Trajectory replay¶

For some applications it is necessary to run calculations on all (or a subset of) frames from a trajectory obtained in another job. Often this is done with a different engine in order to learn something about the differences in the potential energy surface between the engines. This could easily be scripted with PLAMS, but since the 2022.1 release the AMS driver itself also implements a convenient “Replay” task to do this.

The input for the replay task is quite minimal. The following input file would replay every tenth of the first 1000 frames on the trajectory of oldjob with the DFTB engine to calculate nuclear gradients:

Task Replay

Replay
   File oldjob.results
   Frames 1:1000:10
End

Properties
   Gradients True
End

Engine DFTB
EndEngine

Note that it is not necessary to specify a System block in the input of a replay job, as the system will just be loaded from the job to be replayed. What properties are evaluated for each frame in the trajectory can be configured in the Properties block of the input file. For a list of available properties, see for example the the manual pages on Gradients, Hessian, Stress tensor, Elasticity and Dipole moment, Polarizability, Bond orders.

Depending on the kind of job that is being replayed, you will get a slightly different default behavior for the selection of frames to be replayed:

• For PES scans only the frames corresponding to the converged PES points will be replayed. The resulting ams.rkf output file will contain the PESScan section, as if it had been produced by running a PES scan job.
• For Nudged Elastic Band (NEB) jobs only the frames corresponding to the converged images are replayed. The resulting ams.rkf file will have an NEB section that looks like it is from a NEB job that converged immediately.
• All other job types (such as Molecular Dynamics) will just replay whatever is on the History section on the ams.rkf file. By default all frames are replayed, but a subset can be chosen with the Replay%Frames keyword.

Note that the replay task does not use driver level parallelism. The engine is called on each frame in turn, always restarting from the previously calculated frame. (The restarting should not cause problems, but it can be disabled with the EngineDebugging%IgnorePreviousResults option, see developer options).

Details of the Replay task can be configure in the Replay block in the input.

Replay
   File string
   Frames integer_list
   StoreAllResultFiles Yes/No
End

Replay

Type:	Block
Description:	Configures the details of the Replay task.

File

Type:	String
GUI name:	Restart from
Description:	Provide an ams.rkf file (or a .results folder) from a previously run job to replay. The file needs to contain a History section.

Frames

Type:

Integer List

Description:

List of frames from the History section to recompute. If not specified the recomputed frames are determined automatically based on the task of the job that is being replayed: PES scans and NEB calculations will only have the converged points replayed, while all other tasks will have all frames recomputed. Specifying the frames to recompute in the input is probably only useful when replaying trajectories from MolecularDynamics calculations.

StoreAllResultFiles

Type:	Bool
Default value:	No
Description:	If this option is enabled AMS will produce a separate engine output file for every replayed frame. While basic properties like energy, gradients, stress tensor, etc. are stored anyway on the History section in the AMS driver output file (if they were requested in the Properties block), engine specific properties (e.g. excitations energies from ADF) will only be available if the full result files are stored.