The ABINIT Users Mailing List ( CLOSED )

[Masayoshi Mikami <mmikami@rc.m-kagaku.co.jp>]

Dear Denes,

I have little time to look into this, but I may have suggestions ...

> > Which kind of pseudopotentials were adopted ?
>
> The Troullier-Martins pseudopotentials from the abinit web page.

I assume " A. Khein and D.C. Allan" version...
As I said, you might want to check the potentials for simpler cases...
(You can forget about such complex molecular crystal at the first stage.
 You need just to find isolated molecule cases including Mo (Cu, etc.).
 You could ask someone about them?  I am not familiar with such 
materials)

A question: how about the charge states of Mo and Cu ?
Are they d^0 states (Mo6+/Cu+) or not ? (occ/occcopt could be tuned.)

> > Your system has :
> > 4*Mo, 8*P, 2*S, 40*C, 60*H, 3*Cu, 3*I
        (snip)
> > Before doing this calculation, you checked your pseudopotentials
> > with simpler systems ?
>
> no, in fact it's not easy to find a system similar to this case but 
> significantly smaller. I could get rid of maybe 20 C and 40 H atoms 
> (methyl groups on the complexing rings) , but the rest should stay...
> anyway, I need suggestions about what to try on the smaller system.

As Paul suggested, CH3- may be replace H- at the first stage.
Then, separate your molecular system into isolated molecules
 ("metallocene" part) and terminate the cut bond with something
 to keep it neutral (or set charge state).
Then check your Mo PP... and so on...

> > (The pseudopotentials available at the WEB
> > site should be used at your own risk, as it says ...)
>
> What other pseudopotential should I try? Is there a summary about 
> which one is used (normally) for what?

For your complex case, please try literature survey about similar
calculations, or learn by yourself...
Mo might need some tuning. Cu also might.
(Other PPs might also ... for your system ?)

e.g., Mo may have semicore (4s)2(4p)6 as valence states ... zion=14 !
(When I tried calculation of MoO3 long ago, such semicore states needed
 to be valence states to reproduce its experimental structure...
 although I am not sure this is indeed necessary for your systems ...
 It might be worthwhile to try ?)

If necessary, you might want to make your own pseudopotentials
by consulting with documents (tutorials) in FHI98PP and OPIUM etc.
(NB: I have noticed that OPIUM has been updated to v.2.0.1,
 which now produce Troullier-Martins (TM) PP as well.
 The tutorials (at the UCSB Summer School) are also available.
 Thanks a lot, Eric ! (Please wait for photos awhile ... ))

> > I am not sure "ecut 8" is sufficient. It may be more increased ?
> > (it really depends on pseudopotential type...)
>
> Basically I have not enough RAM to increase ecut with this much 
> k-points

In general (as a rule of thumb), TM PP of C needs about 40-45 Ry.
Cu PP (zion 11) available on the WEB apparently needs about 60 Ry.
(ftp://ftp.abinit.org/pub/abinitio/Psps/LDA_TM.psps/29/29cu.pspnc
 You could consult with ~ABINIT/Infos/Psp_infos/psp1.info.
 "psp1.data" is also interesting to read !)
If Mo PP includes the semicore states as valence, "ecut" should get 
larger...
and so on...
OPIUM masters (e.g. Eric) will say,
"Optimized-PPs should be softer than TM-PPs",  I suppose. ;-)
(PAW will ease such heavy calculations ... let us see...
 you might want to try  the tutorials in v.4.6.3 ?)

> any more ideas?

I just keep my fingers crossed for your good luck. ;-)

> arigato gozaimas :)

"Nics mit" :) in Hungarian.
"Doh Itashi mashite" in Japanese ("Not at all")

Have a good weekend to all,
Masayoshi
(I will be absent during the next week ...)