Why the Vitali set defeats any translation-invariant measure on the reals
One of the most profound results in measure theory is that not every subset of ℝ is Lebesgue measurable. This is not a deficiency of the Lebesgue construction — it is an inherent limitation. No measure on ℝ can simultaneously be (1) defined on all subsets, (2) countably additive, (3) translation-invariant, and (4) normalized so that m([0, 1]) = 1. Something must give, and in standard measure theory, we give up measuring all subsets.
The first explicit construction of a non-measurable set was given by Giuseppe Vitali in 1905. His argument is elegant and requires only basic properties of the rationals and the Axiom of Choice. The existence of non-measurable sets is deeply connected to foundational questions in set theory and has inspired extensive research into alternative set-theoretic frameworks.
Understanding why non-measurable sets exist is essential for appreciating whymeasure theory is structured the way it is. The σ-algebra is not an arbitrary restriction — it is the necessary framework that makes consistent measurement possible.
Define an equivalence relation on [0, 1) by x ~ y if and only if x − y ∈ ℚ. This partitions [0, 1) into uncountably many equivalence classes, each of which is a translated copy of ℚ ∩ [0, 1) (modulo 1). By the Axiom of Choice, we can select exactly one representative from each equivalence class to form a set V — this is the Vitali set.
Now consider the translates V + q (mod 1) for each rational q ∈ [0, 1) ∩ ℚ. These translates are pairwise disjoint (if x ∈ (V + q₁) ∩ (V + q₂), then the representatives would be equivalent, contradicting the selection). Moreover, their union is all of [0, 1): every x ∈ [0, 1) is equivalent to some v ∈ V, so x ∈ V + (x − v).
If V were measurable, then by translation invariance each translate V + q would have the same measure m(V). By countable additivity of the disjoint union, 1 = m([0, 1)) = Σ_q m(V + q) = Σ_q m(V). But a countable sum of identical non-negative terms is either 0 (if m(V) = 0) or ∞ (if m(V) > 0). Neither equals 1. This is a contradiction, so V cannot be Lebesgue measurable.
The demonstration below illustrates the key idea behind Vitali's construction. Explore how the equivalence classes partition [0, 1), how selecting one representative from each class forms the Vitali set, and how its rational translates tile the interval — leading to the contradiction.
Say x ~ y if x - y is rational. This partitions [0, 1] into uncountably many equivalence classes. For example: 0 ~ 1/2 ~ 1/3 since their differences are rational, but 0 ≁ √2/2 since √2/2 is irrational.
Uncountably many classes, each containing countably many elements
Key insight: The contradiction arises from the interplay of three properties: translation invariance forces all translates to have the same measure, countable additivity requires the measures to sum to 1, and there are countably many translates. No single number, summed countably many times, can equal 1.
The Axiom of Choice (AC) is essential to Vitali's construction. We need AC to select one representative from each of the uncountably many equivalence classes. Without AC, the argument breaks down — we cannot form V in the first place. This raises a natural question: do non-measurable sets exist without the Axiom of Choice?
The answer, remarkably, is no — at least not necessarily. Robert Solovay showed in 1970 that if an inaccessible cardinal exists, then there is a model of set theory (ZF + Dependent Choice, without full AC) in which every subset of ℝ is Lebesgue measurable. In the Solovay model, there are no Vitali sets, no Banach-Tarski paradoxes, and measure theory works on all subsets.
However, the Solovay model gives up the full Axiom of Choice, keeping only Dependent Choice (which suffices for most of analysis). Working mathematicians generally accept AC and live with non-measurable sets as a necessary price. The Banach-Tarski paradox — which decomposes a ball into finitely many pieces that can be reassembled into two balls of the same size — is another dramatic consequence of AC applied to non-measurable sets in ℝ³.
The existence of non-measurable sets explains why measure theory requires σ-algebras. If we could measure every subset, we would not need to restrict our attention. But since some subsets are inherently non-measurable (assuming AC), we must carefully delineate which sets we allow ourselves to measure. The σ-algebra is precisely this delineation.
The Lebesgue σ-algebra is, in a sense, the largest σ-algebra on which we can define a translation-invariant, countably additive extension of length. Making it any larger would force us to include non-measurable sets, destroying countable additivity. Making it smaller (say, restricting to Borel sets) loses completeness — subsets of measure-zero sets would not all be measurable.
Non-measurable sets also appear naturally in other contexts. In probability theory, the σ-algebra of events must be chosen carefully to avoid paradoxes. In functional analysis, the theory of Lp spaces relies on the completeness of Lebesgue measure, which in turn depends on having the right σ-algebra. The lesson is clear: σ-algebras are not a technicality but a fundamental structural necessity.