[CP2K-user] [CP2K:14980] Problem about the analytical stress tensor for meta-GGA functionals in CP2K 8.1

[Tianhua Wang]

Dear Prof. Hutter,
  Thank you very much for your early reply and detailed guidance. Following 
your guidance, I tried my best to do a series of additional calculations to 
find out the possible problems. For the cell optimizations for cassiterite 
(SnO2), I tested the performance of SCAN functional combining with 
different basis sets and GTH pseudopotentials, as well as the effect of the 
cutoff of the electronic density. Besides, the lattice parameters of 
cassiterite were also optimized by using TPSS functional (combining with 
GTH-PBE pseudopotentials). These results, together with the results in my 
original letter, are tabulated in Table 1 in the *Supplementary Materials* 
attached. An example of my input file can be found in my original letter. 
It seems that basis sets and cutoff values have little influence on the 
results, whereas the effect of pseudopotentials is significant. The 
calculations using SCAN functional with GTH-SCAN pseudopotentials 
underestimated the volumes with relatively large errors, but the 
calculations using SCAN functional with GTH-PBE pseudopotentials turned to 
overestimate the volumes with smaller errors. If this problem really can be 
attributed to the pseudopotentials, would it be possible to obtain better 
results by improving the quality of pseudopotentials? Besides, I noticed 
that in the paper proposing r2SCAN functional by John P. Perdew’s group 
(Furness et al., 2020, *J. Phys. Chem. Lett.*), they mentioned that “the 
SCAN exchange-correlation potential for an atom diverges in the tail of the 
density, making pseudopotential construction difficult.” Will this defect 
influence the quality of GTH-SCAN pseudopotentials?
  The analytic stress tensors derived from the geometry optimizations for 
cassiterite (using the experimental lattice parameters in Bolzan et al. 
(1997, *Acta Cryst. B*)) are tabulated in Table 2 in the *Supplementary 
Materials* attached. The input file for the calculation at the SCAN/TZVP 
level is attached for reference. These results confirm that the 
calculations using SCAN functional with GTH-SCAN pseudopotentials tend to 
underestimate the volumes whereas the calculations using SCAN functional 
with GTH-PBE pseudopotentials show the reverse.
  However, on the other hand, in my another series of additional 
calculations for the cell optimization for rutile (TiO2), SCAN functional 
with GTH-SCAN pseudopotentials reproduced its lattice parameters very well, 
but there were small errors in the calculated interaxial angles. The 
optimized lattice parameters are tabulated in Table 3 in the *Supplementary 
Materials* attached. I tested the system of rutile because it had similar 
structure with cassiterite, i.e., tetragonal system, 4/*mmm* crystal class 
and *P*4(2)/*mnm* space group. The model of rutile was a supercell 
consisting of 2×2×4 unit cells, which was constructed based on the 
single-crystal X-ray diffraction data (Swope et al., 1995, *Am. Mineral.*). 
The input files used are similar to those for the cell optimizations for 
cassiterite. I am confused about the different performance of SCAN 
functional between the cell optimizations for cassiterite and rutile. What 
I can only think of is that the system of rutile has higher electron 
density than cassiterite and liquid water, and SCAN functional may be 
sensitive to electron density.
  Last but not the least, I am sorry to say that I can not do more FPMD 
simulations of liquid water because of the relatively low computational 
efficiency of cp2k-8.1-Linux-x86_64.ssmp. However, it is very important to 
give a good description of aqueous solutions. I am wondering if these tests 
can be helpful for understanding the overestimation of water density at 
ambient conditions when using SCAN functional?
  Thank you for your time and I am looking forward to your reply.
Sincerely,
Tianhua Wang


在2021年3月24日星期三 UTC+8 下午7:07:47<jgh> 写道：

> Hi
>
> it is difficult to say where the problem is. You can check the
> correctness of the analytic stress tensor by running a debug
> run. This will compare the analytic stress to a finite difference
> calculation. You could do this for your SnO2 example.
> We do this for other examples and the tests are passed.
>
> Besides that, it might be also a problem of pseudopotential, basis set
> and/or cutoff. In principle you have to check all of this in 
> independent runs. For this it would be best to have an independent
> result to compare to.
> Another line of tests could be to use another MGGA (like TPSS)
> to see if you get consistent results.
>
> best regards
>
> Juerg Hutter
>
> --------------------------------------------------------------
> Juerg Hutter Phone : ++41 44 635 4491 <+41%2044%20635%2044%2091>
> Institut für Chemie C FAX : ++41 44 635 6838 <+41%2044%20635%2068%2038>
> Universität Zürich E-mail: h... at chem.uzh.ch
> Winterthurerstrasse 190
> CH-8057 Zürich, Switzerland
> ---------------------------------------------------------------
>
> -----c... at googlegroups.com wrote: -----
> To: "cp2k" <c... at googlegroups.com>
> From: "Tianhua Wang" 
> Sent by: c... at googlegroups.com
> Date: 03/22/2021 09:56AM
> Subject: [CP2K:14980] Problem about the analytical stress tensor for 
> meta-GGA functionals in CP2K 8.1
>
> Dear CP2K developers,
> Thank you very much for developing the analytical stress tensor for 
> meta-GGA functionals in CP2K 8.1. By using the cp2k-8.1-Linux-x86_64.ssmp 
> downloaded from the GitHub project page of CP2K, I did two series of 
> calculations in which the analytical stress tensor for SCAN functional was 
> used. However, the results indicated that the calculations using SCAN 
> functional underestimated volumes significantly, which was inconsistent 
> with the good performance of SCAN functional reported in literatures. I am 
> wondering if there could be something wrong with my operations or the 
> analytical stress tensor for meta-GGA functionals in CP2K 8.1. The 
> computational details and results are shown as below.
> Firstly, cell optimizations for cassiterite (SnO2) were performed at 
> different computational levels, i.e., combining three different functionals 
> (i.e., SCAN, LDA (PADE) and PBE) with DZVP/TZVP basis sets and GTH 
> pseudopotentials. The model of cassiterite was a supercell consisting of 
> 2×2×4 unit cells, and its initial structure was constructed based on the 
> powder neutron diffraction data (Bolzan et al., 1997, Acta Cryst. B). The 
> input file for the calculation at the SCAN/TZVP level is attached for 
> reference. From the lattice parameters of optimized structures tabulated in 
> Table 1, it can be seen that the calculations using SCAN functional 
> underestimated volumes significantly, and the relative errors (compared 
> with experimental values) produced by SCAN functional are even larger than 
> those produced by PBE functional.
>
> Table 1. Lattice parameters of optimized structures of cassiterite (SnO2). 
> a
>
>
> Another series of calculations were the FPMD simulations of liquid water 
> in isothermal-isobaric (NPT) ensemble at ambient conditions, in which the 
> temperature of 330 K was used. The simulations were performed by using SCAN 
> and PBE-D3, together with DZVP basis sets and GTH pseudopotentials. For 
> simulations using SCAN, three kinds of DZVP basis sets and GTH 
> pseudopotentials were performed, i.e., (1) DZVP-MOLOPT-SCAN-GTH basis sets 
> and GTH-SCAN pseudopotentials from the GitHub project page of Prof. Jürg 
> Hutter, (2) DZVP-MOLOPT-SR-GTH basis sets in the BASIS_MOLOPT file and 
> GTH-SCAN pseudopotentials mentioned above, and (3) DZVP-MOLOPT-SR-GTH basis 
> sets in the BASIS_MOLOPT file and GTH-PBE pseudopotentials in the 
> GTH_POTENTIALS file. The input file for the calculation using SCAN 
> functional with MOLOPT-SCAN-GTH basis sets and GTH-SCAN pseudopotentials 
> (i.e., (1) mentioned above) is attached for reference. For the simulation 
> using PBE-D3, DZVP-MOLOPT-SR-GTH basis sets in the BASIS_MOLOPT file and 
> GTH-PBE pseudopotentials in the GTH_POTENTIALS file were employed. Each 
> simulation was performed for 10.0 ps, following a classical MD simulation 
> for SPC/E water in NVT ensemble for 10.0 ns. The running average densities 
> derived from the last 8.0 ps of the simulations were shown in Figure 1. I 
> know 10.0 ps is too short to derive a reasonable density, and the data in 
> Figure 1 are far away from convergence. However, on the one hand, the 
> computational efficiency of cp2k-8.1-Linux-x86_64.ssmp is relatively low, 
> and it has taken a long time for these simulations; on the other hand, as 
> shown in Figure 1, the densities derived from the simulations using SCAN 
> functional are all larger than the density derived from the simulation 
> using PBE-D3, which is similar with the phenomenon seen in the cell 
> optimizations for cassiterite.
>
>
> Figure 1. Running average densities as a function of simulation time.
>
> Thank you for your time and I am looking forward to your reply.
> Sincerely,
> Tianhua Wang
>
>
>
> -- 
> 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/4bd9fc54-4beb-4a09-98b7-0044d46ab62fn%40googlegroups.com
> .
>
>
> [attachment "330NPT-1bar_64H2O_SCAN+DZVP-SCAN_800Ry.inp" removed by Jürg 
> Hutter/at/UZH]
> [attachment "CELL_OPT_32SnO2_SCAN_TZVP_0bar.inp" removed by Jürg 
> Hutter/at/UZH]
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210330/c9943d33/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: GEO_OPT_32SnO2_SCAN_GPW_TZVP_V-exp.inp
Type: chemical/x-gamess-input
Size: 6789 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210330/c9943d33/attachment.inp>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Supplementary Materials.docx
Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document
Size: 27065 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210330/c9943d33/attachment.docx>