You are on page 1of 27

International Journal of Mathematical Education in / i ^ \ Taylor & Francis

Science and Technology. Vol. 36, No. 6. 2005. 617 642 /

Duality of polyhedra

P. GAILIUNASt and J. SHARPJ*


t25 Hedley Terrace. Gosforth. Newcastle, NE3 lDP. UK
J20 The Glebe. Watford. Herls.. WD25 OLR. UK

(Received 14 December 2003)

Everyone is familiar with the concept that the cube and octahedron.
dodecahedron and icosahedron are dual pairs, wiih Ihe tetrahedron being self-
dual. On the face of it, the eoneept seems slraighlforward: however, in all but the
most symmetrical cases it is far from clear. By using the computer and three-
dimensional graphics programs, it is possible to elarify the concept and explore
new ideas. Moreover, it is an ideal topic for teaching clear logical thinking.

1. Introduction

Duality is usually taught initially as a property ofthe Platonic solids, by pointing out
that that the number of vertices ofthe cube and the number of faces ofthe octahedron
are equal and vice versa. Similarly, the dodecahedron and icosahedron are dual pairs
and the tetrahedron is self-dual. It is common to see drawings of this with, for
example, a regular octahedron with its vertices sitting at the centre of a eube. This
works for the regular polyhedra because of their symmetry, and is often cited as a
way of obtaining the dual: for example Holden [1. p. 4]. uses it as a means of
introducing duality. However, even the duals ofthe Archimedean solids (the Catalan
solids) cannot be constructed by joining the centroids ofthe faces, but you can jo\n
the points where the faees of a Catalan solid touch the insphere to get its metric dual
(the original Archimedean). To be fair to Holden, he does not discuss the joining of
the centre of laces when describing the Archimedeans and tbeir duals, but he does
not make it explicit in his first definition that his description is a special case. This is a
typical way that the half truth is perpetuated and why we have provided detailed
examples, many of them non-symmetrical.
In discussing why this is the case, and looking through the literature, we have had
diffieulty in finding a clear exposition of the subjeet. Even the relatively recent book
on polyhedra hy Cromwell [2] shies away from considering duality, which seems
most unusual in such a comprehensive book, but is indicative ofthe lack of suitable
souree material. We were surprised that despite the idea being quite simple,
discussion of the finer detail is often confused. There is the further complication
that in some papers, the original defmition is used in a way that does not follow the
mathematical convention, although the subject matter is perfectly valid in its own

*Corresponding author. Email: sliceforms(fl compaserve.eom


liiwriiarioniil Jotirmil of Miillwnianml Edm miim in Sciviur und Tcclmuloay
ISSN (K12O-7.19X prinilSSN I4M-5211 online c 2005 Taylor & Francis
hup;,'; www.iiindf.co.uk. journal
DOI; 10.1080,00207390500064049
618 P. Gailiuna.s and .1. Sharp

right (for example dc Villiers [3] has generalized van Aubel's theorem using a non-
standard concept of duality). Because they are not conventional, we will not cover
them but stick to a discussion of duality which is precise in defining terms and in
using concepts in order to avoid confusion, and possibly errors. Another potential
source of confusion is illustrated by Bakos [4], where he swaps the words cube and
octahedron in discussing some dual configurations, without going into the finer
detail of what he is actually discussing. So he claims that finding a configuration ofa
cube inscribed in an octahedron, with its vertices on the edges ofthe octahedron, is
the dual problem of fmding a configuration of an octahedron inscribed in a cube,
with the octahedral vertices on the edges of the cube. Grunbaum and Shephard [5]
discuss some aspects of these difficulties. However, in their resolution to demonstrate
the extent of the muddle, they have potentially added to it, in some instances by
failing to make clear which meaning they intend, and also by giving the impression
that they accept certain poiyhedra to be duals, based on a mistaken assumption. We
will deal with these cases in detail later.
This description of the duality of poiyhedra has been made much easier by the
use of computers to produce objects we can see and manipulate in three dimensions
on the computer screen. Apart from the visual attraction of this as a teaching tool,
the use of technology is an aid to the use of clear and logical thinking. We had to
produce objects dual to the ones we were studying, so we had to be clear about what
we were trying to achieve, in order to construct an algorithm. This points out
another lesson that having to work from scratch is a better aid to learning than
pressing buttons and entering data in someone else's program.

2. Poiyhedra

Coxeter 16] defines a polyhedron as 'a finite, connected set of plane polygons, such
that every side of each polygon belongs also to just one other polygon". He excludes
anomalies like pairs of pyramids with a common apex by including that the polygons
surrounding each vertex form a single circuit. The polygons arc face.s of the
polyhedron, and to add clarity we will distinguish their side.s from edges of the
polyhedron, since an edge corresponds with sides of two polygons. Similarly corners
of polygons arc distinguished from vertiuw of the polyhedron.
In considering duals of some poiyhedra we will need to include examples where
the faces intersect one another, and where the faces may also have sides that
intersect. This needs much wider definitions of polygons and poiyhedra.
Grunbaum [7] has developed suitable rigorous definitions by treating polygons as
circuits, so that a convex polygon is a circuit where each corner is visited once, and
none of the sides cross, while a star polygon has sides crossing. Poiyhedra are
constructed by joining such polygons along their sides, and Grunbaum gives many
examples of the range of poiyhedra allowable under these definitions. We will also
need to consider polygons having sides of zero length, so that different corners are
not necessarily at distinct points, and polygons that are actually different can look
the same. A hexagon can look like a triangle, for example, if three of its sides are of
zero length.
We will also need to consider the planes that contain the faces, the (infinite) lines
that contain the edges, and the points that contain the vertices of poiyhedra. A point
can have none, one or more vertices.
Duality of polvheclra 619

3. Types of duality

The term duality is used in a range ofcontexts. and its meaning varies, so we need to
be clour about whieh one is being diseussed., since a statement can be valid il' one
definition is used and invalid in another case. Many problems arise by using the word
duality in an imprecise sense and in some cases by using the term in more than one
sense without informing the reader that the sense has changed.

3.1. Combinatovial duality


A simple instanee of duality occurs in the theory of planar graphs. Duality of graphs
is usually defined in set theoretic terms, but there is a more intuitive construction.
Given a eonneeted graph consisting of a set of vertices connected by ares that divide
the plane into several regions, a vertex of the dual is identified with each region ofthe
original graph, and ares ofthe dual constructed between these vertices if, and only If.,
the corresponding regions have a common are as a boundary. This gives a one to one
eorrespondencc between ares ofthe graph and its dual, between regions ofthe graph
and vertices of the dual, and between vertices of the graph and regions of the dual.
This determines the numbers of elements (vertices, edges and so on) and their
connectivity, but not other geometric properties, e.g. convexity.
This idea ean be applied in a very natural way to polyhedra that are simply
connected and have no interseeting faces, since they can be mapped onto a planar
graph (the Schlegel diagram), but il cannot apply to more general cases, for example,
those having faces that are star polygons., sinee such polygons are non-planar. A
natural extension, however, considers a pair of polyhedra to be combinatorial duals
if there is a one to one correspondence between the faces of one and the vertiees of
the other sueh thai faees that share an edge in one correspond to vertices that are
joined by an edge in the other,
A simplistic attempt deriving from these ideas leads to the supposed duals
constructed by identifying the centres of faees with dual vertiees. This will produee
forms of the accepted duals of the Platonic polyhedra, since each has an insphere
that touches its faees at their centres, but fails even with the Arehimedeans, when the
face eentres may not be eoplanar.

3.2. Projective duality


The axioms of projective geometry in the plane that two points are incident with just
one line and that two lines are incident with just one point imply the principle of
duality:
Every definition remains significant and every theorem remains true by virtue
of its dual theorem when the two pairs of concepts pointlline and join/intersect
are exchanged.
The principle of duality in the plane is properiy attributed to Gergonne (1826). Its
extension to three-dimensional spaee and placing on analytical footing is due to
Mobius 18].
Employing the principle to obtain a dual theorem, or deseription we interchange
aeeording to a wider 'dictionary' than the one above. In the plane, the dictionary
620 P. Gailiunas and J. Sharp

includes the following, with an item from either column being replaced by the
corresponding one from the other

point line
passing through lying on
concurrent collinear
join intersection
point on a curve tangent to a curve
range {all Ihc points in a line) pencil (all the lines in a poini)
locus envelope

When we move away from the plane some items in the dictionary change., for
example with three-dimensional space we have:

point plane
Line line
passing through lying on
coplanar points concurrent planes
coaxial planes collinear points
join intersection
range (all the poinls in a line) pencil (all the planes in a line)

Points and planes are now duals, and lines are dual to lines. As wilh combina-
torial duality, we are only establishing a correspondence, albeit a more rigid one. In
combinatorial duality, lines are lopological 'rubber" line segments. In projective
duality, lines are infinite straight lines, what would have been called right lines two
hundred years ago to distinguish them from curved lines. In both cases we can create
figures that are duals, but in neither case can we construct or calculate one from the
other.

3.3. Reciprocity or polarity in a conic or quadric


In projective geometry, it is possible to construct specific figures, which retain the
above "dictionary" properties, by making use ofthe pole and polar construction with
respect to a conic in the plane or quadric surface in space. This is really a fortuitous
coincidence, but as Pedoe [9] in a paper entitled 'Notes on the history of geometrical
ideas: II. The principle of duality", describes, it caused a great deal of confusion in
the nineteenth century. When Gergonne described the principle of duality, Poncelet
(one of the giants of geometry who were responsible for reviving geometry after it
had been subjugated by Cartesian geometry) protested that it was nothing more than
his method of reciprocation (polarity) with respect to a conic. The debate became
very acrimonious with Ponceiet devoting large parts of his books to denouncing all
the major mathematicians for stealing his ideas.
The confusion between duality and reciprocation still exists in many books and is
part of the cause of the problems we are trying to resolve. For a discussion of the
difference see Coxeter [10]. quoted by Pedoe as concluding that. "Since the general
Pappus configuration is self-dtial without being self-polar, the old controversy
between Poncelet and Gergonne is settled in the lattcr's favour.' For our purposes
the important point is that reciprocity is a more stringent requirement than duality,
and figures that are reciprocals are necessarily dual.
Pole and polar are terms which fit in the dictionaries above. In the plane
the relation is between a polar point and a polar line, with respect to a conic.
Duality of poiyhedra 621

Figure 1. Pole and polar.

In three-dimensional space it is between a polar point and polar plane with respect to
a quadric surface. The important point to realise is that in each case there is a one to
one mapping of the points and lines in the plane, and of the points and planes of
three dimensional space. It is not difficult to construct the polar line ofa point with
respect to a conic if the point is outside the conic and it is possible to construct
tangents to it: draw two tangents to the conic from the point and join the points of
tangency. The construction ofthe pole is as easy if the line cuts the conic in two real
points: intersect the tangents at those points. For the general case (figure 1), the
double column method which was popular in projective geometry books shows both
the power of the concept and its elegance:

To find the polar line of point X To find the pole of line x


1. Choose any two lines (ii and b) 1. Choose any two points A and B on
through X whicb intersect the line X which have real tangents with the eonie.
conic in real points.
2. Intersect these lines and the 2. Drtiw the pair of tangents from each
conic in points P. Q (for a) point, p. q (from A) and r. s (from B).
and R, S (lor b).
3. Draw Ihc tangents p. q. r. s at 3. Draw the tangent points of these
P, Q. R and S. lines P. Q. R and S for p. q, r, s.
4. Intersect tangents p, q in A 4. Join points P. Q in line a and R, S
and r. s in B. in line b.
5. Join A and B to give the polar line x. 5. Intersect a and b to give the pole X.

Note how figure I is a self-dual diagram; it applies to both constructions.


Consequently, it is easy to see that the construction is reversible, for example if
you construct the polar line ofa point with respect to a conic and then construct the
pole of a line with respect to the same conic, you will arrive back at the original
point.
This is fine for constructing pole and polar in the plane, and although a similar
construction can be applied in space, drawing in space is not as easy. The
construction of the reciprocal of a polyhedron in general becomes quite involved,
so the normal convention uses a sphere as the reciprocating quadric. This has the
benefit that the line joining the centre of the sphere and the pole is perpendicular to
the polar plane.
622 P. (iaitiuna.s and ./. Sharp

The symmetry of the sphere also makes it easier lo visualize the result of
reciprocation, but applying it to the Platonic solids, and extrapolating the results
is a root of some ofthe problems. In the ease of Platonic poiyhedra the sphere can be
ehosen to be the insphere, so that the vertices of the dual lie at the centroids of the
faces; the circumsphere. so that the faces of the dual lie in the tangent planes at the
vertiees; or the sphere touehing the mid-points of the edges, so that the edges of
the dual pair mutually intersect at right angles. These relations are a consequence of
the high symmetry of the Platonic solids, and it should not be assumed thai they
extend to polyhedra in general.
Duals of symmetrical polyhedra have been widely studied, usually understood to
be reciprocals with respect to a sphere. Wenninger [II]., for instanee, has given
detailed direetions for building models of sueh duals ol' uniform polyhedra.
Even though rcciprocallon would seem to define an essentially unique dual the
situation is not so clear in less symmetrical cases, when, in the absenee of a centre of
symmetry, there may not be an ohvious centre for the sphere. Using a different centre
(or indeed a quadric different from a sphere) produces a polyhedron that has a very
different appearance. Changing the radius of the sphere simply gives a similar
reciprocal, differing only in scale.

4. Misleading cases

We can now return to the examples in the introduetion.

4.1. Joininfi the centres of the faces of a polyhedron gives a dual


It is certainly the ease that joining the centroids of the faces of the Platonic solids
gives the metrieally regular reciprocal dual. For example joining the centres of faces
of a cube gives a regular octahedron, and the insphere of the cube becomes the
circumsphere ofthe octahedron.
Going a stage further to the Arehimedean solids no longer gives the correct
accepted dual (the appropriate Catalan polyhedron). This is because the faces of the
Archimedean solids have no eommon insphere. In faet the eentroids ofthe faees of
an Archimedean polyhedron having a eommon vertex are not in general eoplanar. so
the construction does not even give a polyhedron. Reciprocation will always ensure
that concurrent planes give rise to eoplanar points in the dual, so this problem will
never arise if it is used. In fact this is the only mapping between points and planes
that can achieve this, because any sueh mapping eould then be followed by a
reciprocation, so that concurrent planes would have been mapped to eoplanar
points, and then back to concurrent planes. Similarly eoplanar points would be
mapped to concurrent planes, then baek to eoplanar points. The situation with
collinear points/planes would be similar, so that the combined mapping would
preserve incidences, and be a projective transformation. This means that the original
mapping is equivalent to a projeetive transformation, followed by the inverse of a
reciprocation (whieh is also reciprocation). But this combination would simply be
another reciprocation.
The natural requirement that eoplanar points are dual to concurrent planes has
an important consequence when applied to polyhedra having eoplanar faces, for
example figure 2. Clearly all the vertiees of eoplanar faces are themselves eoplanar.
Duality of polyhedra 623

(a) (b)
Figure 2. Reciprocal ion with copianar faces.

so Ihc corresponding dual faces are concurrent al a point that is dual to the common
plane. This means that the vertices dual to the copianar faces are all at the same
point, although we need to distinguish them since duality is a one to one mapping
between faces and vertices.
In figure 2(a). the two triangles in the same plane at the left become two vertices
in the same point at the lower left of figure 2(b),
Although duals of the Archimedeans cannot be constructed by joining the
centroids of the faces, because their vertices lie on a sphere each Catalan solid has
an insphere. and you can join the points where the faces of a Catalan solid touch the
Insphere to get its metric dual (the original Archimedean).

4.2. The Bakos configurations


By calling them solutions to dual problems Bakos [4] gives the impression that the
following two compound polyhedra are a pair of duals. The first is a regular
octahedron inscribed in a cube so that its vertices are one on each of six edges of
the cube. The second consists of a cube inscribed in a regular octahedron so that its
vertices are one on each of eight edges ofthe octahedron.
Analysing through the logic of transliteration according to the projective
dictionary for his first figure using the "two-column" method yields the following
result:

Hjjure 3(a) Projective dual


1. tach vertex ofthe octahedron lies on the 1. Each face ofthe cube musl lie on the edge
edge of a cube. of an octahedron.
2. Each face of the cube has one edge of the 2. Each vertex ofthe octahedron must lie on
octahedron in common. one edge of the cube,
3. None of the vertices of the cube lie on 3. None ofthe faces ofthe octahedron must
edges of the octahedron. lie on edges of the cube.

Comparing the statements in both columns shows that figure 3(a) is self-dual and
cannot be the dual of figure 3(b). We leave it to the reader to show that figure 3(b) is
also self-dual. In both cases the cube and octahedron have a common centre, so
reciprocation in a sphere can produce an octahedron and a cube simultaneously, and
both compounds are actually self-reeiproeal. It is interesting to note that if you do
perform reciprocation in a sphere, then the resulting figure is rotated. In figure 3(a).
624 P. Gailiunas und J. Sharp

(a) (b)
Figure 3. Bakos configiiralions.

(a) (b)
Figure 4. Griinbaum and Shephard examples.

it is rotated 60 about the common 3-fold axis. In figure 3(b), it is rotated 45 about
the common 4-fold axis.

4.3. The Gruttbaum and Shephard examples


Grunbuum and Shephard [5] consider several poiyhedra and discuss the possibility
of constructing their duals. Their first figure shows a pair of poiyhedra which they
state to be duals of each other. It is easy to see that they arc combinatorial duals
since these poiyhedra are topologically equivalent to the dodecahedron and icosahe-
dron. Figure 4(a) consists of twelve (non-convex) pentagons with three at each vertex
and figure 4(b) of twenty triangles with five faces at each vertex. Although the article
sets out to demonstrate that in general there is no correspondence between the
convexity properties of a polyhedron and its dual, in this case they comment there
Duality of polyhedra 625

are analogous symmetry and convexity properties, and leave the impression that the
examples could be duals in some stronger sense, but they are certainly not reciprocal.
Figure 4(a) can also be considered in the following way as described by Ounsted
112] in a paper entitled 'An unfamiliar dodecahedron". The construction of a
dodecahedron (which goes back to Euclid) consists of adding a set of "roofs' to a
cube. If these roofs are inverted, so that they are subtracted from the cube instead of
adding to it. then the result is figure 4(a). For this reason, we will refer to this
polyhedron as the inverted dodecahedron.
Figure 4(b) can be considered as an icosahedron where three pairs of edges,
defining orthogonal planes, have been pushed towards the centre. Jessen 113] wrote a
paper about orthogonal icosahedra and so it is generally known as the Jessen
icosahedron. We have drawn it slightly differently to the illustration in Grunbaum
and Shcphard [5 figure 13-la]. not only to make it more easily understood, but also
deriving it from a regular icosahedron. so that when we create the reciprocal
(figure 16). we get a polyhedron that is related to a regular dodecahedron. We
have done no more than make a shallower depression than they have.
Grunbaum and Shephard say that "there is no difficulty finding topological
complexes that are duals of any given polyhedron, but finding a dual polyhedron
(their italics) is a much more illusive goal' [5, p 207]. They go on to show that in
general this goal is unattainable if the dual is required to satisfy the usual definition
of a polyhedron. Earlier on. however, they say [5. p 206] "reciprocation is the only
(their italics) known method of actually constructing a polyhedron P* dual to a given
polyhedron P'. and we find it curious that they did not actually try to develop this
approach, because as we show later, it yields interesting and illuminating results. It
shows that figures 4(a) and 4(b) are not reciprocals, although their reciprocals are
interesting in their own right and are shown later (figures 15 and 19).

5. A geometrical construction for reciprocals

There is a geometrical construction for the duals of polyhedra attributed to Dorman


Luke which is mentioned in Cundy and Rollett [14] and described in detail in
Wenninger 111]. It is not an easy method to use and is only suitable for convex
polyhedra being reciprocated in a sphere. It relies on taking the vertex figure (the
polygon obtained by joining points along the edges around a vertex at a fixed
distance from the vertex) and circumscribing the polygon. Tangents are then drawn
at each vertex ofthe polygon (these are the polar lines of these points in the circle) to
construct the reciprocal polygon. This method gives the shape of the faces of the
reciprocal polyhedron, but it does not give the spatial coordinates and much more
work is required to create three dimensional models in a CAD system. Because it is
so restricted, and too cumbersome to allow us to explore reciprocals quickly and
easily, we refer the reader to Wenninger III].

6. Determining the reciprocal by calculation

Cartesian coordinates in Euclidean space use three variables v. v. and r to denote the
position of an absolute position of a point in space relative to an origin. Projective
geometry is not normally concerned with absolute positions, but in geometrical
626 P. Gailiunas and J. Sharp

properties which are invariant under projection, so that, for example, a circle, an
ellipse, a parabola and a hyperbola arc all projcctively equivalent.
To work in projective space, we need to use homogeneous coordinates. Since we
are using this coordinate system as a tool, we will only give some basic background
for their use, and simplify the concepts for our needs without loss of rigour. We
suggest that the reader goes to a book on projective geometry such as Coxeter [10] or
Maxwell [15] for a detailed explanation.
Homogeneous coordinates extend the Euclidean concept of a plane and space to
infinite or ideal positions. If we arc totally concerned with the fmite. we can work
with them so as to maintain an equivalence. Using homogeneous coordinates the
position ofa point in a plane is uniquely defined by the ratio of three coordinates and
in space by four. So in a plane a point's position has Ihree coordinates .v, v, and r. To
convert homogeneous coordinates ofa two dimensional pomt (.v. y, z) to Cartesian
coordinates, simply divide all coordinates by the = value to give (.Y/Z, y/z, 1).
One ofthe beauties of using homogeneous coordinates is that the dual behaves in
the same way. For example a line in the plane ean be also be considered as the ratio
of three coordinates (.V, K, Z). Then one equation surfices to describe dual situations.
For example, the following "linear" equation can take dual meanings:

.vA" + y K + zZ = 0

• if A", y and Z are constant, it describes a line as a range of points (all the points
on the line)
• if .Y, _r and z are constant, it describes a point as a pencil of lines (all the lines
though the point)
For our purposes, dealing with points and planes in space, we need four coordinates
for either. In common with the usual convention in writing homogeneous coordi-
nates we will now use (.Y|. .V^, .Vi, A-4) to denote a point or a plane as we choose. This
has benefits in being extensible to any dimension with the added benefits that you
can quickly see which dimension you are working in. It also makes computer
programs easier to write. Lines in polyhedra are sides of faces; they are drawn
by computer as joins of points and polygons are specified as an ordered array of
points.
In programming terms, the duality of using (.V|, .YI, .V^, .Y4) to denote a point or a
plane as we choose means we can use one subroutine to perform operations on
points and planes in space, (and similarly on points or lines in the plane) as will
becotne clear in the algorithm for calculating reciprocals. Projective geometry using
homogeneous coordinates is very common in computer graphics; for example see
Penna and R Patterson [16].

6.1. Data ased in the algorithm


To calculate the reciprocal, we need the following sets of information which are
specified as a series of arrays.

Data Used for


coordiiiittes of vertices position of vertices
vertices in a face plane coordinates of (aces
order of faces around a vertex order of vertices in a reciproeal face
Duality of polytwdru til

6.2. Alfiovithm for computing the reciprocal


Since CAD programs can perform visualisation with the ability lo alter the view,
perform hidden line resolution and so on. we have programmed to write dala which
is in Ihc CAD program format. The major part ofthe program for creating the file
for loading into a CAD program is concerned with loading the data describing the
original polyhedron and writing it out in the required format. This can take so many
forms (and will change in the future), so we will only provide the essentials part of
the program as the following steps in pseudo-code, together with the algebra for the
calculations. Not everyone has access to a CAD system, but the VRML language is
an alternative whose viewer is obtainable free on the internet (from the Cosmo
software or Cortona sites) and the tile is simple and easy to edit manually. The
calculations themselves are simple enough to be made in a spreadsheet. So we have
provided an appendix on how lo work with a combination of spreadsheet and
VRML which could easily be used in a classroom environment. The algorithm is as
follows:

read in the data as described above


convert the Cartesian coordinates to homogeneous coordinates
for the first three coordinates of each face
calculate the plane coordinates of the face for each face
use this to calculate the coordinates of the reciprocal
point for each vertex
for each face in the ordered set find the reciprocal point
convert the homogeneous coordinates to Cartesian
in the same order assemble the vertices of the reciprocal face
write out the faces as required by CAD programfileformat
Note that to determine the coordinates of a face, you only need three points, just as
three planes uniquely determine a point. Care must be taken to choose three points
that are not all in a line, otherwise the plane cannot be defined. If the face is
degenerate such that there are only three points which He in a line, then the dual
ofthe tace is also degenerate as three planes in a line. If the face is degenerate so that
it shrinks to a single point., then the dual is degenerate as a single plane. In these
cases, a vertex does not have a defined position, but is limited to a line or plane
respectively.
A critical aspect of this algorithm is finding the order ofthe corners of each face
of the dual, which is then determined as a circuit, consistent with Grunbaum's
definition of a polygon mentioned above. It is the same as the order of the faces
around the corresponding vertex in the original polyhedron, which can be deter-
mined by observing that adjacent faces have a common edge.
Because we are discussing duality, it would be equally valid to find the reciprocal
of a point as a plane and assemble the polyhedron with more calculation, in practice,
computer programs accept data as sets of points and so the above algorithm is more
cllicient and easier to program. It is also easier to visuahse and write the algorithm
because o'i this.

6.3. Projective calculations used in the alf^orithm


The three calculations required in the algorithm arc relatively simple. Since they are
tools we will only provide the results. Many standard books on projective geometry
628 P. Gailiunas anil J. Sharp

and homogeneous coordinates such as Smith [17] and Maxwell [15] can be accessed
for further information.
The simplest calculation is the transformation between Cartesian and homo-
geneous coordinates. We simply use the value I for the fourth coordinate. That is a
point whose Cartesian coordinates have values (v. v. ~) has homogeneous coordi-
nates with values (.v, _v, z. 1).
When we have to convert back, we normaHse the fourth coordinate and then only
use the first three as our Cartesian coordinates. That is:
(.Y[, .YT, .V3. .Y4) becomes {.V| /.Y4, .Y2/.V4, .Y3/A4)

Having found homogeneous coordinates for the vertices the next job is to find the
coordinates of the planes containing the faces. If three points have coordinates (fli,
"2> w_i, «4), (/)], /j2, by. h^), (C|. ts, Ci. ('4) then the coordinates of the plane (,YI, .YI, .YI,
X4) they define are given by the following determinants:

fli «4 a\
/)4 /j|

CA C\

XT, = •V4 ^ b,
CA

Determining the reciprocal plane of a point with respect to a quadric is a standard


procedure, and is covered in standard texts, for example Smith [17], so with
eonversion of the result to homogeneous coordinates, we get the result that the
reeiprocal plane of the point (fl|, «2- ^3^ "4) with respect to the general conicoid:

/).Y' + By- + Cz- + Div- + 2Py: + 2Gz.\ + 2H.xy + 2UtL\w + 2 Vyw -\-2W=\v = 0

(where the uppercase letters are constants) has coordinates

iAai + Ha2 + Ga^ + Vw), iHa\ + Bih + Pciy + VOA). {GOX + Pcij + Ca^ + Wu^),

if we simplify this to a sphere centred at the origin

A-- ++ rr -- R^w- = 0
then the reciprocal plane has coordinates

where R is the radius ofthe sphere. Since we are using homogeneous coordinates the
dual operation is identical in form, so if we are finding the reciprocal of a plane with
coordinates {.YI, .Y^, .V^, Y4), this gives us a pole with coordinates (A-|, A^, X^-R^XA) or
Cartesian coordinates (x^j R^x^, X2J-R~X4, x^j-R'^x^).
This result shows that the size of the reciprocating sphere has no effect on the
shape of the reciprocal polyhedron but only on its size by virtue of -R'^ being the
factor which decides the size.
Duality of polyhedra 629

7. Results and some new polyhedra

Now we have a simple program lo calculate the reciprocal ofa polyhedron, we can
explore quickly and easily by small variations in a way that would be difficult and
tedious by conventional geometric methods. But first, to enable interpretation of the
results, we need to consider some more general polyhedra.

7.1. Non-convex polygons and polyhedra


In his consideration of definitions of polyhedra, Cromwell [2. p. 205. figure 5.15]
shows two cases of polyhedra which were recorded in 1830 by the minera-
logist Hessel as counter-examples to Euler's formula which we have redrawn in
figure 5.
Most definitions of a polyhedron, such as Coxeter's. avoid the difficulty by
excluding them. However, if you were to make them as models (for example using
bamboo skewers) they can be seen as being of dilTerent types. The one on the left
would not be rigid, but could flex about edge BD. We need not consider polyhedra of
this type. Polyhedra that have intersecting faces (such as figure 19) look rather like
this, but the intersection is not an edge, and no flexing is possible. The one on the
right (figure 5(b)) could be made so that AB. CD and EF were straight lines such that
AFEB, CEED and ADBC are intersecting quadrilaterals. The figure is constructed
by joining polygons along their sides, and so qualifies as a polyhedron under the
wider defmition mentioned earlier. Point G is the point of intersection, but not
a vertex. The polyhedron is thus a type of prism. It has three four-sided faces and
two triangular ones, six vertices and nine edges and thus conforms to Euler's
formula.
The way the sides ofthe quadrilaterals intersect is analogous to the way the sides
of a star polygon do. Polyhedra with intersecting faces using pentagrams and
triangles are well known in the Kepler Poinsot polyhedra. Such intersecting poly-
hedra will appear when we try to form reciprocal polyhedra.

(a) (b)
Figure 5. Hcssel's counter examples to Euler's formula.
630 P. Gailiuna.s and J. Sharp

(a) (b)
Figure 6. A polyhedron and its reciprocal, taken from Grunbaum and Shephard.

7.2. More Griinhaum and Shephard 'duals'


This problem of intersection arises in Grunbaum and Shephard's next set of
examples [5]. In their figure 13-2 (on p. 208) they show five poiyhedra. which are
apparently simple cases, and discuss their duality. We will start by considering the
first pair of these in figure 6.
Figure 6(a) (ihcir fig. 13-2a) consists of a polyhedron made from a cube with a
square pyramid {part of an octahedron) erected on one face. We can visualize the
dual by recalling that the dual ofa cube is an octahedron and hence vice versa. The
apex (point) of the pyramid gives rise to a new face (plane) of the dual (figure 6(b)),
which is a square, since the vertex is 4-valent, and the edges have square symmetry.
Four faces of the dual become trapezia instead of triangles, corresponding to the
vertices where the pyramid joins the cube, which are 4-valent. The four points at the
base of the cube become four triangles meeting at the point which is dual to the base
of the cube. The resulting dual shown in figure 6(b) (which is Grunbaum and
Shephard"s figure l3-2b), they point out. gives a pair which is combinatorially/
projeetively equivalent, and so they are self-dual. In figure 6(a) we have added a
point to the cube while at the same time losing a face. This means in the dual
(figure 6(b)) we have added a plane to an octahedron which has lost a vertex.
Note that the reciprocal has been calculated with respect to a sphere centred in
the middle of the cube.
If the apex is moved along the axis of symmetry, its dual (the square plane at the
top in figure 6(b)) will move also, intersecting the octahedron at different places to
give a varying square face. The higher the apex, the nearer the face to the mid-plane
of the octahedron. If the apex is lowered so that it sits in the top faee of the cube,
then the plane cuts the octahedron at its vertex, and the four corners ofthe square,
which has sides of zero length, are all at the same point.
If the apex is moved further towards the centre of the cube (giving figure 7(a)
which is like Grunbaum and Shephard's figure 13-2c). the plane moves further away
from the mid-plane of the octahedron, and a polyhedron is produced having faces
that are self-intersecting trapezia.
The reciprocal in figure 7(b) needs to be thought of in the same way as the Hessel
polyhedron in figure 5(b). The 'extra vertex' which looks as if it is the top vertex of
an octahedron is something we can perceive although it is not a vertex of the
Duality of polyhedra 631

(a) (b)
Figure 7. Another Grunbaum and Shcphard example, with its reciprocal.

(b)

(c) (d)
Figure 8. The reciprocal :is the apex in Figure 7(a) is depressed further.

polyhedron we have constructed. Its dual is the top 'extra face' of ihc cube in
tigure 7(a) which we can imagine but not touch.
If the upex is moved to the centre of the cube, and we continue to create the
reciprocal with respect to a sphere centred at the centre ofthe cube, then the dual of
the apex moves to infinity. Figure 8(a) shows one representation of this case. It
would be equally valid to draw the infinite lines to the square at infinity above the
half octahedron.
If we take the apex down past the centre ofthe 'cubic* part, then the square in the
dual comes back from underneath (figure 8(b)). The self-duality then becomes more
evident. If the apex is taken below the base ofthe cube then the dual is similarly self-
dual. Figures H(c) and 8(d) show the same polyhedron without, and then with,
hidden line removal. Figure 8(c) has had no line thickness adjustments to attempt to
show the way the lines are hidden, neither does it show the lines resulting from
intersection with the sqtiare visible in figure 8(d).
632 P. Gailiunas and J. Sharp

Figure 9. Changing the centre of reciprocation in Figure 8(d).

These examples illustrate an aspect of the reciprocation process applied to


poiyhedra that is usually ignored. Reciprocation is a mapping between points, and
infinite lines and planes, but poiyhedra consist of vertices, and (usually) finite line-
segments and faces. Normally the natural identification of an edge with the finite
part of the line between two vertices goes unremarked., and it occurs automatically
when the reciprocation algorithm is executed, but cases such as figure 8(a) make
it obvious that a choice is being made. It is also clear that there is a discontinuity at
this stage, when the set of four semi-infinite edges flips from pointing upwards to
pointing downwards, and the edges that are the finite parts of these lines in figure 8(b)
correspond to the infinite parts that are ignored in figures 6(b) and 7(b).
In figure 7(b), we have polygons which are self-intersecting. This is purely a result
of where the quadric (in this case a sphere) is positioned when reciprocating. By
moving the polyhedron in figure 7(a) up until the centre of the sphere lies outside the
polyhedron, the self-duality is more evident (figure 9). There are now no self-
intersecting polygons.
The polyhedron in figure 7(a) does not have the complete symmetry of the cube,
only a fourfold symmetry about the vertical axis. We could make a polyhedron in
which there are dimples on each face of the cube. If they all reach the centre o\' the
cube then the squares in the reciprocal go to infinity if the centre of our reciprocating
sphere is at the centre ofthe cube. However, if we only make the dimples slight, so
that we get a symmetrical version of figure 7(b). and create the reciprocal, the
polyhedron does exist in the literature. It is figure 104 in Bruckner [18] shown in our
figure 10(a).
We can also consider what happens to reciprocals of poiyhedra formed by adding
square pyramids to each face of a cube when the apices of the pyramids are moved
symmetrically. The measurements for the height ofthe apices are given relative to a
starting cube of side 2 units.
The reciprocal begins as an octahedron with all six vertices truncated, forming
square faces (figure 11 (b)) where the apices in figure 11 (a) are 1,15 units above the
faces. As the apices of the square pyramids are raised, the squares of the truncated
octahedron move down, at one stage forming a regular truncated octahedron
(figure 1 l(d)) when the apices move to 1.5 units as in figure I l(c) which corresponds
to Catalan's tetrakis hexahedron. Continuing to raise the apices to 2 units as in
figure 1 l(e), the triangular faces ofthe pyramids become coplanar in pairs, forming a
rhombic dodecahedron with eaeh face divided into two triangles. In the reciprocal
the vertices of the squares touch to form a polyhedron that looks like a cubocta-
hedron (figure 11(0). but the triangular faces are actually hexagons, having three of
the sides of zero length, and what appear as single vertices in the figure are actually
Duality of polyhedra 633

(a) (b)
Figure 10. Bruckner polyhedra.

pairs of vertices. Raising the apices further is like 'folding' the rhombs of the
rhombic dodecahedron" about the cube edges so that the edge is no longer convex. In
the two examples shown, figure 1 l{g) has the apex at 2.5 units and figure 1 l(h) at 3.
Note that polyhedra like these also exist in Bruckner [18]. The one in figure 10(b)
is figure 1 in table XI of Bruckner.

8. Convexity and figure 4(a) and 4(b)

In the earlier discussion we noted Griinbaum and Shephurd's consideration of the


use of convexity as a means of obtaining duals such as figure 4(a) and 4(b), and iheir
conclusion that 'convexity properties cannot be preserved in duality", which they
illustrate by considering the polyhedron In (our) figure 7(a). They say that it 'has four
non-convex edges meeting at a vertex, so a dual ouiihi (their italics) to have four non-
convex edges bounding a quadrangular face", and observe that no such polyhedron
exists. Indeed there are no known rules to predict the convexity properties of parts of
the dual from parts ofthe original polyhedron, and even the general observation that
convex polyhedra have convex duals breaks down if there are no restrictions on the
centre of the reciprocating sphere. For e.xample, figures 9 and 7(b) are both duals of
figure 7(a) becau.se they have been created as reciprocals, and they differ because the
position of the reciprocating sphere is different. Similarly, the set of 'octahedra* in
figure 12 shows the efTect of reciprocating a cube by successively moving the centre
oi^ the reciprocating sphere up from the centre of the cube.
As the centre of the reciprocating sphere moves up, the top half of the
octahedron becomes more elongated. The top vertex then jumps over infinity and
as in figure 8 we use the finite line segments. The faces that originally formed the
eonvex base now become a concave top. Thus different duals ofthe same polyhedron
may have different convexity properties.
We are now in a position to look at the inverse dodecahedron (figure 4(a)) for
which Grunbaum and Shephard gave the misleading dual (figure 4(b)) and to
determine the reciprocal of figure 4(b). In the following, the duals are created as
634 P. Gaitiuna.s and J. Sharp

(c) (cl)

(e) (0

(g)

Figure II. Cubes with square pyramids added to their forces with their reciprocals.

reciprocals with respect to a sphere whose centre is at the centre of symmetry ofthe
two polyhedra.
Some properties of the dual of the inverse dodecahedron (figure 4(a)) can be
deduced by comparing it with some well-known related polyhedra that have
accepted duals. It can be considered as a creat stellated dodecahedron
Duality of polyhedra 635

(a)

(c) (d)
Figure 12. Reciprocal of a cube wilh ditlerenl positions of reciprocatinj; sphere.

(tigure I3(a)) with twelve ofthe spikes removed. Twelve oTthe crossing points ofthe
star pentagons, which are usually not considered to be vertices and so have no
corresponding faces in the dual, the great icosahedron (figure 13(b)). become true
vertices. The dual will have eight planes of faces in common with the great
icosahedron. corresponding to the eight spikes that are not removed, and since
these vertices form the vertices of the cube that is the convex hull of the inverted
dodecahedron, the dual faces surround an octahedral core.
Viewing the great stellated dodecahedron as an icosahedron with triangular
pyramids (the spikes) stuck on each face makes it clear that the other twelve vertices
ofthe inverted dodecahedron are common with an icosahedron. and so they are dual
to faces lying in the planes of a dodecahedron. Since the inverted dodecahedron has
only one type of face, its dual has one type of vertex, which we have already seen to
be those ofthe great icosahedron. and the vertices ofthe great icosahedron are those
of its convex hull, an icosahedron. So these twelve faces in the dual are those of a
polyhedron that is a facetted icosahedron and a stellated dodecahedron. They are in
fact part of a great dodecahedron (figure 14).
So. we can say that the complete dual has twelve (icosahedral) vertices and
twenty laces, eight are equilateral triangles from the great icosahedron. twelve are
acute isosceles triangles derived from the pentagonal faces of a great dodecahedron.
636 P. Gaitiunas and J. Sharp

(a) (b)
Figure 13. Great stellated dodecahedron and great icosahedron.

Figure 14. Great dodecahedron.

Although it is still difficult to visualise it. the reciprocation algorithm will generate
pictures of the dual, and the foregoing discussion will help in interpreting them.
At first glance the result is not straightforward, and there appear to be faces and
edges missing. Figure 15 is a stereoscopic pair to aid examination. There is a similar
'wedge' shape at the back, completing the symmetry, which is hidden.
There are points that are not vertices, for example where three faces intersect in
the dimples either side of the "wedge" shapes. The true twelve vertices are at the tips
of the wedges.
There appear to be only four faces meeting at a vertex, and because of the
intersections only parts of some faces can be seen. Counting the visible parts gives a
total of 36, so some parts must belong to the same face.
When a dual is calculated by reciprocation, there is an added bonus of numerical
data, for example information about the two types of triangle. The set of twelve are
isosceles triangles with two angles being 72' and sides in the ratio of the golden
Duality of polyhedra 637

(a) (b)
Figure 15. Reciprocal of inverted dodecahedron (figure 5 left).

(a) (b)
Figure 16. Reciprocal or inverted dodecahedron wilh complete edges.

section. The shortest sides of these triangles are the sharp edges of the wedges. The
other triangles are equilateral triangles with three parts of eaeh of them visible. The
edges of the twelve isoseeles triangles sit in the faces of other similar triangles. They
can be drawn in as in figure 16.
This now shows the five faces in each vertex. It also shows that the edge going to
the vertex in the dimple formed by intersections is also a line formed by the
intersection of two equilateral triangles. If this were not so, too many edges would
meet at a vertex. Also, since edges of the polyhedron are common to two faces, the
equilateral triangles must also lie in the faces ofthe isosceles triangles. The edges of
the equilateral triangles are the same as the long edges ofthe isosceles triangles. This
all suggests a complex internal structure to this dual with the "virtual" octahedron
buried in the centre.
Figure 17 shows the result of removing all the isosceles triangles to leave the eight
equilateral triangles as a stellation of the octahedron. This does not fall within the
usual definition of stellations, since the sides ofthe triangles are not common edges;
the only closed stellation of the octahedron is the Stella octangula.
The octahedron can only be seen by slicing to get to the interior as shown in
figure 18.
638 P. Guilitinas mul./. Sharp

(a) (b)
Figure 17. Reciprocal of inverlcd dodecahedron, inlcrnal structure.

Figure 18. Reciprocal of inverlcd dodecahedron, section to show octahedron.

Thus Ihc dualization considered by Grunbaum and Shephard was wide of the
mark, and it is nol even necessary to construct the reciprocal to deduce that
the polyhedron is inconsistent with the well-known duality relations between the
Kepler-Poinsot poiyhedra.
We have only touched on the properties of the inverse dodecahedron and its dual
(which we believe to be a new polyhedron) here and they are obviously polyhedra
which deserve more study.
The dual of the Jessen icosahedron in figure 4(b) is much simpler. Earlier, we said
that the drawing is dift'erent from the one shown by Grunbaum and Shephard.
We have chosen to construct it from two sets of triangles as follows. The external
triangles are equilateral. Those where there is a concavity are isosceles triangles with
two angles of 36 . that is a triangle with sides in the golden section. Figure 4(b) is a
regular icosahedron with wedges cut out of it perpendicular to parallel edges, and so
we might expect to see a regular dodecahedron with some additions at edges. This is
indeed the case as figure 19(a) shows.
Duality of polyhcdra 639

(a) (b)
Figure 19. Reciprocal of Jessen icosahedron {figure 5 right).

Figure IS)(b) shows how ihe pentagons of ihc dodecahedron have been extended
and interseet to form an equilateral {though not equiangular) pentagon. We also
believe this lo he a new polyhedron.

9. Conclusions

The term "duality", meaning a mapping between points and planes (or vertices and
faces), generally occurs in geometries where these are fundamental objects. In
projective geometry they occur symmetrically in the axioms, so that duality is a
natural consequence. Some of the problems that arise when considering dual
poiyhedra occur when there is some further structure. For example Grunbaum
and Shephard [5] define poiyhedra as manifolds and go on to show that in some
circumstances the dual ofa polyhedron does not fit this definition. Another example
is provided by figure 2. Different faces being coplatiar seems unremarkable, since
they occupy different positions in spaee. The dual situation, difierent vertices being
coincident, is certainly contrary to common usage, and is very counter-Intuitive.
Further problems oecur beeause the concept of 'duality" carries a different
meaning in different eontexts. It is important that there should be no ambiguity in
its application if misunderstandings are to be avoided, and care is needed to be sure
of obtaining logically consistent results.
With modern computer graphics and CAD packages it is easy to implement an
algorithm for reciprocation, which not only provides a useful way lo eheck ideas
about poiyhedra and their duals, but also offers much potential for further study and
the discovery of new ones. We hope to provide ftirther examples soon.

Appendix. Using VRML to view poiyhedra

Not everyone can program or use a CAD program, but in learning mathematics it is
reasonable to assume that a spreadsheet could be used to perform calculations and a
640 P. Gaitiuna.s and J. Sharp

B
Figure 20. Square pyrtimidc used in VRML example.

simple VRML file can be used to view the result with free VRML plugins for Internet
browsers being available on the internet for different platforms. VRML (Virtual
Reahty Markup Language) has an easy to use file format which can be written in any
text editor. Providing simple templates into which calculated data can be edited
overcomes any need for learning to program the language. The result can be viewed
in minutes and the polyhedron manipulated in real time. The following example
shows how to analyse a simple polyhedron and use it in a VRML file which can serve
as a template.
As well as being a good training in analysis of data, the calculation in a spreadsheet
provides a useful exercise of organisation ofthe spreadsheet (arranging data to make
it easy to copy formulae) and the use of features like determinants. This spreadsheet
exercise is left to the reader, since the algorithm is very easy to implement.
The example uses a square pyramid with base as shown in figure 20 with
Cartesian coordinates:

A (1. 1.0)
B(l, -1.0)
C ( - l , -1,0)
D ( - l . 1. 0)
E (0. 0. I)

In order to keep the calculation simple, the reciprocating sphere is left at the origin
and an offset added to move the polyhedron. With the coordinates above there is no
offset, so the plane ABCD goes though the centre ofthe sphere and its pole goes to
infinity. The following table shows the calculated eoordinates of the poles of each
plane.

olfset (O.O.I) offset (0.-2.-2)


(6.0, - I ) (0.0.0.5)
(-0.5.0. -0.5) (1,0. 1)
(0, -0.5. -0.5) (0. 0.3333. 0.3333)
(0.5.0, -0.5) (-1.0. 1)
(0,0,5, -0.5) (0. I. - I )

With an offset in the z direction, as in the first set of coordinates, the self duality
is evident as a similar pyramid is produced. The second set of coordinates, with an
offset not along the r axis the base is no longer square. The result is shown in
figure 21.
Duality of polyhedra 641

higLirc 21, Screen dunip Ironi V R M L viewer.

The template VRML tile is as follows:


#VRML V2.0 utf8
# offset 0, -2, -2
Navigationinfo {
type 'EXAMINE'
headlight TRUE
}
Viewpoint {
description 'outside'
position 0 0 2
)
Background {
skyColor [1.0 1.0 1.0]
}
DirectionalLight [
on TRUE
intensity 0 . 5
ambientlntensity 0. 5
color 1.0 1.0 1.0
direction -10 .0 0.0 1.0

Shape {
appearance Appearance {
material Material {
diffuseColor 0.5 0.5 .5 }
1
geometry IndexedFaceSet {
solid FALSE
coord Coordinate {
point [
642 P. Gailiuna.s and J. Sharp

0, 0 , 0 . !D
1, 0, 1
0, 0 .3333, 0.3333
-1 0, 1
0, 1, -1
]

:d] :n dex
0, 1, 4, -1 ,
0, 1, 2, -1 ,
0, 2, 3, -1 ,
0, 3, 4, -1 ,
1, 2, 3, 4, -1 ]
1

It is not neeessary to understand the template in detail in order to use it. The
important part for the purposes of this paper is to see that the coordinates are
entered as a table followed by a table which indexes eaeh plane. So the quadrilateral
is the ordered set of points I, 2, 3, 4 which is then terminated with a —1, as are each
ofthe triangular planes.
There is not space to describe more sophisticated use of VRML with Java scripts
to allow dynamic movement of the polyhedron relative to the sphere, either
animated or under user control.

References

[I] Holden, A.. 1971. Shape.";. Space and Symmetry (New York & London: Columbia
University Press, repiiblishcd by Dover).
[2] Cromwell, P.. 1997. Poiyhedra (Cambridge: Cambridge University Press).
[3] de Villiers. M.. 1998, Dual generalisations of Van Aubel's theorem, Mathematical
Gazette. 495,405^12.
[4] Bakos, T,. 1959. Octahedra inscribed in a cube. Mathematical Gazette, 43, 17 20.
[5] Grunbaum. B. and Shephard, G. C , 1988. Duality of Poiyhedra. In: M. Senechal and G.
Fleck (Eds) Shaping .Space (Boston, MA: Birkhauser).
[6] Coxeter, H.S.M., 1973, Regular Polytope.s (Dover).
[7] Griinbanm, B.. 1993. Poiyhedra with hollow faces. In: T. Bisztriezky, P. McMullen. R.
Schneider and A, Ivic'Wciss (Eds) Proc. NATO-ASI Conference on Polytopes: Abstraet,
Convex and Computational. Toronto, 1993 (Dordrecht; Kluwer Academic), pp. 43-70,
[8] Mobius. A.F., 1831. Analyti.sch-geometrische Entwickclungcn. Vol. 2, 1831.
f9] Pedoe, D., 1975. Notes on the History of Geometrical Ideas II: the principle of Duality,
Mathematics Magazine, Nov-Dec, 274 277.
[10] Coxeter. H.S.M., 1955, The Real Projective Plane (Cambridge: Cambridge University
Press), pp. 74-75.
[11] Wenninger. M.J., 1983. Dual Motlcis (Cambridge: Cambridge University Press).
[12] Ounsted, J.. 1978, An unfamiliar dodecahedron. Mathematics Teaching. 83. 46 47.
[13] .lessen, B., 1967, Orthogonal icosahedra, Nordi.sk Matemati.sk Tidskrift, 15, 90 96.
[14] Cundy. H.M. and Rollett, A.P.. 1961, Mathematical Models (Oxford: Oxford University
Press).
[15] Maxwell. E.A.. 1963, The Methods of Plane Projective Geometry based on the u.se of
General Homogeneous Equatkm.s (Cambridge: Cambridge University Press).
[16] Penna. M, and Patterson, R., 1986, Projeclive Geometry and its Applications to Computer
Graphics (Englewood ClitTs. NJ: Prentice Hall).
[17] Smith, C , 1901, .An Elementary Treatise on Solid Geometry (London: Macmillan).
[18] Bruckner. M., 1900, Vieleckv und Vietflache (Leipzig: Tuebner).

You might also like