Dear LAMMPS developers:
When I use the run_style verlet command, I don’t get an error, but I tried using the run_style respa command and it did get an error. I’m not quite sure why the error occurred. Can you help me.Here is my log file
lax.log (23.8 KB)
Neither do I know the cause and without means to reproduce it, I cannot debug it.
What is worrisome are the many warnings and errors about setting up your InfiniBand interconnect. That is something you need to consult your system administrator(s) or manager(s) about. Most likely there is a hardware failure somewhere.
Please see the guidelines post in this forum how to properly report issues and also consult the LAMMPS manual, especially: 11.3. Reporting bugs — LAMMPS documentation and 11.4. Debugging crashes — LAMMPS documentation
In your specific case you should see if you can reproduce the segfault with just 1 processor and without the replicate command.
Dear akohlmey,
Here is my simulation file, I’m just doing a simple test. Due to the C-H bond, I would like to use run_style respa.
run.in (2.5 KB)
system.data (280.9 KB)
What about the other things that I have asked you for or asked you to do?
Have you even read the sections of the documentation that I pointed out to you?
As for your input. Why do you set the timestep to 1.0 when this is the default?
Also, wouldn’t you get better performance if you just held all bonds including a hydrogen atom rigid with fix shake and keep run style verlet? You should be able to go up to a timestep of 2.0fs with that.
Finally, if using run style respa for performance reasons, you should also consider computing kspace only every other timestep. I.e. set a timestep of 4.0 and then use:
run_style respa 4 2 2 2 bond 1 angle 1 dihedral 2 pair 3 kspace 4
or
run_style respa 3 4 2 bond 1 angle 2 dihedral 2 pair 2 kspace 3
Yes,I have already consider fix shake with bonds that bond of H atom,and set timestep 1.0;There shouldn’t be much difference between 1.0 or 2.0, right？In fact, I want to try calculating angles and bonds every 0.25fs, dihedral angles every 0.5fs, Lj interactions and Kspace every 1.0fs. This is what I have seen in the literature and I want to give it a try. The run_style respa command I wrote should be fine, please confirm.I am still trying to solve the simulation error. I have reviewed the document you mentioned and tried to follow its steps
Then please share what you learned from those steps. If you want help you have to make it easier to help you. Keep in mind that we don’t read minds.
Yes, if I solve it, I will come back to report.Dear akohlmey,I would like to confirm if the run_style respa command I wrote matches my description, to ensure that I did not mistake the explanation provided by Lammps. Thank you
Have you tried removing the redundant timestep command?
LAMMPS should output the assigned timestep values for all r-RESPA levels.
Do you mean I should try removing the command “timestep 1.0”? Sorry, I haven’t tried it before. I’m not quite sure if I understand what you’re saying
Dear akohlmey,When I remove “timestep 1.0”,the error have missed.Thaks a lot!