subroutine dlaqr2	(	logical	WANTT,
		logical	WANTZ,
		integer	N,
		integer	KTOP,
		integer	KBOT,
		integer	NW,
		double precision, dimension( ldh, * )	H,
		integer	LDH,
		integer	ILOZ,
		integer	IHIZ,
		double precision, dimension( ldz, * )	Z,
		integer	LDZ,
		integer	NS,
		integer	ND,
		double precision, dimension( * )	SR,
		double precision, dimension( * )	SI,
		double precision, dimension( ldv, * )	V,
		integer	LDV,
		integer	NH,
		double precision, dimension( ldt, * )	T,
		integer	LDT,
		integer	NV,
		double precision, dimension( ldwv, * )	WV,
		integer	LDWV,
		double precision, dimension( * )	WORK,
		integer	LWORK
	)

DLAQR2 performs the orthogonal similarity transformation of a Hessenberg matrix to detect and deflate fully converged eigenvalues from a trailing principal submatrix (aggressive early deflation).

Download DLAQR2 + dependencies [TGZ] [ZIP] [TXT]

Purpose:

    DLAQR2 is identical to DLAQR3 except that it avoids
    recursion by calling DLAHQR instead of DLAQR4.

    Aggressive early deflation:

    This subroutine accepts as input an upper Hessenberg matrix
    H and performs an orthogonal similarity transformation
    designed to detect and deflate fully converged eigenvalues from
    a trailing principal submatrix.  On output H has been over-
    written by a new Hessenberg matrix that is a perturbation of
    an orthogonal similarity transformation of H.  It is to be
    hoped that the final version of H has many zero subdiagonal
    entries.

Parameters

[in]	WANTT	WANTT is LOGICAL If .TRUE., then the Hessenberg matrix H is fully updated so that the quasi-triangular Schur factor may be computed (in cooperation with the calling subroutine). If .FALSE., then only enough of H is updated to preserve the eigenvalues.
[in]	WANTZ	WANTZ is LOGICAL If .TRUE., then the orthogonal matrix Z is updated so so that the orthogonal Schur factor may be computed (in cooperation with the calling subroutine). If .FALSE., then Z is not referenced.
[in]	N	N is INTEGER The order of the matrix H and (if WANTZ is .TRUE.) the order of the orthogonal matrix Z.
[in]	KTOP	KTOP is INTEGER It is assumed that either KTOP = 1 or H(KTOP,KTOP-1)=0. KBOT and KTOP together determine an isolated block along the diagonal of the Hessenberg matrix.
[in]	KBOT	KBOT is INTEGER It is assumed without a check that either KBOT = N or H(KBOT+1,KBOT)=0. KBOT and KTOP together determine an isolated block along the diagonal of the Hessenberg matrix.
[in]	NW	NW is INTEGER Deflation window size. 1 .LE. NW .LE. (KBOT-KTOP+1).
[in,out]	H	H is DOUBLE PRECISION array, dimension (LDH,N) On input the initial N-by-N section of H stores the Hessenberg matrix undergoing aggressive early deflation. On output H has been transformed by an orthogonal similarity transformation, perturbed, and the returned to Hessenberg form that (it is to be hoped) has some zero subdiagonal entries.
[in]	LDH	LDH is integer Leading dimension of H just as declared in the calling subroutine. N .LE. LDH
[in]	ILOZ	ILOZ is INTEGER
[in]	IHIZ	IHIZ is INTEGER Specify the rows of Z to which transformations must be applied if WANTZ is .TRUE.. 1 .LE. ILOZ .LE. IHIZ .LE. N.
[in,out]	Z	Z is DOUBLE PRECISION array, dimension (LDZ,N) IF WANTZ is .TRUE., then on output, the orthogonal similarity transformation mentioned above has been accumulated into Z(ILOZ:IHIZ,ILO:IHI) from the right. If WANTZ is .FALSE., then Z is unreferenced.
[in]	LDZ	LDZ is integer The leading dimension of Z just as declared in the calling subroutine. 1 .LE. LDZ.
[out]	NS	NS is integer The number of unconverged (ie approximate) eigenvalues returned in SR and SI that may be used as shifts by the calling subroutine.
[out]	ND	ND is integer The number of converged eigenvalues uncovered by this subroutine.
[out]	SR	SR is DOUBLE PRECISION array, dimension (KBOT)
[out]	SI	SI is DOUBLE PRECISION array, dimension (KBOT) On output, the real and imaginary parts of approximate eigenvalues that may be used for shifts are stored in SR(KBOT-ND-NS+1) through SR(KBOT-ND) and SI(KBOT-ND-NS+1) through SI(KBOT-ND), respectively. The real and imaginary parts of converged eigenvalues are stored in SR(KBOT-ND+1) through SR(KBOT) and SI(KBOT-ND+1) through SI(KBOT), respectively.
[out]	V	V is DOUBLE PRECISION array, dimension (LDV,NW) An NW-by-NW work array.
[in]	LDV	LDV is integer scalar The leading dimension of V just as declared in the calling subroutine. NW .LE. LDV
[in]	NH	NH is integer scalar The number of columns of T. NH.GE.NW.
[out]	T	T is DOUBLE PRECISION array, dimension (LDT,NW)
[in]	LDT	LDT is integer The leading dimension of T just as declared in the calling subroutine. NW .LE. LDT
[in]	NV	NV is integer The number of rows of work array WV available for workspace. NV.GE.NW.
[out]	WV	WV is DOUBLE PRECISION array, dimension (LDWV,NW)
[in]	LDWV	LDWV is integer The leading dimension of W just as declared in the calling subroutine. NW .LE. LDV
[out]	WORK	WORK is DOUBLE PRECISION array, dimension (LWORK) On exit, WORK(1) is set to an estimate of the optimal value of LWORK for the given values of N, NW, KTOP and KBOT.
[in]	LWORK	LWORK is integer The dimension of the work array WORK. LWORK = 2*NW suffices, but greater efficiency may result from larger values of LWORK. If LWORK = -1, then a workspace query is assumed; DLAQR2 only estimates the optimal workspace size for the given values of N, NW, KTOP and KBOT. The estimate is returned in WORK(1). No error message related to LWORK is issued by XERBLA. Neither H nor Z are accessed.

Author

Univ. of Tennessee

Univ. of California Berkeley

Univ. of Colorado Denver

NAG Ltd.

Date

September 2012

Contributors:

Karen Braman and Ralph Byers, Department of Mathematics, University of Kansas, USA

Definition at line 280 of file dlaqr2.f.

*
*  -- LAPACK auxiliary routine (version 3.4.2) --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*     September 2012
*
*     .. Scalar Arguments ..
      INTEGER            ihiz, iloz, kbot, ktop, ldh, ldt, ldv, ldwv,
     $                   ldz, lwork, n, nd, nh, ns, nv, nw
      LOGICAL            wantt, wantz
*     ..
*     .. Array Arguments ..
      DOUBLE PRECISION   h( ldh, * ), si( * ), sr( * ), t( ldt, * ),
     $                   v( ldv, * ), work( * ), wv( ldwv, * ),
     $                   z( ldz, * )
*     ..
*
*  ================================================================
*     .. Parameters ..
      DOUBLE PRECISION   zero, one
      parameter                ( zero = 0.0d0, one = 1.0d0 )
*     ..
*     .. Local Scalars ..
      DOUBLE PRECISION   aa, bb, beta, cc, cs, dd, evi, evk, foo, s,
     $                   safmax, safmin, smlnum, sn, tau, ulp
      INTEGER            i, ifst, ilst, info, infqr, j, jw, k, kcol,
     $                   kend, kln, krow, kwtop, ltop, lwk1, lwk2,
     $                   lwkopt
      LOGICAL            bulge, sorted
*     ..
*     .. External Functions ..
      DOUBLE PRECISION   dlamch
      EXTERNAL           dlamch
*     ..
*     .. External Subroutines ..
      EXTERNAL           dcopy, dgehrd, dgemm, dlabad, dlacpy, dlahqr,
     $                   dlanv2, dlarf, dlarfg, dlaset, dormhr, dtrexc
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, dble, int, max, min, sqrt
*     ..
*     .. Executable Statements ..
*
*     ==== Estimate optimal workspace. ====
*
      jw = min( nw, kbot-ktop+1 )
      IF( jw.LE.2 ) THEN
         lwkopt = 1
      ELSE
*
*        ==== Workspace query call to DGEHRD ====
*
         CALL dgehrd( jw, 1, jw-1, t, ldt, work, work, -1, info )
         lwk1 = int( work( 1 ) )
*
*        ==== Workspace query call to DORMHR ====
*
         CALL dormhr( 'R', 'N', jw, jw, 1, jw-1, t, ldt, work, v, ldv,
     $                work, -1, info )
         lwk2 = int( work( 1 ) )
*
*        ==== Optimal workspace ====
*
         lwkopt = jw + max( lwk1, lwk2 )
      END IF
*
*     ==== Quick return in case of workspace query. ====
*
      IF( lwork.EQ.-1 ) THEN
         work( 1 ) = dble( lwkopt )
         RETURN
      END IF
*
*     ==== Nothing to do ...
*     ... for an empty active block ... ====
      ns = 0
      nd = 0
      work( 1 ) = one
      IF( ktop.GT.kbot )
     $   RETURN
*     ... nor for an empty deflation window. ====
      IF( nw.LT.1 )
     $   RETURN
*
*     ==== Machine constants ====
*
      safmin = dlamch( 'SAFE MINIMUM' )
      safmax = one / safmin
      CALL dlabad( safmin, safmax )
      ulp = dlamch( 'PRECISION' )
      smlnum = safmin*( dble( n ) / ulp )
*
*     ==== Setup deflation window ====
*
      jw = min( nw, kbot-ktop+1 )
      kwtop = kbot - jw + 1
      IF( kwtop.EQ.ktop ) THEN
         s = zero
      ELSE
         s = h( kwtop, kwtop-1 )
      END IF
*
      IF( kbot.EQ.kwtop ) THEN
*
*        ==== 1-by-1 deflation window: not much to do ====
*
         sr( kwtop ) = h( kwtop, kwtop )
         si( kwtop ) = zero
         ns = 1
         nd = 0
         IF( abs( s ).LE.max( smlnum, ulp*abs( h( kwtop, kwtop ) ) ) )
     $        THEN
            ns = 0
            nd = 1
            IF( kwtop.GT.ktop )
     $         h( kwtop, kwtop-1 ) = zero
         END IF
         work( 1 ) = one
         RETURN
      END IF
*
*     ==== Convert to spike-triangular form.  (In case of a
*     .    rare QR failure, this routine continues to do
*     .    aggressive early deflation using that part of
*     .    the deflation window that converged using INFQR
*     .    here and there to keep track.) ====
*
      CALL dlacpy( 'U', jw, jw, h( kwtop, kwtop ), ldh, t, ldt )
      CALL dcopy( jw-1, h( kwtop+1, kwtop ), ldh+1, t( 2, 1 ), ldt+1 )
*
      CALL dlaset( 'A', jw, jw, zero, one, v, ldv )
      CALL dlahqr( .true., .true., jw, 1, jw, t, ldt, sr( kwtop ),
     $             si( kwtop ), 1, jw, v, ldv, infqr )
*
*     ==== DTREXC needs a clean margin near the diagonal ====
*
      DO 10 j = 1, jw - 3
         t( j+2, j ) = zero
         t( j+3, j ) = zero
   10 CONTINUE
      IF( jw.GT.2 )
     $   t( jw, jw-2 ) = zero
*
*     ==== Deflation detection loop ====
*
      ns = jw
      ilst = infqr + 1
   20 CONTINUE
      IF( ilst.LE.ns ) THEN
         IF( ns.EQ.1 ) THEN
            bulge = .false.
         ELSE
            bulge = t( ns, ns-1 ).NE.zero
         END IF
*
*        ==== Small spike tip test for deflation ====
*
         IF( .NOT.bulge ) THEN
*
*           ==== Real eigenvalue ====
*
            foo = abs( t( ns, ns ) )
            IF( foo.EQ.zero )
     $         foo = abs( s )
            IF( abs( s*v( 1, ns ) ).LE.max( smlnum, ulp*foo ) ) THEN
*
*              ==== Deflatable ====
*
               ns = ns - 1
            ELSE
*
*              ==== Undeflatable.   Move it up out of the way.
*              .    (DTREXC can not fail in this case.) ====
*
               ifst = ns
               CALL dtrexc( 'V', jw, t, ldt, v, ldv, ifst, ilst, work,
     $                      info )
               ilst = ilst + 1
            END IF
         ELSE
*
*           ==== Complex conjugate pair ====
*
            foo = abs( t( ns, ns ) ) + sqrt( abs( t( ns, ns-1 ) ) )*
     $            sqrt( abs( t( ns-1, ns ) ) )
            IF( foo.EQ.zero )
     $         foo = abs( s )
            IF( max( abs( s*v( 1, ns ) ), abs( s*v( 1, ns-1 ) ) ).LE.
     $          max( smlnum, ulp*foo ) ) THEN
*
*              ==== Deflatable ====
*
               ns = ns - 2
            ELSE
*
*              ==== Undeflatable. Move them up out of the way.
*              .    Fortunately, DTREXC does the right thing with
*              .    ILST in case of a rare exchange failure. ====
*
               ifst = ns
               CALL dtrexc( 'V', jw, t, ldt, v, ldv, ifst, ilst, work,
     $                      info )
               ilst = ilst + 2
            END IF
         END IF
*
*        ==== End deflation detection loop ====
*
         GO TO 20
      END IF
*
*        ==== Return to Hessenberg form ====
*
      IF( ns.EQ.0 )
     $   s = zero
*
      IF( ns.LT.jw ) THEN
*
*        ==== sorting diagonal blocks of T improves accuracy for
*        .    graded matrices.  Bubble sort deals well with
*        .    exchange failures. ====
*
         sorted = .false.
         i = ns + 1
   30    CONTINUE
         IF( sorted )
     $      GO TO 50
         sorted = .true.
*
         kend = i - 1
         i = infqr + 1
         IF( i.EQ.ns ) THEN
            k = i + 1
         ELSE IF( t( i+1, i ).EQ.zero ) THEN
            k = i + 1
         ELSE
            k = i + 2
         END IF
   40    CONTINUE
         IF( k.LE.kend ) THEN
            IF( k.EQ.i+1 ) THEN
               evi = abs( t( i, i ) )
            ELSE
               evi = abs( t( i, i ) ) + sqrt( abs( t( i+1, i ) ) )*
     $               sqrt( abs( t( i, i+1 ) ) )
            END IF
*
            IF( k.EQ.kend ) THEN
               evk = abs( t( k, k ) )
            ELSE IF( t( k+1, k ).EQ.zero ) THEN
               evk = abs( t( k, k ) )
            ELSE
               evk = abs( t( k, k ) ) + sqrt( abs( t( k+1, k ) ) )*
     $               sqrt( abs( t( k, k+1 ) ) )
            END IF
*
            IF( evi.GE.evk ) THEN
               i = k
            ELSE
               sorted = .false.
               ifst = i
               ilst = k
               CALL dtrexc( 'V', jw, t, ldt, v, ldv, ifst, ilst, work,
     $                      info )
               IF( info.EQ.0 ) THEN
                  i = ilst
               ELSE
                  i = k
               END IF
            END IF
            IF( i.EQ.kend ) THEN
               k = i + 1
            ELSE IF( t( i+1, i ).EQ.zero ) THEN
               k = i + 1
            ELSE
               k = i + 2
            END IF
            GO TO 40
         END IF
         GO TO 30
   50    CONTINUE
      END IF
*
*     ==== Restore shift/eigenvalue array from T ====
*
      i = jw
   60 CONTINUE
      IF( i.GE.infqr+1 ) THEN
         IF( i.EQ.infqr+1 ) THEN
            sr( kwtop+i-1 ) = t( i, i )
            si( kwtop+i-1 ) = zero
            i = i - 1
         ELSE IF( t( i, i-1 ).EQ.zero ) THEN
            sr( kwtop+i-1 ) = t( i, i )
            si( kwtop+i-1 ) = zero
            i = i - 1
         ELSE
            aa = t( i-1, i-1 )
            cc = t( i, i-1 )
            bb = t( i-1, i )
            dd = t( i, i )
            CALL dlanv2( aa, bb, cc, dd, sr( kwtop+i-2 ),
     $                   si( kwtop+i-2 ), sr( kwtop+i-1 ),
     $                   si( kwtop+i-1 ), cs, sn )
            i = i - 2
         END IF
         GO TO 60
      END IF
*
      IF( ns.LT.jw .OR. s.EQ.zero ) THEN
         IF( ns.GT.1 .AND. s.NE.zero ) THEN
*
*           ==== Reflect spike back into lower triangle ====
*
            CALL dcopy( ns, v, ldv, work, 1 )
            beta = work( 1 )
            CALL dlarfg( ns, beta, work( 2 ), 1, tau )
            work( 1 ) = one
*
            CALL dlaset( 'L', jw-2, jw-2, zero, zero, t( 3, 1 ), ldt )
*
            CALL dlarf( 'L', ns, jw, work, 1, tau, t, ldt,
     $                  work( jw+1 ) )
            CALL dlarf( 'R', ns, ns, work, 1, tau, t, ldt,
     $                  work( jw+1 ) )
            CALL dlarf( 'R', jw, ns, work, 1, tau, v, ldv,
     $                  work( jw+1 ) )
*
            CALL dgehrd( jw, 1, ns, t, ldt, work, work( jw+1 ),
     $                   lwork-jw, info )
         END IF
*
*        ==== Copy updated reduced window into place ====
*
         IF( kwtop.GT.1 )
     $      h( kwtop, kwtop-1 ) = s*v( 1, 1 )
         CALL dlacpy( 'U', jw, jw, t, ldt, h( kwtop, kwtop ), ldh )
         CALL dcopy( jw-1, t( 2, 1 ), ldt+1, h( kwtop+1, kwtop ),
     $               ldh+1 )
*
*        ==== Accumulate orthogonal matrix in order update
*        .    H and Z, if requested.  ====
*
         IF( ns.GT.1 .AND. s.NE.zero )
     $      CALL dormhr( 'R', 'N', jw, ns, 1, ns, t, ldt, work, v, ldv,
     $                   work( jw+1 ), lwork-jw, info )
*
*        ==== Update vertical slab in H ====
*
         IF( wantt ) THEN
            ltop = 1
         ELSE
            ltop = ktop
         END IF
         DO 70 krow = ltop, kwtop - 1, nv
            kln = min( nv, kwtop-krow )
            CALL dgemm( 'N', 'N', kln, jw, jw, one, h( krow, kwtop ),
     $                  ldh, v, ldv, zero, wv, ldwv )
            CALL dlacpy( 'A', kln, jw, wv, ldwv, h( krow, kwtop ), ldh )
   70    CONTINUE
*
*        ==== Update horizontal slab in H ====
*
         IF( wantt ) THEN
            DO 80 kcol = kbot + 1, n, nh
               kln = min( nh, n-kcol+1 )
               CALL dgemm( 'C', 'N', jw, kln, jw, one, v, ldv,
     $                     h( kwtop, kcol ), ldh, zero, t, ldt )
               CALL dlacpy( 'A', jw, kln, t, ldt, h( kwtop, kcol ),
     $                      ldh )
   80       CONTINUE
         END IF
*
*        ==== Update vertical slab in Z ====
*
         IF( wantz ) THEN
            DO 90 krow = iloz, ihiz, nv
               kln = min( nv, ihiz-krow+1 )
               CALL dgemm( 'N', 'N', kln, jw, jw, one, z( krow, kwtop ),
     $                     ldz, v, ldv, zero, wv, ldwv )
               CALL dlacpy( 'A', kln, jw, wv, ldwv, z( krow, kwtop ),
     $                      ldz )
   90       CONTINUE
         END IF
      END IF
*
*     ==== Return the number of deflations ... ====
*
      nd = jw - ns
*
*     ==== ... and the number of shifts. (Subtracting
*     .    INFQR from the spike length takes care
*     .    of the case of a rare QR failure while
*     .    calculating eigenvalues of the deflation
*     .    window.)  ====
*
      ns = ns - infqr
*
*      ==== Return optimal workspace. ====
*
      work( 1 ) = dble( lwkopt )
*
*     ==== End of DLAQR2 ====
*

Here is the call graph for this function:

Here is the caller graph for this function: