§6

Structure theorem — anatomy of G

Translated into algebra, the three invariants make G a subgroup of index 12 inside the free assembly space:

Theorem 6.1 — Singmaster

G ≅ {(σ, x, τ, y) \in S_{8} \times (Z /3)^{8} \times S_{12} \times (Z /2)^{12} \sum x_{i} \equiv 0, \sum y_{i} \equiv 0, sgn (σ) = sgn (τ)}

More compactly, G contains two semidirect products ("wreath products") as subgroups:

corner sector Z /3 ≀ S_{8} \times edge sector Z /2 ≀ S_{12}

The corner sector has 88,179,840 elements; it is coupled to the edge sector by the single parity lock sgn(cp) = sgn(ep).

The wreath product

A ≀ B

: B "positions" each carrying their own copy of A. B permutes positions (shuffles corners around), A independently rotates within each (twists each corner). For the cube,

B = S_{8}

and

A = Z /3

6.1 Short exact sequence

The cube group fits into a short exact sequence. Let $N = (Z /3)^{7} \times (Z /2)^{11}$ (orientations with the two dependent ones removed), and $P$ the parity-linked permutation pair (the index-2 subgroup of $S_{8} \times S_{12}$ ):

1 ⟶ N ⟶ G ⟶ P ⟶ 1

This says G is a P-extension of N — i.e. G/N ≅ P. Sanity check: |N| = 3⁷ · 2¹¹ = 4,478,976; |P| = 8! · 12! / 2 = 9,656,672,256,000; product = 4.3 × 10¹⁹ = |G|. ✓

6.2 Does it split?

A natural follow-up: does the extension split? i.e. is there an embedding of P into G? Answer: yes. The set "permute cubies while keeping CO = 0 and EO = 0" is an embedded copy of P. The extension splits, so G is a semidirect product:

G ≅ N ⋊ P

This is the algebraic counterpart of "first permute, then twist": every element of G factors uniquely as (orientation) · (permutation). This is the algebraic foundation of the (cp, co, ep, eo) state encoding.

6.3 Direct vs semidirect — the Klein 4 contrast

To see the difference between a semidirect product and a direct product, the Klein 4-group is the cleanest example. Consider $Z /2 \times Z /2$ : this is direct — two independent ℤ/2 factors. Compare against:

Structure	Multiplication	Abelian?	Cube instance
Direct product $A \times B$	$(a_{1}, b_{1}) (a_{2}, b_{2}) = (a_{1} a_{2}, b_{1} b_{2})$	iff A, B abelian	none: corners/edges parity-coupled
Semidirect $A ⋊_{φ} B$	$(a_{1}, b_{1}) (a_{2}, b_{2}) = (a_{1} φ_{b_{1}} (a_{2}), b_{1} b_{2})$	iff φ trivial	$G ≅ N ⋊ P$ ,P acts on orientations by conjugation
Wreath product $A ≀ B$	$A^{B} ⋊ B$ (B permutes \|B\| copies of A)	usually not	$Z /3 ≀ S_{8}$ is the corner sector

Key distinction: in a direct product, B's operations do not affect A. In a semidirect product, B re-interprets A by conjugation — algebraically capturing the cube's "after a U turn, corner orientations are re-labelled relative to the new positions."

6.4 Generators — 18 face turns generate G

The cube group G is generated by the 18 face turns (6 faces × 3 angles: 90°, 180°, 270°). But 6 generators suffice: $S = {U, D, F, B, L, R}$ (clockwise quarter-turns only), since $U^{2}$ and $U^{- 1} = U^{3}$ follow from U. Going further:

Just 2 generators suffice: it is known that G can be generated by 2 specific elements (e.g. $R$ and $U$ together). The proof analyses subgroup chains.
G is not cyclic: 1 generator is never enough, because G is non-abelian.
"Opposite-face equivalence": U and D are conjugate via cube rotation, but the rotation itself is not in G (G is only face-turns), so strictly speaking U and D are not conjugate within G.

This dovetails with the state-vector encoding in §3: a "word" in the generators is a solve sequence. The minimum word length is exactly God's number (§11) — the diameter of G under the generating set S.

∎

cuberoot.me · Rubik's Cube as a Group · 2026