🔢 mathematics

Prescribed realisation of longest runs in continued fractions

This paper demonstrates that for any prescribed partial quotient and admissible growth scale, there exists a set of irrational numbers with full Hausdorff dimension where the longest run of that specific symbol uniquely determines the overall maximum run length, thereby proving that the symbol responsible for the longest run can be fixed in advance without reducing the dimension of the exceptional set.

Original authors: Ying Wai Lee

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ying Wai Lee

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 looking at a very long, never-ending string of numbers. These aren't just random numbers; they are the "partial quotients" that make up the continued fraction of a specific irrational number (like $\pi$ or $\sqrt{2}$ ). You can think of this string as a long line of colored beads, where each bead has a number on it.

In this paper, the author, Ying Wai Lee, is studying a specific game played with these beads: The Longest Run Game.

The Game: Finding the Longest Streak

Imagine you are scanning the bead string looking for the longest consecutive streak of the same number.

If you have a sequence like ... 5, 5, 5, 2, 2, 2, 2, 7 ..., the longest run of the number 5 is 3. The longest run of the number 2 is 4.
The "Overall Longest Run" is simply the biggest streak of any number in that section. In the example above, the overall winner is 2 with a streak of 4.

Usually, mathematicians have studied two separate things:

The Fixed-Symbol Game: "How long is the longest streak of the number 5?"
The Overall Game: "What is the longest streak of any number?"

The Problem: The "Accidental" Winner

In the past, researchers knew that for almost all numbers, these streaks grow in a predictable way. However, there was a gap in our understanding.

If you forced the number 5 to have a very specific, growing streak length, did that guarantee that 5 would always be the winner of the "Overall Longest Run" game?
Or, could a different number (like 7) accidentally sneak in and have an even longer streak, stealing the title from 5?

Before this paper, we didn't know if we could force a specific number to be the "Champion" of the longest runs while also controlling exactly how long those runs were.

The Solution: The "Champion Controller"

Ying Wai Lee's paper says: Yes, you can.

The author proves that you can construct a massive, complex set of numbers (so large it has "full dimension," meaning it's not a tiny, rare exception) where you get to pick:

Which number is the champion (e.g., "I want the number 7 to always win").
How fast the champion's streaks grow (e.g., "I want the streaks to grow exactly as fast as the square root of the total length of the string").

The paper shows that you can build these numbers so that the number 7 not only grows its streaks at your exact prescribed speed, but it always beats every other number. No other number is ever allowed to have a streak as long as 7's.

The Analogy: The Race Track

Think of the continued fraction as a race track with many runners (the numbers 1, 2, 3, etc.).

Old View: We knew that if you watched the race long enough, the runners would generally keep up with a certain average speed. We also knew that if you picked a specific runner (say, Runner 7), you could find a track where Runner 7 ran at a specific speed. But we didn't know if Runner 7 would stay in the lead.
New View (This Paper): Lee shows you can design a track where Runner 7 is not only running at a speed you dictate, but is also guaranteed to be the fastest runner on the track at every single moment. No other runner can ever catch up to Runner 7's longest sprint.

Why This Matters (In Math Terms)

The paper achieves this by building a "Cantor-type set" (a fractal-like structure). They use a construction method where they:

Insert long, perfect blocks of the "Champion" number (e.g., 7, 7, 7...).
Carefully fill the gaps with other numbers, but keep those other numbers' streaks short enough so they never threaten the Champion.
Use "separators" (like 8 and 9) to break up any accidental long streaks of other numbers.

The result is a mathematical proof that the "Champion" of the longest runs can be prescribed in advance. You don't have to hope that a specific number wins; you can mathematically force it to win, while still maintaining the full complexity and size of the set of numbers you are studying.

Summary

The Topic: Continued fractions (a way of writing numbers).
The Puzzle: Can we force a specific number to have the longest streaks, growing at a specific rate, while ensuring no other number ever beats it?
The Answer: Yes. The paper constructs a huge collection of numbers where a chosen number is the undisputed, unique winner of the "Longest Run" game, growing exactly as fast as we tell it to.
The Impact: It unifies two separate mathematical problems (fixed-symbol growth and overall growth) into one powerful result, showing that the "winner" of the game is fully controllable.

Technical Summary: Prescribed Realisation of Longest Runs in Continued Fractions

Problem Statement
The paper investigates the asymptotic behavior of longest runs in the continued fraction expansions of irrational numbers in the unit interval $[0, 1)$ . Two distinct perspectives on longest runs are traditionally studied:

Fixed-symbol longest runs ( $L_n(x, \lambda)$ ): The maximum length of a consecutive run of a specific partial quotient value $\lambda$ within the first $n$ terms.
Overall longest runs ( $R_n(x)$ ): The maximum length of a consecutive run of any partial quotient value within the first $n$ terms.

Previous literature (e.g., Song–Zhou, Wang–Wu) established that for almost every $x$ , these quantities follow specific logarithmic growth laws. Furthermore, exceptional sets of full Hausdorff dimension exist where these growth rates deviate from the typical law or follow prescribed growth scales. However, a gap remained in understanding the interaction between these two prescriptions:

Prescribing the growth of a fixed-symbol run does not guarantee that this specific symbol $\lambda$ is the one responsible for the overall maximum run length.
Conversely, prescribing the growth of the overall maximum does not specify which symbol achieves it.

The central problem addressed is whether one can construct a set of full Hausdorff dimension where a prescribed symbol $\lambda$ not only exhibits a prescribed growth rate for its longest runs but also uniquely dominates all other symbols in terms of run length at every stage.

Methodology
The author employs a constructive approach involving the creation of a Cantor-type set with specific combinatorial and dynamical properties.

Admissible Functions: The growth rates are governed by "admissible" functions $f: \mathbb{N} \to \mathbb{N}$ , defined as non-decreasing, unbounded functions satisfying $\lim_{n \to \infty} \frac{f(n+f(n))}{f(n)} = 1$ . This class generalizes previous results which were often restricted to logarithmic scales.
Cantor Set Construction: A set $\mathcal{K}_{\lambda, f, M, B}$ $K_{λ, f, M, B}$ is constructed by iteratively defining blocks of partial quotients.
- The construction partitions the expansion into stages $k$ .
- At each stage, "free" digits are retained and subdivided.
- Crucially, specific "prescribed" blocks of the target symbol $\lambda$ (of length $s_k \approx f(t_k)$ ) are inserted.
- "Separator" digits ( $\sigma_1 = \lambda+1, \sigma_2 = \lambda+2$ ) are inserted between free segments and the prescribed blocks to prevent accidental long runs of $\lambda$ or other symbols from forming across boundaries.
Dimension Estimation:
- The Hausdorff dimension is estimated using a mass distribution principle adapted to continued fractions.
- A key technical step involves analyzing the continuant function $K(\omega)$ , which determines the denominator of the convergent and the size of cylinder sets.
- The author establishes distortion lemmas (Lemmas 7 and 8) showing that the insertion of separators and prescribed blocks distorts the continuant in a controlled manner. Specifically, the ratio of the continuant of the expanded word to the underlying word is bounded, allowing the dimension of the constructed set to be related to the dimension of a simpler set $\mathcal{F}_B$ (numbers with bounded partial quotients).
- By taking limits as the bound $B \to \infty$ and the parameter $\alpha \to 1$ , the dimension is shown to approach 1.

Key Contributions and Results

Theorem 1 (Main Result):
For any prescribed symbol $\lambda \in \mathbb{N}$ and any admissible growth function $f$ , the set
$E_{\lambda, f} = \left\{ x \in [0, 1) \setminus \mathbb{Q} : \lim_{n \to \infty} \frac{L_n(x, \lambda)}{f(n)} = 1 \text{ and } \forall n, \forall \mu \neq \lambda, L_n(x, \lambda) > L_n(x, \mu) \right\}$
has Hausdorff dimension equal to 1.

This result simultaneously prescribes the growth rate of the longest run for a specific symbol and ensures that this symbol is the unique maximizer of the overall longest run.

Corollaries:

Refined Fixed-Symbol Result: The set of numbers where the longest run of a fixed symbol $\lambda$ grows exactly as $f(n)$ has dimension 1. This generalizes Song–Zhou's result by allowing any admissible function $f$ , not just logarithmic scales.
Refined Overall Result: The set of numbers where the overall longest run grows exactly as $f(n)$ has dimension 1. This recovers Wang–Wu's result but adds the constraint that the maximizing symbol can be pre-specified.
Prescribed Dominance: The set of numbers where a specific symbol $\lambda$ strictly dominates all other symbols in run length (eventually, and in fact for all $n$ in the constructed set) has dimension 1. This is a new phenomenon not previously established in the literature.

Significance
The paper claims to provide a "genuine simultaneous refinement" of existing exceptional-set theorems. By demonstrating that the symbol responsible for the overall longest run can be fixed in advance without any loss of Hausdorff dimension, the work bridges the gap between fixed-symbol and overall longest-run theories. It shows that the "maximizing symbol" in the overall problem is fully prescribable. The results strengthen the understanding of atypical behavior in continued fractions, proving that extreme run-length behaviors can be engineered to be both specific to a symbol and dominant over all others, even when the growth rate is arbitrarily prescribed (within the class of admissible functions).