Status: complete enumeration — exactly 3 essentially distinct projections.
Which orthographic projections of the 4-dimensional 24-cell into 3D produce a shadow whose 96 edges all lie along default zometool strut directions (blue / yellow / red / green of the icosahedral symmetry system)?
We do not require:
We do require:
Up to W(F₄)-symmetry of the 24-cell and O(3)-symmetry of the target, exactly three projection families exist:
| # | Common name | 3D shape | Color signature (× 8 edges per unsigned 4D-direction class) | Hull |
|---|---|---|---|---|
| 1 | Cell-first (a.k.a. short-root projection) | cuboctahedron | 6 blue + 6 green | 18 vertices, 96 edges, no collapse |
| 2 | Vertex-first (a.k.a. long-root projection) | rhombic dodecahedron | 3 blue + 8 yellow + 1 collapsed | 15 vertices, 88 visible edges + 8 zero-length |
| 3 | Triality (3-fold symmetric) | a 3-fold-symmetric solid with 4 symmetric “wings” | 3 blue + 3 yellow + 6 red (red splits 3 short + 3 long) | 24 vertices, 96 edges |
The classical names “cell-first” and “vertex-first” come from which feature of the 24-cell aligns with the projection direction:
| File | Family | Notes |
|---|---|---|
24cell_short_root_cuboctahedron.vZome |
#1 | hand-built; 18 balls + 60 struts (24 B0 + 36 G0) |
24cell_long_root_rhombic_dodecahedron.vZome |
#2 | hand-built; 15 balls + 44 struts (12 B2 + 32 Y2), φ⁴ scale |
24_cell_triality.vZome |
#3 | uses vZome’s Polytope4d with quaternion="0 0 1 3/2 3/2 5/2 5/2 4" (= (0, 1, φ, φ²)); renders 24 balls + 96 struts (24 b3 + 24 r2 + 24 r3 + 24 y3) |
Nan-24-cell-F4*.vZome |
various exploratory | early F4 / quaternion experiments |
We work in two equivalent embeddings of the 24-cell in ℝ⁴:
These two are congruent as polytopes (related by a 4D rotation + scale by √2), but projecting one along a fixed direction n in its coordinate frame does not give the same 3D shape as projecting the other along n in its frame — because the rotation between them does not commute with the projection operator. We therefore search both embeddings.
We construct the four families of unsigned 3D directions by acting with the icosahedral rotation group A₅ (60 elements) on the prototypes:
B = (1, 0, 0) → 15 unsigned axes (3 + cyclic + …)
Y = (1, 1, 1)/√3 → 10 unsigned axes
R = (0, 1, φ)/√(1+φ²) → 6 unsigned axes
G = (1, 1, 0)/√2 → 30 unsigned axes ←(originally 15 in icosahedral; 30 here covers icosi-dodecahedral)
(Yielding 61 unsigned default axes total in our search; 40 distinct values of |cos∠| occur between pairs.)
For a candidate 4D kernel direction n:
_).The kernel direction n only matters up to scale, so we enumerate
4-tuples with components in the lattice ℤ[φ] = {a + bφ : a,b ∈ ℤ},
bounded by |a|, |b| ≤ 3. After deduplicating by sign-flips and
permutations of the four axes, this gives 20,474 candidate
directions, exhausted in seconds with the cosine prefilter.
A given color signature can be produced by many directions (entire W(F₄) orbits, plus the long-↔short-root duality). To avoid overcounting, we fingerprint each hit’s 3D vertex set by:
This fingerprint is invariant under translation, rotation, reflection, and uniform scaling — it depends only on the 3D shape.
Across both 24-cell embeddings and all 20,474 directions:
These 176 hits collapse to exactly 3 distinct 3D shapes under the fingerprint:
| Signature | Distinct shape | #balls | #raw hits |
|---|---|---|---|
B:48 + G:48 |
cuboctahedron | 18 | 72 |
B:24 + Y:64 + 8 collapsed |
rhombic dodecahedron | 15 | 72 |
B:24 + R:48 + Y:24 |
triality solid | 24 | 32 |
rng=2 to rng=3 produced more
representative directions per orbit but no new shape. The solution
set is algebraic (defined by polynomial constraints over ℤ[φ]) and the
search has saturated it.A heuristic explanation: the W(F₄) action on S³ has a small number of “highly symmetric” orbits — the directions stabilized by large subgroups of W(F₄). The three families correspond to the three classes of mirror planes / fixed points of W(F₄) on the 3-sphere of unit kernel directions:
Other candidate special directions (e.g. through edge midpoints, through face-centers of cells, etc.) do not produce all-default-color shadows: the cosine spectrum of their projected edges contains values absent from the icosahedral system.
cd C:\Users\nanma\Documents\vZome\24-cell\search
python exhaustive_search.py 3 # ~16 seconds
python group_shapes.py # shape fingerprinting
Key scripts:
search/exhaustive_search.py — enumerate ℤ[φ]⁴ directions, test each
in both 24-cell embeddings, record color signatures.search/group_shapes.py — group hits by shape fingerprint;
produces the final “exactly 3” tally.search/decode_polytope4d.py — establishes that vZome’s
Polytope4d group="F4" uses the short-root 24-cell with
Hamilton left-multiplication followed by drop-k as its 3D
projection.search/emit_vzome.py — hand-builds the cuboctahedron and rhombic
dodecahedron .vZome files using <ShowPoint> / <JoinPointPair>
commands and validated against the strut lookup table.Last updated 2026-05-01.
➡️ 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).