Annotation of rpl/lapack/lapack/dgtrfs.f, revision 1.18
1.8 bertrand 1: *> \brief \b DGTRFS
2: *
3: * =========== DOCUMENTATION ===========
4: *
1.15 bertrand 5: * Online html documentation available at
6: * http://www.netlib.org/lapack/explore-html/
1.8 bertrand 7: *
8: *> \htmlonly
1.15 bertrand 9: *> Download DGTRFS + dependencies
10: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dgtrfs.f">
11: *> [TGZ]</a>
12: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dgtrfs.f">
13: *> [ZIP]</a>
14: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dgtrfs.f">
1.8 bertrand 15: *> [TXT]</a>
1.15 bertrand 16: *> \endhtmlonly
1.8 bertrand 17: *
18: * Definition:
19: * ===========
20: *
21: * SUBROUTINE DGTRFS( TRANS, N, NRHS, DL, D, DU, DLF, DF, DUF, DU2,
22: * IPIV, B, LDB, X, LDX, FERR, BERR, WORK, IWORK,
23: * INFO )
1.15 bertrand 24: *
1.8 bertrand 25: * .. Scalar Arguments ..
26: * CHARACTER TRANS
27: * INTEGER INFO, LDB, LDX, N, NRHS
28: * ..
29: * .. Array Arguments ..
30: * INTEGER IPIV( * ), IWORK( * )
31: * DOUBLE PRECISION B( LDB, * ), BERR( * ), D( * ), DF( * ),
32: * $ DL( * ), DLF( * ), DU( * ), DU2( * ), DUF( * ),
33: * $ FERR( * ), WORK( * ), X( LDX, * )
34: * ..
1.15 bertrand 35: *
1.8 bertrand 36: *
37: *> \par Purpose:
38: * =============
39: *>
40: *> \verbatim
41: *>
42: *> DGTRFS improves the computed solution to a system of linear
43: *> equations when the coefficient matrix is tridiagonal, and provides
44: *> error bounds and backward error estimates for the solution.
45: *> \endverbatim
46: *
47: * Arguments:
48: * ==========
49: *
50: *> \param[in] TRANS
51: *> \verbatim
52: *> TRANS is CHARACTER*1
53: *> Specifies the form of the system of equations:
54: *> = 'N': A * X = B (No transpose)
55: *> = 'T': A**T * X = B (Transpose)
56: *> = 'C': A**H * X = B (Conjugate transpose = Transpose)
57: *> \endverbatim
58: *>
59: *> \param[in] N
60: *> \verbatim
61: *> N is INTEGER
62: *> The order of the matrix A. N >= 0.
63: *> \endverbatim
64: *>
65: *> \param[in] NRHS
66: *> \verbatim
67: *> NRHS is INTEGER
68: *> The number of right hand sides, i.e., the number of columns
69: *> of the matrix B. NRHS >= 0.
70: *> \endverbatim
71: *>
72: *> \param[in] DL
73: *> \verbatim
74: *> DL is DOUBLE PRECISION array, dimension (N-1)
75: *> The (n-1) subdiagonal elements of A.
76: *> \endverbatim
77: *>
78: *> \param[in] D
79: *> \verbatim
80: *> D is DOUBLE PRECISION array, dimension (N)
81: *> The diagonal elements of A.
82: *> \endverbatim
83: *>
84: *> \param[in] DU
85: *> \verbatim
86: *> DU is DOUBLE PRECISION array, dimension (N-1)
87: *> The (n-1) superdiagonal elements of A.
88: *> \endverbatim
89: *>
90: *> \param[in] DLF
91: *> \verbatim
92: *> DLF is DOUBLE PRECISION array, dimension (N-1)
93: *> The (n-1) multipliers that define the matrix L from the
94: *> LU factorization of A as computed by DGTTRF.
95: *> \endverbatim
96: *>
97: *> \param[in] DF
98: *> \verbatim
99: *> DF is DOUBLE PRECISION array, dimension (N)
100: *> The n diagonal elements of the upper triangular matrix U from
101: *> the LU factorization of A.
102: *> \endverbatim
103: *>
104: *> \param[in] DUF
105: *> \verbatim
106: *> DUF is DOUBLE PRECISION array, dimension (N-1)
107: *> The (n-1) elements of the first superdiagonal of U.
108: *> \endverbatim
109: *>
110: *> \param[in] DU2
111: *> \verbatim
112: *> DU2 is DOUBLE PRECISION array, dimension (N-2)
113: *> The (n-2) elements of the second superdiagonal of U.
114: *> \endverbatim
115: *>
116: *> \param[in] IPIV
117: *> \verbatim
118: *> IPIV is INTEGER array, dimension (N)
119: *> The pivot indices; for 1 <= i <= n, row i of the matrix was
120: *> interchanged with row IPIV(i). IPIV(i) will always be either
121: *> i or i+1; IPIV(i) = i indicates a row interchange was not
122: *> required.
123: *> \endverbatim
124: *>
125: *> \param[in] B
126: *> \verbatim
127: *> B is DOUBLE PRECISION array, dimension (LDB,NRHS)
128: *> The right hand side matrix B.
129: *> \endverbatim
130: *>
131: *> \param[in] LDB
132: *> \verbatim
133: *> LDB is INTEGER
134: *> The leading dimension of the array B. LDB >= max(1,N).
135: *> \endverbatim
136: *>
137: *> \param[in,out] X
138: *> \verbatim
139: *> X is DOUBLE PRECISION array, dimension (LDX,NRHS)
140: *> On entry, the solution matrix X, as computed by DGTTRS.
141: *> On exit, the improved solution matrix X.
142: *> \endverbatim
143: *>
144: *> \param[in] LDX
145: *> \verbatim
146: *> LDX is INTEGER
147: *> The leading dimension of the array X. LDX >= max(1,N).
148: *> \endverbatim
149: *>
150: *> \param[out] FERR
151: *> \verbatim
152: *> FERR is DOUBLE PRECISION array, dimension (NRHS)
153: *> The estimated forward error bound for each solution vector
154: *> X(j) (the j-th column of the solution matrix X).
155: *> If XTRUE is the true solution corresponding to X(j), FERR(j)
156: *> is an estimated upper bound for the magnitude of the largest
157: *> element in (X(j) - XTRUE) divided by the magnitude of the
158: *> largest element in X(j). The estimate is as reliable as
159: *> the estimate for RCOND, and is almost always a slight
160: *> overestimate of the true error.
161: *> \endverbatim
162: *>
163: *> \param[out] BERR
164: *> \verbatim
165: *> BERR is DOUBLE PRECISION array, dimension (NRHS)
166: *> The componentwise relative backward error of each solution
167: *> vector X(j) (i.e., the smallest relative change in
168: *> any element of A or B that makes X(j) an exact solution).
169: *> \endverbatim
170: *>
171: *> \param[out] WORK
172: *> \verbatim
173: *> WORK is DOUBLE PRECISION array, dimension (3*N)
174: *> \endverbatim
175: *>
176: *> \param[out] IWORK
177: *> \verbatim
178: *> IWORK is INTEGER array, dimension (N)
179: *> \endverbatim
180: *>
181: *> \param[out] INFO
182: *> \verbatim
183: *> INFO is INTEGER
184: *> = 0: successful exit
185: *> < 0: if INFO = -i, the i-th argument had an illegal value
186: *> \endverbatim
187: *
188: *> \par Internal Parameters:
189: * =========================
190: *>
191: *> \verbatim
192: *> ITMAX is the maximum number of steps of iterative refinement.
193: *> \endverbatim
194: *
195: * Authors:
196: * ========
197: *
1.15 bertrand 198: *> \author Univ. of Tennessee
199: *> \author Univ. of California Berkeley
200: *> \author Univ. of Colorado Denver
201: *> \author NAG Ltd.
1.8 bertrand 202: *
1.11 bertrand 203: *> \ingroup doubleGTcomputational
1.8 bertrand 204: *
205: * =====================================================================
1.1 bertrand 206: SUBROUTINE DGTRFS( TRANS, N, NRHS, DL, D, DU, DLF, DF, DUF, DU2,
207: $ IPIV, B, LDB, X, LDX, FERR, BERR, WORK, IWORK,
208: $ INFO )
209: *
1.18 ! bertrand 210: * -- LAPACK computational routine --
1.1 bertrand 211: * -- LAPACK is a software package provided by Univ. of Tennessee, --
212: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
213: *
214: * .. Scalar Arguments ..
215: CHARACTER TRANS
216: INTEGER INFO, LDB, LDX, N, NRHS
217: * ..
218: * .. Array Arguments ..
219: INTEGER IPIV( * ), IWORK( * )
220: DOUBLE PRECISION B( LDB, * ), BERR( * ), D( * ), DF( * ),
221: $ DL( * ), DLF( * ), DU( * ), DU2( * ), DUF( * ),
222: $ FERR( * ), WORK( * ), X( LDX, * )
223: * ..
224: *
225: * =====================================================================
226: *
227: * .. Parameters ..
228: INTEGER ITMAX
229: PARAMETER ( ITMAX = 5 )
230: DOUBLE PRECISION ZERO, ONE
231: PARAMETER ( ZERO = 0.0D+0, ONE = 1.0D+0 )
232: DOUBLE PRECISION TWO
233: PARAMETER ( TWO = 2.0D+0 )
234: DOUBLE PRECISION THREE
235: PARAMETER ( THREE = 3.0D+0 )
236: * ..
237: * .. Local Scalars ..
238: LOGICAL NOTRAN
239: CHARACTER TRANSN, TRANST
240: INTEGER COUNT, I, J, KASE, NZ
241: DOUBLE PRECISION EPS, LSTRES, S, SAFE1, SAFE2, SAFMIN
242: * ..
243: * .. Local Arrays ..
244: INTEGER ISAVE( 3 )
245: * ..
246: * .. External Subroutines ..
247: EXTERNAL DAXPY, DCOPY, DGTTRS, DLACN2, DLAGTM, XERBLA
248: * ..
249: * .. Intrinsic Functions ..
250: INTRINSIC ABS, MAX
251: * ..
252: * .. External Functions ..
253: LOGICAL LSAME
254: DOUBLE PRECISION DLAMCH
255: EXTERNAL LSAME, DLAMCH
256: * ..
257: * .. Executable Statements ..
258: *
259: * Test the input parameters.
260: *
261: INFO = 0
262: NOTRAN = LSAME( TRANS, 'N' )
263: IF( .NOT.NOTRAN .AND. .NOT.LSAME( TRANS, 'T' ) .AND. .NOT.
264: $ LSAME( TRANS, 'C' ) ) THEN
265: INFO = -1
266: ELSE IF( N.LT.0 ) THEN
267: INFO = -2
268: ELSE IF( NRHS.LT.0 ) THEN
269: INFO = -3
270: ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
271: INFO = -13
272: ELSE IF( LDX.LT.MAX( 1, N ) ) THEN
273: INFO = -15
274: END IF
275: IF( INFO.NE.0 ) THEN
276: CALL XERBLA( 'DGTRFS', -INFO )
277: RETURN
278: END IF
279: *
280: * Quick return if possible
281: *
282: IF( N.EQ.0 .OR. NRHS.EQ.0 ) THEN
283: DO 10 J = 1, NRHS
284: FERR( J ) = ZERO
285: BERR( J ) = ZERO
286: 10 CONTINUE
287: RETURN
288: END IF
289: *
290: IF( NOTRAN ) THEN
291: TRANSN = 'N'
292: TRANST = 'T'
293: ELSE
294: TRANSN = 'T'
295: TRANST = 'N'
296: END IF
297: *
298: * NZ = maximum number of nonzero elements in each row of A, plus 1
299: *
300: NZ = 4
301: EPS = DLAMCH( 'Epsilon' )
302: SAFMIN = DLAMCH( 'Safe minimum' )
303: SAFE1 = NZ*SAFMIN
304: SAFE2 = SAFE1 / EPS
305: *
306: * Do for each right hand side
307: *
308: DO 110 J = 1, NRHS
309: *
310: COUNT = 1
311: LSTRES = THREE
312: 20 CONTINUE
313: *
314: * Loop until stopping criterion is satisfied.
315: *
316: * Compute residual R = B - op(A) * X,
317: * where op(A) = A, A**T, or A**H, depending on TRANS.
318: *
319: CALL DCOPY( N, B( 1, J ), 1, WORK( N+1 ), 1 )
320: CALL DLAGTM( TRANS, N, 1, -ONE, DL, D, DU, X( 1, J ), LDX, ONE,
321: $ WORK( N+1 ), N )
322: *
323: * Compute abs(op(A))*abs(x) + abs(b) for use in the backward
324: * error bound.
325: *
326: IF( NOTRAN ) THEN
327: IF( N.EQ.1 ) THEN
328: WORK( 1 ) = ABS( B( 1, J ) ) + ABS( D( 1 )*X( 1, J ) )
329: ELSE
330: WORK( 1 ) = ABS( B( 1, J ) ) + ABS( D( 1 )*X( 1, J ) ) +
331: $ ABS( DU( 1 )*X( 2, J ) )
332: DO 30 I = 2, N - 1
333: WORK( I ) = ABS( B( I, J ) ) +
334: $ ABS( DL( I-1 )*X( I-1, J ) ) +
335: $ ABS( D( I )*X( I, J ) ) +
336: $ ABS( DU( I )*X( I+1, J ) )
337: 30 CONTINUE
338: WORK( N ) = ABS( B( N, J ) ) +
339: $ ABS( DL( N-1 )*X( N-1, J ) ) +
340: $ ABS( D( N )*X( N, J ) )
341: END IF
342: ELSE
343: IF( N.EQ.1 ) THEN
344: WORK( 1 ) = ABS( B( 1, J ) ) + ABS( D( 1 )*X( 1, J ) )
345: ELSE
346: WORK( 1 ) = ABS( B( 1, J ) ) + ABS( D( 1 )*X( 1, J ) ) +
347: $ ABS( DL( 1 )*X( 2, J ) )
348: DO 40 I = 2, N - 1
349: WORK( I ) = ABS( B( I, J ) ) +
350: $ ABS( DU( I-1 )*X( I-1, J ) ) +
351: $ ABS( D( I )*X( I, J ) ) +
352: $ ABS( DL( I )*X( I+1, J ) )
353: 40 CONTINUE
354: WORK( N ) = ABS( B( N, J ) ) +
355: $ ABS( DU( N-1 )*X( N-1, J ) ) +
356: $ ABS( D( N )*X( N, J ) )
357: END IF
358: END IF
359: *
360: * Compute componentwise relative backward error from formula
361: *
362: * max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) )
363: *
364: * where abs(Z) is the componentwise absolute value of the matrix
365: * or vector Z. If the i-th component of the denominator is less
366: * than SAFE2, then SAFE1 is added to the i-th components of the
367: * numerator and denominator before dividing.
368: *
369: S = ZERO
370: DO 50 I = 1, N
371: IF( WORK( I ).GT.SAFE2 ) THEN
372: S = MAX( S, ABS( WORK( N+I ) ) / WORK( I ) )
373: ELSE
374: S = MAX( S, ( ABS( WORK( N+I ) )+SAFE1 ) /
375: $ ( WORK( I )+SAFE1 ) )
376: END IF
377: 50 CONTINUE
378: BERR( J ) = S
379: *
380: * Test stopping criterion. Continue iterating if
381: * 1) The residual BERR(J) is larger than machine epsilon, and
382: * 2) BERR(J) decreased by at least a factor of 2 during the
383: * last iteration, and
384: * 3) At most ITMAX iterations tried.
385: *
386: IF( BERR( J ).GT.EPS .AND. TWO*BERR( J ).LE.LSTRES .AND.
387: $ COUNT.LE.ITMAX ) THEN
388: *
389: * Update solution and try again.
390: *
391: CALL DGTTRS( TRANS, N, 1, DLF, DF, DUF, DU2, IPIV,
392: $ WORK( N+1 ), N, INFO )
393: CALL DAXPY( N, ONE, WORK( N+1 ), 1, X( 1, J ), 1 )
394: LSTRES = BERR( J )
395: COUNT = COUNT + 1
396: GO TO 20
397: END IF
398: *
399: * Bound error from formula
400: *
401: * norm(X - XTRUE) / norm(X) .le. FERR =
402: * norm( abs(inv(op(A)))*
403: * ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X)
404: *
405: * where
406: * norm(Z) is the magnitude of the largest component of Z
407: * inv(op(A)) is the inverse of op(A)
408: * abs(Z) is the componentwise absolute value of the matrix or
409: * vector Z
410: * NZ is the maximum number of nonzeros in any row of A, plus 1
411: * EPS is machine epsilon
412: *
413: * The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B))
414: * is incremented by SAFE1 if the i-th component of
415: * abs(op(A))*abs(X) + abs(B) is less than SAFE2.
416: *
417: * Use DLACN2 to estimate the infinity-norm of the matrix
418: * inv(op(A)) * diag(W),
419: * where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) )))
420: *
421: DO 60 I = 1, N
422: IF( WORK( I ).GT.SAFE2 ) THEN
423: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I )
424: ELSE
425: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I ) + SAFE1
426: END IF
427: 60 CONTINUE
428: *
429: KASE = 0
430: 70 CONTINUE
431: CALL DLACN2( N, WORK( 2*N+1 ), WORK( N+1 ), IWORK, FERR( J ),
432: $ KASE, ISAVE )
433: IF( KASE.NE.0 ) THEN
434: IF( KASE.EQ.1 ) THEN
435: *
436: * Multiply by diag(W)*inv(op(A)**T).
437: *
438: CALL DGTTRS( TRANST, N, 1, DLF, DF, DUF, DU2, IPIV,
439: $ WORK( N+1 ), N, INFO )
440: DO 80 I = 1, N
441: WORK( N+I ) = WORK( I )*WORK( N+I )
442: 80 CONTINUE
443: ELSE
444: *
445: * Multiply by inv(op(A))*diag(W).
446: *
447: DO 90 I = 1, N
448: WORK( N+I ) = WORK( I )*WORK( N+I )
449: 90 CONTINUE
450: CALL DGTTRS( TRANSN, N, 1, DLF, DF, DUF, DU2, IPIV,
451: $ WORK( N+1 ), N, INFO )
452: END IF
453: GO TO 70
454: END IF
455: *
456: * Normalize error.
457: *
458: LSTRES = ZERO
459: DO 100 I = 1, N
460: LSTRES = MAX( LSTRES, ABS( X( I, J ) ) )
461: 100 CONTINUE
462: IF( LSTRES.NE.ZERO )
463: $ FERR( J ) = FERR( J ) / LSTRES
464: *
465: 110 CONTINUE
466: *
467: RETURN
468: *
469: * End of DGTRFS
470: *
471: END
CVSweb interface <joel.bertrand@systella.fr>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to dgtrfs.f CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [local] / rpl / lapack / lapack