Minos: Too many major interations

[shu...@gmail.com]

What does the above error indicate? Also, it did not show optimal solution found, so I'm trying to find out what is causing it to have too many iterations. I am using MINOS solver.

[Robert Fourer]

By default MINOS stops after 50 major iterations -- see the "Major_iterations" option on page 8 of http://ampl.com/BOOKLETS/ampl-minos.pdf. You can increase the allowed number of major iterations with an AMPL command such as

option minos_options 'major_iterations=1000';

(or add major_iterations=1000 to a minos_options string that you are already using).

Bob Fourer
am...@googlegroups.com

[Michael Saunders]

If you hit the Major iteration limit, your problem has nonlinear constraints.

In each major iteration, Minos optimizes a certain subproblem (with augmented

Lagrangian as objective, subject to linearized constraints).  Each subproblem

is allowed a limited number of Minor iterations (default 40 after a feasible point

is found for the subproblem).  You probably don't want to change this limit,

but you may need to increase "Iterations", which limits the total number of

minor iterations for all subproblems.  E.g., add 'Iterations=50000' to your

minos_options string.

Without knowing the size of your problem, it's hard to suggest sensible limits.

Michael Saunders

--
You received this message because you are subscribed to the Google Groups "AMPL Modeling Language" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ampl+uns...@googlegroups.com.
To post to this group, send email to am...@googlegroups.com.
Visit this group at https://groups.google.com/group/ampl.
For more options, visit https://groups.google.com/d/optout.

[shu...@gmail.com]

Hi Michael,

Here is the problem size I am dealing with. How do i decide sensible limits, and how will that scale with the size of my problem? I'm currently working on a reduced problem and the original data set is more than 3 orders of magnitude larger.

153 variables:

126 nonlinear variables

27 linear variables

125 constraints; 500 nonzeros

25 nonlinear constraints

100 linear constraints

100 equality constraints

25 inequality constraints

1 nonlinear objective; 51 nonzeros.

If you need any other information do let me know.

Seth

[Michael Saunders]

Hi Seth,

So the problem size is not (yet) a cause for many iterations.

It must be the degree of nonlinearity.  Perhaps just the scaling

of the nonlinear variables (the default Scaling option leaves them alone),

but maybe the functions are unfriendly.

The Print file from Minos (with option Solution=Yes) would tell us more.

Since it's a small model, you could share it with everyone.

Good luck,

Michael

To unsubscribe from this group and stop receiving emails from it, send an email to ampl+unsubscribe@googlegroups.com.

[shu...@gmail.com]

Dear Michael,

I've added the line: option minos_options 'Solution = Yes';

Where do I find the Print file?

This is the output I am getting now:

File exists

You are using the solver minos.

Executing AMPL.

processing data.

processing commands.

Executing on neos-7.neos-server.org

153 variables:

126 nonlinear variables

27 linear variables

125 constraints; 500 nonzeros

25 nonlinear constraints

100 linear constraints

100 equality constraints

25 inequality constraints

1 nonlinear objective; 51 nonzeros.

MINOS 5.51: Solution = Yes

MINOS 5.51: too many major iterations.

383 iterations, objective -28.86297347

Nonlin evals: obj = 965, grad = 964, constrs = 965, Jac = 964.

beta [*] :=

1  -6.61091

2  10.342

;

MINOS 5.51: Solution = Yes

MINOS 5.51: too many major iterations.

363 iterations, objective -28.97460208

Nonlin evals: obj = 860, grad = 859, constrs = 860, Jac = 859.

Seth

[Michael Saunders]

Hi Seth,

You can add option 'Print_file 9' to create a file called fort.9 (or something similar).

You could also add "Summary_file 6' to see the major iterations on the screen

and get more idea of what's happening.

You haven't mentioned the type of nonlinearity in the constraints and objective.

I meant that since your model is small, you could share the model with everyone.

Michael

To unsubscribe from this group and stop receiving emails from it, send an email to ampl+unsubscribe@googlegroups.com.