Why is marginal value for my NLP not available?

Hi,

I am solving a NLP problem by using BARON/GAMS.
After several iterations, the upper bound and lower bound are converged (global optimal), but I found that for simple binding constraints, such as x(i)=l=5, the marginal value is not available (NA) to obtain from solver.
Does anyone know why? I really appreciate for sharing your ideas.


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.

Hi,

There were recent changes in GAMS/BARON and BARON itself that addressed this issue. Make sure you are using the latest version (24.2.1) and post the model, so we can reproduce. In general BARON does not report the duals if the final solve (restart from the best point by the NLP solver) results in a solution with infeasibilities larger than some internal tolerance. You might want to switch the NLP solver used by BARON and see if the problem disappears. BTW, NA in general is not a bad marginal. In GAMS this indicates that the row/column is non-basic with a zero marginal value.

Best,
Michael Bussieck - GAMSWorld Coordinator

On Saturday, February 1, 2014 9:48:17 AM UTC-5, yya…@ualberta.ca wrote:

Hi,

I am solving a NLP problem by using BARON/GAMS.
After several iterations, the upper bound and lower bound are converged (global optimal), but I found that for simple binding constraints, such as x(i)=l=5, the marginal value is not available (NA) to obtain from solver.
Does anyone know why? I really appreciate for sharing your ideas.


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.

Just a quick correction: Michael’s note confused EPS and NA. EPS in general is not a bad marginal, but instead indicates that the row/column is non-basic with a zero marginal value. NA is an indication that no marginal is available. The EPS is quite common. The NA value for a marginal is much less common.

-Steve


On Mon, Feb 3, 2014 at 2:56 AM, Michael Bussieck wrote:

Hi,

There were recent changes in GAMS/BARON and BARON itself that addressed this issue. Make sure you are using the latest version (24.2.1) and post the model, so we can reproduce. In general BARON does not report the duals if the final solve (restart from the best point by the NLP solver) results in a solution with infeasibilities larger than some internal tolerance. You might want to switch the NLP solver used by BARON and see if the problem disappears. BTW, NA in general is not a bad marginal. In GAMS this indicates that the row/column is non-basic with a zero marginal value.

Best,
Michael Bussieck - GAMSWorld Coordinator


On Saturday, February 1, 2014 9:48:17 AM UTC-5, yya…@ualberta.ca wrote:

Hi,

I am solving a NLP problem by using BARON/GAMS.
After several iterations, the upper bound and lower bound are converged (global optimal), but I found that for simple binding constraints, such as x(i)=l=5, the marginal value is not available (NA) to obtain from solver.
Does anyone know why? I really appreciate for sharing your ideas.


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.



\

Steven Dirkse, Ph.D.
GAMS Development Corp., Washington DC
Voice: (202)342-0180 Fax: (202)342-0181
sdirkse@gams.com
http://www.gams.com

Thanks for your replying! I use the newest gams 24.2.1

Since our model is for the refinery optimization, it is too big to post.

If we use the SNOPT instead of BARON, the marginal value is always available. However, we still prefer the BARON because we want the global optimal. For some data scenarios, the BARON can provide the marginal value, but there are some other cases such that BARON only shows NA for the marginal value. BTW, the BARON also uses SNOPT as a NLP solver in my program, do I need to change it?

Based on your analysis, is it possible that the BARON’s solution actually is infeasible?


Regards,

Yu

On Monday, February 3, 2014 2:56:02 AM UTC-5, Michael Bussieck wrote:

Hi,

There were recent changes in GAMS/BARON and BARON itself that addressed this issue. Make sure you are using the latest version (24.2.1) and post the model, so we can reproduce. In general BARON does not report the duals if the final solve (restart from the best point by the NLP solver) results in a solution with infeasibilities larger than some internal tolerance. You might want to switch the NLP solver used by BARON and see if the problem disappears. BTW, NA in general is not a bad marginal. In GAMS this indicates that the row/column is non-basic with a zero marginal value.

Best,
Michael Bussieck - GAMSWorld Coordinator

On Saturday, February 1, 2014 9:48:17 AM UTC-5, yya…@ualberta.ca wrote:

Hi,

I am solving a NLP problem by using BARON/GAMS.
After several iterations, the upper bound and lower bound are converged (global optimal), but I found that for simple binding constraints, such as x(i)=l=5, the marginal value is not available (NA) to obtain from solver.
Does anyone know why? I really appreciate for sharing your ideas.


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.

Yu,

What I said is that the last solve of SNOPT inside BARON might have some difficulties and therefore BARON does not provide marginals. What happens if you restart SNOPT with the primal solution from BARON to get duals? I would try different NLP solvers in BARON (option nlpsol). If you think the model is too big to post here, you should consider contacting GAMS support to get specific help on your problem.

Best,
Michael

On Thursday, February 6, 2014 11:25:36 AM UTC-5, yya…@ualberta.ca wrote:

Thanks for your replying! I use the newest gams 24.2.1

Since our model is for the refinery optimization, it is too big to post.

If we use the SNOPT instead of BARON, the marginal value is always available. However, we still prefer the BARON because we want the global optimal. For some data scenarios, the BARON can provide the marginal value, but there are some other cases such that BARON only shows NA for the marginal value. BTW, the BARON also uses SNOPT as a NLP solver in my program, do I need to change it?

Based on your analysis, is it possible that the BARON’s solution actually is infeasible?


Regards,

Yu

On Monday, February 3, 2014 2:56:02 AM UTC-5, Michael Bussieck wrote:

Hi,

There were recent changes in GAMS/BARON and BARON itself that addressed this issue. Make sure you are using the latest version (24.2.1) and post the model, so we can reproduce. In general BARON does not report the duals if the final solve (restart from the best point by the NLP solver) results in a solution with infeasibilities larger than some internal tolerance. You might want to switch the NLP solver used by BARON and see if the problem disappears. BTW, NA in general is not a bad marginal. In GAMS this indicates that the row/column is non-basic with a zero marginal value.

Best,
Michael Bussieck - GAMSWorld Coordinator

On Saturday, February 1, 2014 9:48:17 AM UTC-5, yya…@ualberta.ca wrote:

Hi,

I am solving a NLP problem by using BARON/GAMS.
After several iterations, the upper bound and lower bound are converged (global optimal), but I found that for simple binding constraints, such as x(i)=l=5, the marginal value is not available (NA) to obtain from solver.
Does anyone know why? I really appreciate for sharing your ideas.


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.

Hi,

I think I have found the reason. The marginal value is NA when .modelstat=7.
I don’t understand why because the lower and upper bounds are close enough since I set optcr=0.00001. May be I need to switch the SNOPT to other NLP solver.

Regards,

Yu


On Thursday, February 6, 2014 12:16:47 PM UTC-5, Michael Bussieck wrote:

Yu,

What I said is that the last solve of SNOPT inside BARON might have some difficulties and therefore BARON does not provide marginals. What happens if you restart SNOPT with the primal solution from BARON to get duals? I would try different NLP solvers in BARON (option nlpsol). If you think the model is too big to post here, you should consider contacting GAMS support to get specific help on your problem.

Best,
Michael

On Thursday, February 6, 2014 11:25:36 AM UTC-5, yya…@ualberta.ca wrote:

Thanks for your replying! I use the newest gams 24.2.1

Since our model is for the refinery optimization, it is too big to post.

If we use the SNOPT instead of BARON, the marginal value is always available. However, we still prefer the BARON because we want the global optimal. For some data scenarios, the BARON can provide the marginal value, but there are some other cases such that BARON only shows NA for the marginal value. BTW, the BARON also uses SNOPT as a NLP solver in my program, do I need to change it?

Based on your analysis, is it possible that the BARON’s solution actually is infeasible?


Regards,

Yu

On Monday, February 3, 2014 2:56:02 AM UTC-5, Michael Bussieck wrote:

Hi,

There were recent changes in GAMS/BARON and BARON itself that addressed this issue. Make sure you are using the latest version (24.2.1) and post the model, so we can reproduce. In general BARON does not report the duals if the final solve (restart from the best point by the NLP solver) results in a solution with infeasibilities larger than some internal tolerance. You might want to switch the NLP solver used by BARON and see if the problem disappears. BTW, NA in general is not a bad marginal. In GAMS this indicates that the row/column is non-basic with a zero marginal value.

Best,
Michael Bussieck - GAMSWorld Coordinator

On Saturday, February 1, 2014 9:48:17 AM UTC-5, yya…@ualberta.ca wrote:

Hi,

I am solving a NLP problem by using BARON/GAMS.
After several iterations, the upper bound and lower bound are converged (global optimal), but I found that for simple binding constraints, such as x(i)=l=5, the marginal value is not available (NA) to obtain from solver.
Does anyone know why? I really appreciate for sharing your ideas.


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.