Updated 9 April 2005.
This essay is unfinished and still very much a work in progress.
Geometrical untidiness is distinguished from from topological degeneracy. Some types of untidiness and degeneracy are discussed. Features located at infinity can have two opposing images.
9 April 2005. Precursors renamed from generators. Odd bits improved and a note on
a forthcoming essay added, as much
to show that I am still making progress on the current topic as for any other reason.
21 March 2004. Split off from "Polytopes, duality and generators", with some new material.
The ideas presented in this essay are generally applicabe to higher polytopes, although the focus here is on three-dimensional polyhedra. For the reader concerned only with the familiar three dimensions, where the word "polytope" is encountered, you may substitute "polyhedron" with no loss of rigour. Conversely, the student of higher polytopes will readily be able to extend the principles described.
Despite over 2,000 years of academic study, mathematicians have yet to come up with a satisfactory understanding of certain properties of the more complicated polyhedra. Some broad problem areas include:
Any coherent understanding of polyhedra and polytopes must take on board a number of theoretical advances which have been made over the centuries. It is worth summarising some of the more important ideas here.
We understand a polytope to be made up of sub-polytopes, or subtopes; a line segment or ditelon is bounded by vertices, a polygon by ditela, a polyhedron by polygons, and so on.
Every polyhedron has an abstract topological structure which may be distinguished from its geometrical form: any given topology may be realized as a wide variety of geometric polyhedra, all having the same underlying form.
Geometrical reciprocation about some sphere twins every polyhedron with a reciprocal figure. Reciprocating the second figure results in the original. The topologies of the two figures are dual to each other. Using a different sphere yields a different reciprocal figure - all reciprocals of a given polyhedron share the same topology. A vertex of one polyhedron is dual to some face of the reciprocal. If a vertex figure of one twin is projected onto the plane of the dual face, the coplanar figures are found to be planar reciprocals. More about vertex figures of this and other types may be found here.
The theories of topological duality and geometric reciprocation are quite general and may be applied to polytopes in any number of dimensions. The insights which they provides are so profound and illuminating that I believe they may be used to resolve many of the issues over whether certain figures are or are not polytopes. I would suggest that any acceptable definition of a polytope must treat as axiomatic, that if a certain figure is a polytope, then its reciprocals are also polytopes. Much detail discussed here on this topic is likely to be superseded by another essay in preparation, in which I dub twin pairs of polyhedra "didymohedra".
Elsewhere  I have used these ideas to define a single precursor, which may manifest as either of two dual topologies. Like polytopes, precursors are hierarchical - a polyhedron precursor is made up of polygon precursors, which are in turn made up of linear precursors.
The precursor gives equal importance to the faces and vertex figures in describing a polyhedron. It is made up of two subsets of precursors, one is manifest as the faces of a polyhedron and the other as its vertex figures. To obtain the dual polyhedron, one simply exchanges the topological manifestations (faces or vertex figures) of the two subsets. A precursor is thus a higher form of abstraction, from which either of two dual abstract polygons, or topologies, may be derived.
We will say that if one manifestation of a precursor is a polyhedron, then all its manifestations are.
It is normal to speak universally of vertices and edges. It will sometimes be convenient here to distinguish between the vertices and edges of a polyhedron on the one hand, and the corners and sides of a polygon on the other.
The term "polyhedron" has been applied to a wide variety of three-dimensional forms. A polyhedron-like structure which does not conform fully to the more standard definitions is sometimes said to be degenerate. However, different authors' ideas of what is standard and what is degenerate tend to differ. This difference is partly due to their differing areas of interest, and partly due to a failure to separate geometric and topological properties. That is to say, traditionally a degenerate structure may have a degenerate topology, or a well-formed topology but degenerate geometry. Notice that if the topology is degenerate, the geometry will necessariliy be so too.
I will say that a tidy polyhedron is one whose geometry conforms to the more standard definitions. That is, it has a finite, irreducible closed set of finite-sized flat faces which bound a connected region of space, and any edge joins exactly two vertices and two faces. This requires its topology to be well-formed. An untidy polyhedron is one whose topology is well-formed but whose geometry is degenerate in some way. From now on, I will apply the term "degenerate" only to topologies.
Characteristics of untidy polyhedra in Euclidean space include:
Curiously, while many authors have described polyhedra having one or more of these untidy properties, a planar figure exhibiting any equivalent properties is seldom accepted as a polygon. There is a lack of consistency in this tradition, which requires cleaning up.
If a polyhedron has two colinear but separate edges, its geometric reciprocal may have these edges superimposed such that four faces meet at the combined edge. For a geometric polyhedron this is traditionally not allowed, or at best regarded as degenerate (though there is of course some debate). Note however that its topology will retain the two distinct edges of the original, and remains well-formed. Consequently we may now understand such forms (for example Skilling's uniform polyhedron [x]) to be genuine, well-formed polyhedra, albeit having an untidy appearance.
In Euclidean space, if a face plane passes through the centre of the reciprocating sphere then the face reciprocates to a vertex located at infinity and its boundary to an infinite prism. Topological duals of such "hemi" forms, and even geometric reciprocals in elliptic space, cause no such problems.
The hemi faceted cube has six "butterfly" cross-quadrilateral faces and three rectangular hemi faces. The crossing points in the middle of each cross-quadriateral are "false vertces," i.e. there is only the illusion of a vertex. This cube is unusual in that its vertices are chiral, with dextro and laevo forms alternating.
|Fig : Hemi faceted cube - 3D view and vertex figure|
The reciprocal of the hemi faceted cube is an infinite stellation of the octahedron. I will call it an infinite octahedron. It has the appearance of six rhombic prisms, extending to infinity. Each face is bounded by three pairs of parallel edges, each from a different prism.
The traditional view is that the prisms extend to infinity in only one direction, which may be chosen at random. However, it is not possible to make a choice which both closes (i.e. fully bounds) the polyhedron and keeps congruency of the infinite faces and symmetry of the polyhedron - conditions which the reciprocal must surely meet - so any such choice is unacceptable.
Fig - Traditional mess
 presented an argument that a vertex at infinity has images in both directions. This provides the required closure, congruence and symmetry. It is unreasonable to treat different polyhedra differently on a whim, so all such prisms, reciprocal to any hemi faces, must have both ends.
In the centre of each infinite face is a triangular region, internal to all three pairs of prisms and having three vertices. At each of these vertices, three edges intersect - one from each pair. Each edge appears to be geometrically trisynaptic, having two finite vertices and a third at infinity along its length. This is untidy. The reciprocal vertex figure is both tidy and chiral, yet the geometric face appears to be neither. However, topologically each edge is connected to only one of the finite vertices. That is, a true vertex and a false vertex coincide. By disconnecting the correct edges, the trimethoric character is removed, and chirality restored. The edges and vertices are topologically tidy but appear geometrically untidy.
Fig - Face of the infinite octahedron
Chirality is generally treated as a geometric property, and not a topological one. Yet we have seen that it does not appear to be preserved under reciprocation. How can this be? Either the form of chirality seen here is topological in nature, or the meeting or missing of an edge passing through a vertex is a geometric property and not merely topological. I have not yet figured the answer to this one. Possibly, this is because I do not understand the difference between projective and other kinds of geometry.
The reciprocal feature to the apparently trisynaptic edges is that along any edge of the original cube, three face planes (but not three faces) share a common line of intersection. We see the hemi solids as geometrically tidy, while their infinite reciprocals are untidy.
Elsewhere  I have mentioned the difficulty of allowing for certain kinds of hole in a polygon or polyhedron. A polygon essentially traces a circuit, which is not particularly useful here when things get complicated: holes can punch through overlapping circuits or loops and only really become significant when viewed from a higher dimension.
Holes are really just the flip side of filling the interior of a polygon or polyhedron. More about the importance of fillings can be found here.
Yet another class of geometrically untidy polyhedra are those with non-planar faces. Here, the edges of a face typically form a skew polygon, and may or may not be straight. Any topologically tidy polyhedron may be represented by a simply-connected graph . A circuit enclosing any region of the graph represents a face - a circuit around an irreducible region represents a face of the outer hull, and a wider circuit around several such regions represents an edge facet of that hull. Topologically this is all nice and tidy, but geometrically the general circuit is skew, leading to skew facets; it is only in certain special cases that the circuit and facet are planar. The skew facets may be thought of in different ways, as empty in the manner of Grünbaum, as filled with a curved minimal surface such as found in certain theoretical bubbles or saddle polyhedra , or as filled with some other specialised surface such as the walls between real bubbles of different sizes, in a foam. Bubbles tend also to have curved edges. One may observe that there are a large number of facetings of a typical polyhedron, of which only a relatively small subset have planar faces. The question arises as to whether the non-planar cases reciprocate to any meaningful polyhedra. If a general curve is treated as a string of arbitrarily short line segments joined by vertices, then its reciprocal is seen to be a string of vertices joined by arbitrarily short line segments, which is a similar construction, viz. another curve. [No doubt the theory of reciprocal curved lines and planes goes a lot further. If reciprocation does break down, how do we account for the dual topology? If not, what is the reciprocal of a bubble?]
Classes of degenerate polyhedra in Euclidean space include:
Tilings are sometimes seen as infinite polytopes, or apeirotopes, of higher dimension, having an infinite number of faces. For example plane tilings are apeirohedra.
Certain non-planar arrangements also extend infinitely, dividing space into two parts. They are also apeirohedra, and have been variously called infinite polyhedra , honeycombs  and pseudopolyhedra [Gott].
We can readily find degenerate polyhedra where faces or vertices are visited by multiple circuits, or edges are visited twice by the same circuit or by different circuits. In certain situations, in a given surface n-space, we may find a compound or a composite where the figure divides up into structurally discrete but geometrically overlapping polyhedra.
Some geometric figures may best be seen as nn-methoric or n-synaptic forms, having n faces or vertices along a given edge. The topology of such figures breaks the dyadic rule, that only two (n-1)-topes meet along any (n-2) tope, so these figures are degenerate. The principles of duality still apply, so precursors remain a useful way of describing them. Indeed, since simple circuits around faces or vertex figures are no longer possible, precursors are one of the few apporaches which do remain of value.
In all these cases, the dual topology is also degenerate. We can say that in general, the dual of a degenerate polyhedron will also be degenerate.
I have previously suggested  that the dual or reciprocal of a tidy polyhedron should also be tidy. However, now that we have made the distinction between degeneracy and untidiness, we can say that while the dual of a degenerate topology will also be degenerate, the reciprocal of an untidy polyhedron may be tidy or untidy depending on the chosen projection (i.e. the chosen realisation and/or the chosen sphere of reciprocation).
Whe have seen how different geometrical figures can be classified according to their degree of degeneracy. For example the infinite "hemi duals" are topologically well-formed, and result from an accident of sphere location which has nothing to do with their underlying form. It is to be hoped that this classification will help bring some kind of order when asking of various figures, "is this a polyhedron or not, and if so then what kind is it?"