Ptychographic Algorithms for Phase Recovery in 4D Scanning Transmission Electron Microscopy

This thesis provides an overview of 4D-STEM ptychography, mathematically explores reconstruction methods like ePIE, WDD, and SSB, and demonstrates their application through simulated MoS₂ data and experimental 4D-STEM recordings.

The Gist

The Big Picture: Seeing the Invisible with a Flashlight

Imagine you are in a dark room trying to figure out what a hidden object looks like. You have a flashlight (the electron beam) and a wall (the detector).

In a standard microscope, you shine the light through the object and look at the shadow on the wall. But here's the problem: shadows only show you the outline (amplitude), not the texture or depth (phase). It's like looking at a shadow puppet; you know the shape, but you can't tell if the puppet is made of wood, plastic, or if it has a smiley face carved into it.

This paper is about a special technique called Ptychography. Instead of just taking one shadow, this method moves the flashlight around in a grid pattern, taking thousands of overlapping pictures. By mathematically comparing how the shadows overlap and interfere with each other, the computer can "solve the puzzle" to reconstruct the hidden texture and depth of the object. This allows scientists to see things much smaller and clearer than ever before.

The Core Concept: The 4D Puzzle

The paper focuses on a specific type of microscope called STEM (Scanning Transmission Electron Microscopy).

• The Old Way: The microscope scans a tiny beam across a sample and records a single number (brightness) for each spot. This creates a 2D image.
• The New Way (4D STEM): Instead of just recording brightness, the microscope records the entire diffraction pattern (a complex starburst of light) for every single spot the beam touches.
  • Analogy: Imagine taking a photo of a room.
    • Standard: You take a photo of the room.
    • 4D STEM: You take a photo of the room, but for every single pixel in that photo, you also record a 3D map of how the light bounced off that specific spot.
  • This creates a massive "4D" dataset (2 dimensions for the scan position + 2 dimensions for the diffraction pattern).

The Problem: The "Phase" Mystery

When electrons pass through a very thin object (like a single layer of atoms), they don't just get blocked; they get delayed. This delay is called phase.

• The Issue: Our detectors are like cameras; they can only see how bright the light is (intensity). They cannot see the delay (phase). It's like trying to hear a song by only looking at the volume meter; you know it's loud, but you can't tell the melody.
• The Solution: Ptychography uses the overlapping data to mathematically calculate the missing "melody" (the phase) so we can see the true structure of the material.

The Tools: How They Solve the Puzzle

The paper discusses different mathematical "recipes" (algorithms) to solve this puzzle.

1. The Iterative Engine (ePIE):

   • Analogy: Imagine trying to guess a secret code. You make a guess, check it against the clues, realize you were wrong, adjust your guess, and try again. You do this thousands of times until the code finally fits perfectly.
   • How it works: The computer starts with a guess of what the object looks like, simulates what the data should look like, compares it to the real data, and tweaks the guess. It repeats this loop until the image is clear.

2. The Direct Method (WDD & SSB):

   • Analogy: Instead of guessing and checking, imagine you have a magic decoder ring that instantly translates the overlapping shadows into the final picture in one step.
   • WDD (Wigner Distribution Deconvolution): This is a fast, direct mathematical trick that separates the "light source" (the probe) from the "object" (the sample) without needing thousands of loops. It's like using a specific filter to instantly remove the glare from a photo.
   • SSB (Single Side-Band): This is a simplified version of WDD. It works best when the object is very thin and transparent (like a ghost). It's a "quick and dirty" method that gives great results for simple materials without needing heavy computing power.

What the Author Actually Did

The paper is a mix of theory and practice. Here is what the author, Amel Shamseldien Ali Alhassan, actually accomplished:

1. The Theory: The author spent time explaining the math behind how electrons interact with matter and how these algorithms work (Sections 1 and 2).
2. The Simulation (MoS2): The author wrote a computer program (in Python) to test the SSB method. They used a fake (simulated) dataset of a material called Molybdenum Disulfide (MoS2).
   • Result: The program successfully turned the raw 4D data into a clear image showing the atoms of the MoS2. This proved the code worked.
3. The Real Data (Gold): The author went to a lab and took real pictures of a Gold specimen using a high-tech microscope.
   • Result: They compared these raw images to images processed by a more advanced team using the "ePIE" method. The paper shows that while the raw images are blurry, the processed images reveal the crystal structure clearly.

The Limitations and Conclusion

The paper concludes with a few honest "fine print" notes:

• It's not magic for everything: This technique works best on very thin samples (2–5 nanometers thick). If the sample is too thick, the electrons bounce around too much (multiple scattering), and the math breaks down.
• Speed: Taking these 4D pictures takes a long time compared to standard photos. The author notes that while we are getting faster, "live" imaging (like watching a movie of atoms moving) is still a future goal, not a current reality.
• The Future: The author suggests that the next logical step is to implement the WDD algorithm on their real-world data to see if it can produce even better results than the SSB method they tested.

In summary: This paper is a guidebook and a proof-of-concept. It explains how to turn a confusing mess of electron diffraction patterns into a crystal-clear 3D map of an atom's structure, and it shows that the author successfully built a tool to do this for simulated materials and real gold samples.

Technical Summary

Technical Summary: Ptychographic Algorithms for Phase Recovery in 4D Scanning Transmission Electron Microscopy

Problem Statement
The paper addresses the fundamental "phase problem" in Scanning Transmission Electron Microscopy (STEM). In conventional STEM imaging modes (such as Bright Field, Annular Dark Field, and Annular Bright Field), detectors integrate intensity over specific angular ranges, discarding the phase information of the electron wave. Since recorded intensity is the square of the wave amplitude, the phase—which carries critical information about the specimen's potential and structure—is lost. This limitation is particularly acute for weak phase objects (e.g., biological specimens or light elements) where phase contrast is the primary source of image formation. Furthermore, conventional methods are often limited by the information limit of the optical setup and are sensitive to noise and aberrations. The paper posits that Momentum-Resolved STEM (4D STEM), which records a full 2D diffraction pattern at every 2D scan position, contains sufficient redundant information to computationally recover both the phase and amplitude of the Object Transfer Function (OTF).

Methodology
The thesis provides a mathematical and algorithmic framework for ptychographic reconstruction, distinguishing between iterative and non-iterative (direct) approaches.

1. Mathematical Framework: The work establishes the interaction between the electron probe and the specimen using the multislice formalism and the object transfer function (OTF), $q(\vec{r})$. It details how the exit wave is formed by the convolution of the illumination and the object, and how the diffraction pattern at the detector is the Fourier transform of this exit wave.
2. Iterative Methods: The paper reviews the Ptychographic Iterative Engine (PIE) and its extensions, such as the extended PIE (ePIE). These algorithms utilize forward and backward propagation loops to iteratively update estimates of the probe function and the object function, enforcing constraints in both real space (support) and reciprocal space (measured intensity).
3. Non-Iterative (Direct) Methods: A significant portion of the methodology focuses on dense sampling techniques, specifically:
   • Wigner Distribution Deconvolution (WDD): This method treats the ptychographic problem as a linear inversion. By taking the Fourier transform of the diffraction intensity with respect to the probe position, the algorithm separates the object and probe contributions into a Wigner distribution. A Wiener filter is applied to deconvolve the probe function, allowing for the direct retrieval of the object's complex transmission function.
   • Single Side-Band (SSB): A simplified version of WDD applicable to weak phase objects. Under the weak phase object approximation, the algorithm isolates the object's phase information by analyzing the overlap of diffraction discs in momentum space, effectively ignoring higher-order scattering terms.

Key Contributions and Results
The paper presents a review of existing implementations and reports on specific reconstruction efforts:

• Theoretical Overview: The work synthesizes the theoretical underpinnings of 4D STEM, detailing the requirements for ptychography (e.g., overlapping illumination areas, dense sampling) and the mathematical derivation of the WDD and SSB algorithms.
• Simulated Reconstruction (MoS2): The author developed a Python script to implement the SSB algorithm. This script was applied to simulated 4D data of a Molybdenum Disulfide (MoS2) monolayer. The results successfully produced amplitude and phase images of the monolayer, demonstrating the algorithm's ability to retrieve structural information from simulated data under aberration-free conditions.
• Experimental Comparison (Gold Specimen): The paper presents Annular Dark Field (ADF) images of a polycrystalline gold specimen acquired during a laboratory session. While the author did not perform the ptychographic reconstruction on this specific dataset within the thesis timeframe, these images are presented for comparison against existing ePIE reconstructions of similar gold grids (referenced from Strauch, 2020). This serves to illustrate the potential of ptychography to retrieve both object and probe functions, including aberration correction, compared to conventional imaging.

Significance and Claims
The paper claims that ptychography offers a robust pathway to super-resolution imaging that exceeds the information limit of the microscope's aperture, limited only by the electron wavelength. Key advantages highlighted include:

• Super-Resolution: Achieving resolutions twice that of the conventional setup (citing early WDD implementations).
• Dose Efficiency: The ability to produce high-resolution phase images with lower electron doses, which is critical for beam-sensitive biological and light-material specimens.
• Robustness: The resilience of WDD and SSB algorithms to corrupt data, missing pixels, and optical aberrations.
• Directness: The computational speed of non-iterative methods (WDD/SSB) compared to iterative schemes, making them suitable for potential real-time applications.

The thesis concludes modestly, noting that while ptychography is a powerful tool for thin specimens (2–5 nm) with low atomic numbers, it faces challenges with multiple scattering in thicker samples. The author notes that the practical implementation of the WDD algorithm on experimental 4D data remains a necessary next step, as the thesis work was limited by time constraints to simulated data and literature review. The work serves as a foundational overview and a partial implementation of SSB reconstruction, setting the stage for further algorithmic development in the field.