LAPACK: SRC/sgegv.f Source File

Go to the documentation of this file.
00001       SUBROUTINE SGEGV( JOBVL, JOBVR, N, A, LDA, B, LDB, ALPHAR, ALPHAI,
00002      $                  BETA, VL, LDVL, VR, LDVR, WORK, LWORK, INFO )
00003 *
00004 *  -- LAPACK driver routine (version 3.2) --
00005 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
00006 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
00007 *     November 2006
00008 *
00009 *     .. Scalar Arguments ..
00010       CHARACTER          JOBVL, JOBVR
00011       INTEGER            INFO, LDA, LDB, LDVL, LDVR, LWORK, N
00012 *     ..
00013 *     .. Array Arguments ..
00014       REAL               A( LDA, * ), ALPHAI( * ), ALPHAR( * ),
00015      $                   B( LDB, * ), BETA( * ), VL( LDVL, * ),
00016      $                   VR( LDVR, * ), WORK( * )
00017 *     ..
00018 *
00019 *  Purpose
00020 *  =======
00021 *
00022 *  This routine is deprecated and has been replaced by routine SGGEV.
00023 *
00024 *  SGEGV computes the eigenvalues and, optionally, the left and/or right
00025 *  eigenvectors of a real matrix pair (A,B).
00026 *  Given two square matrices A and B,
00027 *  the generalized nonsymmetric eigenvalue problem (GNEP) is to find the
00028 *  eigenvalues lambda and corresponding (non-zero) eigenvectors x such
00029 *  that
00030 *
00031 *     A*x = lambda*B*x.
00032 *
00033 *  An alternate form is to find the eigenvalues mu and corresponding
00034 *  eigenvectors y such that
00035 *
00036 *     mu*A*y = B*y.
00037 *
00038 *  These two forms are equivalent with mu = 1/lambda and x = y if
00039 *  neither lambda nor mu is zero.  In order to deal with the case that
00040 *  lambda or mu is zero or small, two values alpha and beta are returned
00041 *  for each eigenvalue, such that lambda = alpha/beta and
00042 *  mu = beta/alpha.
00043 *
00044 *  The vectors x and y in the above equations are right eigenvectors of
00045 *  the matrix pair (A,B).  Vectors u and v satisfying
00046 *
00047 *     u**H*A = lambda*u**H*B  or  mu*v**H*A = v**H*B
00048 *
00049 *  are left eigenvectors of (A,B).
00050 *
00051 *  Note: this routine performs "full balancing" on A and B -- see
00052 *  "Further Details", below.
00053 *
00054 *  Arguments
00055 *  =========
00056 *
00057 *  JOBVL   (input) CHARACTER*1
00058 *          = 'N':  do not compute the left generalized eigenvectors;
00059 *          = 'V':  compute the left generalized eigenvectors (returned
00060 *                  in VL).
00061 *
00062 *  JOBVR   (input) CHARACTER*1
00063 *          = 'N':  do not compute the right generalized eigenvectors;
00064 *          = 'V':  compute the right generalized eigenvectors (returned
00065 *                  in VR).
00066 *
00067 *  N       (input) INTEGER
00068 *          The order of the matrices A, B, VL, and VR.  N >= 0.
00069 *
00070 *  A       (input/output) REAL array, dimension (LDA, N)
00071 *          On entry, the matrix A.
00072 *          If JOBVL = 'V' or JOBVR = 'V', then on exit A
00073 *          contains the real Schur form of A from the generalized Schur
00074 *          factorization of the pair (A,B) after balancing.
00075 *          If no eigenvectors were computed, then only the diagonal
00076 *          blocks from the Schur form will be correct.  See SGGHRD and
00077 *          SHGEQZ for details.
00078 *
00079 *  LDA     (input) INTEGER
00080 *          The leading dimension of A.  LDA >= max(1,N).
00081 *
00082 *  B       (input/output) REAL array, dimension (LDB, N)
00083 *          On entry, the matrix B.
00084 *          If JOBVL = 'V' or JOBVR = 'V', then on exit B contains the
00085 *          upper triangular matrix obtained from B in the generalized
00086 *          Schur factorization of the pair (A,B) after balancing.
00087 *          If no eigenvectors were computed, then only those elements of
00088 *          B corresponding to the diagonal blocks from the Schur form of
00089 *          A will be correct.  See SGGHRD and SHGEQZ for details.
00090 *
00091 *  LDB     (input) INTEGER
00092 *          The leading dimension of B.  LDB >= max(1,N).
00093 *
00094 *  ALPHAR  (output) REAL array, dimension (N)
00095 *          The real parts of each scalar alpha defining an eigenvalue of
00096 *          GNEP.
00097 *
00098 *  ALPHAI  (output) REAL array, dimension (N)
00099 *          The imaginary parts of each scalar alpha defining an
00100 *          eigenvalue of GNEP.  If ALPHAI(j) is zero, then the j-th
00101 *          eigenvalue is real; if positive, then the j-th and
00102 *          (j+1)-st eigenvalues are a complex conjugate pair, with
00103 *          ALPHAI(j+1) = -ALPHAI(j).
00104 *
00105 *  BETA    (output) REAL array, dimension (N)
00106 *          The scalars beta that define the eigenvalues of GNEP.
00107 *          
00108 *          Together, the quantities alpha = (ALPHAR(j),ALPHAI(j)) and
00109 *          beta = BETA(j) represent the j-th eigenvalue of the matrix
00110 *          pair (A,B), in one of the forms lambda = alpha/beta or
00111 *          mu = beta/alpha.  Since either lambda or mu may overflow,
00112 *          they should not, in general, be computed.
00113 *
00114 *  VL      (output) REAL array, dimension (LDVL,N)
00115 *          If JOBVL = 'V', the left eigenvectors u(j) are stored
00116 *          in the columns of VL, in the same order as their eigenvalues.
00117 *          If the j-th eigenvalue is real, then u(j) = VL(:,j).
00118 *          If the j-th and (j+1)-st eigenvalues form a complex conjugate
00119 *          pair, then
00120 *             u(j) = VL(:,j) + i*VL(:,j+1)
00121 *          and
00122 *            u(j+1) = VL(:,j) - i*VL(:,j+1).
00123 *
00124 *          Each eigenvector is scaled so that its largest component has
00125 *          abs(real part) + abs(imag. part) = 1, except for eigenvectors
00126 *          corresponding to an eigenvalue with alpha = beta = 0, which
00127 *          are set to zero.
00128 *          Not referenced if JOBVL = 'N'.
00129 *
00130 *  LDVL    (input) INTEGER
00131 *          The leading dimension of the matrix VL. LDVL >= 1, and
00132 *          if JOBVL = 'V', LDVL >= N.
00133 *
00134 *  VR      (output) REAL array, dimension (LDVR,N)
00135 *          If JOBVR = 'V', the right eigenvectors x(j) are stored
00136 *          in the columns of VR, in the same order as their eigenvalues.
00137 *          If the j-th eigenvalue is real, then x(j) = VR(:,j).
00138 *          If the j-th and (j+1)-st eigenvalues form a complex conjugate
00139 *          pair, then
00140 *            x(j) = VR(:,j) + i*VR(:,j+1)
00141 *          and
00142 *            x(j+1) = VR(:,j) - i*VR(:,j+1).
00143 *
00144 *          Each eigenvector is scaled so that its largest component has
00145 *          abs(real part) + abs(imag. part) = 1, except for eigenvalues
00146 *          corresponding to an eigenvalue with alpha = beta = 0, which
00147 *          are set to zero.
00148 *          Not referenced if JOBVR = 'N'.
00149 *
00150 *  LDVR    (input) INTEGER
00151 *          The leading dimension of the matrix VR. LDVR >= 1, and
00152 *          if JOBVR = 'V', LDVR >= N.
00153 *
00154 *  WORK    (workspace/output) REAL array, dimension (MAX(1,LWORK))
00155 *          On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
00156 *
00157 *  LWORK   (input) INTEGER
00158 *          The dimension of the array WORK.  LWORK >= max(1,8*N).
00159 *          For good performance, LWORK must generally be larger.
00160 *          To compute the optimal value of LWORK, call ILAENV to get
00161 *          blocksizes (for SGEQRF, SORMQR, and SORGQR.)  Then compute:
00162 *          NB  -- MAX of the blocksizes for SGEQRF, SORMQR, and SORGQR;
00163 *          The optimal LWORK is:
00164 *              2*N + MAX( 6*N, N*(NB+1) ).
00165 *
00166 *          If LWORK = -1, then a workspace query is assumed; the routine
00167 *          only calculates the optimal size of the WORK array, returns
00168 *          this value as the first entry of the WORK array, and no error
00169 *          message related to LWORK is issued by XERBLA.
00170 *
00171 *  INFO    (output) INTEGER
00172 *          = 0:  successful exit
00173 *          < 0:  if INFO = -i, the i-th argument had an illegal value.
00174 *          = 1,...,N:
00175 *                The QZ iteration failed.  No eigenvectors have been
00176 *                calculated, but ALPHAR(j), ALPHAI(j), and BETA(j)
00177 *                should be correct for j=INFO+1,...,N.
00178 *          > N:  errors that usually indicate LAPACK problems:
00179 *                =N+1: error return from SGGBAL
00180 *                =N+2: error return from SGEQRF
00181 *                =N+3: error return from SORMQR
00182 *                =N+4: error return from SORGQR
00183 *                =N+5: error return from SGGHRD
00184 *                =N+6: error return from SHGEQZ (other than failed
00185 *                                                iteration)
00186 *                =N+7: error return from STGEVC
00187 *                =N+8: error return from SGGBAK (computing VL)
00188 *                =N+9: error return from SGGBAK (computing VR)
00189 *                =N+10: error return from SLASCL (various calls)
00190 *
00191 *  Further Details
00192 *  ===============
00193 *
00194 *  Balancing
00195 *  ---------
00196 *
00197 *  This driver calls SGGBAL to both permute and scale rows and columns
00198 *  of A and B.  The permutations PL and PR are chosen so that PL*A*PR
00199 *  and PL*B*R will be upper triangular except for the diagonal blocks
00200 *  A(i:j,i:j) and B(i:j,i:j), with i and j as close together as
00201 *  possible.  The diagonal scaling matrices DL and DR are chosen so
00202 *  that the pair  DL*PL*A*PR*DR, DL*PL*B*PR*DR have elements close to
00203 *  one (except for the elements that start out zero.)
00204 *
00205 *  After the eigenvalues and eigenvectors of the balanced matrices
00206 *  have been computed, SGGBAK transforms the eigenvectors back to what
00207 *  they would have been (in perfect arithmetic) if they had not been
00208 *  balanced.
00209 *
00210 *  Contents of A and B on Exit
00211 *  -------- -- - --- - -- ----
00212 *
00213 *  If any eigenvectors are computed (either JOBVL='V' or JOBVR='V' or
00214 *  both), then on exit the arrays A and B will contain the real Schur
00215 *  form[*] of the "balanced" versions of A and B.  If no eigenvectors
00216 *  are computed, then only the diagonal blocks will be correct.
00217 *
00218 *  [*] See SHGEQZ, SGEGS, or read the book "Matrix Computations",
00219 *      by Golub & van Loan, pub. by Johns Hopkins U. Press.
00220 *
00221 *  =====================================================================
00222 *
00223 *     .. Parameters ..
00224       REAL               ZERO, ONE
00225       PARAMETER          ( ZERO = 0.0E0, ONE = 1.0E0 )
00226 *     ..
00227 *     .. Local Scalars ..
00228       LOGICAL            ILIMIT, ILV, ILVL, ILVR, LQUERY
00229       CHARACTER          CHTEMP
00230       INTEGER            ICOLS, IHI, IINFO, IJOBVL, IJOBVR, ILEFT, ILO,
00231      $                   IN, IRIGHT, IROWS, ITAU, IWORK, JC, JR, LOPT,
00232      $                   LWKMIN, LWKOPT, NB, NB1, NB2, NB3
00233       REAL               ABSAI, ABSAR, ABSB, ANRM, ANRM1, ANRM2, BNRM,
00234      $                   BNRM1, BNRM2, EPS, ONEPLS, SAFMAX, SAFMIN,
00235      $                   SALFAI, SALFAR, SBETA, SCALE, TEMP
00236 *     ..
00237 *     .. Local Arrays ..
00238       LOGICAL            LDUMMA( 1 )
00239 *     ..
00240 *     .. External Subroutines ..
00241       EXTERNAL           SGEQRF, SGGBAK, SGGBAL, SGGHRD, SHGEQZ, SLACPY,
00242      $                   SLASCL, SLASET, SORGQR, SORMQR, STGEVC, XERBLA
00243 *     ..
00244 *     .. External Functions ..
00245       LOGICAL            LSAME
00246       INTEGER            ILAENV
00247       REAL               SLAMCH, SLANGE
00248       EXTERNAL           ILAENV, LSAME, SLAMCH, SLANGE
00249 *     ..
00250 *     .. Intrinsic Functions ..
00251       INTRINSIC          ABS, INT, MAX
00252 *     ..
00253 *     .. Executable Statements ..
00254 *
00255 *     Decode the input arguments
00256 *
00257       IF( LSAME( JOBVL, 'N' ) ) THEN
00258          IJOBVL = 1
00259          ILVL = .FALSE.
00260       ELSE IF( LSAME( JOBVL, 'V' ) ) THEN
00261          IJOBVL = 2
00262          ILVL = .TRUE.
00263       ELSE
00264          IJOBVL = -1
00265          ILVL = .FALSE.
00266       END IF
00267 *
00268       IF( LSAME( JOBVR, 'N' ) ) THEN
00269          IJOBVR = 1
00270          ILVR = .FALSE.
00271       ELSE IF( LSAME( JOBVR, 'V' ) ) THEN
00272          IJOBVR = 2
00273          ILVR = .TRUE.
00274       ELSE
00275          IJOBVR = -1
00276          ILVR = .FALSE.
00277       END IF
00278       ILV = ILVL .OR. ILVR
00279 *
00280 *     Test the input arguments
00281 *
00282       LWKMIN = MAX( 8*N, 1 )
00283       LWKOPT = LWKMIN
00284       WORK( 1 ) = LWKOPT
00285       LQUERY = ( LWORK.EQ.-1 )
00286       INFO = 0
00287       IF( IJOBVL.LE.0 ) THEN
00288          INFO = -1
00289       ELSE IF( IJOBVR.LE.0 ) THEN
00290          INFO = -2
00291       ELSE IF( N.LT.0 ) THEN
00292          INFO = -3
00293       ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
00294          INFO = -5
00295       ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
00296          INFO = -7
00297       ELSE IF( LDVL.LT.1 .OR. ( ILVL .AND. LDVL.LT.N ) ) THEN
00298          INFO = -12
00299       ELSE IF( LDVR.LT.1 .OR. ( ILVR .AND. LDVR.LT.N ) ) THEN
00300          INFO = -14
00301       ELSE IF( LWORK.LT.LWKMIN .AND. .NOT.LQUERY ) THEN
00302          INFO = -16
00303       END IF
00304 *
00305       IF( INFO.EQ.0 ) THEN
00306          NB1 = ILAENV( 1, 'SGEQRF', ' ', N, N, -1, -1 )
00307          NB2 = ILAENV( 1, 'SORMQR', ' ', N, N, N, -1 )
00308          NB3 = ILAENV( 1, 'SORGQR', ' ', N, N, N, -1 )
00309          NB = MAX( NB1, NB2, NB3 )
00310          LOPT = 2*N + MAX( 6*N, N*(NB+1) )
00311          WORK( 1 ) = LOPT
00312       END IF
00313 *
00314       IF( INFO.NE.0 ) THEN
00315          CALL XERBLA( 'SGEGV ', -INFO )
00316          RETURN
00317       ELSE IF( LQUERY ) THEN
00318          RETURN
00319       END IF
00320 *
00321 *     Quick return if possible
00322 *
00323       IF( N.EQ.0 )
00324      $   RETURN
00325 *
00326 *     Get machine constants
00327 *
00328       EPS = SLAMCH( 'E' )*SLAMCH( 'B' )
00329       SAFMIN = SLAMCH( 'S' )
00330       SAFMIN = SAFMIN + SAFMIN
00331       SAFMAX = ONE / SAFMIN
00332       ONEPLS = ONE + ( 4*EPS )
00333 *
00334 *     Scale A
00335 *
00336       ANRM = SLANGE( 'M', N, N, A, LDA, WORK )
00337       ANRM1 = ANRM
00338       ANRM2 = ONE
00339       IF( ANRM.LT.ONE ) THEN
00340          IF( SAFMAX*ANRM.LT.ONE ) THEN
00341             ANRM1 = SAFMIN
00342             ANRM2 = SAFMAX*ANRM
00343          END IF
00344       END IF
00345 *
00346       IF( ANRM.GT.ZERO ) THEN
00347          CALL SLASCL( 'G', -1, -1, ANRM, ONE, N, N, A, LDA, IINFO )
00348          IF( IINFO.NE.0 ) THEN
00349             INFO = N + 10
00350             RETURN
00351          END IF
00352       END IF
00353 *
00354 *     Scale B
00355 *
00356       BNRM = SLANGE( 'M', N, N, B, LDB, WORK )
00357       BNRM1 = BNRM
00358       BNRM2 = ONE
00359       IF( BNRM.LT.ONE ) THEN
00360          IF( SAFMAX*BNRM.LT.ONE ) THEN
00361             BNRM1 = SAFMIN
00362             BNRM2 = SAFMAX*BNRM
00363          END IF
00364       END IF
00365 *
00366       IF( BNRM.GT.ZERO ) THEN
00367          CALL SLASCL( 'G', -1, -1, BNRM, ONE, N, N, B, LDB, IINFO )
00368          IF( IINFO.NE.0 ) THEN
00369             INFO = N + 10
00370             RETURN
00371          END IF
00372       END IF
00373 *
00374 *     Permute the matrix to make it more nearly triangular
00375 *     Workspace layout:  (8*N words -- "work" requires 6*N words)
00376 *        left_permutation, right_permutation, work...
00377 *
00378       ILEFT = 1
00379       IRIGHT = N + 1
00380       IWORK = IRIGHT + N
00381       CALL SGGBAL( 'P', N, A, LDA, B, LDB, ILO, IHI, WORK( ILEFT ),
00382      $             WORK( IRIGHT ), WORK( IWORK ), IINFO )
00383       IF( IINFO.NE.0 ) THEN
00384          INFO = N + 1
00385          GO TO 120
00386       END IF
00387 *
00388 *     Reduce B to triangular form, and initialize VL and/or VR
00389 *     Workspace layout:  ("work..." must have at least N words)
00390 *        left_permutation, right_permutation, tau, work...
00391 *
00392       IROWS = IHI + 1 - ILO
00393       IF( ILV ) THEN
00394          ICOLS = N + 1 - ILO
00395       ELSE
00396          ICOLS = IROWS
00397       END IF
00398       ITAU = IWORK
00399       IWORK = ITAU + IROWS
00400       CALL SGEQRF( IROWS, ICOLS, B( ILO, ILO ), LDB, WORK( ITAU ),
00401      $             WORK( IWORK ), LWORK+1-IWORK, IINFO )
00402       IF( IINFO.GE.0 )
00403      $   LWKOPT = MAX( LWKOPT, INT( WORK( IWORK ) )+IWORK-1 )
00404       IF( IINFO.NE.0 ) THEN
00405          INFO = N + 2
00406          GO TO 120
00407       END IF
00408 *
00409       CALL SORMQR( 'L', 'T', IROWS, ICOLS, IROWS, B( ILO, ILO ), LDB,
00410      $             WORK( ITAU ), A( ILO, ILO ), LDA, WORK( IWORK ),
00411      $             LWORK+1-IWORK, IINFO )
00412       IF( IINFO.GE.0 )
00413      $   LWKOPT = MAX( LWKOPT, INT( WORK( IWORK ) )+IWORK-1 )
00414       IF( IINFO.NE.0 ) THEN
00415          INFO = N + 3
00416          GO TO 120
00417       END IF
00418 *
00419       IF( ILVL ) THEN
00420          CALL SLASET( 'Full', N, N, ZERO, ONE, VL, LDVL )
00421          CALL SLACPY( 'L', IROWS-1, IROWS-1, B( ILO+1, ILO ), LDB,
00422      $                VL( ILO+1, ILO ), LDVL )
00423          CALL SORGQR( IROWS, IROWS, IROWS, VL( ILO, ILO ), LDVL,
00424      $                WORK( ITAU ), WORK( IWORK ), LWORK+1-IWORK,
00425      $                IINFO )
00426          IF( IINFO.GE.0 )
00427      $      LWKOPT = MAX( LWKOPT, INT( WORK( IWORK ) )+IWORK-1 )
00428          IF( IINFO.NE.0 ) THEN
00429             INFO = N + 4
00430             GO TO 120
00431          END IF
00432       END IF
00433 *
00434       IF( ILVR )
00435      $   CALL SLASET( 'Full', N, N, ZERO, ONE, VR, LDVR )
00436 *
00437 *     Reduce to generalized Hessenberg form
00438 *
00439       IF( ILV ) THEN
00440 *
00441 *        Eigenvectors requested -- work on whole matrix.
00442 *
00443          CALL SGGHRD( JOBVL, JOBVR, N, ILO, IHI, A, LDA, B, LDB, VL,
00444      $                LDVL, VR, LDVR, IINFO )
00445       ELSE
00446          CALL SGGHRD( 'N', 'N', IROWS, 1, IROWS, A( ILO, ILO ), LDA,
00447      $                B( ILO, ILO ), LDB, VL, LDVL, VR, LDVR, IINFO )
00448       END IF
00449       IF( IINFO.NE.0 ) THEN
00450          INFO = N + 5
00451          GO TO 120
00452       END IF
00453 *
00454 *     Perform QZ algorithm
00455 *     Workspace layout:  ("work..." must have at least 1 word)
00456 *        left_permutation, right_permutation, work...
00457 *
00458       IWORK = ITAU
00459       IF( ILV ) THEN
00460          CHTEMP = 'S'
00461       ELSE
00462          CHTEMP = 'E'
00463       END IF
00464       CALL SHGEQZ( CHTEMP, JOBVL, JOBVR, N, ILO, IHI, A, LDA, B, LDB,
00465      $             ALPHAR, ALPHAI, BETA, VL, LDVL, VR, LDVR,
00466      $             WORK( IWORK ), LWORK+1-IWORK, IINFO )
00467       IF( IINFO.GE.0 )
00468      $   LWKOPT = MAX( LWKOPT, INT( WORK( IWORK ) )+IWORK-1 )
00469       IF( IINFO.NE.0 ) THEN
00470          IF( IINFO.GT.0 .AND. IINFO.LE.N ) THEN
00471             INFO = IINFO
00472          ELSE IF( IINFO.GT.N .AND. IINFO.LE.2*N ) THEN
00473             INFO = IINFO - N
00474          ELSE
00475             INFO = N + 6
00476          END IF
00477          GO TO 120
00478       END IF
00479 *
00480       IF( ILV ) THEN
00481 *
00482 *        Compute Eigenvectors  (STGEVC requires 6*N words of workspace)
00483 *
00484          IF( ILVL ) THEN
00485             IF( ILVR ) THEN
00486                CHTEMP = 'B'
00487             ELSE
00488                CHTEMP = 'L'
00489             END IF
00490          ELSE
00491             CHTEMP = 'R'
00492          END IF
00493 *
00494          CALL STGEVC( CHTEMP, 'B', LDUMMA, N, A, LDA, B, LDB, VL, LDVL,
00495      $                VR, LDVR, N, IN, WORK( IWORK ), IINFO )
00496          IF( IINFO.NE.0 ) THEN
00497             INFO = N + 7
00498             GO TO 120
00499          END IF
00500 *
00501 *        Undo balancing on VL and VR, rescale
00502 *
00503          IF( ILVL ) THEN
00504             CALL SGGBAK( 'P', 'L', N, ILO, IHI, WORK( ILEFT ),
00505      $                   WORK( IRIGHT ), N, VL, LDVL, IINFO )
00506             IF( IINFO.NE.0 ) THEN
00507                INFO = N + 8
00508                GO TO 120
00509             END IF
00510             DO 50 JC = 1, N
00511                IF( ALPHAI( JC ).LT.ZERO )
00512      $            GO TO 50
00513                TEMP = ZERO
00514                IF( ALPHAI( JC ).EQ.ZERO ) THEN
00515                   DO 10 JR = 1, N
00516                      TEMP = MAX( TEMP, ABS( VL( JR, JC ) ) )
00517    10             CONTINUE
00518                ELSE
00519                   DO 20 JR = 1, N
00520                      TEMP = MAX( TEMP, ABS( VL( JR, JC ) )+
00521      $                      ABS( VL( JR, JC+1 ) ) )
00522    20             CONTINUE
00523                END IF
00524                IF( TEMP.LT.SAFMIN )
00525      $            GO TO 50
00526                TEMP = ONE / TEMP
00527                IF( ALPHAI( JC ).EQ.ZERO ) THEN
00528                   DO 30 JR = 1, N
00529                      VL( JR, JC ) = VL( JR, JC )*TEMP
00530    30             CONTINUE
00531                ELSE
00532                   DO 40 JR = 1, N
00533                      VL( JR, JC ) = VL( JR, JC )*TEMP
00534                      VL( JR, JC+1 ) = VL( JR, JC+1 )*TEMP
00535    40             CONTINUE
00536                END IF
00537    50       CONTINUE
00538          END IF
00539          IF( ILVR ) THEN
00540             CALL SGGBAK( 'P', 'R', N, ILO, IHI, WORK( ILEFT ),
00541      $                   WORK( IRIGHT ), N, VR, LDVR, IINFO )
00542             IF( IINFO.NE.0 ) THEN
00543                INFO = N + 9
00544                GO TO 120
00545             END IF
00546             DO 100 JC = 1, N
00547                IF( ALPHAI( JC ).LT.ZERO )
00548      $            GO TO 100
00549                TEMP = ZERO
00550                IF( ALPHAI( JC ).EQ.ZERO ) THEN
00551                   DO 60 JR = 1, N
00552                      TEMP = MAX( TEMP, ABS( VR( JR, JC ) ) )
00553    60             CONTINUE
00554                ELSE
00555                   DO 70 JR = 1, N
00556                      TEMP = MAX( TEMP, ABS( VR( JR, JC ) )+
00557      $                      ABS( VR( JR, JC+1 ) ) )
00558    70             CONTINUE
00559                END IF
00560                IF( TEMP.LT.SAFMIN )
00561      $            GO TO 100
00562                TEMP = ONE / TEMP
00563                IF( ALPHAI( JC ).EQ.ZERO ) THEN
00564                   DO 80 JR = 1, N
00565                      VR( JR, JC ) = VR( JR, JC )*TEMP
00566    80             CONTINUE
00567                ELSE
00568                   DO 90 JR = 1, N
00569                      VR( JR, JC ) = VR( JR, JC )*TEMP
00570                      VR( JR, JC+1 ) = VR( JR, JC+1 )*TEMP
00571    90             CONTINUE
00572                END IF
00573   100       CONTINUE
00574          END IF
00575 *
00576 *        End of eigenvector calculation
00577 *
00578       END IF
00579 *
00580 *     Undo scaling in alpha, beta
00581 *
00582 *     Note: this does not give the alpha and beta for the unscaled
00583 *     problem.
00584 *
00585 *     Un-scaling is limited to avoid underflow in alpha and beta
00586 *     if they are significant.
00587 *
00588       DO 110 JC = 1, N
00589          ABSAR = ABS( ALPHAR( JC ) )
00590          ABSAI = ABS( ALPHAI( JC ) )
00591          ABSB = ABS( BETA( JC ) )
00592          SALFAR = ANRM*ALPHAR( JC )
00593          SALFAI = ANRM*ALPHAI( JC )
00594          SBETA = BNRM*BETA( JC )
00595          ILIMIT = .FALSE.
00596          SCALE = ONE
00597 *
00598 *        Check for significant underflow in ALPHAI
00599 *
00600          IF( ABS( SALFAI ).LT.SAFMIN .AND. ABSAI.GE.
00601      $       MAX( SAFMIN, EPS*ABSAR, EPS*ABSB ) ) THEN
00602             ILIMIT = .TRUE.
00603             SCALE = ( ONEPLS*SAFMIN / ANRM1 ) /
00604      $              MAX( ONEPLS*SAFMIN, ANRM2*ABSAI )
00605 *
00606          ELSE IF( SALFAI.EQ.ZERO ) THEN
00607 *
00608 *           If insignificant underflow in ALPHAI, then make the
00609 *           conjugate eigenvalue real.
00610 *
00611             IF( ALPHAI( JC ).LT.ZERO .AND. JC.GT.1 ) THEN
00612                ALPHAI( JC-1 ) = ZERO
00613             ELSE IF( ALPHAI( JC ).GT.ZERO .AND. JC.LT.N ) THEN
00614                ALPHAI( JC+1 ) = ZERO
00615             END IF
00616          END IF
00617 *
00618 *        Check for significant underflow in ALPHAR
00619 *
00620          IF( ABS( SALFAR ).LT.SAFMIN .AND. ABSAR.GE.
00621      $       MAX( SAFMIN, EPS*ABSAI, EPS*ABSB ) ) THEN
00622             ILIMIT = .TRUE.
00623             SCALE = MAX( SCALE, ( ONEPLS*SAFMIN / ANRM1 ) /
00624      $              MAX( ONEPLS*SAFMIN, ANRM2*ABSAR ) )
00625          END IF
00626 *
00627 *        Check for significant underflow in BETA
00628 *
00629          IF( ABS( SBETA ).LT.SAFMIN .AND. ABSB.GE.
00630      $       MAX( SAFMIN, EPS*ABSAR, EPS*ABSAI ) ) THEN
00631             ILIMIT = .TRUE.
00632             SCALE = MAX( SCALE, ( ONEPLS*SAFMIN / BNRM1 ) /
00633      $              MAX( ONEPLS*SAFMIN, BNRM2*ABSB ) )
00634          END IF
00635 *
00636 *        Check for possible overflow when limiting scaling
00637 *
00638          IF( ILIMIT ) THEN
00639             TEMP = ( SCALE*SAFMIN )*MAX( ABS( SALFAR ), ABS( SALFAI ),
00640      $             ABS( SBETA ) )
00641             IF( TEMP.GT.ONE )
00642      $         SCALE = SCALE / TEMP
00643             IF( SCALE.LT.ONE )
00644      $         ILIMIT = .FALSE.
00645          END IF
00646 *
00647 *        Recompute un-scaled ALPHAR, ALPHAI, BETA if necessary.
00648 *
00649          IF( ILIMIT ) THEN
00650             SALFAR = ( SCALE*ALPHAR( JC ) )*ANRM
00651             SALFAI = ( SCALE*ALPHAI( JC ) )*ANRM
00652             SBETA = ( SCALE*BETA( JC ) )*BNRM
00653          END IF
00654          ALPHAR( JC ) = SALFAR
00655          ALPHAI( JC ) = SALFAI
00656          BETA( JC ) = SBETA
00657   110 CONTINUE
00658 *
00659   120 CONTINUE
00660       WORK( 1 ) = LWKOPT
00661 *
00662       RETURN
00663 *
00664 *     End of SGEGV
00665 *
00666       END