Usage

The calling sequence for the package is

CALL NSPCG ($\langle$ precon $\rangle$ , $\langle$ accel $\rangle$ ,NDIM,MDIM,N,MAXNZ,COEF,JCOEF,P,IP,U,UBAR,RHS,
WKSP,IWKSP,NW,INW,IPARM,RPARM,IER)

where

$\langle$ precon $\rangle$

is the name of a preconditioning routine which must be declared EXTERNAL in the user's calling routine. It must be one of a certain number of choices given in Section 3. [name]

$\langle$ accel $\rangle$

is the name of an acceleration routine which must be declared EXTERNAL in the user's calling routine. It must be one of a certain number of choices given in Section 4. [name]

NDIM

is the row dimension of the COEF array (See Section 8 on Storage Modes for more information). [integer, input]

MDIM

is the column dimension of the COEF array (See Section 8 on Storage Modes for more information). [integer, input]

N

is the order of the linear system. [integer, input]

MAXNZ

is the active column width of the COEF array (See Section 8 on Storage Modes for more information). [integer, input/output]

COEF

is a real array containing the matrix nonzero coefficients in one of several possible formats to be given in Section 8. [real, input/output]

JCOEF

is an integer array containing auxiliary matrix nonzero information corresponding to COEF. [integer, input/output]

P

is an integer vector. If no permutation is to be applied to the matrix, then P is unused and may be dimensioned to be of length one. If the matrix is to be permuted into another form, such as red-black, P is a coloring vector of length N indicating the integer coloring number of each equation. On output, P is then the permutation vector corresponding to the multi-color ordering. (See Sections 11 and 12 for more details.) [integer, input/output]

IP

is an integer vector. If no permutation is to be applied to the matrix, then IP is unused and may be dimensioned to be of length one. If the matrix is to be permuted, IP is an integer workspace vector of length N on input and the inverse permutation vector corresponding to P on output. (See Sections 11 and 12 for more details.) [integer, output]

U

is a real vector of length N containing the current estimate of the solution to the linear system on input. The user must supply an initial guess, which can be all zeros if no information is known. On output, U contains the solution estimate computed by the package. [real, input/output]

UBAR

is a real vector of length N containing the exact solution to the linear system, if known. This vector is used only if the exact stopping test is used. Otherwise, it may be dimensioned to be of length one. (See Section 10 for information on the available stopping tests.) [real, input]

RHS

is a real vector of length N containing the right-hand-side of the matrix problem. [real, input]

WKSP

is a real vector of length NW used for workspace. [real, input]

IWKSP

is an integer vector of length INW used for workspace. [integer, input]

NW

is an integer variable indicating the length of the WKSP vector on input and the amount used on output. If insufficient real workspace is provided, NW is set on output to the amount of workspace needed at the point execution terminated. (See Section 9 for suggested initial values.) [integer, input/output]

INW

is an integer variable indicating the length of the IWKSP vector on input and the amount used on output. If insufficient integer workspace is provided, INW is set on output to the amount of workspace needed at the point execution terminated. (See Section 9 for suggested initial values.) [integer, input/output]

IPARM

is an integer vector of length 25 giving various integer parameters which are described in Section  7. [integer, input/output]

RPARM

is a real vector of length 16 giving various real parameters which are described in Section 7. [real, input/output]

IER

is an error flag. A nonzero value on output indicates a fatal or warning error was detected during the iteration. (See Section 13 for a list of error codes.) [integer, output]