CONOPT Infeasibility

GAMS
1

Hi,

I am solving a GAMS NLP model using CONOPT solver. The Lst file shows the Model Status is Infeasble and the meaasge i got is “Infeasible solution. Reduced gradient less than tolerance.”. I know which equation is causing the infeasibility. However, GAMS returned values for all the variables and by verifying the equations manually, the variable values are correct and all equations are satisfied given the values found by GAMS. I got other warning massages which are:

** Warning ** The variance of the derivatives in the initial
point is large (= 7.9 ). A better initial
point, a better scaling, or better bounds on the
variables will probably help the optimization.


Pre-triangular equations: 4915
Post-triangular equations: 28


** Infeasible solution. Reduced gradient less than tolerance.


** Warning ** The variance of the derivatives in the final
point is large (= 6.2 ). A better scaling or
better bounds on the variables will probably
help the optimization.

I’ve added bounds to my variables but still get the same solution and the same error/warning massage.

I also added the following line to raise the upper bound of CONOPT but still I get warning massages:

$onecho > conopt.opt
Rtmaxv=10e29
$offecho
PSD.optfile = 1;
Also I tried different solvers only conopt return correct values of the variables (correct values are the model solved manually which I use to compare solutions)

Can anyone tell me what the source of infeasibility is? is there anything wrong with the solver since my equations are correct? or can I simply ignore the infeasibility?

Thank you

2

If you need to tell Conopt to accept coefficients/gradients of 1e30 (Rtmaxv=10e29) you have a very poorly scaled model. If you scale your model properly I guess your infeasibility problems (due to scaling) will go away. The Conopt manual has a good section on scaling: https://www.gams.com/latest/docs/S_CONOPT.html#CONOPT_SCALING.

-Michael

3

I have a similar issue while upgrading to CONOPT4 with the following algorithm options, does anyone could help how to upgrade to CONOPT4. How to relax the pre-triangular feasibility tolerance?

rtnwtr = 1.e-3;
lfnicr = 1000;
lfstal = 1000;
lfilog = 500;
rtredg = 1.d-3;
rvhess 100;
lfdegi 500;
lfnsup 7000;
rtmins 0.0009765625;
rtnwmi = 1.d-9;