DLARRB2(3) ScaLAPACK routine of NEC Numeric Library Collection DLARRB2(3) NAME DLARRB2 - does "limited" bisection to refine the eigenvalues of L D L^T, W( IFIRST-OFFSET ) through W( ILAST-OFFSET ), to more accuracy SYNOPSIS SUBROUTINE DLARRB2( N, D, LLD, IFIRST, ILAST, RTOL1, RTOL2, OFFSET, W, WGAP, WERR, WORK, IWORK, PIVMIN, LGPVMN, LGSPDM, TWIST, INFO ) INTEGER IFIRST, ILAST, INFO, N, OFFSET, TWIST DOUBLE PRECISION LGPVMN, LGSPDM, PIVMIN, RTOL1, RTOL2 INTEGER IWORK( * ) DOUBLE PRECISION D( * ), LLD( * ), W( * ), WERR( * ), WGAP( * ), WORK( * ) PURPOSE Given the relatively robust representation(RRR) L D L^T, DLARRB2 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. NOTE: There are very few minor differences between DLARRB from LAPACK and this current subroutine DLARRB2. The most important reason for creating this nearly identical copy is profiling: in the ScaLAPACK MRRR algorithm, eigenvalue computation using DLARRB2 is used for refinement in the construction of the repre- sentation tree, as opposed to the initial computation of the eigenval- ues for the root RRR which uses DLARRB. When profiling, this allows an easy quantification of refinement work vs. computing eigenvalues of the root. ARGUMENTS N (input) INTEGER The order of the matrix. D (input) DOUBLE PRECISION array, dimension (N) The N diagonal elements of the diagonal matrix D. LLD (input) DOUBLE PRECISION array, dimension (N-1) The (N-1) elements L(i)*L(i)*D(i). IFIRST (input) INTEGER The index of the first eigenvalue to be computed. ILAST (input) INTEGER The index of the last eigenvalue to be computed. RTOL1 (input) DOUBLE PRECISION RTOL2 (input) DOUBLE PRECISION Tolerance for the convergence of the bisection intervals. An interval [LEFT,RIGHT] has converged if RIGHT-LEFT.LT.MAX( RTOL1*GAP, RTOL2*MAX(|LEFT|,|RIGHT|) ) where GAP is the (esti- mated) distance to the nearest eigenvalue. OFFSET (input) 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. W (input/output) 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. WGAP (input/output) 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.EQ.ILAST then WGAP(IFIRST-OFF- SET) must be set to ZERO. On output, these gaps are refined. WERR (input/output) 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. WORK (workspace) DOUBLE PRECISION array, dimension (4*N) Workspace. IWORK (workspace) INTEGER array, dimension (2*N) Workspace. PIVMIN (input) DOUBLE PRECISION The minimum pivot in the sturm sequence. LGPVMN (input) DOUBLE PRECISION Logarithm of PIVMIN, precomputed. LGSPDM (input) DOUBLE PRECISION Logarithm of the spectral diameter, precomputed. TWIST (input) 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) INFO (output) INTEGER Error flag. ScaLAPACK routine 31 October 2017 DLARRB2(3)