1: *> \brief \b DPTRFS
2: *
3: * =========== DOCUMENTATION ===========
4: *
5: * Online html documentation available at
6: * http://www.netlib.org/lapack/explore-html/
7: *
8: *> \htmlonly
9: *> Download DPTRFS + dependencies
10: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dptrfs.f">
11: *> [TGZ]</a>
12: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dptrfs.f">
13: *> [ZIP]</a>
14: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dptrfs.f">
15: *> [TXT]</a>
16: *> \endhtmlonly
17: *
18: * Definition:
19: * ===========
20: *
21: * SUBROUTINE DPTRFS( N, NRHS, D, E, DF, EF, B, LDB, X, LDX, FERR,
22: * BERR, WORK, INFO )
23: *
24: * .. Scalar Arguments ..
25: * INTEGER INFO, LDB, LDX, N, NRHS
26: * ..
27: * .. Array Arguments ..
28: * DOUBLE PRECISION B( LDB, * ), BERR( * ), D( * ), DF( * ),
29: * $ E( * ), EF( * ), FERR( * ), WORK( * ),
30: * $ X( LDX, * )
31: * ..
32: *
33: *
34: *> \par Purpose:
35: * =============
36: *>
37: *> \verbatim
38: *>
39: *> DPTRFS improves the computed solution to a system of linear
40: *> equations when the coefficient matrix is symmetric positive definite
41: *> and tridiagonal, and provides error bounds and backward error
42: *> estimates for the solution.
43: *> \endverbatim
44: *
45: * Arguments:
46: * ==========
47: *
48: *> \param[in] N
49: *> \verbatim
50: *> N is INTEGER
51: *> The order of the matrix A. N >= 0.
52: *> \endverbatim
53: *>
54: *> \param[in] NRHS
55: *> \verbatim
56: *> NRHS is INTEGER
57: *> The number of right hand sides, i.e., the number of columns
58: *> of the matrix B. NRHS >= 0.
59: *> \endverbatim
60: *>
61: *> \param[in] D
62: *> \verbatim
63: *> D is DOUBLE PRECISION array, dimension (N)
64: *> The n diagonal elements of the tridiagonal matrix A.
65: *> \endverbatim
66: *>
67: *> \param[in] E
68: *> \verbatim
69: *> E is DOUBLE PRECISION array, dimension (N-1)
70: *> The (n-1) subdiagonal elements of the tridiagonal matrix A.
71: *> \endverbatim
72: *>
73: *> \param[in] DF
74: *> \verbatim
75: *> DF is DOUBLE PRECISION array, dimension (N)
76: *> The n diagonal elements of the diagonal matrix D from the
77: *> factorization computed by DPTTRF.
78: *> \endverbatim
79: *>
80: *> \param[in] EF
81: *> \verbatim
82: *> EF is DOUBLE PRECISION array, dimension (N-1)
83: *> The (n-1) subdiagonal elements of the unit bidiagonal factor
84: *> L from the factorization computed by DPTTRF.
85: *> \endverbatim
86: *>
87: *> \param[in] B
88: *> \verbatim
89: *> B is DOUBLE PRECISION array, dimension (LDB,NRHS)
90: *> The right hand side matrix B.
91: *> \endverbatim
92: *>
93: *> \param[in] LDB
94: *> \verbatim
95: *> LDB is INTEGER
96: *> The leading dimension of the array B. LDB >= max(1,N).
97: *> \endverbatim
98: *>
99: *> \param[in,out] X
100: *> \verbatim
101: *> X is DOUBLE PRECISION array, dimension (LDX,NRHS)
102: *> On entry, the solution matrix X, as computed by DPTTRS.
103: *> On exit, the improved solution matrix X.
104: *> \endverbatim
105: *>
106: *> \param[in] LDX
107: *> \verbatim
108: *> LDX is INTEGER
109: *> The leading dimension of the array X. LDX >= max(1,N).
110: *> \endverbatim
111: *>
112: *> \param[out] FERR
113: *> \verbatim
114: *> FERR is DOUBLE PRECISION array, dimension (NRHS)
115: *> The forward error bound for each solution vector
116: *> X(j) (the j-th column of the solution matrix X).
117: *> If XTRUE is the true solution corresponding to X(j), FERR(j)
118: *> is an estimated upper bound for the magnitude of the largest
119: *> element in (X(j) - XTRUE) divided by the magnitude of the
120: *> largest element in X(j).
121: *> \endverbatim
122: *>
123: *> \param[out] BERR
124: *> \verbatim
125: *> BERR is DOUBLE PRECISION array, dimension (NRHS)
126: *> The componentwise relative backward error of each solution
127: *> vector X(j) (i.e., the smallest relative change in
128: *> any element of A or B that makes X(j) an exact solution).
129: *> \endverbatim
130: *>
131: *> \param[out] WORK
132: *> \verbatim
133: *> WORK is DOUBLE PRECISION array, dimension (2*N)
134: *> \endverbatim
135: *>
136: *> \param[out] INFO
137: *> \verbatim
138: *> INFO is INTEGER
139: *> = 0: successful exit
140: *> < 0: if INFO = -i, the i-th argument had an illegal value
141: *> \endverbatim
142: *
143: *> \par Internal Parameters:
144: * =========================
145: *>
146: *> \verbatim
147: *> ITMAX is the maximum number of steps of iterative refinement.
148: *> \endverbatim
149: *
150: * Authors:
151: * ========
152: *
153: *> \author Univ. of Tennessee
154: *> \author Univ. of California Berkeley
155: *> \author Univ. of Colorado Denver
156: *> \author NAG Ltd.
157: *
158: *> \ingroup doublePTcomputational
159: *
160: * =====================================================================
161: SUBROUTINE DPTRFS( N, NRHS, D, E, DF, EF, B, LDB, X, LDX, FERR,
162: $ BERR, WORK, INFO )
163: *
164: * -- LAPACK computational routine --
165: * -- LAPACK is a software package provided by Univ. of Tennessee, --
166: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
167: *
168: * .. Scalar Arguments ..
169: INTEGER INFO, LDB, LDX, N, NRHS
170: * ..
171: * .. Array Arguments ..
172: DOUBLE PRECISION B( LDB, * ), BERR( * ), D( * ), DF( * ),
173: $ E( * ), EF( * ), FERR( * ), WORK( * ),
174: $ X( LDX, * )
175: * ..
176: *
177: * =====================================================================
178: *
179: * .. Parameters ..
180: INTEGER ITMAX
181: PARAMETER ( ITMAX = 5 )
182: DOUBLE PRECISION ZERO
183: PARAMETER ( ZERO = 0.0D+0 )
184: DOUBLE PRECISION ONE
185: PARAMETER ( ONE = 1.0D+0 )
186: DOUBLE PRECISION TWO
187: PARAMETER ( TWO = 2.0D+0 )
188: DOUBLE PRECISION THREE
189: PARAMETER ( THREE = 3.0D+0 )
190: * ..
191: * .. Local Scalars ..
192: INTEGER COUNT, I, IX, J, NZ
193: DOUBLE PRECISION BI, CX, DX, EPS, EX, LSTRES, S, SAFE1, SAFE2,
194: $ SAFMIN
195: * ..
196: * .. External Subroutines ..
197: EXTERNAL DAXPY, DPTTRS, XERBLA
198: * ..
199: * .. Intrinsic Functions ..
200: INTRINSIC ABS, MAX
201: * ..
202: * .. External Functions ..
203: INTEGER IDAMAX
204: DOUBLE PRECISION DLAMCH
205: EXTERNAL IDAMAX, DLAMCH
206: * ..
207: * .. Executable Statements ..
208: *
209: * Test the input parameters.
210: *
211: INFO = 0
212: IF( N.LT.0 ) THEN
213: INFO = -1
214: ELSE IF( NRHS.LT.0 ) THEN
215: INFO = -2
216: ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
217: INFO = -8
218: ELSE IF( LDX.LT.MAX( 1, N ) ) THEN
219: INFO = -10
220: END IF
221: IF( INFO.NE.0 ) THEN
222: CALL XERBLA( 'DPTRFS', -INFO )
223: RETURN
224: END IF
225: *
226: * Quick return if possible
227: *
228: IF( N.EQ.0 .OR. NRHS.EQ.0 ) THEN
229: DO 10 J = 1, NRHS
230: FERR( J ) = ZERO
231: BERR( J ) = ZERO
232: 10 CONTINUE
233: RETURN
234: END IF
235: *
236: * NZ = maximum number of nonzero elements in each row of A, plus 1
237: *
238: NZ = 4
239: EPS = DLAMCH( 'Epsilon' )
240: SAFMIN = DLAMCH( 'Safe minimum' )
241: SAFE1 = NZ*SAFMIN
242: SAFE2 = SAFE1 / EPS
243: *
244: * Do for each right hand side
245: *
246: DO 90 J = 1, NRHS
247: *
248: COUNT = 1
249: LSTRES = THREE
250: 20 CONTINUE
251: *
252: * Loop until stopping criterion is satisfied.
253: *
254: * Compute residual R = B - A * X. Also compute
255: * abs(A)*abs(x) + abs(b) for use in the backward error bound.
256: *
257: IF( N.EQ.1 ) THEN
258: BI = B( 1, J )
259: DX = D( 1 )*X( 1, J )
260: WORK( N+1 ) = BI - DX
261: WORK( 1 ) = ABS( BI ) + ABS( DX )
262: ELSE
263: BI = B( 1, J )
264: DX = D( 1 )*X( 1, J )
265: EX = E( 1 )*X( 2, J )
266: WORK( N+1 ) = BI - DX - EX
267: WORK( 1 ) = ABS( BI ) + ABS( DX ) + ABS( EX )
268: DO 30 I = 2, N - 1
269: BI = B( I, J )
270: CX = E( I-1 )*X( I-1, J )
271: DX = D( I )*X( I, J )
272: EX = E( I )*X( I+1, J )
273: WORK( N+I ) = BI - CX - DX - EX
274: WORK( I ) = ABS( BI ) + ABS( CX ) + ABS( DX ) + ABS( EX )
275: 30 CONTINUE
276: BI = B( N, J )
277: CX = E( N-1 )*X( N-1, J )
278: DX = D( N )*X( N, J )
279: WORK( N+N ) = BI - CX - DX
280: WORK( N ) = ABS( BI ) + ABS( CX ) + ABS( DX )
281: END IF
282: *
283: * Compute componentwise relative backward error from formula
284: *
285: * max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) )
286: *
287: * where abs(Z) is the componentwise absolute value of the matrix
288: * or vector Z. If the i-th component of the denominator is less
289: * than SAFE2, then SAFE1 is added to the i-th components of the
290: * numerator and denominator before dividing.
291: *
292: S = ZERO
293: DO 40 I = 1, N
294: IF( WORK( I ).GT.SAFE2 ) THEN
295: S = MAX( S, ABS( WORK( N+I ) ) / WORK( I ) )
296: ELSE
297: S = MAX( S, ( ABS( WORK( N+I ) )+SAFE1 ) /
298: $ ( WORK( I )+SAFE1 ) )
299: END IF
300: 40 CONTINUE
301: BERR( J ) = S
302: *
303: * Test stopping criterion. Continue iterating if
304: * 1) The residual BERR(J) is larger than machine epsilon, and
305: * 2) BERR(J) decreased by at least a factor of 2 during the
306: * last iteration, and
307: * 3) At most ITMAX iterations tried.
308: *
309: IF( BERR( J ).GT.EPS .AND. TWO*BERR( J ).LE.LSTRES .AND.
310: $ COUNT.LE.ITMAX ) THEN
311: *
312: * Update solution and try again.
313: *
314: CALL DPTTRS( N, 1, DF, EF, WORK( N+1 ), N, INFO )
315: CALL DAXPY( N, ONE, WORK( N+1 ), 1, X( 1, J ), 1 )
316: LSTRES = BERR( J )
317: COUNT = COUNT + 1
318: GO TO 20
319: END IF
320: *
321: * Bound error from formula
322: *
323: * norm(X - XTRUE) / norm(X) .le. FERR =
324: * norm( abs(inv(A))*
325: * ( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X)
326: *
327: * where
328: * norm(Z) is the magnitude of the largest component of Z
329: * inv(A) is the inverse of A
330: * abs(Z) is the componentwise absolute value of the matrix or
331: * vector Z
332: * NZ is the maximum number of nonzeros in any row of A, plus 1
333: * EPS is machine epsilon
334: *
335: * The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B))
336: * is incremented by SAFE1 if the i-th component of
337: * abs(A)*abs(X) + abs(B) is less than SAFE2.
338: *
339: DO 50 I = 1, N
340: IF( WORK( I ).GT.SAFE2 ) THEN
341: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I )
342: ELSE
343: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I ) + SAFE1
344: END IF
345: 50 CONTINUE
346: IX = IDAMAX( N, WORK, 1 )
347: FERR( J ) = WORK( IX )
348: *
349: * Estimate the norm of inv(A).
350: *
351: * Solve M(A) * x = e, where M(A) = (m(i,j)) is given by
352: *
353: * m(i,j) = abs(A(i,j)), i = j,
354: * m(i,j) = -abs(A(i,j)), i .ne. j,
355: *
356: * and e = [ 1, 1, ..., 1 ]**T. Note M(A) = M(L)*D*M(L)**T.
357: *
358: * Solve M(L) * x = e.
359: *
360: WORK( 1 ) = ONE
361: DO 60 I = 2, N
362: WORK( I ) = ONE + WORK( I-1 )*ABS( EF( I-1 ) )
363: 60 CONTINUE
364: *
365: * Solve D * M(L)**T * x = b.
366: *
367: WORK( N ) = WORK( N ) / DF( N )
368: DO 70 I = N - 1, 1, -1
369: WORK( I ) = WORK( I ) / DF( I ) + WORK( I+1 )*ABS( EF( I ) )
370: 70 CONTINUE
371: *
372: * Compute norm(inv(A)) = max(x(i)), 1<=i<=n.
373: *
374: IX = IDAMAX( N, WORK, 1 )
375: FERR( J ) = FERR( J )*ABS( WORK( IX ) )
376: *
377: * Normalize error.
378: *
379: LSTRES = ZERO
380: DO 80 I = 1, N
381: LSTRES = MAX( LSTRES, ABS( X( I, J ) ) )
382: 80 CONTINUE
383: IF( LSTRES.NE.ZERO )
384: $ FERR( J ) = FERR( J ) / LSTRES
385: *
386: 90 CONTINUE
387: *
388: RETURN
389: *
390: * End of DPTRFS
391: *
392: END
CVSweb interface <joel.bertrand@systella.fr>