The 8-cell (tesseract) is the only convex regular 4-polytope whose set of zomeable orthographic projections is infinite. Verified by exhaustive search over kernel directions n ∈ ℤ[φ]⁴ at rng = 1, 2, 3 (yielding 8, 32, 129 distinct shapes — the count grows without bound).
A fully inheritance-free, kernel-free confirmation was added 2026-05-08
via the strut-quadruple iso-frame enumerator
(ongoing_work/strut_iso_enumerator.py), which constructs the 4 strut
vectors (s₁, s₂, s₃, s₄) directly in ℤ[φ]³ — each as
(natural ℤ[φ]³ generator) × (positive ℤ[φ] scalar) — and verifies the
orthographic isotropy condition M Mᵀ = c·I₃ numerically. At scalar
bound = 12 the enumerator finds 15 non-face-first fingerprints:
1 cube + 1 rhombic dodec + 1 phi-oblique + 12 nondegenerate inf-family
members (12 distinct (a:b) ratios). The collapsed face-first B/G cuboid
(a,b)=(1,1) is also part of the support-2 family and is now emitted
explicitly as a notable named sample. No additional sporadic types appear,
confirming the taxonomy below.
This document gives the complete, rigorous classification.
For a unit kernel direction n ∈ S³ ⊂ ℝ⁴, the orthogonal projection of the regular 8-cell sends the four edge directions e_i ↦ c_i = e_i − n_i n. The image vectors satisfy:
| c_i | ² = 1 − n_i², c_i · c_j = −n_i n_j (i ≠ j), Σ n_i c_i = 0. |
The model has 16 ball positions V(s) = Σ s_i c_i’ for s ∈ {±1}⁴ (where c_i’ is c_i rotated into 3D = n^⊥), and 32 struts in 4 parallel classes.
The projection is zometool-axis aligned (search-engine criterion) iff each direction c_i’/|c_i’| equals a default zometool axis (B/Y/R/G).
It is additionally vZome-embeddable (in the standard icosahedral ℤ[φ]-field) iff after rotation R the c_i’ lie in ℤ[φ]³ up to a global scale, which requires every length ratio |c_i|/|c_j| ∈ ℚ(φ).
A 4D unit kernel direction n falls into one of four cases by support (number of nonzero components), and the family structure is:
| Support of n | Zometool-aligned? | vZome-embeddable? | Family structure |
|---|---|---|---|
| 1 | Yes | Yes | Cell-first sporadic: the cube (V=8). Also the b → 0 degenerate limit of the support-2 inf-family (4 of the 16 balls collapse pairwise onto the other 4). |
| 2 | Always | Iff the required metric ratios match the available zome color/scale ratios | The infinite family: split cuboids parametrized by (a:b); face-first cuboid (a,b)=(1,1) is a notable V=12 blue/green golden-field case |
| 3 | Discrete subset | One φ-rational direction yes + 2 cubic-only directions | Phi-oblique sporadic (V=16, vZome ℤ[φ]); edge-first hex prism (V=14, cubic Z[√2,√3]); face-first-like BYR shape (V=16, cubic) |
| 4 | Discrete subset | Only n = ±(1,1,1,1) up to ±-flips | Vertex-first sporadic: rhombic dodecahedron (V=15) |
So in vZome strict (icosahedral ℤ[φ] coords): there is
1 infinite family + 3 sporadic shapes (cube, phi-oblique, rhombic dodec).
We list the cube as a sporadic even though it is technically the b → 0
degenerate boundary of the inf-family: at that limit the strut count and
ball count drop (V=8, E=12 instead of 16/32), making it visually and
combinatorially distinct from every interior member of the family.
The “phi-oblique” sporadic (kernel support 3, three image vectors coplanar in 3D — none zero — with the 4th independent) was added 2026-05-08 after the inheritance-free matrix sweep and confirmed by the strut-quadruple iso-frame enumerator. The original 2-sporadic count missed it.
In the broader “zometool-axis-aligned” sense (any rotation of the 4
image vectors onto default zometool directions, regardless of length
field), the edge-first hex prism (V=14) and face-first-like BYR
shape (V=16, kernel ~ (0, 1/φ, −φ, −1)) still fall outside the strict
golden-field inventory: their ratios involve √2/√3 or √(2+φ). The
ordinary face-first cuboid is different: although its support-2
parameter has a=b=1 and c=sqrt(2), that √2 ratio is exactly the
standard blue-to-green zome strut-length ratio, so it is a golden-field
B/G model.
For n = (a, b, 0, 0) (and B₄-images thereof), with a, b ∈ ℤ[φ]:
| c_1’ = (b, 0, 0) (B-axis, magnitude | b | ) |
| c_2’ = (−a, 0, 0) (B-axis, magnitude | a | , antiparallel to c_1’) |
The 16 ball positions form a 4 × 2 × 2 cuboid lattice with x-spacings (2|b|, 2|a−b|, 2|b|), y-spacing 2c, z-spacing 2c.
Parameter: pair (a, b) ∈ ℤ[φ]²₊ modulo overall scale and modulo a ↔ b swap. The old description treated the transverse axes as blue axes only. The correct zome-frame analysis is color-aware.
For a split cuboid, the two longitudinal image vectors are antiparallel on one zome axis, and the two transverse image vectors must be perpendicular to that axis, perpendicular to each other, and equal in Euclidean length. Enumerating the default icosahedral zome axes gives, up to icosahedral symmetry, the strict golden support-2 frames:
| longitudinal axis | equal transverse axes | equation for scale parameters |
|---|---|---|
| B | B/B | c² = a² + b² |
| B | G/G | 2c² = a² + b² |
There is also a perpendicular color pattern G with B/G transverse axes, but the two transverse axes have an unavoidable √2 length ratio; making them equal would require a non-golden scalar. It is therefore not a strict golden split-cuboid frame in this support-2 family.
vZome-embeddability constraint. Write a = a₁ + a₂φ, b = b₁ + b₂φ with aᵢ, bᵢ ∈ ℤ. Then
a² + b² = (a₁² + a₂² + b₁² + b₂²) + [a₂(2a₁ + a₂) + b₂(2b₁ + b₂)] · φ
The B/B subfamily has c = √(a²+b²) ∈ ℤ[φ]. That splits into two regimes:
(i) Trace-zero regime — the φ-coefficient of a²+b² vanishes:
a₂(2a₁ + a₂) + b₂(2b₁ + b₂) = 0 (trace-zero condition)
Then a² + b² is a positive integer S, and we need S to be a perfect square or 5·m² in ℤ for c = √S to lie in ℤ[φ]. The two integer sub-families therefore arise:
(ii) Genuinely-φ regime — a²+b² has a nonzero φ-coefficient, but c ∈ ℤ[φ] picks up a matching φ-coefficient so that c² = a²+b². For example, (a, b) = (1+3φ, 4) gives a²+b² = 26+15φ, which equals c² for c = −1+5φ ∈ ℤ[φ]. The 8-cell projection then has y- and z-edge length |c| = √(26+15φ), perfectly valid in vZome.
These ℤ[φ]-Pythagorean triples give the B/B emitted samples — both trace-zero and genuinely-φ — modulo (a:b) projective equivalence and the swap a ↔ b.
The B/G subfamily is governed instead by a²+b² = 2c² in ℤ[φ].
Its smallest integral solution is the notable face-first projection:
(a,b)=(1,1), so c=√2. In a blue-only chart
this looks non-ℤ[φ], but in the zome palette the √2 ratio is exactly
the B/G length ratio. It emits as 12 balls and 20 struts:
B2:8, G2:12.Primitive integer (a, b ∈ ℤ, so trivially trace-zero) solutions on the 5m² branch come from factoring in ℤ[i]: writing a + b·i = (2 + i)·(p + q·i)² gives
a = 2p² − 2pq − 2q², b = p² + 4pq − q², m = p² + q².
First few primitive integer (a, b, m): (1, 2, 1), (2, 11, 5), (19, 22, 13), (2, 29, 13), (22, 31, 17), …
Genuinely-φ generators (a₂ ≠ 0 or b₂ ≠ 0) also exist; the integer case is just the (a₂ = b₂ = 0) sub-locus. The smallest non-integer generator points (canonicalised so a ≥ b > 0, then by ascending S) are:
| (a, b) | regime | a² + b² | c | notes |
|---|---|---|---|---|
| (√5, 2) = (−1+2φ, 2) | (i) trace-zero | 9 | 3 | emitted as 8cell_inf_family_phi_aSqrt5_b2.vZome |
| (3 + 2φ, 4φ − 4) | (i) trace-zero | 45 | 3√5 | 5m² branch (m=3) |
| (4φ, 5 − 2φ) | (i) trace-zero | 45 | 3√5 | Galois conjugate of the previous row |
| (4, 1+3φ) | (ii) genuinely-φ | 26+15φ | −1+5φ ≈ 7.09 | emitted as 8cell_inf_family_phi_a4_b1plus3phi.vZome |
rng=5/6 audit (2026-05-10). A re-search at rng ∈ {4, 5, 6} with
B₄-symmetry-deduplicated kernel directions
(gen_dirs(permute_dedup=True), exact for the tesseract) finds
exactly the same 4 shape types — cube, rhombic dodec, phi-oblique,
inf-family — and 2 inf-family parameter rows not previously emitted
(the (3+2φ, 4φ−4) and (4φ, 5−2φ) rows above). Both snap cleanly to
ℤ[φ]³ with integer coordinates and single-colour B struts. No new
sporadic types appear. Audit driver:
ongoing_work/probes/tesseract_audit_rng5.py.
Strut-iso enumeration status. The strut-quadruple iso-frame enumerator
(ongoing_work/strut_iso_enumerator.py) at scalar bound 12 found
12 distinct (a:b) ratios in the inf-family at V=16: a few integer
Pythagorean branch members ((1,2,√5), (3,4,5), …), one (2,√5,3) trace-zero
ℤ[φ] member, and 8 additional ratios on the genuinely-φ branch. The
complete list is in ongoing_work/strut_iso_frames_b12.json. Pushing
the bound higher adds more (a:b) points but produces no new shape type
— the cube + rhombic dodec + phi-oblique + inf-family taxonomy is
closed under all icosahedral-palette zomeable orthographic projections
of the 8-cell.
Recommended six examples:
8cell_face_first_cuboid_a1_b1.vZome — B/G face-first, collapsed 12-ball cuboid.8cell_inf_family_a1_b2.vZome — smallest B/B 5m² branch sample.8cell_inf_family_a3_b4.vZome — familiar B/B Pythagorean sample.8cell_inf_family_phi_aSqrt5_b2.vZome — first non-integer ℤ[φ] B/B sample.8cell_inf_family_phi_a4_b1plus3phi.vZome — genuinely-φ B/B sample.8cell_inf_family_BG_phi_a3plus3phi_bminus3plus5phi.vZome — non-collapsed B/G sample.The emitted gallery is deliberately limited to these six curated examples; older extra samples were removed after this family split was clarified.
| (a, b) | a² + b² | c | branch | filename suffix |
|---|---|---|---|---|
| (1, 1) | 2 | √2 | B/G support-2 | face_first_cuboid_a1_b1 |
| (1, 2) | 5 | 2φ − 1 = √5 | 5m² (m=1) | a1_b2 |
| (3, 4) | 25 | 5 | Pythagorean | a3_b4 |
| (√5, 2) | 9 | 3 | trace-zero ℤ[φ] | phi_aSqrt5_b2 |
| (4, 1 + 3φ) | 26 + 15φ | −1 + 5φ | genuinely-φ B/B | phi_a4_b1plus3phi |
| (3 + 3φ, −3 + 5φ) | 2(5 + φ)² | 5 + φ | genuinely-φ B/G | BG_phi_a3plus3phi_bminus3plus5phi |
Other valid integer (a, b) include (5, 12, 13), (7, 24, 25), (20, 21, 29), … on the Pythagorean branch and (2, 11, 5), (19, 22, 13), (22, 31, 17), … on the 5m² branch. Trivially-related cases: (2, 4) ≡ (1, 2) (same ratio, scaled); (1, 0) → cube; (a, a) → degenerate 12-ball collapse.
Kernel n = (0, 0, 0, 1) (or any e_i). The kernel is perpendicular to a cubic cell of the 8-cell, so the projection drops one full cell onto its 3-cube image. The 4th edge direction collapses: c_4 = 0.
8cell_cell_first_cube.vZomeThis is also the b → 0 degenerate boundary of the inf-family
(support-2) parametrization: setting (a, b) = (1, 0) gives kernel
n = (1, 0, 0, 0), with c_4 = 0 collapsing 4 of the 16 inf-family balls
onto their c_3 partners. We list it as a sporadic anyway because the
collapse changes both the ball count (16 → 8) and the strut count
(32 → 12), so visually and combinatorially it is a distinct shape, not
a generic family member.
Kernel n = (1, −1, −1, 1)/2. The kernel passes through two opposite vertices of the 8-cell (a body diagonal). All four |c_i|² = 3/4 (equal), and the 4 image vectors land tetrahedrally on the cube body-diagonals (Y axes (1,1,1), (1,−1,−1), (−1,1,−1), (−1,−1,1)).
8cell_vertex_first_rhombic_dodec.vZomeThis is the only support-4 vZome-embeddable shape, because among 4-vectors of equal magnitude in ℝ³ at tetrahedral angles, only the cube-diagonal frame has ℤ[φ] coordinates.
8cell_phi_oblique.vZome (kernel support 3)Discovered 2026-05-08 by the inheritance-free matrix sweep
(tools/inheritance_free_sweep.py) and confirmed by the strut-quadruple
iso-frame enumerator at scalar bound 12. Kernel (0, 1/φ², 1, −1/φ)
(a support-3 direction whose three nonzero components are all
φ-rational). Geometric interpretation: the three c_iʹ corresponding to
the nonzero kernel entries lie coplanar in 3D — none of them is zero
— while the c_iʹ for the zero kernel entry is independent of that plane.
This is the support-3 case the original analysis missed: unlike the two
non-embeddable support-3 directions below, this one does embed
cleanly in ℤ[φ]³. Full V=16, E=32 with 4 distinct edge-length classes
(1.0×8, 1.473×8, φ×8, 1.701×8). Stage-B sig prefix 88c5a53810074918.
(The earlier-discovered kc_triage_2c105cd47b.vZome from the M19 ZZ[φ,√2]
sweep — kernel (0, 1/φ², 1, −1/φ) — is the same shape under tesseract
B₄ symmetry; both are absorbed into this orbit.)
These are additional canonical-looking X-first projections that traditional 4D-polytope literature calls out. Unlike the ordinary face-first cuboid, they fall outside the icosahedral ℤ[φ] field and so are excluded from the “1 infinite family + 3 sporadics” count above. They are still legitimate zometool sculptures — just in algebraic-field extensions beyond the current output convention.
These two live in algebraic-field extensions beyond the current
icosahedral Q(phi) output convention, so they are not emitted here.
For the 5/16/24/120/600-cells, edge directions are not aligned with any 4D coordinate basis, so no analog of the “support-2 kernel ⇒ automatic zomeability” argument exists. Each cell type yields a finite list of zomeable projections (4, 6, 3, 1, 1 respectively in our enumeration).
8cell_cell_first_cube.vZome ← cell-first sporadic (cube, 8 balls)
8cell_vertex_first_rhombic_dodec.vZome ← vertex-first sporadic (rhombic dodec, 15 balls)
8cell_phi_oblique.vZome ← phi-oblique sporadic, support-3 (16 balls, added 2026-05-08)
8cell_face_first_cuboid_a1_b1.vZome ← face-first support-2 cuboid, (a,b)=(1,1), 12 balls, B2/G2 struts
8cell_inf_family_a1_b2.vZome ← infinite family, 5m² branch (m=1)
8cell_inf_family_a3_b4.vZome ← infinite family, Pythagorean (3-4-5)
8cell_inf_family_phi_aSqrt5_b2.vZome ← infinite family, Pythagorean (√5, 2, 3) — first ℤ[φ] generator (added 2026-05-08)
8cell_inf_family_phi_a4_b1plus3phi.vZome ← infinite family, genuinely-φ B/B sample (4, 1+3φ, −1+5φ)
8cell_inf_family_BG_phi_a3plus3phi_bminus3plus5phi.vZome ← infinite family, non-collapsed B/G sample
The emitted samples use webapp-safe integer golden-coordinate pairs, with manual physical scaling and view framing chosen so the models are readable in the browser.
Generator: lib/emit_8cell.py. To add more cuboids from the infinite
family, append to the examples list with any (a, b, c) satisfying
a² + b² = c² in ℤ (Pythagorean triple) or a² + b² = 5m² (then
c = m·(2φ−1)).
➡️ Open this page on GitHub Pages to interact with the 3D models below (the embeds only render when this file is served via GitHub Pages, not in github.com’s markdown preview).