Purpose
To solve a set of systems of linear equations, A'*A*X = B, or, in the implicit form, f(A)*X = B, with A'*A or f(A) positive definite, using symmetric Gaussian elimination.Specification
SUBROUTINE MB02XD( FORM, STOR, UPLO, F, M, N, NRHS, IPAR, LIPAR,
$ DPAR, LDPAR, A, LDA, B, LDB, ATA, LDATA, DWORK,
$ LDWORK, INFO )
C .. Scalar Arguments ..
CHARACTER FORM, STOR, UPLO
INTEGER INFO, LDA, LDATA, LDB, LDPAR, LDWORK, LIPAR, M,
$ N, NRHS
C .. Array Arguments ..
DOUBLE PRECISION A(LDA,*), ATA(*), B(LDB,*), DPAR(*), DWORK(*)
INTEGER IPAR(*)
Arguments
Mode Parameters
FORM CHARACTER*1
Specifies the form in which the matrix A is provided, as
follows:
= 'S' : standard form, the matrix A is given;
= 'F' : the implicit, function form f(A) is provided.
If FORM = 'F', then the routine F is called to compute the
matrix A'*A.
STOR CHARACTER*1
Specifies the storage scheme for the symmetric
matrix A'*A, as follows:
= 'F' : full storage is used;
= 'P' : packed storage is used.
UPLO CHARACTER*1
Specifies which part of the matrix A'*A is stored, as
follows:
= 'U' : the upper triagular part is stored;
= 'L' : the lower triagular part is stored.
Function Parameters
F EXTERNAL
If FORM = 'F', then F is a subroutine which calculates the
value of f(A) = A'*A, for given A.
If FORM = 'S', then F is not called.
F must have the following interface:
SUBROUTINE F( STOR, UPLO, N, IPAR, LIPAR, DPAR, LDPAR, A,
$ LDA, ATA, LDATA, DWORK, LDWORK, INFO )
where
STOR (input) CHARACTER*1
Specifies the storage scheme for the symmetric
matrix A'*A, as follows:
= 'F' : full storage is used;
= 'P' : packed storage is used.
UPLO (input) CHARACTER*1
Specifies which part of the matrix A'*A is stored,
as follows:
= 'U' : the upper triagular part is stored;
= 'L' : the lower triagular part is stored.
N (input) INTEGER
The order of the matrix A'*A. N >= 0.
IPAR (input) INTEGER array, dimension (LIPAR)
The integer parameters describing the structure of
the matrix A.
LIPAR (input) INTEGER
The length of the array IPAR. LIPAR >= 0.
DPAR (input) DOUBLE PRECISION array, dimension (LDPAR)
The real parameters needed for solving the
problem.
LDPAR (input) INTEGER
The length of the array DPAR. LDPAR >= 0.
A (input) DOUBLE PRECISION array, dimension
(LDA, NC), where NC is the number of columns.
The leading NR-by-NC part of this array must
contain the (compressed) representation of the
matrix A, where NR is the number of rows of A
(function of IPAR entries).
LDA (input) INTEGER
The leading dimension of the array A.
LDA >= MAX(1,NR).
ATA (output) DOUBLE PRECISION array,
dimension (LDATA,N), if STOR = 'F',
dimension (N*(N+1)/2), if STOR = 'P'.
The leading N-by-N (if STOR = 'F'), or N*(N+1)/2
(if STOR = 'P') part of this array contains the
upper or lower triangle of the matrix A'*A,
depending on UPLO = 'U', or UPLO = 'L',
respectively, stored either as a two-dimensional,
or one-dimensional array, depending on STOR.
LDATA (input) INTEGER
The leading dimension of the array ATA.
LDATA >= MAX(1,N), if STOR = 'F'.
LDATA >= 1, if STOR = 'P'.
DWORK DOUBLE PRECISION array, dimension (LDWORK)
The workspace array for subroutine F.
LDWORK (input) INTEGER
The size of the array DWORK (as large as needed
in the subroutine F).
INFO INTEGER
Error indicator, set to a negative value if an
input scalar argument is erroneous, and to
positive values for other possible errors in the
subroutine F. The LAPACK Library routine XERBLA
should be used in conjunction with negative INFO.
INFO must be zero if the subroutine finished
successfully.
Parameters marked with "(input)" must not be changed.
Input/Output Parameters
M (input) INTEGER
The number of rows of the matrix A. M >= 0.
N (input) INTEGER
The order of the matrix A'*A, the number of columns of the
matrix A, and the number of rows of the matrix X. N >= 0.
NRHS (input) INTEGER
The number of columns of the matrices B and X. NRHS >= 0.
IPAR (input) INTEGER array, dimension (LIPAR)
If FORM = 'F', the integer parameters describing the
structure of the matrix A.
This parameter is ignored if FORM = 'S'.
LIPAR (input) INTEGER
The length of the array IPAR. LIPAR >= 0.
DPAR (input) DOUBLE PRECISION array, dimension (LDPAR)
If FORM = 'F', the real parameters needed for solving
the problem.
This parameter is ignored if FORM = 'S'.
LDPAR (input) INTEGER
The length of the array DPAR. LDPAR >= 0.
A (input) DOUBLE PRECISION array,
dimension (LDA, N), if FORM = 'S',
dimension (LDA, NC), if FORM = 'F', where NC is
the number of columns.
If FORM = 'S', the leading M-by-N part of this array
must contain the matrix A.
If FORM = 'F', the leading NR-by-NC part of this array
must contain an appropriate representation of matrix A,
where NR is the number of rows.
If FORM = 'F', this array is not referenced by this
routine itself, except in the call to the routine F.
LDA INTEGER
The leading dimension of array A.
LDA >= MAX(1,M), if FORM = 'S';
LDA >= MAX(1,NR), if FORM = 'F'.
B (input/output) DOUBLE PRECISION array, dimension
(LDB, NRHS)
On entry, the leading N-by-NRHS part of this array must
contain the right hand side matrix B.
On exit, if INFO = 0 and M (or NR) is nonzero, the leading
N-by-NRHS part of this array contains the solution X of
the set of systems of linear equations A'*A*X = B or
f(A)*X = B. If M (or NR) is zero, then B is unchanged.
LDB INTEGER
The leading dimension of array B. LDB >= MAX(1,N).
ATA (output) DOUBLE PRECISION array,
dimension (LDATA,N), if STOR = 'F',
dimension (N*(N+1)/2), if STOR = 'P'.
The leading N-by-N (if STOR = 'F'), or N*(N+1)/2 (if
STOR = 'P') part of this array contains the upper or lower
triangular Cholesky factor of the matrix A'*A, depending
on UPLO = 'U', or UPLO = 'L', respectively, stored either
as a two-dimensional, or one-dimensional array, depending
on STOR.
LDATA INTEGER
The leading dimension of the array ATA.
LDATA >= MAX(1,N), if STOR = 'F'.
LDATA >= 1, if STOR = 'P'.
Workspace
DWORK DOUBLE PRECISION array, dimension (LDWORK)
LDWORK INTEGER
The length of the array DWORK.
Error Indicator
INFO INTEGER
= 0: successful exit;
< 0: if INFO = -i, the i-th argument had an illegal
value;
> 0: if INFO = i, then the (i,i) element of the
triangular factor of the matrix A'*A is exactly
zero (the matrix A'*A is exactly singular);
if INFO = j > n, then F returned with INFO = j-n.
Method
The matrix A'*A is built either directly (if FORM = 'S'), or implicitly, by calling the routine F. Then, A'*A is Cholesky factored and its factor is used to solve the set of systems of linear equations, A'*A*X = B.References
[1] Golub, G.H. and van Loan, C.F.
Matrix Computations. Third Edition.
M. D. Johns Hopkins University Press, Baltimore, 1996.
[2] Anderson, E., Bai, Z., Bischof, C., Blackford, Demmel, J.,
Dongarra, J., Du Croz, J., Greenbaum, A., Hammarling, S.,
McKenney, A., Sorensen, D.
LAPACK Users' Guide: Third Edition, SIAM, Philadelphia, 1999.
Numerical Aspects
For speed, this routine does not check for near singularity of the
matrix A'*A. If the matrix A is nearly rank deficient, then the
computed X could be inaccurate. Estimates of the reciprocal
condition numbers of the matrices A and A'*A can be obtained
using LAPACK routines DGECON and DPOCON (DPPCON), respectively.
The approximate number of floating point operations is
(M+3)*N**2/2 + N**3/6 + NRHS*N**2, if FORM = 'S',
f + N**3/6 + NRHS*N**2, if FORM = 'F',
where M is the number of rows of A, and f is the number of
floating point operations required by the subroutine F.
Further Comments
NoneExample
Program Text
NoneProgram Data
NoneProgram Results
None