[molpro-user] DFT calculations produced different energy with a different guesses
David Danovich
david.danovich at gmail.com
Wed Nov 3 20:45:21 GMT 2010
Dear Kirk,
In general, you are right and it is always possible to converge to different
electronic state, especially for the systems with transition metals. For DFT
methods I always do TDDFT calculations in order to be sure that the state I
get is a lower one (for such systems with 3 transition metal atoms I have no
ideas how to do CASSCF with 33 valence electrons and I do not know a priory
which d electrons are important). But for this particular case I think I
have the same state with the same occupancy as you can see below and also
with almost the same eigenvectors. But as I said before difference in total
energy is 3.8 kcal/mol. What is also strange that for irreducible
representation A1 Aufbau principle is not satisfy. You may see that orbital
11.1 has much lower energy than orbitals 6.1, 7.1, 8.1, 9.1, 10.1 for both
calculations. My feeling is that something goes wrong with DFT calculations
in Molpro. I will try to carry out this type of calculations with other
programs.
Thank you David
Case 1)
Final alpha occupancy: 12 5 9 4
Final beta occupancy: 10 5 8 4
Orbital energies:
1.1 2.1 3.1 4.1 5.1
6.1
7.1 8.1 9.1 10.1
-4.671414 -4.118564 -3.065611 -2.328939 -2.323338
-0.438461 -0.379638 -0.343993 -0.313299 -0.288287
11.1 12.1 13.1 14.1
-0.509678 -0.218071 -0.055982 -0.024797
1.2 2.2 3.2 4.2 5.2
6.2
7.2
-3.028087 -2.319030 -0.478500 -0.324411 -0.306541
-0.066386 0.004508
1.3 2.3 3.3 4.3 5.3
6.3 7.3 8.3 9.3 10.3
-4.118395 -3.028437 -2.326183 -2.321552 -0.481322
-0.301593 -0.288912 -0.282174 -0.196122 -0.041215
11.3
0.035134
1.4 2.4 3.4 4.4 5.4 6.4
-2.318707 -0.440376 -0.298955 -0.284142 0.018513
0.139495
Case 3)
Final alpha occupancy: 12 5 9 4
Final beta occupancy: 10 5 8 4
1.1 2.1 3.1 4.1 5.1
6.1 7.1 8.1 9.1 10.1
-4.673062 -4.117760 -3.014878 -2.327371 -2.323105
-0.441349 -0.360882 -0.345206 -0.319350 -0.288610
11.1 12.1 13.1 14.1
-0.514110 -0.216705 -0.051875 -0.024126
1.2 2.2 3.2 4.2 5.2
6.2 7.2
-3.061378 -2.318440 -0.460149 -0.323223 -0.306348
-0.070494 0.001935
1.3 2.3 3.3 4.3 5.3
6.3
7.3 8.3 9.3 10.3
-4.117591 -3.050251 -2.324891 -2.321034 -0.451631
-0.300020 -0.288446 -0.281090 -0.195381 -0.042282
11.3
0.035544
1.4 2.4 3.4 4.4 5.4 6.4
-2.318118 -0.492930 -0.298691 -0.283211 0.018322
0.139423
On Wed, Nov 3, 2010 at 9:40 PM, Kirk Peterson <kipeters at wsu.edu> wrote:
> David,
>
> it is certainly not always the case that the resulting HF or DFT solution
> is independent of the initial guess. In a perfect world that would be nice,
> but often a poorly constructed initial guess (or just bad luck) can result
> in convergence to a local minimum, i.e., an upper electronic state. This
> happens quite often for transition metal complexes because of the plethora
> of possible d-orbital occupations. The only way to get a feeling for what
> is going on is to print the orbitals and inspect the higher lying occupied
> ones. Is the bonding as you might expect? Another option is to carry out a
> CASSCF calculation as I suggested before in order to determine what the
> character of the higher lying occupied orbitals should really be. One can
> always then use the rotate command to swap in the correct orbitals in your
> HF or DFT calculation.
>
> -Kirk
>
> On Nov 3, 2010, at 9:03 AM, David Danovich wrote:
>
> 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/20101103/9ba88742/attachment.html>
More information about the Molpro-user
mailing list