16.01.2015 Views

GAMS — The Solver Manuals - Available Software

GAMS — The Solver Manuals - Available Software

GAMS — The Solver Manuals - Available Software

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

154 CONOPT<br />

Option Description Default<br />

rtpiva Absolute pivot tolerance. A pivot element is only considered acceptable if its 1.e-10<br />

absolute value is larger than rtpiva. <strong>The</strong> default value is 1.e-10. You may have<br />

to decrease this value towards 1.e-11 or 1.e-12 on poorly scaled models.<br />

rtpivr Relative pivot tolerance. A pivot element is only considered acceptable relative 0.05<br />

to other elements in the column if its absolute value is at least rtpivr * the<br />

largest absolute value in the column. <strong>The</strong> default value is 0.05. You may have<br />

to increase this value towards one on poorly scaled models. Increasing rtpivr<br />

will result in denser L and U factors of the basis.<br />

rtpivt Triangular pivot tolerance. A nonlinear triangular pivot element is considered 1.e-7<br />

acceptable if its absolute value is larger than rtpivt. <strong>The</strong> default value is 1.e-7.<br />

Linear triangular pivot must be larger than rtpiva.<br />

rtredg Optimality tolerance. <strong>The</strong> reduced gradient is considered zero and the solution 9.e-8<br />

optimal if the largest superbasic component is less than rtredg. <strong>The</strong> default<br />

value depends on the machine, but is usually around 9.e-8. If you have problems<br />

with slow progress or stalling you may increase rtredg. This is especially<br />

relevant for very large models.<br />

rvspac A space allocation factor that sometime can speed up the solution of square<br />

systems. CONOPT will tell you if it is worth while to set this parameter to a<br />

non-default value for your class of model.<br />

rvstlm Step length multiplier. <strong>The</strong> step length in the one-dimensional line search is not 4<br />

allowed to increased by a factor of more than rvstlm between steps for models<br />

with nonlinear constraints and a factor of 100 * rvstlm for models with linear<br />

constraints. <strong>The</strong> default value is 4.<br />

dohess A logical variable that controls the creation of the Hessian (matrix of second<br />

derivatives). <strong>The</strong> default value depends on the model. If the number of equalities<br />

is very close to the number of non-fixed variables then the solution is<br />

assumed to be in a corner point or in a very low dimensional space where second<br />

derivatives are not needed, and dohess is initialized to false. Otherwise<br />

dohess is initialized to true. If dohess is false you will not get statistics about<br />

the Hessian in the listing file.<br />

It takes some time to generate second order information and it uses some space.<br />

If CONOPT3 generates this information for your model but it does not use it,<br />

i.e. if you see that no time is spend on 2nd derivative evaluations, then you may<br />

experiment with dohess turned off. If the number of Hessian elements is very<br />

large you may also try turning dohess off. Note that CONOPT3 still can use<br />

directional second derivatives and therefore use its SQP algorithm in the cases<br />

where the Hessian is not available. (CONOPT3 only).<br />

rvhess A real number that controls the space available for creation of the Hessian. <strong>The</strong> 10<br />

maximum number of nonzero elements in the Hessian and in some intermediate<br />

terms used to compute it is limited by Rvhess times the number of Jacobian<br />

elements (first derivatives). <strong>The</strong> default value of Rvhess is 10, which means that<br />

the Hessian should not be denser than 10 second derivatives per first derivative.<br />

(CONOPT3 only).<br />

Gcform Defines the functional form used to implement Cone constraints as a nonlinear<br />

inequality constraint using a 0-1 value.<br />

0<br />

0: <strong>The</strong> Cone constraints are implemented as sqr(x) =G= sum(i, sqr( y(i)<br />

) ) for the quadratic cone and 2*x1*x2 =G= sum(i, sqr( y(i) ) ) for<br />

the rotated or hyperbolic cone.<br />

1: <strong>The</strong> cone constraints are implemented as x+GCPtb2 =G=<br />

sqrt(GCPtb1+sum(i,sqr(y(i)))) for the quadratic cone and<br />

(GCPtb1+2*x1*x2) =G= Sqrt(GCptb1+sum(i,sqr(y(i)))) for the<br />

rotated or hyperbolic cone.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!