[CP2K-user] [CP2K:13207] CELL Optimization - Energy not decreasing

Thomas Kühne tku... at gmail.com
Thu Apr 30 16:14:22 UTC 2020


Dear Jan, 

thanks for putting this to our attention, the deviation looks indeed on the upper end. 
However, computing numerical derivatives is a bit tricky and has to be done with 
maximum accuracy, since relative deviations of small values are reported. 
So this may be an error, but can you please increase EPS_DEFAULT to 10E-16 
and EPS_SCF to 10E-7 and try this also with DX to 5.0E-4? 

Cheers, 
Thomas

P.S. BTW, I did a similar test with the sole PADE XC functional, which is supposed 
to mimic PW92 LDA many years ago and it passed the test ...

> Am 30.04.2020 um 17:30 schrieb Jan Elsner <janel... at gmail.com>:
> 
> Dear all, 
> 
> I thought I should post an update on this. 
> 
> It seems that there is a problem with using PW92 as the local correlation part of XC functional. When I use VWN instead, the analytical and numerical forces are equal. I attach input and output files for DEBUG runs on a single H20 molecule using optPBE in combination with PW92 and VWN. In the case of PW92, there is a mismatch between analytical and numerical forces. 
> 
> I note that the same applies for the some of the other functionals listed here: https://github.com/cp2k/cp2k/tree/master/data/xc_section <https://github.com/cp2k/cp2k/tree/master/data/xc_section> (at least, I can confirm that optB88 and C09 also give a mismatch between numerical and analytical forces using the settings listed here). 
> 
> Best wishes, 
> 
> Jan
> 
> 
> 
> 
> On Friday, April 24, 2020 at 10:43:00 AM UTC+1, Jan Elsner wrote:
> Dear Juerg, 
> 
> Many thanks for the quick response! Unfortunately, decreasing EPS_SCF (all the way to 1.0E-10) does not solve the problem. 
> 
> I have done a DEBUG run which results in the following error (see end of attached file debug.out): 'Abort: A mismatch between the analytical and the numerical stress tensor has been detected. Check the implementation of the stress tensor'. I'm not entirely sure how I should proceed from here / what the problem is exactly - any insight would be very much appreciated. 
> 
> If it helps shed some light on the problem, I also attach a figure showing Total Energy, RMS Gradient and Cell Volume as a function of optimization step (plotted from the original cell.out file I sent). The optimiser seems to overshoot the minimum energy configuration. 
> 
> One other question which comes to mind is whether it's ok to use GTH-PBE or GTH-PADE pseudopotentials with optPBE-vdW?
> 
> Best wishes, 
> 
> Jan
> 
> 
> 
> 
> 
> On Monday, April 20, 2020 at 10:20:34 AM UTC+1, jgh wrote:
> Hi 
> 
> I don't know how well a setup as this is tested. However, from 
> your output I would guess the problem could be related to your 
> SCF convergence. With the methods you are using you have to set 
> the convergence criteria much tighter in order to get converged 
> energy and forces. I would suggest to use at least 1.E-8. 
> 
> The ultimate test would be to do a DEBUG run in order to verify 
> forces and stress tensor for the setting of options. 
> 
> regards 
> 
> Juerg Hutter 
> -------------------------------------------------------------- 
> Juerg Hutter                         Phone : ++41 44 635 4491 
> Institut für Chemie C                FAX   : ++41 44 635 6838 
> 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: "Jan Elsner" 
> Sent by: c... at googlegroups.com <> 
> Date: 04/18/2020 08:33PM 
> Subject: [CP2K:13145] CELL Optimization - Energy not decreasing 
> 
> Dear all, 
> 
> I am trying to run a Cell Optimisation on a periodic molecular crystal (beta-resorcinol: C6H6O2, 4 molecules in unit cell) using K-points and BFGS as optimiser. I am using the vdW-DF-optPBE functional (https://github.com/cp2k/cp2k/commit/74ef05b96a46523a0daf7da04e75804420b5cd5e <https://github.com/cp2k/cp2k/commit/74ef05b96a46523a0daf7da04e75804420b5cd5e>). After 83 optimisation steps, my system does converge to the required MOTION thresholds, however from step 7 onwards the total energy increases at every step i.e.: 
> 
> --------  Informations at step =     7 ------------ 
>   Optimization Method        =                 BFGS 
>   Total Energy               =      -280.1489665596 
>   Internal Pressure [bar]    =     -2667.4573178334 
>   Real energy change         =         0.0001037473 
>   Predicted change in energy =        -0.0001063956 
>   Scaling factor             =         0.0000000000 
>   Step size                  =         0.0124910496 
>   Trust radius               =         0.4724315332 
>   Decrease in energy         =                   NO 
>   Used time                  =               60.978 
> 
>   Convergence check : 
>   Max. step size             =         0.0124910496 
>   Conv. limit for step size  =         0.0010000000 
>   Convergence in step size   =                   NO 
>   RMS step size              =         0.0029192108 
>   Conv. limit for RMS step   =         0.0010000000 
>   Convergence in RMS step    =                   NO 
>   Max. gradient              =         0.0034738629 
>   Conv. limit for gradients  =         0.0001000000 
>   Conv. for gradients        =                   NO 
>   RMS gradient               =         0.0006857738 
>   Conv. limit for RMS grad.  =         0.0001000000 
>   Conv. for gradients        =                   NO 
>   Pressure Deviation [bar]   =     -2668.4705678334 
>   Pressure Tolerance [bar]   =       200.0000000000 
>   Conv. for  PRESSURE        =                   NO 
>  --------------------------------------------------- 
> 
> The resulting cell-optimised structure is therefore not a minimum in energy. I've attached the input and output files. Some points which may (or may not) be relevant: 
> I did not encounter this problem when using a different functional (PBE + D3 instead of optPBE-vdW as used here), but otherwise identical settings. 
> I encountered the same issue (energy increasing at every step) using CUTOFF/REL_CUTOFF = 800/60 
> I initially tried to converge CUTOFF and REL_CUTOFF as in the tutorial ( 
> https://www.cp2k.org/howto:converging_cutoff <https://www.cp2k.org/howto:converging_cutoff>), however my single point calculations did not converge quite as nicely as in the tutorial - instead oscillating with an amplitude of order ~ e-4 Ha (I've attached a figure showing this). This is why I use such large CUTOFF/REL_CUTOFF values. I note that this was also the case when I ran PBE + D3 calculations, however in the latter case my cell optimisation did work. 
> 
> Any input would be greatly appreciated. 
> 
> Best wishes, 
> 
> Jan 
> 
> 
>   
>   -- 
>  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 c... at googlegroups.com <>. 
>  To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/e97788c6-1c43-453d-893e-6b973a977f9d%40googlegroups.com <https://groups.google.com/d/msgid/cp2k/e97788c6-1c43-453d-893e-6b973a977f9d%40googlegroups.com>. 
>   
> 
> [attachment "cell.inp" removed by Jürg Hutter/at/UZH] 
> [attachment "cell.out" removed by Jürg Hutter/at/UZH] 
> [attachment "E_vs_cutoff.png" removed by Jürg Hutter/at/UZH] 
> [attachment "subsys.include" removed by Jürg Hutter/at/UZH] 
> 
> -- 
> 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 <mailto:cp... at googlegroups.com>.
> To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/48cb2870-f660-437b-83b2-4a9e54e15b20%40googlegroups.com <https://groups.google.com/d/msgid/cp2k/48cb2870-f660-437b-83b2-4a9e54e15b20%40googlegroups.com?utm_medium=email&utm_source=footer>.
> <pw92.inp><pw92.out><vwn.inp><vwn.out>



==============================
Thomas D. Kühne
Dynamics of Condensed Matter
Chair of Theoretical Chemistry
University of Paderborn
Warburger Str. 100
D-33098 Paderborn
Germany
tdku... at mail.upb.de
+49/(0)5251/60-5726

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200430/70624b09/attachment.htm>


More information about the CP2K-user mailing list