What do I intrepret from this resut about the problem? Is MINOS able to solve the problem? shall I change some options?

196 views
Skip to first unread message

Harprinder

unread,
Jul 1, 2019, 4:24:10 PM7/1/19
to AMPL Modeling Language
Hi

Here is the display

'MINOS 5.51: major_iterations=10000000
minor_iterations=10000000
MINOS 5.51: the objective has not changed for the last 358 iterations.
1417 iterations, objective 8098.368778
Nonlin evals: obj = 11286, grad = 11285, constrs = 11286, Jac = 11285.'

Thanks
Harprinder

AMPL Google Group

unread,
Jul 2, 2019, 6:41:47 AM7/2/19
to AMPL Modeling Language
MINOS has given the termination message "the objective has not changed for the last 358 iterations." This means that MINOS has become stuck at a solution that does not satisfy the local optimality conditions.

Thus MINOS is unable to solve the problem from the given starting point. A carefully chosen starting point may give better results; see the discussion at https://ampl.com/BOOK/CHAPTERS/21-nonlinear.pdf#page=8. (Note that if you do not specify any starting point, a start with all variables at zero is assumed; for many problems, this is not a good choice.)

Also you may want to try other nonlinear solvers. In particular, the algorithm implemented in MINOS is best for problems that have a lot of linear or near-linear constraints, while other solvers may do better on problems that mostly have highly nonlinear constraints.

--
Robert Fourer
am...@googlegroups.com
{#HS:893181639-48098#}

Harprinderjot Singh

unread,
Jul 2, 2019, 9:38:53 AM7/2/19
to am...@googlegroups.com
Ohh

My problem is highly nonlinear with non linear constraints and nonlinear objective function. I tried to use other solvers but those didn't work. E.g. Knitro gave me an error that it can't handle non linear quadratic constraint. I also try using SNOPT (i hope i spelled it right) which doesn't seems better than MINOS. Other than that I tried different other solver which are not able to solve it. Can you suggest some solver.Even with MINOS sometimes it says infeasible problem or bad starting guess. I have been trying to reduce my problem size as small as possible and trying various ways.

Please suggest something in this regard.

Thanks
Harprinder

--
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.
To view this discussion on the web visit https://groups.google.com/d/msgid/ampl/reply-77152-893181639-2502481763-1562064102-1514101211%40helpscout.net.
For more options, visit https://groups.google.com/d/optout.

Harprinderjot Singh

unread,
Jul 2, 2019, 12:50:11 PM7/2/19
to am...@googlegroups.com
Hi

This is my problem. 

'Presolve eliminates 344 constraints and 438 variables.
Substitution eliminates 978 variables.
Adjusted problem:
134 variables:
126 nonlinear variables
8 linear variables
52 constraints; 838 nonzeros
38 nonlinear constraints
14 linear constraints
20 equality constraints
32 inequality constraints
1 nonlinear objective; 134 nonzeros.'

The problem also has ceiling function with variable inside it and different places.

Thanks
Harprinder

AMPL Google Group

unread,
Jul 5, 2019, 4:10:07 AM7/5/19
to AMPL Modeling Language
You mention that your model has a "ceiling function with a variable inside it." In general, none of our current solvers can handle the ceiling function applied to variables. These solvers assume smooth and continuous functions, while ceiling is discontinuous. There are some exceptions where these solvers can find a locally optimal solution even though the model applies ceiling to variables, but such exceptions are very rare and are unlikely to occur in problem. Instead you will see various nonoptimal termination messages, such as the message "the objective has not changed" from MINOS.

Currently, to handle models involving ceiling (and related functions floor, round, and trunc) applied to variables, it's necessary to reformulate in terms of smooth functions and integer variables, and then send the problem to a mixed-integer nonlinear solver such as Knitro, BARON, or Couenne. It is not possible to explain the reformulation in general; it depends on how you have used the ceiling function in your model.

--
Robert Fourer
am...@googlegroups.com
{#HS:893181639-48098#}
On Tue, Jul 2, 2019 at 4:50 PM UTC, AMPL Modeling Language <am...@googlegroups.com> wrote:
Hi

This is my problem.

'Presolve eliminates 344 constraints and 438 variables.
Substitution eliminates 978 variables.
Adjusted problem:
134 variables:
126 nonlinear variables
8 linear variables
52 constraints; 838 nonzeros
38 nonlinear constraints
14 linear constraints
20 equality constraints
32 inequality constraints
1 nonlinear objective; 134 nonzeros.'

The problem also has ceiling function with variable inside it and different places.

Thanks
Harprinder


On Tue, Jul 2, 2019 at 1:38 PM UTC, AMPL Modeling Language <am...@googlegroups.com> wrote:
Ohh

My problem is highly nonlinear with non linear constraints and nonlinear objective function. I tried to use other solvers but those didn't work. E.g. Knitro gave me an error that it can't handle non linear quadratic constraint. I also try using SNOPT (i hope i spelled it right) which doesn't seems better than MINOS. Other than that I tried different other solver which are not able to solve it. Can you suggest some solver.Even with MINOS sometimes it says infeasible problem or bad starting guess. I have been trying to reduce my problem size as small as possible and trying various ways.

Please suggest something in this regard.

Thanks
Harprinder


On Tue, Jul 2, 2019 at 10:41 AM UTC, AMPL Google Group <am...@googlegroups.com> wrote:
MINOS has given the termination message "the objective has not changed for the last 358 iterations." This means that MINOS has become stuck at a solution that does not satisfy the local optimality conditions.

Thus MINOS is unable to solve the problem from the given starting point. A carefully chosen starting point may give better results; see the discussion at https://ampl.com/BOOK/CHAPTERS/21-nonlinear.pdf#page=8. (Note that if you do not specify any starting point, a start with all variables at zero is assumed; for many problems, this is not a good choice.)

Also you may want to try other nonlinear solvers. In particular, the algorithm implemented in MINOS is best for problems that have a lot of linear or near-linear constraints, while other solvers may do better on problems that mostly have highly nonlinear constraints.

--
Robert Fourer
am...@googlegroups.com

Reply all
Reply to author
Forward
0 new messages