non-bonded cutoffs

[pak - at - cgl.ucsf.EDU (Peter Kollman)]

I wanted to add to the non-bond cutoff debate.  Even though one typically
 uses a non-bonded cutoff that is too small, e.g. 8 angstoms, in many
 macromolecular simulations, it is likely better to use ewald summation than
 no cutoff in crystal simulations..(what does infinite cutoff really mean
 when one has periodic boundary conditions..it is really undefined until you
 define your unit cell, which you can make larger and larger)...even in
 solution simulations without periodic boundary conditions, with many
 formal charges, I would argue that infinite cutoffs might lead to
 artifacts because one can't run the simulations long enough to get appropriate
 averaging of all these formal charges to give realistic ion atmosphere
 effects...the cure may be worse than the disease...i do think longer than
 8 angstrom cutoffs are appropriate and recommend dual cutoffs(c.f. Berendsen,
 van Gunsteren,Darden,Charifson) which can consider effects out to, e.g. 15
 to 20 angstroms at little computational overhead...but in summary, there are
 many things more important in reducing simulation artefacts than switching
 to an infinite non-bonded cutoff--even if we could, i don't think we should.