temper N M temp fix-ID seed1 seed2 index
N = total # of timesteps to run
M = attempt a tempering swap every this many steps
temp = initial temperature for this ensemble
fix-ID = ID of the fix that will control temperature during the run
seed1 = random # seed used to decide on adjacent temperature to partner with
seed2 = random # seed for Boltzmann factor in Metropolis swap
index = which temperature (0 to N-1) I am simulating (optional)
temper 100000 100 $t tempfix 0 58728 temper 40000 100 $t tempfix 0 32285 $w
Run a parallel tempering or replica exchange simulation using multiple replicas (ensembles) of a system. Two or more replicas must be used.
Each replica runs on a partition of one or more processors. Processor partitions are defined at run-time using the -partition command-line switch. Note that if you have MPI installed, you can run a multi-replica simulation with more replicas (partitions) than you have physical processors, e.g you can run a 10-replica simulation on one or two processors. You will simply not get the performance speed-up you would see with one or more physical processors per replica. See the Howto replica doc page for further discussion.
Each replica’s temperature is controlled at a different value by a fix with fix-ID that controls temperature. Most thermostat fix styles (with and without included time integration) are supported. The command will print an error message and abort, if the chosen fix is unsupported. The desired temperature is specified by temp, which is typically a variable previously set in the input script, so that each partition is assigned a different temperature. See the variable command for more details. For example:
variable t world 300.0 310.0 320.0 330.0 fix myfix all nvt temp $t $t 100.0 temper 100000 100 $t myfix 3847 58382
would define 4 temperatures, and assign one of them to the thermostat used by each replica, and to the temper command.
As the tempering simulation runs for N timesteps, a temperature swap between adjacent ensembles will be attempted every M timesteps. If seed1 is 0, then the swap attempts will alternate between odd and even pairings. If seed1 is non-zero then it is used as a seed in a random number generator to randomly choose an odd or even pairing each time. Each attempted swap of temperatures is either accepted or rejected based on a Boltzmann-weighted Metropolis criterion which uses seed2 in the random number generator.
As a tempering run proceeds, multiple log files and screen output files are created, one per replica. By default these files are named log.lammps.M and screen.M where M is the replica number from 0 to N-1, with N = # of replicas. See the -log and -screen command-line swiches for info on how to change these names.
The main screen and log file (log.lammps) will list information about which temperature is assigned to each replica at each thermodynamic output timestep. E.g. for a simulation with 16 replicas:
Running on 16 partitions of processors Step T0 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11 T12 T13 T14 T15 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 500 1 0 3 2 5 4 6 7 8 9 10 11 12 13 14 15 1000 2 0 4 1 5 3 6 7 8 9 10 11 12 14 13 15 1500 2 1 4 0 5 3 6 7 9 8 10 11 12 14 13 15 2000 2 1 3 0 6 4 5 7 10 8 9 11 12 14 13 15 2500 2 1 3 0 6 4 5 7 11 8 9 10 12 14 13 15 ...
The column headings T0 to TN-1 mean which temperature is currently assigned to the replica 0 to N-1. Thus the columns represent replicas and the value in each column is its temperature (also numbered 0 to N-1). For example, a 0 in the 4th column (column T3, step 2500) means that the 4th replica is assigned temperature 0, i.e. the lowest temperature. You can verify this time sequence of temperature assignments for the Nth replica by comparing the Nth column of screen output to the thermodynamic data in the corresponding log.lammps.N or screen.N files as time proceeds.
You can have each replica create its own dump file in the following manner:
variable rep world 0 1 2 3 4 5 6 7 dump 1 all atom 1000 dump.temper.$rep
Each replica’s dump file will contain a continuous trajectory for its atoms where the temperature varies over time as swaps take place involving that replica. If you want a series of dump files, each with snapshots (from all replicas) that are all at a single temperature, then you will need to post-process the dump files using the information from the log.lammps file. E.g. you could produce one dump file with snapshots at 300K (from all replicas), another with snapshots at 310K, etc. Note that these new dump files will not contain “continuous trajectories” for individual atoms, because two successive snapshots (in time) may be from different replicas. The reorder_remd_traj python script can do the reordering for you (and additionally also calculated configurational log-weights of trajectory snapshots in the canonical ensemble). The script can be found in the tools/replica directory while instructions on how to use it is available in doc/Tools (in brief) and as a README file in tools/replica (in detail).
The last argument index in the temper command is optional and is used when restarting a tempering run from a set of restart files (one for each replica) which had previously swapped to new temperatures. The index value (from 0 to N-1, where N is the # of replicas) identifies which temperature the replica was simulating on the timestep the restart files were written. Obviously, this argument must be a variable so that each partition has the correct value. Set the variable to the N values listed in the log file for the previous run for the replica temperatures at that timestep. For example if the log file listed the following for a simulation with 5 replicas:
500000 2 4 0 1 3
then a setting of
variable w world 2 4 0 1 3
would be used to restart the run with a tempering command like the example above with $w as the last argument.
This command can only be used if LAMMPS was built with the REPLICA package. See the Build package doc page for more info.