Re: [gmx-users] Twin range cut-off's: rlist and rvdw
On Wed, May 27, 2009 at 3:05 PM, Jussi Lehtola wrote: > On Wed, 2009-05-27 at 14:33 +0200, Yan Chai wrote: > > > It seems that in the twin range cut-off method, rlist does not only > > play a role as a cut-off for neighbor searching, but also as a cut-off > > for short-range interactions. Do I understand correctly? > > If you're using Coulombic cutoffs, then yes. If you're using PME, then > AFAIK the part r > rcut will be performed in Fourier space. Excuse me, to make sure whether my understanding is correct or not, might I check the following two examples? Case 1, let's consider PME and set rlist=rcoulomb. At time t=0, there are a particle O at the original point, a outward moving particle A within the radius of rlist=rcoulomb, and a inward moving particle B out of the radius of rlist=rcoulomb, as shown below O -- A -- rlist -- B, for t=0 I set nstlist=10. At t=0, the neighborlist of particle O is created and particle A is on the list but particle B is not. The Coulomb forces on particle O is also calculated, and F_OA is calculated in real space while F_OB is calculated in Fourier space: Fcoulomb=Fcoulomb_OA_real+Fcoulomb_OB_fourier, for t=0. Let's imagine when the time is between t=0 and t=10, before the second time for neighborlist update, particle A moves outside of the radius of rlist=rcoulomb and particle B moves inside of the radius of rlist=rcoulomb, as shown below O -- B -- rlist -- A, for 0___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php
RE: [gmx-users] Twin range cut-off's: rlist and rvdw
Hi, I don't think it matters much if you put the cut-off at a minimum in the rdf. The real issue is that the force is not updated beyond the cut-off, while the atoms there still move. You want as little motion, of better as little change in force due to the motion of the atoms as possible. The solution to this is to increase the cut-off as much as possible, but this is computationally unfavourable. I should also note the all versions of Gromacs up till now use a very simplistic twin-cut off setup with an integrator which is not reversible. This can lead to significant energy increase in the system. Therefore twin-range cut-off should not be used for Coulomb interactions, only for Lennard-Jones, where the forces are anyhow very small. We might consider implementing a proper multiple time step integrator for version 4.1 or 4.2. Berk > Subject: Re: [gmx-users] Twin range cut-off's: rlist and rvdw > From: jussi.leht...@helsinki.fi > To: gmx-users@gromacs.org > Date: Wed, 27 May 2009 16:05:25 +0300 > > On Wed, 2009-05-27 at 14:33 +0200, Yan Chai wrote: > > > It seems that in the twin range cut-off method, rlist does not only > > play a role as a cut-off for neighbor searching, but also as a cut-off > > for short-range interactions. Do I understand correctly? > > If you're using Coulombic cutoffs, then yes. If you're using PME, then > AFAIK the part r > rcut will be performed in Fourier space. > > > Do people think the effect of these inward moving and outward moving > > particles is small and the error for calculation can be tolerated? > > When particles move in and out of the cutoff range it induces static in > the properties. Ideally for monocomponent systems (such as metals) the > cutoff should be in an (absolute) minimum of the pair correlation > function, i.e. there aren't many particles in the region where the > cutoff is made. For molecular systems the decision of a cutoff is > harder, since some pair correlations stay quite constant and there is no > clear cut way to make the decision. > > The farther you go the fewer static you get. Roughly speaking, the > number of particles at cutoff range grows as r_cut^2, but for > Lennard-Jones the interaction weakens as r_cut^-6, resulting in an error > proportional to r_cut^-4. > -- > -- > Jussi Lehtola, FM, Tohtorikoulutettava > Fysiikan laitos, Helsingin Yliopisto > jussi.leht...@helsinki.fi, p. 191 50632 > -- > Mr. Jussi Lehtola, M. Sc., Doctoral Student > Department of Physics, University of Helsinki, Finland > jussi.leht...@helsinki.fi > -- > > ___ > gmx-users mailing listgmx-users@gromacs.org > http://www.gromacs.org/mailman/listinfo/gmx-users > Please search the archive at http://www.gromacs.org/search before posting! > Please don't post (un)subscribe requests to the list. Use the > www interface or send it to gmx-users-requ...@gromacs.org. > Can't post? Read http://www.gromacs.org/mailing_lists/users.php _ Express yourself instantly with MSN Messenger! Download today it's FREE! http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] Twin range cut-off's: rlist and rvdw
On Wed, 2009-05-27 at 14:33 +0200, Yan Chai wrote: > It seems that in the twin range cut-off method, rlist does not only > play a role as a cut-off for neighbor searching, but also as a cut-off > for short-range interactions. Do I understand correctly? If you're using Coulombic cutoffs, then yes. If you're using PME, then AFAIK the part r > rcut will be performed in Fourier space. > Do people think the effect of these inward moving and outward moving > particles is small and the error for calculation can be tolerated? When particles move in and out of the cutoff range it induces static in the properties. Ideally for monocomponent systems (such as metals) the cutoff should be in an (absolute) minimum of the pair correlation function, i.e. there aren't many particles in the region where the cutoff is made. For molecular systems the decision of a cutoff is harder, since some pair correlations stay quite constant and there is no clear cut way to make the decision. The farther you go the fewer static you get. Roughly speaking, the number of particles at cutoff range grows as r_cut^2, but for Lennard-Jones the interaction weakens as r_cut^-6, resulting in an error proportional to r_cut^-4. -- -- Jussi Lehtola, FM, Tohtorikoulutettava Fysiikan laitos, Helsingin Yliopisto jussi.leht...@helsinki.fi, p. 191 50632 -- Mr. Jussi Lehtola, M. Sc., Doctoral Student Department of Physics, University of Helsinki, Finland jussi.leht...@helsinki.fi -- ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] Twin range cut-off's: rlist and rvdw
On Wed, May 27, 2009 at 1:36 PM, Jussi Lehtola wrote: > In the twin range method [1] interactions that are in the range > rlist..rvdw are only calculated during neighborlist updates. In the mean > time they are considered to stay constant. > > The idea behind this is that when r>rlist the interactions are weaker > and also vary more slowly, thus one doesn't need to be as precise in > taking those interactions into account. > > [1] http://dx.doi.org/10.1002/prot.340060203 Thanks for your explanation and the link for the reference. I think you are right. The interactions between pairs within the range rlist..rvdw are computed and stored when neighborlists are updated, and added as constant to short-range forces within rlist at every step between successive neighborlist updates. It seems that in the twin range cut-off method, rlist does not only play a role as a cut-off for neighbor searching, but also as a cut-off for short-range interactions. Do I understand correctly? I also agree with you that interactions are weak when r>rlist. So it seems reasonable to use this twin range cut-off algorithm. However, if we compare the original Verlet list algorithm and the twin range cut-off algorithm, I think there is a shorting coming of the twin range cut-off algorithm to consider the particles which diffuse and across the boundary of rlist between successive neighborlist updates. The particles which move into the radius of rlist will not be calculated between successive neighborlist updates. On the other hand, some particles can move out of the radius of rlist between successive neighborlist updates. If rlist is really a cut-off for short range interaction, such out moving particle will not bring too much trouble for the calculation of forces. However, if rlist is only a cut-off for neighbor searching but not for short range interactions, such out moving particles will give a particle-based cut-off for short range interaction rather than a well-defined distance-based cut-off for forces. Do people think the effect of these inward moving and outward moving particles is small and the error for calculation can be tolerated? Thanks! Yan ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] Twin range cut-off's: rlist and rvdw
On Wed, 2009-05-27 at 11:00 +0200, Yan Chai wrote: > If my understanding above is correct, it seems that the concept or > the algorithm of neighborlist for the twin range cut-off's in Gromacs > is different from the original concept of Verlet neighborlist which is > discussed in the textbook on MD simulations, for instance, in the book > Understanding Molecular Simulation by Frenkel and Smit. The original > algorithm of Verlet list needs rlist>rvdw in order to contain all > particles which might have interactions. However, the twin range > cut-off's in Gromacs with rvdw>=rlist seems to be contradict to this > original idea of Verlet list and to miss some contribution to the > interaction by the particles which stay between rlist and rvdw in the > normal steps without updating the neighborlist. > > So my question is: what is the idea for Gromacs to take such > algorithm of twin range cut-off's? In the twin range method [1] interactions that are in the range rlist..rvdw are only calculated during neighborlist updates. In the mean time they are considered to stay constant. The idea behind this is that when r>rlist the interactions are weaker and also vary more slowly, thus one doesn't need to be as precise in taking those interactions into account. [1] http://dx.doi.org/10.1002/prot.340060203 -- -- Jussi Lehtola, FM, Tohtorikoulutettava Fysiikan laitos, Helsingin Yliopisto jussi.leht...@helsinki.fi, p. 191 50632 -- Mr. Jussi Lehtola, M. Sc., Doctoral Student Department of Physics, University of Helsinki, Finland jussi.leht...@helsinki.fi -- ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php
[gmx-users] Twin range cut-off's: rlist and rvdw
Dear Gromacs users, I have a question about the twin range cut-off's in Gromacs. If I choose vdwtype as Cut-off, the manual on run parameters tells me that I need to choose rvdw>=rlist in this case. I have read the section on the treatment of cutoffs in the manual and also the mailing list. My understanding on the twin range cut-off's is following: 1, Every nstlist steps, the neighborlist is updated according to the cut-off of neighborlist rlist and all interactions are calculated according to the cut-off of forces, such as rvdw; 2, Every step, all interactions of atom pairs in the neighborlist are calculated. If my understanding above is correct, it seems that the concept or the algorithm of neighborlist for the twin range cut-off's in Gromacs is different from the original concept of Verlet neighborlist which is discussed in the textbook on MD simulations, for instance, in the book Understanding Molecular Simulation by Frenkel and Smit. The original algorithm of Verlet list needs rlist>rvdw in order to contain all particles which might have interactions. However, the twin range cut-off's in Gromacs with rvdw>=rlist seems to be contradict to this original idea of Verlet list and to miss some contribution to the interaction by the particles which stay between rlist and rvdw in the normal steps without updating the neighborlist. So my question is: what is the idea for Gromacs to take such algorithm of twin range cut-off's? Thanks! Yan ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php