Introduction

It is well known for that all regular polytopes (including the star polytopes) and regular compounds the vertices can be found in the vertices of the regular convex polytopes. Also the number of regular polytopes is known for all dimensions. The number of compounds is not really known. I describe some new ones, but I now know that there are a huge number that is only partly described, it is doubtful whether a full description ever will become available. The table below summarises what I now know:

dimensionconvex polytopesstar polytopescompoundssub-compounds
35432
461020>645
>430**

Note: there are no compounds in dimensions 5 and 6, 3 in dimension 7 and 6 in dimension 8. I do not know whether the exact number in higher dimensions is actually known, it has all to do with some kinds of orthogonal matrices.

The two sub-compounds in 3D are mirror reflections of each other, so the number of actually different sub-compounds is one. In 4D there are at least many sub-compounds that are not mirror reflections, so the number of described actually different sub-compounds is huge, the number given (645) is of the kind of compounds I cover (but I am still counting). See below in the section on compounds what subcompounds I do *not* cover.

Schläfli symbol

I almost everywhere give the Schläfli symbol of a regular object to indicate the object. It is a pair of braces with a sequence of numbers between them. For plane convex polygons it is a symbol {n} where n is the number of edges. So {4} is a square. (And as an extension {} is a line segment.)

Polygrams (e.g. the pentagram) you get from a polygon by connecting vertices with the m-th vertex in a clockwise (or counter-clockwise) fashion. It is indicated by {n/m}. If m > n/2, m can be replaced by n - m, for they give the same figure. When m equals n/2, it is just the set of lines connecting opposite vertices, which is not particularly interesting. When m divides n the figure is a set of n/m-gons (a compound). In these pages you will find the following plane figures:

figurename
{3}triangle
{4}square
{5}pentagon
{5/2}pentagram
{6}hexagon
{6/2}compound of two triangles
{10}decagon
{10/2}compound of two pentagons
{10/3}decagram
{10/4}compound of two pentagrams

For mathematical reasons other regular plane figures do not exist in the polytopes.

In 3D a regular object is given by a pair of numbers between braces. The first number gives the form of the plane figure, the second number the form of the vertex figure. To get the vertex figure you start with a vertex, take all edges that emanate from it and take the midpoints of those edges. The figure made by those figures is the vertex figure. So in a cube (which has square faces) you start at a vertex, go along the edges halfway to the next vertices, connect them and you find a triangle. So the symbol for the cube is {4, 3}. (In a regular polytope the vertex figure is the same for all vertices.)

In higher dimensional objects you do something similar. In the n-th dimension the symbol consists of n-1 numbers between braces. The first n-2 symbols describe the objects that are the enclosing cells (or hypercells), and the last n-2 symbols describe the vertex figure. In this case the vertex figure is the figure described by the cells that coincide at a particular vertex. There is obvious overlap in those numbers, but a polytopes can only be regular if the overlap coincides.

Groups

When you look at 3D regular objects (like the cube) you will find that you can rotate it so that after the rotations the object seems to be the same, but vertices have interchanged, as have edges and faces. The possible ways to rotate form the rotation group of the object.

There is however no way to rotate the cube in such a way that all four pairs of opposite vertices change place. To do that you need a central inversion. Adding that inversion to the rotation group gives the symmetry group. This is a quite important group on these objects. (But of course, for the tetraeder and its generalisations central inversion is not possible. For those the symmetry group is the same as the rotation group.)

It is known (see Coxeter for details) that the symmetry groups of the n-dimensional regular polytopes can be generated by n (n-1-dimensional) mirrors. Such a mirror can be indicated by the single vector that is orthogonal to that mirror. The Schläfli symbol connects in a simple way to the angles between the mirrors of the symmetry group (at least for the regular polytopes). Whenever the n-th number in the Schläfli symbol is k, the angle between vector n and n + 1 is π/k. The angle between non-adjacent vectors is π/2.

Applying such a mirroring to a point given in coordinates is quite simple. Let's have a vector orthogonal to the mirror and normed to length 1, the following Householder matrix gives the proper mirroring:
    I - 2vvT
That this is indeed a mirroring is outlined in many books on numerical algebra (where those matrices are used very much).

Getting back from the symmetry group to the rotation group when you only have the mirrors is also simple. Just apply an even number of mirrorings.

Sections

Starting with a regular convex polytope, you can note the vertices at the enclosing hypersphere. On the other hand (as I do) you can start with those points on the hypersphere and connect each point to every other point on it. This gives the full graph of edges between those points. Next you can look at the Euclidean lengths of those edges and you will find that they fall in a number of sections, where each section contains the portion of the graph with equal-distance edges. I number those sections, starting at 1, for increasing edge-length. So for a 3D cube the edges have length 2 (I normalise at a smallest length of 2), 2.sqrt(2) (the edges along the diagonals of the faces) and 2.sqrt(3) (the diagonals), so there are three sections of edges.

Additionally you can look at the graph-distance of vertices, i.e. the minimal number of steps you have to take along minimum-length edges to get from one point to the other. I do not use them (they are less coarse than the Euclidean distance), but on occasion I refer to them.

Transitivity

In many cases I talk about transitivity. Transitivity means that the symmetry group (or perhaps some other group) is able to transmit each object from a set to each other object in the set. For instance, the rotation group of the cube is transitive on its vertices; using rotations you can position each vertex at each other vertex. When that is not possible, the set of objects can be divided in subsets (orbits) such that the group is transitive on those subsets.

Obviously you can not expect transitivity between objects sitting in different sections. The symmetry group does not change Euclidean length. Just as obviously, the symmetry group can not change the graph-distance. On the other hand, if there are more similar objects within a section than the order of the symmetry group, transitivity does not hold either.

In almost all cases transitivity by the symmetry group holds for all regular objects. There are however exceptions from 4D onwards. But the most interesting are the intransitivities found on edges in some sections of {5, 3, ,3} (there are no intransitivities on edges for other polytopes). Some are because the edges in a sections do not have the same graph-distrance, others are because the number of edges exceeds the number of elements in the symmetry group. And some for no apparent reason at all.

However, in all cases you will find the reason for the intransitivity somewhere in the information provided.

Reciprocation

Getting reciprocals of polytopes is quite easy. Suppose you have an n-dimensional polytopy. Consider the hypersphere that contains all of the vertices. Next take the centres of the bounding cells of the polytope and project them on the enclosing sphere from the origin. These are the vertices of the reflection. Next when two cells meet at a sub-cell, connect the vertices of the reflection with an edge, and go on. What you get is the reflection, and if the initial polytope is regular, the result is also regular, and has as Schläfly symbol the reverse of the symbol of the original polytope.

You can try it with a cube to find that the reflection is the octahedron. Take the midpoints of the six faces. Connect two points if the faces have a common edge, and you will see an emerging octahedron.

Compounds

A compound is a collection of identical regular polytopes such that every vertex of some basic polytope is covered and that all vertices of the components are a vertex of that basic polytope (and also every vertex of the basic polytope should have equally many covers). For instance, when you look at the cube and take as edges all diagonals across the faces you will find that they form two interleaved tetrahedra. And you will find that that collection of tetrahedra covers each vertex of the cube exactly once. (Note that Coxeter considers these vertex-regular compounds and also considers cell-regular compounds. As compounds that are only cell-regular are the reciprocals of compounds that are vertex-regular, they are not considered, nor named, here. A compound that is both cell-regular and vertex-regular can be called fully-regular.) When you look at the regular figures in the various sections of the polytopes you can find many compounds. But more can be said about this.

Take for instance section 3 of the dodecahedron. In it you will see 10 tetrahedra. You can split those ten tetrahedra in two subsets of five so that each subset covers all of the vertices of the dodecaheder, so they are also compounds (in my pages I will in general call them sub-compounds). An interesting feature of the two subsets is that the rotation group of the dodecahedron will not transform one into the other, you need a reflection to do that, so you can say that one set is left-handed and the other right-handed. But not always when you can split are the subsets distinguishable in that way.

On the other hand, stranger things do happen. In section 2 of {3, 3, 5} you can find 25 {3, 4, 3}'s. You can group them in 5 subsets of 5 each, such that each subset covers all the vertices. The rotation group interchanges the subsets. An interesting feature is that you can do this grouping in two ways. But the subsets you get in one way are mirror reflections of the subsets you get in the other way. So while in the previous case a left-handed set and a right-handed set together form the complete compound, here you get a complete compound by either combining five left-handed sets or by combining five right-handed sets.

And as a last example, in section 18 of {5, 3, 3} you can find 840 different {3, 3, 3}'s. But the symmetry group splits them in two subsets, one with 120 elements and the other with 720 elements. So also here we have two essentially different sub-compounds, but here there is no left- or right-handedness.

I am mainly interested in minimal compounds, i.e. compounds that do not contain sub-compounds. If a compounds comtains a sub-compound, the complement of that sub-compound is also a compound. And the union of two disjoint compounds also forms a compound. So with the minimal compounds you can show everything.

On the other hand, in some sections of the {5, 3, 3} the symmetry group naturally splits the constituent polytopes that form the compound, in these cases also the maximal compound within each subset is also considered (and this maximal compound always contains all elements within a subgroup). Minimal compounds consist either of polytopes of a single class, or of polytopes of both classes (there are never more than two classes).

I am trying to cover all, but I will have missed some. Note section 5 of {5, 3, 3} (??? section 5 is empty, maybe he means section 8 --MN) where there is a huge number of sub-compounds. The reason is that the {5, 3, 3} contains a compound of {3, 3, 5}'s, {3, 3, 5}'s contain a compound of {3,&nsbp;4, 3}'s and {3, 4, 3}'s contain compounds of either {4, 3, 3)'s or of {3, 3, 4}'s. Obviously the elements of compounds from compounds also can form major compounds, and it would be nice if all such major compounds would contain only compounds of compounds. That is, alas, not the case. The situation is still worse here, were we have elements of compounds from compounds from compounds. I do not cover that level, it gives too many different compounds (at a conservative estimate: over 480 million).

On the other hand, I think Coxeter was wrong on the compounds in section 18 of {5, 3, 3}.

I may note that not all compounds are fully-regular, nor are all sub-compounds only vertex-regular. Indeed you will find examples of compounds that are not cell-regular with sub-compounds that are fully-regular. In general I will not use Coxeter's notation for compounds, with one exception, the page covering the compounds in particular.