Dvoretzky Covering Problem

• Dvoretzky covering problem is a probabilistic model that examines when a space—typically the unit circle—is fully covered by random intervals based on prescribed lengths and statistical laws.
• Extensions involve general measures and higher-dimensional settings, using potential theory and capacity estimates to establish sharp coverage thresholds.
• Martingale and harmonic analysis techniques reveal that the structure of non-covered sets is intricately linked to Fourier asymptotics and multifractal properties.

The Dvoretzky covering problem addresses the probabilistic covering of sets—most notably, the unit circle—by randomly placed intervals with prescribed lengths and statistical laws governing their centers. Initially posed by A. Dvoretzky, the question is: for which sequences of lengths, and under what statistical framework for the placement of intervals, is every point of a space almost surely covered infinitely often or at least once? The problem links extremal probability, Fourier analysis, potential theory, and multifractal geometry, and has seen extensions from the uniform circle model to arbitrary measures and spaces of higher dimension.

1. Classical Dvoretzky Covering Model and Shepp’s Criterion

In the canonical setting, the space is $\mathbb{T} = \mathbb{R}/\mathbb{Z}$, the unit circle. Let $(\omega_n)_{n=1}^\infty$ be i.i.d. random variables, each uniform on $\mathbb{T}$, denoting arc centers. Given a fixed, decreasing sequence $(\ell_n)$ with $\ell_n \to 0$, define random arcs $I_n := (\omega_n,\,\omega_n+\ell_n)$ (mod 1).

A central object is the limsup set: $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$

The core question is: Under what conditions on $(\ell_n)$ does

$$\mathbb{P}\left( \mathbb{T} = \limsup I_n \right) = 1$$

hold? Shepp (1972) gave a sharp necessary and sufficient condition: $$\sum_{n=1}^\infty n^{-2} \exp\left(\ell_1 + \cdots + \ell_n\right) = \infty \iff \text{full covering a.s.}$$ When this condition fails, a random exceptional set $(\omega_n)_{n=1}^\infty$0 remains, whose properties have been the subject of intense analysis (Tan, 12 Nov 2025, Fan et al., 2019).

2. Extensions to General Measures and Geometric Sets

The model has evolved to include non-uniform laws for centers and to address covering problems on the real line and higher dimensions. The law of arc centers can be any Borel probability measure $(\omega_n)_{n=1}^\infty$1, not necessarily uniform. In such general settings, the random covering set is

$(\omega_n)_{n=1}^\infty$2

with $(\omega_n)_{n=1}^\infty$3 a sequence of radii and $(\omega_n)_{n=1}^\infty$4 i.i.d. from $(\omega_n)_{n=1}^\infty$5.

Anttila and Myllyoja (Anttila et al., 16 Jan 2026) provide a comprehensive solution for arbitrary analytic target sets $(\omega_n)_{n=1}^\infty$6 and arbitrary Borel measures:

• Define $(\omega_n)_{n=1}^\infty$7 ("thin points").
• Introduce a capacity $(\omega_n)_{n=1}^\infty$8 based on the energy

$(\omega_n)_{n=1}^\infty$9

Main theorem: $\mathbb{T}$0 This both generalizes and sharpens the potential-theoretic criterion by Kahane for the uniform case.

3. Fine Structure and Harmonic Analysis of the Exceptional Set

If Shepp’s covering criterion is not met, the residual non-covered set $\mathbb{T}$1 exhibits intricate harmonic and geometric structure (Tan, 12 Nov 2025). Notably:

• When $\mathbb{T}$2 ($\mathbb{T}$3), almost surely $\mathbb{T}$4.
• The work (Tan, 12 Nov 2025) establishes that for a large class of $\mathbb{T}$5, the set $\mathbb{T}$6 supports a natural multiplicative chaos measure $\mathbb{T}$7. This measure is constructed as the weak a.s. limit of product-martingale densities

$\mathbb{T}$8

The measure $\mathbb{T}$9 is supported on $(\ell_n)$0 and is absolutely continuous under suitable convolution powers.

• Crucially, $(\ell_n)$1 is a Rajchman measure, i.e., its Fourier coefficients vanish at infinity,

$(\ell_n)$2

so $(\ell_n)$3 is a set of multiplicity ($(\ell_n)$4-set); if $(\ell_n)$5 is continuous, $(\ell_n)$6 is an $(\ell_n)$7-set. This connects probabilistic covering phenomena to questions of uniqueness in Fourier series.

4. Potential Theory, Capacity, and Critical Exponents

Potential-theoretic methods are central to all modern solutions. Kahane’s $(\ell_n)$8-capacity for a kernel

$(\ell_n)$9

governs, for probability measure $\ell_n \to 0$0,

$\ell_n \to 0$1

A compact $\ell_n \to 0$2 is covered a.s. iff $\ell_n \to 0$3 (Anttila et al., 16 Jan 2026).

For general measures, the capacity is replaced by the generalized $\ell_n \to 0$4 energy. The "thin point" contribution appears—if $\ell_n \to 0$5 sees only finitely many overlapping balls in the second moment sense, then $\ell_n \to 0$6 is not covered.

A key implication is that for polynomially decaying radii $\ell_n \to 0$7, there is a sharp phase transition:

• If $\ell_n \to 0$8, full measure is covered a.s.
• If $\ell_n \to 0$9, covering fails a.s.
• At the threshold $I_n := (\omega_n,\,\omega_n+\ell_n)$0 the outcome is dictated by the constant $I_n := (\omega_n,\,\omega_n+\ell_n)$1 and multifractal structure—specifically, the spectrum of average densities (Anttila et al., 16 Jan 2026).

5. Coverings with Non-Uniform Laws: Sharp Thresholds and Genericity

The extension to non-uniform laws for arc centers employs the essential infimum $I_n := (\omega_n,\,\omega_n+\ell_n)$2 of a density $I_n := (\omega_n,\,\omega_n+\ell_n)$3 and the set $I_n := (\omega_n,\,\omega_n+\ell_n)$4 of points on which $I_n := (\omega_n,\,\omega_n+\ell_n)$5 attains this value. The principal threshold, for $I_n := (\omega_n,\,\omega_n+\ell_n)$6 and $I_n := (\omega_n,\,\omega_n+\ell_n)$7, is: $I_n := (\omega_n,\,\omega_n+\ell_n)$8 This extends Shepp’s criterion and shows that the box-counting dimension of $I_n := (\omega_n,\,\omega_n+\ell_n)$9 governs the transition: when $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$0 is "small," non-uniformity can dramatically lower the threshold for covering (Hirayama et al., 2021, Fan et al., 2019).

The Menshov-type theorem further demonstrates that, except in the uniform case, one can force covering by arbitrarily small perturbations of $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$1 if $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$2, indicating a strong form of genericity in the critical regime (Hirayama et al., 2021).

6. High-Dimensional and Convex Geometric Analogues

Dvoretzky-type covering phenomena appear in high-dimensional convex geometry, notably in questions of covering the unit ball $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$3 of a Banach space by finitely many closed convex sets (Raja, 2024). Here, principal results include:

• For fixed $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$4, any covering of $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$5 in sufficiently high dimension contains, among its pieces, an $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$6-dimensional ball of radius arbitrarily close to 1.
• In infinite-dimensional $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$7, given mild translation conditions, one finds an infinite-dimensional ball of radius $$\limsup_n I_n = \{ t \in \mathbb{T} : t \in I_n \text{ for infinitely many } n \}.$$8 in one of the covering sets.

These structural insights extend the reach of Dvoretzky’s classical theorem on almost Euclidean sections and tie covering problems in probability to the geometry of Banach spaces.

7. Proof Methodologies and Martingale Techniques

A consistent theme in both classical and modern work is the use of second-moment (Billard) methods and martingale convergence, now refined with products of survival indicators and multiplicative chaos tools. Weak convergence arguments, energy/capacity estimates, and measurable-selection principles (e.g., Jankov–von Neumann uniformization in (Anttila et al., 16 Jan 2026)) are employed to rigorously establish sufficiency and necessity of coverage criteria. In harmonic settings, convolution powers and Fourier asymptotics validate singularity or Rajchman properties essential for multiplicity assertions (Tan, 12 Nov 2025).

---

The Dvoretzky covering problem, from its probabilistic-geometric origins, now embodies a rich interplay between probability, real and harmonic analysis, fractal geometry, and potential theory. Its modern complete characterizations for general measures and analytic sets illuminate both probabilistic covering and deep structural properties of exceptional sets (Anttila et al., 16 Jan 2026, Tan, 12 Nov 2025, Hirayama et al., 2021, Fan et al., 2019, Raja, 2024).