[CP2K-user] [CP2K:19853] Cons. Qty for different ensembles + Pressure fluctuation

[Léon Luntadila Lufungula]

Dear Matthias,

I tried continuing the calculation with the 20 Å water layer, because this 
was the only one that looked like it was going towards a point where the 
constant of motion would also be equilibrated. However, upon restarting the 
calculation directly from the .restart file and only adjusting the maximum 
number of steps, so that everything is restarted (including the barostat), 
I immediately see larger fluctuations in the pressure and a resulting drift 
of the constant of motion... (see below) The only things I can think of 
that are inherently different between the two calculations is the 
number/type of nodes I'm running on and the SCF_GUESS. The 1st calculation 
was ran on 4 normal nodes and the wavefunction for the first step was 
restarted from a .wfn file of a geometry optimization which was performed 
before the MD run, while the restarted calculation was on 2 high-memory 
nodes and the wavefunction was started from a superposition of atomic 
wavefunctions as I read in the exercise on AIMD 
<https://www.cp2k.org/exercises:2016_summer_school:aimd> that storing the 
wavefunction can slow down your MD run, so this was not stored during the 
AIMD run. I'm currently running a calculation with the same node setup as 
the first calculation, but I was wondering if it could be possible that 
these changes are due to the fact that I am not starting from the .wfn file 
of the 1st calculation? 
[image: a101-solv20.md-NPT_F-eq2.png]
I can't do these calculations in one go as they take longer than the 
maximum wall time on my HPC cluster, so if I can't properly restart them I 
can't equilibrate my system, let alone do a production run...

Kind regards,
Léon
On Monday 29 January 2024 at 12:24:28 UTC+1 Léon Luntadila Lufungula wrote:

> Dear Matthias,
>
>    1. The values plotted are indeed the conserved quantity that is 
>    printed in the 6th column of the .ener file.
>    2. Since there are still relatively large fluctuations in the cell 
>    volume, should I then continue the equilibration until the cell volume is 
>    constant? I can confirm that the computational cost of AIMD is quite large, 
>    but I would rather avoid cutting corners, because I don't want 
>    deterioration of the sampling quality as you point out.
>    3. If changing the thermostat/barostat parameters will change the 
>    value of the constant of motion, can you still use the constant of motion 
>    as a measure of system equilibration when equilibrating in two parts: 
>    strong early equilibration (small timecons: T50, B100) + weaker late 
>    equilibration (larger timecons: T100, B1000)?
>    4. Regarding EPS_DEFAULT, I did test the energy conservation of an NVE 
>    trajectory for several combinations of EPS_SCF/EPS_DEFAULT on a slab 
>    without solvent and 1.0E-7/1.0E-10 (first picture) gave similar results as  
>    1.0E-7/1.0E-12 (second picture), but perhaps I should indeed reduce it 
>    further to ensure stable dynamics.
>    [image: default-10.png][image: default-12.png]
>    5. I agree that a noticeable energy drift will always occur if you 
>    just simulate long enough, but I wouldn't expect such a large drift (0.08 
>    a.u. for the water box and 0.4 a.u. for the solvated slab if you disregard 
>    the jump when changing the thermostat/barostat) to already pop up when 
>    simulating for only 20 ps...
>    6. For the solvated slab, I was testing how thick I needed to make the 
>    water layer to correctly model the structure of liquid water so I simulated 
>    an NPT_F trajectory for a water layer of 10, 20, 30 and 40 Å. The 10 Å 
>    water layer is the one I showed in the previous message with the large 
>    energy drift, but the 20 Å water layer does seem to show an equilibrating 
>    constant of motion (see below). The input parameters are all the same and 
>    the structures were all solvated in a similar way by adding a vacuum layer 
>    to the slab with ASE and solvating the vacuum layer with Gromacs (gmx 
>    solvate). Any idea why the 20 Å trajectory does seem to equilibrate very 
>    easily, while the 10, 30 and 40 Å trajectories all show these weird jumps 
>    in the constant of motion?
>    [image: solv20.png]
>    
>
> As always, thanks a lot for all the help and my apologies for the barrage 
> of questions, these calculations are really hard to setup correctly...
>
> Kind regards,
> Léon 
>
>
> On Monday 29 January 2024 at 11:46:24 UTC+1 Krack Matthias wrote:
>
>> Dear Léon
>>
>>  
>>
>> MD based on DFT compared to classical potentials suffers more from 
>> numerical noise in the forces. Therefore, you should reduce EPS_DEFAULT 
>> from 1.0E-10 to at least 1.0E-12 in order to reduce that noise. 
>> Nevertheless, you will always observe a noticeable energy drift when you 
>> run long enough.
>>
>>  
>>
>> HTH
>>
>>  
>>
>> Matthias
>>
>>  
>>
>> *From: *cp... at googlegroups.com <cp... at googlegroups.com> on behalf of 
>> Léon Luntadila Lufungula <Leon.luntad... at uantwerpen.be>
>> *Date: *Monday, 29 January 2024 at 11:03
>> *To: *cp2k <cp... at googlegroups.com>
>> *Subject: *Re: [CP2K:19840] Cons. Qty for different ensembles + Pressure 
>> fluctuation
>>
>> Dear Matthias,
>>
>>  
>>
>> My apologies, I forgot to add the input files. You can find them in 
>> attachment to this message.
>>
>>  
>>
>> Kind regards,
>>
>> Léon
>>
>> On Monday 29 January 2024 at 10:36:30 UTC+1 Léon Luntadila Lufungula 
>> wrote:
>>
>> Dear Matthias,
>>
>>  
>>
>> I've also tried to equilibrate a box of 216 water molecules with DFT and 
>> while the temperature and pressure seem to equilibrate quite easily, the 
>> constant of motion shows large jumps despite the potential and kinetic 
>> energy also being quite constant... If I'm not mistaken, the constant of 
>> motion in the NPT ensemble is the energy of the system + the energy of the 
>> thermostat + energy of barostat, as shown in this lecture 
>> <https://www.cp2k.org/_media/events:2016_summer_school:cp2k_moving_atoms.pdf> 
>> from 2016, so this would suggest that the fluctuations are due to the 
>> thermostat or barostat, right? Like the TIP3P NPT run, I used a thermostat 
>> timecon of 100 fs and a barostat timecon of 1000 fs and after 21 125 fs 
>> (blue vertical line in figures), I restarted the calculation and increased 
>> the thermostat timecon to 500 fs. I accidentaly set RESTART_BAROSTAT and 
>> RESTART_BAROSTAT_THERMOSTAT to FALSE upon restarting, but I'm unsure if 
>> this might have lead to the large jumps in the constant of motion later at 
>> t > 28 000 fs... 
>>
>> Same goes for the NPT_F calculation of my solvated TiO2 slab. I did a 10 
>> ps simulation with thermostat timecon 50 and barostat timecon 100 and 
>> everything seems to equilibrate except for the constant of motion which 
>> seems to steadily rise. After 10 ps of simulation I restarted the 
>> calculation and again accidentaly messed up my input and set 
>> RESTART_BAROSTAT and RESTART_BAROSTAT_THERMOSTAT to FALSE but this time I 
>> didn't even change the timecons of the thermostat and barostat, but this 
>> time I immediately see a significant jump in the constant of motion after 
>> which it continues to increase... Any idea how I can fix this so that the 
>> constant of motion stays...well...constant?
>>
>>  
>>
>> Kind regards,
>>
>> Léon
>>
>>  
>>
>> On Tuesday 23 January 2024 at 14:41:46 UTC+1 Léon Luntadila Lufungula 
>> wrote:
>>
>> Dear Matthias,
>>
>>  
>>
>> My AIMD calculations are on a slab wherein I keep my bottom layer fixed, 
>> but apparently REGION MASSIVE is not compatible with constraints, so I'll 
>> have to use REGION GLOBAL. Either way, thanks a lot for the detailed 
>> answer! For a novice MD practicioner as myself this is already enough 
>> information to get started! I'll try the values you suggested during my 
>> production run while keeping an eye on the temperature and pressure during 
>> the simulation; I guess when I've chosen my TIMECON too large, this will 
>> result in the system drifting away from the equilibrium values.
>>
>>  
>>
>> As always, thanks a lot for all the help!
>>
>>  
>>
>> Kind regards,
>>
>> Léon
>>
>> On Tuesday 23 January 2024 at 14:21:05 UTC+1 Krack Matthias wrote:
>>
>> Dear Léon
>>
>>  
>>
>> Using a MASSIVE CSVR thermostat with a small TIMECON of about 50 is 
>> appropriate and efficient for equilibration. Thereafter, the TIMECON for 
>> the thermostat can be increased to 500-1000 for the sampling (production) 
>> runs and also a GLOBAL instead of a MASSIVE thermostat can be used, if 
>> possible, to reduce the impact of the thermostat on the dynamics. You will 
>> find different opinions about the absolute and relative values for the 
>> TIMECONs of thermostat and barostat in the web or literature. I suggest to 
>> use a two to five times larger TIMECON for the barostat compared to the 
>> thermostat. You have, however, to keep in mind that the appropriate (best) 
>> parameters for baro- and thermostat depend on the type of system and the 
>> selected conditions eventually.
>>
>>  
>>
>> HTH
>>
>>  
>>
>> Matthias  
>>
>>  
>>
>> *From: *cp... at googlegroups.com <cp... at googlegroups.com> on behalf of 
>> Léon Luntadila Lufungula <Leon.luntad... at uantwerpen.be>
>> *Date: *Tuesday, 23 January 2024 at 11:47
>> *To: *cp2k <cp... at googlegroups.com>
>> *Subject: *Re: [CP2K:19801] Cons. Qty for different ensembles + Pressure 
>> fluctuation
>>
>> Hi Matthias,
>>
>>  
>>
>> So I started an NpT directly from the inputs you gave me with a box 
>> containing 216 water molecules. After a 300ps equilibration (plots thus 
>> starrt at 0.3 ns), the pressure fluctuations are again in the 10kbar range 
>> which is okay as you pointed out, while the average pressure is around -14 
>> bar, so tha'ts also within the 20 bar range you mentioned. The energy drift 
>> per atom is also below 1K so everything seems to work reasonably. 
>>
>>  
>>
>>
>> I'm starting my AIMD simulations as we speak, so hopefully all goes well 
>> there too. I read that it is best to do a short equilibration with a strong 
>> coupling to the thermostat (e.g. CSVR, TIMECON 20 and REGION MASSIVE), 
>> before going to a weaker coupling (e.g. CSVR, TIMECON 200 and REGION 
>> GLOBAL), but is there also a rule of thumb w.r.t. the barostat TIMECON? I 
>> regularly see TIMECON 1000, but is it useful to set it to TIMECON 100 in 
>> this early equilibration?
>>
>>  
>>
>> Thanks for all the help already, it really helps me a lot given the fact 
>> that I don't have a senior researcher with a lot of hands-on CP2K 
>> experience around!
>>
>>  
>>
>> Kind regards,
>>
>> Léon
>>
>>  
>>
>> -- 
>>
>> 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+uns... at googlegroups.com.
>> To view this discussion on the web visit 
>> https://groups.google.com/d/msgid/cp2k/c498aa90-e014-4484-869e-a94877dd9813n%40googlegroups.com 
>> <https://groups.google.com/d/msgid/cp2k/c498aa90-e014-4484-869e-a94877dd9813n%40googlegroups.com?utm_medium=email&utm_source=footer>
>> .
>>
>

-- 
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/28bfa788-4f09-498f-b760-615702976dfdn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240130/3c05d706/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: a101-solv20.md-NPT_F-eq2.png
Type: image/png
Size: 492458 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240130/3c05d706/attachment-0001.png>