[QE-users] help with turbo-davidson not converging

[balducci at units.it]

hello

I'm having convergence problems with a turbo-davidson calculation on
an isolated (neutral) molecule of a Ru complex:

http://www.dscf.units.it/~balducci/tmp/turbo-davidson-problem/XEDFAQ_1.png

I have limited experience with this kind of calcs, but some time ago I
did similar calcs on another molecule of the same type without any
apparent problem (see also below).

So far my protocol is:
=> geometry relaxation of the isolated molecule (cubic cell (a=21.50
   ang) with a minimum inter-image distance of 10.1 ang, gamma
   point)
=> scf calculation with nbnd=120 (there are 96 occupied bands)
   http://www.dscf.units.it/~balducci/tmp/turbo-davidson-problem/scf.in.txt
=> turbo-davidson run:
   http://www.dscf.units.it/~balducci/tmp/turbo-davidson-problem/davidson.in.txt

    ----8<----
    &lr_input
        outdir='./tmp',
        restart=.false.
    /
    &lr_dav
        p_nbnd_occ=10
        p_nbnd_virt=10
        num_init=60
        num_eign=25
        num_basis_max=200
        residue_conv_thr=1.0E-4
        start=0.18866813     ! 483 nm
        finish=0.23792873    ! 383 nm
        step=0.001
        reference=0.21021155 ! 433.5 nm
        broadening=0.005
        poor_of_ram=.true.
        poor_of_ram2=.true.
        lplot_drho=.true.
    /
    ----8<----

As said, the turbo-davidson calculation doesn't seem to be going
anywhere; here is a typical pattern of the "Largest residue" figure vs
n. of iterations:

http://www.dscf.units.it/~balducci/tmp/turbo-davidson-problem/convergence.png

and here is an output:

http://www.dsch.univ.trieste.it/~balducci/tmp/turbo-davidson-problem/davidson.out.txt

Things that I have tried (with no joy, and, I must admit, without
knowing very well what I was doing):
=> increase num_init
=> decrease num_eign
=> decrease the energy range (start,finish)
=> increase num_basis_max
=> switch vdW-DF off in the scf run
=> increase p_nbnd_occ and p_nbnd_virt

The only way I could get a converged result so far has been to request
a single eigenvalue (num_eign=1): however, I guess this is simply
extreme tweaking of the calculation: even in this conditions, it took
some 22 iterations to converge, which seems too much to me.

I have tried to reproduce my previous calc (the one which worked
smoothly) since qe version at the time was different (I'm presently
using qe-6.5); the molecule was this (ionic species, charge +2):

http://www.dscf.units.it/~balducci/tmp/turbo-davidson-problem/fbt1805_1.png

With the following input:


    ----8<----
    &lr_input
        outdir='./tmp',
        restart=.false.
    /
    &lr_dav
        p_nbnd_occ=10
        p_nbnd_virt=10
        num_init=60
        num_eign=25
        num_basis_max=200
        residue_conv_thr=1.0E-4
        start=0.15
        finish=0.185
        step=0.001
        broadening=0.005
        reference=0.1675
        poor_of_ram=.true.
        poor_of_ram2=.true.
        lplot_drho=.true.
    /
    ---->8----

the calculation converges straightforwardly with qe-6.5:

    ----8<----
    Largest residue:           0.051299398950
    Largest residue:           0.014049254805
    Largest residue:           0.005898374852
    Largest residue:           0.002998212328
    Largest residue:           0.003119825980
    Largest residue:           0.001028311667
    Largest residue:           0.000741910032
    Largest residue:           0.000399386757
    Largest residue:           0.000545167736
    Largest residue:           0.000146498010
    Largest residue:           0.000098106072
    ---->8----

I'm out of ideas and hoping that someone on the list can give me a
clue (apologies: I realize that the statement of the problem is rather
vague, but I don't know how to be more precise than this)

Needless to say, I will be happy to provide any further information
which might be useful

thanks a lot in advance for any hint/suggestion

ciao
-gabriele


-- 
Gabriele Balducci - Dipartimento di Scienze Chimiche e Farmaceutiche - Via
L. Giorgieri 1 I-34127 TRIESTE tel: I-040-5583957 fax: I-040-5583903 e-mail:
balducci at units.it Please, if possible, don't send me MS Word or PowerPoint
attachments Why? See: http://www.gnu.org/philosophy/no-word-attachments.html