The inverse problem of convex polygon coordinates

This paper focuses on convex polygons in the plane by summarizing and comparing Gibbs coordinates (based on entropy maximization and exponential functions) with Wachspress coordinates (rational functions), identifying their points of agreement and divergence, and demonstrating how Gibbs coordinates for polygons with rational vertices can be interpreted as algebraic functions.

The Gist

Imagine you have a convex polygon (like a square, a triangle, or a weirdly shaped four-sided figure) drawn on a piece of paper. Inside this shape, there is a specific point, let's call it The Target.

The big question this paper asks is: "How can we describe The Target using only the corners of the shape?"

In math, we do this by saying The Target is a "mixture" of the corners. If the corners are ingredients (like flour, sugar, and eggs), The Target is the cake. The "recipe" (the amounts of each ingredient) are called Barycentric Coordinates.

The problem is: There isn't just one recipe. You can make the same cake in many different ways. This paper compares two famous, very different recipes for finding the "perfect" mixture.

The Two Chefs: Gibbs vs. Wachspress

The authors compare two different "chefs" (mathematical methods) who try to find the best recipe for any point inside the polygon.

1. Chef Gibbs: The Entropy Maximizer (The "Smooth" Approach)

Chef Gibbs is a statistician. He believes the best recipe is the one that is most random or most balanced.

• The Analogy: Imagine you are trying to guess the weather. If you have no idea, you assume every possibility is equally likely. Gibbs wants the "least biased" mix.
• How it works: He uses a special formula involving exponentials (like $e^x$). It's like a complex, high-tech blender that smooths everything out perfectly.
• The Catch: Because it uses exponentials, the math is "transcendental." It's hard to calculate exactly without a computer, and the numbers can get messy. It's like trying to bake a cake using a recipe that requires "a pinch of the square root of pi."

2. Chef Wachspress: The Geometric Rationalist (The "Clean" Approach)

Chef Wachspress is a geometer. He believes the best recipe comes from simple shapes and areas.

• The Analogy: Imagine the polygon is a pizza. To find the weight of a slice, you just look at the area of the triangle formed by the point and the two corners.
• How it works: He uses rational functions (fractions of polynomials). It's like a standard kitchen scale. You measure, you divide, you get a clean fraction. No weird exponentials.
• The Catch: While the math is "cleaner" (easier for computers to handle exactly), it might not be the "most random" or balanced mix in the statistical sense.

The Great Showdown: When do they agree?

The authors spent the paper figuring out when these two chefs produce the exact same cake.

• The "Simple" Cases (Agreement):
  If your shape is a triangle or a parallelogram (a "semisimplex"), both chefs agree! They produce the exact same recipe. In these simple shapes, the "most random" mix is the same as the "geometric area" mix.

• The "Weird" Cases (Disagreement):
  If your shape is a generic quadrilateral (a four-sided shape that isn't a parallelogram), the chefs disagree.

  • Chef Gibbs might say: "This point is 22% Corner A, 24% Corner B..."
  • Chef Wachspress might say: "No, it's 18% Corner A, 26% Corner B..."
  • The Result: The point is the same, but the "recipe" is different.

The "Equator" and the Discrepancy

The paper introduces some cool concepts to map out this disagreement:

1. The Discrepancy Field: Imagine a wind map over the polygon. The wind shows you how much the two recipes differ at every single point.
2. The Equator: This is the most fascinating part. Even though the chefs disagree almost everywhere, there is a special curved line running through the middle of the shape (connecting two opposite corners) where they suddenly agree.
   • Think of it like a "peace treaty line." If you stand on this line, Chef Gibbs and Chef Wachspress shake hands and give you the exact same recipe.
   • The authors actually calculated the exact equation for this line (it's a bit complex, but it exists!).

Why Does This Matter?

You might ask, "Who cares about two different ways to mix corners?"

• Computer Graphics & Video Games: When rendering 3D models, computers need to know how to color a point inside a triangle or polygon based on the colors of the corners. Wachspress coordinates are great because they are easy for computers to calculate quickly (rational functions). Gibbs coordinates are great for physics simulations where "entropy" or randomness matters.
• Mathematical Unity: The authors show that these two seemingly different worlds (statistics/entropy and geometry/areas) are actually connected. They are both trying to solve the same problem: "How do we describe a point using its boundaries?"

The Bottom Line

This paper is a detective story about coordinates.

• The Mystery: How do we describe a point inside a shape using its corners?
• The Suspects: The "Entropy Chef" (Gibbs) and the "Geometry Chef" (Wachspress).
• The Verdict: They are best friends in simple shapes (triangles/parallelograms) but have a rocky relationship in complex shapes. However, they always find a middle ground on a special "Equator" line running through the shape.

The authors essentially built a dictionary to translate between the language of Statistics (entropy, probabilities) and the language of Geometry (areas, determinants), showing us that even in math, there are many ways to slice the same pizza.

Technical Summary

Here is a detailed technical summary of the paper "The Inverse Problem of Convex Polygon Coordinates" by Romanowska, Smith, and Zamojska-Dzienio.

1. Problem Statement

The paper addresses the inverse problem of barycentric coordinates: given a point $x$ within the convex hull of a set of extreme points (vertices) $V$, how can one uniquely express $x$ as a convex combination of the elements of $V$?
$$x = \sum_{v \in V} p_v v, \quad \text{where } \sum p_v = 1, p_v \geq 0.$$
While this representation is unique for simplices (triangles in 2D, tetrahedra in 3D), it is generally non-unique for convex polytopes with more vertices than the dimension of the space plus one (e.g., a quadrilateral in 2D). The authors focus on determining specific, well-defined coordinate systems for convex polygons in the plane, comparing two prominent approaches: Gibbs coordinates and Wachspress coordinates.

2. Methodology and Theoretical Framework

The authors employ an algebraic perspective based on barycentric algebras, moving away from traditional affine geometry to treat convex sets as intrinsic algebraic structures.

• Barycentric Algebras: The paper defines a barycentric algebra as a set equipped with binary operations $xy_p$ (for $p \in (0,1)$) satisfying idempotence, skew-commutativity, and skew-associativity. This framework allows the authors to discuss convex sets without relying on an ambient vector space.
• Free Barycentric Algebras: Simplices are identified as free barycentric algebras generated by their vertices.
• Two Coordinate Systems:
  1. Gibbs Coordinates: Derived from statistical mechanics, these coordinates maximize the entropy of the probability distribution representing the point. They are defined via the Gibbs distribution (softmax function) involving exponential functions and potentials.
  2. Wachspress Coordinates: A well-established system in computer graphics and finite element analysis. For polygons, these are defined using rational functions involving signed areas (determinants) and are interpreted as a "bulk-boundary correspondence" related to discrete boundary curvature.

3. Key Contributions and Results

A. Algebraic Unification

The paper establishes that both coordinate systems can be viewed as kernel functions (or partitions of unity) within the category of barycentric algebras.

• Gibbs Coordinates: Defined via the maximization of entropy $H(\alpha) = \sup \{-\sum p_i \log p_i\}$. The coordinates are given by $q_\beta(x) = \frac{e^{-x\beta}}{Z(\beta)}$, where $\beta$ is a potential function and $Z$ is the partition function.
• Wachspress Coordinates: Defined geometrically. For a polygon vertex $v_i$, the weight is proportional to the ratio of the area of the triangle formed by neighbors to the product of areas of triangles formed with the interior point.

B. Comparison and Coincidence

The authors rigorously compare the two systems:

• Coincidence on Semisimplices: Theorem 5.5 proves that Gibbs and Wachspress coordinates coincide on semisimplices (polytopes that are direct products of simplices, e.g., triangles and parallelograms in 2D). In these cases, the Wachspress coordinates also maximize entropy.
• Divergence on General Polygons: For general quadrilaterals (and higher-order polygons), the coordinates differ. The paper provides a specific counter-example (Example 5.6) showing distinct coordinate vectors for the same interior point.

C. The G-W Discrepancy Field

To quantify the difference, the authors introduce the (G-W) discrepancy field:

• Defined as an $(n-1)$-dimensional vector field over the polygon, where the $i$-th component is the difference between the $i$-th Gibbs coordinate and the $i$-th Wachspress coordinate.
• Proposition 5.8: These discrepancy vectors lie in a vector space of dimension $n-3$. For a quadrilateral ($n=4$), the discrepancy is a 1-dimensional vector (a scalar magnitude), which can be visualized as a contour plot (Figure 5).
• Boundary Behavior: The coordinates agree on the boundary of the polygon (where the discrepancy is zero).

D. The Equator

For a general quadrilateral, the set of points where Gibbs and Wachspress coordinates coincide forms an algebraic curve called the equator.

• The authors derive the explicit equation for this equator (Theorem 5.10) for a specific quadrilateral, showing it connects two opposite vertices through the interior.
• This demonstrates that while the systems differ generally, they intersect along specific algebraic manifolds.

E. Algebraic Nature of Gibbs Coordinates

A significant technical contribution is the demonstration that for polygons with rational vertices, Gibbs coordinates can be construed as algebraic functions rather than transcendental ones.

• By utilizing the specific algebraic relations of the vertices (rational convex combinations), the exponential terms in the Gibbs definition can be eliminated, allowing the coordinates to be expressed using roots and rational operations (Section 5.4.1).

4. Significance and Implications

• Bridging Disciplines: The paper successfully bridges statistical mechanics (Gibbs/entropy), algebraic geometry (Wachspress/rational functions), and universal algebra (barycentric algebras).
• Generalization: It provides a unified framework for understanding barycentric coordinates that extends beyond Euclidean affine spaces to abstract barycentric algebras.
• Computational Geometry: The comparison highlights the trade-offs between the two systems:
  • Wachspress: Computationally efficient (rational functions), geometrically intuitive (curvature-based), but does not maximize entropy.
  • Gibbs: Maximizes entropy (statistically "most likely" distribution), but typically involves transcendental functions (exponentials), though the paper shows they can be algebraic in specific rational cases.
• Future Directions: The authors suggest applying threshold convexity to handle numerical noise (distinguishing zero from non-zero coordinates) and using modal theory to approximate continuous convex sets via polytopes.

Conclusion

The paper resolves the inverse problem for convex polygons by characterizing the relationship between the entropy-maximizing Gibbs coordinates and the rational Wachspress coordinates. It proves they are identical only for semisimplices and diverges otherwise, quantifying this divergence via a discrepancy field and identifying the "equator" where they intersect. This work offers a robust algebraic foundation for coordinate systems used in computer graphics, finite element methods, and geometric modeling.