Linear fractals of the Besicovitch-Eggleston type

This paper investigates the topological, metric, and fractal properties of subsets of the interval $[0,1]$ defined by specific asymptotic means of digits in their ternary representations, while exploring their relationship to numbers characterized by fixed digit frequencies.

The Gist

Imagine you have a magical number line stretching from 0 to 1. Now, imagine you have a special camera that can zoom in on any number on this line and see its "DNA." In mathematics, this DNA is called a representation. Just as you can write the number 1/3 as 0.333... in our normal decimal system, you can write any number using a ternary (base-3) system.

In this ternary system, every number is built from a sequence of three possible "building blocks" or digits: 0, 1, and 2.

For example, the number 0.121022... is just a long string of these three digits.

The Big Question: What's the Average?

The paper by Pratsiovytyi and Klymchuk asks a simple but deep question: If you look at the infinite string of digits for a specific number, what is the "average" value of those digits?

• If a number is mostly made of 0s, its average is close to 0.
• If it's mostly 2s, the average is close to 2.
• If it's a perfect mix of 0s, 1s, and 2s, the average is 1.

The authors call this the "Asymptotic Mean." It's like asking: "If I flip a coin forever, what percentage of the time will it land on heads?" Here, instead of a coin, we are looking at the digits of a number.

The "Normal" vs. The "Weird"

Most numbers you encounter in real life are what mathematicians call "Normal."

• The Analogy: Imagine a perfectly shuffled deck of cards. If you draw cards forever, you will eventually see an equal number of Aces, Kings, and Queens.
• In Math: For almost all numbers, the digits 0, 1, and 2 appear exactly one-third of the time each. The average is exactly 1.

However, the paper is interested in the weird, rare numbers that don't follow this rule.

• What if a number has 90% zeros and 10% ones? Its average would be 0.1.
• What if a number has a chaotic pattern where the average keeps jumping up and down and never settles on a single number?

The "Fractal" Discovery

The authors study the collection of all numbers that have a specific, pre-chosen average. Let's say you want to find every single number whose digits average out to exactly 0.5.

You might think this collection is just a random scattering of points. But the paper reveals something magical: These numbers form a "Fractal."

• The Metaphor: Imagine a coastline. From far away, it looks like a smooth line. But as you zoom in, you see bays, rocks, and tiny pebbles. No matter how much you zoom in, the coastline is still jagged and complex.
• The Result: The set of numbers with a specific average (like 0.5) is a "jagged coastline" of numbers. It is a Linear Fractal. It has a specific "dimension" (a measure of how "full" or "complex" it is) that is different from a simple line or a solid block.

Key Findings in Simple Terms

1. The "Frequency" Connection:
   The authors show a clever link between the average of the digits and how frequently each digit appears.

   • If you know how often 0s, 1s, and 2s appear, you can calculate the average.
   • Conversely, if you fix the average, you can figure out the possible combinations of frequencies for the digits.

2. The "Chaos" of the Function:
   They studied a function that tells you the average of a number's digits.

   • The Metaphor: Imagine a map where every point has a color representing its average. If you move your finger just a tiny bit on the map, the color might suddenly jump from "Red" to "Blue" to "Green" with no smooth transition.
   • The Finding: This function is everywhere discontinuous. It's like a landscape made entirely of cliffs. There are no smooth slopes; it's jagged everywhere.

3. The "Size" of the Sets:

   • If the average is 1 (the "Normal" case): The set of numbers is huge. It contains almost all the numbers on the line (in terms of standard size/measure).
   • If the average is anything else (e.g., 0.5): The set is "invisible" in terms of standard size (it has zero length), but it is infinite and complex. It's a "dust" of numbers that is so intricate it has a fractal dimension.

4. The "Super-Fractal" of Chaos:
   There is a special group of numbers where the average never settles down. It keeps fluctuating forever. The authors note that this group of "chaotic" numbers is actually the biggest and most complex fractal of all, filling the space in a way that standard math struggles to describe.

Why Does This Matter?

This isn't just about playing with numbers. This research helps us understand:

• Randomness: How true randomness behaves versus structured patterns.
• Complexity: How simple rules (like "digits must average to 0.5") can create incredibly complex, self-similar structures (fractals).
• Information Theory: How we encode information. If you try to compress data based on the frequency of digits, understanding these "weird" sets helps define the limits of what is possible.

In a Nutshell:
The paper takes a simple idea—averaging the digits of a number—and discovers that the collection of numbers sharing a specific average forms a beautiful, infinitely complex, and jagged "fractal" shape hidden inside the number line. It's a map of the hidden, chaotic, and beautiful structures that exist within the fabric of mathematics.

Technical Summary

Here is a detailed technical summary of the paper "Linear Fractals of the Besicovitch–Eggleston Type" by M. V. Pratsiovytyi and S. O. Klymchuk.

1. Problem Statement

The paper investigates the topological, metric, and fractal properties of subsets of the unit interval $[0, 1]$ defined by the asymptotic mean of digits in their ternary (3-adic) representation.

While classical fractal geometry (specifically Besicovitch–Eggleston sets) focuses on sets defined by the asymptotic frequencies of specific digits (e.g., the limit of the count of digit $i$ divided by the total length), this study shifts the focus to the asymptotic mean of the digits. For a number $x$ with ternary digits $\alpha_n \in \{0, 1, 2\}$, the asymptotic mean is defined as:
$$r(x) = \lim_{n \to \infty} \frac{1}{n} \sum_{i=1}^n \alpha_i(x)$$
The authors aim to characterize the set $S_a = \{x \in [0, 1] : r(x) = a\}$ for a given parameter $a \in [0, 2]$. Specifically, they analyze the relationship between the existence of digit frequencies and the existence of the asymptotic mean, and determine the Hausdorff–Besicovitch dimension of these sets.

2. Methodology

The authors employ a combination of number-theoretic analysis, measure theory, and fractal geometry techniques:

• Definitions and Conventions: They establish a unique ternary representation for all numbers by excluding expansions ending in infinite periods of $(2)$, using only those ending in $(0)$. This ensures the digit function $\alpha_n(x)$ is well-defined.
• Relationship Analysis: The core methodological step involves deriving algebraic relationships between the asymptotic mean $r(x)$ and the digit frequencies $\nu_i(x)$ (where $\nu_i(x) = \lim_{n \to \infty} \frac{N_i(x, n)}{n}$).
  • They prove that for ternary representations, if frequencies exist, $r(x) = \nu_1(x) + 2\nu_2(x)$.
  • They investigate the logical dependencies: if the frequency of one digit does not exist, does the mean exist? (Lemma 3 and Lemma 6).
• Set Decomposition: The set $S_a$ is decomposed into two disjoint subsets:
  1. $K_a$: Numbers where the frequencies of all digits exist.
  2. $M_a$: Numbers where the frequency of at least one digit does not exist (but the mean $a$ exists).
• Fractal Dimension Calculation: To find the Hausdorff–Besicovitch dimension ($\alpha_0$) of $K_a$, the authors utilize the known formula for Besicovitch–Eggleston sets and apply optimization techniques (Lagrange multipliers or calculus) to maximize the dimension function subject to the constraints imposed by the fixed mean $a$.
• Topological Proofs: They use constructive arguments involving "cylinders" (intervals defined by finite prefixes) to prove density and cardinality properties.

3. Key Contributions and Results

A. Relationship Between Mean and Frequencies

• Lemma 2: For ternary numbers, if all digit frequencies exist, the asymptotic mean is linearly related to the frequencies: $r(x) = \nu_1(x) + 2\nu_2(x)$.
• Lemma 6: A critical result establishing that if the asymptotic mean $r(x)$ exists and the frequency of digit 0 ($\nu_0$) exists, then the frequencies of digits 1 and 2 must also exist. Conversely, if $\nu_0$ does not exist but $r(x)$ does, then $\nu_1$ and $\nu_2$ cannot exist.
• Lemma 3: If the frequency of one digit in a ternary representation does not exist, the frequency of at least one other digit also fails to exist.

B. Properties of the Digit Frequency Function

• Theorem 2: The digit frequency function $\nu_i(x)$ is bounded on $[0, 1]$ but is discontinuous at every point in the interval.
• Lemma 5: Even at points where the frequency is defined, the limit of the frequency function as $x$ approaches that point does not exist. This highlights the highly irregular, "fractal-like" nature of the frequency function itself.

C. Analysis of the Set $K_a$ (Where Frequencies Exist)

The paper focuses primarily on the subset $K_a \subset S_a$ where all digit frequencies exist.

• Cardinality and Density: $K_a$ is a continuum (cardinality of the real numbers) and is everywhere dense in $[0, 1]$.
• Lebesgue Measure:
  • If $a = 1$, the set $K_1$ has full Lebesgue measure ($\lambda(K_1) = 1$). This is because $a=1$ corresponds to the set of normal numbers (where $\nu_0=\nu_1=\nu_2=1/3$), which constitute almost all real numbers.
  • If $a \neq 1$, the set $K_a$ has zero Lebesgue measure.
• Hausdorff–Besicovitch Dimension:
  The authors derive an explicit formula for the fractal dimension of $K_a$. By maximizing the entropy function subject to the constraints $\nu_0 + \nu_1 + \nu_2 = 1$ and $\nu_1 + 2\nu_2 = a$, they obtain:
  $$\alpha_0(K_a) = -\frac{1}{\ln 3} \ln \left[ \left(\frac{t-1}{3}\right)^{\frac{t-1}{3}} \left(\frac{7-3a-t}{6}\right)^{\frac{7-3a-t}{6}} \left(\frac{1+3a-t}{6}\right)^{\frac{1+3a-t}{6}} \right]$$
  where $t = \sqrt{1 + 6a - 3a^2}$.
  • Corollary 3: For the specific case $a=1$, the dimension is $\alpha_0(K_1) = 1$, consistent with the set having full measure.

D. The Set $M_a$ (Where Frequencies Do Not Exist)

The paper notes that the set of numbers where frequencies do not exist (but the mean might) forms a "superfractal" set with Hausdorff dimension 1, though the primary analytical focus remains on $K_a$.

4. Significance

• Extension of Classical Theory: This work extends the classical theory of Besicovitch–Eggleston sets (which are defined by fixed digit frequencies) to sets defined by the asymptotic mean of digits. This provides a new class of "linear fractals."
• Characterization of Singular Sets: It clarifies the structure of sets where statistical properties of digits behave differently from the "normal" case. It demonstrates that fixing the mean $a$ (where $a \neq 1$) creates a set of zero measure but potentially high fractal dimension, distinct from the set of normal numbers.
• Topological Insight: The proof that the frequency function is discontinuous everywhere and that its limit does not exist even at defined points reinforces the pathological nature of digit distribution functions, a key concept in metric number theory.
• Explicit Dimension Formula: The derivation of the closed-form expression for the Hausdorff dimension of $K_a$ provides a precise tool for quantifying the "size" of these fractal sets in terms of their parameter $a$.

In summary, the paper successfully characterizes the set of numbers with a prescribed asymptotic mean of ternary digits, proving that while these sets are topologically large (dense and continuum), they are metrically small (measure zero) unless the mean corresponds to the normal distribution, and providing a precise formula for their fractal dimension.