[Paper Review] Fuglede's conjecture is false in 5 and higher dimensions

Terence Tao|ArXiv.org|Jun 8, 2003

Mathematical Analysis and Transform Methods12 references4 citations

TL;DR

This paper disproves Fuglede's conjecture in dimensions 5 and higher by constructing a finite union of unit cubes in ℝ⁵ that admits an orthonormal basis of exponentials (i.e., is spectral) but does not tile ℝ⁵ by translation. The counterexample is derived from a finite abelian group counterexample using Hadamard matrices in ℤ₂¹² and ℤ₃⁶, then transferred to ℝ⁵ via transference principles, showing that spectral sets and translational tiles are not equivalent in high dimensions.

ABSTRACT

We give an example of a set $Ω\subset \R^5$ which is a finite union of unit cubes, such that $L^2(Ω)$ admits an orthonormal basis of exponentials $\{\frac{1}{|Ω|^{1/2}} e^{2πi ξ_j \cdot x}: ξ_j \in Λ\}$ for some discrete set $Λ\subset \R^5$, but which does not tile $\R^5$ by translations. This answers a conjecture of Fuglede in the negative, at least in 5 and higher dimensions.

Motivation & Objective

• To disprove Fuglede's conjecture, which posits that a set is spectral (admits an orthonormal basis of exponentials) if and only if it tiles ℝⁿ by translation.
• To construct a concrete counterexample in ℝⁿ for n ≥ 5 where a set is spectral but does not tile, thereby refuting the conjecture in high dimensions.
• To extend finite abelian group counterexamples (in ℤ₂¹² and ℤ₃⁶) to the continuous setting ℝ⁵ via transference techniques.
• To clarify the limitations of the spectral set conjecture and identify conditions under which it may still hold, such as in lower dimensions or under convexity constraints.

Proposed method

• Construct a finite set Ω₀ ⊂ ℤ₂¹² as the standard basis vectors, which does not tile ℤ₂¹² because 12 does not divide 2¹².
• Use a 12×12 Hadamard matrix to define a spectrum Λ₀ ⊂ ℤ₂¹² such that the exponentials {e²πiξ·x / √12} form an orthonormal basis for l²(Ω₀).
• Transfer the finite group counterexample to ℤ⁵ by using a 6-element subset Ω₁ ⊂ ℤ₃⁵ that is spectral but does not tile ℤ⁵ due to the non-divisibility of 6 into 3⁵.
• Apply a transference argument to lift the discrete counterexample in ℤ⁵ to ℝ⁵ by defining Ω₂ = Ω₁ + [0,1)⁵ and Λ₂ = Λ₁ + ℤ⁵, preserving the spectral property.
• Prove that Ω₂ does not tile ℝ⁵ by contradiction: assuming a tiling leads to a cardinality mismatch in local density estimates, exploiting boundary effects and cube covering arguments.
• Use measure-theoretic and combinatorial estimates involving annuli and cube overlaps to show that the local density of Ω₂ is 6/3⁵ = 2/81, which is incompatible with tiling due to irrationality in the density ratio.

Experimental results

Research questions

• RQ1 Does every set that admits an orthonormal basis of exponentials (i.e., is spectral) necessarily tile ℝⁿ by translation, as Fuglede's conjecture claims?
• RQ2 Can a finite union of unit cubes in ℝⁿ be spectral without being a translational tile, particularly in dimensions n ≥ 5?
• RQ3 What role do Hadamard matrices over finite fields play in constructing counterexamples to the spectral set conjecture in finite abelian groups?
• RQ4 To what extent can finite group counterexamples be transferred to the continuous setting ℝⁿ via transference principles?
• RQ5 Is Fuglede’s conjecture still valid in lower dimensions, such as ℝ¹ or ℝ², or under additional geometric constraints like convexity?

Key findings

• The paper constructs an explicit set Ω₂ ⊂ ℝ⁵ that is a finite union of unit cubes and admits an orthonormal basis of exponentials, proving it is spectral.
• The set Ω₂ does not tile ℝ⁵ by translation, as shown by a contradiction in local density estimates when assuming a tiling exists.
• The counterexample is derived from a 6-element subset of ℤ₃⁵ that is spectral but does not tile ℤ₃⁵, due to the fact that 6 does not divide 3⁵.
• The spectral property is preserved under the transference construction, so Λ₂ = Λ₁ + ℤ⁵ forms a spectrum for Ω₂ = Ω₁ + [0,1)⁵ in ℝ⁵.
• The argument extends to all dimensions n ≥ 5, showing Fuglede’s conjecture fails in all such cases.
• The result implies that the spectral and tiling properties are not equivalent in high-dimensional Euclidean spaces, even for simple geometric sets like unions of cubes.