Annotation of rpl/lapack/lapack/dtbrfs.f, revision 1.18
1.9 bertrand 1: *> \brief \b DTBRFS
2: *
3: * =========== DOCUMENTATION ===========
4: *
1.15 bertrand 5: * Online html documentation available at
6: * http://www.netlib.org/lapack/explore-html/
1.9 bertrand 7: *
8: *> \htmlonly
1.15 bertrand 9: *> Download DTBRFS + dependencies
10: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dtbrfs.f">
11: *> [TGZ]</a>
12: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dtbrfs.f">
13: *> [ZIP]</a>
14: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dtbrfs.f">
1.9 bertrand 15: *> [TXT]</a>
1.15 bertrand 16: *> \endhtmlonly
1.9 bertrand 17: *
18: * Definition:
19: * ===========
20: *
21: * SUBROUTINE DTBRFS( UPLO, TRANS, DIAG, N, KD, NRHS, AB, LDAB, B,
22: * LDB, X, LDX, FERR, BERR, WORK, IWORK, INFO )
1.15 bertrand 23: *
1.9 bertrand 24: * .. Scalar Arguments ..
25: * CHARACTER DIAG, TRANS, UPLO
26: * INTEGER INFO, KD, LDAB, LDB, LDX, N, NRHS
27: * ..
28: * .. Array Arguments ..
29: * INTEGER IWORK( * )
30: * DOUBLE PRECISION AB( LDAB, * ), B( LDB, * ), BERR( * ),
31: * $ FERR( * ), WORK( * ), X( LDX, * )
32: * ..
1.15 bertrand 33: *
1.9 bertrand 34: *
35: *> \par Purpose:
36: * =============
37: *>
38: *> \verbatim
39: *>
40: *> DTBRFS provides error bounds and backward error estimates for the
41: *> solution to a system of linear equations with a triangular band
42: *> coefficient matrix.
43: *>
44: *> The solution matrix X must be computed by DTBTRS or some other
45: *> means before entering this routine. DTBRFS does not do iterative
46: *> refinement because doing so cannot improve the backward error.
47: *> \endverbatim
48: *
49: * Arguments:
50: * ==========
51: *
52: *> \param[in] UPLO
53: *> \verbatim
54: *> UPLO is CHARACTER*1
55: *> = 'U': A is upper triangular;
56: *> = 'L': A is lower triangular.
57: *> \endverbatim
58: *>
59: *> \param[in] TRANS
60: *> \verbatim
61: *> TRANS is CHARACTER*1
62: *> Specifies the form of the system of equations:
63: *> = 'N': A * X = B (No transpose)
64: *> = 'T': A**T * X = B (Transpose)
65: *> = 'C': A**H * X = B (Conjugate transpose = Transpose)
66: *> \endverbatim
67: *>
68: *> \param[in] DIAG
69: *> \verbatim
70: *> DIAG is CHARACTER*1
71: *> = 'N': A is non-unit triangular;
72: *> = 'U': A is unit triangular.
73: *> \endverbatim
74: *>
75: *> \param[in] N
76: *> \verbatim
77: *> N is INTEGER
78: *> The order of the matrix A. N >= 0.
79: *> \endverbatim
80: *>
81: *> \param[in] KD
82: *> \verbatim
83: *> KD is INTEGER
84: *> The number of superdiagonals or subdiagonals of the
85: *> triangular band matrix A. KD >= 0.
86: *> \endverbatim
87: *>
88: *> \param[in] NRHS
89: *> \verbatim
90: *> NRHS is INTEGER
91: *> The number of right hand sides, i.e., the number of columns
92: *> of the matrices B and X. NRHS >= 0.
93: *> \endverbatim
94: *>
95: *> \param[in] AB
96: *> \verbatim
97: *> AB is DOUBLE PRECISION array, dimension (LDAB,N)
98: *> The upper or lower triangular band matrix A, stored in the
99: *> first kd+1 rows of the array. The j-th column of A is stored
100: *> in the j-th column of the array AB as follows:
101: *> if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j;
102: *> if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd).
103: *> If DIAG = 'U', the diagonal elements of A are not referenced
104: *> and are assumed to be 1.
105: *> \endverbatim
106: *>
107: *> \param[in] LDAB
108: *> \verbatim
109: *> LDAB is INTEGER
110: *> The leading dimension of the array AB. LDAB >= KD+1.
111: *> \endverbatim
112: *>
113: *> \param[in] B
114: *> \verbatim
115: *> B is DOUBLE PRECISION array, dimension (LDB,NRHS)
116: *> The right hand side matrix B.
117: *> \endverbatim
118: *>
119: *> \param[in] LDB
120: *> \verbatim
121: *> LDB is INTEGER
122: *> The leading dimension of the array B. LDB >= max(1,N).
123: *> \endverbatim
124: *>
125: *> \param[in] X
126: *> \verbatim
127: *> X is DOUBLE PRECISION array, dimension (LDX,NRHS)
128: *> The solution matrix X.
129: *> \endverbatim
130: *>
131: *> \param[in] LDX
132: *> \verbatim
133: *> LDX is INTEGER
134: *> The leading dimension of the array X. LDX >= max(1,N).
135: *> \endverbatim
136: *>
137: *> \param[out] FERR
138: *> \verbatim
139: *> FERR is DOUBLE PRECISION array, dimension (NRHS)
140: *> The estimated forward error bound for each solution vector
141: *> X(j) (the j-th column of the solution matrix X).
142: *> If XTRUE is the true solution corresponding to X(j), FERR(j)
143: *> is an estimated upper bound for the magnitude of the largest
144: *> element in (X(j) - XTRUE) divided by the magnitude of the
145: *> largest element in X(j). The estimate is as reliable as
146: *> the estimate for RCOND, and is almost always a slight
147: *> overestimate of the true error.
148: *> \endverbatim
149: *>
150: *> \param[out] BERR
151: *> \verbatim
152: *> BERR is DOUBLE PRECISION array, dimension (NRHS)
153: *> The componentwise relative backward error of each solution
154: *> vector X(j) (i.e., the smallest relative change in
155: *> any element of A or B that makes X(j) an exact solution).
156: *> \endverbatim
157: *>
158: *> \param[out] WORK
159: *> \verbatim
160: *> WORK is DOUBLE PRECISION array, dimension (3*N)
161: *> \endverbatim
162: *>
163: *> \param[out] IWORK
164: *> \verbatim
165: *> IWORK is INTEGER array, dimension (N)
166: *> \endverbatim
167: *>
168: *> \param[out] INFO
169: *> \verbatim
170: *> INFO is INTEGER
171: *> = 0: successful exit
172: *> < 0: if INFO = -i, the i-th argument had an illegal value
173: *> \endverbatim
174: *
175: * Authors:
176: * ========
177: *
1.15 bertrand 178: *> \author Univ. of Tennessee
179: *> \author Univ. of California Berkeley
180: *> \author Univ. of Colorado Denver
181: *> \author NAG Ltd.
1.9 bertrand 182: *
183: *> \ingroup doubleOTHERcomputational
184: *
185: * =====================================================================
1.1 bertrand 186: SUBROUTINE DTBRFS( UPLO, TRANS, DIAG, N, KD, NRHS, AB, LDAB, B,
187: $ LDB, X, LDX, FERR, BERR, WORK, IWORK, INFO )
188: *
1.18 ! bertrand 189: * -- LAPACK computational routine --
1.1 bertrand 190: * -- LAPACK is a software package provided by Univ. of Tennessee, --
191: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
192: *
193: * .. Scalar Arguments ..
194: CHARACTER DIAG, TRANS, UPLO
195: INTEGER INFO, KD, LDAB, LDB, LDX, N, NRHS
196: * ..
197: * .. Array Arguments ..
198: INTEGER IWORK( * )
199: DOUBLE PRECISION AB( LDAB, * ), B( LDB, * ), BERR( * ),
200: $ FERR( * ), WORK( * ), X( LDX, * )
201: * ..
202: *
203: * =====================================================================
204: *
205: * .. Parameters ..
206: DOUBLE PRECISION ZERO
207: PARAMETER ( ZERO = 0.0D+0 )
208: DOUBLE PRECISION ONE
209: PARAMETER ( ONE = 1.0D+0 )
210: * ..
211: * .. Local Scalars ..
212: LOGICAL NOTRAN, NOUNIT, UPPER
213: CHARACTER TRANST
214: INTEGER I, J, K, KASE, NZ
215: DOUBLE PRECISION EPS, LSTRES, S, SAFE1, SAFE2, SAFMIN, XK
216: * ..
217: * .. Local Arrays ..
218: INTEGER ISAVE( 3 )
219: * ..
220: * .. External Subroutines ..
221: EXTERNAL DAXPY, DCOPY, DLACN2, DTBMV, DTBSV, XERBLA
222: * ..
223: * .. Intrinsic Functions ..
224: INTRINSIC ABS, MAX, MIN
225: * ..
226: * .. External Functions ..
227: LOGICAL LSAME
228: DOUBLE PRECISION DLAMCH
229: EXTERNAL LSAME, DLAMCH
230: * ..
231: * .. Executable Statements ..
232: *
233: * Test the input parameters.
234: *
235: INFO = 0
236: UPPER = LSAME( UPLO, 'U' )
237: NOTRAN = LSAME( TRANS, 'N' )
238: NOUNIT = LSAME( DIAG, 'N' )
239: *
240: IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN
241: INFO = -1
242: ELSE IF( .NOT.NOTRAN .AND. .NOT.LSAME( TRANS, 'T' ) .AND. .NOT.
243: $ LSAME( TRANS, 'C' ) ) THEN
244: INFO = -2
245: ELSE IF( .NOT.NOUNIT .AND. .NOT.LSAME( DIAG, 'U' ) ) THEN
246: INFO = -3
247: ELSE IF( N.LT.0 ) THEN
248: INFO = -4
249: ELSE IF( KD.LT.0 ) THEN
250: INFO = -5
251: ELSE IF( NRHS.LT.0 ) THEN
252: INFO = -6
253: ELSE IF( LDAB.LT.KD+1 ) THEN
254: INFO = -8
255: ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
256: INFO = -10
257: ELSE IF( LDX.LT.MAX( 1, N ) ) THEN
258: INFO = -12
259: END IF
260: IF( INFO.NE.0 ) THEN
261: CALL XERBLA( 'DTBRFS', -INFO )
262: RETURN
263: END IF
264: *
265: * Quick return if possible
266: *
267: IF( N.EQ.0 .OR. NRHS.EQ.0 ) THEN
268: DO 10 J = 1, NRHS
269: FERR( J ) = ZERO
270: BERR( J ) = ZERO
271: 10 CONTINUE
272: RETURN
273: END IF
274: *
275: IF( NOTRAN ) THEN
276: TRANST = 'T'
277: ELSE
278: TRANST = 'N'
279: END IF
280: *
281: * NZ = maximum number of nonzero elements in each row of A, plus 1
282: *
283: NZ = KD + 2
284: EPS = DLAMCH( 'Epsilon' )
285: SAFMIN = DLAMCH( 'Safe minimum' )
286: SAFE1 = NZ*SAFMIN
287: SAFE2 = SAFE1 / EPS
288: *
289: * Do for each right hand side
290: *
291: DO 250 J = 1, NRHS
292: *
293: * Compute residual R = B - op(A) * X,
1.8 bertrand 294: * where op(A) = A or A**T, depending on TRANS.
1.1 bertrand 295: *
296: CALL DCOPY( N, X( 1, J ), 1, WORK( N+1 ), 1 )
297: CALL DTBMV( UPLO, TRANS, DIAG, N, KD, AB, LDAB, WORK( N+1 ),
298: $ 1 )
299: CALL DAXPY( N, -ONE, B( 1, J ), 1, WORK( N+1 ), 1 )
300: *
301: * Compute componentwise relative backward error from formula
302: *
303: * max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) )
304: *
305: * where abs(Z) is the componentwise absolute value of the matrix
306: * or vector Z. If the i-th component of the denominator is less
307: * than SAFE2, then SAFE1 is added to the i-th components of the
308: * numerator and denominator before dividing.
309: *
310: DO 20 I = 1, N
311: WORK( I ) = ABS( B( I, J ) )
312: 20 CONTINUE
313: *
314: IF( NOTRAN ) THEN
315: *
316: * Compute abs(A)*abs(X) + abs(B).
317: *
318: IF( UPPER ) THEN
319: IF( NOUNIT ) THEN
320: DO 40 K = 1, N
321: XK = ABS( X( K, J ) )
322: DO 30 I = MAX( 1, K-KD ), K
323: WORK( I ) = WORK( I ) +
324: $ ABS( AB( KD+1+I-K, K ) )*XK
325: 30 CONTINUE
326: 40 CONTINUE
327: ELSE
328: DO 60 K = 1, N
329: XK = ABS( X( K, J ) )
330: DO 50 I = MAX( 1, K-KD ), K - 1
331: WORK( I ) = WORK( I ) +
332: $ ABS( AB( KD+1+I-K, K ) )*XK
333: 50 CONTINUE
334: WORK( K ) = WORK( K ) + XK
335: 60 CONTINUE
336: END IF
337: ELSE
338: IF( NOUNIT ) THEN
339: DO 80 K = 1, N
340: XK = ABS( X( K, J ) )
341: DO 70 I = K, MIN( N, K+KD )
342: WORK( I ) = WORK( I ) + ABS( AB( 1+I-K, K ) )*XK
343: 70 CONTINUE
344: 80 CONTINUE
345: ELSE
346: DO 100 K = 1, N
347: XK = ABS( X( K, J ) )
348: DO 90 I = K + 1, MIN( N, K+KD )
349: WORK( I ) = WORK( I ) + ABS( AB( 1+I-K, K ) )*XK
350: 90 CONTINUE
351: WORK( K ) = WORK( K ) + XK
352: 100 CONTINUE
353: END IF
354: END IF
355: ELSE
356: *
1.8 bertrand 357: * Compute abs(A**T)*abs(X) + abs(B).
1.1 bertrand 358: *
359: IF( UPPER ) THEN
360: IF( NOUNIT ) THEN
361: DO 120 K = 1, N
362: S = ZERO
363: DO 110 I = MAX( 1, K-KD ), K
364: S = S + ABS( AB( KD+1+I-K, K ) )*
365: $ ABS( X( I, J ) )
366: 110 CONTINUE
367: WORK( K ) = WORK( K ) + S
368: 120 CONTINUE
369: ELSE
370: DO 140 K = 1, N
371: S = ABS( X( K, J ) )
372: DO 130 I = MAX( 1, K-KD ), K - 1
373: S = S + ABS( AB( KD+1+I-K, K ) )*
374: $ ABS( X( I, J ) )
375: 130 CONTINUE
376: WORK( K ) = WORK( K ) + S
377: 140 CONTINUE
378: END IF
379: ELSE
380: IF( NOUNIT ) THEN
381: DO 160 K = 1, N
382: S = ZERO
383: DO 150 I = K, MIN( N, K+KD )
384: S = S + ABS( AB( 1+I-K, K ) )*ABS( X( I, J ) )
385: 150 CONTINUE
386: WORK( K ) = WORK( K ) + S
387: 160 CONTINUE
388: ELSE
389: DO 180 K = 1, N
390: S = ABS( X( K, J ) )
391: DO 170 I = K + 1, MIN( N, K+KD )
392: S = S + ABS( AB( 1+I-K, K ) )*ABS( X( I, J ) )
393: 170 CONTINUE
394: WORK( K ) = WORK( K ) + S
395: 180 CONTINUE
396: END IF
397: END IF
398: END IF
399: S = ZERO
400: DO 190 I = 1, N
401: IF( WORK( I ).GT.SAFE2 ) THEN
402: S = MAX( S, ABS( WORK( N+I ) ) / WORK( I ) )
403: ELSE
404: S = MAX( S, ( ABS( WORK( N+I ) )+SAFE1 ) /
405: $ ( WORK( I )+SAFE1 ) )
406: END IF
407: 190 CONTINUE
408: BERR( J ) = S
409: *
410: * Bound error from formula
411: *
412: * norm(X - XTRUE) / norm(X) .le. FERR =
413: * norm( abs(inv(op(A)))*
414: * ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X)
415: *
416: * where
417: * norm(Z) is the magnitude of the largest component of Z
418: * inv(op(A)) is the inverse of op(A)
419: * abs(Z) is the componentwise absolute value of the matrix or
420: * vector Z
421: * NZ is the maximum number of nonzeros in any row of A, plus 1
422: * EPS is machine epsilon
423: *
424: * The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B))
425: * is incremented by SAFE1 if the i-th component of
426: * abs(op(A))*abs(X) + abs(B) is less than SAFE2.
427: *
428: * Use DLACN2 to estimate the infinity-norm of the matrix
429: * inv(op(A)) * diag(W),
430: * where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) )))
431: *
432: DO 200 I = 1, N
433: IF( WORK( I ).GT.SAFE2 ) THEN
434: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I )
435: ELSE
436: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I ) + SAFE1
437: END IF
438: 200 CONTINUE
439: *
440: KASE = 0
441: 210 CONTINUE
442: CALL DLACN2( N, WORK( 2*N+1 ), WORK( N+1 ), IWORK, FERR( J ),
443: $ KASE, ISAVE )
444: IF( KASE.NE.0 ) THEN
445: IF( KASE.EQ.1 ) THEN
446: *
1.8 bertrand 447: * Multiply by diag(W)*inv(op(A)**T).
1.1 bertrand 448: *
449: CALL DTBSV( UPLO, TRANST, DIAG, N, KD, AB, LDAB,
450: $ WORK( N+1 ), 1 )
451: DO 220 I = 1, N
452: WORK( N+I ) = WORK( I )*WORK( N+I )
453: 220 CONTINUE
454: ELSE
455: *
456: * Multiply by inv(op(A))*diag(W).
457: *
458: DO 230 I = 1, N
459: WORK( N+I ) = WORK( I )*WORK( N+I )
460: 230 CONTINUE
461: CALL DTBSV( UPLO, TRANS, DIAG, N, KD, AB, LDAB,
462: $ WORK( N+1 ), 1 )
463: END IF
464: GO TO 210
465: END IF
466: *
467: * Normalize error.
468: *
469: LSTRES = ZERO
470: DO 240 I = 1, N
471: LSTRES = MAX( LSTRES, ABS( X( I, J ) ) )
472: 240 CONTINUE
473: IF( LSTRES.NE.ZERO )
474: $ FERR( J ) = FERR( J ) / LSTRES
475: *
476: 250 CONTINUE
477: *
478: RETURN
479: *
480: * End of DTBRFS
481: *
482: END
CVSweb interface <joel.bertrand@systella.fr>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to dtbrfs.f CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [local] / rpl / lapack / lapack