CPM Workflow Overview

What is conventional ptychography?

Conventional Ptychographic Microscopy (CPM) is a computational imaging technique where a coherent probe beam is scanned across a sample at overlapping positions. At each position, a diffraction pattern (intensity only) is recorded. An iterative algorithm then recovers both the complex-valued object (amplitude and phase) and the illumination probe from these intensity measurements alone.

Data flow

HDF5 File
    |
    v
ExperimentalData   (loads diffraction patterns, geometry, scan positions)
    |
    v
Reconstruction     (holds mutable state: object, probe, positions)
    |
    v
Engine.reconstruct()   (iterative phase retrieval: ePIE, mPIE, mqNewton, ...)
    |
    v
Monitor            (real-time visualization of object, probe, error)

Step-by-step reconstruction workflow

1. Load experimental data

Diffraction patterns and experimental geometry are loaded from an HDF5 file into an ExperimentalData object. The required fields are: ptychogram (diffraction intensities), wavelength, encoder (scan positions), dxd (detector pixel size), and zo (sample-to-detector distance).

See Data Format for the full HDF5 specification.

2. Initialize reconstruction state

A Reconstruction object is created from the experimental data. It computes coordinate grids and allocates arrays for the object and probe:

• Object: complex-valued 2D array representing the sample's transmission function
• Probe: complex-valued 2D array representing the illumination beam

Initialization types:

Type	Description
"circ"	Circular aperture with random noise
"circ_smooth"	Smoothed circular aperture
"ones"	All-ones array with random noise
"rand"	Fully random initialization
"gaussian"	Gaussian distribution
"upsampled"	Upsampled from diffraction patterns

3. Configure parameters

The Params object holds shared settings: propagator type, constraint switches, regularization options, position ordering, and more. See Configuration Reference for all options.

4. Choose and configure an engine

An engine implements a specific reconstruction algorithm. All engines share the same interface:

engine = Engines.mPIE(reconstruction, experimentalData, params, monitor)
engine.numIterations = 50
for loop, posLoop in engine.reconstruct():
    pass

See Engines for the full list and selection guide.

5. Iterate

During engine.reconstruct(), the algorithm loops through scan positions. At each position:

1. Extract the object patch at the current scan position
2. Compute the exit surface wave (ESW): probe × object_patch
3. Propagate the ESW to the detector plane
4. Apply the intensity constraint (replace computed amplitude with measured amplitude)
5. Back-propagate to the sample plane
6. Update the object and probe estimates

The Monitor shows live plots of reconstruction progress.

6. Save results

reconstruction.saveResults("result.hdf5")

Reconstructed output

The reconstructed output is a 6D array of shape (nlambda, nosm, npsm, nslice, No, No):

Dimension	Meaning
nlambda	Number of wavelengths (polychromatic reconstruction)
nosm	Object state mixture (incoherent object modes)
npsm	Probe state mixture (incoherent probe modes)
nslice	Depth slices (multislice for thick samples)
No	Output frame size (pixels)

Key concepts

Term	Description
Object	Complex-valued transmission function of the sample
Probe	Complex-valued illumination beam at the sample plane
Ptychogram	Stack of 2D diffraction intensity patterns, one per scan position
Encoder / Scan positions	Physical (x, y) coordinates where the probe illuminates the sample
Exit surface wave (ESW)	Product of probe and object patch — the field immediately behind the sample
Propagator	Method to compute wave propagation between sample and detector planes
Intensity constraint	Replacing computed amplitude with measured amplitude while retaining the phase

Multi-modal capabilities

PtyLab.py supports advanced reconstruction modes by adjusting dimension parameters on the Reconstruction object:

Parameter	Default	Description
npsm	1	Number of incoherent probe modes (partial coherence)
nosm	1	Number of incoherent object modes
nlambda	1	Number of wavelengths (polychromatic)
nslice	1	Number of object slices (multislice, thick samples)

Note

Polychromatic reconstruction requires spectralDensity in the experimental data. Multislice reconstruction requires setting nslice > 1 and using the e3PIE engine. See Advanced Topics.