LAPACK  3.9.0
LAPACK: Linear Algebra PACKage

◆ dlarrb()

subroutine dlarrb ( integer  N,
double precision, dimension( * )  D,
double precision, dimension( * )  LLD,
integer  IFIRST,
integer  ILAST,
double precision  RTOL1,
double precision  RTOL2,
integer  OFFSET,
double precision, dimension( * )  W,
double precision, dimension( * )  WGAP,
double precision, dimension( * )  WERR,
double precision, dimension( * )  WORK,
integer, dimension( * )  IWORK,
double precision  PIVMIN,
double precision  SPDIAM,
integer  TWIST,
integer  INFO 
)

DLARRB provides limited bisection to locate eigenvalues for more accuracy.

Download DLARRB + dependencies [TGZ] [ZIP] [TXT]

Purpose:
 Given the relatively robust representation(RRR) L D L^T, DLARRB
 does "limited" bisection to refine the eigenvalues of L D L^T,
 W( IFIRST-OFFSET ) through W( ILAST-OFFSET ), to more accuracy. Initial
 guesses for these eigenvalues are input in W, the corresponding estimate
 of the error in these guesses and their gaps are input in WERR
 and WGAP, respectively. During bisection, intervals
 [left, right] are maintained by storing their mid-points and
 semi-widths in the arrays W and WERR respectively.
Parameters
[in]N
          N is INTEGER
          The order of the matrix.
[in]D
          D is DOUBLE PRECISION array, dimension (N)
          The N diagonal elements of the diagonal matrix D.
[in]LLD
          LLD is DOUBLE PRECISION array, dimension (N-1)
          The (N-1) elements L(i)*L(i)*D(i).
[in]IFIRST
          IFIRST is INTEGER
          The index of the first eigenvalue to be computed.
[in]ILAST
          ILAST is INTEGER
          The index of the last eigenvalue to be computed.
[in]RTOL1
          RTOL1 is DOUBLE PRECISION
[in]RTOL2
          RTOL2 is DOUBLE PRECISION
          Tolerance for the convergence of the bisection intervals.
          An interval [LEFT,RIGHT] has converged if
          RIGHT-LEFT < MAX( RTOL1*GAP, RTOL2*MAX(|LEFT|,|RIGHT|) )
          where GAP is the (estimated) distance to the nearest
          eigenvalue.
[in]OFFSET
          OFFSET is INTEGER
          Offset for the arrays W, WGAP and WERR, i.e., the IFIRST-OFFSET
          through ILAST-OFFSET elements of these arrays are to be used.
[in,out]W
          W is DOUBLE PRECISION array, dimension (N)
          On input, W( IFIRST-OFFSET ) through W( ILAST-OFFSET ) are
          estimates of the eigenvalues of L D L^T indexed IFIRST through
          ILAST.
          On output, these estimates are refined.
[in,out]WGAP
          WGAP is DOUBLE PRECISION array, dimension (N-1)
          On input, the (estimated) gaps between consecutive
          eigenvalues of L D L^T, i.e., WGAP(I-OFFSET) is the gap between
          eigenvalues I and I+1. Note that if IFIRST = ILAST
          then WGAP(IFIRST-OFFSET) must be set to ZERO.
          On output, these gaps are refined.
[in,out]WERR
          WERR is DOUBLE PRECISION array, dimension (N)
          On input, WERR( IFIRST-OFFSET ) through WERR( ILAST-OFFSET ) are
          the errors in the estimates of the corresponding elements in W.
          On output, these errors are refined.
[out]WORK
          WORK is DOUBLE PRECISION array, dimension (2*N)
          Workspace.
[out]IWORK
          IWORK is INTEGER array, dimension (2*N)
          Workspace.
[in]PIVMIN
          PIVMIN is DOUBLE PRECISION
          The minimum pivot in the Sturm sequence.
[in]SPDIAM
          SPDIAM is DOUBLE PRECISION
          The spectral diameter of the matrix.
[in]TWIST
          TWIST is INTEGER
          The twist index for the twisted factorization that is used
          for the negcount.
          TWIST = N: Compute negcount from L D L^T - LAMBDA I = L+ D+ L+^T
          TWIST = 1: Compute negcount from L D L^T - LAMBDA I = U- D- U-^T
          TWIST = R: Compute negcount from L D L^T - LAMBDA I = N(r) D(r) N(r)
[out]INFO
          INFO is INTEGER
          Error flag.
Author
Univ. of Tennessee
Univ. of California Berkeley
Univ. of Colorado Denver
NAG Ltd.
Date
June 2017
Contributors:
Beresford Parlett, University of California, Berkeley, USA
Jim Demmel, University of California, Berkeley, USA
Inderjit Dhillon, University of Texas, Austin, USA
Osni Marques, LBNL/NERSC, USA
Christof Voemel, University of California, Berkeley, USA

Definition at line 198 of file dlarrb.f.

198 *
199 * -- LAPACK auxiliary routine (version 3.7.1) --
200 * -- LAPACK is a software package provided by Univ. of Tennessee, --
201 * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
202 * June 2017
203 *
204 * .. Scalar Arguments ..
205  INTEGER IFIRST, ILAST, INFO, N, OFFSET, TWIST
206  DOUBLE PRECISION PIVMIN, RTOL1, RTOL2, SPDIAM
207 * ..
208 * .. Array Arguments ..
209  INTEGER IWORK( * )
210  DOUBLE PRECISION D( * ), LLD( * ), W( * ),
211  $ WERR( * ), WGAP( * ), WORK( * )
212 * ..
213 *
214 * =====================================================================
215 *
216 * .. Parameters ..
217  DOUBLE PRECISION ZERO, TWO, HALF
218  parameter( zero = 0.0d0, two = 2.0d0,
219  $ half = 0.5d0 )
220  INTEGER MAXITR
221 * ..
222 * .. Local Scalars ..
223  INTEGER I, I1, II, IP, ITER, K, NEGCNT, NEXT, NINT,
224  $ OLNINT, PREV, R
225  DOUBLE PRECISION BACK, CVRGD, GAP, LEFT, LGAP, MID, MNWDTH,
226  $ RGAP, RIGHT, TMP, WIDTH
227 * ..
228 * .. External Functions ..
229  INTEGER DLANEG
230  EXTERNAL dlaneg
231 *
232 * ..
233 * .. Intrinsic Functions ..
234  INTRINSIC abs, max, min
235 * ..
236 * .. Executable Statements ..
237 *
238  info = 0
239 *
240 * Quick return if possible
241 *
242  IF( n.LE.0 ) THEN
243  RETURN
244  END IF
245 *
246  maxitr = int( ( log( spdiam+pivmin )-log( pivmin ) ) /
247  $ log( two ) ) + 2
248  mnwdth = two * pivmin
249 *
250  r = twist
251  IF((r.LT.1).OR.(r.GT.n)) r = n
252 *
253 * Initialize unconverged intervals in [ WORK(2*I-1), WORK(2*I) ].
254 * The Sturm Count, Count( WORK(2*I-1) ) is arranged to be I-1, while
255 * Count( WORK(2*I) ) is stored in IWORK( 2*I ). The integer IWORK( 2*I-1 )
256 * for an unconverged interval is set to the index of the next unconverged
257 * interval, and is -1 or 0 for a converged interval. Thus a linked
258 * list of unconverged intervals is set up.
259 *
260  i1 = ifirst
261 * The number of unconverged intervals
262  nint = 0
263 * The last unconverged interval found
264  prev = 0
265 
266  rgap = wgap( i1-offset )
267  DO 75 i = i1, ilast
268  k = 2*i
269  ii = i - offset
270  left = w( ii ) - werr( ii )
271  right = w( ii ) + werr( ii )
272  lgap = rgap
273  rgap = wgap( ii )
274  gap = min( lgap, rgap )
275 
276 * Make sure that [LEFT,RIGHT] contains the desired eigenvalue
277 * Compute negcount from dstqds facto L+D+L+^T = L D L^T - LEFT
278 *
279 * Do while( NEGCNT(LEFT).GT.I-1 )
280 *
281  back = werr( ii )
282  20 CONTINUE
283  negcnt = dlaneg( n, d, lld, left, pivmin, r )
284  IF( negcnt.GT.i-1 ) THEN
285  left = left - back
286  back = two*back
287  GO TO 20
288  END IF
289 *
290 * Do while( NEGCNT(RIGHT).LT.I )
291 * Compute negcount from dstqds facto L+D+L+^T = L D L^T - RIGHT
292 *
293  back = werr( ii )
294  50 CONTINUE
295 
296  negcnt = dlaneg( n, d, lld, right, pivmin, r )
297  IF( negcnt.LT.i ) THEN
298  right = right + back
299  back = two*back
300  GO TO 50
301  END IF
302  width = half*abs( left - right )
303  tmp = max( abs( left ), abs( right ) )
304  cvrgd = max(rtol1*gap,rtol2*tmp)
305  IF( width.LE.cvrgd .OR. width.LE.mnwdth ) THEN
306 * This interval has already converged and does not need refinement.
307 * (Note that the gaps might change through refining the
308 * eigenvalues, however, they can only get bigger.)
309 * Remove it from the list.
310  iwork( k-1 ) = -1
311 * Make sure that I1 always points to the first unconverged interval
312  IF((i.EQ.i1).AND.(i.LT.ilast)) i1 = i + 1
313  IF((prev.GE.i1).AND.(i.LE.ilast)) iwork( 2*prev-1 ) = i + 1
314  ELSE
315 * unconverged interval found
316  prev = i
317  nint = nint + 1
318  iwork( k-1 ) = i + 1
319  iwork( k ) = negcnt
320  END IF
321  work( k-1 ) = left
322  work( k ) = right
323  75 CONTINUE
324 
325 *
326 * Do while( NINT.GT.0 ), i.e. there are still unconverged intervals
327 * and while (ITER.LT.MAXITR)
328 *
329  iter = 0
330  80 CONTINUE
331  prev = i1 - 1
332  i = i1
333  olnint = nint
334 
335  DO 100 ip = 1, olnint
336  k = 2*i
337  ii = i - offset
338  rgap = wgap( ii )
339  lgap = rgap
340  IF(ii.GT.1) lgap = wgap( ii-1 )
341  gap = min( lgap, rgap )
342  next = iwork( k-1 )
343  left = work( k-1 )
344  right = work( k )
345  mid = half*( left + right )
346 
347 * semiwidth of interval
348  width = right - mid
349  tmp = max( abs( left ), abs( right ) )
350  cvrgd = max(rtol1*gap,rtol2*tmp)
351  IF( ( width.LE.cvrgd ) .OR. ( width.LE.mnwdth ).OR.
352  $ ( iter.EQ.maxitr ) )THEN
353 * reduce number of unconverged intervals
354  nint = nint - 1
355 * Mark interval as converged.
356  iwork( k-1 ) = 0
357  IF( i1.EQ.i ) THEN
358  i1 = next
359  ELSE
360 * Prev holds the last unconverged interval previously examined
361  IF(prev.GE.i1) iwork( 2*prev-1 ) = next
362  END IF
363  i = next
364  GO TO 100
365  END IF
366  prev = i
367 *
368 * Perform one bisection step
369 *
370  negcnt = dlaneg( n, d, lld, mid, pivmin, r )
371  IF( negcnt.LE.i-1 ) THEN
372  work( k-1 ) = mid
373  ELSE
374  work( k ) = mid
375  END IF
376  i = next
377  100 CONTINUE
378  iter = iter + 1
379 * do another loop if there are still unconverged intervals
380 * However, in the last iteration, all intervals are accepted
381 * since this is the best we can do.
382  IF( ( nint.GT.0 ).AND.(iter.LE.maxitr) ) GO TO 80
383 *
384 *
385 * At this point, all the intervals have converged
386  DO 110 i = ifirst, ilast
387  k = 2*i
388  ii = i - offset
389 * All intervals marked by '0' have been refined.
390  IF( iwork( k-1 ).EQ.0 ) THEN
391  w( ii ) = half*( work( k-1 )+work( k ) )
392  werr( ii ) = work( k ) - w( ii )
393  END IF
394  110 CONTINUE
395 *
396  DO 111 i = ifirst+1, ilast
397  k = 2*i
398  ii = i - offset
399  wgap( ii-1 ) = max( zero,
400  $ w(ii) - werr(ii) - w( ii-1 ) - werr( ii-1 ))
401  111 CONTINUE
402 
403  RETURN
404 *
405 * End of DLARRB
406 *
Here is the caller graph for this function:
dlaneg
integer function dlaneg(N, D, LLD, SIGMA, PIVMIN, R)
DLANEG computes the Sturm count.
Definition: dlaneg.f:120