How to Learn Ptychography and Coherent Diffraction Imaging: From Overlapping Diffraction Patterns and Phase Retrieval to 3D Nanostructure, Electron Ptychography and Computational Microscopy

## Wait, What? Ptychography Makes an Image Without an Imaging Lens A conventional microscope uses a lens to map points on the sample to points on the detector. Ptychography can do something stranger. A coherent probe illuminates one region of the sample. The detector records only a **diffraction pattern**. Then the probe moves to an overlapping region and records another diffraction pattern. Repeat this many times. None of those detector frames resembles the sample. A reconstruction algorithm uses the overlap constraints to recover complex object transmission, the illumination probe and sometimes position errors and partial coherence. > **ptychography is not a camera with better software; it is a deliberately overdetermined phase-retrieval experiment.** ## The One-Sentence Answer **Learn ptychography by tracing coherent probe → overlapping scan positions → far- or near-field diffraction intensities → iterative phase retrieval → complex object/probe reconstruction, then add scan-position error, partial coherence, dynamical scattering, dose and algorithmic ambiguity before interpreting recovered phase as quantitative structure.** # Beginner Layer — Why Diffraction Loses Phase ## Stage 1: A Coherent Wave Interacts With the Object Represent the object by a complex transmission function **O(r)** encoding amplitude attenuation and phase shift. ## Stage 2: A Probe Illuminates Only Part of the Object At scan position Rj: **ψj(r) = P(r − Rj) O(r)**, where P is the probe. ## Stage 3: The Wave Propagates to a Detector In far-field ptychography, the exit wave is approximately Fourier transformed and the detector records **Ij(q) = |F{ψj}|²**. ## Stage 4: The Detector Loses the Fourier Phase It measures intensity, not complex amplitude. This is the phase problem. # Why Overlap Solves More Than One Diffraction Pattern ## Stage 5: Adjacent Probe Positions Overlap The same piece of object appears in several diffraction patterns. ## Stage 6: Overlap Creates Redundant Constraints The algorithm must find one object/probe pair that explains all measurements simultaneously. ## Stage 7: Redundancy Makes Extended Objects Possible Unlike isolated-object coherent diffraction imaging, the sample need not fit inside one finite support. ## Stage 8: Too Little Overlap Weakens Reconstruction Sparse scans reduce redundancy and can create artifacts or non-uniqueness. # Iterative Reconstruction Layer ## Stage 9: Start With a Guess Guess object O and probe P. ## Stage 10: Predict a Diffraction Pattern For one scan position, calculate the exit wave and propagate it. ## Stage 11: Enforce the Measured Amplitude Replace predicted detector amplitude with **√Imeasured** while retaining the current phase estimate. ## Stage 12: Back-Propagate and Update Object/Probe Algorithms such as ePIE iteratively adjust estimates across all scan positions. ## Stage 13: Convergence Is Not Proof of Uniqueness An algorithm can converge to a local minimum, biased solution or one member of an ambiguity family. # Probe Retrieval Layer ## Stage 14: Ptychography Can Reconstruct the Illumination Probe Overlap redundancy allows simultaneous inference of object and probe. ## Stage 15: Probe Retrieval Can Absorb Sample Errors If the forward model is wrong, object structure can leak into the probe estimate or vice versa. ## Stage 16: Inspect the Retrieved Probe Physically A bizarre probe shape or phase is often a warning that the model is compensating for something else. # Scan-Position Layer ## Stage 17: Reconstruction Assumes Probe Positions Are Known Mechanical stages have drift, scale error, rotation and vibration. ## Stage 18: Position Error Produces Blurring and Ghost Structure The overlap constraint becomes inconsistent. ## Stage 19: Algorithms Can Refine Positions From the Diffraction Data Position correction improves reconstruction; 2026 metrology work demonstrates the value of independent interferometric position measurement. ## Stage 20: Computational Correction Does Not Make Mechanical Stability Irrelevant If motion is too large or fast, information is lost. # Raster-Grid Pathology ## Stage 21: Perfectly Periodic Scan Grids Can Create Ambiguities Blind ptychography has known raster-grid pathologies. ## Stage 22: Slightly Irregular Scan Positions Can Improve Uniqueness Experimental design can improve mathematical identifiability. # Partial-Coherence Layer ## Stage 23: Real Beams Are Not Perfectly Coherent Sources have finite size, bandwidth and temporal fluctuation. ## Stage 24: Motion During Fly Scans Mimics Partial Coherence Several probe positions may be averaged during one exposure. ## Stage 25: Mixed-State Ptychography Models Multiple Incoherent Probe Modes Instead of one probe P, reconstruct several modes. ## Stage 26: More Modes Can Overfit Noise Mode count should be justified by weights, residual improvement and source physics. # Near-Field Versus Far-Field ## Stage 27: Far-Field Ptychography Records Fraunhofer-Like Diffraction The detector is sufficiently far that Fourier propagation is appropriate. ## Stage 28: Near-Field Ptychography Records Fresnel Propagation The detector is closer and the pattern can look more image-like. ## Stage 29: Near-Field Methods Can Cover Large Fields With Fewer Positions Extended probes and lower detector dynamic-range demands can be useful, but the forward model must match the regime. # Resolution Layer ## Stage 30: Resolution Is Not Defined by a Lens Numerical Aperture Alone Recoverable spatial frequency depends on scattering angle, wavelength, signal-to-noise, detector and reconstruction. ## Stage 31: Reconstruction Pixel Size Is Not Resolution A 2 nm pixel grid does not prove 2 nm resolving power. ## Stage 32: Fourier Ring/Shell Correlation Can Estimate Resolution Split-data or independent-reconstruction approaches test reproducibility of spatial frequencies. ## Stage 33: Resolution Is Local and Contrast Dependent Weak-phase objects may lose fine features before strongly scattering regions. # Dose Layer ## Stage 34: More Scattered Photons Improve High-Frequency Information But radiation-sensitive samples can change during acquisition. ## Stage 35: Ptychography Can Be Dose Efficient It uses recorded diffraction information efficiently and can outperform some scanning-imaging workflows for specific dose-limited samples. ## Stage 36: Dose Efficiency Is Not Zero Dose Damage can still be the limiting resolution. Frame-order analysis helps detect specimen evolution. # X-Ray Ptychography ## Stage 37: X-Ray Ptychography Is Naturally Quantitative in Phase For sufficiently thin/simple objects, recovered phase can relate to projected electron density. ## Stage 38: Hard X-Rays Penetrate Thick Samples This supports materials, devices and biological specimens. ## Stage 39: Phase-to-Density Conversion Requires Material/Wavelength Models Quantitative phase is powerful but is not automatically one elemental density. # Spectroscopic Ptychography ## Stage 40: Repeat the Measurement Across Photon Energy Near an absorption edge, complex refractive index changes. ## Stage 41: Each Pixel Can Contain an X-Ray Absorption Spectrum The result becomes **x × y × photon energy** at nanoscale resolution. ## Stage 42: Chemical Mapping Requires Spectral Unmixing The chemical map is an inverse problem layered on top of an inverse reconstruction. # Ptychographic Tomography ## Stage 43: Rotate the Sample and Reconstruct Projections At each angle, ptychography recovers a quantitative projected phase. ## Stage 44: Tomography Reconstructs a 3D Volume Ptychographic X-ray CT has demonstrated nanometre-scale 3D resolution. ## Stage 45: Tomography Adds Alignment and Missing-Wedge Errors Projection alignment, rotation-axis error and incomplete angular coverage can limit the volume. ## Stage 46: Phase Retrieval Error Propagates Into CT The two inverse problems are nested. # Electron Ptychography ## Stage 47: 4D-STEM Naturally Produces Ptychographic Data Scan a focused electron probe across x–y and record a 2D diffraction pattern at every point, yielding **x × y × qx × qy**. ## Stage 48: Electron Ptychography Can Recover Very High Spatial Frequencies Modern direct detectors capture overlapping diffraction disks with high dynamic range. ## Stage 49: Mixed-State Electron Ptychography Improved Low-Dose Atomic Imaging Partial-coherence modelling enables high precision at low dose. ## Stage 50: 20 keV Electron Ptychography Has Reached Sub-Ångström Resolution A 2025 result demonstrated about 0.67 Å resolution using a transmission-mode SEM and multislice ptychography. # Thick-Sample / Multislice Layer ## Stage 51: A Thin Phase-Object Model Fails for Thick Crystals Electrons undergo strong multiple/dynamical scattering. ## Stage 52: Multislice Ptychography Represents the Object in Depth Slices Propagate the wave through successive layers to recover depth-dependent information. ## Stage 53: Multislice Inversion Is More Powerful and Less Constrained Thickness, aberrations and slice structure can trade off. ## Stage 54: 2026 Work Pushes Toward Thick-Object Atomic Metrology Current work reports atomic-scale information in thick specimens and 3D device metrology for gate-all-around structures. # 2026 Atomic-Dynamics Frontier ## Stage 55: Electron Ptychography Is Beginning to Measure Thermal Vibrations A June 2026 *ACS Nano* study recovered temperature-dependent atomic vibration amplitudes during in-situ heating. ## Stage 56: Correlated Atomic Motion Can Be Reconstructed A 10 June 2026 *Nature Communications* study introduced correlated lattice-vibration imaging at atomic scale. ## Stage 57: Imaging and Spectroscopy Are Converging A 10 August 2026 *Physical Review Applied* accepted paper proposes multislice hollow ptychography compatible with simultaneous EELS while retaining most electrons for spectroscopy. # Semiconductor Metrology ## Stage 58: Ptychography Can Image Buried Device Structure A 2026 *Nature Communications* study used multislice electron ptychography to map 3D strain relaxation and roughness in gate-all-around transistors. ## Stage 59: EUV Reflection Ptychography Targets Next-Generation Lithography Metrology NIST and others develop reflective DUV/EUV ptychography for nanoscale semiconductor structures. # AI and Fast Phase Retrieval ## Stage 60: Deep Networks Can Produce Fast Initial Reconstructions Transformer and neural methods can reduce iterative computation. ## Stage 61: Neural Phase Retrieval Learns a Prior It can help with sparse scans and also hallucinate expected features. ## Stage 62: Hybrid AI + Iterative Physics Is More Defensible Use a network for initialization, denoising or scan steering, then require consistency with measured diffraction intensities. ## Stage 63: Raw Diffraction Consistency Is the Final Court A beautiful image that does not forward-project into the measured diffraction data is not a valid ptychographic solution. # Professional Layer ## Stage 64: Separate Three Objects 1. **measured diffraction intensities** 2. **forward propagation model** 3. **reconstructed complex object/probe** The third is not the first. ## Stage 65: Professional Ptychography Is a Redundancy–Coherence–Phase-Retrieval Problem > **Which amplitude, phase, depth or atomic feature remains identifiable after scan-position error, partial coherence, detector limits, radiation damage, propagation-model error and alternative phase-retrieval solutions are all allowed to explain the same diffraction dataset?** # Evidence: What Makes a Ptychographic Claim Strong? Stronger evidence combines independent position metrology, raw diffraction preservation, residual/likelihood maps, split-data resolution tests, multiple reconstructions/seeds, probe-mode inspection, scan perturbation, orthogonal TEM/STXM/XRF/XRD and dose-series checks. # Misconceptions Worth Hunting – Ptychography directly photographs the sample. – The reconstruction algorithm creates phase from nothing. – More overlap always guarantees a unique solution. – A smaller reconstruction pixel means higher resolution. – More probe modes always improve physical accuracy. – Position refinement removes all stage instability. – Phase is automatically electron density. – Ptychographic tomography has only one inverse problem. – Electron ptychography avoids dynamical scattering. – Deep learning phase retrieval cannot hallucinate structure. – A low reconstruction error proves the structure uniquely. # Transfer Check A reconstruction sharpens dramatically after position refinement. Did the sample become more ordered? **No. Scan geometry was corrected.** A mixed-state model improves residuals but requires ten probe modes of similar weight. Is that automatically realistic? **No. It may be overfitting or hiding other model errors.** A neural network predicts a feature absent when the reconstructed object is forward-projected into the measured diffraction frames. Is the feature observed? **No.** A thin-object reconstruction fails for a 60 nm crystal but multislice succeeds. Did the detector change? **No. The forward model became physically appropriate to multiple scattering.** # How We Know the Learning Has Held A learner should be able to explain phase loss, overlapping scans, object/probe reconstruction, ePIE-style constraint cycling, position error, raster ambiguity, partial coherence, near/far-field propagation, resolution/dose, spectroscopic/tomographic ptychography, 4D-STEM electron ptychography, multislice thick-sample modelling and AI validation by diffraction consistency. # Model Limits Ptychography is computationally powerful because the experiment is redundant. It can fail when coherence is poorly modelled, positions are unstable, detector saturates, dose changes the sample or the forward model is wrong. Professional ptychography keeps **raw diffraction + scan positions + probe modes + propagation model + reconstruction + resolution test + dose + orthogonal image/structure** visible together. # Teaching Guide Teach in this order: **coherent wave → diffraction intensity → phase problem → local probe → overlapping scan → iterative constraints → probe retrieval → position error → raster pathology → partial coherence → near/far field → resolution/dose → X-ray spectro/tomo → 4D-STEM electron ptychography → multislice → semiconductor/atomic dynamics → AI → validation.** > “If every detector frame is only a diffraction pattern and contains no recognizable object image, where does the final real-space picture come from?” # Connect This to the eduKate Learning Estate – https://edukatesengkang.com/2026/08/29/how-to-learn-x-ray-diffraction-crystallography/ – https://edukatesengkang.com/2026/08/28/how-to-learn-microscopy-scientific-imaging-super-resolution-image-evidence/ # Research Foundations and Further Learning – *Ptychography at all wavelengths* — Nature Reviews Methods Primers, 30 October 2025. – Foundational ePIE/difference-map and blind-ptychography uniqueness literature. – NIST DUV/EUV nanoscopy and reflective ptychography metrology programme. – *Sub-ångström resolution ptychography in a scanning electron microscope at 20 keV* — 2025. – *Quantitative Atomic Resolution Electron Ptychography of Thermal Vibrations Under In Situ Heating* — ACS Nano, 2 June 2026. – *Electron ptychography reveals correlated lattice vibrations at atomic resolution* — Nature Communications, 10 June 2026. – *3D atomic-scale metrology of strain relaxation and roughness in Gate-All-Around transistors via electron ptychography* — 2026. – *Multislice hollow ptychography for simultaneous atomic-layer-resolved 3D imaging and spectroscopy* — Physical Review Applied, accepted 10 August 2026. – HEPS ptychography interferometric-position metrology — online 16 July 2026. # The Quiet Ending The beginner asks: “How did diffraction become an image?” The developing computational microscopist asks: “Which overlap constraints recovered the missing phase?” The advanced learner asks: “Could scan error or partial coherence create the same structure?” > **Which reconstructed feature survives when the forward diffraction model—not image plausibility—is treated as the final test?**