infeasible problem due to constraints

422 views
Skip to first unread message

k.ka...@googlemail.com

unread,
Feb 20, 2009, 7:27:15 PM2/20/09
to AMPL Modeling Language
Hello everyone,

I have a problem with constraints as part of a complicated model. I
have narrowed down the problem to just a single constraint:

-bd - v <= w. where b and d are variables and v and w are
parameters.

If I put comment before this constraint, the problem runs or even if i
change the sign of w. But this means that the problem has changed.

The error I get is

MINOS 5.5: infeasible problem (or bad starting guess).
60 iterations.

Please how can I reformat the constraint to get a feasible solution.

Please help.

Michael A. Saunders

unread,
Feb 20, 2009, 11:03:37 PM2/20/09
to AMPL Modeling Language
Dear k,

With nonlinear constraints, much depends on your starting point.
If the initial values of b and d are both zero, you are likely
to such results. Try initializing b and d to be 1, say.

Take care with the bounds on b and d also.
Are you allowing b or d to be zero?
If you expect them both to be larger than 0.5 (say),
then definitely give them such a lower bound.

Michael

k.ka...@googlemail.com

unread,
Feb 21, 2009, 6:59:52 PM2/21/09
to AMPL Modeling Language
Thanks Michael.

I tried initializing b and d to be both 0.01 and then 1 and i got

presolve: constraint NN[1,1] cannot hold:
body <= 0.003 cannot be >= 0.02; difference = -0.017.
This is when i initialize b and d with 0.01.

When i initilize b and d with 1, i got

presolve: constraint NN[1,1] cannot hold:
body <= 0.003 cannot be >= 2; difference = -1.997.

I actually, want b and d to be as close to 0 (zero) as possible.

How can i resolve this issue?

Please help.

On Feb 21, 4:03 am, "Michael A. Saunders" <saund...@stanford.edu>
wrote:
> Dear k,
>
> With nonlinear constraints, much depends on your starting point.
> If the initial values of b and d are both zero, you are likely
> to such results.  Try initializing b and d to be 1, say.
>
> Take care with the bounds on b and d also.
> Are you allowing b or d to be zero?
> If you expect them both to be larger than 0.5 (say),
> then definitely give them such a lower bound.
>
> Michael
>

Michael A. Saunders

unread,
Feb 21, 2009, 10:37:55 PM2/21/09
to AMPL Modeling Language
Dear k,

We need to know more facts about your constraint

-bd - v <= w, where b and d are variables and v and w are parameters.

Are v and w positive parameters?
(If so, you would want to initialize b and d with opposite sign.)
Are b and d allowed to be both positive or negative?
Can you set meaningful upper and lower bounds on them?

If you can post your model, someone more expert in AMPL might be
able to help.

Michael
Reply all
Reply to author
Forward
0 new messages