📊 statistics

Inverting Poisson-Laguerre tessellations

This paper addresses the challenge of inverting Poisson-Laguerre tessellations by characterizing generator configurations that yield identical tessellations and proposing a consistent method to retrieve the original weighted generators from observed cells, thereby enabling statistical inference and nonparametric estimation of the weight distribution.

Original authors: Thomas van der Jagt, Geurt Jongbloed, Martina Vittorietti

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Thomas van der Jagt, Geurt Jongbloed, Martina Vittorietti

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 looking at a patchwork quilt made of different colored shapes. In the world of mathematics, this is called a tessellation. Specifically, this paper deals with a special kind of quilt called a Laguerre tessellation.

Here is the simple breakdown of what the authors did, using everyday analogies:

1. The Setup: The "Growing Balloon" Quilt

To understand this quilt, imagine a room filled with invisible balloons.

The Generators: Each balloon has a specific center point (where it is tied down) and a specific "weight" (which determines how fast it starts growing).
The Growth: At time zero, all balloons start inflating. However, the heavier balloons start inflating later than the lighter ones.
The Result: As they grow, they bump into each other. The space where a balloon stops growing because it hit a neighbor becomes a "cell." The final pattern of these cells is the tessellation.

Usually, if you know where the balloons are and how heavy they are, you can easily calculate what the final quilt looks like. This is the "forward" problem, and mathematicians have known how to solve it for a long time.

2. The Problem: The "Reverse Puzzle"

This paper asks the reverse question: If you are handed the finished quilt (the cells), can you figure out exactly where the original balloons were and how heavy they were?

The authors discovered a tricky problem: The quilt is "overparameterized."
Think of it like a magic trick. You can change the size of the room, move the balloons, or adjust their weights in very specific, complex ways, and the final quilt pattern looks exactly the same.

The Analogy: Imagine you see a shadow on a wall. You can't tell if the object casting the shadow is a small toy close to the light or a giant statue far away. Similarly, many different sets of balloons can cast the exact same "shadow" (tessellation). You cannot uniquely reverse-engineer the original setup just by looking at the cells.

3. The Solution: Finding the "Most Likely" Balloons

Since you can't find the one true answer, the authors asked: Can we find the "best guess" that is statistically most likely to be the original setup?

They focused on a specific type of random quilt called a Poisson-Laguerre tessellation. This is like a quilt where the balloons are scattered randomly across a huge field, rather than being placed in a specific pattern.

Their Method: The "Center of Gravity" Trick
The authors proposed a clever way to estimate the original balloon positions:

Look at the Cells: For every cell in the observed quilt, find its center of mass (its geometric middle).
The Intuition: In a random tessellation, the original balloon (generator) is usually very close to the center of its own cell.
The Math: They created a mathematical formula (a "least-squares" calculation) that tries to find a set of balloon positions that minimizes the distance between the balloons and the centers of their cells.
The Result: By solving this equation, they can "invert" the tessellation. They get a set of estimated balloon positions and weights that, when used to grow new balloons, would recreate a quilt almost identical to the one they started with.

4. The Catch: The "Window" Effect

In the real world, we can't see the entire infinite field; we only see a square window (like looking through a picture frame).

The Edge Problem: The balloons right at the edge of the window are cut off. We can't see their full cells, so we can't calculate their true centers perfectly. This introduces some "noise" or error into the calculation.
The Fix: The authors showed that as you make your observation window bigger and bigger (looking at a larger and larger piece of the quilt), your guess for the original balloons gets closer and closer to the truth. They proved mathematically that with a big enough window, the method works consistently.

5. The Simulation: Does it Work?

The authors ran computer simulations to test their idea.

They created random quilts with known balloon positions.
They "hid" the balloon positions and only showed the cells.
They ran their inversion algorithm to guess the positions back.
The Outcome: The guesses were incredibly accurate. The estimated balloons overlapped almost perfectly with the real ones.

They also used these estimated balloons to try and guess the "weight distribution" (how heavy the balloons usually are). While the estimates were good, they were slightly less accurate than if they had known the true balloon positions from the start. This makes sense: if you have to guess the starting point, your final calculation will have a tiny bit more error.

Summary

The paper solves a "reverse engineering" puzzle for a specific type of random geometric pattern.

The Issue: You can't uniquely figure out the cause (balloons) from the effect (cells) because many different causes create the same effect.
The Fix: They developed a statistical method to find the most likely cause by looking at the centers of the cells.
The Proof: They proved mathematically that this method works better the larger the area you observe, and simulations confirmed it produces very accurate results.

This is useful for scientists who see a microscopic image of a material (like a metal or rock) and want to understand the underlying random process that created its structure, even if they don't know the original "seeds" that started the growth.

Technical Summary: Inverting Poisson-Laguerre Tessellations

Problem Statement
While algorithms exist to compute the cells of a Laguerre tessellation given a set of weighted generator points, the inverse problem—retrieving the weighted generators from an observed tessellation—is not straightforward. Unlike Voronoi tessellations, Laguerre tessellations are "overparameterized," meaning multiple distinct configurations of weighted points can yield the exact same tessellation. This non-uniqueness poses a significant challenge for statistical inference, where the standard approach assumes the underlying point process of weighted generators is known to estimate model parameters. This paper addresses the inversion of Poisson-Laguerre tessellations, aiming to provide a consistent method for approximating the original weighted generators from observed cells, thereby enabling statistical inference without prior knowledge of the generators.

Methodology
The authors approach the problem in two stages: characterizing the non-uniqueness in general Laguerre tessellations and proposing a consistent inversion procedure for the Poisson-Laguerre case.

Characterization of Overparameterization:
The paper first establishes a rigorous characterization of all configurations of weighted points that generate the same Laguerre tessellation. Building on Lemma 1 from [17], the authors prove (Theorem 2) that if two sets of weighted points $\phi$ and $\psi$ generate the same tessellation (under regularity and general position conditions), then $\psi$ must be a specific transformation of $\phi$ . Specifically, the generators $(x'_i, h'_i)$ in $\psi$ are related to $(x_i, h_i)$ in $\phi$ by:
$x'_i = \lambda x_i + c$
$h'_i = \lambda h_i + \lambda(\lambda - 1)\|x_i\|^2 + 2\lambda\langle x_i, c\rangle + z$
where $\lambda > 0$ , $c \in \mathbb{R}^d$ , and $z \in \mathbb{R}$ are constants. This result implies that while exact reconstruction is impossible, the true generators can be recovered up to this specific transformation group.
Inversion via Weighted Least Squares:
For Poisson-Laguerre tessellations, where generators are a realization of a Poisson process $\eta$ on $\mathbb{R}^d \times (0, \infty)$ , the authors propose an inversion procedure to estimate the unknown transformation parameters $(\lambda_0, c_0)$ that map the observed (transformed) generators back to the original configuration.
- Criterion Function: They define a criterion function $T_n(\lambda, c)$ based on the sum of squared distances between the transformed generators and the centroids of their corresponding Laguerre cells, weighted by cell volumes.
- Optimization: The estimators $(\hat{\lambda}_n, \hat{c}_n)$ are obtained by minimizing $T_n$ . The authors derive explicit formulas for these estimators, showing they correspond to a weighted least-squares solution.
- Consistency: Theoretical analysis (Theorems 3 and 4) demonstrates that as the observation window $W_n$ expands unboundedly, the estimators converge to the true parameters $(\lambda_0, c_0)$ . This consistency relies on the ergodic properties of the Poisson process and the stabilization of cell volumes.
Statistical Application:
The paper investigates whether the inverted (approximated) generators can be used to non-parametrically estimate the weight distribution function $F$ of the underlying Poisson process. They adapt the estimator $\hat{F}^0_n$ from [28] to use the approximated generators $\hat{\eta}^*_n$ instead of the true generators.

Key Results

Theoretical Characterization: The paper provides the first explicit characterization of the class of all weighted point configurations yielding the same Laguerre tessellation (Theorem 2).
Consistent Inversion: Under conditions where the weight distribution $F$ is concentrated on a bounded interval (or via stabilization techniques), the proposed inversion procedure is shown to be consistent. The estimators $(\hat{\lambda}_n, \hat{c}_n)$ converge in probability (and almost surely under stronger conditions) to the true transformation parameters as the observation window grows.
Simulation Performance:
- In simulations with $d=2$ , the inversion procedure yields highly accurate approximations of the original generator configuration. The mean absolute errors for $\lambda$ and $c$ decrease as the number of observed points increases.
- The method performs better when the underlying distribution $F$ results in more uniform cell sizes (e.g., $F_1$ ) compared to distributions with high variance in cell sizes (e.g., $F_2$ ).
- When using the inverted generators to estimate the distribution function $F$ , the resulting estimates are accurate but slightly less precise than those obtained using the true generators. The estimates tend to be smoothed versions of the true distribution, with reduced accuracy in the tails (particularly for large weights).

Significance and Claims
The authors claim that their work provides a necessary bridge for statistical inference in Poisson-Laguerre tessellations. Since the true generators are often unknown in practical applications (e.g., when fitting tessellations to image data), the ability to consistently invert the tessellation allows researchers to proceed with parameter estimation.

The paper modestly notes that while the inversion procedure yields a "very close approximation" of the original generators, the subsequent estimation of the weight distribution function $F$ is less accurate when based on these approximations compared to using the true generators. The authors explicitly state that whether the estimators for $F$ based on inverted generators are generally consistent remains an "important open problem." Furthermore, they acknowledge that their current procedure does not account for edge effects in the observation window, though they provide a practical variant in simulations that excludes boundary cells to mitigate this issue.

In summary, the paper establishes that while Laguerre tessellations are inherently overparameterized, the specific structure of Poisson-Laguerre tessellations allows for a consistent recovery of the underlying generator configuration up to a known transformation, enabling statistical inference in scenarios where the generators are not directly observable.