Different ZPE between small and big basis sets

[quan.phung]

Dear Turbomole user,

I'm doing some calculations on Ferrocene with DFT/B3LYP. After geometry optimization, I use aoforce to compute ZPE. I use 2 types of basis sets:
- Small: def2-TZVP on C and H; def2-QZVPP on Fe. I got ZPE=0.21 Ha.
- Big: def2-QZVPPD on both C, H and Fe atoms. I got ZPE=0.50 Ha.
As you can see that there is a big difference between two ZPEs. I think there is some problems with the big basis set result since it give totally wrong binding energy of Ferrocene. Could you please tell me what's wrong here?

Thank you.

[quan.phung]

Dear Turbomole users,

I computed again with def2-QZVPP for all atoms and the result of ZPE is around 0.21 Ha again. I think the problem must come from the diffuse part of basis set def2-QZVPPD.

Best regards

[Arnim]

Hi,

when I do the calculation, I get 0.17 H with the small basis set and 0.13 H with the big basis set.
Maybe you forgot to optimize the structure?

Cheers,

Arnim

[quan.phung]

Hi Arnim,

Thank you for your reply. Yes i optimized the structure.
In fact yesterday, I did the calculation again with serial optimization and aoforce. I got 0.17 Ha for small basis set.
I don't know if this is a bug or there is problem with our computers or I did wrong. I used 2 procedures:
- Old one: geometry optimization with jobex parallel. After that compute aoforce (serial) in the same directory. I have 0.21 Ha.
- New one: geometry optimization with jobex serial and then compute aoforce (serial) in same directory. I have 0.17 Ha.
Also, total energy compute with 2 procedures are the same, except ZPE and frequencies.

Best regards,
Quan

 

[Arnim]

Hi Quan,

the serial and the parallel version give the same results. This is assured with TURBOTEST.

I suggest you check the two aoforce output for differences. I still think, you are using wrong coordinates somehow.
Maybe your PBS submit scripts do not copy the optimized coordinates back or something like that.

Cheers,

Arnim

[quan.phung]

Dear Arnim,

I checked both of output files. I use both Turbomole 6.2 and 6.3 but I think there is no much different between them. This is the comparison.
http://diffchecker.com/7G5m1lb

You can see that the difference start from Block Davidson iteration and Integral of ground state density.

Best regards,
Quan

[Arnim]

Interesting. aoforce from TM 6.2 on your node20 produces inconsistent numbers.
I just run different versions on different platforms and they all gave the correct result.
Could you run TTEST on that specific computer and, if errors occur, report them to turbomole@cosmologic.de ?

Best,

Arnim