This paper, published in 2002 by Keith Briggs, explores a classic problem in number theory: how "badly" can we approximate a pair of irrational numbers using fractions with a common denominator? While the one-dimensional version of this problem was solved over a century ago, the two-dimensional case remains one of the great mysteries of Diophantine approximation. Briggs combines deep theoretical results with high-precision computation to find specific pairs of numbers that are harder to approximate than any previously known.
TL;DR
Briggs identifies specific pairs of irrational numbers that are exceptionally difficult to approximate with rational fractions, coming closer than ever to a conjectured theoretical limit. By calculating over 2 million terms of a continued fraction and applying Cusick’s theorem, he discovered a pair with an approximation constant of $\approx 0.2857082$, just a hair's breadth away from the suspected maximum of $2/7$.
What Problem Does This Paper Address?
In the world of Diophantine approximation, we want to know how closely we can "mimic" an irrational number $\alpha$ using a fraction $p/q$. Dirichlet’s theorem guarantees that we can always find infinitely many fractions such that the error is less than $1/q^2$. Some numbers, like the golden ratio, are "badly approximable"—they are the "most irrational" because they resist being close to any fraction.
When you move to two dimensions, you try to approximate a pair $(\alpha_1, \alpha_2)$ using two fractions $(p_1/q, p_2/q)$ with the same denominator $q$. The "goodness" of this approximation is measured by a constant $c(\alpha)$. For a single number, the best possible constant is $1/\sqrt{5}$. For two numbers, the analogous constant $c_2$ is unknown. Mathematical "folklore" suggests it is $2/7$, but proving this—or even finding pairs that get close to it—is notoriously difficult.
Methods
Briggs uses a specialized "hunting" strategy based on the work of T.W. Cusick.
- Cubic Fields: The search is narrowed down to the cubic number field $\mathbb{Q}(\theta)$, where $\theta = 2\cos(2\pi/7)$. This specific field is chosen because it has the smallest possible positive discriminant (49), making it a prime candidate for finding "worst-case" numbers.
- Pattern Matching: Cusick’s theorem provides a shortcut: if the continued fraction expansion of $\theta$ contains specific patterns (like large numbers separated by "1, 1" or "2"), one can construct an integral basis $\{1, \alpha, \beta\}$ that is very badly approximable.
- Brute Force Computation: Briggs computed over 2 million partial quotients of the continued fraction of $\theta$. This is a massive task requiring high-precision arithmetic, as the numbers involved grow exponentially.
- Verification: To ensure the results weren't just numerical artifacts, Briggs used the Furtwängler algorithm—a tool for finding best simultaneous rational approximations—to verify the behavior of the discovered pairs.
Datasets
The "dataset" in this paper is not a collection of external files, but rather the mathematical properties and continued fraction expansion of the algebraic number $\theta = 2\cos(2\pi/7)$, derived from the cubic equation $x^3 + x^2 - 2x - 1 = 0$.
Key Insights and Ideas
- The Rarity of Badness: Even in a field primed for "bad" numbers, the required patterns in the continued fraction occur "very infrequently." Briggs had to dig through millions of terms to find just a handful of candidates.
- The "Long Initial Transient": One of the most interesting findings is that these pairs don't look badly approximable at first. If you only check small denominators ($q$), they seem easy to approximate. Their "badness" (the large approximation constant) only emerges after an "extremely long initial transient." You have to look at denominators of staggering size before the true nature of the pair is revealed.
- Cusick’s Construction: The paper demonstrates that Cusick’s theoretical framework is not just an abstract existence proof but a practical recipe for finding record-breaking constants.
Key Results
Briggs identifies seven specific pairs (labeled A through G) with exceptionally large approximation constants.
- The Record Holder: Pair (G) yielded an approximation constant $c^* \approx 0.2857082$.
- Proximity to the Limit: This value is remarkably close to the conjectured limit of $2/7 \approx 0.2857143$.
- Scale of Computation: In the most extreme case (G), the integers $p, q, r, s$ used to define the pair are of the order $2^{3,000,000}$. These are numbers with nearly a million digits.
Limitations and Future Work
- The $2/7$ Mystery: While the results strongly support the $2/7$ conjecture, they don't prove it. The value $c_2$ remains unknown for the general case of all irrational pairs.
- Restricted Search: The method is specifically tied to the cubic field $\mathbb{Q}(\theta)$. It remains an open question whether even "worse" pairs exist outside of this field (though Davenport's theorem suggests the supremum over all pairs might be the same as the supremum over this field).
- Computational Limits: Finding even better pairs would require searching even deeper into the continued fraction, which becomes computationally expensive due to the precision required.
Summary updated