GAMS — The Solver Manuals - Available Software

CONOPT 151
10 APPENDIX B - CR-Cells
The CR-Cells that ordinary GAMS users can access are listed below. CR-Cells starting on R assume real values,
CR-Cells starting on LS assume logical values (TRUE, T, FALSE, or F), and all other CR-Cells starting on L
assume integer values. Several CR-Cells are only used in several versions of CONOPT in which case it will be
mentioned below. However, these CR- Cells can still be defined in an options file for the old CONOPT, but they
will silently be ignored:
Option Description Default
lfilog Iteration Log frequency. A log line is printed to the screen every lfilog iterations
(see also lfilos). The default value depends on the size of the model: it
is 10 for models with less than 500 constraints, 5 for models between 501 and
2000 constraints and 1 for larger models. The log itself can be turned on and
off with the Logoption (LO) parameter on the GAMS call.
Lfilos Iteration Log frequency for SLP and SQP iterations. A log line is printed to the
screen every lfilos iterations while using the SLP or SQP mode. The default
value depends on the size of the model: it is 1 for large models with more than
2000 constraints or 3000 variables, 5 for medium sized models with more than
500 constraints or 1000 variables, and 10 for smaller models.
lfderr The Function and Derivative Debugger (by default used with external equations)
10
will not write more than lfderr error messages independent of the num-
ber of errors found.
lfmxns Limit on new superbasics. When there has been a sufficient reduction in the 5
reduced gradient in one subspace, CONOPT tests if any nonbasic variables
should be made superbasic. The ones with largest reduced gradient of proper
sign are selected, up to a limit of lfmxns. The default value of lfmxns is 5.
The limit is replaced by the square root of the number of structural variables if
lfmxns is set to zero.
lfnicr Limit for slow progress / no increase. The optimization is stopped with a 12
”Slow Progress” message if the change in objective is less than 10 * rtobjr *
max(1,abs(FOBJ)) for lfnicr consecutive iterations where FOBJ is the value
of the current objective function. The default value of lfnicr is 12.
lfnsup Maximum Hessian dimension. If the number of superbasics exceeds lfnsup 500
CONOPT will no longer store a Reduced Hessian matrix.
CONOPT2 will switch to a steepest descend approach, independent of the degree
of nonlinearity of the model. The default value of lfnsup is 500. If lfnsup
is increased beyond its default value the default memory allocation may not be
sufficient, and you may have to include a ”.WORKSPACE = xx.x;” statement
in your GAMS source file, where ”model” represent the GAMS name of
the model. You should try to increase the value of lfnsup if CONOPT performs
many iterations in Phase 4 with the number of superbasics (NSB) larger
than lfnsup and without much progress. The new value should in this case be
larger than the number of superbasics.
CONOPT3 will also refrain from using a reduced Hessian. However, it can still
use second derivatives in combination with a conjugate gradient algorithm. It is
usually not a good idea to increase lfnsup much beyond its default value of 500
with CONOPT3. The time used to manipulate a very large reduced Hessian
matrix is often large compared to the potential reduction in the number of
iterations.
(Note: CONOPT2 and CONOPT3 react very differently to changes in lfnsup.)
lfscal Frequency for scaling. The scale factors are recomputed after lfscal recomputations
of the Jacobian. The default value is 20. Not CONOPT1
20