[gmx-developers] Question about ewald_LRcorrection() function

[Sheng Bi]

Sorry, some type-offs in previous question.
In gmx-4.6.7, ewald_LRcorrection(). We loop over excluded neighbours only in condition calc_excl_corr=1. But when using Verlet cut-off scheme, calc_excl_corr will be 0.


--
PhD student
State Key Laboratory of Coal Combustion,
School of Energy and Power Engineering,
Huazhong University of Science and Technology (HUST),
Room 302 Power Building, 
1037 Luoyu Road, Wuhan, Hubei 430074 China
Email: chrishengbee at hust.edu.cn
Phone: (86)27-87548122 
http://itp.energy.hust.edu.cn 

From: Sheng Bi
Sent: Tuesday, June 7, 2016 1:28 AM
To: gmx-developers at gromacs.org
Subject: [gmx-developers] Question about ewald_LRcorrection() function

Dear GMX developers:
               I find a abnormal phenomenon when using gromacs-4.6.7. I set up a system containing two walls(consist of freeze atoms), one wall has positively charges on atoms, another has negatively charges.  I put an water molecule(rigid spc/spce/tip3p model) between these two walls. Whole system seems like below:
               ****************************************
               * -          H                           +                                          *             
               * -             \                         +             vacuum                              *
              * -               O                     +                                          *
               * -             /                        +                                          *
               * -          H                           +                                          *
               ****************************************
               I use Verlet cut-off scheme, PME method, and ewald-geometry=3DC. Temperature of whole system is at 0K, controlled by v-rescale method. I run whole simulation only in OPENMP parallel.
               One can have following expected behavior of water molecule without MD simulation: Since it can be seen as an infinite panel capacitor, there is constant electric field between two walls. So the water will change it orientation, but won’t change it’s position(Since 0K).
               This phenomenon can be confirmed by gromcas-3.3.1 (although there is position shift, but quite small). Also, if one use higher version of gromacs, such as 4.6.7 or higher with ewald-geometry=3D, either no problem. However, if one use 3DC (and it should be), you can see an obvious force on water molecule.
               I hacked into source code, and find this might due to ewald_LRcorrection() function:
               Firstly, in gmx-4.6.7, ewald_LRcorrection(). We loop over excluded neighbours only in condition calc_excl_corr=0. But when using verlet cut-off scheme, calc_excl_corr will be 1. In gmx-3.3.1, we do loop over excluded neighbours without additional if() statement.
               Secondly, in gmx-4.6.7, , ewald_LRcorrection(). When do dipole correction on force, we loop over excl_load (Exclusion load distribution over the threads). In fact, most water model will have nrexel=2. So, as a result, we only do dipole correction on O atom, but do nothing on H atoms and we will see an obvious force on the whole water molecule. I think this is not purpose of 3DC。
               Best regards！
Sheng Bi
               
---
PhD student
State Key Laboratory of Coal Combustion,
School of Energy and Power Engineering,
Huazhong University of Science and Technology (HUST),
Room 302 Power Building, 
1037 Luoyu Road, Wuhan, Hubei 430074 China
Email: chrishengbee at hust.edu.cn
Phone: (86)27-87548122 
http://itp.energy.hust.edu.cn 


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20160607/5893628a/attachment-0001.html>