Add a tutorial on time-resolved TEM from molecular dynamics

[TomaSusi]

Adopts the documentation contributed by @giolajide and @tgmaxson in #4, which shows how the finite exposure time of a TEM image averages over atomic motion. A Cu adatom on Cu(111) is imaged both from a single molecular dynamics snapshot and as an average over a 10 ps trajectory, and the underlying study (Olajide, Maxson & Szilvási, ACS Catal. 15, 18908, 2025) is added to the bibliography and cited.

The page goes under the advanced tutorials, and the frozen phonons walkthrough now links to it — the two are the same ensemble average over atomic configurations, differing only in what the configurations represent.

Changes to the contributed notebook

The physics is unchanged; the adaptations are for reproducibility, runtime, and house style.

• Reproducibility. EMT (parametrized for Cu) replaces the MACE/EMT fallback and the GPU autodetection, so the notebook runs anywhere. The molecular dynamics, the frozen phonons, and the shot noise are all seeded — two full re-executions agree to within 1.7% of pixels, inside the 3% tolerance used by the notebook tests.
• Runtime. The trajectory is sampled every 90 fs instead of every 6 fs, giving 112 snapshots instead of 1667. The whole notebook executes in about a minute, with no visible change to the averaged image.
• One physics fix. The original applied frozen phonons to all atoms, including the two layers held fixed during the MD, which left the substrate blurrier in the static panel than in the time-averaged one. Per-atom sigmas that vanish on the fixed layers make the two panels directly comparable.
• Figures. abtem.stack(...).show(explode=True, common_color_scale=True) replaces hand-transposed arrays and a magic /2e2 rescaling, the field of view is cropped to the surface (it was ~60% vacuum), and a line profile through the adatom quantifies the effect: a 19% peak in the short exposure becomes a broad 3% ripple after 10 ps.
• Prose. The "static" case is reframed as the short-exposure limit — long compared to a phonon period, too short for the adatom to move — which is what the frozen-phonon average actually represents. Absolute readthedocs URLs became internal cross-references, the to_zarr/from_zarr step became a tip rather than executed cells writing into the docs tree, and the MD trajectory is kept in memory so nothing is written to disk.

Checks

jb build . succeeds and adds no new warnings; both citations and all cross-references resolve; the notebook widget lint passes.

Closes #4

🤖 Generated with Claude Code