🔢 mathematics

Scattering Measures and Information Geometry

This paper formulates scattering experiments using parameterized finite measures to establish an information geometry framework based on the Fisher metric, which decomposes information into scattering strength and shape redistribution, unifies measurement processes as contracting kernels, and connects these concepts to optimal experimental design and quantum measurement theory.

Original authors: Tatsuro Oda

Published 2026-09-29

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Tatsuro Oda

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 a scientist standing before a beam of particles, watching them collide with a target and scatter in every direction. The goal is to learn something about the target—perhaps its internal structure, its temperature, or the forces holding it together. To do this, the scientist must count how many particles arrive at different angles and with different energies. In the traditional view of such experiments, the data is often treated as a collection of probabilities: the chance that a particle will end up in a specific spot. However, this approach often separates two crucial pieces of information. One piece is the total number of particles that scatter, which tells us about the overall strength of the interaction. The other is the pattern of where those particles land, which reveals the shape of the distribution. In standard practice, these two aspects are frequently analyzed as if they were independent, or the total count is treated merely as a scaling factor rather than a source of deep information in its own right.

This separation creates a blind spot when scientists try to design the perfect experiment. They might adjust their instruments to sharpen the image of the scattering pattern, only to lose the ability to measure the total intensity accurately, or vice versa. The core question becomes: how can we view the entire experiment—the physical scattering, the imperfect detectors, and the final data reduction—as a single, unified process? By treating the scattering event not just as a set of probabilities, but as a finite amount of "stuff" distributed across possible outcomes, researchers can map the flow of information from the initial collision to the final recorded data. This perspective allows for a precise calculation of how much we can learn about the target's properties and how different experimental choices might help or hinder that learning.

In a new formulation, a researcher at the University of Tokyo has reimagined the scattering experiment by starting with the concept of a "scattering measure." Instead of focusing immediately on the density of particles or the probability of a specific outcome, this approach treats the scattering event as a finite quantity of strength assigned to all possible outcomes. Think of this quantity as a total amount of scattering power that the experiment generates. This total amount can change, and the way it is spread out among the different outcomes can also change. The researcher developed a mathematical framework that tracks both the total amount and the shape of the distribution simultaneously. This framework reveals that the information carried by the experiment naturally splits into two distinct parts: one part comes from changes in the total strength of the scattering, and the other comes from changes in the normalized shape of the distribution. These two parts are mathematically independent, meaning they provide separate clues about the physical properties being studied.

The paper demonstrates that this unified view is essential for understanding how experimental tools affect the data. Real-world instruments are never perfect. Detectors might miss some particles, or they might blur the precise location where a particle landed. The new framework models these imperfections as transformations that act on the scattering measure. When a detector misses a particle, it is not just losing a count; it is losing a specific piece of information. The research shows that these losses can be quantified precisely. If an instrument blurs the data, it reduces the ability to distinguish between different shapes of the scattering pattern. If it rejects certain events based on their energy, it might distort the total strength information. By treating the entire chain of events—from the physical collision to the final recorded numbers—as a sequence of transformations, the researcher can calculate exactly how much information is lost at each step.

A key finding of this work is that the loss of information is not random; it follows a strict geometric rule. When an experiment transforms the data, the information that remains is simply the part of the original signal that can be predicted from the new, recorded outcome. Any detail that cannot be predicted is lost forever. This insight allows scientists to design experiments that maximize the information they get. For instance, in a triple-axis neutron spectrometer, which is used to study the vibrations of atoms in a crystal, there are many ways to set up the instrument to measure the same physical point. The new framework shows that different settings, even if they target the same point, will preserve different amounts of information depending on how they handle resolution and efficiency. One setting might be better at measuring the size of an energy gap, while another is better at measuring the strength of the interaction, even if they are looking at the same spot.

The paper also connects this classical description of scattering to the quantum world. In quantum mechanics, a measurement produces probabilities for different outcomes. The researcher shows that the finite scattering measure used in this new framework naturally arises when a quantum measurement is incomplete or when only a specific sector of outcomes is observed. This clarifies the hierarchy between the quantum world, where states are described by complex wave functions, and the classical world of experimental data, where we count particles. The framework bridges the gap by showing how the quantum probabilities, when filtered through a specific measurement setup, become the finite measures that drive the classical analysis. This means the tools developed for analyzing classical scattering data are directly applicable to understanding the information extracted from quantum experiments, provided the measurement process is accounted for.

Ultimately, this work provides a new language for experimental design. Instead of guessing which instrument settings are best, scientists can now calculate the "Fisher information," a measure of how distinguishable two different physical models are, for any proposed setup. This calculation takes into account the total count rate, the shape of the distribution, the resolution of the detector, and the efficiency of the system. It allows researchers to optimize their experiments to answer specific questions. If the goal is to measure a specific parameter, such as the temperature of a material, the framework can identify the instrument settings that preserve the most information about that temperature while ignoring irrelevant noise. It reveals that there is no single "best" resolution for an experiment; the optimal setting depends entirely on what parameter is being measured and how the different sources of information interact. By unifying the treatment of total strength and distribution shape, and by rigorously accounting for information loss, this approach offers a powerful tool for extracting the maximum possible knowledge from the complex and often imperfect data of scattering experiments.

Technical Summary: Scattering Measures and Information Geometry

Problem Statement

Conventional descriptions of scattering experiments often treat the physical scattering distribution, instrumental response, counting statistics, and data reduction as separate mathematical entities. Specifically, the distinction between the total scattering strength (mass) and the normalized distribution of outcomes (shape) is frequently obscured, as is the treatment of information loss during measurement transformations. Furthermore, the mathematical hierarchy connecting quantum measurements (via the Born rule) to classical outcome statistics lacks a unified framework that explicitly handles non-normalized finite measures, which are intrinsic to scattering cross-sections.

This work addresses the need for a unified mathematical structure that treats the scattering process as a sequence of transformations on finite measures. It seeks to define an intrinsic information geometry for parameterized scattering measures that is independent of specific density representations (e.g., differential cross-sections) and counting models, while clarifying how experimental operations contract statistical information.

Methodology

The paper formulates scattering experiments using the theory of finite positive measures on outcome spaces, integrating concepts from measure theory, information geometry, and statistical inference.

Scattering Measures: The physical distribution of scattering strength is defined as a finite positive measure $\sigma$ on a measurable outcome space $(X, \mathcal{X})$ . Differential cross-sections are treated not as primary objects but as Radon–Nikodym densities ( $s = d\sigma/d\mu$ ) with respect to a chosen reference measure $\mu$ .
Measure Transformations: Experimental processes are modeled as transformations of these measures:
- Deterministic operations (e.g., coordinate changes, binning) are represented as pushforwards ( $T_\#$ ) of measures.
- Stochastic processes (e.g., detector resolution, channel mixing, inefficiency) are represented by probability or sub-probability kernels ( $K$ ). The induced measure is $\nu = K_\# \sigma$ .
Intrinsic Fisher Geometry: For a parameterized family of measures $\{\sigma_\lambda\}$ , the paper defines an intrinsic score $u_\lambda = d\dot{\sigma}_\lambda / d\sigma_\lambda$ and a Fisher metric directly on the space of finite measures, avoiding dependence on a specific density representation.
Mass–Shape Decomposition: The geometry of finite measures is shown to decompose orthogonally into a "mass" component (changes in total scattering strength) and a "shape" component (changes in the normalized distribution).
Poisson Realization: The intrinsic Fisher information is linked to Poisson point process statistics, showing that the Fisher information of the observed counting process per unit exposure coincides with the intrinsic metric of the intensity measure.
Information Contraction: The paper analyzes how measurement kernels contract Fisher information, deriving that the output score is the conditional expectation of the input score, and information loss corresponds to the unresolved conditional variance.
Quantum Connection: The framework is connected to quantum measurement theory by treating the scattering sector as a sub-probability measure derived from a complete POVM via the Born rule, establishing a hierarchy from quantum states to classical finite measures.

Key Contributions and Results

1. Intrinsic Fisher Geometry of Finite Measures

The paper establishes that the Fisher metric for finite measures admits a Pythagorean decomposition:
$I(\sigma)(\lambda) = \frac{\dot{r}_\lambda^2}{r_\lambda} + r_\lambda I(P)(\lambda)$
where $r_\lambda = \sigma_\lambda(X)$ is the total mass and $P_\lambda$ is the normalized shape.

Geometric Interpretation: The space of finite measures forms a metric cone over the space of normalized probability measures. The radial direction corresponds to mass variations, while tangential directions correspond to shape variations.
Orthogonality: Mass and shape variations are orthogonal in the Fisher geometry. This decomposition is intrinsic and does not rely on a specific counting model.

2. Information Contraction under Kernels

For a parameter-independent probability kernel $K$ , the Fisher information of the transformed measure $\nu = K_\# \sigma$ satisfies $I(\nu) \leq I(\sigma)$ .

Mechanism: The output score is the conditional expectation of the input score given the recorded outcome. The information loss is exactly the expected conditional variance of the input score that remains unresolved by the measurement.
Consequences: This formalizes information loss in discrete channel mixing, finite resolution (Gaussian broadening), and coarse-graining. It distinguishes between the loss of distinctions among retained events (shape contraction) and the loss of events themselves (mass reduction).

3. Experimental Design as Optimization

The framework reframes experimental design as the optimization of the observed measure's Fisher information.

Design Variables: Instrumental settings (e.g., resolution width, acceptance) and data reduction choices are treated as parameters that define the measurement kernel and exposure.
Optimality Criteria: The paper demonstrates how standard design criteria (D-, A-, E-, c-optimality) apply to the Fisher information matrix of the observed measure.
Triple-Axis Spectroscopy Example: The paper illustrates that different instrumental settings mapping to the same nominal kinematic point (a "kinematic fiber") can yield different Fisher information due to variations in resolution kernels. Optimization involves selecting the setting within the fiber that maximizes the efficient Fisher information for the parameter of interest, accounting for nuisance parameters.
Resolution–Throughput Trade-off: In a diffuse-sphere model, the paper shows that the optimal resolution depends on the specific inferential objective (e.g., D-optimality vs. c-optimality) and the throughput law, demonstrating that no single "best resolution" exists independent of the scientific goal.

4. Connection to Quantum Measurement

The paper clarifies the position of the scattering measure within the quantum-to-classical hierarchy.

Born Rule Mapping: A quantum measurement (POVM) maps a parameterized quantum state $\rho_\lambda$ to a classical probability measure. The scattering sector corresponds to a sub-probability measure (finite measure) $\zeta_\lambda = q_\lambda P_\lambda$ , where $q_\lambda$ is the probability of a scattering event.
Hierarchical Decomposition: The Fisher information of the complete quantum measurement outcome decomposes into contributions from scattering occupancy (Bernoulli process), discrete sector labels, and within-sector shape distributions.
Rare-Event Limit: The Poisson point process model used in scattering experiments is recovered as the rare-event limit of repeated independent trials, where the no-scattering contribution vanishes, leaving the finite intensity measure $\Lambda_\lambda$ .

Significance and Claims

The paper claims to provide a unified information-theoretic framework that connects the physical scattering process, instrumental response, and data reduction into a single sequence of measure transformations.

Unification: It unifies the treatment of scattering strength (mass) and distribution (shape), showing they are orthogonal sources of parameter distinguishability. This clarifies why probe-dependent weighting can convert shape sensitivity into mass sensitivity.
Intrinsic Definition: By defining Fisher geometry at the measure level, the formulation avoids artifacts introduced by specific density representations or coordinate choices.
Design Framework: It places experimental design within a common framework where the "best" measurement is defined by the preservation of parameter-sensitive distinctions, quantified by the contraction of Fisher information.
Quantum-Classic Bridge: It explicitly locates the finite scattering measure as the classical outcome of a quantum measurement with incomplete detection, separating quantum limitations from subsequent classical information loss.

The work does not propose new physical scattering laws or specific new experimental setups but rather offers a rigorous mathematical language to analyze existing scattering experiments and optimize their design based on information geometry principles. The significance lies in the structural clarity it brings to the hierarchy of information flow from quantum states to recorded data.