Hi Axel,
I vote for a 2.0*r default cutoff for the harmonic bonds. This should be large
enough
for most cases. As Christoph proposed, we could use a negative cut off for "no cut
off".
Then we don't need an additional switch for this.
Torsten.
Axel Arnold wrote:
Hi all,
currently, there is a problem with the harmonic potential on multi-processor
systems: to calculate any bonded potential, both bonding partners need to be
on the same processor, at least as a ghost. To ensure this, the maximal
cutoff needs to be larger than the distance of the two bonding partners. For
the harmonic potential, this distance can in general be infinite; but
currently, the maximal cutoff is only set to the _minimum_ of the harmonic
potential.
There are some solutions:
- use the skin to increase the max_cutoff: Bad idea, since Espresso will then
create huge Verlet lists, and on the other hand, particles can move at most
skin/2 before the ghost shells are recalculated, leaving essentially the same
problem.
- give the harmonic potential also a breaking distance, not only a minimum:
that is, we specify a second length, after which the potential breaks. That
will however change the syntax of the harmonic potential, although we could
provide a default (e.g. 2*r). In addition, we could also provide a switch to
turn off breaking. At least on single processors, a harmonic bond would never
break.
Any votes? Suggestions?
Axel