Re: MD neighbor lists
Hi, are you going for a massively parallel setp such as on a connection
machine, or just vectorizing as on a Cray?
One obvious scheme for 1-2 exclusions and such is to store a bit string on
each processor for each of the atoms it handles which tells it whether to
include or exclude such interactions. Since you will have at most 12 such
exclusions unless you go to highly coordinated systems, you could also just
store the specific interactions which are that way and either check for them
or do them twice, subtracting them from the sum on the second try which would
have the effect of not having done them in the first place.
I just use boxes for neighbor lists myself. But communicating this data could
be troublesome in a massively parallel environment....