[molpro-user] DFT calculations produced different energy with a different guesses

David Danovich david.danovich at gmail.com
Wed Nov 3 16:03:21 GMT 2010


Dear Kirk,

You are right, the calculations converge to different structures. But the
question is why. In all calculations as you can see in the input I sent in
my original post I used the same starting geometry, just guess wave
functions for B3LYP 4B2 state were different. I got different energy already
in the first step (at the point where geometry was the same for all three
cases). It means that even for single pint calculations solution is
depending on starting guess. In general, I can use any guess wave function
and should get the same converged result. But as you see I got different
results. Occupancy and eigenvalues are more or less similar for the cases 1)
and 3) but in any case the difference in energy is around 3.8 kcal/mol. How
one can be sure that he really get the lower solution (for particular
geometry)? Do you have any suggestion.

Thank you  David


On Wed, Nov 3, 2010 at 5:01 PM, Kirk Peterson <kipeters at wsu.edu> wrote:

> David,
>
> it looks to me that you're getting stuck into a few local minima.  Do these
> different calculations converge to different structures?  How different are
> the energies at the starting geometries?  I would strongly recommend
> printing the orbitals and taking a careful look.   You could also try doing
> a CASSCF calculation (print the orbitals and CI vector)  to see how things
> should really be occupied.
>
> regards,
>
> Kirk
>
>
>   On Nov 3, 2010, at 6:19 AM, David Danovich wrote:
>
>    Hello,
>
> I am calculating CuAu2 molecule in high spin state (4B2 state) using B3LYP
> method. Below you can find three different possibilities I have used in the
> calculations. In the first one 1) I have calculated first HF wave
> function and used it as a guess for DFT calculation. Energy I got was
> -468.86467917 au. In the second calculation 2) I directly have
> calculated B3LYP without HF wave function. The energy I got was
> -468.83995914 au. In the third calculation 3) I first have done B3LYP
> calculation for state (4A2) and then used it as a guess for the calculations
> of 4B2. The energy I got was
> -468.85824088 au. As you can see there is quite large difference in energy
> for the same state. I was trying to use different grids but it does not
> solved the problem. In my opinion result should not depend on the guess so
> drastically. What can be a solution for this problem?
>
> Thank you in advance  David
>
> 1)
>
> ***, Peterson PP tz basis set
> memory,250,m
>  r =   2.72684681 ang;
>  a = 60.65125012 degree;
> geometry={Cu1;              !z-matrix geometry input
>           Au2,Cu1,r;
>           Au3,Cu1,r,Au2,a;
>           }
> basis=cc-pVTZ-PP
>
> {hf,maxit=500;
> wf,57,3,3}
>
> {rks,b3lyp3,maxit=500;
> wf,57,3,3}
>
> optg
>
> Total energy  -468.86467917 au
> ____________________________________________________
>
> 2)
> ***, Peterson PP tz basis set
> memory,250,m
>  r =   2.72684681 ang;
>  a = 60.65125012 degree;
> geometry={Cu1;              !z-matrix geometry input
>           Au2,Cu1,r;
>           Au3,Cu1,r,Au2,a;
>           }
> basis=cc-pVTZ-PP
>
> {ks,b3lyp3,maxit=500;
> shift,-0.1,0.0;
> wf,57,3,3}
>
> optg
>
> Total energy -468.83995914 au
>
> ______________________________________________________
>
> 3)
> ***, Peterson PP tz basis set
> memory,250,m
>  r =   2.72684681 ang;
>  a = 60.65125012 degree;
> geometry={Cu1;              !z-matrix geometry input
>           Au2,Cu1,r;
>           Au3,Cu1,r,Au2,a;
>           }
>
> ! cc-pVTZ-PP
>
> basis=cc-pVTZ-PP
>
> {ks,b3lyp3,maxit=500;
> wf,57,4,3}
>
> {ks,b3lyp3,maxit=500;
> shift,-1.0,0.0;
> wf,57,3,3}
>
> optg
>
> Total energy -468.85824088 au
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20101103/0c7df6dd/attachment.html>


More information about the Molpro-user mailing list