1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
|
SUBROUTINE SPTT05( N, NRHS, D, E, B, LDB, X, LDX, XACT, LDXACT,
$ FERR, BERR, RESLTS )
*
* -- LAPACK test routine (version 3.0) --
* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
* Courant Institute, Argonne National Lab, and Rice University
* February 29, 1992
*
* .. Scalar Arguments ..
INTEGER LDB, LDX, LDXACT, N, NRHS
* ..
* .. Array Arguments ..
REAL B( LDB, * ), BERR( * ), D( * ), E( * ),
$ FERR( * ), RESLTS( * ), X( LDX, * ),
$ XACT( LDXACT, * )
* ..
*
* Purpose
* =======
*
* SPTT05 tests the error bounds from iterative refinement for the
* computed solution to a system of equations A*X = B, where A is a
* symmetric tridiagonal matrix of order n.
*
* RESLTS(1) = test of the error bound
* = norm(X - XACT) / ( norm(X) * FERR )
*
* A large value is returned if this ratio is not less than one.
*
* RESLTS(2) = residual from the iterative refinement routine
* = the maximum of BERR / ( NZ*EPS + (*) ), where
* (*) = NZ*UNFL / (min_i (abs(A)*abs(X) +abs(b))_i )
* and NZ = max. number of nonzeros in any row of A, plus 1
*
* Arguments
* =========
*
* N (input) INTEGER
* The number of rows of the matrices X, B, and XACT, and the
* order of the matrix A. N >= 0.
*
* NRHS (input) INTEGER
* The number of columns of the matrices X, B, and XACT.
* NRHS >= 0.
*
* D (input) REAL array, dimension (N)
* The n diagonal elements of the tridiagonal matrix A.
*
* E (input) REAL array, dimension (N-1)
* The (n-1) subdiagonal elements of the tridiagonal matrix A.
*
* B (input) REAL array, dimension (LDB,NRHS)
* The right hand side vectors for the system of linear
* equations.
*
* LDB (input) INTEGER
* The leading dimension of the array B. LDB >= max(1,N).
*
* X (input) REAL array, dimension (LDX,NRHS)
* The computed solution vectors. Each vector is stored as a
* column of the matrix X.
*
* LDX (input) INTEGER
* The leading dimension of the array X. LDX >= max(1,N).
*
* XACT (input) REAL array, dimension (LDX,NRHS)
* The exact solution vectors. Each vector is stored as a
* column of the matrix XACT.
*
* LDXACT (input) INTEGER
* The leading dimension of the array XACT. LDXACT >= max(1,N).
*
* FERR (input) REAL array, dimension (NRHS)
* The estimated forward error bounds for each solution vector
* X. If XTRUE is the true solution, FERR bounds the magnitude
* of the largest entry in (X - XTRUE) divided by the magnitude
* of the largest entry in X.
*
* BERR (input) REAL array, dimension (NRHS)
* The componentwise relative backward error of each solution
* vector (i.e., the smallest relative change in any entry of A
* or B that makes X an exact solution).
*
* RESLTS (output) REAL array, dimension (2)
* The maximum over the NRHS solution vectors of the ratios:
* RESLTS(1) = norm(X - XACT) / ( norm(X) * FERR )
* RESLTS(2) = BERR / ( NZ*EPS + (*) )
*
* =====================================================================
*
* .. Parameters ..
REAL ZERO, ONE
PARAMETER ( ZERO = 0.0E+0, ONE = 1.0E+0 )
* ..
* .. Local Scalars ..
INTEGER I, IMAX, J, K, NZ
REAL AXBI, DIFF, EPS, ERRBND, OVFL, TMP, UNFL, XNORM
* ..
* .. External Functions ..
INTEGER ISAMAX
REAL SLAMCH
EXTERNAL ISAMAX, SLAMCH
* ..
* .. Intrinsic Functions ..
INTRINSIC ABS, MAX, MIN
* ..
* .. Executable Statements ..
*
* Quick exit if N = 0 or NRHS = 0.
*
IF( N.LE.0 .OR. NRHS.LE.0 ) THEN
RESLTS( 1 ) = ZERO
RESLTS( 2 ) = ZERO
RETURN
END IF
*
EPS = SLAMCH( 'Epsilon' )
UNFL = SLAMCH( 'Safe minimum' )
OVFL = ONE / UNFL
NZ = 4
*
* Test 1: Compute the maximum of
* norm(X - XACT) / ( norm(X) * FERR )
* over all the vectors X and XACT using the infinity-norm.
*
ERRBND = ZERO
DO 30 J = 1, NRHS
IMAX = ISAMAX( N, X( 1, J ), 1 )
XNORM = MAX( ABS( X( IMAX, J ) ), UNFL )
DIFF = ZERO
DO 10 I = 1, N
DIFF = MAX( DIFF, ABS( X( I, J )-XACT( I, J ) ) )
10 CONTINUE
*
IF( XNORM.GT.ONE ) THEN
GO TO 20
ELSE IF( DIFF.LE.OVFL*XNORM ) THEN
GO TO 20
ELSE
ERRBND = ONE / EPS
GO TO 30
END IF
*
20 CONTINUE
IF( DIFF / XNORM.LE.FERR( J ) ) THEN
ERRBND = MAX( ERRBND, ( DIFF / XNORM ) / FERR( J ) )
ELSE
ERRBND = ONE / EPS
END IF
30 CONTINUE
RESLTS( 1 ) = ERRBND
*
* Test 2: Compute the maximum of BERR / ( NZ*EPS + (*) ), where
* (*) = NZ*UNFL / (min_i (abs(A)*abs(X) +abs(b))_i )
*
DO 50 K = 1, NRHS
IF( N.EQ.1 ) THEN
AXBI = ABS( B( 1, K ) ) + ABS( D( 1 )*X( 1, K ) )
ELSE
AXBI = ABS( B( 1, K ) ) + ABS( D( 1 )*X( 1, K ) ) +
$ ABS( E( 1 )*X( 2, K ) )
DO 40 I = 2, N - 1
TMP = ABS( B( I, K ) ) + ABS( E( I-1 )*X( I-1, K ) ) +
$ ABS( D( I )*X( I, K ) ) + ABS( E( I )*X( I+1, K ) )
AXBI = MIN( AXBI, TMP )
40 CONTINUE
TMP = ABS( B( N, K ) ) + ABS( E( N-1 )*X( N-1, K ) ) +
$ ABS( D( N )*X( N, K ) )
AXBI = MIN( AXBI, TMP )
END IF
TMP = BERR( K ) / ( NZ*EPS+NZ*UNFL / MAX( AXBI, NZ*UNFL ) )
IF( K.EQ.1 ) THEN
RESLTS( 2 ) = TMP
ELSE
RESLTS( 2 ) = MAX( RESLTS( 2 ), TMP )
END IF
50 CONTINUE
*
RETURN
*
* End of SPTT05
*
END
|
