🔢 mathematics

Osculating Geometry and Higher-Order Distance Loci

This paper investigates higher-order distance loci by defining the higher-order distance degree as an intersection-theoretic invariant involving jet bundles and polar loci, deriving closed formulas for various embeddings, analyzing the impact of the Bombieri-Weyl metric, and introducing a tropical framework for combinatorial computation in toric settings.

Original authors: Sandra Di Rocco, Kemal Rose, Luca Sodomaco

Published 2026-03-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sandra Di Rocco, Kemal Rose, Luca Sodomaco

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 standing in a vast, dark room filled with invisible, complex shapes made of light. You have a flashlight (a data point), and your goal is to find the spot on one of these shapes that is closest to your light. This is a classic problem in math and data science called distance optimization.

Usually, mathematicians ask: "Where is the closest point?" They look at the tangent line—the line that just barely touches the shape at that point. If your light is perpendicular to that tangent line, you've found a "critical point" (a candidate for the closest spot). The number of these candidates is called the Euclidean Distance Degree (EDD). It tells you how complicated the problem is.

But this paper, Osculating Geometry and Higher-Order Distance Loci, asks a much deeper question: "What if the shape isn't just touching the light, but hugging it?"

Here is the breakdown of their discovery, using simple analogies:

1. From a Handshake to a Hug (Osculating Spaces)

In the old way (first-order geometry), we only care about the tangent. Imagine shaking hands with a curve; the tangent is the direction your hand is pointing.

The authors introduce Osculating Spaces. Think of this as a "hug."

Order 1 (Tangent): A handshake.
Order 2 (Osculating Plane): A hug that follows the curve's bend.
Order 3+: A full-body embrace that matches the curve's twist and torsion.

The paper studies what happens when we require our "closest point" to not just be perpendicular to a tangent, but to be perpendicular to this higher-order "hug." This is crucial for things like Computer Vision (where a camera needs to understand how a car's surface curves) or Manifold Learning (where AI tries to understand the shape of data).

2. The "Data Locus" (The Map of Critical Points)

Imagine you have a specific shape (like a twisted ribbon). You want to know: "From which positions in the room can I shine my light so that it hits the ribbon at a point where the ribbon is 'hugging' the light perfectly?"

The collection of all those special positions in the room is called the Distance Locus.

For simple tangents (Order 1), this locus is usually the entire room.
For higher-order hugs (Order 2, 3, etc.), the locus becomes a smaller, more specific shape (like a thin sheet or a wire) floating in the room.

The authors calculate the Higher-Order Distance Degree, which is essentially a measure of the "size" or "complexity" of this specific sheet. It tells you how many different ways the geometry can align perfectly.

3. The Metric Matters (The Shape of the Room)

One of the paper's big surprises is that the answer depends entirely on the metric (the rule for measuring distance).

Standard Ruler (Euclidean): If you measure distance with a standard ruler, you get one set of answers.
Bombieri-Weyl Metric: This is a special, weighted ruler often used in polynomial math.

The authors show that switching from a standard ruler to this special weighted ruler changes the geometry completely.

Analogy: Imagine trying to fit a square peg into a round hole. With a standard ruler, the peg fits perfectly. With the weighted ruler, the peg suddenly looks like a triangle, and it fits differently.
In their example, using the special metric made the "hug" happen in fewer places (lower degree) but made the relationship between the light and the shape more complex (the map wasn't "one-to-one" anymore).

4. The Tropical Toolbox (Solving with Sand)

Calculating these complex shapes is incredibly hard, like trying to solve a 4D puzzle. The authors introduce a Tropical Geometry approach.

The Metaphor: Imagine turning your complex, curvy algebraic shapes into a pile of sand or a Lego structure made of flat blocks. In "Tropical" math, curves become straight lines, and multiplication becomes addition.
The Result: They showed that you can compute these complex "Distance Degrees" by looking at how these flat blocks intersect. It's like solving a difficult geometry problem by just counting how many times two stacks of blocks cross each other. They even built a software tool (in the Julia programming language) that does this counting automatically for specific types of shapes called Toric varieties.

5. Why Does This Matter? (The Real World)

Why should a non-mathematician care?

Better AI: When AI tries to learn the shape of data (like recognizing a face or a car), it often assumes the data lies on a smooth surface. Understanding "higher-order" contact helps the AI understand curvature and twists, leading to more accurate models.
Computer Vision: When a camera tries to reconstruct a 3D scene from 2D photos, it solves a distance problem. If the math accounts for higher-order curves, the reconstruction is more stable and less prone to errors.
Faster Algorithms: The paper connects these geometric concepts to the Gauss-Newton method, a popular algorithm for optimization. They found that when a solution satisfies these "higher-order" conditions, the algorithm becomes incredibly fast and stable. It's like finding a shortcut that guarantees your GPS won't get stuck in traffic.

Summary

This paper is about upgrading our mathematical "ruler." Instead of just asking "How close is it?" (First-order), they ask "How perfectly does it fit?" (Higher-order). They discovered that the answer changes depending on the ruler you use, and they invented a new "sandcastle" method (Tropical Geometry) to calculate these answers efficiently. This helps us build better AI, sharper cameras, and faster optimization tools.

1. Problem Statement and Motivation

Core Problem:
The paper addresses the problem of optimizing the distance function from a given data point $u$ to an algebraic variety $X$ (or a morphism $f: X \to \mathbb{P}^n$ ). While classical Metric Algebraic Geometry focuses on the Euclidean Distance Degree (EDD)—counting the number of critical points where the distance vector is orthogonal to the tangent space (first-order contact)—this work generalizes the problem to higher-order contact.

Motivation:
In many applications (manifold learning, computer vision, signal processing, trajectory planning), minimizing Euclidean distance alone is insufficient. Accurate modeling often requires the data point to be critical with respect to higher-order geometric features, such as curvature, torsion, or higher-order derivatives. This necessitates studying points where the distance vector is orthogonal not just to the tangent space, but to the $k$ -th order osculating space ( $T^k_p f$ ).

Key Questions:

How do we define and compute the Higher-Order Distance Degree (HODD), denoted $DD_k(f, Q)$ , which counts critical points of order $k$ ?
What is the geometry of the Higher-Order Distance Locus ( $DL_k$ ), the set of data points admitting such critical points?
How do these invariants depend on the choice of metric (quadric $Q$ ) and the specific geometry of the variety (e.g., Veronese, Toric, Affine)?

2. Methodology and Framework

The authors develop a comprehensive algebro-geometric framework combining intersection theory, jet bundles, and tropical geometry.

A. Geometric Setup

Osculating Spaces: For a morphism $f: X \to \mathbb{P}^n$ , the $k$ -th osculating space $T^k_p(f)$ at a point $p$ is the projective span of partial derivatives of $f$ up to order $k$ . It is the image of the $k$ -th jet map $j_{k,p}$ .
Higher-Order Normality: Given a non-singular quadric $Q$ (defining the metric), the $k$ -th order normal space $N^k_p(f, Q)$ is defined as the projective span of the polar subspace of $T^k_p(f)$ with respect to $Q$ and the point $f(p)$ itself.
Distance Correspondence: The incidence variety $DC_k(f, Q)$ consists of pairs $([v], [u])$ where $[v] = f(p)$ is a critical point of order $k$ for the distance function $d_u$ .
Distance Locus: The projection of $DC_k$ onto the space of data points $u$ yields the Higher-Order Distance Locus $DL_k(f, Q)$ .

B. Algebraic Invariants

Higher-Order Distance Degree ( $DD_k$ ): Defined as $\deg(DL_k) \cdot \deg(\phi_{2,k})$ , where $\phi_{2,k}$ is the projection from the correspondence to the locus.
Generic Distance Degree ( $gDD_k$ ): The value of $DD_k$ when the pair $(f, Q)$ is in "general $k$ -osculating position" (transversal intersection). This is shown to be independent of the specific quadric $Q$ .
Polar Classes: The authors utilize higher-order polar classes $p_{k,i}(f)$ (introduced by Piene) to compute these degrees. The generic distance degree is the sum of the degrees of these polar classes:
$gDD_k(f) = \sum_{i=0}^m \mu_{k,i}(f)$
where $\mu_{k,i}$ are the degrees of the $i$ -th polar class.

C. Computational Approaches

The paper employs three distinct methods to compute these invariants:

Intersection Theory (Chern Classes): For $k$ -regular embeddings (where osculating spaces have maximal dimension), the authors derive recursive formulas for the Chern classes of the jet bundles $P_k(f)$ . This allows for closed-form polynomial expressions in terms of Chern classes of $X$ and the hyperplane class.
Tropical Geometry: For toric varieties, the authors introduce a tropical framework. They tropicalize the higher-order conormal variety $W_k(f, Q)$ . The higher-order distance degrees are computed as stable intersections between the tropicalized conormal variety and Bergman fans of uniform matroids. This provides a combinatorial algorithm for effective computation.
Affine Adaptation: The theory is extended to affine morphisms $f: \mathbb{A}^m \to \mathbb{A}^n$ , adapting the definitions of osculating spaces and normal bundles to the affine setting.

3. Key Contributions and Results

A. General Formulas for Generic Maps

Theorem 1.1: Provides a closed formula for the generic $k$ -th order distance degree of a generic polynomial map $f: \mathbb{P}^m \to \mathbb{P}^n$ .
$gDD_k(f) = \sum_{i=0}^m \binom{\binom{m+k}{k}}{i} (d-k)^i d^{m-i}$
This generalizes the classical EDD formula for generic maps.

B. Veronese Embeddings and the Bombieri-Weyl Metric

The paper investigates the specific case of Veronese embeddings $\nu^d_m$ (representing rank-1 symmetric tensors) under the Bombieri-Weyl metric (a specific inner product on polynomials).

Theorem 1.2: Derives explicit formulas for $DD_k(\nu^d_m, Q_{BW})$ .
Key Finding: The choice of metric profoundly affects the result. For the Bombieri-Weyl metric, the degree can be strictly lower than the generic degree ( $DD_k < gDD_k$ ), and the projection map $\phi_{2,k}$ may fail to be birational (degree $> 1$ ). This reveals "defects" in the distance degree similar to classical ED defects.
Connection to Tensors: The authors link these critical points to higher-order eigenvectors of symmetric tensors, generalizing the concept of tensor eigenvectors.

C. Toric Varieties and Tropical Computation

Theorem 1.3: Establishes that the generic $k$ -th order distance degree of a toric embedding can be computed as a sum of stable intersections between the tropical conormal variety and specific Bergman fans.
Algorithm 1: Presents a combinatorial algorithm (implemented in Julia) to compute these degrees for toric varieties, handling cases where the embedding is not $k$ -regular (e.g., Segre-Veronese embeddings).
Example 7.10: Demonstrates the computation of polar degrees for a Segre-Veronese embedding where classical Chern class formulas (Section 6) are not directly applicable due to lack of regularity.

D. Affine Morphisms and Numerical Optimization

Section 8: Extends the theory to affine varieties.
Proposition 8.8: Shows a crucial link between second-order criticality and numerical optimization. If a point is a second-order critical point for the distance function, the Hessian of the objective function equals $2 J_f^T Q J_f$ . This implies that for data points in the second-order distance locus, the Gauss-Newton method (a common approximation in numerical optimization) becomes exact.
Example 8.9: Applies this to the multi-view triangulation problem in computer vision, computing the 2nd-order distance locus for a specific toric multiview variety and showing how it certifies data configurations solvable by Gauss-Newton.

4. Significance and Impact

Generalization of Metric Algebraic Geometry: The paper successfully moves beyond first-order tangency, providing the theoretical foundation for "higher-order" metric geometry. This is essential for applications requiring curvature-aware fitting.
Metric Dependence: It highlights that algebraic invariants like distance degrees are not absolute but depend heavily on the choice of the metric (quadric). The distinction between generic metrics and specific ones (like Bombieri-Weyl) is a major theoretical insight.
Computational Tools: By bridging intersection theory with tropical geometry, the authors provide practical tools (Macaulay2 and Julia implementations) to compute these complex invariants for toric varieties, which are ubiquitous in applied mathematics.
Numerical Analysis Connection: The connection between higher-order distance degrees and the exactness of the Gauss-Newton method provides a rigorous algebraic justification for the behavior of numerical solvers in specific geometric configurations.
Tensor Approximation: The results on Veronese embeddings and Bombieri-Weyl metrics offer new perspectives on the best rank-1 tensor approximation problem, a central issue in data science.

In summary, this work unifies advanced algebraic geometry (jet bundles, polar classes, tropicalization) with practical optimization problems, offering both deep theoretical formulas and effective computational algorithms for analyzing the complexity of distance optimization on algebraic varieties.