The FFEA runner has the possibility to continue or extend a simulation that was completed or even crashed. In the case of calc_noise being enabled we take care that random number generator we use, RngStreams, recovers the state of the last time step calculated, ensuring that unwanted correlations arise.
When a simulation is to be continued, one will need:
The previous run will have provided this set of files, their names being specified in the FFEA input file through keywords:
trajectory_out_fnamecheckpoint_out , with default name <ffea-input-file>.fcpmeasurement_out_fnamedet_measurement_out_fname, with default name <ffea-measurement-file>.fdm, only if:ffea was not run with the -d flag orblob.In addition one needs to modify the already used FFEA input file, to indicate that the will to extend to simulations. Specifically one needs to set:
< restart = 1 >< checkpoint_in = <valid-checkpoint-file> > while keeping trajectory_out_fname and measurement_out_fname unchanged.There is a last rule that applies. Values for files checkpoint_out and checkpoint_in must differ. Therefore, if in the previous run checkpoint_out was not specified we recommend to rename the old checkpoint_out file with its default name to something else and then use it as checkpoint_in.
Output files will never be overwritten by the FFEA runner, but will be renamed instead. This way, if a new simulation were launched by error in the same folder where a previous simulation was run, every one of the following files:
checkpoint_outtrajectory_out_fnamemeasurement_out_fnamedet_measurement_out_fnamewill be renamed to __<old-file-name>__bckp.<N>, where N is the smallest natural number found so that the resulting text does not match the name of any existing file in the working folder.