[CP2K-user] [CP2K:16831] Re: xTB descrepancy between CP2K and DFTB+

[Xavier Bidault]

I used the driver/optimizer couple GO/rational for DFTB+ and explored a few 
SCC epsilon parameters (expressed below as 1e-n). For CP2K with BFGS 
optimizer, I similarly explored various values for SCF epsilon. Epsilons 
for geometry of 1e-5 or 1e-6 yields very similar results.

CP2K (all other epsilons are default):
n   Volume(A3)
5   1870.51
6   1863.46
7   1873.30
8   1759.92
9   1834.73
10 1944.21
11 1761.18
12 1846.56
13 1860.54
14 1826.25
15 1852.36

The differences are large, and non-monotonic. There is, imo, no obvious 
choice. I also quickly checked that without constraining the monoclinic 
geometry, it was actually preserved.

DFTB+:
5   1901.99
6   1859.22
7   1904.44
8   1902.72
9   1900.05
10 1900.89
11 1901.01
12 1900.90
13 1900.35

Note that the only one preserving the monoclinic geometry is 1e-6. For all 
the other epsilons, Alpha and Gamma are around 90.2° and 90.7°, 
respectively, instead of 90°. But all those other volumes are very similar. 
Would it mean we have to avoid 1e-6 for DFTB+? Or the contrary?

Any remarks that could help?
On Thursday, April 7, 2022 at 5:05:34 PM UTC-5 Xavier Bidault wrote:

> Hello,
>
> I just posted also on DFTB+ issues (
> https://github.com/dftbplus/dftbplus/issues/1001), including the input 
> script for DFTB+
>
> Xavier
>
> On Tuesday, April 5, 2022 at 3:22:02 PM UTC-5 Xavier Bidault wrote:
>
>> Hello,
>>
>> I run a variable-cell optimization of a high-energy molecular crystal, 
>> beta-HMX, with CP2K 9.1 and DFTB+ 21.2, both using GFN1-xTB, which includes 
>> D3(BJ) correction. To compare apples to apples, I optimized the same 
>> supercell, using the Gamma point. The initial (experimental) volume for the 
>> supercell is 2069.80 A3. The resulting volume is:
>> 1950.83 A3 with DFTB+
>> 1759.92 A3 with CP2K
>>
>> In order to check if I did something wrong in the configuration files, I 
>> used DFTB + D3(0) with the default DFTB parameter file from CP2K for both 
>> CP2K and DFTB+, and I find:
>> 1467.79 A3 with DFTB+
>> 1467.14 A3 with CP2K
>> So, DFTB + D3(0) is consistent between CP2K and DFTB+, but not GFN1-xTB.
>>
>> IMO, the dispersion correction of xTB is wrong somewhere in CP2K. Looking 
>> at the difference between CP2K and DFTB+, could it be that D3(BJ) 
>> correction is counted twice in CP2K? Or with some wrong units we don't have 
>> access to?
>>
>> I attached the CP2K input file for the GFN1-xTB calculation.
>>
>> Please, let me know.
>>
>> Thank you,
>> Xavier
>>
>>
>>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/96beee3f-c95d-4bf5-a2da-2dbe9a3a210cn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20220410/86bb79c5/attachment-0001.htm>