BARON ,Infeasible

[samaneh]

Dear all,

 

Previousely I solve my model by using AOA with multi- start algorithm and recieve a local optimal solution. now I would like to get a global optimal solution by using BARON.

I use BARON solver to solve my model which is MINLP, concave function with linear constraints. but after run , it reture this warning:

Warning: BARON returning 'infeasible' might indicate that the value of the option 'maximal variable bound' was too low.

what the reason for infeasibility? how can I modify it?

Thanks for your help and guidance.

 

Best regards,

Samaneh

 

[Marcel Hunting]

Hi,

BARON requires that all nonlinear variables and expressions in the mathematical program are bounded from below and above. It is important that the user provides finite lower and upper bounds on all variables. If not, global optimality of the solutions provided is not guaranteed.

The BARON option 'Maximal Variable Bound' is used to make sure that variables without a bound get a bound. You can increase the value of this option by using menu Settings - Option Tree - Specific solvers - Baron - General -  Maximal Variable Bound (see also the AIMMS-BARON help for this option). Or better, assign real upper and lower bounds to all variables (except the objective variable which must be free).

Marcel Hunting

AIMMS Software Developer

[José Magalhaes Jr.]

Helllo Folks !

Could you help me do find out the problem in my Model?

When I run the Main Execution, I get the following messages:

Warning: The columns "x(3,8,1)" and "x(2,8,1)" are equal in the generated mathematical program ..... see the help associated with the option "Warning_duplicate_column".

Warning: The columns "x(2,8,1)" and "x(1,8,1)" are equal in the generated mathematical program...... see the help associated with the option "Warning_duplicate_column".

Please,

What do these messages mean? 

Is it a problema in one of my constraints?

Another point

I´d like to check if I wrote correctly my constraint. 

Is it possible to check How the constrainsts are being generated ?!

For example, I have

Identifier: Const4

Index: k

Definition: sum((i,j)|val(i)=1 and (val(j)>1 AND val(j)<=(BaseFinal/2)),x(k,i,j)) + sum((i,j)|val(i)=1 and val(j)=BaseFinal,x(k,i,j))=1

I´d like to see the equations associated to this constraint:

k=1:  X(1,2,2) + X (1,2,3) +......  =1

k=2:  X(2,2,2) + X (2,2,3) +......  =1

k=3:  X(3,2,2) + X (3,2,3) +......  =1

    and so on

I´d like to see all equations to verify if the constraint is being generated as I guess.

Is it possible?

Thanks for all help.

Regards,

    José Magalhães.

[samaneh bagheri]

Dear Marcel,

 

thanks for your help. I increase the variable bound. however, it takes a long time (more than 3 hours) to solve model,while by using GMP-AOA with multi start algorithm, the running time is less than 2 minuets. is there any way to decrease the run time in BARON? 

 

Another question,When I combine GMP-AOA with multi start algorithm the solution is the same as solution of GMP-AOA , can I conclude that this is a optimal solution or not?( remark that my model  is MINLP, concave function with linear constraints) 

 

 

Best regards,

Samaneh

--
You received this message because you are subscribed to the Google Groups "AIMMS - The Modeling System" group.
To view this discussion on the web visit https://groups.google.com/d/msg/aimms/-/ZYUpCN5CToYJ.
To post to this group, send email to ai...@googlegroups.com.
To unsubscribe from this group, send email to aimms+un...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/aimms?hl=en.

[Marcel Hunting]

Hi,

BARON is a global solver that will try to find the global optimum and prove that it is the global optimum. That is much more difficult than "just" finding a good local solution as local solvers like GMP-AOA do. GMP-AOA can only guarantee to find the global optimum if the model is convex, otherwise it cannot (also not if multi-start is used).

You could try to improve the performance of BARON by switching on the general solvers option 'Nonlinear presolve'.  You also might try playing with the BARON options 'Point Selection Strategy' and 'Variable selection strategy'.

Best regards,

Marcel

[samaneh bagheri]

Hi Marcel,

 

 

Thanks for your reply. As  do not have any experiance in  using BARON, I would be grateful if you please suggest me some references or examples about BARON  'Point Selection Strategy' and 'Variable selection strategy'.

 

Thanks in advance,

Samaneh

 

 

To view this discussion on the web visit https://groups.google.com/d/msg/aimms/-/psPAorwBgjAJ.

[Guido Diepen]

Hi Jose,

next time please post the question as a separate question in the group instead of replying to another message. This helps to keep everything structured and easy to find for other people.

Regarding your question, these are not errors, but they are warnings that something might be wrong in your formulation.

if you have the very simple mathematical program:

max x1 + x2

s.t.  x1 + x2 <= 5

then x1 and x2 are exactly the same columns: they have the exact same coefficients in the objective and each of the constraints. this means that a lot of solutions exist. As stated in the warning, you can find more information in the AIMMS help. You can view this by pressing F1, go to the tab page Search, and type "warning duplicate column". This will show you the information about this particular warning.

You can verify the constraints that are generated in two ways:

* Math Program Inspector (Menu > Diagnostic Tools > Math Program Inspector). You can find more information about the MPI in the AIMMS Users Guide

* Constraint Listing. You can instruct AIMMS to create a textual representation of the model that is generated and sent to the solver. Instructions on how to do this, you can find at the knowledge base article http://www.aimms.com/knowledgebase/kb000017.html.

Guido Diepen

AIMMS Specialist

[samaneh bagheri]

Dear Marcel,

 

I am switching on the general solvers option 'Nonlinear presolve' and also put the smaller upperbound for variables than before, but when I run model by using BARON it takes several houres as the same as before, and finally I egnore to continue it and pouse the running process.

do you think your suggestion about using BARON options 'Point Selection Strategy' and 'Variable selection strategy' could help me to run model faster and receive results?

How can I modify this model to receive results?

 

Thanks in advance,

Samaneh

To view this discussion on the web visit https://groups.google.com/d/msg/aimms/-/psPAorwBgjAJ.

[Luis Pinto]

Hello,

On a side note, and out of curiosity, can AIMMS do the solver tunning for these kind of problems as well?
If not, to which types are they limited to?

Cheers,

Luis Pinto

www.unisoma.com.br

[Marcel Hunting]

Hi,

I am not surprised that BARON takes much longer as finding a global optimum is much more difficult then finding a local optimum. And problems in which a concave function is minimized are especially hard for BARON.

It would be interesting to know how the lower bound and upper bound in the file 'baron_sum.lst' (in the log folder; after switching on BARON option 'Keep Summary File') behave. Is the gap decreasing slowly or is the gap already small after 30 minutes?

There is no guarantee that setting the values of the BARON options 'Point Selection Strategy' and 'Variable selection strategy' to non-default values will results in a better performance. You can set a time limit of, e.g., 5 minutes and solve the problem for different combinations of 'Point Selection Strategy' and 'Variable selection strategy' settings, and check the gap between the lower and upper bound in the progress window (or in the 'baron_sum.lst' file) after 5 minutes. For the most promising setting you could then do a run with the time limit removed. There is no automatic way to tune these BARON options.

Best regards,

Marcel

[Marcel Hunting]

Hi Luis,

AIMMS leaves the solver tuning to the solvers. Currently CPLEX is the only solver in AIMMS that can do tuning.

Note that solvers have often many options that can influence the performance, and using an algorithm that simply tries all combinations of option settings might take a very long time to run. CPLEX uses a more clever approach such that the total tuning time is usually limited.

Regards,

Marcel

[attia...@yahoo.com]

Dear samaneh

I have used baron to solve a MINL problem and faced the same problems you mentioned . The only solution worked for my problem was to run the program with additional constrains to introduce a starting point in the form of fixed values or ranges for some variables.The tutorials said a starting point will help get a solution .Then remove these constraints and run again.Best wishes.

Sent from my iPad

To view this discussion on the web visit https://groups.google.com/d/msg/aimms/-/Gv60L4NSYCgJ.

[Marcel Hunting]

Hi Samaneh,

Could you try running your project with BARON after making the variables 'pipeline' and 'safety' Inline (by specifying Inline in the Property attribute of these variables)?

Marcel