[CP2K-user] [CP2K:19162] Re: GEMC module does not keep stable liquid and vapor phases for water

['sing...@umn.edu' via cp2k]

Thanks for providing your files Slava.  I will run for this case.

I am not sure if FIST can handle polarizable FF.  I guess you can still 
estimate and come up with a non- polarizable FF to be used in FIST 
(pre-sampling).
I will let you know what I get from running your water files.

On Saturday, July 29, 2023 at 1:02:50 PM UTC-5 Vyacheslav Bryantsev wrote:

> Hi Ramanish,
>
> Please find a complete input file attached to this email.
> It does take some time to see the problem. After a few hours run on 3 
> intel nodes (36 cores) 12 water molecules moved from one box (liquid) to 
> the other (gas).
> This would continue until about an equal density in each box. 
>
> I get your point about having a decent force field for doing pre-sampling. 
> Would it be possible to use a polarizable force filed (PIM model) in the 
> FIST module of CP2K? This is for modeling  liquid AlCl3 consisting of pure 
> Al2Cl6 dimers. The PIM force field is available for Al2Cl6.
>
> Thank you,
> Slava
> On Friday, July 28, 2023 at 10:55:32 PM UTC-4 sing... at umn.edu wrote:
>
>> Moreover, I would like to add that people experienced in FPMC using CP2K 
>> have told me that it does not matter much if you do not have a very good 
>> force field for your molecules. The biasing potential (ff) is more or less 
>> used to avoid overlaps during pre-sampling. You could estimate the epsilon, 
>> sigma, and charges using established FFs and the bond and angle constants 
>> can be estimated by fitting a parabolic function to the bond length and 
>> angle energy scan (you could do that in Gaussian). 
>>
>> But, as I do not know about your system, I do not know if it would still 
>> be applicable for the molecules you want to simulate.
>>
>> Best,
>> Ramanish
>>
>> On Friday, July 28, 2023 at 9:49:56 PM UTC-5 sing... at umn.edu wrote:
>>
>>> Hi Slava, 
>>>
>>> Thanks for explaining the issue. I think I can try to run for this case 
>>> and see if I face the same issue. Can you please tell me after how many 
>>> steps you see the densities of the two boxes becoming nearly equal?
>>>
>>> Best,
>>> Ramanish
>>>
>>> On Friday, July 28, 2023 at 7:37:39 PM UTC-5 Vyacheslav Bryantsev wrote:
>>>
>>>> Hi Ramanish,
>>>>
>>>> Thank you for your comments and a suggestion. 
>>>> I am looking for a way to run GEMC simulations in the case when we do 
>>>> not have a good force field.
>>>> After I turned off the  LBIAS and made the DFT method exactly the same 
>>>> in two boxes, I still experience the same problem of getting water 
>>>> distributed about equally between the two phases. 
>>>> The two boxes input files along with the output *mc_molecules* file 
>>>> showing a drift of water molecules from one box to the other are attached. 
>>>> Again, if the force field is used in the second box, as in the provided 
>>>> example, https://www.cp2k.org/howto:gemc,  there is no issue. 
>>>>
>>>> Therefore, the problem of doing all the MC moves with DFT still 
>>>> persists. 
>>>>
>>>> Thank you,
>>>> Slava
>>>>
>>>>
>>>>
>>>> On Friday, July 28, 2023 at 1:36:05 PM UTC-4 sing... at umn.edu wrote:
>>>>
>>>>> Hi Vyacheslav,
>>>>>
>>>>> Could you please elaborate why you want to use DFT as biasing 
>>>>> potential instead of a force field (FF) (for e.g. TIP3P)?
>>>>> My understanding of the biasing potential is that a FF can be used to 
>>>>> conduct some computationally cheap steps between two expensive DFT energy 
>>>>> evaluations. If LBIAS is on, there will be NMOVES number of steps conducted 
>>>>> using the FF you define in the biasing potential. 
>>>>>
>>>>>  [image: Capture.PNG]
>>>>> At the end of FF pre-sampling NMOVES, there is a comparison between 
>>>>> the change in DFT energy and FF energy b/w the initial and final stages to 
>>>>> determine if the final state would be accepted or not. 
>>>>>
>>>>> So, if you want to use DFT as your potential for pre-sampling, then it 
>>>>> doesn't make sense to have LBIAS on. You can set LBIAS to FALSE, and do all 
>>>>> the MC moves based on the DFT potential you define in box1.inp and box2.inp.
>>>>> Let me know if this makes sense.
>>>>>
>>>>> Best,
>>>>> Ramanish
>>>>> On Wednesday, May 31, 2023 at 6:54:19 AM UTC-5 Vyacheslav Bryantsev 
>>>>> wrote:
>>>>>
>>>>>> Dear CP2K Developers,
>>>>>>
>>>>>> I am looking into using the Gibbs Ensemble Monte Carlo (GEMC) methos 
>>>>>> in CP2K to compute the liquid-vapor phase diagram using a fully DFT-based 
>>>>>> method.
>>>>>>
>>>>>> However, I have problems stabilizing liquid and vapor phases when 
>>>>>> doing simulations fully ab intio, as described below.
>>>>>>
>>>>>> By taking input files for water from
>>>>>>
>>>>>> https://www.cp2k.org/howto:gemc 
>>>>>>
>>>>>> and minimally changing them, I was able to run the GEMC model and get 
>>>>>> sensible results for water. This input utilizes the BLYP functional and an 
>>>>>> empirical model for pre-sampling, employing the FIST methods as specified 
>>>>>> in the bias_template.inp file. The input files and the output mc_molecules 
>>>>>> file are attached. The latter file  shows how many molecules are present 
>>>>>> during the GEMC run in the liquid phase (between 64 and 62 water molecules) 
>>>>>> and the gas phase (between 0 and 2 water molecules).
>>>>>>
>>>>>> The problem starts where I substitute the empirical potential for 
>>>>>> water with the actual BLYP method in the bias_template.inp file (I renamed 
>>>>>> that file to be bias_template-DFT.inp). Please note that I keep all other 
>>>>>> settings the same as before, but only substituting the empirical water 
>>>>>> model with the DFT model.  Now, the results of simulations have no sense. 
>>>>>> The water molecules from the liquid box move to another gas phase box until 
>>>>>> two phases have roughly the same number of water molecules. The  
>>>>>> mc_molecules-DFT shows this problem. Using fully DFT calculations is 
>>>>>> supposed to give very similar or slightly improved results, but instead it 
>>>>>> produces very different and unexpected results. It looks like that 
>>>>>> replacing the FIST model with the DFT model causes some changes in the 
>>>>>> program that treats swapping molecules between the two boxes differently 
>>>>>> (almost like allowing them to swap only one way). 
>>>>>>
>>>>>> At this stage, it is crucial to consult CP2K developers to identify 
>>>>>> the factors that contribute to a significant change in the software's 
>>>>>> behavior when transitioning from the utilization of the empirical potential 
>>>>>> in bias_template.inp to using DFT. 
>>>>>>
>>>>>>
>>>>>> Thank you,
>>>>>>
>>>>>> Slava 
>>>>>>
>>>>>> Vyacheslav Bryantsev
>>>>>>
>>>>>> Chemical Separations Group
>>>>>>
>>>>>> Oak Ridge National Laboratory
>>>>>>
>>>>>

-- 
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/a1e3532b-396f-49e5-bc75-ea6fb18c395cn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20230731/59469bb3/attachment.htm>