🔢 mathematics

A zero-one law for improvements to Dirichlet's theorem in arbitrary dimension

This paper establishes a zero-one law for the Lebesgue measure of the set of $\psi$ -Dirichlet matrices in arbitrary dimensions by removing a technical condition from prior work and employing a dynamical approach involving shrinking targets in the space of lattices, utilizing a new family of carefully chosen subsets and short-range mixing estimates.

Original authors: Andreas Strömbergsson, Shucheng Yu

Published 2026-02-20

📖 5 min read🧠 Deep dive

CC0 1.0

Original authors: Andreas Strömbergsson, Shucheng Yu

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ 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: The "Perfect Fit" Game

Imagine you are playing a game with an infinite grid of dots (like graph paper, but in many dimensions). You have a mysterious machine that spits out a random pattern of numbers (a matrix $A$ ).

The Original Rule (Dirichlet's Theorem):
The old rule says: "No matter what pattern your machine spits out, you can always find a pair of dots on the grid that are very close to the pattern." Specifically, if you look at a circle of radius $t$ , you can find a dot within a distance of $1/t$ .

The Challenge:
Mathematicians asked: "Can we do better? Can we find dots that are even closer?"
Instead of a distance of $1/t$ , can we find dots within a distance of $\psi(t)$ , where $\psi(t)$ is a tiny, shrinking function (like $1/t^2$ or $1/(t \log t)$ )?

The answer depends on the "pattern" (the matrix $A$ ). Some patterns are "lucky" and allow for these super-precise fits. Others are "unlucky" and stick to the original rule.

The Main Question: Is it All or Nothing?

The paper asks a fundamental question about the "lucky" patterns:
If we pick a pattern completely at random (like picking a point in a vast ocean), what are the odds it is "lucky"?

The Zero-One Law: The authors prove that the answer is binary. The probability is either 0% (almost no patterns are lucky) or 100% (almost all patterns are lucky). There is no "50-50" middle ground.

The only thing that decides which side you fall on is a specific mathematical formula involving the "tightness" of the rule ( $\psi$ ). If the formula adds up to a finite number, you get 100%. If it adds up to infinity, you get 0%.

The Problem with Previous Research

Before this paper, mathematicians (Kleinbock, Shi, and Yu) had already solved this for many cases, but they had a "technical crutch." They had to assume that the "tightness" of the rule behaved in a very smooth, predictable way (like a ball rolling down a smooth hill).

If the rule behaved erratically—wiggling up and down a bit while generally going down—the old math broke down. They couldn't prove the Zero-One Law for those "wiggly" rules.

The Authors' Solution: A New Way to Look at the Grid

Strömbergsson and Yu (the authors) removed that crutch. They proved the law holds even if the rule is "wiggly" (as long as it doesn't wiggle too wildly).

Here is how they did it, using analogies:

1. The "Shrinking Target" Analogy

Imagine a dartboard where the bullseye is shrinking every second.

The Dart: Represents the mathematical pattern (the matrix).
The Bullseye: Represents the "lucky" condition (finding a dot that fits the tight rule).
The Goal: Does the dart hit the bullseye infinitely many times as it shrinks?

The authors needed to calculate how often a random dart hits these shrinking targets.

2. The "Crowded Room" Problem (The Old Way)

In the previous research, the "bullseyes" were like large, fuzzy blobs. When the blobs shrank, they sometimes overlapped in complicated ways. To prove the Zero-One Law, the authors had to assume the blobs shrank in a perfectly smooth way so they could calculate the overlaps. If the shrinking was jerky, the overlaps became impossible to calculate.

3. The "Laser Pointer" Innovation (The New Way)

The authors' key innovation was to change the shape of the bullseye. Instead of a fuzzy blob, they defined a very specific, narrow "laser beam" target.

Why? They realized that if you aim for a very specific, narrow slice of the grid, these targets become disjoint (they don't overlap) for a much longer time, even if the shrinking is jerky.
The Metaphor: Imagine trying to catch raindrops in a bucket. If the bucket is wide and wobbly, it's hard to predict when it will catch a drop. But if you use a narrow, rigid tube, you know exactly when a drop will pass through, even if the rain is falling erratically.

By using these "narrow tubes" (mathematically called subsets of the shrinking targets), they could prove that the targets don't interfere with each other. This allowed them to use a powerful statistical tool called Mixing.

4. The "Mixing" Magic

In physics, "mixing" is like stirring milk into coffee. Eventually, the milk and coffee are so well-mixed that if you take a spoonful, it represents the whole cup perfectly.

In this math paper, the "mixing" theorem says that if you wait long enough, the random patterns (darts) become so thoroughly scrambled that hitting a target at time $T$ has almost nothing to do with hitting a target at time $T+100$ .

Because the authors' "narrow tube" targets don't overlap, they could prove that the "mixing" happens fast enough to ignore the "wiggles" in the rule. This allowed them to calculate the total probability without needing the smoothness assumption.

The Conclusion

The paper is a triumph of precision.

Old View: "We can only predict the outcome if the rules are smooth and predictable."
New View: "We can predict the outcome even if the rules are a bit messy, as long as we look at the problem through a sharper, more specific lens."

They proved that for almost any dimension (2D, 3D, or 100D), the universe of numbers follows a strict Zero-One law: either the "perfect fit" is possible for almost everyone, or it's impossible for almost everyone. There is no middle ground.

In short: They took a complex, high-dimensional math problem, removed a restrictive assumption that limited previous solutions, and used a clever geometric trick (narrowing the target) to show that the fundamental laws of randomness hold true even in messy situations.

1. Problem Statement

The paper addresses the metric theory of Diophantine approximation, specifically focusing on improvements to Dirichlet's Theorem in arbitrary dimensions ( $m \times n$ matrices).

Classical Context: Dirichlet's Theorem states that for any real matrix $A \in M_{m,n}(\mathbb{R})$ and any $t > 1$ , there exist integer vectors $p \in \mathbb{Z}^m$ and $q \in \mathbb{Z}^n \setminus \{0\}$ such that $\|Aq - p\| < 1/t$ and $\|q\| < t$ .
The Question: Can this bound be improved? Specifically, for a given decreasing function $\psi(t)$ , what is the Lebesgue measure of the set of matrices $A$ for which the system $\|Aq - p\| < \psi(t)$ and $\|q\| < t$ has solutions for all sufficiently large $t$ ?
The Gap: Previous work by Kleinbock, Shi, and Yu [KSY22] established a partial zero-one law for this problem. They proved that the measure is either 0 or 1 based on the convergence or divergence of a specific series involving $\psi$ $ψ$ . However, their result required two technical conditions:
1. A monotonicity assumption on $t\psi(t)$ (specifically, that $t \mapsto t\psi(t)$ is increasing).
2. A technical condition on the partial sums of the series (Condition 1.8 in the paper) to handle the divergence case.
Goal: The authors aim to remove the technical condition (1.8) and weaken the monotonicity assumption, thereby establishing a complete zero-one law for a broader class of approximating functions.

2. Methodology

The proof relies on the homogeneous dynamics approach, specifically the Dani correspondence, which translates the Diophantine approximation problem into a shrinking target problem on the space of unimodular lattices $X_d = SL_d(\mathbb{R})/SL_d(\mathbb{Z})$ (where $d=m+n$ ).

Key Dynamical Tools:

Dani Correspondence: Relates the existence of good approximations for a matrix $A$ $A$ to the trajectory of the lattice $\Lambda_A$ $Λ_{A}$ under the action of a one-parameter diagonal subgroup $g_s$ $g_{s}$ .
- $A$ is $\psi$ -Dirichlet if and only if the trajectory $g_s \Lambda_A$ avoids certain "shrinking target" sets for all large $s$ .
Effective Equidistribution and Mixing: The authors utilize effective equidistribution and doubly mixing estimates for the action of $g_s$ on the submanifold of lattices corresponding to matrices. These estimates (from Kleinbock-Margulis and Kleinbock-Shi-Weiss) provide error bounds for the correlation of functions along the trajectory, crucial for applying probabilistic limit theorems.
Borel-Cantelli Lemmas:
- Convergence Case: Uses the standard Borel-Cantelli lemma. If the sum of measures of the target sets converges, the measure of matrices hitting the targets infinitely often is zero.
- Divergence Case: Requires a divergent Borel-Cantelli lemma. This typically requires proving that the hitting events are "quasi-independent" (i.e., the correlation between events decays sufficiently fast).

3. Key Contributions and Innovations

The paper introduces several novel technical ingredients to overcome the limitations of [KSY22]:

A. Relaxation of Monotonicity

Instead of requiring $t\psi(t)$ to be strictly increasing, the authors introduce a "quasi-decreasing" condition (Condition 1.9). This allows the function $F_\psi(t) = 1 - t\psi(t)$ to fluctuate within a controlled range over logarithmic intervals. This is achieved by defining a new auxiliary function $r(s)$ that tracks the shrinking rate and proving it satisfies a similar quasi-monotonicity property.

B. Removal of the Technical Condition (1.8)

The most significant contribution is the removal of the condition (1.8) required for the divergence case. In [KSY22], this condition was necessary to ensure the targets were sufficiently "spread out" to guarantee independence.

New Strategy: The authors construct a carefully chosen subset of the shrinking targets, denoted $\Delta'_r$ .
Definition of $\Delta'_r$ : These are subsets of the standard critical locus $\Delta^{-1}[0, r]$ intersected with a specific geometric region $H(R_r)$ , where $R_r$ is a thin rectangular box in $\mathbb{R}^d$ .
Disjointness Property (Proposition 4.4): The authors prove a crucial disjointness result: For a specific range of time steps $k$ $k$ , the sets $g^k \Delta'_r$ $g^{k} Δ_{r}^{'}$ are pairwise disjoint.
- This disjointness is stronger than what was available for the full set $\Delta^{-1}[0, r]$ .
- It allows the authors to bound the correlation terms in the variance calculation (the "second moment") without needing the restrictive Condition (1.8).

C. Smooth Approximations and Measure Estimates

The authors provide precise measure estimates for the new targets $\Delta'_r$ and their thickened versions $\tilde{\Delta}'_r$ . They prove that the measure of these new sets is comparable to the original sets (Theorem 2.2) but possess better geometric properties for the mixing estimates. They also construct smooth approximations ( $C^\infty_c$ functions) for these sets with controlled norms, which are essential for applying the effective mixing theorems.

4. Main Results

Theorem 3 (Main Theorem):
Let $\psi$ be a continuous, decreasing function satisfying $\psi(t) < 1/t$ and the quasi-decreasing condition (1.9). Let $F_\psi(t) = 1 - t\psi(t)$ . Then the Lebesgue measure of the set of $\psi$ -Dirichlet matrices, denoted $DI_{\alpha,\beta}(\psi)$ , satisfies a zero-one law:

$\text{Leb}(DI_{\alpha,\beta}(\psi)) = \begin{cases} 1 & \text{if } \sum_{k} k^{-1} F_\psi(k)^{\kappa_d} \log^{\lambda_d}\left(1 + \frac{1}{F_\psi(k)}\right) < \infty \\ 0 & \text{if } \sum_{k} k^{-1} F_\psi(k)^{\kappa_d} \log^{\lambda_d}\left(1 + \frac{1}{F_\psi(k)}\right) = \infty \end{cases}$

Where:

$\kappa_d = \frac{d^2+d-4}{2}$
$\lambda_d = \frac{d(d-1)}{2}$
$d = m+n$

Significance of the Series:
The series determines the measure. If it converges, almost no matrices are $\psi$ -Dirichlet (measure 0). If it diverges, almost all matrices are $\psi$ -Dirichlet (measure 1). The authors show that under their weaker assumptions, the divergence of this series is sufficient to guarantee measure 1, without the extra partial sum condition required in [KSY22].

5. Significance and Impact

Completeness of the Theory: This paper completes the metric theory for improvements to Dirichlet's theorem in arbitrary dimensions. It resolves the open problem of removing the technical condition (1.8), which was a barrier to a full zero-one law.
Generalization: The result applies to weighted Diophantine approximation (using quasi-norms defined by weight vectors $\alpha, \beta$ ), covering the standard case ( $m=n=1$ ) and higher dimensions.
Methodological Advancement: The introduction of the specific subsets $\Delta'_r$ and the proof of their disjointness under the flow $g_s$ represents a significant technical innovation in the field of homogeneous dynamics. It demonstrates how refining the geometry of the target sets can bypass difficult analytic conditions.
Robustness: By weakening the monotonicity assumption to a "quasi-decreasing" condition, the theorem applies to a wider class of functions $\psi$ , including those that may oscillate or have flat regions, which are common in number-theoretic applications.

In summary, Strömbergsson and Yu have successfully generalized the zero-one law for Dirichlet improvements by combining refined geometric constructions of shrinking targets with powerful dynamical mixing estimates, effectively removing the last major technical obstruction in the field.