[CPMD-list] P(111) - O Questions

[Eduardo J. Lamas]

Dear List members : I have been trying for some time (a long long time :) )
to reproduce published results in atomic oxygen adsorption on a Pt (111)
surface but I was not able to achieve convergent results with CPMD. So I am
wondering if I can get your opinion about what could be wrong in my input or
pseudopotentials that is causing this problems (I also saw some discussions
about whether CPMD can handle this system and that's worries me a little
bit! :)).
 
In the inputs I was trying up to now I have a Pt (111) surface with three
layers (+ 4 vacuum layers) and an O atom located on bridge or top sites, and
I am attempting only wavefunction optimizations.
 
The system seems to be less problematic in the isolated Pt surface without
the O.
For example I reached this point:
 
 ==     NFI=     29      ETOT=   -321.259906   TCPU=   1393.16 ==
 == DRHOMAX= 7.941E-04  DETOT=    -1.552E-03    THL= 7.598E-01 ==


 
After adding the oxygen it looks like is going to find a minimum and then
the solver seems to get confused and starts oscillating  on energy. At the
beginning of the calculations I also tried several solid Pt runs that seemed
to be going fairly well (in these cased I only used different TM BLYP PP)
and I got overestimated unit cell parameter but I thought they weren't that
bad.The energies are given bellow. For the O PP I also tried geometry
optimizations to see if I can get the bond distance and the results were not
excellent but not so bad (1.19083 A with a TM PP). Another issue about
convergence problems is that for TM when I start having them I obtain
adsorption energies that are about the published values, but that is not the
case for the revPBE PP (that is supposed to perform better) where I got
energy differences of about 1 RY!!!.
 
 
   a (A)	   E (Ry) TM BLYP 1	   E (Ry) TM BLYP 3	   E (Ry) TM
BLYP 5	
3.7639	 -106.699024	 -106.3210598	 -106.032463	
3.962	 -106.796925	 -106.424957	 -106.135927	
4.011525	 -106.803911	 -106.4329388	 -106.144288	
4.06105	 -106.805899	 -106.4357779	 -106.147094	
4.1601	 -106.797795	 -106.4289586	 -106.139742	
4.3	 -106.766884	 -106.3991177	 -106.110298	
 
 
 
The pseudopotentials I was using are all MT with different functionals ( I
tried BLYP, revPBE (if I can make this work I prefer this one for
adsorption) and BP) I tried different core radius and different radius for
the NLCC, but in general I based the configurations in the one gold PP that
is available for download at the CPMD web site and used Haman code to
generate the pseudos. In order to obtain the final PP file I mixed the PP
files based on different reference state (with  0.25 charge and net positive
charge on lower levels to obtain the p wave, that seems that localize the p
wave closer to the core) in order to do it after getting the two configs I
just cut and pasted the column in the wave and pot sections for the p
channel on the corresponding column used to obtain the s and d channels.
 
I am sending attached the complete input file, together with the INFO
sections on the PP used. If somebody want to try the PP I can send the
complete PPs. I am sending the complete input because I say an example of a
Pt 111 surface where if I calculate the distance between the    
 
I am using GAUSS HERMIT=15 to integrate the non local part of the PP but I
noticed that I I try to increase it to 20 the program will stop with the
following error (CPMD still defaults to zero points if you don't input
anything maybe it was not changed because of this error):
 
 LANCZOS DIAGONALIZATION (KRYLOV SUBSPACE)
    MAX. KRYLOV BLOCK SIZE                                     64
  PARAMETER NHX TOO SMALL :          117          100
 

 PROGRAM STOPS IN SUBROUTINE SETSYS|   [PROC=   0]
p0_10459:  p4_error: : 999
p7_11279:  p4_error: net_recv read:  probable EOF on socket: 1
p0_10459: (0.480469) net_send: could not write to fd=4, errno = 32
 
Is GAUSS HERMIT=15 accurate enough ? I also noticed that when I use GAUSS
HERMIT=15 I have the following output:
 
 *                                  GH INTEGRATION POINTS:   11 *

Is that correct ? 
 
Sorry about this long e-mail and thanks for reading ! 
 
 
Best regards,
 
Eduardo
 
P.S.: if somebody needs more details like complete PP or ploits about
convergence I am willing to send it either privatly or through the list this
e-mail is too long already for a distribution list.
 
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20041214/12fb2021/attachment.html 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: INFO_Pt_TM_BP5spd.txt
Url: http://cpmd.org/pipermail/cpmd-list/attachments/20041214/12fb2021/attachment.txt 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: INFO_Pt_TM_BP5spd.txt
Url: http://cpmd.org/pipermail/cpmd-list/attachments/20041214/12fb2021/attachment-0001.txt 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: info_Pt_TM_BLYPsd5.txt
Url: http://cpmd.org/pipermail/cpmd-list/attachments/20041214/12fb2021/attachment-0002.txt 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: info_O_revPBE.txt
Url: http://cpmd.org/pipermail/cpmd-list/attachments/20041214/12fb2021/attachment-0003.txt 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: BkPtnew.in
Type: application/octet-stream
Size: 1568 bytes
Desc: not available
Url : http://cpmd.org/pipermail/cpmd-list/attachments/20041214/12fb2021/attachment.obj