SYNOPSIS

SUBROUTINE PDSYEV(

JOBZ, UPLO, N, A, IA, JA, DESCA, W, Z, IZ, JZ, DESCZ, WORK, LWORK, INFO )

    

CHARACTER JOBZ, UPLO

    

INTEGER IA, INFO, IZ, JA, JZ, LWORK, N

    

INTEGER DESCA( * ), DESCZ( * )

    

DOUBLE PRECISION A( * ), W( * ), WORK( * ), Z( * )

    

INTEGER BLOCK_CYCLIC_2D, DLEN_, DTYPE_, CTXT_, M_, N_, MB_, NB_, RSRC_, CSRC_, LLD_

    

PARAMETER ( BLOCK_CYCLIC_2D = 1, DLEN_ = 9, DTYPE_ = 1, CTXT_ = 2, M_ = 3, N_ = 4, MB_ = 5, NB_ = 6, RSRC_ = 7, CSRC_ = 8, LLD_ = 9 )

    

DOUBLE PRECISION ZERO, ONE

    

PARAMETER ( ZERO = 0.0D+0, ONE = 1.0D+0 )

    

INTEGER ITHVAL

    

PARAMETER ( ITHVAL = 10 )

    

LOGICAL LOWER, WANTZ

    

INTEGER CONTEXTC, CSRC_A, I, IACOL, IAROW, ICOFFA, IINFO, INDD, INDD2, INDE, INDE2, INDTAU, INDWORK, INDWORK2, IROFFA, IROFFZ, ISCALE, IZROW, J, K, LCM, LCMQ, LDC, LLWORK, LWMIN, MB_A, MB_Z, MYCOL, MYPCOLC, MYPROWC, MYROW, NB, NB_A, NB_Z, NN, NP, NPCOL, NPCOLC, NPROCS, NPROW, NPROWC, NQ, NRC, QRMEM, RSRC_A, RSRC_Z, SIZEMQRLEFT, SIZEMQRRIGHT

    

DOUBLE PRECISION ANRM, BIGNUM, EPS, RMAX, RMIN, SAFMIN, SIGMA, SMLNUM

    

INTEGER DESCQR( 10 ), IDUM1( 3 ), IDUM2( 3 )

    

LOGICAL LSAME

    

INTEGER ILCM, INDXG2P, NUMROC, SL_GRIDRESHAPE

    

DOUBLE PRECISION PDLAMCH, PDLANSY

    

EXTERNAL LSAME, ILCM, INDXG2P, NUMROC, SL_GRIDRESHAPE, PDLAMCH, PDLANSY

    

EXTERNAL BLACS_GRIDEXIT, BLACS_GRIDINFO, CHK1MAT, DCOPY, DESCINIT, DGAMN2D, DGAMX2D, DSCAL, DSTEQR2, PCHK2MAT, PDELGET, PDGEMR2D, PDLASCL, PDLASET, PDORMTR, PDSYTRD, PXERBLA

    

INTRINSIC DBLE, ICHAR, MAX, MIN, MOD, SQRT

PURPOSE

PDSYEV computes all eigenvalues and, optionally, eigenvectors of a real symmetric matrix A by calling the recommended sequence of ScaLAPACK routines.

In its present form, PDSYEV assumes a homogeneous system and makes no checks for consistency of the eigenvalues or eigenvectors across the different processes. Because of this, it is possible that a heterogeneous system may return incorrect results without any error messages.

NOTES

A description vector is associated with each 2D block-cyclicly dis- tributed matrix. This vector stores the information required to establish the mapping between a matrix entry and its corresponding process and memory location.

In the following comments, the character _ should be read as "of the distributed matrix". Let A be a generic term for any 2D block cyclicly distributed matrix. Its description vector is DESCA:

NOTATION STORED IN EXPLANATION
--------------- -------------- --------------------------------------
DTYPE_A(global) DESCA( DTYPE_) The descriptor type.
CTXT_A (global) DESCA( CTXT_ ) The BLACS context handle, indicating
                               the BLACS process grid A is distribu-
                               ted over. The context itself is glo-
                               bal, but the handle (the integer
                               value) may vary.
M_A (global) DESCA( M_ ) The number of rows in the distributed
                               matrix A.
N_A (global) DESCA( N_ ) The number of columns in the distri-
                               buted matrix A.
MB_A (global) DESCA( MB_ ) The blocking factor used to distribute
                               the rows of A.
NB_A (global) DESCA( NB_ ) The blocking factor used to distribute
                               the columns of A.
RSRC_A (global) DESCA( RSRC_ ) The process row over which the first
                               row of the matrix A is distributed.
CSRC_A (global) DESCA( CSRC_ ) The process column over which the
                               first column of A is distributed.
LLD_A (local) DESCA( LLD_ ) The leading dimension of the local
                               array storing the local blocks of the
                               distributed matrix A.
                               LLD_A >= MAX(1,LOCr(M_A)).

Let K be the number of rows or columns of a distributed matrix, and assume that its process grid has dimension p x q. LOCr( K ) denotes the number of elements of K that a process would receive if K were distributed over the p processes of its process column.
Similarly, LOCc( K ) denotes the number of elements of K that a process would receive if K were distributed over the q processes of its process row.
The values of LOCr() and LOCc() may be determined via a call to the ScaLAPACK tool function, NUMROC:

        LOCr( M ) = NUMROC( M, MB_A, MYROW, RSRC_A, NPROW ),

        LOCc( N ) = NUMROC( N, NB_A, MYCOL, CSRC_A, NPCOL ).

ARGUMENTS

   NP = the number of rows local to a given process.
   NQ = the number of columns local to a given process.

JOBZ (global input) CHARACTER*1
        Specifies whether or not to compute the eigenvectors:
        = 'N':  Compute eigenvalues only.
        = 'V':  Compute eigenvalues and eigenvectors.

UPLO (global input) CHARACTER*1
        Specifies whether the upper or lower triangular part of the
        symmetric matrix A is stored:
        = 'U':  Upper triangular
        = 'L':  Lower triangular

N (global input) INTEGER
        The number of rows and columns of the matrix A.  N >= 0.

A (local input/workspace) block cyclic DOUBLE PRECISION array,
        global dimension (N, N), local dimension ( LLD_A,
        LOCc(JA+N-1) )

        On entry, the symmetric matrix A.  If UPLO = 'U', only the
        upper triangular part of A is used to define the elements of
        the symmetric matrix.  If UPLO = 'L', only the lower
        triangular part of A is used to define the elements of the
        symmetric matrix.

        On exit, the lower triangle (if UPLO='L') or the upper
        triangle (if UPLO='U') of A, including the diagonal, is
        destroyed.

IA (global input) INTEGER
        A's global row index, which points to the beginning of the
        submatrix which is to be operated on.

JA (global input) INTEGER
        A's global column index, which points to the beginning of
        the submatrix which is to be operated on.

DESCA (global and local input) INTEGER array of dimension DLEN_.
        The array descriptor for the distributed matrix A.
        If DESCA( CTXT_ ) is incorrect, PDSYEV cannot guarantee
        correct error reporting.

W (global output) DOUBLE PRECISION array, dimension (N)
        On normal exit, the first M entries contain the selected
        eigenvalues in ascending order.

Z (local output) DOUBLE PRECISION array,
        global dimension (N, N),
        local dimension ( LLD_Z, LOCc(JZ+N-1) )
        If JOBZ = 'V', then on normal exit the first M columns of Z
        contain the orthonormal eigenvectors of the matrix
        corresponding to the selected eigenvalues.
        If JOBZ = 'N', then Z is not referenced.

IZ (global input) INTEGER
        Z's global row index, which points to the beginning of the
        submatrix which is to be operated on.

JZ (global input) INTEGER
        Z's global column index, which points to the beginning of
        the submatrix which is to be operated on.

DESCZ (global and local input) INTEGER array of dimension DLEN_.
        The array descriptor for the distributed matrix Z.
        DESCZ( CTXT_ ) must equal DESCA( CTXT_ )

WORK (local workspace/output) DOUBLE PRECISION array,
        dimension (LWORK)
        Version 1.0:  on output, WORK(1) returns the workspace
        needed to guarantee completion.
        If the input parameters are incorrect, WORK(1) may also be
        incorrect.

        If JOBZ='N' WORK(1) = minimal=optimal amount of workspace
        If JOBZ='V' WORK(1) = minimal workspace required to
           generate all the eigenvectors.

LWORK (local input) INTEGER
        See below for definitions of variables used to define LWORK.
        If no eigenvectors are requested (JOBZ = 'N') then
           LWORK >= 5*N + SIZESYTRD + 1
        where
           SIZESYTRD = The workspace requirement for PDSYTRD
                       and is MAX( NB * ( NP +1 ), 3 * NB )
        If eigenvectors are requested (JOBZ = 'V' ) then
           the amount of workspace required to guarantee that all
           eigenvectors are computed is:

           QRMEM = 2*N-2
           LWMIN = 5*N + N*LDC + MAX( SIZEMQRLEFT, QRMEM ) + 1

        Variable definitions:
           NB = DESCA( MB_ ) = DESCA( NB_ ) =
                DESCZ( MB_ ) = DESCZ( NB_ )
           NN = MAX( N, NB, 2 )
           DESCA( RSRC_ ) = DESCA( RSRC_ ) = DESCZ( RSRC_ ) =
                            DESCZ( CSRC_ ) = 0
           NP = NUMROC( NN, NB, 0, 0, NPROW )
           NQ = NUMROC( MAX( N, NB, 2 ), NB, 0, 0, NPCOL )
           NRC = NUMROC( N, NB, MYPROWC, 0, NPROCS)
           LDC = MAX( 1, NRC )
           SIZEMQRLEFT = The workspace requirement for PDORMTR
                         when it's SIDE argument is 'L'.

        With MYPROWC defined when a new context is created as:
           CALL BLACS_GET( DESCA( CTXT_ ), 0, CONTEXTC )
           CALL BLACS_GRIDINIT( CONTEXTC, 'R', NPROCS, 1 )
           CALL BLACS_GRIDINFO( CONTEXTC, NPROWC, NPCOLC, MYPROWC,
                                MYPCOLC )

        If LWORK = -1, the LWORK is global input and a workspace
        query is assumed; the routine only calculates the minimum
        size for the WORK array.  The required workspace is returned
        as the first element of WORK and no error message is issued
        by PXERBLA.

INFO (global output) INTEGER
        = 0:  successful exit
        < 0:  If the i-th argument is an array and the j-entry had
              an illegal value, then INFO = -(i*100+j), if the i-th
              argument is a scalar and had an illegal value, then
              INFO = -i.
        > 0:  If INFO = 1 through N, the i(th) eigenvalue did not
              converge in DSTEQR2 after a total of 30*N iterations.
              If INFO = N+1, then PDSYEV has detected heterogeneity
              by finding that eigenvalues were not identical across
              the process grid.  In this case, the accuracy of
              the results from PDSYEV cannot be guaranteed.