[gmx-developers] virial symmetry

[Alexander Herz]

Hi,

no I haven't changed the walls.

the virial contains : r_x*F_z, r_y*F_z and r_z*F_z in components other
then [ZZ][ZZ]
But the only correction made to the virial for the walls is for the
zz,zz component (last line of the wall code you copied)!
So (unlike for the rest of the pair potentials like lennard jones,
estat, etc.) the wall contribution to the virial via the atom forces is
position dependent!

To reproduce the problem:
make a small sim with 2 walls which are identical except density (make
em differ by a factor of 10 or so)
and then look at the average virial at the end of the md.log, it will be
asymmetric (so theta_ij!=theta_ji).

to fix the problem I'd suggest:

inside the force calculating loop:
instead of

Fwall[w]+=(-F);

do:
//force MUST act on the same x/y coord as the particle otherwise newtons
3rd law is violated
fr->vir_wall[XX]+=-0.5*x[i][XX]*(-F);
fr->vir_wall[YY]+=-0.5*x[i][YY]*(-F);
fr->vir_wall[XX]+=-0.5*box_zz*(-F);

and inside sim_util.c:

vir_part[XX][XX]+=fr->vir_wall[XX];
vir_part[YY][ZZ]+=fr->vir_wall[YY];
vir_part[ZZ][ZZ]+=fr->vir_wall[ZZ];

Alex


Berk Hess wrote:
> Hi,
>
> At the end of do_walls it says:
>
>  /* Since the lower wall is at z=0 it does not contribute
>   * to the single sum virial.
>   */
>  if (nwall == 2)
>    fr->vir_wall_zz = -0.5*box_zz*Fwall[1];
>
> This line should ensure that the virial only depends on the distance
> between the particles and the upper wall.
>
> I don't understand what you mean with asymmetric.
> The wall only acts on the z-component of the forces and both the force
> and distance have no x and y
> components, therefore the wall forces only occur in the zz element of
> the virial tensor.
>
> Have you modified the code such that this correction no longer holds?
>
> Berk
>
> Alexander Herz wrote:
>> Having looked a bit closer into this, it seams that the definition of
>> the virial used in gromacs
>> (using absolute particle coords) is obviously only translation invariant
>> as long newtons 3rd law
>> is obeyed inside the system and no external forces act on the system (so
>> all forces sum to zero).
>> Apparently this condition is violated when using 2 walls with different
>> parameters like density???
>>
>> Alex
>>
>> Alexander Herz wrote:
>>  
>>> Hi guys,
>>>
>>> I'm having some problems with the walls feature (new in gromacs 4.0).
>>>
>>> If I understand correctly then I do get a symmetric virial tensor
>>> even for
>>> systems which are inhomogenous as long only pair potentials are used
>>>
>>> (Theta=-0.5*sum(r_ij*F_ij) where r_ij is r_j-r_i (so the distance
>>> between to particles acting on each other)
>>> and F_ij is the force acting on them;with F_ij=-F_ji the contributions
>>> cancel out and the viral is symmetric )
>>>
>>> Now if I use 2 walls with different density I get an asymmetric virial
>>> and looking at the wall's source
>>> it seams that for walls not r_ij is used but absolute wall coordinate
>>> (box_zz for wall2 and 0 for wall 1
>>> giving a zero contribution to the viral for wall1). Also for the
>>> individual atoms it seams the absolute position
>>> is used rather than r_ij (looking at calc_vir() in calcvir.c).
>>>
>>> So why do the walls use the absolute z coordinate instead of
>>> particle_z-wall_z and why
>>> do I get an asymmetric virial for walls with different density but not
>>> for the same
>>> system without walls which is still asymmetric in z (some crystal
>>> occupies the lower z half off the system, the upper
>>> half is filled with water).
>>>
>>> Thx,
>>> Alex
>>>
>>> _______________________________________________
>>> gmx-developers mailing list
>>> gmx-developers at gromacs.org
>>> http://www.gromacs.org/mailman/listinfo/gmx-developers
>>> Please don't post (un)subscribe requests to the list. Use the www
>>> interface or send it to gmx-developers-request at gromacs.org.
>>>       
>>
>> _______________________________________________
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://www.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the www
>> interface or send it to gmx-developers-request at gromacs.org.
>>   
>
> _______________________________________________
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the www
> interface or send it to gmx-developers-request at gromacs.org.