🔬 optics

Multiparameter Maximum Information States for Coherent Diffraction Measurements

This paper extends the concept of maximum information states from single-parameter to multiparameter estimation in coherent diffraction measurements by optimizing scalar functions of the Fisher information matrix and developing strategies to mitigate the impact of nuisance parameters, with findings validated through numerical simulations of 2D coupled dipoles.

Original authors: Bram Verreussel, Jacob Seifert, Allard Pieter Mosk

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Bram Verreussel, Jacob Seifert, Allard Pieter Mosk

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

The Big Picture: The "Smart Flashlight"

Imagine you are trying to take a picture of a very complex, foggy object in the dark. You have a flashlight (the light source), but the fog scatters the light everywhere, making it hard to see details.

In the world of physics, scientists want to measure tiny things—like the exact position, height, or tilt of a microscopic object. The problem is that light is noisy (like static on a radio). To get a clear answer, you usually need to shine a lot of light or wait a long time. But sometimes, you can't shine too much light because it might damage the object, or you don't have enough time.

This paper is about building a "Smart Flashlight." Instead of shining a standard, boring beam of light, the authors show how to shape the light waves perfectly so that every single photon carries the maximum amount of useful information about the object.

The Old Way vs. The New Way

The Old Way (Single Parameter):
Previously, scientists figured out how to shape light to measure one thing perfectly.

Analogy: Imagine you want to know exactly how far a car is from a wall. You can tune your flashlight to bounce off the car in a way that tells you the distance perfectly. But if you try to measure the car's speed at the same time, your "perfect distance" light might be terrible at measuring speed.

The New Way (Multiparameter):
This paper asks: "What if we need to measure three things at once? (e.g., distance, height, and tilt)."

The Problem: If you tune the light for distance, you might lose information about height. If you tune it for tilt, you lose the others. It's like trying to tune a radio to three different stations at once; usually, you just get static.
The Solution: The authors developed a mathematical recipe to find a "Super State" of light. This is a specific pattern of light waves that balances the information. It doesn't give you the absolute best result for just one thing, but it gives you the best combined result for all three things simultaneously.

The "Scattering Matrix" (The Object's Fingerprint)

To create this Smart Flashlight, the scientists need to know how the object messes with the light. They use something called a Scattering Matrix.

Analogy: Think of the object as a complex maze. The Scattering Matrix is a complete map of the maze. It tells you: "If I send a beam of light in Door A, it will come out of Window B with a specific twist."
The paper's breakthrough is using this map to calculate exactly how to shape the light entering the maze so that the light coming out tells us everything we need to know about the maze's layout.

The "Nuisance" Problem

Sometimes, you want to measure two things (like the car's distance and speed), but there is a third thing you don't care about (like the color of the car) that still messes up your measurement. In science, this is called a Nuisance Parameter.

The Dilemma: If you ignore the nuisance parameter, your measurement of the important things might get worse because the "noise" from the nuisance parameter interferes with the signal.
The Paper's Strategy: The authors tested different ways to handle this. They found that the best strategy is to acknowledge the nuisance parameter exists and mathematically "cancel out" its interference, rather than trying to measure it perfectly or ignoring it completely. It's like wearing noise-canceling headphones: you don't need to know the exact pitch of the background noise, you just need to cancel it out so you can hear the music (your data) clearly.

The Results: What Did They Find?

The authors tested their theory using a computer simulation of a grid of tiny mirrors (dipoles).

Better Together: They found that the "Smart Flashlight" designed for multiple parameters was much better at measuring everything at once than a flashlight designed for just one parameter.
The Trade-off: They discovered that trying to make everything perfect at once is hard. Sometimes, you have to accept that one measurement isn't quite as perfect as it could be on its own, just so the others can be good.
The "Pareto Frontier": They drew a graph showing the best possible balance. It's like a curve on a map: you can't get better at measuring distance without getting slightly worse at measuring speed, unless you use their specific "Smart Flashlight" method.

Why Does This Matter?

The paper doesn't claim to cure diseases or build new phones yet. Instead, it provides a mathematical toolkit.

It tells scientists how to calculate the perfect light pattern for any complex object they are studying.
It proves that by shaping light intelligently, you can get more precise measurements without needing more light or more time.
It solves the tricky problem of measuring several things at once without them interfering with each other.

In short: The authors figured out how to write the "perfect instructions" for a light beam so that when it hits a messy, complex object, the light bounces back carrying the clearest possible story about what that object looks like.

Technical Summary: Multiparameter Maximum Information States for Coherent Diffraction Measurements

Problem Statement
In optical metrology, the precision of parameter estimation is fundamentally limited by photon noise, quantified by the Fisher Information (FI). While previous work established "maximum information states" for single-parameter estimation by maximizing the FI using the system's scattering matrix, practical applications often require the simultaneous estimation of multiple parameters. In the multiparameter regime, the FI becomes a matrix (the Fisher matrix), where off-diagonal elements represent "cross-information" that can degrade the precision of individual estimates. Furthermore, experimental scenarios often involve "nuisance parameters"—variables that must be accounted for but are not of direct interest—which can further degrade the precision of the parameters of interest. Existing approaches for multiparameter optimization using the scattering matrix were lacking, particularly regarding how to handle nuisance parameters and how to balance precision across multiple parameters.

Methodology
The authors extend the single-parameter framework to multiparameter estimation using a 2D system of coupled dipoles as a numerical testbed. The core methodology involves:

Fisher Tensor Formulation: The authors derive an expression for the multiparameter Fisher matrix $J_{\mu\nu}$ based on the scattering matrix $S$ of the system. For an input state $|E_{in}\rangle$ , the Fisher matrix is given by $J_{\mu\nu} = \langle E_{in} | (\partial_\mu S^\dagger \partial_\nu S + \partial_\nu S^\dagger \partial_\mu S) | E_{in} \rangle$ . This formulation allows for model-free optimization, as it relies solely on the measured scattering matrix without prior knowledge of the system's internal structure.
Optimality Criteria: To optimize the Fisher matrix for multiple parameters, the authors evaluate several scalar functions (optimality criteria) commonly used in experimental design:
- D-optimality: Maximizing the determinant of the Fisher matrix.
- A-optimality: Minimizing the trace of the inverse Fisher matrix (average joint Cramér-Rao Lower Bound).
- E-optimality: Maximizing the smallest eigenvalue of the Fisher matrix.
- Normalized Trace: A novel criterion proposed by the authors, defined as the trace of the Fisher matrix normalized by the maximum eigenvalue of each parameter's block. This allows optimization via eigendecomposition rather than gradient descent, offering a faster, deterministic solution.
Nuisance Parameter Strategies: The paper investigates three strategies for handling nuisance parameters ( $\xi$ $ξ$ ) when estimating parameters of interest ( $\vartheta$ $ϑ$ ):
- Full Fisher Matrix: Optimizing the entire matrix including nuisance parameters.
- Subblock Fisher Information: Ignoring nuisance parameters entirely by optimizing only the submatrix $J_{\vartheta\vartheta}$ .
- Partial Fisher Information: Using the Schur complement $J_{\vartheta\vartheta \cdot \xi} = J_{\vartheta\vartheta} - J_{\vartheta\xi}J_{\xi\xi}^{-1}J_{\xi\vartheta}$ , which effectively accounts for the correlation between nuisance and interest parameters while removing the direct optimization of the nuisance parameters.

Key Results
Numerical simulations on a system of coupled dipoles (estimating horizontal translation $x$ , vertical translation $z$ , and rotation $\phi$ ) yielded the following findings:

Multiparameter vs. Scalar Optimization: Multiparameter optimization goals (D-, A-, E-optimality) generally outperformed scalar optimization goals when evaluated using the joint Cramér-Rao Lower Bound (CRLB). While scalar optimization is optimal for a single parameter in isolation, multiparameter strategies can improve the joint CRLB by reducing cross-information correlations or increasing information on other parameters.
Optimality Criteria Performance: D- and A-optimality effectively minimized the joint CRLBs. E-optimality tended to equalize CRLBs across parameters but sometimes sacrificed overall optimality. The proposed Normalized Trace criterion provided a fast, eigenvalue-based optimization method that yielded balanced results, though the resulting states were not strictly Pareto efficient.
Nuisance Parameter Handling:
- Partial Fisher Information proved highly effective. When used with D- or A-optimality, the resulting states consistently landed on the Pareto frontier (the set of optimal trade-offs between parameters), indicating that this strategy successfully directs resources toward parameters of interest without wasting effort on nuisance parameters.
- Subblock Fisher Information (ignoring nuisance parameters) performed similarly to or better than the full matrix approach in many cases. However, the authors note that when cross-correlations are strong, ignoring nuisance parameters can be suboptimal because the optimization fails to utilize the Fisher information on the nuisance parameter to reduce cross-correlations.
Bias Analysis: The study examined the bias of estimators as a function of nuisance parameter deviation. The bias was found to be well-approximated by cubic functions of the nuisance parameter, remaining small within the tested range.

Significance and Claims
The paper claims to provide a general framework for multiparameter maximum information states that relies on the scattering matrix, enabling model-free optimization for complex scattering environments. The primary contributions are:

Generalization: Extending the concept of maximum information states from single to multiple parameters, explicitly addressing the role of the Fisher matrix's off-diagonal elements.
Nuisance Parameter Framework: Systematically comparing strategies for dealing with nuisance parameters, demonstrating that Partial Fisher Information combined with standard optimality criteria (like D- or A-optimality) is a robust method for achieving Pareto-efficient estimation of parameters of interest.
Computational Efficiency: Introducing the Normalized Trace criterion, which allows for the rapid calculation of maximum information states via eigendecomposition, avoiding the computational cost of gradient descent required by other criteria.

The authors emphasize that while their approach requires measuring the scattering matrix and its derivatives (which can be time-consuming), the rank of the derivative matrix is often lower than the full scattering matrix, offering potential avenues for efficiency. They conclude that while time-sharing (sequentially illuminating with different optimized states) is a potential future direction, their single-illumination multiparameter approach offers a strong baseline for precise estimation in photon-noise-limited regimes.