🔢 mathematics

The Euler Stratification for $\mathbb{P}^1 \times \mathbb{P}^1 \times \mathbb{P}^n$

This paper investigates the Euler stratification of hypersurfaces in $\mathbb{P}^1 \times \mathbb{P}^1 \times \mathbb{P}^n$ linked to three-way independence models in algebraic statistics, demonstrating that while the Euler characteristic is determined by principal $A$ -determinant vanishing patterns only for $n=1$ , all positive integers up to the maximum ML degree are realizable for any $n \geq 1$ , with complete stratification results provided for $n=1$ and partial results for $n=2$ .

Original authors: Serkan Hoşten, Vadym Kurylenko, Elke Neuhaus, Nikolas Rieke

Published 2026-03-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Serkan Hoşten, Vadym Kurylenko, Elke Neuhaus, Nikolas Rieke

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 an architect designing a complex, multi-dimensional sculpture. This sculpture isn't made of stone or steel, but of mathematical relationships between numbers. Specifically, you are building a shape based on a grid of numbers (a tensor) that represents how three different variables interact.

In the world of algebraic statistics, this shape represents a "model" of independence. Think of it like a giant, invisible web connecting three sets of data: two binary switches (like a light being on or off) and one dial with many settings (like a volume knob with $n+1$ positions).

The paper you asked about is a deep dive into understanding the "shape" and "complexity" of this web as you tweak the numbers in your grid. Here is the story of what the authors discovered, broken down into simple concepts.

1. The Core Concept: The "Euler Characteristic" as a Complexity Score

Imagine your mathematical sculpture is a landscape. Sometimes it's a smooth hill; sometimes it's a jagged mountain range with holes, tunnels, and peaks.

The authors are interested in a specific number called the Euler characteristic. Think of this as a "Complexity Score" for your landscape.

A simple, smooth sphere might have a score of 2.
A shape with a hole (like a donut) has a different score.
A shape with many holes and twists has a very different score.

In this paper, this "Complexity Score" is directly linked to something called the Maximum Likelihood Degree (ML Degree). In the real world, statisticians use this to figure out how hard it is to find the "best fit" for their data.

Low ML Degree: The data is easy to understand; there are only a few possible answers.
High ML Degree: The data is tricky; there are many confusing "local peaks" where the answer could hide, making it hard to find the true best answer.

2. The Big Question: Can We Predict the Score?

The authors asked a fundamental question: "If we know which parts of our mathematical grid are 'broken' (vanishing), can we predict the Complexity Score?"

In simpler terms: If you know which specific connections in your web have snapped, can you tell exactly how complicated the remaining web is?

The Old Belief (The Conjecture): For simpler shapes (like a flat sheet), the answer was "Yes." If you knew which specific "factors" (the mathematical equivalent of broken springs) were zero, you knew the score.
The New Discovery: For the 3D shape they studied ( $P1 \times P1 \times Pn$ ), the answer is "Not always."

They found a surprise: You can have two different sculptures where the exact same springs are broken, yet one sculpture is a simple hill and the other is a twisted maze. The "broken springs" don't tell the whole story anymore. Sometimes, you need to look at the rank (the underlying strength) of the connections, not just whether they are broken or not.

3. The "Stratification" Map

Because the "broken springs" rule doesn't always work, the authors had to draw a new map. They called this the Euler Stratification.

Imagine a giant map of all possible ways to build your sculpture. The map is divided into regions (strata).

Region A: No springs are broken. The complexity score is at its maximum (6 for the simplest 3D case).
Region B: One specific spring is broken. The score drops to 5.
Region C: Two specific springs are broken. The score drops to 4.
And so on...

For the simplest 3D case ( $P1 \times P1 \times P1$ ), they mapped out 41 distinct regions. They figured out exactly which combinations of broken springs lead to which complexity scores. It's like having a complete instruction manual that says, "If you break these specific 3 springs, your model will have a complexity of 3."

4. The "Realizability" Discovery: Filling the Gap

The final part of the paper answers a very practical question: "Can we build a model with any complexity score we want?"

The maximum possible score for a model of this size is known (it's like the highest peak on the mountain). The minimum is 1 (a flat, simple line).

The Question: Can we hit every number in between? (e.g., can we get a score of 3? 4? 5?)
The Answer: Yes.

The authors proved that for any integer between 1 and the maximum, there is a specific way to set your numbers (your "scaling tensor") to create a model with exactly that complexity. They provided a recipe to construct these models. It's like saying, "No matter what difficulty level you want for your puzzle, we can build one that is exactly that hard."

Summary of the Analogy

Think of the paper as a guide for a Lego architect:

The Goal: Build a structure where the number of "tricky spots" (complexity) depends on how you arrange the bricks.
The Surprise: You thought that if you removed specific bricks (the "factors"), you'd know exactly how tricky the structure is. But for 3D structures, sometimes removing the same bricks leads to different levels of trickiness depending on how the remaining bricks are stacked.
The Map: The authors drew a complete map of all the ways to remove bricks and what the resulting "trickiness" score will be.
The Power: They proved you can build a structure with any trickiness score you desire, from the simplest to the most complex.

Why Does This Matter?

In the real world, statisticians and data scientists use these models to analyze data (like medical trials or election polls). Knowing the "Complexity Score" (ML Degree) tells them:

How many solutions might exist?
How hard will it be for a computer to find the best answer?
Is the model too complicated to be useful?

By understanding these rules, the authors help scientists build better, more predictable models for understanding the world.

1. Problem Statement

The paper investigates the Maximum Likelihood Degree (ML degree) of three-way independence models in algebraic statistics. Specifically, it studies the hypersurface $Y_{W,n} \subset (\mathbb{C}^*)^{n+2}$ defined by a polynomial $f_W$ whose monomial support corresponds to the lattice points of the polytope $\Delta_1 \times \Delta_1 \times \Delta_n$ (the product of two 1-simplices and one $n$ -simplex).

The central problem is to understand how the Euler characteristic $\chi(Y_{W,n})$ (which equals the signed ML degree of the associated scaled toric variety $X_{W,n}$ ) varies as the coefficients $W$ of the defining polynomial change.

A fundamental question addressed is Conjecture 1.3 (from CHKO24): Does the vanishing pattern of the factors of the principal $A$ -determinant $E_A$ uniquely determine the Euler characteristic $\chi(Y_{W,n})$ ?

This conjecture holds for Segre products $\mathbb{P}^m \times \mathbb{P}^n$ .
The authors aim to determine if it holds for the three-way product $\mathbb{P}^1 \times \mathbb{P}^1 \times \mathbb{P}^n$ .

2. Methodology

The authors employ a combination of algebraic geometry, combinatorics, and computational algebra:

Euler Characteristic via Inclusion-Exclusion: They utilize the relationship between the ML degree and the Euler characteristic of the complement of the hypersurface. Using results from Huh and TW24, they express the ML degree as a signed Euler characteristic of the intersection of quadrics in the torus.
Principal $A$ -Determinant ( $E_A$ ): They analyze the factorization of $E_A$ $E_{A}$ for the polytope $\Delta_1 \times \Delta_1 \times \Delta_n$ $Δ_{1} \times Δ_{1} \times Δ_{n}$ . The factors include:
- 2-minors: Determinants of $2 \times 2$ submatrices of the coefficient tensor $W$ .
- $2 \times 2 \times 2$ Hyperdeterminants: Polynomials of degree 4 associated with $2 \times 2 \times 2$ subtensors.
- $2 \times 2 \times 3$ Hyperdeterminants: Resultants associated with $2 \times 2 \times 3$ subtensors.
Geometric Analysis of Quadrics: The problem is reduced to analyzing the intersection patterns of $n+1$ quadric curves ( $Q_0, \dots, Q_n$ ) in $(\mathbb{C}^*)^2$ . The authors classify the intersections based on the rank of the coefficient matrices and the vanishing of the hyperdeterminants.
Stratification and Counterexamples: They systematically enumerate the possible vanishing patterns of the factors of $E_A$ for $n=1$ and $n=2$ . For $n \ge 2$ , they construct specific counterexamples using tensor ranks to show that vanishing patterns alone are insufficient to determine the ML degree.
Realizability Construction: To prove that all integer ML degrees are achievable, they construct specific scaling tensors $W$ using "alternating hook minors" to control the number of intersection points between quadrics.

3. Key Contributions and Results

A. Refutation of Conjecture 1.3 for $n \ge 2$

The paper proves that the vanishing pattern of the factors of the principal $A$ -determinant does not uniquely determine the Euler characteristic (and thus the ML degree) for $n \ge 2$ .

Theorem 1.6:
- For $n=1$ , the conjecture holds: $\chi(Y_{W,1})$ is determined solely by the vanishing factors of $E_A$ .
- For $n \ge 2$ , the conjecture fails. The Euler characteristic depends not only on which factors vanish but also on the rank of the flattening matrices of the subtensors (specifically the $2 \times 2 \times 3$ subtensors).
Counterexample (Example 3.16): The authors provide two tensors $W$ and $W'$ where the exact same set of factors of $E_A$ vanish, yet $\chi(Y_{W,2}) = -8$ (ML degree 8) and $\chi(Y_{W',2}) = -9$ (ML degree 9). The difference arises because the rank of the $2 \times 2 \times 3$ flattening matrix differs between the two cases, even though the discriminantal factors vanish in both.

B. Complete Euler Stratification for $\mathbb{P}^1 \times \mathbb{P}^1 \times \mathbb{P}^1$

For the case $n=1$ , the authors provide a complete classification of the Euler stratification (Theorem 1.7). The principal $A$ -determinant has 7 factors (6 minors and 1 hyperdeterminant). The 41 strata are categorized as follows:

No factors vanish: $\chi = 6$ (Max ML degree).
1 factor vanishes: $\chi = 5$ (7 strata).
2 factors vanish: $\chi = 4$ (21 strata).
3 factors vanish: $\chi = 3$ (8 strata, corresponding to "corners" of the tensor).
4 factors vanish (including hyperdeterminant): $\chi = 2$ (3 strata, corresponding to "cubic frames").
All 7 factors vanish: $\chi = 1$ .

C. Realizability of ML Degrees

The authors settle the question of whether every integer between 1 and the maximum possible ML degree can be realized.

Theorem 1.8: For any $n \ge 1$ and any integer $r$ such that $1 \le r \le \deg(X_n) = (n+1)(n+2)$ , there exists a scaling tensor $W$ such that $\text{mldeg}(X_{W,n}) = r$ .
Method: They construct $W$ by selecting a subset of "alternating hook minors" to vanish. This controls the number of intersection points of the quadrics in the torus, allowing for a precise tuning of the Euler characteristic.

4. Significance

Algebraic Statistics: The results clarify the limitations of using the principal $A$ -determinant as a complete invariant for the ML degree in three-way independence models. It highlights that for $n \ge 2$ , the geometry of the solution space is more complex than the discriminantal variety suggests, requiring rank conditions on flattenings.
Toric Geometry: The paper provides a rare complete description of the Euler stratification for a non-Segre product toric variety ( $\mathbb{P}^1 \times \mathbb{P}^1 \times \mathbb{P}^1$ ), offering a blueprint for analyzing higher-dimensional products.
Combinatorial Algebra: The construction of the "alternating hook minors" to achieve every possible ML degree demonstrates a deep connection between the combinatorial structure of the coefficient tensor and the topological properties of the resulting hypersurface.

In summary, the paper resolves the realizability of ML degrees for this family of models and demonstrates that while the principal $A$ -determinant governs the stratification for $n=1$ , it is insufficient for $n \ge 2$ without additional rank information.