Py4dstem

py4DSTEM

What this library is for

py4DSTEM is a toolkit for reading, calibrating, visualizing, and analyzing 4D-STEM diffraction datacubes. It provides domain objects and workflows for virtual imaging, Bragg disk detection, center-of-mass/DPC analysis, strain and orientation mapping, phase retrieval, ptychography, and calibration.

When to use this vs. alternatives

• Use py4DSTEM for 4D-STEM datacubes, diffraction-pattern stacks, scan/detector axis handling, virtual images/diffraction, Bragg vectors, COM/DPC, strain, orientation mapping, and ptychographic phase retrieval.
• Use HyperSpy when the task is broader multidimensional microscopy/spectroscopy signal handling, EELS/EDX workflows, lazy loading, or interactive signal decomposition outside py4DSTEM-specific diffraction workflows.
• Use NumPy/scikit-image only for narrow image-processing steps after py4DSTEM has handled datacube loading, calibration, slicing, and diffraction semantics.
• Do not treat a 4D-STEM file as an arbitrary 4D array without documenting scan axes, detector axes, reciprocal-space calibration, beam center, and masks.

Canonical workflow

Start by loading a datacube, inspecting dimensions, setting or checking calibration, building virtual detectors/images, then moving to Bragg/COM/strain workflows only after the raw data and masks make sense.

import py4DSTEM

py4DSTEM.print_h5_tree("scan_4dstem.h5")
datacube = py4DSTEM.read("scan_4dstem.h5", datapath="root/datacube")
# For non-native microscope formats, use py4DSTEM.import_file(...) instead.

print(datacube.data.shape)  # typically (scan_y, scan_x, q_y, q_x)
datacube.calibration

# Average diffraction pattern and virtual bright-field image.
dp_mean = datacube.get_dp_mean()
bf = datacube.get_virtual_image(
    mode="circle",
    geometry=((0, 0), 20),
    centered=True,
)

# Inspect before quantitative analysis.
py4DSTEM.show(dp_mean)
py4DSTEM.show(bf)

# Bragg disk workflows need a probe/kernel and detection parameters chosen from
# the actual diffraction pattern, not copied blindly from an example.
probe = datacube.get_vacuum_probe()
bragg_peaks = datacube.find_Bragg_disks(template=probe, corrPower=1.0, sigma=2)

For deeper examples, read:

• Tutorial notebooks: https://github.com/py4dstem/py4DSTEM_tutorials
• First steps: https://py4dstem.readthedocs.io/en/latest/examples/first_steps.html
• Virtual imaging: https://py4dstem.readthedocs.io/en/latest/examples/virtual_imaging.html
• Bragg disk detection: https://py4dstem.readthedocs.io/en/latest/examples/bragg_disk_detection.html

Key conventions and gotchas

• py4DSTEM v0.14 is a major workflow/API boundary. Older pre-0.14 examples can be structurally misleading; check the docs version before copying code.
• Phase-retrieval APIs were reorganized in v0.14.9 with shortened class names. If ptychography examples fail at import time, verify the installed py4DSTEM version and current phase-retrieval class names.
• Datacube axes are domain-significant. Confirm scan axes and diffraction axes before indexing, reshaping, summing, or exporting arrays.
• Virtual detector geometry is in detector/reciprocal-space pixel coordinates unless calibration-specific code says otherwise. Do not use a copied radius or center without checking the beam center and diffraction pattern scale.
• Quantitative Bragg/strain/orientation workflows depend on probe/kernel choice, thresholds, masks, calibration, elliptical distortion correction, and scan distortions. Do not report quantitative strain from raw peaks without these checks.
• py4DSTEM data can be large. Prefer the package's readers, tree objects, and tutorial patterns over loading entire HDF5 datasets into ad hoc arrays.

Anti-patterns

• Do not write generic NumPy code that assumes shape order without printing and naming (R_y, R_x, Q_y, Q_x) or the corresponding py4DSTEM dimensions.
• Do not run Bragg disk detection with tutorial thresholds on a new dataset. Inspect the mean/max diffraction pattern, mask saturated/hot pixels, choose a probe/template, and validate peaks visually.
• Do not average or crop diffraction patterns before recording calibration and beam-center assumptions.
• Do not mix py4DSTEM v0.13 notebooks with v0.14+ code unless deliberately porting the workflow.
• Do not use phase-retrieval examples without checking whether the class names match the installed version.

Diagnostic checks

Before trusting outputs, the agent should:

• Log py4DSTEM version, file path, tree keys, datacube shape, dtype, and whether data are loaded eagerly or lazily.
• Show the mean or max diffraction pattern and at least one scan-position diffraction pattern before quantitative processing.
• Verify scan/detector axis order, beam center, calibration units, detector mask, and virtual detector geometry.
• For Bragg workflows, overlay detected peaks on representative diffraction patterns and record probe/template, thresholds, and masks.
• For strain/orientation outputs, record calibration, reference lattice/peaks, distortion corrections, and uncertainty or residual checks.
• Save intermediate py4DSTEM objects or output files with enough metadata to reproduce detector geometry and calibration choices.

Pointers to deeper material

• Documentation: https://py4dstem.readthedocs.io/
• Tutorial repository: https://github.com/py4dstem/py4DSTEM_tutorials
• Source repository: https://github.com/py4dstem/py4DSTEM
• Paper: Savitzky et al. (2021), "py4DSTEM: A Software Package for Four- Dimensional Scanning Transmission Electron Microscopy Data Analysis", Microscopy and Microanalysis 27, 712-743. https://doi.org/10.1017/S1431927621000477