMIN 855434.1617756 0.05479452 0.01 15.14004623 4 integer solution

OQNLP 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

SBB 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution

KNITRO 855434.1617677 0.054794521 0.01 15.14004631 4 integer solution

DICOPT 855434.1617686 0.054794521 0.01 15.14004623 4 integer solution


Best regards.

–
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.