Domain-Decomposition Run Fails But Share-Memory Ends Successfully


My attempts to run MD simulations of simple proteins in a cubic water box as spatial domain decomposition (single node for testing) jobs failed because of either a LINCS error or a warning: ‘Triclinic box is too skewed’. The same md.tpr binaries ran successfully as shared memory jobs (also with gpu assistance), resulting in normally behaved MD simulations.

I wanted to ask you if domain decomposition jobs require a more rigorous minimization and equilibration, and thus if adjustments are needed of the input .mdp files? If no, could this be caused by improper configuration the software/hardware?