1: SUBROUTINE DSPRFS( UPLO, N, NRHS, AP, AFP, IPIV, B, LDB, X, LDX,
2: $ FERR, BERR, WORK, IWORK, INFO )
3: *
4: * -- LAPACK routine (version 3.2) --
5: * -- LAPACK is a software package provided by Univ. of Tennessee, --
6: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
7: * November 2006
8: *
9: * Modified to call DLACN2 in place of DLACON, 5 Feb 03, SJH.
10: *
11: * .. Scalar Arguments ..
12: CHARACTER UPLO
13: INTEGER INFO, LDB, LDX, N, NRHS
14: * ..
15: * .. Array Arguments ..
16: INTEGER IPIV( * ), IWORK( * )
17: DOUBLE PRECISION AFP( * ), AP( * ), B( LDB, * ), BERR( * ),
18: $ FERR( * ), WORK( * ), X( LDX, * )
19: * ..
20: *
21: * Purpose
22: * =======
23: *
24: * DSPRFS improves the computed solution to a system of linear
25: * equations when the coefficient matrix is symmetric indefinite
26: * and packed, and provides error bounds and backward error estimates
27: * for the solution.
28: *
29: * Arguments
30: * =========
31: *
32: * UPLO (input) CHARACTER*1
33: * = 'U': Upper triangle of A is stored;
34: * = 'L': Lower triangle of A is stored.
35: *
36: * N (input) INTEGER
37: * The order of the matrix A. N >= 0.
38: *
39: * NRHS (input) INTEGER
40: * The number of right hand sides, i.e., the number of columns
41: * of the matrices B and X. NRHS >= 0.
42: *
43: * AP (input) DOUBLE PRECISION array, dimension (N*(N+1)/2)
44: * The upper or lower triangle of the symmetric matrix A, packed
45: * columnwise in a linear array. The j-th column of A is stored
46: * in the array AP as follows:
47: * if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j;
48: * if UPLO = 'L', AP(i + (j-1)*(2*n-j)/2) = A(i,j) for j<=i<=n.
49: *
50: * AFP (input) DOUBLE PRECISION array, dimension (N*(N+1)/2)
51: * The factored form of the matrix A. AFP contains the block
52: * diagonal matrix D and the multipliers used to obtain the
53: * factor U or L from the factorization A = U*D*U**T or
54: * A = L*D*L**T as computed by DSPTRF, stored as a packed
55: * triangular matrix.
56: *
57: * IPIV (input) INTEGER array, dimension (N)
58: * Details of the interchanges and the block structure of D
59: * as determined by DSPTRF.
60: *
61: * B (input) DOUBLE PRECISION array, dimension (LDB,NRHS)
62: * The right hand side matrix B.
63: *
64: * LDB (input) INTEGER
65: * The leading dimension of the array B. LDB >= max(1,N).
66: *
67: * X (input/output) DOUBLE PRECISION array, dimension (LDX,NRHS)
68: * On entry, the solution matrix X, as computed by DSPTRS.
69: * On exit, the improved solution matrix X.
70: *
71: * LDX (input) INTEGER
72: * The leading dimension of the array X. LDX >= max(1,N).
73: *
74: * FERR (output) DOUBLE PRECISION array, dimension (NRHS)
75: * The estimated forward error bound for each solution vector
76: * X(j) (the j-th column of the solution matrix X).
77: * If XTRUE is the true solution corresponding to X(j), FERR(j)
78: * is an estimated upper bound for the magnitude of the largest
79: * element in (X(j) - XTRUE) divided by the magnitude of the
80: * largest element in X(j). The estimate is as reliable as
81: * the estimate for RCOND, and is almost always a slight
82: * overestimate of the true error.
83: *
84: * BERR (output) DOUBLE PRECISION array, dimension (NRHS)
85: * The componentwise relative backward error of each solution
86: * vector X(j) (i.e., the smallest relative change in
87: * any element of A or B that makes X(j) an exact solution).
88: *
89: * WORK (workspace) DOUBLE PRECISION array, dimension (3*N)
90: *
91: * IWORK (workspace) INTEGER array, dimension (N)
92: *
93: * INFO (output) INTEGER
94: * = 0: successful exit
95: * < 0: if INFO = -i, the i-th argument had an illegal value
96: *
97: * Internal Parameters
98: * ===================
99: *
100: * ITMAX is the maximum number of steps of iterative refinement.
101: *
102: * =====================================================================
103: *
104: * .. Parameters ..
105: INTEGER ITMAX
106: PARAMETER ( ITMAX = 5 )
107: DOUBLE PRECISION ZERO
108: PARAMETER ( ZERO = 0.0D+0 )
109: DOUBLE PRECISION ONE
110: PARAMETER ( ONE = 1.0D+0 )
111: DOUBLE PRECISION TWO
112: PARAMETER ( TWO = 2.0D+0 )
113: DOUBLE PRECISION THREE
114: PARAMETER ( THREE = 3.0D+0 )
115: * ..
116: * .. Local Scalars ..
117: LOGICAL UPPER
118: INTEGER COUNT, I, IK, J, K, KASE, KK, NZ
119: DOUBLE PRECISION EPS, LSTRES, S, SAFE1, SAFE2, SAFMIN, XK
120: * ..
121: * .. Local Arrays ..
122: INTEGER ISAVE( 3 )
123: * ..
124: * .. External Subroutines ..
125: EXTERNAL DAXPY, DCOPY, DLACN2, DSPMV, DSPTRS, XERBLA
126: * ..
127: * .. Intrinsic Functions ..
128: INTRINSIC ABS, MAX
129: * ..
130: * .. External Functions ..
131: LOGICAL LSAME
132: DOUBLE PRECISION DLAMCH
133: EXTERNAL LSAME, DLAMCH
134: * ..
135: * .. Executable Statements ..
136: *
137: * Test the input parameters.
138: *
139: INFO = 0
140: UPPER = LSAME( UPLO, 'U' )
141: IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN
142: INFO = -1
143: ELSE IF( N.LT.0 ) THEN
144: INFO = -2
145: ELSE IF( NRHS.LT.0 ) THEN
146: INFO = -3
147: ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
148: INFO = -8
149: ELSE IF( LDX.LT.MAX( 1, N ) ) THEN
150: INFO = -10
151: END IF
152: IF( INFO.NE.0 ) THEN
153: CALL XERBLA( 'DSPRFS', -INFO )
154: RETURN
155: END IF
156: *
157: * Quick return if possible
158: *
159: IF( N.EQ.0 .OR. NRHS.EQ.0 ) THEN
160: DO 10 J = 1, NRHS
161: FERR( J ) = ZERO
162: BERR( J ) = ZERO
163: 10 CONTINUE
164: RETURN
165: END IF
166: *
167: * NZ = maximum number of nonzero elements in each row of A, plus 1
168: *
169: NZ = N + 1
170: EPS = DLAMCH( 'Epsilon' )
171: SAFMIN = DLAMCH( 'Safe minimum' )
172: SAFE1 = NZ*SAFMIN
173: SAFE2 = SAFE1 / EPS
174: *
175: * Do for each right hand side
176: *
177: DO 140 J = 1, NRHS
178: *
179: COUNT = 1
180: LSTRES = THREE
181: 20 CONTINUE
182: *
183: * Loop until stopping criterion is satisfied.
184: *
185: * Compute residual R = B - A * X
186: *
187: CALL DCOPY( N, B( 1, J ), 1, WORK( N+1 ), 1 )
188: CALL DSPMV( UPLO, N, -ONE, AP, X( 1, J ), 1, ONE, WORK( N+1 ),
189: $ 1 )
190: *
191: * Compute componentwise relative backward error from formula
192: *
193: * max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) )
194: *
195: * where abs(Z) is the componentwise absolute value of the matrix
196: * or vector Z. If the i-th component of the denominator is less
197: * than SAFE2, then SAFE1 is added to the i-th components of the
198: * numerator and denominator before dividing.
199: *
200: DO 30 I = 1, N
201: WORK( I ) = ABS( B( I, J ) )
202: 30 CONTINUE
203: *
204: * Compute abs(A)*abs(X) + abs(B).
205: *
206: KK = 1
207: IF( UPPER ) THEN
208: DO 50 K = 1, N
209: S = ZERO
210: XK = ABS( X( K, J ) )
211: IK = KK
212: DO 40 I = 1, K - 1
213: WORK( I ) = WORK( I ) + ABS( AP( IK ) )*XK
214: S = S + ABS( AP( IK ) )*ABS( X( I, J ) )
215: IK = IK + 1
216: 40 CONTINUE
217: WORK( K ) = WORK( K ) + ABS( AP( KK+K-1 ) )*XK + S
218: KK = KK + K
219: 50 CONTINUE
220: ELSE
221: DO 70 K = 1, N
222: S = ZERO
223: XK = ABS( X( K, J ) )
224: WORK( K ) = WORK( K ) + ABS( AP( KK ) )*XK
225: IK = KK + 1
226: DO 60 I = K + 1, N
227: WORK( I ) = WORK( I ) + ABS( AP( IK ) )*XK
228: S = S + ABS( AP( IK ) )*ABS( X( I, J ) )
229: IK = IK + 1
230: 60 CONTINUE
231: WORK( K ) = WORK( K ) + S
232: KK = KK + ( N-K+1 )
233: 70 CONTINUE
234: END IF
235: S = ZERO
236: DO 80 I = 1, N
237: IF( WORK( I ).GT.SAFE2 ) THEN
238: S = MAX( S, ABS( WORK( N+I ) ) / WORK( I ) )
239: ELSE
240: S = MAX( S, ( ABS( WORK( N+I ) )+SAFE1 ) /
241: $ ( WORK( I )+SAFE1 ) )
242: END IF
243: 80 CONTINUE
244: BERR( J ) = S
245: *
246: * Test stopping criterion. Continue iterating if
247: * 1) The residual BERR(J) is larger than machine epsilon, and
248: * 2) BERR(J) decreased by at least a factor of 2 during the
249: * last iteration, and
250: * 3) At most ITMAX iterations tried.
251: *
252: IF( BERR( J ).GT.EPS .AND. TWO*BERR( J ).LE.LSTRES .AND.
253: $ COUNT.LE.ITMAX ) THEN
254: *
255: * Update solution and try again.
256: *
257: CALL DSPTRS( UPLO, N, 1, AFP, IPIV, WORK( N+1 ), N, INFO )
258: CALL DAXPY( N, ONE, WORK( N+1 ), 1, X( 1, J ), 1 )
259: LSTRES = BERR( J )
260: COUNT = COUNT + 1
261: GO TO 20
262: END IF
263: *
264: * Bound error from formula
265: *
266: * norm(X - XTRUE) / norm(X) .le. FERR =
267: * norm( abs(inv(A))*
268: * ( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X)
269: *
270: * where
271: * norm(Z) is the magnitude of the largest component of Z
272: * inv(A) is the inverse of A
273: * abs(Z) is the componentwise absolute value of the matrix or
274: * vector Z
275: * NZ is the maximum number of nonzeros in any row of A, plus 1
276: * EPS is machine epsilon
277: *
278: * The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B))
279: * is incremented by SAFE1 if the i-th component of
280: * abs(A)*abs(X) + abs(B) is less than SAFE2.
281: *
282: * Use DLACN2 to estimate the infinity-norm of the matrix
283: * inv(A) * diag(W),
284: * where W = abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) )))
285: *
286: DO 90 I = 1, N
287: IF( WORK( I ).GT.SAFE2 ) THEN
288: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I )
289: ELSE
290: WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I ) + SAFE1
291: END IF
292: 90 CONTINUE
293: *
294: KASE = 0
295: 100 CONTINUE
296: CALL DLACN2( N, WORK( 2*N+1 ), WORK( N+1 ), IWORK, FERR( J ),
297: $ KASE, ISAVE )
298: IF( KASE.NE.0 ) THEN
299: IF( KASE.EQ.1 ) THEN
300: *
301: * Multiply by diag(W)*inv(A').
302: *
303: CALL DSPTRS( UPLO, N, 1, AFP, IPIV, WORK( N+1 ), N,
304: $ INFO )
305: DO 110 I = 1, N
306: WORK( N+I ) = WORK( I )*WORK( N+I )
307: 110 CONTINUE
308: ELSE IF( KASE.EQ.2 ) THEN
309: *
310: * Multiply by inv(A)*diag(W).
311: *
312: DO 120 I = 1, N
313: WORK( N+I ) = WORK( I )*WORK( N+I )
314: 120 CONTINUE
315: CALL DSPTRS( UPLO, N, 1, AFP, IPIV, WORK( N+1 ), N,
316: $ INFO )
317: END IF
318: GO TO 100
319: END IF
320: *
321: * Normalize error.
322: *
323: LSTRES = ZERO
324: DO 130 I = 1, N
325: LSTRES = MAX( LSTRES, ABS( X( I, J ) ) )
326: 130 CONTINUE
327: IF( LSTRES.NE.ZERO )
328: $ FERR( J ) = FERR( J ) / LSTRES
329: *
330: 140 CONTINUE
331: *
332: RETURN
333: *
334: * End of DSPRFS
335: *
336: END
CVSweb interface <joel.bertrand@systella.fr>