The simulation runs as well. However, I do not know why am I getting such these warnings at the beginning of output file:
[cn02.cluster] pmix_mca_base_component_repository_open: unable to open mca_pnet_opa: libpsm2.so.2: cannot open shared object file: No such file or directory (ignored)
[cn02.cluster] mca_base_component_repository_open: unable to open mca_oob_ud: libosmcomp.so.3: cannot open shared object file: No such file or directory (ignored)
I searched and understood it is not related to the Lammps and maybe originated from OPENMP. But there is a warning indicating unreliability of the data which is my main concern here:
WARNING: Using I/O redirection is unreliable with parallel runs. Better to use the -in switch to read input files. (src/lammps.cpp:548)
I would be grateful, if you write any comments on them
That is only half correct. The messages originate from OpenMPI and NOT from OpenMP. That one letter makes a very big difference.
That warning is self-explanatory and more details can be found in the LAMMPS manual. In general, before posting questions here, please have the courtesy to search the manual and the forum for possible answers and think.
Using I/O redirection is a legacy feature in LAMMPS that goes back to the (ancient) times when LAMMPS was written in Fortran and there was no portable way in the Fortran standards (at that time) to process arguments to a command. This situation no longer applies and support for I/O redirection for reading LAMMPS input may be removed in the future to simplify a currently rather complex piece of code. Besides, there are technical reasons to not use it and for multi-replica simulations the -in flag is already required.
It is not about unreliability of data, it is about functionality. If your input would have loops with “jump SELF label” commands, for example, there is a good chance that it will fail. Also, not all MPI implementation libraries will reliably replicate stdin from rank 0 to all MPI ranks. To me personally, just expecting that this will work is asking the MPI library to do something that it should not be doing and that is technically very challenging. More importantly, it is ridiculously easily handled by replacing “<” with " -in ".