TrajectoryObserver

Records the state of an atomic structure at each recorded step of a relaxation or MD run.

TrajectoryObserver is attached as a callback to an ASE Optimizer (via BaseCalculator.relax()) or an ASE MD integrator (via BaseMDCalculator.run()); ASE invokes it every interval steps, and it appends the current energies, forces, stress, cell, and atomic positions/species to internal lists. For MD runs it additionally records temperature and velocities. It behaves like a read-only Sequence: indexing or iterating over it yields the per-step property tuple. It can be exported as a pandas.DataFrame (as_pandas()) or pickled to disk (save()).

You should not construct or attach a TrajectoryObserver yourself. relax() and run() already construct and attach one internally.

For relax(), the returned dict's "trajectory" key is the TrajectoryObserver instance, so you can call its methods directly on the result:

calc = SomeCalculator(fmax=0.05, steps=500)
result = calc.relax(structure)

obs = result["trajectory"]
df = obs.as_pandas()
print(df["potential_energies"])
obs.save("relaxation.traj.pkl")

For run() (molecular dynamics), the recorded properties are already unpacked into plain lists on the results dict (e.g. result["total_energy"], result["forces"]). There's no "trajectory" key to pull an observer instance from.

See the API Reference if you need the full list of properties a TrajectoryObserver records.