[CPMD-list] ADJMU| BISECTION COULD NOT

[Eung-Gun Kim]

Dear List Subscribers,

When I calculate eigenvalues of individual states, I normally perform
wavefunction optimization using ODIIS and then do one step of LANCZOS diago.  I
haven't had any trouble doing this in the past (with v3.9.1 on IBM P4), but
lately I found myself in the following situation when certain values of KPOINTS
are used:

   1) All calculations go smoothly with ODIIS, independent of KPOINTS.
   2) Restarting with the LANCZOS diago goes fine all the way through the very
      last k point, and then crashes with
      "...SUBROUTINE ADJMU| BISECTION COULD NOT CONVERGE...".
   3) This error sometimes goes away when k point grid ratios are adjusted, say,
      from 2 8 14 to either 2 8 16 or 1 4 8.
   4) Similar errors were also encountered in a system with different contents
      and supercell size.
   5) The error message complains about a small number of STATES. I feed here
      with the total number of filled states, and a larger number is not allowed
      by the code when switching from ODIIS to LANCZOS.  And I don't see a
      reason why I need a larger STATES for my organic crystals with sizable
      band gaps.

I am attaching portions of the input and output files of a representative diago
run below for your reference.  Thanks very much for reading.

Regards,

EG Kim

###############################################################################
input file --------------------------------------------------------------------

&CPMD
  RESTART COORDINATES WAVEFUNCTION
  OPTIMIZE WAVEFUNCTION
  LANCZOS DIAGONALISATION
  LANCZOS PARAMETER
    200  8  0  1.D-8
  STORE WAVEFUNCTIONS OFF
  MAXSTEP
    1
&END

&DFT
  FUNCTIONAL BLYP
&END

&SYSTEM
  ANGSTROM
  SYMMETRY
    ORTHORHOMBIC
  CELL ABSOLUTE DEGREE
    7.92534743  10.6994  6.8   90.0 90.0 90.0
  CUTOFF
    70
  CHARGE
    0
  STATES
    92
  KPOINTS MONKHORST-PACK
    6  4  6
......
........

output file -------------------------------------------------------------------

........
......

<<36:36<<<<<<<<<<<< LANCZOS DIAGONALIZATION <<<<<<<<<<<<<<<<<<<<
 >> TIME FOR INITIAL SUBSPACE DIAGONALIZATION:                .70
 >> CYCLE     NCONV        B2MAX        B2MIN     #HPSI      TIME
        1        92    2.248E-13    1.633E-15      1.00       .11
  ADJMU| EIGENVALUES:
 1 1 -0.744303257893736525
 1 2 -0.744207855529486761
 1 3 -0.744131382450865297
 1 4 -0.744046961929959005
 1 5 -0.713127925040847876
 1 6 -0.713072937556081698
 1 7 -0.712971653074167544
 1 8 -0.712869568517556229

....
....
....

36 92 0.123545374312123368
         AMU1= -0.745159946143003404        RHINT1= -183.944444444444457
         AMU2= 0.133314578748302381        RHINT2= -0.960653778747655451E-11
         DAMU= 0.878474524891305730        RHINT = -183.944444444444457
         RHINT-NEL= -183.944444444444457
  ADJMU! THE NUMBER OF STATES [ 92 ] IS TOO SMALL


 PROGRAM STOPS IN SUBROUTINE ADJMU| BISECTION COULD NOT CONVERGE [PROC=   0]