[gmx-users] DPPS/Na in water system

[Erik Lindahl]

David L. Bostick wrote:

>Hello,
>
>I have written emails earlier about equilibration of a DPPS bilayer from
>scratch.  I have gromacs 3.0 installed in single precision and am running a
>bilayer with 2624 waters , 128 sodiums, and 128 DPPS molecules.  
>
>The topology for a DPPS molecule has a net charge of EXACTLY -1.  So the
>entire system should have a net charge of exactly zero because of the
>Sodium counterions.  However upon grompp, I get,
>
>checking input for internal consistency...
>Excluding 3 bonded neighbours for DPS 128
>Excluding 2 bonded neighbours for SOL 2624
>Excluding 1 bonded neighbours for NA 128
>Warning 1 [file "../ps.top", line 22]:
>  System has non-zero total charge: 2.437830e-05
> 
>processing coordinates...
>double-checking input for internal consistency...
>
>Granted this is a very tiny charge, why is it there?  How will it affect
>calculation? 
>
Hi David,

Actually I'm not sure it is there... The problem is that most atomic 
charges can't be
represented represented exactly in binary power-2 floating-point. This 
means there
is an error of about 1e-7 for each atom. When you add the charge from 
10000 atoms
or so, the error will increase as sqrt(N), so something around 1e-5 is 
reasonable.

Now, we never *use* this total charge in the simulation, and the error 
it might cause
in the PME routines is much lower than the inherent accuracy of PME 
which is around
1e-3 to 1e-4 depending on your grid spacing. I've seen exactly the same 
warning for
my uncharged DPPC molecules.

I actually halfheartedly tried to fix it once, but it's not trivial 
since it should depend on the system
size, but we still want to catch a unit charge in a very large system. 
Don't worry if it's 1e-5 though!

>
>
>I am also still interested in why the harmonic bonds in my system seem more
>stable in 3.0 (allowing me to take a 2 fs timestep) than in 3.1 (allowing
>only .1 fs timestep) during my bilayer construction.
>


This I'm not sure about - haven't had time to speak to Berk yet :-)

Cheers,

Erik