Hi Fancy,
This is a significantly more complex question than you realize. To begin
with when you run periodic boundary calculations in AMBER they use the
Particle Mesh Ewald method by default. I suggest looking up how this method
works. Essentially it gives you infinite electrostatics without doing
infinite work. In this situation the cutoff merely serves to divide up the
calculation between a direct space part and a reciprocal space part. Hence
in terms of electrostatics you really do not gain much by adjusting the
cutoff. You'd don't improve the accuracy by much, you simply need to do more
direct space calculations, which are expensive, and less FFT calculations
for the reciprocal part which are cheap.
The VDW interactions, however, are truncated at the cutoff and there is not
a logical equivalent to PME for dealing with these. However, VDW
interactions, due to the 1/r^6 dependence fall off very quickly and are
typically negligible by 8 angstroms. Hence the reason the cut off is set to
8 angstroms by default. Setting the cutoff below this value will lead to
large errors due to truncation of the VDW so you should never set it less
than 8 angstroms. You can gain a little bit by going to 9 angstroms and
maybe a tiny bit more by going to 10 angstroms. The difference is likely,
though, to be negligible.
Thus simply asking what the cutoff should be based on the system you are
simulating is largely meaningless. Certainly you cannot make things worse by
increasing the cutoff, you will always improve things but in smaller and
smaller increments. Meanwhile your calculation time will increase
considerably. Thus you really have to make a decision between whether you
gain more by using an 8 Angstrom cutoff and getting more sampling done than
using a 10 angstrom cutoff and getting less sampling done. Typically you
will improve your results more by doing more sampling.
If, however, you are doing an implicit solvent or gas phase calculation then
there is no PME and the electrostatics are truncated at the cutoff. In this
case you should typically not use a cutoff, set cut to larger than your
system size, in order to avoid artefacts due to truncation of the charge
interactions.
I hope this helps,
All the best
Ross
> -----Original Message-----
> From: amber-bounces.ambermd.org [mailto:amber-bounces.ambermd.org] On
> Behalf Of fancy2012
> Sent: Saturday, June 05, 2010 6:21 AM
> To: AMBER Mailing List
> Subject: Re:Re: [AMBER] cutoff
>
> Dear Bill,
>
> Thanks very much for your explanation! what I study is doing MD
> simulation of EGFR in complex with an inhibitor, so is cutoff of 8
> angstrom enough? Or 10 is better? Thanks!
>
> All the best,
> fancy
>
>
>
> $B:_(B2010-06-05$B!$(B"Bill Ross" <ross.cgl.ucsf.EDU> $B<LF;!'(B
> >> When we do MD simulations using amber, does cutoff = 10 angstrom
> will
> >> make much difference from 8 angstrom?
> >
> >In principle over a long simulation it might. Another aspect is that
> >box clearance should be > cutoff, so with 10A cut you would need a
> >bigger box, so the expense increases due to more water as well as
> >more pair evaluations.
> >
> >> When choosing 8 angstrom, the calculation will be much faster than
> >> 10 angstrom?
> >
> >Yes, you can try it and see what the difference is for your system.
> >
> >> How should we choose?
> >
> >I would read the literature for experiments similar to yours.
> >Without knowing what you are studying, there is no way to know.
> >
> >Bill
> >
> >_______________________________________________
> >AMBER mailing list
> >AMBER.ambermd.org
> >http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jun 05 2010 - 12:30:03 PDT
