🔬 condensed matter

CoBarS: Fast reweighted sampling for polygon spaces in any dimension

This paper introduces CoBarS, the first algorithm that efficiently and correctly samples random closed $n$ -gons with fixed edge lengths in any dimension by generating open polygons, applying an $O(n)$ Möbius transformation to close them, and computing explicit reweighting factors to transform the resulting distribution into standard measures like symplectic volume.

Original authors: Jason Cantarella, Henrik Schumacher

Published 2026-08-03

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jason Cantarella, Henrik Schumacher

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 a detective trying to solve a mystery about how things move and connect. In the world of science, there is a special branch called statistical physics that studies how giant chains of atoms, like the plastic in a water bottle or the DNA in your cells, wiggle and fold. These chains are often modeled as "polygons"—shapes made of straight sticks connected end-to-end. If the chain is a ring, the first stick must connect back to the last one, forming a closed loop. Scientists also care about these shapes in robotics, where a robot arm with joints that can spin in any direction is essentially a polygon trying to reach a target without breaking its own rules.

The big challenge is that these shapes can twist into an almost infinite number of configurations. To understand them, scientists need to take a "snapshot" of a random shape, but not just any random shape—they need a snapshot that is perfectly fair, representing every possible shape with the correct probability. Think of it like rolling a die: if you want to know the average roll, you need a fair die where every number has an equal chance. If your die is weighted (biased), your answer will be wrong. For decades, scientists had trouble creating a "fair die" for these complex, multi-dimensional shapes, especially when the sticks were different lengths or the shapes existed in spaces with more than three dimensions.

This is where the paper "CoBarS: Fast reweighted sampling for polygon spaces in any dimension" steps in. The authors, Jason Cantarella and Henrik Schumacher, have invented a clever new method to generate these random shapes quickly and fairly. Their approach is like a magic trick involving a rubber sheet and a magnet.

First, they start with an "open" polygon—a chain of sticks that doesn't quite close the loop. They pick the directions of these sticks randomly, like spinning a compass needle for each stick. Usually, this open chain won't close; the end will be far from the start. To fix this, they use a mathematical tool called a Möbius transformation. You can imagine this as a special kind of lens or a stretchy rubber sheet that bends space. They use this lens to warp the open chain until the "center of gravity" of the sticks lands exactly on the origin, effectively pulling the ends together to close the loop.

Here is the catch: simply warping the shape changes the odds. Some shapes become more likely to appear, and others become rare, just like stretching a rubber sheet makes some parts thicker and others thinner. If you just took these warped shapes as your final answer, your "die" would be weighted, and your scientific results would be biased.

The paper's main breakthrough is figuring out exactly how biased the results are and providing a formula to fix it. They calculated "reweighting factors"—mathematical correction numbers that tell you how much to trust each shape you generated. By multiplying the results by these factors, they cancel out the bias, ensuring that every possible shape is counted with the perfect, fair probability.

The authors proved that their method works for polygons in any number of dimensions (not just the 3D world we live in) and for any set of stick lengths, as long as the sticks aren't so uneven that they can't possibly close. They also showed that this process is incredibly fast. While older methods took time that grew quadratically (getting much slower as the number of sticks increased), their new method, called CoBarS, scales linearly. This means if you double the number of sticks, the time it takes only doubles, not quadruples.

In their experiments, they tested their method against known mathematical answers for hexagons (six-sided shapes) and tetragons (four-sided shapes). The results matched the theory perfectly, even when they used a million random samples. They also compared their speed to other popular methods, finding that for shapes with more than about 50 sticks, their new algorithm was significantly faster at giving accurate answers.

The paper doesn't just offer a new way to guess shapes; it offers a way to prove those guesses are correct. They provided an open-source code so anyone can use this "fair die" to study everything from how polymers fold to how robot arms move. While they didn't solve every mystery about these shapes, they gave scientists a powerful, fast, and reliable tool to explore the vast, twisting world of polygon configurations with confidence.

Technical Summary: CoBarS – Fast Reweighted Sampling for Polygon Spaces

Problem Statement
The paper addresses the challenge of efficiently sampling random configurations of closed $n$ -gons with fixed edge lengths $r_1, \dots, r_n$ in arbitrary dimension $d$ . Existing algorithms for this problem suffer from significant limitations: many are restricted to dimensions $d=2$ or $d=3$ , lack proofs of sampling from the correct probability measure, or are limited to equilateral polygons. The configuration space of such polygons is relevant to the statistical physics of ring polymers (freely-jointed chain models), robotics (kinematic configuration spaces of closed-loop arms), and differential geometry/topology. The goal is to develop a method that is fast, mathematically proven to sample from standard measures (including symplectic volume), and capable of handling arbitrary dimensions and edge lengths.

Methodology
The authors propose Conformal Barycenter Sampling (CoBarS), a two-stage process involving the generation of open polygons followed by a geometric closure and reweighting procedure.

Generation of Open Polygons:
The algorithm begins by sampling an open "arm" of $n$ unit vectors $x = (x_1, \dots, x_n)$ from a product of spheres $S^{d-1}$ . These vectors are weighted by the fixed edge lengths $r_i$ . While these open configurations satisfy the edge length constraints, they generally do not close (i.e., $\sum r_i x_i \neq 0$ ).
Conformal Closure:
To close the polygon, the authors utilize the concept of the conformal barycenter from hyperbolic geometry. Viewing the unit ball $B$ as the Poincaré disk model of hyperbolic space, they define a map that finds a unique point $w^* \in B$ (the conformal barycenter) such that the weighted sum of the "directors" (unit tangent vectors to geodesics from $w^*$ to $x_i$ ) vanishes.
A specific hyperbolic isometry (a "shift map" $\sigma$ ) is then applied to the open configuration $x$ to move the conformal barycenter $w^*$ to the origin. This transformation, $\sigma(-w^*, x)$ , maps the open arm to a closed polygon $y$ in the configuration space $\text{Pol}$ . This closure map is computable in $O(n)$ time using existing algorithms.
Reweighting:
The direct application of the closure map introduces a sampling bias because the pushforward measure from the open space does not match the standard probability measure on the closed polygon space. The core theoretical contribution is the explicit derivation of reweighting factors (Jacobian determinants) that correct this bias.
The authors derive a closed-form expression for the Jacobian determinant $J_{op}$ of the inverse "opening" map. By combining this with a carefully chosen weight function $\chi(w)$ on the hyperbolic ball, they construct a weight function $K(w, y)$ that transforms the biased samples into samples distributed according to the target measure (e.g., the symplectic volume or the Riemannian volume on the quotient space).
Handling Quotient Spaces:
The method is extended to the moduli space of polygon shapes (quotient by the rotation group $SO(d)$). The authors derive additional correction factors involving the eigenvalues of the inertia tensor to account for the volume of the $SO(d)$ orbits, allowing for sampling of shape distributions rather than just poses.

Key Contributions

First Proven Algorithm: CoBarS is presented as the first algorithm capable of sampling closed $n$ -gons with arbitrary fixed edge lengths in any dimension $d$ that is rigorously proven to sample from standard probability measures.
Explicit Reweighting Factors: The paper provides an explicit, computable formula for the reweighting factors required to eliminate sampling bias. Crucially, these factors can be computed in $O(n)$ time for fixed dimension $d$ , avoiding the $O(n^3)$ complexity typically associated with Jacobian determinants of high-dimensional maps.
Generality: Unlike previous methods (such as Action-Angle or Progressive Action-Angle methods), CoBarS is not restricted to $d=3$ or equilateral polygons. It accommodates arbitrary edge lengths and dimensions.
Open-Source Implementation: The authors provide open-source reference implementations in C++ and Mathematica.

Results

Accuracy: Experimental results demonstrate that the reweighted sampling accurately recovers theoretical probability density functions. The authors tested the method on hexagons with both equilateral and non-equilateral edge lengths in $R^3$ , comparing the distribution of chord lengths against known analytical solutions. The histograms of 1 million samples matched the theoretical curves with high precision.
Efficiency: While the reweighted sampling requires slightly more samples (approx. 25% more) than the Action-Angle method to estimate the mean radius of gyration to a fixed confidence interval, the computational complexity per sample is superior. CoBarS operates in $O(n)$ time, whereas the Action-Angle method requires $O(n^{2.5})$ and the Progressive Action-Angle method requires $O(n^2)$ . Consequently, CoBarS becomes faster overall for $n \gtrsim 50$ .
Robustness: The method handles arbitrary edge lengths and dimensions, provided the edge lengths satisfy the triangle inequality condition (no single edge is longer than the sum of the others).

Significance
The paper claims that CoBarS fills a critical gap in the literature by providing a fast, general, and mathematically controlled method for polygon sampling. Its ability to handle arbitrary dimensions and edge lengths makes it a versatile tool for researchers in statistical physics, robotics, and geometry. The authors suggest that this efficiency and generality open new avenues for investigating the topology and geometry of polygon spaces, such as computing volumes of these spaces or using topological data analysis to study their cohomology, questions that have been difficult to address computationally in higher dimensions. The method effectively bridges the gap between theoretical geometric measures and practical computational sampling.

More like this