Wythoffian operation

A Wythoffian operation is a polytope operation that can be represented on a conventional Coxeter-Dynkin diagram.

Applied to the regular polytope s in a given dimension, the Wythoffian operations produce the "Archimedean polytopes" of that dimension (named as an extension of the Archimedean solids - convex, finite, non-prismatic, and maintaining the regulars' symmetry). Wythoffian operations can also be applied to some uniform polytope s to produce other uniform polytopes, although this may produce degenerate cases such as multiple covers.

History from Stott to Coxeter
Alicia Boole Stott (the daughter of George Boole, of Boolean arithmetic fame), provided the necessary insight into a new construction of polytopes, involving a process of expansion. (1910) The process is to draw some rank of element of a regular polytope outwards, but without changing the size. The effect is to introduce new elements to fill in the hull. For example, the edges of a cube can be radially moved, so that the square faces of the cube become octagons, and new triangle faces appear at the direction of the cube's vertices. The cube is thence converted into a truncated cube. Mrs Stott used this construction to find all but one of the uniform polychora. The notation that Stott used is e_{1}P_6 (ie the cube or hedahedron, expanded by the edges). The e_0 replacement for contract a cube to zero appeared later.

Wythoff restructed the Stott expansions into mirrors. (1912) In essence, an expand preserving the edges was to push the edge further from the centre of the kaleidoscope, without altering its size. The polytope by this construction, is the result of dropping half-edges to various mirrors.

H S M Coxeter had devised a representation of Lie groups, where each point represented an order-two operation (eg AA = 1), and each drawn branch represented a non-communitive operator, the most common being an order-3 branch, of the form ABA=BAB. The Lie groups thus generated included the symmetric mirror groups. Dynkins and de Witt also independently devised this representation.

The insight in the next step is to connect Wythoff's construction as a motif to the Lie-group graph in a way to represent the construction of bulk of the uniform polytopes. (1938). Because all of the mirror groups are representable by Lie groups, this allowed the representation of the polytopes by marking nodes on the graph. None the same, the inline name of the polytope followed the modified Stott notation, applied to the Schläfli symbol, eg t_{0,1} {4,3}. Coxeter is now invoking Kepler's woodworking idioms onto the matter. His 1938 paper (Wythoff Construction) provides the necessary description using the omnitruncated 4_21.

Alicia Stott provided the necessary insight into snubs by the manner of alternation of vertices. In this case, the node is first marked, and then the mirror-point removed, leaving a hollow ring.

Shorthand and naming
Each named Wythoffian operation (except for loose cases like "expanded" or "omnitruncated") corresponds to a specific ringing of nodes and how far it is from the terminus of a Coxeter diagram, regardless of the diagram's size. That is, rectification always means node #1 is ringed, and bitruncation always means nodes #1 and #2 are ringed. (This may become confusing on branched and looped diagrams - for instance, the birectified pentacontatetrapeton has two #2 nodes ringed because the indexing starts at the single-node branch in the middle.)

Which nodes are ringed in a Coxeter diagram can be represented by numbers corresponding to their indices in the diagram, starting at #0. Thus we can say "t012" instead of "the #0, #1, and #2 nodes are ringed" or drawing out the entire diagram.

Since a polytope's Coxeter diagram has a number of nodes equal to the polytope's dimension, an n-dimensional polytope can only use operations with "prefix numbers" (positions of ringed nodes) less than n. (It makes sense since 0 is included in those numbers.) In addition, the names of some operations are typically not used until even higher dimensions than this rule may suggest. For example, rectified polygons are simply called "dual" even though a polygon's diagram can technically be rectified, and even though node 2 of the rectified hexacosichoron is ringed it is not referred to as a "birectified hecatonicosachoron."

When the ringing of a Coxeter diagram has a symmetry to it (ignoring the weights of the connections), the name for the resulting figure can be a combination of the regular and its dual. Examples include the icosidodecahedron (3D, ) and the hexeractihexacontatetrapeton (6D, ).

Examples of Wythoffian operations and their CDs
The Johnson/Ruen adjectives are meant to describe the "distances" between the first ringed node and each other ringed node. A prefix indicates which node is the first one ringed, then the largest distance between rings is listed (the distance between the first and last rings), followed by each other ring's distance from the first.

When the Johnson/Ruen adjectives differ from the Bowers adjectives, the former is more descriptive of the Coxeter diagram.

The Coxeter diagram can also be read in the opposite direction. This is often done when most of the nodes are on the far side. When this is done, the polytope's name is based on the dual. An operation with three or more rings is reduced to multiple two-ring operations that all start at the first ringed node.

There can, of course, be bi-, tri-, or even further versions of these from the terminus of the diagram. Some operations do not correspond to an exact number and position of ringed nodes, but are well-defined nonetheless.

It would not make sense for a polytope to be bidualed, triexpanded, or even biomnitruncated. If one wanted to describe the polytope ... as "biomnitruncated" instead of writing out all the operations in full, they could just call it a "t1,2,3,...,(n-1)-simplex,". In fact this is a case showing up the precedence of the Bowers terminology above the Johnson one. In fact simply turn over the graph into ... and use the appropriate t0,(n-2) prefix and replace therein the "small" by "great" and you are done.

Abstract definition
Wythoffian operations may be generalized to abstract polytopes, which need not be regular. The simplest version of this generalization takes as input a rank-$$n$$ abstract polytope $$\mathcal{P}$$ and a set $$T \subseteq \{1, \ldots, n\}$$ of unringed nodes in the Coxeter diagram, which is assumed to be linear and numbered left to right. For example, $$T=\{1, \ldots, n-1\}$$would represent the dual, as the only ringed node would be node $$n$$. The unringed nodes are given rather than the ringed ones because, instead of performing Stott expansion on the original polytope by "adding" its various rectates, this definition begins with the omnitruncate (where all nodes are ringed) and then "collapses" elements to lower ranks based on the unringed nodes.

To illustrate how this collapse works, consider the truncated icosahedron, with CD diagram. However, the orientation of the diagram in this case would suggest a bitruncated dodecahedron instead, where $$n=3$$, $$\mathcal{P}=$$ doe, and $$T=\{1\}$$. The omnitruncate of doe is grid, which may be constructed abstractly from doe as follows.


 * First, the elements of grid (except the bottom) are defined to be the chains of doe which do not include the "improper" top and bottom elements. In this case:
 * the top element of grid is the empty set (which is vacuously a chain of doe);
 * the faces of grid are chains containing a single proper element of doe (the 20 hexagons are vertices, the 30 squares are edges, and the 12 decagons are faces);
 * the edges of grid are chains with two proper elements of doe (namely 60 vertex-edge chains, 60 vertex-face chains, and 60 edge-face chains);
 * and the vertices of grid are 3-element chains i.e. flags of doe (minus the improper elements), of which there are 120.
 * Next, the chains are ordered by reverse inclusion, i.e. for all chains $$A$$, $$B$$ of doe, $$A$$ is contained in $$B$$ within grid if and only if $$B \subseteq A$$.

Now, node 1 of the diagram must be "collapsed" to obtain the desired polytope ti. Concretely, this looks like shortening the edges of grid that lie between squares and decagons until they become points. Abstractly, it is necessary to understand what happens to each element of grid during this process. The elements of a Wythoffian truncate may be found by taking subsets $$X$$ of the nodes $$\{1, \ldots, n\}$$, where each $$X$$ corresponds to an element type of the omnitruncate. Specifically, this element type is a chain containing elements of $$\mathcal{P}$$ whose ranks are in $$\{1, \ldots, n\} \setminus X = \overline{X}$$. Next, $$X$$ is broken into contiguous runs, which amounts to "factoring" the element as a prism product of other (non-prismatic) polytopes. The "factors" which contain only unringed nodes (i.e. the runs which are subsets of $$T$$) are effectively points, which do not contribute to the rank of the collapsed element. As a result, these are discarded, leaving the $$\Gamma$$-minimal set $$X_\Gamma$$ with respect to the linear diagram $$\Gamma$$. The size of $$X_\Gamma$$ gives the rank of the collapsed element, and for all $$X, Y \subseteq \{1, \ldots, n\}$$, if $$X_\Gamma = Y_\Gamma$$, then the element types of the omnitruncate corresponding to $$X$$ and $$Y$$ collapse to the same thing. The following table continues the example for ti: From the resulting set of possible $$X_\Gamma$$, it can be seen that ti has one vertex type $$\varnothing$$, two edge types $$\{2\}$$ and $$\{3\}$$, two face types $$\{1, 2\}$$ and $$\{2, 3\}$$, and one top element $$\{1, 2, 3\}$$. However, since multiple different element types of grid collapse to the same type of the truncate, the duplicate elements must be removed. For this, we find the maximum $$X$$ that collapses to a given $$X_\Gamma$$. In this case, it is easy to look through the table and find the maxima, but it is more convenient in general to construct them from $$X_\Gamma$$ itself. This may be done by adding back all nodes which are both unringed (i.e. elements of $$T$$) and not adjacent to any element of $$X_\Gamma$$ (as otherwise, it would change some of the retained contiguous runs), which form the $$\Gamma$$-complement of $$X$$, denoted $$X^\Gamma$$. The maximum set $$X_\Gamma \cup X^\Gamma = X\langle\Gamma\rangle$$ is the $$\Gamma$$-completion of $$X$$, which determines which element type of the omnitruncate should become the element of the truncate given by $$X_\Gamma$$. The element type is given in the same way as a generic $$X$$ in the omnitruncate: as a chain whose element ranks are in $$\overline{X\langle\Gamma\rangle}$$. The ti example is once again continued in the next table: Thus:
 * Finally, a bottom element is added to obtain the full abstract polytope of grid. Although the bottom element may be any set other than the aforementioned chains, the set of all elements of doe is a good choice, as the ordering definition naturally extends to include it.


 * the vertices of ti are the 60 edge-face chains of doe;
 * the edges of ti are the 60 vertex-face chains and 30 edges of doe;
 * the faces of ti are the 12 faces and 20 vertices of doe;
 * and the top element of ti is the empty chain of doe.

Any two given chains are ordered by inclusion of their corresponding $$X_\Gamma$$ and by whether the chains' union is itself a chain, so:


 * the vertices of ti are contained by both edge types, since $$\varnothing \subseteq \{2\}$$ and $$\varnothing \subseteq \{3\}$$;
 * the first edge type of ti is contained by both face types, since $$\{2\} \subseteq \{1, 2\}$$ and $$\{2\} \subseteq \{2, 3\}$$;
 * the second edge type of ti is contained by only the second face type, since $$\{3\} \subseteq \{2, 3\}$$ but not $$\{1, 2\}$$;
 * and both face types of ti are contained by its top element, since $$\{1, 2\} \subseteq \{1, 2, 3\}$$ and $$\{2, 3\} \subseteq \{1, 2, 3\}$$.

Finally, as with the omnitruncate, a bottom element is added to complete the abstract polytope for ti.

This construction may be further generalized to non-linear Coxeter diagrams $$\Gamma$$, simply by replacing "contiguous run" by "connected component of $$\Gamma$$" when finding $$X_\Gamma$$ and specifying "adjacent" to mean "connected by an edge of $$\Gamma$$" when finding $$X^\Gamma$$. However, not all possible graphs on $$n$$ vertices produce valid polytopes.