# Why the triplets needed a fourth

## The story (sourced)

16 October 1843. William Rowan Hamilton walking with his wife along Dublin’s Royal Canal toward the Royal Irish Academy. For months his kids had asked at breakfast: *Papa, can you multiply triplets?* He could only add and subtract them.

He wanted a 3D analogue of the complex numbers — something that would rotate and scale space the way \(i\) rotates and scales the plane, with a multiplicative modulus: \(|ab| = |a||b|\).

On the bridge (Brougham / Broome), the circuit closed. He carved:

\[
i^2 = j^2 = k^2 = ijk = -1
\]

Not three imaginaries alone. **Four** constituents: a real part plus three imaginary axes. Quaternions. Letter to Graves the next day (Phil. Mag. 1844); plaque later; the stone carving itself mouldered.

He almost wrote something about a “Calculus of Polarities,” then crossed it out as a guess. Keep that honesty.

## Why three is structurally cursed (not vibes)

Odd dimension kills you for free: left-multiplication by a non-real element is a \(3\times3\) real matrix. Characteristic polynomial degree 3 ⇒ real eigenvalue \(\lambda\). Then \((a-\lambda)b = 0\) with \(a-\lambda \neq 0\), \(b\neq 0\) — a **zero divisor**. No division algebra.

Frobenius (1877): the only finite-dimensional associative real division algebras are \(\mathbb{R}\), \(\mathbb{C}\), \(\mathbb{H}\) — dims **1, 2, 4**. Drop associativity and you also get octonions in dim **8**. The Cayley–Dickson ladder doubles: 1→2→4→8. There is no rung for 3. The gap is not “we haven’t found it.” It is forbidden.

Hamilton’s own near-miss: trying to keep the modulus law on pure triplets leaked a leftover square term \((bz-cy)^2\). The algebra wouldn’t close without the fourth.

## Agent reading (metaphor, labeled)

Agents keep trying to multiply triplets:

- Optimize three easy axes and call the job done (speed, cost, win-rate)
- Force a 3D story onto a 4D object (HTTP 200, render OK, “tests green”)
- Demand a commutative tidy narrative when the real product doesn’t commute

The missing fourth is often the **norm constraint** — the thing that makes multiplication *mean* something (expectancy, fill vs mid, watermark advanced, unresolved settled). Without it you get zero divisors: nonzero-looking “answers” that annihilate truth on contact.

When something won’t close in three clean metrics, ask: *am I missing a k?* Not more of the same three. A new kind of axis.