The Little Symbol That Almost Saved Mathematics
The Barber Who Broke Mathematics
Imagine a village with a single barber. He shaves every man who does not shave himself, and only those men. So who shaves the barber? If he shaves himself, he shouldn’t—he only shaves the self‑non‑shavers. If he doesn’t shave himself, then he must, because he shaves all who don’t shave themselves. Either way, you get a contradiction.
That tiny story is harmless fun. But in 1901 the philosopher and logician Bertrand Russell found a mathematical version: the set of all sets that don’t contain themselves. If that set contains itself, it must not; if it doesn’t, it must. The discovery, called Russell’s paradox, shook mathematics. It suggested that the most basic rules of logic could produce nonsense—and that maybe all of mathematics was built on sand.
David Hilbert (who worked in the early twentieth century) was determined to fix this. He wanted a foolproof way to show that mathematics could never, ever produce a contradiction like “0 = 1.” To do that, he invented a tiny Greek letter with a mighty job: ε.
Hilbert’s Game Plan: A Perfect Formal System
Hilbert’s plan was bold: treat all of mathematics as a kind of game. You begin with a handful of perfectly clear axioms—basic rules that everyone agrees on—and then you only make moves that follow rigid mechanical steps, like moving pieces on a chessboard.
If you can prove that no sequence of legal moves ever leads to the forbidden result “0 = 1,” then mathematics is safe forever. And Hilbert insisted that the proof itself must be finitary—meaning it uses only simple, concrete steps that can be checked by anyone with finite time and patience. No hand‑waving, no infinite leaps.
The tricky part was dealing with statements like “there exists a number with property P.” When you use such a statement in a proof, you might pick a specific witness—say, the number 2 for “there exists an even prime.” But the proof doesn’t care which witness you choose, as long as one exists. Hilbert realized he needed a symbolic operator that could stand for “choose one for me.”
He picked the Greek letter ε (epsilon). The epsilon operator works like this: if you have a formula (A(x)) that might be true for some (x), then the term (\varepsilon x A(x)) names some object that makes (A) true—if any such object exists. Its defining rule is: if (A) holds for any (x) at all, then (A(\varepsilon x A(x))) is true. In short, ε picks a witness whenever there is one.
Even better, you can define the familiar quantifiers using epsilon alone. “Exists x A(x)” becomes just (A(\varepsilon x A)), and “For all x A(x)” becomes (A(\varepsilon x (\neg A))). That means you can rewrite every logical sentence with quantifiers as a quantifier‑free sentence full of epsilon terms. The language becomes simpler and more concrete—exactly what Hilbert needed for his finitary proofs.
How to Erase the Magic Without Losing Truth
But wait—if you add this magical “pick one” operator, aren’t you bringing in new, suspicious ideal objects? Hilbert’s answer came in two results called the epsilon theorems.
The first epsilon theorem says: Suppose you prove a plain, quantifier‑free statement (like “2 + 3 = 5”) from a set of equally plain axioms, but in your proof you used epsilon terms. Then you can eliminate every epsilon and give a new proof that stays entirely inside the safe, quantifier‑free language. The epsilon symbols were just helpful ideal scaffolding—you can knock them away and the result still stands.
The second epsilon theorem extends this to any statement that doesn’t contain epsilon. If you prove it using epsilons, you can prove it without them. In other words, anything you derive in the richer epsilon calculus that talks only about ordinary mathematics is already provable in ordinary first‑order logic.
These theorems were at the heart of Hilbert’s program. They showed that his “ideal elements” (the epsilon terms) never trick you into believing a false concrete fact. If you end up with a statement about real, finite arithmetic, the epsilons have done no harm. That gave him confidence that a finitary consistency proof was possible.
The Machine That Corrects Itself—and the Dream That Faded
The next big challenge was to actually prove that no proof in the epsilon‑enhanced arithmetic ends with “0 = 1.” Hilbert and his students, especially Wilhelm Ackermann, devised a remarkable method called epsilon substitution.
Think of it as a trial‑and‑error machine. You start by guessing values for all the epsilon terms—set them to 0, for instance. Then you check all the axioms of arithmetic that contain those terms. If any axiom is violated (say, the rule that ε should pick a witness when one exists, but your guess gave the wrong witness), you adjust the value to a better one and restart. If the machine always stops after finitely many corrections and reaches an assignment that makes all axioms true, then the theory is consistent—you can never derive “0 = 1.”
Ackermann extended this method to handle nested epsilons (epsilons inside other epsilons) and even tried to apply it to second‑order arithmetic, which allows quantifying over sets of numbers. For a while, the Hilbert school believed a full finitary consistency proof was within reach.
Then in 1930, the young mathematician Kurt Gödel announced his incompleteness theorems. One consequence was that no finitary method—at least none that could be formalized inside the system—could prove the consistency of ordinary arithmetic. The grand aim of Hilbert’s program was impossible.
The story didn’t end there. Gerhard Gentzen later proved the consistency of arithmetic using a method that went beyond the finitary (it used transfinite induction up to a number called ε₀). Ackermann adapted his epsilon substitution method to work along similar lines, and proof theorists ever since have used epsilon‑style ideas to measure the strength of mathematical theories by assigning ordinal numbers to their procedures.
So Hilbert’s dream didn’t come true, but the ε‑operator survived—and found new homes nobody expected.
Donkeys, Computers, and the Afterlife of a Symbol
You might think all this talk of epsilons is only for mathematicians. But the epsilon operator also helps solve a puzzle about ordinary language. Consider the sentence:
Every farmer who owns a donkey beats it.
What does “it” refer to? If you try to write this in standard first‑order logic, you get something like: “For every farmer x and every donkey y, if x owns y then x beats y.” That would mean a farmer who owns three donkeys beats all three of them. But the original sentence suggests the farmer beats a single donkey—the one they own. How can we capture that?
Linguists noticed that the epsilon operator is exactly the right tool. The phrase “a donkey” introduces an existential quantifier. Using epsilon, we can represent it as (\varepsilon y , \text{Donkey}(y)), which picks a particular donkey. Then the pronoun “it” anaphorically points back to that very same epsilon‑chosen donkey. By adding choice functions that depend on context, linguists can elegantly explain how pronouns link up with the right object, even when standard quantifiers fail.
And there’s more. In computer science, automated reasoning systems like HOL and Isabelle use epsilon’s expressive power to represent arbitrary choices without having to introduce new function symbols. The same little ε that Hilbert drew on his chalkboard now helps computers verify that programming code and mathematical proofs are correct.
So a small Greek letter, born from a deep worry about paradoxes, ended up inside our laptops—and inside the way we understand stories about farmers and donkeys.
Think about it
- If someone showed you a proof that mathematics could never contain a contradiction, would you feel 100% certain, or would you still worry that the proof itself might have a hidden mistake? Why?
- When you say “I saw a dog and it was chasing a squirrel,” how does your brain know what “it” refers to? Could a machine ever figure that out the same way you do?
- Hilbert wanted a completely safe foundation for math. Is it more important that mathematical ideas be absolutely certain, or that they be useful and surprising—even if we can’t guarantee they’ll never produce a contradiction?
