F07BPF (ZGBSVX) (PDF version)
F07 Chapter Contents
F07 Chapter Introduction
NAG Library Manual

NAG Library Routine Document

F07BPF (ZGBSVX)

Note:  before using this routine, please read the Users' Note for your implementation to check the interpretation of bold italicised terms and other implementation-dependent details.

+ Contents

    1  Purpose
    7  Accuracy

1  Purpose

F07BPF (ZGBSVX) uses the LU factorization to compute the solution to a complex system of linear equations
AX=B ,  ATX=B   or   AHX=B ,
where A is an n by n band matrix with kl subdiagonals and ku superdiagonals, and X and B are n by r matrices. Error bounds on the solution and a condition estimate are also provided.

2  Specification

SUBROUTINE F07BPF ( FACT, TRANS, N, KL, KU, NRHS, AB, LDAB, AFB, LDAFB, IPIV, EQUED, R, C, B, LDB, X, LDX, RCOND, FERR, BERR, WORK, RWORK, INFO)
INTEGER  N, KL, KU, NRHS, LDAB, LDAFB, IPIV(*), LDB, LDX, INFO
REAL (KIND=nag_wp)  R(*), C(*), RCOND, FERR(NRHS), BERR(NRHS), RWORK(max(1,N))
COMPLEX (KIND=nag_wp)  AB(LDAB,*), AFB(LDAFB,*), B(LDB,*), X(LDX,*), WORK(2*N)
CHARACTER(1)  FACT, TRANS, EQUED
The routine may be called by its LAPACK name zgbsvx.

3  Description

F07BPF (ZGBSVX) performs the following steps:
  1. Equilibration
    The linear system to be solved may be badly scaled. However, the system can be equilibrated as a first stage by setting FACT='E'. In this case, real scaling factors are computed and these factors then determine whether the system is to be equilibrated. Equilibrated forms of the systems AX=B , ATX=B  and AHX=B  are
    DR A DC DC-1X = DR B ,
    DR A DC T DR-1 X = DC B ,
    and
    DR A DC H DR-1 X = DC B ,
    respectively, where DR  and DC  are diagonal matrices, with positive diagonal elements, formed from the computed scaling factors.
    When equilibration is used, A will be overwritten by DR A DC  and B will be overwritten by DR B  (or DC B  when the solution of ATX=B  or AHX=B  is sought).
  2. Factorization
    The matrix A, or its scaled form, is copied and factored using the LU decomposition
    A=PLU ,
    where P is a permutation matrix, L is a unit lower triangular matrix, and U is upper triangular.
    This stage can be by-passed when a factored matrix (with scaled matrices and scaling factors) are supplied; for example, as provided by a previous call to F07BPF (ZGBSVX) with the same matrix A.
  3. Condition Number Estimation
    The LU factorization of A determines whether a solution to the linear system exists. If some diagonal element of U is zero, then U is exactly singular, no solution exists and the routine returns with a failure. Otherwise the factorized form of A is used to estimate the condition number of the matrix A. If the reciprocal of the condition number is less than machine precision then a warning code is returned on final exit.
  4. Solution
    The (equilibrated) system is solved for X ( DC-1X  or DR-1X ) using the factored form of A ( DRADC ).
  5. Iterative Refinement
    Iterative refinement is applied to improve the computed solution matrix and to calculate error bounds and backward error estimates for the computed solution.
  6. Construct Solution Matrix X
    If equilibration was used, the matrix X is premultiplied by DC  (if TRANS='N') or DR  (if TRANS='T' or 'C') so that it solves the original system before equilibration.

4  References

Anderson E, Bai Z, Bischof C, Blackford S, Demmel J, Dongarra J J, Du Croz J J, Greenbaum A, Hammarling S, McKenney A and Sorensen D (1999) LAPACK Users' Guide (3rd Edition) SIAM, Philadelphia http://www.netlib.org/lapack/lug
Golub G H and Van Loan C F (1996) Matrix Computations (3rd Edition) Johns Hopkins University Press, Baltimore
Higham N J (2002) Accuracy and Stability of Numerical Algorithms (2nd Edition) SIAM, Philadelphia

5  Parameters

1:     FACT – CHARACTER(1)Input
On entry: specifies whether or not the factorized form of the matrix A is supplied on entry, and if not, whether the matrix A should be equilibrated before it is factorized.
FACT='F'
AFB and IPIV contain the factorized form of A. If EQUED'N', the matrix A has been equilibrated with scaling factors given by R and C. AB, AFB and IPIV are not modified.
FACT='N'
The matrix A will be copied to AFB and factorized.
FACT='E'
The matrix A will be equilibrated if necessary, then copied to AFB and factorized.
Constraint: FACT='F', 'N' or 'E'.
2:     TRANS – CHARACTER(1)Input
On entry: specifies the form of the system of equations.
TRANS='N'
AX=B (No transpose).
TRANS='T'
ATX=B (Transpose).
TRANS='C'
AHX=B (Conjugate transpose).
Constraint: TRANS='N', 'T' or 'C'.
3:     N – INTEGERInput
On entry: n, the number of linear equations, i.e., the order of the matrix A.
Constraint: N0.
4:     KL – INTEGERInput
On entry: kl, the number of subdiagonals within the band of the matrix A.
Constraint: KL0.
5:     KU – INTEGERInput
On entry: ku, the number of superdiagonals within the band of the matrix A.
Constraint: KU0.
6:     NRHS – INTEGERInput
On entry: r, the number of right-hand sides, i.e., the number of columns of the matrix B.
Constraint: NRHS0.
7:     AB(LDAB,*) – COMPLEX (KIND=nag_wp) arrayInput/Output
Note: the second dimension of the array AB must be at least max1,N.
On entry: the n by n coefficient matrix A.
The matrix is stored in rows 1 to kl+ku+1, more precisely, the element Aij must be stored in
ABku+1+i-jj  for ​max1,j-kuiminn,j+kl.
See Section 8 for further details.
If FACT='F' and EQUED'N', A must have been equilibrated by the scaling factors in R and/or C.
On exit: if FACT='F' or 'N', or if FACT='E' and EQUED='N', AB is not modified.
If EQUED'N' then, if no constraints are violated, A is scaled as follows:
  • if EQUED='R', A=DrA;
  • if EQUED='C', A=ADc;
  • if EQUED='B', A=DrADc.
8:     LDAB – INTEGERInput
On entry: the first dimension of the array AB as declared in the (sub)program from which F07BPF (ZGBSVX) is called.
Constraint: LDABKL+KU+1.
9:     AFB(LDAFB,*) – COMPLEX (KIND=nag_wp) arrayInput/Output
Note: the second dimension of the array AFB must be at least max1,N.
On entry: if FACT='N' or 'E', AFB need not be set.
If FACT='F', details of the LU factorization of the n by n band matrix A, as computed by F07BRF (ZGBTRF).
The upper triangular band matrix U, with kl+ku superdiagonals, is stored in rows 1 to kl+ku+1 of the array, and the multipliers used to form the matrix L are stored in rows kl+ku+2 to 2kl+ku+1.
If EQUED'N', AFB is the factorized form of the equilibrated matrix A.
On exit: if FACT='F', AFB is unchanged from entry.
Otherwise, if no constraints are violated, then if FACT='N', AFB returns details of the LU factorization of the band matrix A, and if FACT='E', AFB returns details of the LU factorization of the equilibrated band matrix A (see the description of AB for the form of the equilibrated matrix).
10:   LDAFB – INTEGERInput
On entry: the first dimension of the array AFB as declared in the (sub)program from which F07BPF (ZGBSVX) is called.
Constraint: LDAFB2×KL+KU+1.
11:   IPIV(*) – INTEGER arrayInput/Output
Note: the dimension of the array IPIV must be at least max1,N.
On entry: if FACT='N' or 'E', IPIV need not be set.
If FACT='F', IPIV contains the pivot indices from the factorization A=LU, as computed by F07BDF (DGBTRF); row i of the matrix was interchanged with row IPIVi.
On exit: if FACT='F', IPIV is unchanged from entry.
Otherwise, if no constraints are violated, IPIV contains the pivot indices that define the permutation matrix P; at the ith step row i of the matrix was interchanged with row IPIVi. IPIVi=i indicates a row interchange was not required.
If FACT='N', the pivot indices are those corresponding to the factorization A=LU of the original matrix A.
If FACT='E', the pivot indices are those corresponding to the factorization of A=LU of the equilibrated matrix A.
12:   EQUED – CHARACTER(1)Input/Output
On entry: if FACT='N' or 'E', EQUED need not be set.
If FACT='F', EQUED must specify the form of the equilibration that was performed as follows:
  • if EQUED='N', no equilibration;
  • if EQUED='R', row equilibration, i.e., A has been premultiplied by DR;
  • if EQUED='C', column equilibration, i.e., A has been postmultiplied by DC;
  • if EQUED='B', both row and column equilibration, i.e., A has been replaced by DRADC.
On exit: if FACT='F', EQUED is unchanged from entry.
Otherwise, if no constraints are violated, EQUED specifies the form of equilibration that was performed as specified above.
Constraint: if FACT='F', EQUED='N', 'R', 'C' or 'B'.
13:   R(*) – REAL (KIND=nag_wp) arrayInput/Output
Note: the dimension of the array R must be at least max1,N.
On entry: if FACT='N' or 'E', R need not be set.
If FACT='F' and EQUED='R' or 'B', R must contain the row scale factors for A, DR; each element of R must be positive.
On exit: if FACT='F', R is unchanged from entry.
Otherwise, if no constraints are violated and EQUED='R' or 'B', R contains the row scale factors for A, DR, such that A is multiplied on the left by DR; each element of R is positive.
14:   C(*) – REAL (KIND=nag_wp) arrayInput/Output
Note: the dimension of the array C must be at least max1,N.
On entry: if FACT='N' or 'E', C need not be set.
If FACT='F' or EQUED='C' or 'B', C must contain the column scale factors for A, DC; each element of C must be positive.
On exit: if FACT='F', C is unchanged from entry.
Otherwise, if no constraints are violated and EQUED='C' or 'B', C contains the row scale factors for A, DC; each element of C is positive.
15:   B(LDB,*) – COMPLEX (KIND=nag_wp) arrayInput/Output
Note: the second dimension of the array B must be at least max1,NRHS.
On entry: the n by r right-hand side matrix B.
On exit: if EQUED='N', B is not modified.
If TRANS='N' and EQUED='R' or 'B', B is overwritten by DRB.
If TRANS='T' or 'C' and EQUED='C' or 'B', B is overwritten by DCB.
16:   LDB – INTEGERInput
On entry: the first dimension of the array B as declared in the (sub)program from which F07BPF (ZGBSVX) is called.
Constraint: LDBmax1,N.
17:   X(LDX,*) – COMPLEX (KIND=nag_wp) arrayOutput
Note: the second dimension of the array X must be at least max1,NRHS.
On exit: if INFO=0 or N+1, the n by r solution matrix X to the original system of equations. Note that the arrays A and B are modified on exit if EQUED'N', and the solution to the equilibrated system is DC-1X if TRANS='N' and EQUED='C' or 'B', or DR-1X if TRANS='T' or 'C' and EQUED='R' or 'B'.
18:   LDX – INTEGERInput
On entry: the first dimension of the array X as declared in the (sub)program from which F07BPF (ZGBSVX) is called.
Constraint: LDXmax1,N.
19:   RCOND – REAL (KIND=nag_wp)Output
On exit: if no constraints are violated, an estimate of the reciprocal condition number of the matrix A (after equilibration if that is performed), computed as RCOND=1.0/A1 A-11 .
20:   FERR(NRHS) – REAL (KIND=nag_wp) arrayOutput
On exit: if INFO=0 or N+1, an estimate of the forward error bound for each computed solution vector, such that x^j-xj/xjFERRj where x^j is the jth column of the computed solution returned in the array X and xj is the corresponding column of the exact solution X. The estimate is as reliable as the estimate for RCOND, and is almost always a slight overestimate of the true error.
21:   BERR(NRHS) – REAL (KIND=nag_wp) arrayOutput
On exit: if INFO=0 or N+1, an estimate of the component-wise relative backward error of each computed solution vector x^j (i.e., the smallest relative change in any element of A or B that makes x^j an exact solution).
22:   WORK(2×N) – COMPLEX (KIND=nag_wp) arrayWorkspace
23:   RWORK(max1,N) – REAL (KIND=nag_wp) arrayOutput
On exit: if INFO=0, RWORK1 contains the reciprocal pivot growth factor maxaij/maxuij. If RWORK1 is much less than 1, then the stability of the LU factorization of the (equilibrated) matrix A could be poor. This also means that the solution X, condition estimator RCOND, and forward error bound FERR could be unreliable. If the factorization fails with INFO>0 and INFON, RWORK1 contains the reciprocal pivot growth factor for the leading INFO columns of A.
24:   INFO – INTEGEROutput
On exit: INFO=0 unless the routine detects an error (see Section 6).

6  Error Indicators and Warnings

Errors or warnings detected by the routine:
INFO<0
If INFO=-i, the ith argument had an illegal value. An explanatory message is output, and execution of the program is terminated.
INFO>0 and INFON
If INFO=i, uii is exactly zero. The factorization has been completed, but the factor U is exactly singular, so the solution and error bounds could not be computed. RCOND=0.0 is returned.
INFO=N+1
The triangular matrix U is nonsingular, but RCOND is less than machine precision, meaning that the matrix is singular to working precision. Nevertheless, the solution and error bounds are computed because there are a number of situations where the computed solution can be more accurate than the value of RCOND would suggest.

7  Accuracy

For each right-hand side vector b, the computed solution x^ is the exact solution of a perturbed system of equations A+Ex^=b, where
EcnεPLU ,
cn is a modest linear function of n, and ε is the machine precision. See Section 9.3 of Higham (2002) for further details.
If x is the true solution, then the computed solution x^ satisfies a forward error bound of the form
x-x^ x^ wc condA,x^,b
where condA,x^,b = A-1 A x^ + b / x^ condA = A-1 A κ A. If x^  is the j th column of X , then wc  is returned in BERRj  and a bound on x - x^ / x^  is returned in FERRj . See Section 4.4 of Anderson et al. (1999) for further details.

8  Further Comments

The band storage scheme for the array AB is illustrated by the following example, when n=6 , kl=1 , and ku=2 . Storage of the band matrix A  in the array AB:
* * a13 a24 a35 a46 * a12 a23 a34 a45 a56 a11 a22 a33 a44 a55 a66 a21 a32 a43 a54 a65 *
The total number of floating point operations required to solve the equations AX=B  depends upon the pivoting required, but if nkl + ku  then it is approximately bounded by O n kl kl + ku  for the factorization and O n 2 kl + ku r  for the solution following the factorization. The condition number estimation typically requires between four and five solves and never more than eleven solves, following the factorization. The solution is then refined, and the errors estimated, using iterative refinement; see F07BVF (ZGBRFS) for information on the floating point operations required.
In practice the condition number estimator is very reliable, but it can underestimate the true condition number; see Section 15.3 of Higham (2002) for further details.
The real analogue of this routine is F07BBF (DGBSVX).

9  Example

This example solves the equations
AX=B ,
where A  is the band matrix
A = -1.65+2.26i -2.05-0.85i 0.97-2.84i 0.00i+0.00 6.30i -1.48-1.75i -3.99+4.01i 0.59-0.48i 0.00i+0.00 -0.77+2.83i -1.06+1.94i 3.33-1.04i 0.00i+0.00 0.00i+0.00 4.48-1.09i -0.46-1.72i
and
B = -1.06+21.50i 12.85+02.84i -22.72-53.90i -70.22+21.57i 28.24-38.60i -20.73-01.23i -34.56+16.73i 26.01+31.97i .
Estimates for the backward errors, forward errors, condition number and pivot growth are also output, together with information on the equilibration of A .

9.1  Program Text

Program Text (f07bpfe.f90)

9.2  Program Data

Program Data (f07bpfe.d)

9.3  Program Results

Program Results (f07bpfe.r)


F07BPF (ZGBSVX) (PDF version)
F07 Chapter Contents
F07 Chapter Introduction
NAG Library Manual

© The Numerical Algorithms Group Ltd, Oxford, UK. 2012