CUTEr tools

Problems are fully described in a Standard Input Format (SIF) file. This file may subsequently be decoded to provide data and Fortran subroutines for input to a nonlinear programming package.

Table 1.2: The unconstrained minimization CUTEr tools as of March 24, 2005.

Tool name	Brief description
ubandh	extract a banded matrix out of the Hessian matrix,
udh	evaluate the Hessian matrix,
udimen	get the number of variables involved,
udimse	determine the number of nonzeros required to store the
	sparse Hessian matrix in finite element format,
udimsh	same as udimse, in coordinate format,
ueh	evaluate the sparse Hessian matrix in finite element format,
ufn	evaluate function value,
ugr	evaluate gradient,
ugrdh	evaluate the gradient and Hessian matrix,
ugreh	evaluate the gradient and Hessian matrix in finite element format,
ugrsh	evaluate the gradient and Hessian matrix in coordinate format,
unames	obtain the names of the problem and its variables,
uofg	evaluate function value and possibly gradient,
uprod	form the matrix-vector product of a vector with the Hessian matrix,
usetup	set up the data structures for unconstrained minimization,
ush	evaluate the sparse Hessian matrix,
uvarty	determine the type of each variable.
ureprt	obtain statistics concerning function evaluation and CPU time used,

Table 1.3: The constrained minimization CUTEr tools as of March 24, 2005.

Tool name	Brief description
ccfg	evaluate constraint functions values and possibly gradients,
ccfsg	same as ccfg, in sparse format,
ccifg	evaluate a single constraint function value and possibly gradient,
ccifsg	same as ccifg, in sparse format,
cdh	evaluate the Hessian of the Lagrangian,
cdimen	get the number of variables and constraints involved,
cdimse	determine number of nonzeros to store the Lagrangian Hessian,
	in finite element format,
cdimsh	determine number of nonzeros to store the Lagrangian Hessian,
	in coordinate format,
cdimsj	determine number of nonzeros to store the matrix of gradients of
	the objective function and constraints, in sparse format,
ceh	evaluate the sparse Lagrangian Hessian in finite element format,
cfn	evaluate function and constraints values,
cgr	evaluate constraints gradients and objective/Lagrangian gradient,
cgrdh	same as cgr, plus Lagrangian Hessian,
cidh	evaluate the Hessian of a problem function,
cish	same as cidh, in sparse format,
cnames	obtain the names of the problem and its variables,
cofg	evaluate function value and possibly gradient,
cprod	form the matrix-vector product of a vector with the Lagrangian Hessian,
cscfg	evaluate constraint functions values and possibly gradients in sparse format,
cscifg	same as cscfg, for a single constraint,
csetup	set up the data structures for constrained minimization,
csgr	evaluate constraints and objective/Lagrangian function gradients,
csgreh	evaluate both the constraint gradients, the Lagrangian Hessian
	in finite element format and the gradient of the
	objective/Lagrangian in sparse format,
csgrsh	same as csgreh, in sparse format instead of finite element format,
csh	evaluate the Hessian of the Lagrangian, in sparse format,
cvarty	determine the type of each variable,
creprt	obtain statistics concerning function evaluation and CPU time used,

Here we describe auxiliary subroutines which are available for users to manipulate the decoded data. The Fortran source of these programs, along with the subroutines obtained when decoding the SIF file, should be compiled with the user's optimization package.

The CUTEr tools are described in the man pages, category 3, which may be viewed using the man command, its X interface xman, or, on LINUX systems, by issuing less manpage where manpage is the man page to be viewed. The man pages are stored in $CUTER/common/man, and this directory should appear in the user's MANPATH. Table 1.2 contains those tools related to unconstrained or bound-constrained minimization as of March 24, 2005, along with a brief description, and Table 1.3 contains a list of CUTEr tools for constrained minimization. The ureprt and creprt tools produce statistics about a particular run. Users of the previous versions of CUTE will notice the strong similarity in the tools names.

Whenever the description states that the Hessian matrix of either the objective or the Lagrangian function is in sparse format, it is implicitly understood that it is stored in coordinate format. Explicit mentions appear whenever this matrix is stored in finite-elements format.