The Vitali Set: Why No Notion of Length Can Cover Every Subset of the Line — Epoche C1
There is a subset of the real line, first written down by Giuseppe Vitali in 1905, to which no sensible notion of length can be attached — not because its length is hard to compute, but because four properties that any notion of length must have force this one set to have length zero and length greater than zero at the same time. The belief that every set of real numbers has a length is therefore not merely unproved; it is refutable. The argument is short and entirely deductive: four premises, one set, and a contradiction. What makes it worth working through slowly is that the contradiction does not tell you which premise to abandon, and the reasoning by which mathematics settled on abandoning one of them in particular — rather than another that looks equally optional — is more interesting than the paradox itself. What a length must do Before anything can be shown impossible, the demands must be written down. Let $m(A)$ denote the proposed length of a set $A$ of real numbers. Four premises capture what the word ought to mean. Totality. $m(A)$ is defined, as a value in $[0,\infty]$, for every subset $A$ of the line. Normalisation. An interval keeps its usual length: $m([a,b]) = b-a$. Translation invariance. Sliding a set along the line does not change its length: $m(A+t)=m(A)$, where $A+t=\{a+t : a\in A\}$. Countable additivity. If $A_1, A_2, \dots$ are pairwise disjoint — no two of them share a point — then $m\big(\bigcup_{k=1}^{\infty}A_k\big)=\sum_{k=1}^{\infty}m(A_k)$. Three consequences will be needed later, and each follows in a line. First, the empty set has length zero: the interval $[0,1]$ is the disjoint union of itself and countably many copies of $\emptyset$, so normalisation gives $1 = 1 + \sum_{k\ge 2}m(\emptyset)$, which is possible only if $m(\emptyset)=0$. Second, countable additivity then implies finite additivity, since a finite disjoint family can be padded out to an infinite one with empty sets. Third, monotonicity: if $A \subseteq B$ then $B$ is the disjoint union of $A$ and $B\setminus A$, so $m(B)=m(A)+m(B\setminus A)\ge m(A)$, the last step using only that lengths are non-negative. The fourth premise is the one a sceptic will challenge, since the finite version is obvious and the infinite version is a genuine extension. It earns its place by doing the work that makes integration possible. Given finite additivity, countable additivity is equivalent to continuity from below: if $A_1 \subseteq A_2 \subseteq \cdots$ is an increasing chain, then $m\big(\bigcup_n A_n\big) = \lim_{n} m(A_n)$. The equivalence is immediate in one direction — write the union as the disjoint union of $A_1$, $A_2\setminus A_1$, $A_3\setminus A_2$ and so on, and the series of their measures has the $m(A_n)$ as its partial sums. That statement is precisely "the measure passes through limits", and it is the property on which the monotone and dominated convergence theorems of integration depend. Drop it and one loses the ability to interchange a limit with an integral, which is most of what integration theory is for. A concrete illustration of the gap between the finite and the infinite rule: the rational numbers in $[0,1]$ form a countable set, a union of countably many single points, each of which has length zero by normalisation and monotonicity, since a point sits inside intervals of arbitrarily small length. Countable additivity forces the whole set to have length zero. Finite additivity alone forces nothing at all about it beyond the bounds $0$ and $1$. The difference between the two premises is not cosmetic. Vitali's construction The set that defeats these four premises is built by cutting $[0,1)$ into pieces and choosing one point from each piece. Call two real numbers equivalent when their difference is rational: $x \sim y$ when $x - y \in \mathbb{Q}$. This is an equivalence relation, and each of the three checks is a line of school algebra: $x-x=0$ is rational; if $x-y$ is rational so is $y-x$; and if $x-y$ and $y-z$ are rational then so is their sum $x-z$. Being an equivalence relation, it partitions the reals into classes, each of which is a coset of the rationals — the set $x+\mathbb{Q}=\{x+q : q \in \mathbb{Q}\}$ for some real $x$. Two cosets are either identical or disjoint, since sharing a point would make every element of one differ from every element of the other by a rational. Every coset meets $[0,1)$, and this is worth checking rather than assuming, because the whole construction takes place inside the unit interval. Given any real $x$, the number $x - \lfloor x \rfloor$ lies in $[0,1)$ and differs from $x$ by the integer $\lfloor x\rfloor$, which is rational; so it belongs to the same coset. The cosets therefore partition $[0,1)$ as well as the line. Now choose exactly one representative from each coset within $[0,1)$, and call the resulting set $V$. Two features of this step deserve emphasis. The first is that no rule performs the selection. Each coset is a dense countable scattering of points and nothing distinguishes one of its members from another; there is no smallest element, no canonical form, no formula. The second is the sheer scale of the choosing. Each coset is countable, being a translate of $\mathbb{Q}$, while the line is uncountable, so the number of cosets is $2^{\aleph_0}$ — as many as there are real numbers. The selection is licensed only by the axiom of choice, the set-theoretic principle stating that from any family of non-empty sets one may form a set containing one element of each. That the principle is doing genuinely infinite and genuinely arbitrary work here is the reason the final section of this essay exists. The contradiction With $V$ in hand, the contradiction comes from translating it by rationals and squeezing the result between two intervals. The translations used are by the rationals in $[-1,1]$, which form a countable set and can therefore be listed as an infinite sequence $q_1, q_2, q_3, \dots$. The choice of that particular range is not arb