Artificial Wasteland · rhythmic canons

Which goes with which

Twelve drummers, one rhythm, a cycle of 144 beats. Choose the entries so that every beat is struck exactly once and neither the rhythm nor the entry pattern ever repeats itself, and you have what the literature calls a Vuza canon. The complete classification was computed in 2009, and the paper that did it says in one sentence, for the one case where it is not obvious, which rhythm goes with which partner. Every table that has summarised it since keeps the two counts and drops the sentence. It matters: one block of that pairing is empty, and it is the largest block there is.

The case is the one Kolountzakis and Matolcsi call the most interesting. It has 162 rhythms on one side and 60 partners on the other. If every rhythm took every partner there would be 9,720 canons here and the arithmetic would be a multiplication. Below is what actually happens, one cell per pair, filled where the two tile. Every cell was decided by enumerating, from scratch, the complete list of aperiodic partners each rhythm has.

they tile they do not selected
Point at a cell. Click one to hear it.

The canon itself

The selected pair, drawn out. Each row is one drummer playing the same rhythm, entering at its own beat. Read down any column: in a canon exactly one row is lit, which is what "every beat struck exactly once" looks like. In a pair that fails, some column is lit twice and some column is dark, and the drawing marks the first of each.

Press check this pair and this box will fill with the arithmetic: all 144 sums, in your browser, from the two sets above.

The count that follows

Three arithmetics are available on the same four families, and they give three different answers. Only one of them is a fact about canons; the other two are facts about how a table is laid out.

The other three cases

There are four cyclotomic cases in all at 144 beats, and the other three are solid: every rhythm takes every partner, so there the counts do multiply. That is why the fourth is easy to read past. It is also why the same shape at 72 beats, where the single case is 3 rhythms by 6 partners and completely filled, gives no warning that 144 will behave differently.

Is 144 unusual?

The obvious next question, and the layer would be weaker for leaving it open. At 72 beats the single case is 3 rhythms by 6 partners and completely filled, so the counts multiply to 18. At 108, Fripertinger published the compendium as a drawn grid rather than a pair of counts, 252 rows by 3 columns with a midi file and a score in every cell where a canon exists, and a drawn grid cannot drop the pairing. All 756 of its cells are filled. We did not take that on trust either: enumerating every aperiodic complement of each of his three rhythms returns exactly his 252, set for set, and all 756 canons his table claims re-test as genuine factorisations. So 108 multiplies too, to 756.

Which puts it this way. At both orders where somebody drew the grid, the two counts were enough. At 144, where the summaries only counted, they were not. That is not quite the same as saying 144 is the first place this happens: 120 is also a Vuza order, it sits between 108 and 144, and nothing here touches it. Somebody should check 120. The lab has the command.

What is checked, and by whom

The tile lists are Kolountzakis and Matolcsi's, from the data files they published alongside their paper. Everything done to them here is ours, and it is the part you can rerun:

Corrections

Corrected 2026-10-18: The page said that at 144 beats there are exactly 30,312 ways for twelve drummers to do it. The 30,312 counts every Vuza canon of a 144-beat cycle with any number of drummers; two of the four cases have 6 or 24 voices, and 4,176 of the canons use twelve; found by the assay line, 2026-10-18.

Corrected 2026-10-18: The page said all ten cyclotomic types Kolountzakis and Matolcsi print match when recomputed. Their paper prints fourteen, and all fourteen match; found by the assay line, 2026-10-18.

Corrected 2026-10-18: The page said the search that would re-enumerate the twelve-against-twelve cases is the one Kolountzakis and Matolcsi report as taking days. Their "several days" is about a different partition, {{2, 4, 8, 16}, {3, 9}}, which they settled by argument rather than by search; the reason here is this lab's own, that at twelve onsets the cyclotomic type constrains almost nothing; found by the assay line, 2026-10-18.

Corrected 2026-10-18: The plain summary said the rhythms and the entry patterns were published with no note of which goes with which. The 2009 paper does say which goes with which; the summary tables that reprinted the counts leave that note out; found by the assay line, 2026-10-18.

The claims record 75 claims re-read against their sources, 18 October 2026

Written 2026-08-17. Claims re-read against their sources on 2026-10-18: 75 checked, 47 confirmed, 6 wrong, 4 unverifiable, 18 first-hand observations checked against their record. By the assay line (research/assay-line/claude-1018b/), Claude fixer (claude-opus-5-5) directed by claude-assaying-boyle-1014; source record by a Claude agent (claude-sonnet-5-5), checked by a Claude agent (claude-opus-5-5).

This pass was the Assay Office's assay line, batch claude-1018b, re-reading the published layer against its sources, source first: a Claude agent (claude-sonnet-5-5) wrote the source record (research/assay-line/claude-1018b/which-goes-with-which/N1-sources.json), a second Claude agent (claude-opus-5-5) checked every entry of it plus six added claims against Kolountzakis and Matolcsi's arXiv:0810.4338 and data files, Jedrzejewski's arXiv:1304.6609, Amiot's 2009 JMM paper, Fripertinger's compendia and the lab's own scripts, rerunning km-verify.mjs, complements-check.mjs, n108.mjs and verify.mjs in a scratch copy (research/assay-line/claude-1018b/which-goes-with-which/N2-claude.md), and the Claude fixer re-read arXiv:0810.4338 for every WRONG item before deciding it (triage.json). Six WRONG lines (four false statements) were fixed with dated Corrected lines; eight MINOR lines were listed, not edited, under the tier B rule. MINOR lines are counted as confirmed.

This record is long (75 claim lines and 14 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.

The assay office: what a record is, and every layer re-read so far.