Khinchin's Constant: The Number Hidden in Most

The magic trick Richard Elwes opens with is deliberately bad, and deliberately good. He asks someone to pick any irrational number from a list, performs a sequence of mathematical operations on it, then reveals that the answer — 2.685 — was written down before the number was chosen. The reveal lands precisely because the trick is immediately exposed: it doesn't matter which number you pick. The answer is almost always the same.

That "almost always" is doing enormous work, and unpacking it is the actual subject of the recent Numberphile episode in which Elwes, a mathematician based at the University of Leeds, walks through one of the stranger and less-celebrated constants in mathematics: Khinchin's constant.

What a Continued Fraction Actually Is

To understand what Khinchin proved, you need to understand continued fractions, which Elwes takes some care to explain without assuming prior familiarity.

Take the cube root of two — approximately 1.2599. It's "one and a bit." That bit is a number less than one, so you can write it as one over something. Take the reciprocal of that fractional part, note the integer, and repeat the process indefinitely. Because the cube root of two is irrational — its decimal expansion never terminates and never settles into a repeating loop — this process never ends. You get an infinite sequence of integers: the continued fraction representation.

The notation looks like [1; 3, 1, 5, 1, 1, ...] where the semicolon separates the integer part from everything that follows. Every irrational number has a unique continued fraction expansion, and the entries in that expansion are the raw material for what comes next.

The Geometric Mean, and Why It Matters Here

The ordinary average of a list of numbers — add them up, divide by the count — is familiar enough. The geometric mean is its multiplicative cousin: multiply the numbers together and take the nth root, where n is how many numbers you have. For two numbers, that's the square root of their product. For three, the cube root. And so on.

The geometric mean of an infinite sequence is a trickier object. It doesn't necessarily exist. What you can do is compute the geometric mean of the first two terms, then the first three, then the first four, and watch whether those values converge toward something. If they do, that limit is, sensibly enough, the geometric mean of the whole infinite sequence.

Elwes computed the geometric mean of the first 367 terms of the cube root of two's continued fraction and got 2.685. Which matched his pre-written answer. Which is the point: that number wasn't a guess about the cube root of two. It was Khinchin's constant.

What Khinchin Actually Proved

The Soviet mathematician Alexander Khinchin established, in the early twentieth century, that the geometric mean of the continued fraction entries of almost any real number converges to the same value: approximately 2.6854520010... According to Wikipedia's entry on the Wiener-Khinchin theorem, this result dates to 1934. Wolfram MathWorld notes that the constant's digits are known to very high precision through abstract probabilistic arguments — the kind of computation that doesn't require you to actually know the number you're working with.

That last part sounds paradoxical, but it's the key to both the result's power and its strangeness.

The 100% That Isn't Everything

Elwes is precise about the probability claim, and it's worth sitting with his exact formulation: "If you pick a number, a real number, so a decimal string at random, you have 100% probability that if you work this out, you'll get Khinchin's constant out of it. 100% probability."

A hundred percent. And yet exceptions exist. Lots of them.

In ordinary language, a hundred percent probability means certainty. In measure theory — the branch of mathematics that formalizes what we mean by "probability" over infinite sets — it means something more specific and more slippery. A set of exceptions can be real, even infinite, while still being so sparse relative to the full number line that they occupy, in a technical sense, zero space. The probability of landing on one, if you're picking truly at random, is zero. But zero probability is not the same as impossibility.

The exceptions include some of the most famous numbers in mathematics. The golden ratio, phi, has a continued fraction made entirely of ones — [1; 1, 1, 1, 1, ...] — and the geometric mean of an infinite sequence of ones is just one, not 2.685. Square roots of whole numbers (what mathematicians call quadratic irrationals) also fail, because their continued fractions eventually become periodic — they loop. Elwes deliberately excluded the square root of two from his list of choices, for exactly this reason. And e, the base of natural logarithms, is another exception: according to Wikipedia's treatment of representations of e, its continued fraction follows a recognizable arithmetic pattern, which is enough to disqualify it.

So we have a class of mathematical celebrities — phi, e, square roots of integers — that conspicuously don't behave like most numbers. And we have a guarantee, in the form of Khinchin's theorem, that essentially all numbers do. The tension between those two facts is genuine and unresolved.

The Proof Gap Nobody Has Closed

Here is where the conversation between Elwes and host Brady Haran becomes genuinely surprising, at least to anyone who assumed that mathematical facts this clean must be well-understood at every level.

They're not. The theorem tells us that 100% of numbers satisfy Khinchin's constant. Computation confirms it for specific numbers like the cube root of two, to the limits of calculation. But "computation confirms" and "mathematically proven" are different things, and nobody has bridged that gap for any specific number you might name — pi, the cube root of two, almost anything.

As Elwes puts it: "We know lots of numbers individually that don't give it. We know that 100% of numbers do give it. We actually don't know individually any number that does give it."

Haran's follow-up sharpens the edge of this: "We know what the number is. We know what Khinchin's constant is." And yes — the constant itself is known to high precision. Wolfram MathWorld documents its digits in detail. The abstract argument that pins down its value is solid. What's missing is the ability to point at a specific number and say: this one, provably, satisfies the theorem.

The obstacle is fundamental. As Elwes explains, the problem with numerical verification is that you cannot run any calculation forever. And the problem with abstract proof is that the structure of most numbers' continued fractions is genuinely chaotic — there's no underlying pattern to grab onto. The patterns that make phi and e exceptions are also what make them tractable. The "generic" numbers that satisfy Khinchin's constant are, in a sense, too wild to prove things about individually.

"Any proof like that is really hard. No one's managed to put one together," Elwes says.

The Shape of the Puzzle

What this Numberphile episode maps, underneath the magic trick framing, is a peculiar inversion of the usual mathematical order. Normally you'd expect that understanding a universal law would give you leverage on individual cases. If something is true of 100% of numbers, surely you can verify it for any particular number you care about?

Khinchin's constant suggests otherwise. The universality of the result is established through a measure-theoretic argument that says nothing about any individual number. The exceptions are the ones we can analyze. The rule, which covers almost everything, can't be confirmed for anything specific.

That is either a deep feature of mathematics at the infinite scale, or a gap that future proof techniques will eventually close. It's been open since the early twentieth century.

By Priya Sharma