[CP2K-user] [CP2K:15704] Wrong water density wih xTB

[Guillaume Stirnemann]

Dear Pedro,

Unfortunately, this is a known issue with tight-binding methods such as
DFTB or xTB. It has been reported before, see e.g.
https://pubs.acs.org/doi/abs/10.1021/jp400953a, although it does not seem
to be widely communicated. This info had been nicely shared by Thomas Kühne
and Juerg Hutter here some time ago (see Dec 2018).
Systems shrink unreasonably in the NPT ensemble, and usually, bubbles
appear if the volume is kept constant to satisfy the experimental density.
It is pretty clear by looking at the radial distribution functions that the
structure is usually pretty badly reproduced.

In my group we did many many tests using different generations of TB
methods in cp2k, including xTB. The only set of parameters that seem to
give "stable" results for water (in the NVT ensemble) are the SCC-DFTB,
sometimes called DFTB2, using the mio-1-1 parameter set. Thomas also shared
some parameters that were reoptimized to better reproduce water structure,
see in https://arxiv.org/pdf/1408.5161.pdf.
These parameters do not lead to voids in the simulation box, but I never
tested them in NPT.

It seems that later generations of xTB seem give much more accurate results
in general (no idea how they behave for liquid water), but I don't think
they are currently implemented in cp2k.

Best wishes,

Guillaume Stirnemann


On Wed, Jul 7, 2021 at 7:01 PM Pedro Maximiano <pedro... at gmail.com> wrote:

> Dear CP2K users and developers
>
> I'm a relatively new CP2K user, and I'm trying to do MD simulations of a
> dimer of a small molecule in water with the GFN-xTB method. To prepare the
> simulations I first minimize and then equilibrate the boxes in a NPT
> ensemble (300 K, 1 bar) with the AMBER force field, followed by a short NVT
> run (also with AMBER FF). I then start the xTB simulation from the final
> point of the classical MD run, in the NVT ensemble (300 K).  However, I
> noticed that voids appear in the xTB simulation boxes, probably indicative
> of a wrong system density.
>
> To reproduce the issue in a simpler system, I ran a NPT simulation with
> xTB (300 K, 1 bar) on a box of 126 water molecules in CP2K. The box was
> pre-equilibrated in a NPT run (also 300 K, 1 bar), with the AMBER FF TIP3P
> parameters, yielding a density of 1.023 g/cm3. This box was then subjected
> to a minimization and a short NVT equilibration with xTB, prior to the NPT
> equilibration. Indeed, I observed that the xTB NPT run compressed the box,
> to a final density of 1.481 g/cm3.
>
> I'm attaching here the input for the xTB NPT simulation of water. The
> sections that are commented are settings which I also tried, without
> success, namely:
> - Adding a NONBONDED potential to xTB (taken from the test file
> xTB/regtest-debug/h2o-atprop_nonbonded.inp)
> - Changing the settings in the POISSON/EWALD section to non default values
> (up to RCUT = 10.0 Angstrom)
>
> Note that I also ran the test xTB/regtest-debug/h2o-atprop_nonbonded.inp
> for 20000 steps and obtained an abnormally high density (1.722 g/cm3).
> All runs were performed in CP2K v8.2, but I also noticed the same issues
> when running in version 7.1.
>
> Could you please take a look at the input file and tell me if I'm missing
> something?
>
> Best regards
> Pedro Maximiano
>
> --
> 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 cp... at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/7e0dd338-885f-469f-acdd-2a32e8dda682n%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/7e0dd338-885f-469f-acdd-2a32e8dda682n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210708/937e02b9/attachment.htm>