🔢 mathematics

Khintchine dichotomy and Schmidt estimates for self-similar measures on $\mathbb{R}^d$

This paper extends the classical Khintchine and Schmidt theorems in metric Diophantine approximation to self-similar measures on $\mathbb{R}^d$ by establishing effective equidistribution of associated random walks on $\text{SL}_{d+1}(\mathbb{R})/\text{SL}_{d+1}(\mathbb{Z})$ through novel techniques involving a bootstrap scheme and refined Dani's correspondence.

Original authors: Timothée Bénard, Weikun He, Han Zhang

Published 2026-02-24

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Timothée Bénard, Weikun He, Han Zhang

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

The Big Picture: Finding Patterns in Chaos

Imagine you are standing in a vast, foggy forest. You want to know if you can find a specific type of tree (let's call it a "Rational Tree") that is very close to where you are standing.

In mathematics, this is called Diophantine approximation. It asks: How well can we approximate a random point in space using simple fractions (rational numbers)?

For a long time, mathematicians knew the answer for "normal" points (like points on a smooth sheet of paper). This is the famous Khintchine Theorem. It says:

If the "target size" (how close you need to be) shrinks very fast, you will almost never hit a Rational Tree.
If the target size shrinks slowly, you will hit them infinitely often.

The Problem: What if you aren't standing on a smooth sheet of paper? What if you are standing on a fractal?
Think of a fractal like the Cantor Set (a line with holes punched out of it) or a Sierpiński Triangle. These shapes are jagged, self-repeating, and "rough." They are very different from a smooth sheet. For decades, mathematicians wondered: Does the Khintchine rule still apply to these weird, jagged shapes?

The Solution: This paper, by Bénard, He, and Zhang, says YES. They proved that even on these complex, self-similar fractals, the same rules apply. If the target is big enough, you will hit rational points infinitely often; if it's too small, you won't.

The Main Characters and Tools

To solve this, the authors had to build a new set of tools. Here is how they did it, using analogies:

1. The Self-Similar Measure (The "Fractal Dust")

Imagine a machine that takes a shape, shrinks it, rotates it, and scatters copies of it everywhere. If you repeat this forever, you get a "fractal dust."

The Math: This is a self-similar measure. It's a way of saying, "If I pick a random point from this fractal, where is it likely to be?"
The Goal: They wanted to know if these random points behave like normal points when it comes to finding rational approximations.

2. The "Magic Elevator" (Dani's Correspondence)

This is the most brilliant part of the paper. The authors use a trick discovered by mathematician S.G. Dani.

The Analogy: Imagine you have a point $s$ on your fractal. You want to know if it's close to a rational number. Instead of looking at the number directly, you put it on an elevator that moves up and down a very tall, strange building (a mathematical space called $SL_{d+1}(\mathbb{R})/SL_{d+1}(\mathbb{Z})$ ).
The Magic:
- If the point $s$ is far from all rational numbers, the elevator stays low in the building.
- If the point $s$ is very close to a rational number, the elevator shoots high up into the sky (into a "cusp").
The Strategy: Instead of counting fractions, the authors just watch the elevator. If they can prove the elevator spends most of its time in the "middle" of the building (not too high, not too low), they know the point is behaving normally.

3. The Random Walk (The "Drunkard's Stroll")

To prove the elevator behaves, they modeled the movement of the fractal points as a random walk.

The Analogy: Imagine a drunk person walking through the building. Every step they take is determined by the rules of the fractal (shrinking and rotating).
The Challenge: In 1D (a line), this walk is easy to predict. But in higher dimensions (2D, 3D, etc.), the walk gets stuck in "algebraic traps." It's like the drunk person keeps walking in circles around a specific pillar and never explores the rest of the building.
The Breakthrough: The authors developed a "Bootstrap Scheme."
- Step 1: Show the walk moves a little bit (gains a tiny bit of "dimension").
- Step 2: Use a clever "multislicing" technique (cutting the building into thin slices) to prove that even if the walk gets stuck, it eventually breaks free and spreads out evenly across the whole building.
- Step 3: Repeat this until the walk is so spread out that it looks like a uniform cloud of fog.

4. The "Siegel Transform" (The "Lattice Counter")

When the elevator goes high, they need to count how many rational points are nearby. They use a tool called the Siegel Transform.

The Analogy: Imagine a net with holes. You throw a ball (the rational point) at the net. The Siegel Transform counts how many balls get caught in the net.
The Difficulty: In 1D, the net is easy to count. In higher dimensions, the net gets messy, and the number of balls can explode. The authors had to invent a way to "trim" the net (truncate the transform) so they could count the balls without the numbers going to infinity.

The Two Main Results

The paper proves two specific things about these fractal points:

The Dichotomy (The "All or Nothing" Rule):
Just like on a smooth sheet of paper, if you set a target size for rational approximations:
- If the sum of the target sizes is finite (they get tiny very fast), the fractal points will never hit the target (probability 0).
- If the sum is infinite (they stay big enough), the fractal points will hit the target infinitely often (probability 1).
- Why this matters: It confirms that fractals, despite their jagged nature, follow the same fundamental laws of randomness as smooth shapes.
The Schmidt Estimate (The "Counting" Rule):
Not only do they hit the target, but the authors can also predict exactly how many times they hit it.
- They proved that the number of hits grows at the exact same rate as the "volume" of the target area.
- Analogy: If you are throwing darts at a moving target, they can tell you exactly how many darts will hit, not just "some" or "many."

Why This Matters

Before this paper, we had to assume the fractal was "nice" (simple, 1-dimensional, or had specific symmetries) to prove these rules. This paper removes all those restrictions.

It works for any dimension (2D, 3D, 100D).
It works for any self-similar fractal (even the messy ones).
It works for any target size function.

In Summary:
The authors took a chaotic, jagged fractal, put it on a magical elevator, watched it dance around a complex building using a random walk, and proved that despite its weird shape, it follows the same predictable rules of chance as a smooth, boring line. They solved a 40-year-old mystery by building a bridge between the jagged world of fractals and the smooth world of classical number theory.

1. Problem Statement and Context

The Core Problem:
The paper addresses the metric Diophantine approximation properties of points distributed according to self-similar measures on $\mathbb{R}^d$ . Specifically, it seeks to generalize the classical Khintchine Theorem and Schmidt's Counting Theorem from the Lebesgue measure to these fractal measures.

Khintchine's Theorem (Classical): For the Lebesgue measure, the set of $\psi$ -approximable vectors $W(\psi)$ (vectors $s$ such that $\|qs - p\| < \psi(q)$ for infinitely many $(p,q)$ ) has either zero or full measure. This zero-one law depends solely on the convergence or divergence of the series $\sum \psi(q)^d$ .
Schmidt's Theorem (Classical): In the divergent case, Schmidt provided an asymptotic formula for the number of solutions to the Diophantine inequality up to a bound $N$ .
The Challenge: While these results were known for Lebesgue measure and certain "friendly" measures, extending them to general self-similar measures (e.g., Cantor measures, Sierpiński gaskets) in arbitrary dimensions $d \ge 1$ $d \geq 1$ has been a major open problem. Previous works (e.g., Khalil-Luethi, Simmons-Weiss) required restrictive conditions such as:
- Dimension $d=1$ .
- Rational IFS with equal contraction ratios ("carpet" type).
- Large dimension assumptions.
- Restrictions on the approximation function $\psi$ (e.g., $\psi(q) < 1/q$ ).

The Goal: To establish the Khintchine dichotomy and Schmidt's counting estimates for arbitrary irreducible self-similar measures on $\mathbb{R}^d$ with finite exponential moments, for any non-increasing approximation function $\psi$ , and in arbitrary dimension $d$ .

2. Methodology and Technical Framework

The authors employ a dynamical systems approach, leveraging the connection between Diophantine approximation and the dynamics of flows on homogeneous spaces.

A. Dani's Correspondence

The paper utilizes Dani's correspondence principle, which translates the Diophantine properties of a vector $s \in \mathbb{R}^d$ into the behavior of the trajectory $(a(t)u(s)x_0)_{t>1}$ in the homogeneous space $X = \text{SL}_{d+1}(\mathbb{R}) / \text{SL}_{d+1}(\mathbb{Z})$ .

$a(t)$ is a diagonal flow expanding in the first $d$ directions and contracting in the last.
$u(s)$ is a unipotent translation.
The condition $\|qs - p\| < \psi(q)$ corresponds to the lattice $a(t)u(s)\mathbb{Z}^{d+1}$ intersecting a specific shrinking target set.

B. Effective Equidistribution of Expanding Fractals

The central technical achievement is proving an effective equidistribution theorem for the pushforward of the self-similar measure $\sigma$ under the flow $a(t)u(s)$ .

Theorem 1.2: For a self-similar measure $\sigma$ , the integral $\int f(a(t)u(s)x) d\sigma(s)$ converges to the Haar measure average $m_X(f)$ with an exponential rate of decay $O(t^{-c})$ , with an error term depending on the injectivity radius of the base point $x$ .
Connection to Random Walks: The authors relate the expanding translates $a(t)u(s)d\sigma(s)$ to the $n$ -step distribution of a random walk on $X$ driven by a measure $\mu$ derived from the IFS (Lemma 8.2). This reduces the problem to studying the equidistribution of random walks.

C. Dimensional Bootstrap and Non-Concentration

To prove the equidistribution of the random walk, the authors must show that the measure acquires "high dimension" (approaching the dimension of the ambient space $X$ ) under convolution.

The Obstacle (Algebraic): In dimensions $d \ge 2$ , standard multislicing techniques (used in $d=1$ ) fail due to algebraic obstructions. Specifically, the random subspaces generated by the walk do not satisfy the strong non-concentration hypothesis required by previous projection theorems (Lemma 7.3).
The Solution (MNC): The authors introduce a Mild Non-Concentration (MNC) property. They prove that while strong non-concentration fails, a weaker form holds, sufficient to drive the dimension up.
- Regularity of Self-Similar Measures: A key intermediate result (Section 4) establishes that self-similar measures are Hölder regular near algebraic subvarieties (Theorem 4.4). This means the measure does not concentrate near proper algebraic sets, which is crucial for the bootstrap.
- Bootstrap Scheme: Using the MNC property and linear multislicing, they iteratively increase the dimension of the measure until it is close to $\dim X$ .
Spectral Gap: Once the measure has high dimension, the spectral gap property of the convolution operator on $L^2(X)$ ensures exponential equidistribution (Theorem 1.3).

D. Handling Siegel Transforms

To derive the counting estimates (Schmidt's theorem), the authors analyze Siegel transforms (sums over lattice points).

Challenge: For $d=1$ , the Siegel transform of a ball is not in $L^2$ , making standard variance estimates fail.
Resolution:
- For $d \ge 2$ , they use standard moment estimates.
- For $d=1$ , they introduce restricted Siegel transforms counting only lattice points with bounded GCD (Proposition 9.17). They prove that the contribution of non-primitive points is negligible asymptotically, allowing them to recover the full counting result.

3. Key Contributions and Results

Main Theorem (Theorem 1.1)

Let $\sigma$ be an irreducible self-similar measure on $\mathbb{R}^d$ with finite exponential moment. Let $\psi: \mathbb{N} \to \mathbb{R}_{\ge 0}$ be non-increasing.

Khintchine Dichotomy:
$\sigma(W(\psi)) = \begin{cases} 0 & \text{if } \sum \psi(q)^d < \infty \\ 1 & \text{if } \sum \psi(q)^d = \infty \end{cases}$
Schmidt Counting (Divergent Case): If the series diverges, for $\sigma$ -almost every $s$ :
$\left| \{ (p, q) \in \mathbb{Z}^d \times [0, N] : \|qs - p\|_\infty < \psi(q) \} \right| \sim_{s, \psi} 2^d \sum_{q=0}^N \psi(q)^d$
(A similar asymptotic holds for primitive solutions).

Key Technical Results

Theorem 1.2 (Effective Equidistribution): Establishes exponential decay of correlations for expanding translates of self-similar measures on $X$ , with explicit dependence on the injectivity radius.
Theorem 1.3 (Effective Equidistribution of Random Walks): Proves exponential convergence of the $\mu$ -random walk to the Haar measure.
Theorem 4.4 (Hölder Regularity): Proves that self-similar measures do not concentrate near algebraic subvarieties of $\mathbb{R}^d$ . This is a result of independent interest in fractal geometry.
Dimensional Bootstrap (Section 7): Develops a new scheme to overcome algebraic obstructions in $d \ge 2$ by utilizing the "Mild Non-Concentration" property, extending the multislicing method to higher dimensions.

4. Significance and Impact

Resolution of a Long-Standing Conjecture: The paper provides a complete affirmative answer to the question of whether self-similar measures satisfy the Khintchine dichotomy, removing all previous restrictions on dimension, the type of IFS (rational/irrational, equal/unequal ratios), and the approximation function $\psi$ .
Generalization of Schmidt's Theorem: It extends Schmidt's quantitative counting estimates to the fractal setting, a significant step beyond the qualitative "zero-one" laws.
Methodological Breakthrough:
- The development of the Mild Non-Concentration (MNC) framework allows for the application of dimension-boosting techniques in settings where strong non-concentration fails due to algebraic geometry.
- The Hölder regularity near subvarieties (Theorem 4.4) is a powerful new tool for analyzing the distribution of self-similar measures, applicable to other problems in Diophantine approximation on fractals.
Unification: The work unifies the treatment of $d=1$ and $d \ge 2$ , resolving the specific difficulties of the $d=1$ case (infinite second moment of Siegel transforms) and the $d \ge 2$ case (algebraic obstructions) within a single coherent framework.

In summary, this paper represents a major advancement in the field of metric Diophantine approximation on fractals, combining deep techniques from homogeneous dynamics, probability theory, and algebraic geometry to solve a fundamental problem in the field.