Alexander Grothendieck

K is for Klasse

In 1957 Grothendieck was proving a generalisation of the Riemann–Roch theorem, and ran into an obstruction that is almost embarrassingly plain. Coherent sheaves on a variety form a monoid under direct sum: you can add them, and you cannot subtract them. But the theorem he wanted is an equality between alternating sums, and an alternating sum needs negatives. There was no group to write it in.

The move is the one a schoolchild makes to get the integers from the counting numbers, performed at the right level of generality. Take formal differences: pairs $(a,b)$ read as “$a$ minus $b$”, with $(a,b) \sim (c,d)$ when $a+d+k = c+b+k$ for some $k$. The result is a group, the map from the monoid into it is universal among all maps into groups, and the classes of sheaves can now cancel. He called it the Klassengruppe, and the letter stuck: $K$. The Grothendieck–Riemann–Roch theorem is then a statement that a certain square commutes, and the whole of topological and algebraic K-theory grows out of the same two-line definition.

The simplest instance is the one every later statement leans on. For a field, $K_0$ is $\mathbb{Z}$, and the class of a vector space is its dimension. The content of that is not the answer but the universality: dimension is additive on short exact sequences, and every additive invariant factors through it. There is nothing else to measure. That is what makes the group worth naming rather than merely constructing.

A measurement, and what it found missing

WP-82 · The Missing Floor held this corpus against an external ruler of mathematical rungs and counted its own vocabulary, one spelling per concept, at a named commit. The result is worth repeating because it is unflattering and was published anyway: noncommutative geometry appears in nine files, Connes in fifteen, spectral triples in seven — and K-theory in none.

That distribution is upside down, and the reason it matters is not bookkeeping. A spectral triple $(\mathcal{A}, \mathcal{H}, D)$ is not a description of anything; it is an instrument. The point of packaging an algebra, a Hilbert space and a Dirac operator together is that pairing $D$ against a K-homology class yields an integer, and that integer is the invariant. Remove K-theory and the index theorem and what is left is notation with nothing to compute. So fifteen files invoking Connes are not wrong; they are ungrounded, resting on semantics that live one rung below, on a rung never written. Grothendieck’s group is that rung's floor.

Something that is nearly an index, and the way it is not

WP-82 gives the unwritten Volume XI a single concrete seed: the transverse Floquet multiplier of the dm³ limit cycle, $e^{\mu_{\max}T^{*}} = e^{-4\pi}$, offered as an analytic index. It is the only candidate for an index anywhere in this corpus, so it is the thing any K-theory volume would have to ground first. It has now been computed against the exact equations of Volume II §4.3 rather than quoted, and the result is two-sided.

Every exact statement checks out. $\Gamma = \{r=1\}$ is invariant — the radial field vanishes there at every height. The transverse eigenvalue is exactly $\lambda(z) = -2(1-e^{-z})$, to fourteen digits, with $\lambda(0) = 0$ at the neutral height. $\lambda \to -2$ as $z \to \infty$. The contact form is non-degenerate for every $r > 0$. With $T^{*} = 2\pi$, the product $\mu_{\max}T^{*}$ is indeed $-4\pi$.

The defining property of an index is not that it is computable. It is that it does not move — invariance under continuous deformation is exactly what allows Atiyah–Singer to compute the same number analytically on one side and topologically on the other. The quantity above moves. It is a smooth, strictly monotone function of the base point, it takes a continuum of values, and it is not an integer at any of them. Whatever it is, it is not yet an index, and no amount of K-theory will turn a continuously varying real number into one.

Addendum · the division nobody performed

Everything above computes decay rates. The exponent E(z₀) = −4π + 2e^−z₀(1−e^−2π) is a rate; so is λ(z) = −2(1−e^−z); so is every eigenvalue in this corpus. A search of every file in the series finds the phrase quality factor exactly once, in a chapter written this week. The corpus has never divided a decay rate by its own period.

Do it. The logarithmic decrement per turn is Λ(z) = |E(z)| and Q = π/Λ. Three things fall out, and none of them required new mathematics.

Verified in book7/dm3-q-factor-verify.py — four blocks, standard library only, no new model and no new assumption. One first-draft constant was wrong in its sixth digit and is corrected in place. See ch-nachbin for where the quantity came from.

Place in the Series

Grothendieck appears twice more in this gallery and is the same person each time. The Weil chapter records that he built étale cohomology to get the Riemann hypothesis for varieties out of it a particular way, that Deligne got it in 1974 by a route he did not regard as the right one, and that his own intended route — the standard conjectures on algebraic cycles — is still open, fifty-seven years on. Volume XIII is the category-theory volume, where the operator chain is finally asked what category it lives on, and where the corpus's use of his machinery would have to be made under a kernel; that volume exists because Mathlib has zero K-theory files and 1089 under CategoryTheory/, which is to say: because this rung is unbuilt and that one is not. Connes is the ceiling the floor is for.

The pattern across the three is one pattern. Grothendieck is the figure this corpus quotes furthest above where it stands — the standard conjectures unproved, the étale machinery unused, the K-groups unwritten — and the remedy is never to quote him less. It is to go down a rung and build.

References