Re: [Towhee-users] VOLNPT output: an error message?
Brought to you by:
marcus_martin
Menu
▾
▴
Re: [Towhee-users] VOLNPT output: an error message?
|
From: Aric N. <agn...@en...> - 2014-10-07 09:32:59
|
Dear Marcus and Mike, I have had a second day to think over both or your comments and I believe that (perhaps) all has been illuminated. I provide this follow-up reply for the discussion board and future users who may need to search the knowledge base in the future. If I understand correctly: 1) bonds across the periodic boundaries are in violation of the VOLNPT.f because a cross-boundary bond cannot maintain a constant interatomic distance between two atoms with a volume change in the simulation cell. and 2) the VOLNPT move scales the position of the center of mass of the molecule with respect to the volume change and leaves the interatomic positions unchanged. These two points are related, but with respect to the montmorillonite sheets, if each sheet is defined as a molecule with bonds across the boundaries, then it violates 1) and 2) above because of the bonds. I have simulated each montmorillonite sheet without bonds (that is how the non-bonded CLAYFF forcefield intends) but received a line in the towhee_output that reads, “unit 2 not connected via bonding graph to atom 1 therefore MC moves will not be allowed to alter atom positions for this molecule”. This output was the reason for creating the model with bonds that violates 1) and 2) above. However, even if I could move each nonbonded montmorillonite sheet as a molecule, I would still have issues with VOLNPT.f because it adjusts the center of mass of the molecule while maintaining the interatomic distances in the molecule (i.e. cracks/ gaps would develop at the boundaries of the simulation cell with each pmcell move because the interatomic distances are unchanged). If I define each atom in the montmorillonite sheet as a molecule (as Marcus suggested), then the VOLNPT.f will adjust/scale the position of each atom/ molecule with each pmcell move and “fill” the simulation cell as the atom/molecule positions are scaled based on their center-of mass (and the montmorllonite relaxes). This simulation approach with each atom as a molecule is the closest to the simulation design that I am trying to execute, but I will have to accept a much more rigid montmorillonite framework and the density fluctuations. Other MD-NPT simulations of phyllosilicates (including montmorillonite) and CLAYFF have provided reasonable experimental data for the mineral unit cell parameters, interlayer d-spacing, and density [Cygan et al. 2004, J. Phys. Chem B. 108(4)]. Is there something about the anisotropic NPT MC algorithm that I am overlooking that is going to result in density fluctuations that are significantly greater than in previous anisotropic MD-NPT simulations? I don’t expect the simulation cell size to diverge significantly from the equilibrium values as determined via MD. Is this an unrealistic expectation? Thank you, Aric Newton On Oct 6, 2014, at 4:35 PM, Aric Newton <agn...@en...> wrote: > Dear Marcus, > > Thank you for your reply. I understand that the errant output is a result of the model violating the constraints of the pmcell move by having bonds in each montmorillonite layer that cross the boundary. The harmonic bonds that cross the boundary have spring constants of zero. Thus, the energy of these bonds is independent of the positions of the atoms (i.e. it’s always zero). This condition is a result of the parameters in the CLAYFF forcefield. I have only “created” these bonds because I would like each montmorillonite layer to be able to translate (as a unit) in the x-, y-, or z-direction as part of the MC moves. > > I can create an input file without these zero-force constant “bonds”, in which case each atom in the montmorillonite is a molecule, but I have been unable to group these individual molecules into a super-molecule that I can then translate as a unit. My starting simulation cell size is the equilibrium cell size from a 1.0 ns MD-NPT in LAMMPS, so I am not too worried about the changes in the MMT density. With an anisotropic MC-NPT that only has a five percent probability of performing a pmcell move, I suspect that density fluctuations in the montmorillonite should be limited and, I am assuming, negligible. Is this a reasonable assumption? > > These simulations are part of a thermodynamic integration. The model has only one box and the equilibrium cell parameters from the MD-NPT simulation confirm that my layer spacing and water content are consistent with the experimental condition that I am trying to simulate. So, if the pmcell move adjusts the center of mass of each atom in the montmorillonite sheets in a manner appropriate with the volume change (adjusting interatomic distances), then that is what I want. However, is there a way to also get the center of mass of each sheet to translate as part of the pmtracm command when in fact each sheet is composed of many molecules? The pmplane move sounds close to what I want, but I am unsure how to define a dynamic definition of the plane so that it always includes only a montmorillonite sheet regardless of previous translation moves during the simulation. Do you have any suggestions or is simulating with rigid montmorillonite layers the only option with the current move algorithms? > > Thank you, > > Aric Newton > > > On Oct 6, 2014, at 10:02 AM, Marcus Martin <me...@gm...> wrote: > >> On Fri, Oct 3, 2014 at 4:22 AM, Aric Newton <agn...@en...> wrote: >> >> I am using towhee for a one box MC-NPT simulation of two montmorillonite layers with an interlayer space containing the counter ions and water molecules. Each montmorillonite layer is treated as an individual molecule and moves as a unit via the pmtracm move probability. I have previously equilibrated the system at high temperature in the NVT ensemble without any errors, but the output that I am getting now that I have switched to the NPT ensemble leaves me a bit uncertain. In the towhee_output file, there are about 10,000 lines that read as below. >> >> VOLNPT: move attemped in box that has a periodic bonded molecule >> Molecule type: -31414 >> Chain number: 1 >> Box: 2 >> ... >> >> The *.pdb snapshots are being output during the simulation, but the rest of the output appears to be non-standard/ non-sensical- there is only one box in the simulation and the negative number as a molecule type seems odd. >> >> >> Indeed the error messages do mean there is a problem. When you perform a volume move in Towhee it scales the center of mass position of all the molecules (in a manner appropriate for the volume change) and then keeps all of the intramolecular distances the same. This is not possible when the "molecule" spans the periodic images with its bonding pattern. >> >> You would need a completely different move that scaled the intramolecular distances for the volume change. That would then lead to problems with any molecule that has a rigid bond length (as the attempt to change that bond length would result in an infinite energy) and also would be fairly inefficient as the intramolecular distances are quite sensitive. >> >> The volnpt.f file description from the towhee code manual notes that “you cannot use this move with a molecule that is bonded through the periodic boundary condition (like a zeolite).” I suspect that the bonds in the montmorillonite molecules are across the periodic boundary and trigger the VOLNPT output message. However, this same note is listed for the volnvt.f file and there are no warnings generated. Are these simulation results corrupt? Is there a way to salvage this simulation or work around this problem? The forcefield for the interactions is CLAYFF. So, the bonds, angles, torsions, and improper in the montmorillonite layers are zero energy. I had to define these terms so that each montmorillonite layer can be a molecule that can move/ translate during the simulation. Is there an alternate way to conduct this simulation and avoid the VOLNPT violation? >> >> Only if every atom in the montmorillonite was a separate molecule. However, be aware you are then changing the density of your montmorillonite and it is quite likely that is not what you are intending to sample. >> >> I'm not sure if you were intending to sample the layer separation distance, but in that case you would need a new implementation of the volume move that only scale the volume in one of the dimensions instead of uniformly. >> >> Marcus >> >> -- >> Marcus G. Martin >> 88 Martinez Road >> Edgewood NM 87015-8222 >> land (505) 286 4457 >> cell (505) 363 3179 >> www.photobirder.com >> towhee.sourceforge.net >> > > Aric Newton > > Laboratory of Nuclear and Environmental Materials > Faculty of Engineering > Hokkaido University > Sapporo, Japan > > > > > > ------------------------------------------------------------------------------ > Slashdot TV. Videos for Nerds. Stuff that Matters. > http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk_______________________________________________ > Towhee-users mailing list > Tow...@li... > https://lists.sourceforge.net/lists/listinfo/towhee-users Aric Newton Laboratory of Nuclear and Environmental Materials Faculty of Engineering Hokkaido University Sapporo, Japan |
