🔢 mathematics

A Database of Continued Fractions of Polynomial Type

This paper presents a comprehensive database of 1,883 continued fractions with polynomial coefficients, including over 1,600 new entries for interesting constants and transcendental functions, which is provided directly within the paper's LaTeX source code.

Original authors: Henri Cohen

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Henri Cohen

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 have a massive, dusty library filled with ancient scrolls. Each scroll contains a secret recipe for calculating a specific number (like $\pi$ or $\log(2)$ ) or a mathematical function. These recipes are written in a very specific, complex shorthand called Continued Fractions.

For centuries, mathematicians have found a few hundred of these recipes. They are beautiful, but they are scattered, hard to find, and often missing the "instruction manual" on how fast they work or where they came from.

Henri Cohen's paper is essentially the construction of a giant, organized digital warehouse for these recipes.

Here is a breakdown of what this paper does, using simple analogies:

1. The "Recipe Book" (The Database)

Cohen has compiled a database of 1,883 of these mathematical recipes.

The Old vs. The New: Before this, only about 250 were known. Cohen used a powerful computer program (a "kitchen robot" called Pari/GP) to invent over 1,600 new recipes.
The Format: Think of these recipes not as a paragraph of text, but as a structured list of ingredients. Each recipe tells you how to build a number by stacking fractions on top of each other forever.
The "Polynomial" Twist: Most of these recipes use "polynomial coefficients." Imagine a recipe that says, "Add the number 5, then add 10, then 15, then 20." That's a simple pattern. Cohen's database is full of recipes where the numbers follow complex, predictable algebraic patterns (like $n^2$ or $2n+1$ ) rather than just random numbers.

2. The "Speedometer" (Convergence Rates)

If you are baking a cake, you want to know how close you are to being done. In math, this is called convergence.

The Problem: Most old recipe books just give you the recipe. They don't tell you, "If you stop after 10 steps, you are 90% accurate. If you stop after 100 steps, you are 99.9% accurate."
The Solution: Cohen calculated the "speedometer" for almost every single recipe in his database. He figured out exactly how fast the recipe approaches the true answer. He even wrote down the "error formula" (like a map showing how far off you are if you stop early).

3. The "Instruction Manual" (Origin and Connections)

Cohen didn't just list the recipes; he tried to explain where they came from.

The "Apéry" Connection: He notes which recipes were made using a special technique called "Apéry acceleration" (a method to make recipes converge faster, like adding a secret ingredient to speed up baking).
The "Hypergeometric" Link: He connects many of these fraction recipes to a different type of math called "hypergeometric series." It's like showing that a French recipe is actually just a variation of a Chinese one, just written differently.

4. How to Read the "Code"

The paper admits that the database is written in a computer language (Pari/GP) that looks like code to a human.

The Analogy: Imagine a library where the books are written in binary code. You can't read them with your eyes, but a computer can read them instantly.
The Solution: The paper provides a "decoder ring." It explains exactly how the computer stores these recipes (using vectors and closures) so that other mathematicians can plug this data into their own computers.
The "Beautification": Cohen admits the raw code is ugly. He shows how to translate one of these code blocks into a pretty, readable mathematical formula that a human can actually understand.

5. The "Mystery Box" (Open Questions)

At the end of the paper, Cohen acts like a detective who has found a treasure chest but still has a few loose ends. He lists questions he couldn't answer, such as:

"Why do some of these recipes seem to have an infinite loop of simplification?"
"Why are there so many recipes for some numbers but almost none for others?"
"Can we find a 'sister' recipe for this one that we haven't found yet?"

The Bottom Line

This paper is not a story about a new discovery that solves a real-world problem (like curing a disease or building a bridge). Instead, it is a reference tool.

It is the "Yellow Pages" or the "Google Maps" for a very specific, niche part of mathematics. Cohen says, "Here is the map. Here are the coordinates. Here is how fast you get there. Now, you mathematicians, go explore."

Crucial Note from the Author:
Cohen is very honest about the limitations. He admits that while he found these 1,600+ new recipes using powerful computer searches, he doesn't always have the formal mathematical proof for why they work. He says, "I found them, and I'm 99% sure they are right, but I haven't written the proof for every single one yet." He invites other mathematicians to check his work and fill in the missing proofs.

Technical Summary: A Database of Continued Fractions of Polynomial Type

Problem Statement
The paper addresses the scarcity of accessible, comprehensive resources regarding continued fractions (CFs) with polynomial coefficients, particularly those associated with interesting mathematical constants and transcendental functions. While approximately 250 such CFs existed in the literature prior to this work, the author notes that the field lacks a unified database that includes precise convergence rates, asymptotic expansions, and origins (such as Apéry acceleration techniques). Furthermore, standard literature often omits the precise rate of convergence of partial quotients, a detail the author deems essential for a complete understanding of these objects.

Methodology
The author developed a specialized Pari/GP software package (accessible via the henri-ellCF2 GIT branch) to systematically generate, analyze, and format continued fractions. The methodology involves:

Generation: Starting from roughly 250 known CFs with polynomial coefficients, the author applied algorithmic techniques, including Apéry acceleration and Bauer–Muir methods, to derive over 1,600 new CFs.
Convergence Analysis: For nearly every entry, the author computed the precise rate of convergence of the partial quotients ( $p(n)/q(n)$ ) and derived the first few terms of their asymptotic expansions.
Classification and Formatting: The CFs were categorized based on their coefficients. A CF is defined as having "polynomial coefficients" if, for a period $T$ (usually 1 or 2), the coefficients $a(n)$ and $b(n)$ become rational functions of $n$ for sufficiently large $n$ . The author established a specific vector-based format for computer representation in Pari/GP, distinguishing between standard CFs, parametrized families, and definitions.
Database Construction: The resulting data, comprising 1,883 entries, was compiled into a machine-readable file embedded within the paper's TeX source.

Key Contributions

The Database: The primary contribution is a database of 1,883 continued fractions with polynomial coefficients. Of these, more than 1,600 are new additions to the literature.
Convergence Data: Unlike previous works, this database provides the exact speed of convergence and asymptotic expansions for the error term $L - p(n)/q(n)$ for almost every entry. The error is expressed in the form $C / (F \cdot E \cdot n^D \cdot n! \cdot P) \cdot A$ , where $A$ is an asymptotic series in powers of $1/n$ or $1/\sqrt{n}$ .
Hypergeometric Connections: The database links many CFs to explicit hypergeometric series, providing the series representation for both the CF and its inverse where applicable.
Apéry Acceleration Tracking: The author documents the behavior of CFs under Apéry acceleration processes, noting whether they lead to other labeled CFs, become self-dual, or result in overly complex expressions.
Software Integration: The paper provides a specific, transparent data format readable by Pari/GP (and other software via interface), allowing for immediate computational verification and further exploration.

Results
The database is organized into 1883 entries, indexed and labeled (e.g., "1.1.0.5", "3.2.4"). The entries cover a wide spectrum of mathematical objects:

Constants: Including roots of integers, $\pi$ , $\log(2)$ , $\zeta(k)$ , $G$ (Catalan's constant), and linear combinations of Zeta and L-values.
Functions: Exponential, logarithmic, power, hyperbolic, trigonometric, and their inverses.
Special Functions: Gamma functions, Digamma functions ( $\psi$ ), Beta functions, Bessel functions ( $J_\nu, I_\nu, K_\nu$ ), and Elliptic integrals.
Parametric Families: Several families of CFs depending on parameters (e.g., $k, u$ ) are included, particularly for periods of degree 2 and 3.

The author notes that while many new CFs were derived rigorously, others were obtained "à la Ramanujan" (heuristic methods) and are presented with high confidence but without formal proof. The database also highlights specific anomalies, such as CFs that are "infinitely contractible" (maintaining the same bidegree upon contraction) and gaps in the literature regarding specific functions like $\beta_2(z)$ or Bessel functions of the second kind ( $Y_\nu$ ).

Significance and Claims
The author explicitly states that this is "not quite a standard mathematical paper" but aims to serve as a "useful" and "standard reference" for mathematicians interested in continued fractions. The significance lies in the sheer volume of new data and the inclusion of convergence details rarely found in texts.

Utility: The paper claims the document is "useless" without the accompanying TeX source file, which contains the actual database.
Modesty: The author is modest regarding the proofs, admitting that for many entries, the origin is heuristic and proofs are not provided. The convergence speeds and constants were often determined using heuristic methods and linear dependence algorithms (e.g., Pari/GP's lindep), and while believed to be correct, they are not rigorously proven within the text.
Future Work: The author envisions the database being hosted on a website for community updates but acknowledges a lack of technical resources to do so personally. The paper invites corrections and additions, noting that the current file is a starting point for a potential larger book containing "beautified" human-readable versions of the CFs.