A conversation about a quiet number
Two voices, one polynomial. After the small counterexamples to the Gaussian Moments Conjecture (C. D. Long, arXiv:2607.18186, 2026).
Doubter. Hand me a random quantity. Build it however you like out of a few draws from a bell curve. I ask one thing of it: its average is zero.
Maker. Easy.
Doubter. Now the average of its square is also zero. And its cube. And every power you can name, all the way up, each one averaging to exactly nothing. Tell me what it is.
Maker. You want me to say zero.
Doubter. It has to be zero. You have measured it at every order and come back empty each time. You have drained it dry. There is nothing left over for it to be.
Maker. That was my first instinct too, and for a single real draw the instinct hardens into a theorem. Square a real number, take the average, and if you get zero then the thing was zero to begin with, because a real square is never negative and cannot cancel against itself. So the moments really do pin it down.
Doubter. Then we agree and we are finished.
Maker. We are finished for as long as the recipe stays real. Let me break that one rule and keep everything else.
Doubter. Go on.
Maker. Take two independent bell-curve draws and fold them into one complex number,
Z = (X₁ + iX₂)/√2. Its mirror image W is just the conjugate. Take one more real draw
and call it T. Here is the whole animal:
P = (1 + Z)·(W − ½(2 + Z)T²).
Five terms. Nothing hidden.
Doubter. An ugly little thing.
Maker. Watch it work. Its average is zero. Square it and average: zero. Cube: zero. I can climb as high as you have patience for and every rung is nothing.
Doubter. So you have dressed up a zero in five terms. I told you.
Maker. Did I. Multiply it by Z first, and now take the average.
Doubter. …one. Do the square.
Maker. Two. Then six. Then twenty-four. It is counting. It gives back m factorial,
exactly, and it never stops.
Doubter. A zero cannot count. Nothing times anything is still nothing.
Maker. No, it can’t. Which is the point. So P is not zero.
Doubter. Then the test was broken. Averaging the powers was supposed to see the whole of it.
Maker. Here is the proof it is alive, in one number. Multiply P by its own conjugate
and average; that is the average of |P|², the honest size of the thing. You get
49/4. Twelve and a quarter. A real, positive weight. P was never small. It was only
invisible to its own powers.
Doubter. How does a thing that large hide from every one of its moments?
Maker. Look at what a power of P is really made of. The mirror parts W do a
peculiar job: averaging drags a W down onto a matching power of Z and hands back a
plain coefficient, scaled by a factorial. The lone real draw T contributes the tail of
a chi-square, which in a generating function is a square-root factor, (1 + u)^{−1/2}.
And I fed that square root the argument u = z(2 + z).
Doubter. And?
Maker. And 1 + z(2 + z) is (1 + z)². The square root of a perfect square. It
comes back as (1 + z)^{−1} and eats exactly one power of (1 + z). So the number I am
chasing turns out to be the coefficient of zᵐ in (1 + z)^{m−1}, a binomial that runs
off the end of the polynomial. There is no such term. It is zero, and it is zero for the
same one-line reason at every power.
Doubter. A lucky little identity.
Maker. The opposite of luck. It is a single point, and I can show you the edge of it.
Change that 2 inside to any other number c and run the square again. The average of
P² comes out 1 − c. It is zero when c is one, and at no other value anywhere on the
line. The silence is not some broad habit the polynomial has. It is a knife-edge it
balances on, held there by that one identity, and it topples the instant you lean it
either way.
Doubter. So the moments lie.
Maker. The moments tell the truth about a smaller question than we thought we were
asking. They see a real quantity all the way to the floor. They cannot always see a
complex one, because a complex thing can be enormous and still have every one of its
straight powers cancel. The size lives in how P pairs against its own conjugate, and
the powers of P never turn to look there.
Doubter. Where does the old certainty give out, then.
Maker. With one draw it holds; that stays a real theorem. Fold in a second complex pair and the argument survives on weight alone. But at three draws it fails, and this five-term thing is the failure, small enough to carry in a pocket. And between the one that is safe and the three that is broken sits the case of exactly two, where the same question has been asked and no one yet knows the answer. Somewhere across that short gap the bedrock stops being bedrock. The exact place is still dark.
