Well-Orderings
Line the natural numbers up: 0, 1, 2, 3, \dots. Pick any non-empty
bunch of them you like — the primes, the multiples of seven, a random handful — and there is always a
smallest one. That single, humble property has a name: \mathbb{N}
is well-ordered. It is the secret engine behind proof by induction, and generalising
it is how mathematics learned to count past infinity.
A well-ordering of a set is a way of lining up its elements — a
relation
< that is a total order (any two elements are comparable) — with
one extra magic ingredient:
- A relation < on a set S is a
well-order if it is a total order in which every non-empty subset has a
least element.
- Equivalently: there is no infinite strictly-decreasing sequence
x_0 > x_1 > x_2 > \cdots — you can never fall forever.
- (\mathbb{N}, <) is the model example; from it grow the
ordinal numbers.
Well-ordered, or not?
The figure lines up two familiar orderings. The naturals climb from a definite floor; the integers
run off to negative infinity with no bottom. Play through it, then re-read the definition.
So (\mathbb{N}, <) is a well-order, but (\mathbb{Z}, <)
with its usual ordering is not — the whole set \mathbb{Z}
has no least element, and neither does the subset \{\dots, -2, -1, 0\}. It's
the "least element of every subset" clause that fails; being a total order is not enough.
Re-ordering can rescue a set
Here's the subtle part: whether a set is well-ordered depends on the ordering, not the set.
The integers under their usual order fail — but you can re-arrange them into
0,\ 1,\ -1,\ 2,\ -2,\ 3,\ -3,\ \dots
and this ordering of \mathbb{Z} is a well-order: it is just a copy
of \mathbb{N}'s ordering, so every subset has a least element (the one that
appears earliest in the list). The remarkable
Well-Ordering Theorem
says this is always possible: every set can be well-ordered — even the real
numbers. (Nobody can write such an ordering of \mathbb{R} down; its
existence rests on the Axiom of Choice.)
// Re-order Z as 0,1,-1,2,-2,... — a well-ordering (a copy of N's order).
function zAt(rank: number): number {
// rank 0,1,2,3,4,... -> 0,1,-1,2,-2,...
return rank % 2 === 1 ? (rank + 1) / 2 : -rank / 2;
}
const listed: number[] = [];
for (let r = 0; r < 9; r++) listed.push(zAt(r));
console.log("Z well-ordered: " + listed.join(", "));
// The least element of ANY subset = whichever appears earliest in this list.
console.log("least of {-2, 3, -1} in this order: " + [-1, -2, 3]
.sort((a, b) => listed.indexOf(a) - listed.indexOf(b))[0]);
Suppose some non-empty subset A had no least element. Then pick
any x_0 \in A; since it isn't least, something smaller
x_1 \in A exists; that isn't least either, so
x_2 < x_1… and you build an infinite descending chain
x_0 > x_1 > x_2 > \cdots. Conversely an infinite descending chain is
a non-empty set with no least element. So "every subset has a least element" and "no infinite
descent" are two faces of one coin — and the second is exactly why
transfinite
induction can't run forever.
-
Total order ≠ well-order. (\mathbb{Z}, <) and
(\mathbb{Q}, <) and (\mathbb{R}, <) are
all perfectly good total orders, yet none is a well-order — each has subsets with no least
element (the open interval (0,1) has no smallest rational, for
instance).
-
"Least", not "greatest". A well-order needs a smallest element in every subset;
it says nothing about greatest ones. \mathbb{N} has a least element but
no greatest — and that's fine.