I'm running a small test coarse-grain simulation and after about 370000
steps I get this error:
ERROR: Lost atoms: original 404 current 402
The system is periodic in all three dimensions. How can I lose atoms?
If you blow an atom way outside the box on a step between re-neighborings,
then you can lose it, especially in parallel.
Steve
Is there any way to prevent this? I've tried setting a larger skin value
(4.0 in real units), but it just happens at a later point in time.
Naveen
The fact that atoms are being blown out of the box is
an indication of a bug or physics problem, not something
that a bigger neighbor skin will fix. So you need to find
what atom is disappearing and why. You could try reneighboring
every timestep and dumping coords, forces, etc every step
to try and track it.
Steve
The problem is that it happens after some 1 million time steps, and
everything before that looks fine (ie the energies are stable, the
trajectory looks normal). Is there a way to generate a one frame
trajectory that is rewritten every time? I guess I can write out a restart
file. Is there a tool within pizza that will take an input file and data
file and calculate all instantaneous forces and energies? I've noticed
there is for pair energies but not for the entire forcefield.
Naveen
Assuming it is reproducible, I would write a restart file right
before the crash. Then you can restart and run very short
simulations (that hopefully will still crash) to debug
what is happening ...
Steve
Hello,
Are there any general guidelines for when to turn newton pair or newton
bond on or off? Ie, we have a cluster with relatively fast processors
(dual 2.4 GHz) butslow interconnect (gigabit ethernet). Would it be faster
to turn off the newton pair and bond communications?
Naveen
I would just try a short simulation both ways and see what timings
you get.
Steve