Geometry : Non-composite polyhedra with convex regular faces
Created on June 20, 2024. Last update on November 20, 2026.
TL;DR The goal is to gather all non-composite non-self-intersecting, convex and non-convex, regular-faced polyhedra.
With this website, I am trying to share my hobby on polyhedra through making interactive viewers but my main topic of interest, my personal quest, is to find all non-composite non-self-intersecting (convex and non-convex) regular-faced polyhedra. To me, this is the category that lies after the (convex but not strictly) regular-faced polyhedra with conditional edges and after the regular-faced polyhedra with parquet faces (or conditional vertices). It simply lets loose of the convexity constraints. This quest is infinite but I am exploring it, one step at a time. By the way, I never displayed any solid with conditional vertices on this website, this is a task for future me, good luck my friend.
The first idea is to reduce the numbers of solids concerned by this search. The idea developed by Zalgaller is to only look for non-composite solids. This website explains it all, though in my case they are non-convex most of the time. The second idea is to start looking for solids by their vertex number. The difficulty is not to miss any in the process so I decided to base my search on polyhedral nets which are easily obtained using plantri as they are simple planar graphs. Then a Tutte's embedding is used to lay the graph in the 2D plane and some projection on the sphere serves as a starting position to find the 3D embedding. This step is about solving for the vertices positions so that the faces are flat and regular, and the solid itself does not self-intersect. Unfortunately, the embedding step may fail to find a proper embedding and may discard a valid polyhedron. In the end, there is still a lot of manual checking and discarding of wrong ones, sometimes even retrieving good ones from the garbage. The whole process is far for being perfect yet.
Since I use plantri to generate the nets, nets with a genus greater than 0 are not considered. Though I would also like to explore this family of polyhedra. Moreover, self intersecting solids are also not considered and solids sith zero dihedral angles as well.
The 3D embedding process may find multiple valid solutions. I have an example with a low number of vertices (if I forget to show it here, please remind me). My goal is not to find them all but at least one. If a conformal version is a composite solid then maybe we should not count it. That's still open for debate though. The alternative is to count all the conformal versions which is hard but knowing for sure that there is no conformal composite version is also hard.
The number of nets to check increases exponentially with the number of vertices, maybe even faster. See OEIS A000944 So some improvements is needed on the discarding step and the 3d embedding step. However, I have noticed that deltahedra (polyhedra with only triangular faces) become the majority of the valid solids for higher vertex numbers. I dare to say that for an arbitrary large n, they represent 99,9999...% of the valid solids. For the same number of vertices, there is a lot less deltahedral nets to check than polyhedral nets, see OEIS A000109. So I can find more solids with higher vertex numbers. See the table below for the numbers. I will keep a row for polyhedra (deltahedra included) and one for deltahedra. I also dare to say that for large n polyhedral nets are mainly nets that have K4 as a subnet, which are automatically disqualified from my enumeration. Unfortunately, plantri cannot generate nets without K4 as a subnet so they have to be filtered out.
The polyhedra are named with two different naming conventions. The first is a simple name based on the order in which they were found/processed based on the IUPAC systematic element naming for atomic elements prepended with the number of vertices unless a proper name already exists. I may have missed some though. The second is based on the graph6 format that encodes the upper triangular matrix of the adjacency matrix of the graph. However, instead of using the 64 character set of the graph6 format, I use the base64 character set which produces more path friendly names.
The following table sums up my findings for the first few vertex numbers. Values with a > are incomplete. The residuals on the polyhedra counted here are below 1e-16 which does not prove anything mathematically of course. Other solutions were very close with around 1e-7 residual values but they are not counted here. They could be called near-misses. More efforts will be put into finding the exact algebraic coordinates.
| # vertices | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| A000944 | 1 | 2 | 7 | 34 | 257 | 2606 | 32300 | 440564 | 6384634 | 96262938 | 1496225352 | 23833988129 |
| Polyhedra | 1 | 1 | 2 | 0 | 3 | 4 | >12 | >22 | >65* | >2 | >1 | >2 |
| A000109 | 1 | 1 | 2 | 5 | 14 | 50 | 233 | 1249 | 7595 | 49566 | 339722 | 2406841 |
| Deltahedra | 1 | 0 | 0 | 0 | 1 | 2 | >5 | >14 | >51 | todo | todo | todo |
* I have checked only deltahedral nets and polyhedral nets with 4-connected vertices.
Each model view is interactive, the controls are the following:
- Click and drag with the mouse right click to rotate the model.
- Click and drag with the mouse left click to pan the model.
- The mouse middle click zooms in or out.
- Press R to reset the model position or triple click.
- Press F to trigger fullscreen or double click.
- Press P to pause/resume the animation.
4 vertices
5 vertices
6 vertices
7 vertices
None