The logfiles in this directory represent known good output. 
When running with 64 bit precision, deviations should be limited to the last digit of each number if run in MPI only mode.
During the first 1000 timestep the output does not depend on the run mode (such as number of MPI processes and 
OpenMP ranks or neighborlist style), it should only depend on the size of the system. For longer simulations the non associative 
and commutative nature of floating point operations can have an influence. In the long run the difference between two simulations
should be similar to the fluctuations within one run. This effect is stronger the more threads are used.
In single precision runs on GPUs (with the KokkosArray variant) this point is often reached after a few thousand 
timesteps. Indeed those deviations can already be seen by running the same simulation twice on the same GPU. Note that 
the automatic test is taking this into account for its judgmenet whether a run PASSES. 

To generate similar output use the provided input scripts 
in.lj.miniMD and in.eam.miniMD and the -s flag.

The different sizes are:
  4k : -s 10
 16k : -s 16
 32k : -s 20
108k : -s 30
256k : -s 40
864k : -s 60

