Notes on operations on polyhedra¶
Operations on polyhedra to produce other polyhedra date back as far as Kepler. Conway defined a set of operations that could be performed on the Platonic solids to obtain the Archimedean and Catalan solids, and others added operators after him. Initially there was not much theory supporting operations on polyhedra, but [Brinkmann]’s paper provides a framework.
This text is an attempt to use Brinkmann’s work to find ways to quantify,
analyze, and expand these operators. In particular, it focuses on operators on
that can be described in terms of a linear operator on the counts of vertices,
edges, and faces. These linear operators can be used to examine the composition
and decomposition of operations on polyhedra. Such operators do not constitute
all possible operations on polyhedra, or even all those that can be represented
by wythoff
in [Antiprism], but they are an interesting subset of those
operators with many nice aesthetic and geometric qualities.
Preliminaries¶
This assumes some familiarity with basic graph theory, solid geometry, and
Conway operators. See [HartConway] for a basic overview of Conway operators, or
better still, spend some time playing with [Polyhedronisme] (a web app) or
conway
in [Antiprism]. Some paper to doodle on is helpful too. In general,
this text uses the same terms as conway
. Also beware that the term “Conway
operator” is not welldefined; it can refer to any operation on a polyhedron,
Conway’s original set, operations that retain the symmetry of the seed
polyhedron like Conway’s operators, etc. depending on your source.
By Steinitz’s theorem, the edges of a convex polyhedron form a 3vertexconnected planar graph, and all 3vertexconnected planar graphs can be realized as a convex polyhedron. Convex polyhedra are topologically equivalent to spheres. The convex polyhedra are a subset of the nonselfintersecting polyhedra, or (in Grunbaum’s terminology) acoptic polyhedra, where faces may be concave (but still no holes or selfintersections). There are also other surfaces such as toroidal polyhedra (containing topological holes). In general, we’ll limit ourselves to operations on convex polyhedra. However, it can be useful to visualize operators on a planar grid, which can be thought of as a zoomedin section of a large convex polyhedra. We may also need to make use of looser sets of abstract polyhedra containing digons (two edges that connect the same vertices) or degree2 vertexes.
Faces with k sides may be called kdegree faces, by analogy with kdegree vertices.
Chamber structure¶
[Brinkmann] et al. observed that Conway’s operators, and operators like it,
can be described in terms of chambers. Each face may be divided into chambers
by identifying the face center and drawing lines from there to each vertex and
edge midpoint, as in facechambers
. Similarly, each vertex of degree
n is surrounded by n white and n grey chambers. Each edge has a white and
grey chamber on each side of the edge, as shown in edgechambers
. The
operator may then be specified by a structure of vertices and edges within
those chambers, possibly with edges crossing from one chamber to another.
[Brinkmann] et al. note that for all operators that can be expressed in terms
of these chambers, the number of edges in the result polyhedron are an integer
multiple of those in the seed polyhedron. They call this the inflation rate,
and we’ll denote it \(g\). It turns out that an edgefocused view of
these operators is fruitful: we can view it as replacing each edge and its
surroundings with a structure like that in edgechambers
, possibly
rotated or stretched, but maintaining orientation with respect to the
polyhedron. Therefore (and since Brinkmann et al. don’t actually introduce an
overarching term for these operators) we’ll call them edge replacement
operators, or EROs. If an operator’s grey chamber is a reflection of the white
chamber, we call it achiral: otherwise the operator is chiral. (Brinkmann et al.
call these local symmetrypreserving operations (LSP) and local operations
that preserve orientationpreserving symmetries (LOPSP), respectively.)
The chamber structure of the composed operator xy can be drawn by applying x to the edges of the chamber structure of y. In particular, for a given ERO x, the chamber structure of xd is simply the chamber structure of x rotated one quarter turn.
There is some freedom in where vertices are placed within the chambers.
This is more apparent with chiral EROs. Often an ERO is drawn
so that most of the vertices lie on the seed edge, but this is not necessary.
For instance, see propeller
for two topologically equivalent ways to
draw George Hart’s propeller operator (see [HartPropeller]).
George Hart’s original drawing  Drawing emphasizing relationship with a square grid 
Particular sets of edgereplacement operators¶
Conway’s original set of operations is denoted with the letters abdegjkmost. Some of these are reducible: e=aa, o=jj, m=kj, and b=ta. Borrowing an idea from ring theory, we refer to d (dual) and S (seed, identity) as the units of the EROs, and operators that are related by d are called associates. At most 4 operators can be associated with each other, corresponding to x, xd, dx, and dxd. Because these operators are so closely related to each other, in the listing at the end of this text one operator from a set of associates is chosen to represent all of them. Conway’s operators are associated as so:
 j=jd, a=dj=djd
 k, t=dkd (as well as n=kd and z=dk)
 g, s=dgd (as well as rgr=gd and rsr=sd)
 m=md, b=dm=dmd
 o=od, e=do=dod
The GoldbergCoxeter operations \(\Box_{a,b}\) and \(\Delta_{a,b}\) described in GoldbergCoxeter Operations on Polyhedra and Tilings can be fairly simply extended to a ERO. In terms of the complex plane used in Master polygons, the chamber structure of \(\Box_{a,b}\) is the section contained in the quadrilateral \(0, x(1i)/2, x, x(1+i)/2\) of a square grid on the Gaussian integers, where \(x=a+bi\). For \(\Delta_{a,b}\), the chamber structure is the quadrilateral section \(0, x(2u)/3, x, x(1+u)/3\) of a triangular grid on the Eisenstein integers, where \(x=a+bu\) and \(u=\exp(i \pi /3)\). GC operators have an invariant T, the “trianglation number”, which is identical to the inflation factor g.
 \(\Box_{a,b}\): \(g = T = a^2 + b^2\)
 \(\Delta_{a,b}\): \(g = T = a^2 + ab + b^2\)
All of the nice qualities of GC operators carry over to this extension; for instance, \(\Box_{a,b}\) operators commute with each other, as do \(\Delta_{a,b}\) operators, and the operators can be decomposed in relation to the Gaussian or Eisenstein integers respectively. Except for g and s, all of Conway’s original operators are GC operations, related by duality, or compositions of GC operators or their duals.
The simplest operators (aside from the identity) are \(\Box_{1,1} = j\) and \(\Delta_{1,1} = n = kd\). One useful relation is that if \(a=b \mod 3\), \(\Delta_{a,b} = n \Delta_{(2a+b)/3, (ba)/3}\), and if \(a=b \mod 2\), \(\Box_{a,b} = j \Box_{(a+b)/2,(ba)/2}\). (These formula may result in negative values, which should be interpreted as per Master polygons.)
Alternating operators¶
In [Coxeter8] (specifically section 8.6), Coxeter defines an alternation operation h on regular polyhedra with only evensided faces. (He actually defines it on general polytopes, but let’s not complicate things by considering higher dimensions.) Each face is replaced with a face with half as many sides, and alternate vertices are either retained as part of the faces or converted into vertices with number of sides equal to the degree of the seed vertex. (He also defines a snub operation in section 8.4, different from the s snub Conway defined, that is equivalent to ht.) The alternation operation converts quadrilateral faces into digons. Usually the digons are converted into edges, but for now, let digons be digons.
This motivates the definition of “alternating operators” and an “alternating
chamber” structure, as depicted in facealtchambers
and
edgealtchambers
. Like earlier, we can think of this as replacing each
edge with edgealtchambers
, stretched or rotated but maintaining
orientation with respect to the polyhedron, so we can call these operators AEROs
(alternating EROs) for short. This structure is only applicable to polyhedra
with evensided faces. The dual operators of those are applicable to polyhedra
with evendegree vertices, and should be visualized as having chambers on the
left and right rather than top and bottom. Like EROs, the chamber
structure of xd is that of x rotated a quarter turn; but now, the direction
of rotation matters, and depends on how the alternating vertices (or faces) of
the underlying polyhedron are specified. For the sake of simplicity, we’ll only
look at AEROs on evensided faces (vertexAEROs, or VAEROs) instead of on
evendegree vertices (faceAEROs, or FAEROs).
VAEROs depend on the ability to partition vertices into two disjoint sets, none of which are adjacent to a vertex in the same set; i.e. it applies to bipartite graphs. We’ll denote those sets as \(+\) and \(\). By basic graph theory, planar bipartite graphs have faces of even degree. However, this does not mean that the two sets of vertexes have the same size, let alone that the sets of vertices of a given degree will have a convenient partition. The cube and many other small evenfaced polyhedra do partition into two equal sets of vertices, so beware that examining simple, highlysymmetric polyhedra can be misleading. (A section on AEROs briefly appeared on the Wikipedia page for Conway operators. It made some errors that seemed to result from assuming that the partitions were of equal size.)
Strictly, since AEROs map polyhedra with evensided faces to arbitrary polyhedra, they are not operators in the strict mathematical sense. (In particular, since AEROs do not necessarily produce evensided faces or evendegree vertices, they cannot be composed together arbitrarily.) However, calling them “transformations” instead felt awkward, since the term “operator” is so commonly used. You can call them AERTs, VAERTs, and FAERTs instead if you like.
Digons and degree2 vertices are an unavoidable fact of certain VAEROs, particularly on quadrilateral faces. Two important special cases are where the seed polyhedron has only quadrilateral faces, and when it has only faces of degree 6 or more (although the latter case only appears in infinite tilings). In the former case, the degree2 features can be uniformly smoothed out. In the latter, degree2 features are not created.
Other Operators¶
There are some important operations on polyhedra that don’t fit into the edgereplacement schema.
 r, the reflection operator. This produces the mirror image of the polyhedron. If an operator x is chiral, rxr is its chiral pair.
 $, the smoothing operator (newly defined here). This operator smooths degree2 vertices and digons, as produced by some AEROs. This operator is recursive, and will smooth features until there are no degree2 features left to smooth. For instance, two vertices may be connected by one edge and another edge split by a degree2 vertex; one smoothing iteration would smooth that degree2 vertex into a single edge, creating a digon, and the next would reduce the digon into a single edge.
 @, the alternation operator (newly defined here). This operator just exchanges the \(+\) and \(\) partitions. Applied to an operator, it reflects its chamber structure horizontally.
There are some operators that can be described like an ERO but violate one
or another constraint. One example is the lozenge operator depicted in
lozenge
. Its problem is that its result is never a 3vertexconnected
graph: it is only 2connected. Thus, the result is not a convex polyhedron:
it is, however, an acoptic polyhedron of genus 0. We’ll call operators
like this pseudoEROs, and denote them with latin letters rotated 180 degrees,
so lozenge is ⅂. (This is to emphasize their separateness, and not at all
because we’re running out of letters.)
Representations of operators¶
In abstract algebraic terms, EROs form a monoid: a group without an inverse, or a semigroup with an identity element. Let \([v,e,f]\) be the count of vertices, edges, and faces of the seed, and \(v_i\) and \(f_i\) be the count of vertices/faces of degree \(i\) such that \(\sum v_i = v\) and \(\sum f_i = f\). There is a series of monoids and homomorphisms between the monoids, as so:
 ERO x (acts on polyhedra)
 Infinitedimensional linear operator \(L_x\) (acts on \(v_i, e, f_i\))
 3x3 matrix \(M_x\) (acts on \([v,e,f]\))
 Inflation factor g (acts on \(e\)) and operator outline
AEROs do not form a monoid (since in general they cannot be composed together) but do admit a similar representation. For VAEROs, the count of vertices of degree \(i\) in the \(+\) partition are denoted \(v^+_i\) and those in the \(\) partition as \(v^_i\). \(\sum v^+_i = v^+\), and similarly for \(\). \(v^+_i + v^_i = v_i\), and \(v^+ + v^ = v\). Partitions of \(f\) for FAEROs are denoted similarly.
Each bullet will be handled in turn.
The action of an ERO on the vertices of degree \(i\), edges, and faces with \(i\) sides can be described with an infinite linear operator \(L_x\). This operator can be determined by counting elements off the chamber structure. Step by step:
 Seed vertices are either retained or converted into faces centered on that vertex. (Other options are precluded by symmetry). Let \(a = 1\) if the seed vertices are retained, and 0 otherwise. Also, the degree of the vertex or face is either the same as the seed vertex, or a multiple of it; let \(k\) be that multiple.
 Seed face centers are either retained (possibly of in a smaller face) or converted into vertices. (Again, other options are precluded by symmetry). Let \(c = 0\) if the seed faces are retained, and 1 otherwise. Let \(\ell\) serve a similar role as \(k\) above: the degree of the vertex or face corresponding to the seed face center is \(k\) times the degree of the seed vertex.
 Except for the faces or vertices corresponding to the seed vertices and face centers, the added elements are in proportion to to the number of edges in the seed. \(g\) is the count of added edges (the edge multiplier or inflation rate), \(b_i\) is the number of vertices of degree \(i\) added, and \(b'_i\) is the number of faces of degree \(i\) added.
Count elements lying on or crossing the outer edge of the chamber structure as half. It may help to draw an adjacent chamber, particularly when determining the number of sides on a face. The result of the counting process can be described in the following operator form; variables in capital letters are the result of the operator.
where \(a\), \(a'\), c, and \(c'\) are either 0 or 1, g is a positive integer, all \(b_i\) and \(b'_i\) are nonnegative integers, and \(k\) and \(\ell\) are positive integers. The subscripted values like \(v_{i/k}\) should be interpreted as 0 if \(i/k\) is not an integer.
The only alteration needed to accommodate VAEROs is that the action on seed vertices may be different depending on which partition they are in. (Counting elements may be more complicated: it’s possible to have an edge pass through one chamber without meeting any vertices.)
\(a^+\), \(a^\), \(a'^+\), and \(a'^\) are either 0 or 1. \(k^+\), \(k^\) are positive integers and \(\ell\) may take values in \(\mathbb{N}/2 = \{1/2, 1, 3/2, 2, ...\}\).
We’ll refer to \(g, a, a', b_i, b'_i, c, c', k, \ell\) as the invariants of an ERO, and \(g, a^+, a'^+, b_i, b'_i, c, c', k^+, k^, \ell\) as the invariants of a VAERO. If \(a^+ = a^\) both may be written as \(a\), and similarly for \(a'\) and \(k\). \(a'\) and \(c'\) may be omitted since they can be calculated from \(a\) and \(c\). FAEROs would be described correspondingly.
Explicitly the composition of two EROs xy can be described as so. Let \(g, a, a', b_i, b'_i, c, c' k, \ell\) be the invariants for \(L_y\); \(G, A, A', B_i, B'_i, C, C', K, L\) for \(L_x\); and \(\gamma, \alpha, \alpha', \beta_i, \beta'_i, \sigma, \sigma', \kappa, \lambda\) for \(L_{xy}\):
Under the constraint that an ERO preserves the Euler characteristic, it can be shown that \(a + a' = 1\), \(c + c' = 1\), and \(g= b + b' + 1\) where \(\sum b_i = b\) and \(\sum b'_i = b'\). For VAEROs, \(a^+ + a'^+ = 1\) and \(a^ + a'^ = 1\). Also, since \(b_i\) and \(b'_i\) are nonnegative integers, only a finite number of their values can be nonzero. This makes the operator form more manageable than the term “infinite linear operator” may suggest; in reality, nearly all applications will only use a finite number of different vertex and face degrees.
Applying the handshake lemma gives relations between the values for EROs:
or for VAEROs:
For EROs, these relations can be manipulated into the form
which is interesting because it eliminates g, a and c, and because it suggests that features with degree 5 or more exist in balance with features of degree 3 (triangles and degree3 vertices), and that in some sense degree 4 features come “for free”. The relationship for VAEROs is the same except replace \(2k\) with \(k^+ + k^\). (For FAEROs, replace \(2\ell\) with \(\ell^+ + \ell^\).)
With these relations, and the assumption that there are no degree 2 features and therefore \(i \ge 3\), a series of inequalities can be derived for EROs:
and for VAEROs:
The dual ERO \(L_d\) has the form \(E = e, V_i = f_i, F_i = v_i\). With a little manipulation, it is easy to see that if \(L_x\) has invariants a, \(b_i\), c, etc, then applications of the dual operator have related forms. \(L_x L_d\)’s invariants exchange a with c, \(a'\) with \(c'\), and k with \(\ell\). \(L_d L_x\)’s invariants exchange a with \(a'\), c with \(c'\), and each \(b_i\) with each \(b'_i\). Finally, \(L_d L_x L_d\)’s invariants exchange a with \(c'\), and \(a'\) with c, k with \(\ell\), and each \(b_i\) with each \(b'_i\).
For EROs, the matrix form \(M_x\) can be obtained from \(L_x\) by summing \(\sum v_i = v\) and \(\sum f_i = f\), or from counting elements directly from the chamber structure without distinguishing between vertices and faces of different degrees. (The conversion from \(L_x\) to \(M_x\) is itself a linear operator.) The matrix takes the form:
The matrix for the identity operator S is just the 3x3 identity matrix. The matrix for the dual operator is the reverse of that:
The dual matrix operates on other matrices by mirroring the values either horizontally or vertically.
VAEROs with \(a^+ = a^\) can also be written as a 3x3 matrix. In general, VAEROs can be written as a 4x3 matrix mapping \([v^+,v^,e,f]\) to \([v,e,f]\). FAEROs can be written as a 4x3 matrix as well, but that one mapping \([v,e,f^+,f^]\) to \([v,e,f]\). Since the \(e\) row is zero except for the value \(g\) in the \(e\) column, there shouldn’t be much ambiguity.
It can be seen from the composition equations that for an ERO xy, the expansion factor g is the product of the g invariants for operators x and y. It can also be seen that \(a, a', c, c'\) form their own linear system, a submatrix of \(M_x\): let \(\Lambda_x = \begin{bmatrix} a & c \\ a' & c' \end{bmatrix}\), then \(\Lambda_{xy} = \Lambda_x \Lambda_y\). \(\Lambda_x\) represents the effect of the operator on the seed faces and vertices: this can also be represented as a drawing of those seed faces and vertices, called the “outline” of the operator. By cofactor expansion, \(\det (M_x) = g \det (\Lambda_x)\). \(\Lambda_x\) has a determinant of 1, 0, or 1. (In fact, \(\Lambda_x\) has two eigenvalues, one of which is always 1, and one of which may be 1, 0, or 1. \(M_x\) has three eigenvalues: two it shares with \(\Lambda_x\), and one is g.) The dual operator has \(\det (M_x) = \det (\Lambda_x) = 1\), and it is easy to see that of the four possible \(\Lambda_x\), the first two and last two in the table below are related by the dual operator. With that motivation, we define the “Type” of the operator as the absolute value of the determinant of \(\Lambda_x\).
Like earlier, VAEROs with \(a^+ = a^\) are also associated with a 2x2 matrix \(\Lambda_x\). All VAEROs are associated with a 3x2 matrix \(\Lambda_x = \left[\begin{array}{ccc}a^+ & a^ & c \\ a'^+ & a'^ & c'\end{array}\right]\). FAEROs are associated with a 3x2 matrix \(\Lambda_x = \left[\begin{array}{ccc}a & c^+ & c^ \\ a' & c'^+ & c'^\end{array}\right]\). To reduce ambiguity, a vertical bar is included to separate the \(a\) values from the \(c\) values. VAEROs and FAEROs with \(a^+ \ne a^\) can be shoehorned into the 2x2 matrix form if the matrix is allowed to have undefined values for its entries, treated like NaN in floatingpoint numbers, which is denoted \(?\). 3x2 matrixes don’t have determinants, so the type of a VAERO with \(a^+ \ne a^\) is not defined.
Outline  Kind & Type  2x2 Matrix  3x2 Matrix 

Any  1  \(\begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix}\)  \(\left[\begin{array}{ccc}1 & 1 & 0 \\ 0 & 0 & 1\end{array}\right]\) or \(\left[\begin{array}{ccc}1 & 0 & 0 \\ 0 & 1 & 1\end{array}\right]\)  
Any  1  \(\begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix}\)  \(\left[\begin{array}{ccc}0 & 0 & 1 \\ 1 & 1 & 0\end{array}\right]\) or \(\left[\begin{array}{ccc}0 & 1 & 1 \\ 1 & 0 & 0\end{array}\right]\)  
Any  0  \(\begin{bmatrix} 1 & 1 \\ 0 & 0 \end{bmatrix}\)  \(\begin{bmatrix} 1 & 1 & 1 \\ 0 & 0 & 0 \end{bmatrix}\)  
Any  0  \(\begin{bmatrix} 0 & 0 \\ 1 & 1 \end{bmatrix}\)  \(\begin{bmatrix} 0 & 0 & 0 \\ 1 & 1 & 1 \end{bmatrix}\)  
VAERO  \(\begin{bmatrix} ? & 0 \\ ? & 1 \end{bmatrix}\)  \(\left[\begin{array}{ccc}1 & 0 & 0 \\ 0 & 1 & 1\end{array}\right]\)  
VAERO  \(\begin{bmatrix} ? & 1 \\ ? & 0 \end{bmatrix}\)  \(\left[\begin{array}{ccc}0 & 1 & 1 \\ 1 & 0 & 0\end{array}\right]\)  
VAERO  \(\begin{bmatrix} ? & 1 \\ ? & 0 \end{bmatrix}\)  \(\left[\begin{array}{ccc}1 & 0 & 1 \\ 0 & 1 & 0\end{array}\right]\)  
VAERO  \(\begin{bmatrix} ? & 0 \\ ? & 1 \end{bmatrix}\)  \(\left[\begin{array}{ccc}0 & 1 & 0 \\ 1 & 0 & 1\end{array}\right]\)  
FAERO  \(\begin{bmatrix} 0 & ? \\ 1 & ? \end{bmatrix}\)  \(\left[\begin{array}{ccc}0 & 1 & 0 \\ 1 & 0 & 1\end{array}\right]\)  
FAERO  \(\begin{bmatrix} 1 & ? \\ 0 & ? \end{bmatrix}\)  \(\left[\begin{array}{ccc}1 & 0 & 1 \\ 0 & 1 & 0\end{array}\right]\)  
FAERO  \(\begin{bmatrix} 1 & ? \\ 0 & ? \end{bmatrix}\)  \(\left[\begin{array}{ccc}1 & 1 & 0 \\ 0 & 0 & 1\end{array}\right]\)  
FAERO  \(\begin{bmatrix} 0 & ? \\ 1 & ? \end{bmatrix}\)  \(\left[\begin{array}{ccc}0 & 0 & 1 \\ 1 & 1 & 0\end{array}\right]\) 
The composition of EROs affects their outlines like so:
In general, AEROs cannot be composed together, but the result of an AERO is just another polyhedron, so any AERO can be composed with an ERO on the left.
For EROs, the parity of the invariants \(g\) and \(b\) also describe the center of the chamber structure. In particular, an ERO with both \(g\) and \(b\) odd is not possible. (This does not apply to AEROs, which have different symmetry structure.)
\(g\)  \(b\)  Description 

Even  Even  A face with even degree lies at the center 
Even  Odd  A vertex with even degree lies at the center 
Odd  Even  An edge crosses the center 
Odd  Odd  Excluded by symmetry 
Unfortunately, all these relations taken together are insufficient to discern invariants that do or do not correspond to an actual ERO. For instance, \(g=4, a=1, c=0, b_4 = 1, b'_3=2, k=2, \ell=1\) satisfies the relations, but doesn’t appear to correspond to any ERO. Furthermore, the pseudoERO Lozenge satisfies the relations. (The inequalities given above only rely on the graph having minimum degree 3, which is necessary for a graph to be 3vertexconnected but not sufficient.)
Furthermore, none of these homomorphisms are injections: there are certain
\(L_x\) or \(M_x\) that correspond to more than one EROs.
Examples for \(M_x\) are easy to come by: where n = kd, \(M_k = M_n\).
For an example where the operators are not related by duality,
\(M_l = M_p\). For \(L_x\), \(L_{prp} = L_{pp}\) but prp is not
the same as pp (one’s chiral, one’s not). For the operator depicted in
waffle
, \(W \ne Wd\), but \(L_W = L_{Wd}\).
(This is a newly named operator, introduced in this text.) A general
counterexample would be operators with sufficiently large g based on
\(\Box_{a,b}\), with a single square face (not touching the seed vertices
or face centers) divided into two triangles:
the counts of vertices of each degree, faces of each degree, and edges would be
the same no matter which faces was chosen, but the operators would be different.
With this construction, it is possible (with a sufficiently large g) to
create arbitrarily large sets of operators with the same invariants.
Chirality¶
It may be possible to introduce another invariant into these operators and distinguish operators not discerned by \(L_x\) or \(M_x\). The most desirable may be a measure for chirality; in theory that would distinguish, e.g. pp vs prp. However, this does not appear as simple as assigning achiral operators to 0 and \(\pm 1\) to chiral operators. The composition of a chiral operator and an achiral operator is always chiral, but:
 Two chiral operators can produce an achiral operator: prp
 Two chiral operators can produce another chiral operator: pp, pg, prg, gg, grg
Further confusing things are chiral EROs where r and d interact. Some chiral
EROs have xd = x, while some others have xd = rxr. (Some have x = dxd,
but none with rxr = dxd have been observed or proven/disproven to exist.)
The gyro operator is one example of the latter, and the bowtie operator
in bowtie
is another, maybe easiertovisualize example.
(Bowtie is a newly named operator, introduced in this text.)
Operators that produce alternating polyhedra¶
The alternation operator @ just exchanges \(+\) and \(\), so its matrix form is a simple permutation matrix.
\([v^+,v^,e,f]\) to \([v^+,v^,e,f]\)  \([v,e,f^+,f^]\) to \([v,e,f^+,f^]\) 
\[\begin{split}\mathbf{M}_@ = \begin{bmatrix}
0 & 1 & 0 & 0 \\
1 & 0 & 0 & 0 \\
0 & 0 & 1 & 0 \\
0 & 0 & 0 & 1 \end{bmatrix}\end{split}\]

\[\begin{split}\mathbf{M}_@ = \begin{bmatrix}
1 & 0 & 0 & 0 \\
0 & 1 & 0 & 0 \\
0 & 0 & 0 & 1 \\
0 & 0 & 1 & 0 \end{bmatrix}\end{split}\]

When considered with the bipartite structure, the dual operator d can be considered to transform polyhedra with bipartite vertices into polyhedra with bipartite faces and vice versa. On operators, it converts VAEROs to FAEROs (and vice versa). Its matrix is also a simple permutation matrix.
\([v^+,v^,e,f]\) to \([v,e,f^+,f^]\)  \([v,e,f^+,f^]\) to \([v^+,v^,e,f]\) 
\[\begin{split}\mathbf{M}_d = \begin{bmatrix}
0 & 0 & 0 & 1 \\
0 & 0 & 1 & 0 \\
1 & 0 & 0 & 0 \\
0 & 1 & 0 & 0 \end{bmatrix}\end{split}\]

\[\begin{split}\mathbf{M}_d = \begin{bmatrix}
0 & 0 & 1 & 0 \\
0 & 0 & 0 & 1 \\
0 & 1 & 0 & 0 \\
1 & 0 & 0 & 0 \end{bmatrix}\end{split}\]

The join operator j produces quadrilateral faces only. In fact, all type 0 \(\Box_{a,b}\) operators produce quadrilateral faces, but those can be reduced into \(j\Box_{c,d}\) for some \(c, d\), so it’s enough to look at j for those operators. One way to assign a bipartite structure to the vertices of j is to mark the seed vertices as \(+\) and the vertices corresponding to the seed faces as \(\). Expressed as a matrix from \([v,e,f]\) to \([v^+,v^,e,f]\):
The opposite bipartite structure would simply be the same matrix, flipped from left to right. This corresponds to applying the dual operator on the right: jd = @j, so the relation gets a little more complicated when considering alternating operators. The ambo operator produces bipartite faces, and since a=dj, it can be expressed in terms of j, d, and @.
There are some tilings where an bipartite structure can be defined on both the vertices and the faces. The square grid is one, as well as some regular hyperbolic tilings (in general, any regular tiling with Schläfli symbol {n,m} where n and m are both even). However, we haven’t defined any operators that require both vertices and faces to have an bipartite structure, so it’s enough to consider one at a time.
All EROs can be expressed with smoothing, an AERO, and the join operator¶
The operator $xj, where x is a VAERO, is an ERO. If x is type 0 or 1 VAERO, then $xj is a type 0 operator. If x has undefined type, then $xj is a type 1 operator. Although $ does not in general have a \(M_x\) form, in the expression $xj it either does nothing, removes an edge and a vertex, or removes an edge and a face. These operations can be represented by taking the matrix form of xj and subtracting the zero matrix or these two following matrices, respectively:
In fact, all EROs y can be expressed as y = $xj, where x is some VAERO or ERO. This is easier to see by going backwards from the operator. As mentioned earlier, if g is odd, there is an edge that lies on or crosses the center point of the seed edge in the chamber structure. Otherwise g is even and either a vertex lies there or a face contains the center point. If g is odd, either split the edge with a degree2 vertex at the center point, or replace the edge with a digon. Then the alternating chamber structure of x is just the white and grey chambers of y, stacked along their long edge. More specifically, given an ERO y, if g is even, then y = xj for an ERO or VAERO x: if g is odd, then y = $xj for (at least) two VAEROs x corresponding to splitting the edge with a vertex or replacing an edge with a digon. (Even though it can be reduced further in a larger set of operators, the ERO form is usually preferable because including all those $ and j operators would get tedious.) A VAERO x may be named “pre(Name)” where (Name) is the name of y.
Note that since xjd = x@j, the ERO of the dual corresponds to the oppositepartition VAERO. EROs may also be decomposed into FAEROs with the form y = $xa, but since a = dj and xd has the chamber structure of x rotated, it’s simpler to just look at VAEROs.
Decomposition¶
An operator that cannot be expressed in terms of operators aside from d and r is “irreducible”. For instance, k (Kis) and j (Join) are irreducible in terms of EROs, but m (Meta) is not (it is equal to kj). A polyhedron that cannot be expressed in terms of another polyhedron and one or more EROs other than the units S and d is an irreducible polyhedron. An interesting fact: the only platonic solid that is irreducible is the tetrahedron; the others can be expressed as some operation on the tetrahedron (O = aT, C = jT, I = sT, D = gT). Consequently, all of the Archimedean and Catalan solids can be expressed as some series of operators and T.
It is an open question whether the decomposition of an ERO in terms of other EROs is unique (up to associates). However, if we allow other operators, there are examples of a nonunique decomposition.
 xj = jk, where x is the PreJoinStake VAERO.
 j⅂ = lj, where ⅂ is the lozenge operator (a pseudoERO).
The relations defined earlier can be used to help reduce an operator, with some caveats. The above representations do not give us a 100% reliable way to decompose an arbitrary operator into a sequence of operators, it does suggest a (trialanderror filled) heuristic to reduce an operator into two operators by starting at the bottom of the homomorphism chain and going up.
 Determine the \(g\) of the two operators from the factors of the \(g\) of the operator to be factored.
 Determine the outline (\(a, a', c, c'\)) of the two operators.
 Determine \(b, b'\) for the two operators.
 Determine \(k, \ell, b_i, b'_i\). for the two operators.
 Figure out if the representations you’ve produced actually corresponds to an ERO.
Some facts relating to decomposition that can be derived from what we have so far:
 If a polyhedron has a prime number of edges, it is irreducible.
 Operators where g is a prime number are irreducible.
 If x=xd or rxr=xd, x has type 0.
 If x=dxd or rxr=dxd, x has type 1, \(g\) is odd, and \(b=b'\) is even.
 If an ERO has type 1, its decomposition cannot contain any EROs of type 0. Correspondingly, if an ERO has type 0, its decomposition must contain at least one type 0 ERO.
 There are no type 1 EROs with \(g=2\), so therefore type 1 EROs with \(g=2p\), where p is prime, are irreducible in terms of EROs. (However, see the section below, All EROs can be expressed with smoothing, an AERO, and the join operator.)
 \(\Box_{a,b}\) that correspond to the Gaussian primes, and \(\Delta_{a,b}\) that correspond to the Eisenstein primes, are irreducible in terms of EROs. (Proof below.) As a consequence of this, there are an infinite number of irreducible EROs.
Proof of the last statement: A Gaussian integer \(a + bi\) is prime if its square norm \(a^2 + b^2\) is prime or the square of a prime. In the first case, that prime has the form \(p=4n+1\); in the latter, \(p=4n+3\). Remember that the squared norm of the integer is just the inflation factor g for the corresponding operator. If g is prime, the operator is irreducible. If g is the square of a prime, the operator \(\Box_{a,b}\) is type 1, specifically, \(\det(\Lambda_{\Box_{a,b}}) = 1\). Suppose the operator can be decomposed into \(\Box_{a,b} = xy\), where x and y both have inflation factor \(g' = \sqrt(g)\). Without loss of generality, assume \(\det(\Lambda_x) = \det(\Lambda_y) = 1\). Their matrix forms are:
therefore, \(B+bg' = B'+b'g'\). It can be demonstrated using the ERO invariant inequalities from earlier that the only solution to this that could correspond to an actual ERO is \(b=b'\) and \(B=B'\). \(g' = p = 4n + 3\), so \(b, b', B, B'\) must all be odd. As mentioned earlier, there are no EROs with both b and g odd, so we have a contradiction, and \(\Box_{a,b}\) is irreducible.
The proof for \(\Delta_{a,b}\) is analogous. An Eisenstein integer \(a + bu\), \(u=\exp(\pi i/3)\), is prime if its square norm \(a^2 + ab + b^2\) is prime or the square of a prime. The prior (except for \((1 + u)\), which we corresponds to the ERO n which we already know is irreducible) have the form \(p=3n+1\); the latter, \(p=3n+2\). When the prime is of the latter form, the ERO is type 1 with \(\det(\Lambda_{\Delta_{a,b}}) = 1\) and its matrix form is:
Define x and y as before: then \(2(B+bg') = B'+b'g'\). Using the inequalities to exclude other choices, \(B' = 2B\) and \(b' = 2b\). g = 3n + 2, but g = b+ b’ + 1 = 3b+1: there is no simultaneous integer solution to both equations, so we have a contradiction, and \(\Delta_{a,b}\) is irreducible.
Extension  Operations on different polyhedra¶
The chamber diagram makes the assumption that each edge is acted on uniformly, and each edge of the seed polyhedron has exactly two adjacent faces. With some care, operators can be applied to any closed polyhedron or tiling that meets that criteria; toruses, polyhedra with multiple holes, planar tilings, hyperbolic tilings, and even nonorientable polyhedra, although the latter is restricted to the achiral operators. The graph of the polyhedron must be embeddable on a certain surface. Planar tilings may be easier to analyze by taking a finite section and treating it as a torus. It’s worth noting that applying \(\Delta\) to the regular triangular grid on the plane, or \(\Box\) to the regular square grid on the plane, just creates a topologically equivalent grid on the plane.
Convex polyhedra may be put into “canonical form” such that all faces are flat, all edges are tangent to the unit sphere, and the centroid of the polyhedron is at the origin. As a consequence, all faces are convex. There is no canonical form guaranteed to exist for general nonconvex polyhedra, however: in particular, there may be no position of the vertices such that all the faces are flat or convex.
Some operators can be applied to degenerate spherical polyhedra (dihedra and hosohedra) with a result that is a convex polyhedron. Specifically, operators with \(k > 1\) may create a convex polyhedron from a dihedron, and operators with \(\ell > 1\) may create a convex polyhedron from a hosohedron. (This is not guaranteed. For instance, try the lozenge operator on a dihedron: the result won’t even be 3connected!) For instance, a nbipyramid is a kis ndihedron, and (applying the dual) an nprism is a truncated nhosohedron. Therefore the octahedron is a kis 4dihedron and the cube is a truncated 4hosohedron. This is interesting because the octahedron is also an ambo tetrahedron, and the cube a join tetrahedron: if we admit degenerate polyhedra, there are some polyhedra with two unequal reductions into operators and seeds. We can also apply AEROs: an nsided pyramid is an alternatingtruncated nhosohedron. Thus a tetrahedron is an alternatingtruncated 3hosohedron (and a cube is a join alternatingtruncated 3hosohedron, etc.)
EROs may also be applied to surfaces with boundary, although the behavior
of the operator at the boundary needs to be specified. In general, this amounts
to dropping incomplete faces or faces that cross over the boundary, and dropping
some related edges and vertices. We lose the relationship with \(L_x\) and
\(M_x\) because not every edge is transformed in the same way: edges
adjacent to one face are different than edges adjacent to two. In general,
operators applied to a surface with boundary are not even associative. For
example, start with a single square face. Applying o splits the square into
four squares, but jj either annihilates the face entirely or creates stuff
outside the boundary of the original face, depending. (conway
in Antiprism
does the former.) EROs are also problematic on kaleidoscopes or other
tilings where 3 or more faces meet at an edge. (However, see the section
Extensions  Multiple chambers.)
Extension  Operations that alter topology¶
In the topology of surfaces, the connected sum A#B of two surfaces A and B can be thought of as removing a disk from A and B and stitching them together along the created boundary. If B has the topology of a sphere, then A#B has the topology of A: a connected sum with a sphere does not change the topology. The classification theorem of closed surfaces states that closed surfaces have the topology of either a sphere or a connected sum of a number of toruses and/or crosscaps.
In a topological sense, EROs and AEROs can be thought of as removing a disk from a surface and replacing it with the chamber structure. In a more elaborate sense, we can think of the operator chamber diagrams we’ve described so far (even the alternating ones) as having the topology of a sphere: identify the two edges on the left and the two edges on the right. Then, the operation can be described as taking the connected sum of the operator chamber diagrams with each face of the seed polyhedron. Thus one of the assumptions for EROs: taking the connected sum with a sphere does not change the topology, so the operation does not change the Euler characteristic.
However, operators that alter the topology can be described, introducing holes or other features to a polyhedron. This may require us to think of the chamber structure as having been extruded from a square into a square prism. One simple operator of this kind makes nested or offset copies of the polyhedron, which we’ll call “copy” and denote \(ɔ_n\). Obviously, this has \(M_ɔ = n M_S = n I_3\) where n is the number of copies produced, and \(k = \ell = 1\). As expected, the Euler characteristic of the result is the Euler characteristic of the seed times n. The n=0 case implies an “annihilate” operator that returns the empty “polyhedron” with 0 vertices, edges, and faces.
Another operator is the skeletonize operator depicted in skeleton
,
denoted ƨ (a reversed s).
Edges and vertices are retained, but faces are removed. The red crosses
indicate that the base faces are not retained or replaced with vertices: they
are removed entirely. If G is the genus of the seed polyhedron, the genus of
the resulting “polyhedron” (inasmuch as an object with no faces can be
considered a polyhedron) is G  f. The \(M_x\) form is obvious:
and \(k = \ell = 1\). (Technically \(\ell\) could be any value, but it makes sense to retain it as a measure of the hole created.)
Instead of annihilating the face completely, one can hollow out a space in its
center and leave behind a solid border. This can be done with the leonardo
command in Antiprism, or the hollow/skeletonize/h operator in Polyhedronisme
(not to be confused with the skeletonize defined above, or the semi operator
from the last section). Although the operations differ in exactly how the new
faces are specified, topologically they both resemble a process like so:
 Duplicate the polyhedron as a slightly smaller polyhedron inside itself.
 For each face, remove the corresponding faces of the larger and smaller polyhedra. Take a torus and remove its outer half. Stitch the upper and lower boundary circles of this torus to the larger and smaller polyhedra where the faces were.
To represent this, we have to extrude the chamber structure out into a sort of
3d prism. We’ll call this operator “hollow”, and notate it ɥ. It is not the
operation performed by leonardo
or Polyhedronisme,
but it can be represented in wythoff
using the notation
[V,VF]0_1v1_0v,1v1f,1V
). In terms of invariants, \(k=2\),
\(\ell=1\), \(b_5 = 2\), \(b'_4 = 3\), and \(M_x\) is:
If the seed polyhedron has Euler characteristic 2 (genus 0), the result has Euler characteristic 42f. The genus is f1, not f, because one torus is needed to connect the two copies of the sphere into a (topologically) spherical surface.
One could also create operators that add arbitrary numbers of holes per edge. (Operators that add crosscaps, e.g. based on a star polyhedron with Euler characteristic 1 such as the tetrahemihexahedron, may be possible. Such operators probably have more theoretical uses than aesthetic or practical ones, and good luck getting the faces to be flat and not intersect awkwardly.)
We can think of these more general topological operators as having the structure of a \(\mathbb{N}\)module over a semiring, where addition is the disjoint union of polyhedra, multiplication is composition of operators, and a coefficient from \(\mathbb{N} = {0, 1, ...}\) represents creating disjoint copies of a polyhedron. Then, the module has an obvious homomorphism with the 3x3 matrices.
Extensions  Multiple chambers¶
The concept of AEROs could be extended to kpartite graphs. \(k(k1)/2\) interrelated chamber structures would have to be specified, which would get a little unmanageable for large k. For example, if k=3, there would need to be 3 chambers: one on edges from set 1 to set 2, one from set 2 to set 3, and one from set 1 to 3. By the fourcolor theorem, the largest k that is necessary for a spherical tiling is 4, although larger k could be used.
Some EROs have forms where they are applied to only vertices or faces of a certain order, such as \(t_3\) to truncate vertices of order 3. These could be described by a set of 3 chamber structures: on an edge between order3 vertices, on an edge from an order3 vertex to a nonorder3 vertex (or vice versa), and on an edge between nonorder3 vertices.
It would also be possible to describe operations on surfaces with boundary this way: a chamber structure for an edge adjacent to 1 face, and a chamber structure for an edge adjacent to 2 faces. Kaleidoscopes where more than two faces meet at an edge are still problematic.
Because they don’t treat every edge uniformly, none of these schemes can be represented in the \(L_x\) or \(M_x\) forms defined earlier.
Listing of operators and transformations¶
Where not specified, \(k\) and \(\ell\) are 1, and
\(b_i\) and \(b'_i\) are 0. Remember that these lists only pick one
out of each set of associated operators. For operators that aren’t already
implemented in Antiprism, an input string to wythoff
is included in the
Notes section.
Operator x  Chiral?  Chambers of x  Matrix \(M_x\)  \(k, \ell\), \(b_i\), \(b'_i\)  Chambers of dx  Notes 

S (Seed, Identity)  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 0 \\
0 & 1 & 0 \\
0 & 0 & 1 \end{bmatrix}\end{split}\]

rr = S, dd = S  
j (Join)  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 1 \\
0 & 2 & 0 \\
0 & 1 & 0 \end{bmatrix}\end{split}\]

\(b'_4=1\)  j = jd = da = dad (jd=@j and ad=@a if considering partitions)  
k (Kis)  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 1 \\
0 & 3 & 0 \\
0 & 2 & 0 \end{bmatrix}\end{split}\]

\(k=2\), \(b'_3=2\)  k = nd = dz = dtd  
g (Gyro)  Y 
\[\begin{split}\begin{bmatrix}
1 & 2 & 1 \\
0 & 5 & 0 \\
0 & 2 & 0 \end{bmatrix}\end{split}\]

\(b_3=2\), \(b'_5=2\)  g = rgdr = ds = rdsdr  
p (Propeller)  Y 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 5 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(b_4=2\), \(b'_4=2\)  p = dpd  
c (Chamfer)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 4 & 0 \\
0 & 1 & 1 \end{bmatrix}\end{split}\]

\(b_3=2\), \(b'_6=1\)  c = dud  
l (Loft)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 5 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(k=2\), \(b_3=2\), \(b'_4=2\)  
q (Quinto)  N 
\[\begin{split}\begin{bmatrix}
1 & 3 & 0 \\
0 & 6 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(b_3=2\), \(b_4=1\), \(b'_5=2\)  
\(L_0\) (Joinlace)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 6 & 0 \\
0 & 3 & 1 \end{bmatrix}\end{split}\]

\(k=2\), \(b_4=2\), \(b'_3=2\), \(b'_4=1\)  
\(L\) (Lace)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 7 & 0 \\
0 & 4 & 1 \end{bmatrix}\end{split}\]

\(k=3\), \(b_4=2\), \(b'_3=4\)  
\(L_{1}\) (OppositeLace) (New)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 7 & 0 \\
0 & 4 & 1 \end{bmatrix}\end{split}\]

\(k=2\), \(b_5=2\), \(b'_3=4\)  [V, E2F] 1F, 1e1_0e, 0_1f1f, 1E  
\(K\) (Stake)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 1 \\
0 & 7 & 0 \\
0 & 4 & 0 \end{bmatrix}\end{split}\]

\(k=3\), \(b_3=2\), \(b'_3=2\), \(b'_4=2\)  
\(w\) (Whirl)  Y 
\[\begin{split}\begin{bmatrix}
1 & 4 & 0 \\
0 & 7 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(b_3=4\), \(b'_6=2\)  
\(E\) (Ethel) (New)  Y 
\[\begin{split}\begin{bmatrix}
1 & 4 & 0 \\
0 & 8 & 0 \\
0 & 3 & 1 \end{bmatrix}\end{split}\]

\(b_3=2\), \(b_4=2\), \(b'_4=2\), \(b'_6=1\)  [V, VE, VF] 0_1_2e1e, 2F, 1_2v2f  
\(J=(kk)_0\) (Joinkiskis)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 1 \\
0 & 8 & 0 \\
0 & 5 & 0 \end{bmatrix}\end{split}\]

\(k=3\), \(\ell=2\), \(b_3=2\), \(b'_3=4\), \(b'_4=1\)  
\(X\) (Cross)  N 
\[\begin{split}\begin{bmatrix}
1 & 3 & 1 \\
0 & 10 & 0 \\
0 & 6 & 0 \end{bmatrix}\end{split}\]

\(k=2\), \(b_4=2\), \(b_6=1\), \(b'_3=4\), \(b'_4=2\)  
\(W\) (Waffle) (New)  N 
\[\begin{split}\begin{bmatrix}
1 & 4 & 1 \\
0 & 9 & 0 \\
0 & 4 & 0 \end{bmatrix}\end{split}\]

\(b_3=2\), \(b_4=2\), \(b'_4=2\), \(b'_5=2\)  [V, E, F, V2E, VF] 0_4_3f4f, 2_4_3v3_4v, 3E  
\(B\) (Bowtie) (New)  Y 
\[\begin{split}\begin{bmatrix}
1 & 5 & 1 \\
0 & 10 & 0 \\
0 & 4 & 0 \end{bmatrix}\end{split}\]

\(b_3=4\), \(b_4=1\), \(b'_3=2\), \(b'_7=2\)  rBr=Bd [V, E, F, VE, EF] 1_3_4, 0_3_4_2e4_1_3e 
Operator x  Chiral?  Matrix \(M_x\)  \(k, \ell\), \(b_i\), \(b'_i\)  Useful relations 

\(m_n\) (Meta)  N 
\[\begin{split}\begin{bmatrix}
1 & n & 1 \\
0 & 3n+3 & 0 \\
0 & 2n+2 & 1 \end{bmatrix}\end{split}\]

\(k=2\), \(\ell=n+1\), \(b_4=n\), \(b'_3=2n+2\)  \(m_1 = m = kj\) 
\(M_n\) (Medial)  N 
\[\begin{split}\begin{bmatrix}
1 & n & 1 \\
0 & 3n+1 & 0 \\
0 & 2n & 1 \end{bmatrix}\end{split}\]

\(\ell=n\), \(b_4=n\), \(b'_3=2n2\), \(b'_4=2\)  \(M_1 = o = jj\) 
\(\Delta_{a,b}\) if T divisible by 3  If \(a \ne b\) and \(b \ne 0\) 
\[\begin{split}\begin{bmatrix}
1 & T/31 & 1 \\
0 & T & 0 \\
0 & 2T/3 & 0 \end{bmatrix}\end{split}\]

\(b_6=b\), \(b'_3=b'\)  \(\Delta_{1,1} = n\), \(\Delta_{a,b}\) \(= n \Delta_{(2a+b)/3, (ba)/3}\) 
\(\Delta_{a,b}\) if T not divisible by 3  If \(a \ne b\) and \(b \ne 0\) 
\[\begin{split}\begin{bmatrix}
1 & (T1)/3 & 0 \\
0 & T & 0 \\
0 & 2(T1)/3 & 1 \end{bmatrix}\end{split}\]

\(b_6=b\), \(b'_3=b'\)  \(\Delta_{2,0} = u\), \(\Delta_{2,1} = dwd\) 
\(\Box_{a,b}\) if T even  If \(a \ne b\) and \(b \ne 0\) 
\[\begin{split}\begin{bmatrix}
1 & T/21 & 1 \\
0 & T & 0 \\
0 & T/2 & 0 \end{bmatrix}\end{split}\]

\(b_4=b\), \(b'_4=b'\)  \(\Box_{a,b} = \Box_{a,b}d\), \(\Box_{1,1} = j\), \(\Box_{2,0} = o = j^2\), \(\Box_{a,b}\) \(= j\Box_{(a+b)/2,(ba)/2}\), (\(\Box_{a,b}d = @\Box_{a,b}\) if alternating vertices) 
\(\Box_{a,b}\) if T odd  If \(a \ne b\) and \(b \ne 0\) 
\[\begin{split}\begin{bmatrix}
1 & (T1)/2 & 0 \\
0 & T & 0 \\
0 & (T1)/2 & 1 \end{bmatrix}\end{split}\]

\(b_4\) \(=b'_4\) \(=b\) \(=b'\)  \(\Box_{a,b} = d\Box_{a,b}d\), \(\Box_{1,2} = p\) 
Operator x  Chiral?  Chambers of x  Matrix \(M_x\)  \(k, \ell\), \(b_i\), \(b'_i\)  Chambers of dx  Useful relations 

ɐ (Annihilate)  N  blank 
\[\begin{split}\begin{bmatrix}
0 & 0 & 0 \\
0 & 0 & 0 \\
0 & 0 & 0 \end{bmatrix}\end{split}\]

blank  ɐx = ɐ for all x  
ƨ (Skeleton)  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 0 \\
0 & 1 & 0 \\
0 & 0 & 0 \end{bmatrix}\end{split}\]

n/a  
ǝ (Edged)  N 
\[\begin{split}\begin{bmatrix}
0 & 2 & 0 \\
0 & 4 & 0 \\
1 & 0 & 1 \end{bmatrix}\end{split}\]

[VF] 0F, 0V  
⅂ (Lozenge)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 5 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(k=2\), \(\ell=2\), \(b_3=2\), \(b'_3=2\)  ⅂ = d⅂d [V, EF] 0_1F, 1_0f1f, 1E  
ɥ (Hollow)  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 7 & 0 \\
1 & 3 & 0 \end{bmatrix}\end{split}\]

\(k=2\), \(b_5 = 2\), \(b'_4 = 3\)  [V, VF] 0_1v1_0v, 1v1f, 1V  
\(ɔ_n\) (Copy)  N 
\[\begin{split}\begin{bmatrix}
n & 0 & 0 \\
0 & n & 0 \\
0 & 0 & n \end{bmatrix}\end{split}\]

\(ɔ_0 =ɐ\) 
In the following two tables, when \(k^+=k^\), both are written as just \(k\).
Operator  Degree2?  Chambers of x  Matrix  \(k_i, \ell_i\), \(b_i\), \(b'_i\)  Chambers of dx  Useful relations 

Alternation, Hemi, Semi  Digons 
\[\begin{split}\begin{bmatrix}
1 & 0 & 0 & 0 \\
0 & 0 & 1 & 0 \\
0 & 1 & 0 & 1 \end{bmatrix}\end{split}\]

\(k^+ = 2\), \(\ell = 1/2\)  $xj = S, $dxj = d  
Alternating Truncate (PreChamfer)  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 1 & 0 \\
0 & 0 & 2 & 0 \\
0 & 1 & 0 & 1 \end{bmatrix}\end{split}\]

\(\ell = 3/2\), \(b_3=1\)  xj = c, dxjd = u  
Prekis  Digons 
\[\begin{split}\begin{bmatrix}
1 & 0 & 0 \\
0 & 2 & 0 \\
0 & 1 & 1 \end{bmatrix}\end{split}\]

\(b'_3 = 1\), \(\ell = 1/2\), \(k^+ = 3\)  $xj = k  
PreJoinStake  N 
\[\begin{split}\begin{bmatrix}
1 & 1 & 0 \\
0 & 3 & 0 \\
0 & 1 & 1 \end{bmatrix}\end{split}\]

\(k^+=2\), \(b_3=1\), \(b'_4=1\)  xj = jk  
Alternating Subdivide  N 
\[\begin{split}\begin{bmatrix}
1 & 1 & 0 \\
0 & 3 & 0 \\
0 & 1 & 1 \end{bmatrix}\end{split}\]

\(\ell = 3/2\), \(b_4=1\), \(b'_3=1\)  
PreGyro  Degree2 vertices 
\[\begin{split}\begin{bmatrix}
1 & 1 & 1 \\
0 & 3 & 0 \\
0 & 1 & 0 \end{bmatrix}\end{split}\]

\(\ell = 1/2\), \(b_3=1\), \(b'_6=1\)  $xj = g. Not the same as PreJoinLace of dual.  
PreJoinLace  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 1 & 0 \\
0 & 0 & 3 & 0 \\
0 & 1 & 1 & 1 \end{bmatrix}\end{split}\]

\(k^+=2\), \(b_4=1\), \(b'_3=1\)  \(xj = L_0\). Not the same as pregyro of dual.  
PreJoinKisKis  N 
\[\begin{split}\begin{bmatrix}
1 & 1 & 0 \\
0 & 4 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(k^+=3\), \(k^=2\), \(b_3=1\), \(b'_3=2\)  \(xj = (kk)_0\)  
PreEthel  N 
\[\begin{split}\begin{bmatrix}
1 & 0 & 2 & 0 \\
0 & 0 & 4 & 0 \\
0 & 1 & 1 & 1 \end{bmatrix}\end{split}\]

\(b_3=1\), \(b_4=1\), \(b'_4=1\)  \(xj = E\).  
PreCross  N 
\[\begin{split}\begin{bmatrix}
1 & 1 & 1 \\
0 & 5 & 0 \\
0 & 3 & 0 \end{bmatrix}\end{split}\]

\(k^+=1\), \(k^=2\), \(\ell = 3/2\), \(b_4=1\), \(b'_3=2\), \(b'_4=1\)  xj = X  
Alternating Meta/Join  N 
\[\begin{split}\begin{bmatrix}
1 & 1 & 1 \\
0 & 5 & 0 \\
0 & 3 & 0 \end{bmatrix}\end{split}\]

\(k^+=1\), \(k^=2\), \(\ell = 2\), \(b_3=1\), \(b'_3=3\)  
Alternating Subdivide/Quinto  N 
\[\begin{split}\begin{bmatrix}
1 & 2 & 0 \\
0 & 5 & 0 \\
0 & 2 & 1 \end{bmatrix}\end{split}\]

\(b_3=1\), \(b_5=1\), \(b'_4=2\)  xj = jg 
Open questions¶
 Are there any irreducible EROs other than j that produce only quad faces?
 Are there any chiral EROs such that rxr = dxd? (They would have to be type 1 operators.)
 Are there other conditions that can be added to the invariants for \(L_x\) to make the set of conditions sufficient as well as necessary?
 Is there an invariant related to the chirality of an operator?
 What other invariants need to be added to fully characterize EROs and AEROs?
 Are there any EROs that can be reduced two (or more) unequal ways into EROs?