When I use deform to do the stretching simulation, there is always atomic loss, for example, ‘ERROR on proc 109: Bond atoms 80449 80427 missing on proc 109 at step 886047 (…/neigh_bond.cpp:65)’, or ‘ERROR on proc 51: Angle atoms 80423 80424 80425 missing on proc 51 at step 467916 (…/neigh_bond.cpp:283)’. I have tried to solve this problem in three ways:
reduced timestep;
changed neighbor list bin to larger ones;
update neighbor list more frequently,like neighbor_modify every 2
But all failed, and the problem still exists. I don’t have a better way to solve this problem now, I look forward to your help.
This error would usually happen, if you are stretching bonds too much, so that some that cross subdomain boundaries are are too long to fit into the ghost atom distance.
this can have two origins: 1) you are deforming your system too fast or too much, or 2) your non-bonded cutoff is rather short and thus also the ghost atom cutoff.
case 1) is a physics/model problem and cannot be solved by changing keywords or adding settings. 2) could be adjusted by setting a larger communication cutoff.
Thank you for your advice. But this problem occurs regardless of whether the stretch is fast or slow, much or less,so i guess the atomic loss is not the first cause.For the second case you said, I have tried to increase the cutoff radius, but the atom will still be lost. So the two case all failed.
I have tried another method to add the command “velocity all create 350 1234 dist gaussia” in the restart input file, the temperature 350k is larger than my initial temperature(“fix md all nvt temp 300 300 10 drag 0.2”). It can run for a while, but in the end it is still failed, and the thermodynamic data of the output is quite different from the previous. I think this method may not be suitable.
Now, the problem is still exists, i do not know how to solve it. I need your help, thank you.
I cannot provide more advice, since there is too little tangible information from you. also, you are drawing conclusions based on reasoning that i cannot follow.