[CPMD-list] convergence problem on iron porphyrin
Yong Zhang
zhangymall11 at gmail.com
Sun Jul 30 18:50:38 CEST 2006
Thank you Axel and Pradip,
I tested several options: VDB/MT; LOC=S or not; PCG or not; DUAL=4,5 or 6.
Finally only those options suggested by Pradip work.
Axel is absolutely correct, "LOC=S" does nothing when VDBpp is used and PCG
is essential. (I just want to double check it by my own, please do not mind
^_^) But the wavefunction optimization runs using MT (with PCG and LOC=S)
died finally. Does this simply mean that MT is not appropriate for this
calculation or I need to take care of other settings?
Both Dual=5 and Dual=6 work. And dual=5 gave a lower energy after geometry
optimization. I suppose this is good enough.
Thank Axel for the link. I guess the problem is related to the
adiabatic/nonadiabatic dynamics of DFT method. But, to be honest, I do not
know how to prove (or define) that my system is nonadiabatic :-(
I did test the different spin states for my system. For spin=1,3 or 5, all
the wavefunction jobs completed without problem. But in the GEO
optimization, only spin=3 completed with reasonable result. This is
consistent with previous references saying that spin=3 is the stablest state
for this complex. But this is only a static calculation, I do not know what
will happen if I run a dynamics.
For spin=1 or 5, I observed a total energy jump which I believe is caused by
the rapid increase in kinetic energy. But I do not understand why this
happened?
The "jump part" for spin=5 is like this:
###############
49 6.952E-05 7.506E-06 -286.501141 8.912E-05 16.50
50 1.485E-04 1.408E-05 -286.500750 3.907E-04 16.50
51 4.059E-04 3.516E-05 -286.497778 2.972E-03 16.50
52 1.682E-03 1.557E-04 -286.430725 6.705E-02 16.50
53 1.276E-02 2.070E-03 -275.098803 1.133E+01 16.50
54 1.683E-01 1.332E-02 7.105774 2.822E+02 16.46
55 1.687E-01 1.298E-02 9.960731 2.855E+00 16.46
56 1.687E-01 1.298E-02 9.937762 -2.297E-02 16.46
57 1.686E-01 1.298E-02 9.910858 -2.690E-02 16.46
...
116 1.650E-01 1.352E-02 -2.314784 -6.991E-01 16.46
117 1.651E-01 1.352E-02 -3.051419 -7.366E-01 16.45
118 1.651E-01 1.351E-02 -3.826305 -7.749E-01 16.45
================================================================
= END OF GEOMETRY OPTIMIZATION =
================================================================
################
and the final summary is:
################
ELECTRONIC GRADIENT:
MAX. COMPONENT = 1.65130E-01 NORM = 1.35140E-02
NUCLEAR GRADIENT:
MAX. COMPONENT = 6.73460E+00 NORM = 1.63507E+00
TOTAL INTEGRATED ELECTRONIC DENSITY
IN G-SPACE = 128.000000
IN R-SPACE = 128.000000
TOTAL INTEGRATED SPIN DENSITY -4.000000
TOTAL INTEGRATED ABSOLUTE VALUE OF SPIN DENSITY 9.373111
VANDERBILT AUGMENTATION CHARGES (MEAN VALUE PER ATOM)
ATOM TYPE NR. OF ATOMS CHARGE
H 12 .020
C 20 -1.374
N 4 -.242
Fe 1 4.451
(K+E1+L+N+X) TOTAL ENERGY = -3.82630538 A.U.
(K) KINETIC ENERGY = 418.55865163 A.U.
(E1=A-S+R) ELECTROSTATIC ENERGY = -192.60043071 A.U.
(S) ESELF = 228.72690743 A.U.
(R) ESR = 7.52699158 A.U.
(L) LOCAL PSEUDOPOTENTIAL ENERGY = -148.70431024 A.U.
(N) N-L PSEUDOPOTENTIAL ENERGY = -28.39383909 A.U.
(X) EXCHANGE-CORRELATION ENERGY = -52.68637696 A.U.
GRADIENT CORRECTION ENERGY = -3.62222815 A.U.
################
the final summary for spin=3 is:
#################
ELECTRONIC GRADIENT:
MAX. COMPONENT = 1.06400E-05 NORM = 5.14668E-07
NUCLEAR GRADIENT:
MAX. COMPONENT = 5.59494E-02 NORM = 8.90548E-03
TOTAL INTEGRATED ELECTRONIC DENSITY
IN G-SPACE = 128.000000
IN R-SPACE = 128.000000
TOTAL INTEGRATED SPIN DENSITY -2.000000
TOTAL INTEGRATED ABSOLUTE VALUE OF SPIN DENSITY 2.335721
VANDERBILT AUGMENTATION CHARGES (MEAN VALUE PER ATOM)
ATOM TYPE NR. OF ATOMS CHARGE
H 12 .032
C 20 .116
N 4 .296
Fe 1 9.654
(K+E1+L+N+X) TOTAL ENERGY = -286.52297826 A.U.
(K) KINETIC ENERGY = 105.06259988 A.U.
(E1=A-S+R) ELECTROSTATIC ENERGY = -212.06028870 A.U.
(S) ESELF = 228.72690743 A.U.
(R) ESR = 7.47507791 A.U.
(L) LOCAL PSEUDOPOTENTIAL ENERGY = -156.32548960 A.U.
(N) N-L PSEUDOPOTENTIAL ENERGY = 38.68448417 A.U.
(X) EXCHANGE-CORRELATION ENERGY = -61.88428402 A.U.
GRADIENT CORRECTION ENERGY = -2.58770500 A.U.
###################
Thank you!
Best wishes
Yong
On 7/28/06, Axel Kohlmeyer <akohlmey at cmm.upenn.edu> wrote:
>
> On 7/28/06, Yong Zhang <zhangymall11 at gmail.com> wrote:
> >
> > It is working now. Thanks a lot.
> > I tried both VDB pp and MT pp, both work. So I guess the "LOC=S" for Fe
> is
> > the point.
>
> LOC=S is ignored for VDB. but for the Fe-MT potential essential,
> so you don't have ghost states. PCG and/or PCG MINIMIZE is
> the ticket for systems with tricky electron structures, which is
> truely the case here.
I tested this and
> By the way, do you have any special reason for using VDB? What is the
> > difference between these PPs? Sorry if this is a naive question. I am a
> very
> > beginner of CMPD. :-p
>
> in that case, you should probably start with a simpler system.
> there are references for both types of pseudopotentials in the manual.
>
> VDB pseudopotentials need a (much) lower plane wave energy cutoff,
> due to not releasing the norm-conservation constraint on the
> pseudopotential.
>
> that has several implications, especially with the VDB implementation in
> CPMD.
> since CPMD does not have a working dual grid for the uspps, you have
> to crank up the density cutoff (best by using DUAL with a value of 5.0
> of even 6.0)
> so you get consistend DFT results and less problems with placing the
> augmentation charges on the grid). also, apart from basic operations like
> MD, geometry optimization, constraints, many advanced features in CPMD
> do not work with ultra-soft pseudopotentials. in the case of iron,
> however,
> the uspps should give a 'better structure' than the MT potential, provided
>
> you take care of the other technical issues.
>
> finally a word of caution. for the kind of system you are dealing with
> certain configurations can be quite tricky and DFT may not always
> give accurate results. please have a look at, e.g., page 18 of
> http://www.vlab.msi.umn.edu/events/download/LDA+U.pdf
> this is done with a different code, and there are other attempts
> to solve this DFT related problem as well, but you should check
> your results carefully.
>
> cheers,
> axel.
>
> >
> > Yong
> >
> > On 7/28/06, Pradip Kumar Biswas <p.biswas at csuohio.edu> wrote:
> > > Hi,
> > >
> > > I had a similar problem with convergence; PCG minimize helped.
> However,
> > > this is to mention that I used VDB pp with a w.fn cutoff of 25 Ry,
> dual
> > > 5.0 and for iron I use LMAX=D LOCAL=S.
> > >
> > > pb.
> >
> >
> > On Jul 28, 2006, at 2:05 PM, Yong Zhang wrote:
> >
> > > Hello CPMDers,
> > >
> > > I am trying to run some simulation on a Fe-porphyrin complex. The
> > > wavefunction optimation did not tend to converge at all. I took an
> > > optimized structure from MM calculation as the initial input. The
> > > triplet was reported before as the stable state of this complex and
> > > singlet and quintet are also possible. I tried all the three states
> > > but the results are similar.
> > >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20060730/16a062ef/attachment.html
More information about the CPMD-list
mailing list