🔬 materials science

Topology of Shape and Data in Material Microstructures

This paper proposes a novel framework that integrates Topological Data Analysis with non-Euclidean shape distances and separable shape tensors to rigorously quantify and visualize the complex topology, shape, and spatial arrangements of material microstructures, offering enhanced tools for electron backscatter diffraction analysis and broader imaging science applications.

Original authors: Jeanie Schreiber, Zachary Grey, Adam Creuziger

Published 2026-07-31

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jeanie Schreiber, Zachary Grey, Adam Creuziger

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery, but instead of looking for fingerprints, you are looking at the microscopic world inside a piece of metal or a chunk of ice. This is the world of materials science, where the strength and behavior of a material depend entirely on the tiny "grains" that make it up. Think of these grains like the individual tiles in a mosaic or the bubbles in a foam. Scientists have long known how to measure the size of these tiles or count how many there are, but they have struggled to describe how the tiles are arranged relative to one another. It's like knowing you have a pile of red and blue marbles, but not knowing if the red ones are clustered in the center or scattered randomly. This arrangement, or "topology," is crucial because it dictates whether a bridge will hold or a jet engine will crack under pressure. To solve this, researchers are turning to a branch of mathematics called Topological Data Analysis (TDA). You can think of TDA as a way of counting the holes in a shape. If you have a donut, it has one hole; if you have a figure-eight, it has two. By counting these holes at different scales, scientists can get a "fingerprint" of the material's internal structure that goes far beyond simple measurements.

The paper "Topology of Shape and Data in Material Microstructures" by Jeanie Schreiber, Zachary Grey, and Adam Creuziger tackles the challenge of quantifying these complex arrangements, specifically looking at "necklacing"—a pattern where tiny grains form a ring around a larger grain, much like a necklace of pearls around a central gem. The authors argue that simply measuring the size of the grains isn't enough; you need to understand how their shapes and their positions work together. To do this, they combine two powerful tools. First, they use a method called Separable Shape Tensors (SST) to mathematically describe the shape of each grain boundary, stripping away its position and rotation to focus purely on its form. Second, they use TDA to look at the spatial connections between these grains. By merging these two ideas, they create a "dual-parameter filtration." Imagine a 3D map where one axis tells you how far a grain's shape is from the "average" shape, and the other axis tells you how far apart the grains are in space. By scanning this map, they can spot patterns that would be invisible to traditional methods.

The researchers tested their new framework on images of ice samples that had been frozen, squished, and allowed to recrystallize under different amounts of pressure. They looked at four specific samples with increasing amounts of strain: 0.03 mm/mm, 0.08 mm/mm, 0.12 mm/mm, and 0.20 mm/mm. Their method revealed a clear story. At low strain (0.03 to 0.08 mm/mm), the ice grains were large and relatively uniform, and the "necklace" patterns were weak or non-existent. However, between 0.08 mm/mm and 0.12 mm/mm, the data showed a sharp, sudden change. The number of persistent "holes" (rings of small grains) increased dramatically, and these rings lasted longer across different scales. This suggests that a critical transition happened right around 0.12 mm/mm, where the material began to reorganize itself into these complex necklace structures. As the strain increased further to 0.20 mm/mm, the pattern stabilized, suggesting the material had reached a new, saturated state of organization.

The authors suggest that this approach offers a much more rigorous way to describe what materials scientists call "texture." Unlike older methods that rely on picking specific numbers (like average grain size) or using "black box" artificial intelligence that is hard to interpret, their method provides a clear, visual, and mathematical map of the microstructure. They explicitly note that while standard methods might miss the difference between a random scattering of grains and a structured necklace, their dual-parameter map highlights these differences instantly. The paper does not claim to have solved all problems in materials science, nor does it suggest this is a magic bullet for predicting every material failure. Instead, it presents a new, principled tool that allows scientists to see and measure the "shape of the arrangement" with unprecedented clarity. By turning the complex visual patterns of ice (and potentially metals and ceramics) into a quantifiable "topological texture," this work offers a new way to understand how materials evolve under stress, potentially helping engineers design stronger, safer materials in the future.

Technical Summary: Topology of Shape and Data in Material Microstructures

Problem Statement
The precise quantification of material microstructures remains a challenge in image analysis, particularly regarding the rigorous characterization of shape, size, and spatial arrangement beyond simple average values. While geometric descriptors (e.g., grain size, volume fraction) and statistical ensembles are well-established, they often fail to capture the global connectivity and topological arrangement of grains. Existing standards, such as ASTM E1181, define topological features like "necklacing" (fine grains surrounding coarse grains) and "banding" qualitatively, relying on visual inspection rather than quantitative metrics. Furthermore, current Topological Data Analysis (TDA) applications typically rely on single-parameter filtrations (usually spatial proximity), which may miss patterns arising from the interplay between spatial arrangement and intrinsic geometric shape. Conversely, advanced shape analysis methods like Separable Shape Tensors (SST) characterize grain geometry rigorously but remain agnostic to the spatial distribution of those shapes. The authors identify a gap in tools capable of simultaneously quantifying the intrinsic geometry of grain boundaries and their spatial topology.

Methodology
The paper proposes a novel framework that merges Topological Data Analysis (TDA) with shape analysis using a dual-parameter filtration approach. The methodology proceeds as follows:

Data Representation: The study utilizes Electron Backscatter Diffraction (EBSD) images where grain boundaries are extracted as closed planar curves. Grain centroids serve as the point cloud for spatial analysis.
Shape Distance (SST): The authors employ Separable Shape Tensors (SSTs) to represent grain boundaries. SSTs map curves onto product manifolds ( $Gr(d, q) \times S^d_{++}$ ), separating the curve into an undulation factor (nonlinear oscillations) and a scale factor (anisotropic size). A product metric is defined to compute the distance of each grain shape from an intrinsic Fréchet mean (the "origin" shape) within this manifold.
Bifiltration Construction: The core innovation is a sublevel Rips bifiltration.
- Parameter 1 (Spatial Scale, $\epsilon$ ): A standard Rips-Vietoris filtration is constructed on grain centroids. As $\epsilon$ increases, edges and simplices form based on Euclidean distance, capturing spatial connectivity.
- Parameter 2 (Shape Distance, $\ell$ ): A sublevel filtration is applied based on the SST distance from the intrinsic mean. As $\ell$ increases, the analysis includes grains that are increasingly dissimilar in shape from the average.
- The Bifiltration: The framework constructs a grid of simplicial complexes where each point $(\epsilon, \ell)$ represents a Rips complex built only on the subset of grains with shape distance $\le \ell$ at spatial scale $\epsilon$ .
Topological Descriptors: The authors track the evolution of the first Betti number ( $\beta_1$ ), which counts the number of independent 1-dimensional loops (holes). In the context of microstructures, a persistent $\beta_1$ corresponds to a ring of grains surrounding a central grain (necklacing).
Scalar Summary: To provide a single metric for comparison, the authors define a scalar summary $I$ , calculated as the integral of the Betti-1 count over the entire bifiltration domain:
$I = \int_{0}^{\ell_{max}} \int_{0}^{\epsilon_{max}} \beta_1(\epsilon, \ell) \, d\epsilon \, d\ell$
This value accumulates the total topological persistence across all shape similarities and spatial scales.

Key Results
The methodology was applied to EBSD scans of ice microstructures (samples PIL254, PIL184, PIL185, PIL255) subjected to increasing axial strain (0.03 mm/mm to 0.20 mm/mm).

Visualizing Topological Texture: The bifiltration contour maps revealed distinct topological signatures. Samples with "necklacing" structures exhibited elevated $\beta_1$ counts forming plateaus or "hotspots" at intermediate spatial scales and specific shape distances, indicating robust ring-like patterns. In contrast, samples with random or uniform distributions showed smoother, less persistent topological features.
Strain Sensitivity: The scalar summary $I$ $I$ demonstrated a monotonic increase with applied axial strain.
- Low Strain (3%–8%): A modest increase in $I$ , corresponding to minimal changes in grain topology.
- Critical Transition (8%–12%): A sharp, abrupt increase in $I$ , coinciding with the onset of dynamic recrystallization and the formation of core-and-mantle (necklace) structures. This aligns with prior observations of grain size collapse and increased grain density in this strain range.
- High Strain (12%–20%): A plateau in $I$ , suggesting a saturation of the topological signature of necklacing, even as recrystallization continues.
Differentiation: The bifiltration successfully distinguished between microstructures (e.g., PIL184 vs. PIL185) that might appear similar under standard statistical summaries but possess distinct spatial arrangements of grain shapes.

Significance and Claims
The paper claims that this framework offers a rigorous, interpretable, and quantitative alternative to qualitative assessments of microstructural texture.

Interpretability: Unlike "black box" machine learning approaches with ambiguous latent spaces, every axis and summary statistic in this construction is directly traceable to physically interpretable quantities (spatial scale and geometric shape distance). A materials scientist can identify precisely which grains and spatial scales drive a signal.
Quantifying ASTM Concepts: The method provides a quantitative basis for concepts like "necklacing" defined in ASTM E1181, moving beyond reliance on example figures and qualitative descriptions.
Decoupling Shape and Space: By using a radial shape distance from an intrinsic mean rather than pairwise shape distances, the approach effectively disentangles the impact of spatial arrangement from the underlying distribution of shape variations.
Generalizability: While demonstrated on ice, the framework relies only on grain-centroid positions and SST-derived shape distances, making it transferable to metals, ceramics, and composites for characterizing duplex grain sizes, banding, and other topological patterns.

The authors position this work not as an optimization for specific property prediction tasks, but as a demonstration of how multi-parameter persistent homology, grounded in shape-distance filtrations, enables the quantitative and interpretable mapping of diverse topological features in scientific imaging.

More like this