New York Journal of Mathematics and Brandy Smith Mayeld

advertisement
New York Journal of Mathematics
New York J. Math. 6 (2000) 237{283.
Divisible Tilings in the Hyperbolic Plane
S. Allen Broughton, Dawn M. Haney, Lori T. McKeough,
and Brandy Smith Mayeld
Abstract. We consider triangle-quadrilateral pairs in the hyperbolic plane
which \kaleidoscopically" tile the plane simultaneously. In this case the tiling
by quadrilaterals is called a divisible tiling. All possible such divisible tilings
are classied. There are a nite number of 1, 2, and 3 parameter families as
well as a nite number of exceptional cases.
Contents
1. Introduction
2. Tilings and Tiling Groups
2.1. Tiling Groups
2.2. Divisible Tilings
3. Overview of Quadrilateral Search
3.1. Free Vertices
3.2. Constrained Vertices
3.3. The Two Search Methods
4. Direct Construction Method (K 12)
4.1. Polygon Construction and Elimination|No Interior Hubs
4.2. Computer Algorithm and Extension to Interior Hubs.
5. Boundary Construction Method (K > 12)
5.1. Geometric Quadrilateral Test
5.2. Boundary Word Test
5.3. Example: Failure of (2 3 7) to Tile (7 7 7 7)
5.4. Example: Successful Tiling of (5 5 5 5) by (2 4 5)
6. Catalogue of Divisible Tilings
Appendix A. Triangles with area 4
References
238
240
241
243
246
246
248
250
251
253
256
257
259
259
262
262
265
282
283
Received October 11, 1999.
Mathematics Subject Classication. 05B45, 29H10, 20H15, 51F15, 52C20, 51M10.
Key words and phrases. tiling, Fuchsian groups, reection groups, crystallographic groups,
hyperbolic plane.
The last three authors were supported by NSF grant DMS-9619714.
237
ISSN 1076-9803/00
238
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
1. Introduction
Let be a polygon in one of the three two-dimensional geometries: the sphere
the Euclidean plane E or the hyperbolic plane H . Suppose also that each
interior angle of the polygon at vertex Pi has measure si where si is an integer.
The polygon generates a tiling of the plane by repeated reections in the sides
of the polygon. Examples are the icosahedral tiling of the sphere by 36 -60 90 triangles in Figure 1.1, and the partially shown tilings of the Euclidean plane
by 45 -45 -90 triangles in Figure 1.2 and the hyperbolic plane by 36 -36 -90
triangles in Figure 1.3. These tilings are called geodesic, kaleidoscopic tilings since
the tiling may be generated by reections in a single tile. We explain the term
geodesic soon. A denizen of the two dimensional geometry could view the tiling by
constructing a polygon of mirrors meeting at the appropriate angles { assuming that
light travels in straight lines in the geometry! In the Euclidean case the mirrored
polygons can actually be physically constructed and the tiling viewed for the three
Euclidean kaleidoscopic triangles (30 -60 -90 45 -45 -90 60 -60 -60 ) and the
one 1-parameter family of kaleidoscopic rectangles.
The plane may be kaleidoscopically tiled in several dierent ways as Figure 1.2
shows. One way is by triangles. A second way is by squares consisting of four
triangles meeting at the center of the square. Yet a third way is the tiling by
squares formed from two triangles meeting along a hypotenuse. The tilings by
squares are both rened or subdivided by the tiling of triangles, i.e., each square
is a union of either four non-overlapping triangles or two non-overlapping triangles
in the two cases. We say that the tiling by squares is divisible or that the tiling by
squares is subdivided by the tiling by triangles. If are such a triangle and
square respectively we call ( ) a divisible tiling pair.
There are no divisible quadrilateral tilings of the sphere. There are innitely
many dierent divisible quadrilateral tilings of the Euclidean plane but they are all
found in the 45 -45 -90 tiling in Figure 1.2. In this paper we turn our attention to
the much richer case of divisible quadrilateral tilings of the hyperbolic plane. The
reader is invited to nd the tiling by quadrilaterals hidden in Figure 1.3 without
\cheating" by looking at the \answers" in the tables in Section 6. Each quadrilateral
has 12 triangles.
The main result of the paper, Theorem 6.2, is a complete catalogue of all divisible quadrilateral tilings of the plane which may be subdivided by a triangle tiling.
For completeness we also include the less complex cases of triangle tilings which
subdivide triangle tilings (Theorem 6.1) and quadrilateral tilings which subdivide
quadrilateral tilings (Theorem 6.3). The classication of tilings of quadrilaterals by
triangles is broken up into two categories, constrained and free : The main dierence between the two is that the free tilings occur in innite families with simple
parametrizations by integers, but there are only a nite number of constrained divisible tilings. The complete lists of the four types of divisible tiling pairs described
above are given in Tables 6.1{6.4 in Section 6. More importantly, pictures of all
the various tiling pairs are given in Tables 6.5{6.8 of the same section.
To put our results in a broader context we sketch an application of the results
to a problem in the classication of Fuchsian groups, which in turn is relevant to
the singularity structure of moduli spaces of Riemann surfaces for certain genera.
Let 1 and 2 be two Fuchsian groups such that 1 has signature (0 l m n) and
S2,
Divisible Tilings in the Hyperbolic Plane
239
2 has signature (0 s t u v) and let H be the disc model of the hyperbolic plane.
Thus the projections H ! H =1 and H ! H =2 are branched covers of the sphere
branched over 3 points of orders l m n and 4 points of orders s t u v respectively.
One may ask under what conditions does 2 1 : The tiling polygons and
generate groups 1 and 2 such that the conformal subgroups 1 1 and
2 2 of index 2, are of the type specied in the classication problem. The two
groups constructed are real, i.e., the tiling polygons may be translated such that
1 and 2 are real, i.e., both groups are invariant under conjugation. In turn this
implies that the points 1 and 2 correspond to singular points on the moduli space
of Riemann surfaces with real dening equations for certain genera. The details
of this will be discussed in a subsequent paper 4] which examines divisible tilings
on surfaces. The problem of determining pairs 2 1 with an equal number
of branch points was solved in 10]. The corresponding tiling problem is tilings
of triangles by triangles and quadrilaterals by quadrilaterals. The solution of this
problem follows easily from 10], though the classication has not been published,
to our knowledge. We include both of these results for completeness.
The remainder of this paper is structured as follows. In Section 2 we introduce
the necessary background on planar tilings, divisible tilings and the tiling groups.
In Sections 3, 4 and 5 we introduce and discuss the two computer algorithms for
determining the divisible tilings and illustrate them with some sample calculations.
Finally all results, including gures are listed in tables in Section 6. Throughout
the discussion, the reader is encouraged to look at these gures to gain a clearer
idea of the denitions and the discussion.
We will use the disc model for the hyperbolic plane H in which the points are
in the interior of the unit disc, the lines are the unit disc portions of circles and
lines perpendicular to the boundary of the unit disc, and reections are inversions
in the circles dening the lines. We will denote the hyperbolic distance between
two points z1 and z2 by (z1 z2): All properties we use about hyperbolic geometry,
in particular the area formula for polygons, may be found in the text by Beardon
1].
Acknowledgments. The initial part of this research work was conducted during the
NSF-REU program at Rose-Hulman Institute of Technology in the summer of 1997
(Haney and McKeough 8]) and continued in 1998 (Smith 12]) under the direction of Allen Broughton. The 1997 project worked out the classication under a
restrictive hypothesis called the corner condition and yielded 13 of the constrained
cases. A subdivided quadrilateral satises the corner condition if each corner of
the quadrilateral contains a single triangle. For example cases C1 and C2 in Table 6.7 satisfy this property though cases C5 and C9 do not. Many of the symmetry
properties of the tilings were examined, and therefore the symmetry groups of the
tiling pairs are included in the tables in Section 6. The 1997 group also undertook some of the preliminary work in determining divisible tilings on surfaces, the
results of which will appear in a forthcoming publication 4]. The 1998 project
took a dierent approach that yielded many of the free cases with a small number
of triangles. The present work combines and extends both approaches to yield a
complete classication of planar hyperbolic divisible tilings by quadrilaterals.
We thank the numerous participants of the 1997 and 1998 programs for the useful
conversations and encouragement. All numerical calculations were performed using
240
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Maple 13]. The gures were produced with Maple and Matlab 14]. All Maple and
Matlab scripts used, as well as images, are available at the tilings website 15].
Figure
Figure
1.1. Icosahedral (2 3 5) tiling of S2
1.2. (2 4 4) tiling of E
2. Tilings and Tiling Groups
Figure
1.3. (3 3 4) tiling of H
A tiling of the spherical, Euclidean or hyperbolic plane is a collection T of
polygons, called tiles, that completely cover the plane without overlaps or gaps.
The sides or edges of the tiles are called the edges of the tiling and the vertices of
the tiles are called the vertices of the tiling. Let E and V denote the collection of
edges and vertices of the tiling.
Divisible Tilings in the Hyperbolic Plane
241
Denition 2.1. A tiling T of a plane is said to be a kaleidoscopic tiling if the
following condition is met:
1. For each edge e 2 E of the tiling the reection re in the edge e is an isometry
of the plane that maps tiles to tiles. In particular it interchanges the two tiles
whose common edge is e.
A tiling T is called a geodesic, kaleidoscopic tiling if in addition we have
the following condition.
2. The xed line or mirror fx 2 S : re (x) = xg of each reection re is the union
of edges of the tiling. Such a line is called a line of the tiling.
The tiling of the Euclidean plane by hexagons or the dodecahedral tiling of the
sphere by pentagons are examples of kaleidoscopic tilings which are not geodesic.
For the remainder of the paper, unless specied otherwise, all tilings we discuss will
satisfy Denition 2.1. The following proposition allows us to easily identify which
polygons give rise to the desired tilings. It is easily proven using the Poincare
Polygon Theorem 1, p. 249].
Proposition 2.2. Let = P1P2 Pn be a n-gon. Then generates a kaleidoscopic tiling of the plane by repeated reection in its sides only if the interior
angles at the vertices of the polygon have measure 2ni where ni is an integer. If in
addition each ni is even, say ni = 2mi so 2ni = mi then generates a geodesic,
kaleidoscopic tiling.
We shall call a polygon kaleidoscopic if it generates a kaleidoscopic tiling. Throughout the paper we shall only consider kaleidoscopic tilings that generate geodesic
tilings, i.e., the angles have the form mi :
Notation 2.3. A polygon P1P2 Pn such that the interior angle at Pi has radian
measure mi is called an (m1 m2 : : : mn )-polygon. Note that 2mi tiles meet at the
vertex Pi : Hence we easily identify the tiles of the icosahedral tiling in Figure 1.1
as (2 3 5)-triangles.
2.1. Tiling Groups. The reections in the edges of a tiling generate a group of
isometries of the tiling, called the tiling group. We describe this group in some
detail now for the case of a triangle. The generalization to the group of a general
polygon easily follows from the triangle discussion. It is easy to show that every
tile in the plane is the image, by some element of the tiling group, of a single tile,
called the master tile, pictured in Figure 2.1. The sides of the master tile, 0 are
labeled p, q, and r, and we denote the vertices opposite these sides by P Q and R
respectively. We also denote by p, q, and r the reection in corresponding side. We
assume that 0 is an (l m n)-triangle so that the angles at R P and Q have size
l radians, m radians, and n radians, respectively, where l m and n are integers
2 (see Figure 2.1). At each of the vertices of the triangle, the product of the
two reections in the sides of the triangle meeting at the vertex is a rotation xing
the vertex. The angle of rotation is twice the angle at this vertex. For example
the product p q , a reection rst through q then through p, is a counter-clockwise
rotation through 2l radians. We will refer to this rotation as a = pq and use it to
label the vertex in Figure 2.1. Rotations around each of the other corners can be
dened in the same way, so that b = qr and c = rp are counter-clockwise rotations
through 2m radians and 2n radians, respectively.
242
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Q
c
π /n
p
r
Δ0
π /l
R
π /m
a
b
q
P
2.1. The master tile and generators of T and T From the geometry of the master tile, we can derive relations among these group
elements. It is clear that since p, q, and r are reections, the order of each of these
elements is 2:
(1)
p2 = q2 = r2 = 1:
From the observations about rotations above, it is also clear that the orders are
given by
(2)
o(a) = l o(b) = m o(c) = n
and
(3)
abc = pqqrrp = 1:
The reections generate a group T = hp q ri and the rotations generate a subgroup T = ha b ci which includes only the orientation-preserving isometries in T :
The subgroup T is of index 2 in T and hence is also normal in T . Here are some
well-known basic facts about T and T:
Proposition 2.4. Let T and T be derived from a tiling T as above. Then the
following hold.
1. The groups T and T have the following presentations
(4)
T = p q r : p2 = q2 = r2 = (pq)l = (qr)m = (rp)n = 1
Figure
Divisible Tilings in the Hyperbolic Plane
and
243
(5)
T = a b c : al = bm = cn = abc = 1 :
2. The full tiling group T acts simply transitively on the tiles of T .
Proof. These facts are well-known, though we give a brief proof sketch for completeness. According to the Poincare Polygon Theorem 1, p. 249], T is a Fuchsian
group, i.e., a discrete group of isometries, and a tile is a fundamental region for T .
Any group element xing a tile must therefore x all the interior points of the
tile, and equals the identity. Thus, simple transitivity is proven. Now for the rst
part. First construct the dual graph by joining incentres of adjacent triangles by a
segment meeting the common edge of the triangles at a right angle. It is easy to
see that every word in T corresponds to an edge path in the dual graph starting
at the incentre of the master tile. Words which equal the identity correspond to
closed paths. Now suppose we have a word corresponding to a closed path. We
must show that we can reduce it to the identity by the relations above. Since H is
simply connected a closed edge path in the dual graph is homotopic to the identity.
Using the homotopy a closed edgepath can be deformed to the identity in a series of
moves of the following type: a) introduce or eliminate a path that crosses a tile edge
and then goes back, b) replace a path which makes a partial clockwise turn around
a vertex with the complimentary counter-clockwise turn around the same vertex.
These two types of homotopies correspond to the relations p2 = q2 = r2 = 1 and
(pq)l = (qr)m = (rp)n = 1 respectively. This proves that T has no other relations.
A similar proof works for T:
Because of the simple transitivity, given a master tile 0 there is a unique
isometry g 2 T such that = g0 : This then allows us to identify a vertex x as
being a vertex of type P , Q, or R depending on which vertex it is equivalent to
in the master tile. The same applies to edges. Of course, in scalene triangles the
vertex type and edge type are easily identied by angle measure and side length.
However in the case of an isosceles triangle it is necessary to use the tiling group
action to dene types. Similar remarks apply to the quadrilateral with respect to
the corresponding tiling groups of an (s t u v)-quadrilateral, which we call Q and
Q:
2 = x2 = y 2 = z 2 =
w
x
y
z
:
w
(6)
Q = (wx)s = (xy)t = (yz )u = (zw)v = 1
and
(7)
Q = d e f g : ds = et = f u = gv = defg = 1 :
2.2. Divisible Tilings.
Denition 2.5. A kaleidoscopic tiling is said to be divisible if it can be kaleidoscopically divided into a ner tiling. Thus we have two tiles both of which
generate a kaleidoscopic tiling of the plane. Each tile of the -tiling is a union of
polygons from the -tiling. We say that the -tiling subdivides the -tiling. We
call ( ) a divisible tiling pair.
We have seen an example of a divisible hyperbolic quadrilateral tiling in Figure 1.3 and all examples of divisible tilings of the hyperbolic plane are given in the
244
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
gures in Section 6. In each of the gures we only give one quadrilateral, though it
may be easily extended to the plane through reections in the sides of the quadrilateral. In fact the following is easily proven by using reections in the sides of
.
Lemma 2.6. Suppose that is a kaleidoscopic tile and that is a larger
polygon such that is a union of the triangles of the tiling dened by and
that the angles of have measure mi for various integers mi : Then ( ) is a
kaleidoscopic tiling pair.
For the remainder of the paper we shall concentrate on the case where both
and are triangles or quadrilaterals and the tilings are geodesic. In almost
every case will be a triangle and will be a quadrilateral.
Remark 2.7. There are examples of kaleidoscopic tiling pairs where the tilings
may not be geodesic. For example the Euclidean \honeycomb" tiling by hexagons
may be subdivided into a tiling by equilateral triangles.
Remark 2.8. For an (l m n)-triangle to tile an (s t u v)-quadrilateral each
of s t u and v must be a divisor of one of l m or n: For, some multiple of an angle
of must t into each corner of the quadrilateral .
The number of triangles. Let the (l m n)-triangle and the (s t u v)-quadrilateral
with form a divisible tiling pair. The areas of the triangle, At , and
quadrilateral, Aq are given (see 1, 150] and 1, 153]) by:
1
1
1
A = 1 ; ; ; = (l m n)
t
l
m n
Aq = 2 ; 1s ; 1t ; u1 ; v1 = (s t u v)
for some positive rationals (l m n) and (s t u v). Note that we must therefore
have
1 + 1 + 1 < 1 and 1 + 1 + 1 + 1 < 2:
l
m n
s t u v
A table of all values of possible (l m n) with (l m n) 14 is given in Appendix A.
These are the only values we shall need for our study. Now is a union of triangles
congruent to let K denote the number of triangles. We have Aq = KAt or,
1
1
1
1
1
1
1
(8)
2; ; ; ; =K 1; ; ; s
t
u v
l
m
n
or alternatively:
2 ; 1s ; 1t ; u1 ; v1 (s t u v)
= (l m n) :
1 ; 1l ; m1 ; n1
It turns out that K has an upper bound of 60 for all triangles in the hyperbolic
plane.
(9)
K=
Divisible Tilings in the Hyperbolic Plane
245
Proposition 2.9. Suppose the (l m n)-triangle tiles the (s t u v)-quadrilateral
: Then,
K = ((sltmunv)) =
2 ; 1s ; 1t ; u1 ; v1
1 ; 1l ; m1 ; n1
60:
Proof. Fix l m and n: To maximize K we need to maximize the area of a quadri-
lateral. Thus each integer s, t, u, and v, should be made as large as possible.
Since each of s, t, u, and v must divide one of l, m, or n, then the largest possible
quadrilateral is a (b b b b)-quadrilateral such that b is the largest integer selected
from l, m, and n. Now the smallest possible value of (l m n) on the hyperbolic
plane is 421 for a (2 3 7)-triangle, according to the table in Appendix A. Picking a
(7 7 7 7)-quadrilateral as suggested above, we get K = 107 = 421 = 60: For any other
triangle we have (l m n) 241 with (l m n) = 241 realized for a (2 3 8) triangle.
But now
2 ; 1s ; 1t ; u1 ; v1
< 1=224 = 48:
K=
1
1
1
1; l ; m ; n
Hubs. Let be an arbitrary quadrilateral and let v be any vertex of the (l m n)tiling contained in the interior or on the boundary of . Assume for the moment
that v is of type R. The collection of (l m n)-triangles with common vertex at v
and contained in will be called an R-hub. If the hub occurs at a corner in then the number of triangles divides l: If the hub occurs on an edge, but not at
a corner, then there are exactly l triangles in the hub and a hub occurring in the
interior of the quadrilateral has 2l triangles. To distinguish the three types of hubs
we call them corner hubs, edge hubs, and interior hubs, respectively. Since we are
mainly concerned about edge hubs we just call them hubs, if it will not cause any
confusion. An interior hub may be considered to be a union of two edge hubs, then
it is called a double hub. The P -hubs and a Q-hubs are dened in the same fashion.
When l m and n are all distinct we refer to the R-hubs, P -hubs and Q-hubs as
l-hubs m-hubs and n-hubs respectively. The various types of hubs are illustrated
in the gures in Table 3.1 in the next section.
The number of R-hubs (edge hubs and interior hubs, with interior hubs counted
as two hubs) in a subdivided quadrilateral is the number hR , given by:
(10)
hR = K ;l cR where cR is the number of triangles occurring in the corner R-hubs: We can clearly
see that hR must be an integer since the remaining edge hubs have l triangles, the
interior hubs have 2l triangles, and every triangle belongs to a unique R-hub. The
number cR is usually easily determined from the numbers s t u v: Similar formulas
hold for P -hubs and Q-hubs:
(11)
cP
K ; cQ
hP = K ;
m hQ = n :
246
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
3. Overview of Quadrilateral Search
We shall employ two dierent types of search algorithms depending on whether
K is large or small. For low values of K we directly construct tilings of quadrilaterals without worrying what the angles are. For large values of K we will develop
an algorithm that starts with a specic (l m n)-triangle and determines all
(s t u v)-quadrilaterals that the triangles can tile. To understand the rationale
of splitting the search into two approaches, we need to dene constrained and free
vertices. We will also obtain constraints and bounds on l m n and K that are
helpful in restricting the search. These bounds are important for otherwise the
computer implementation of the search is impractical.
Denition 3.1. Let generate a divisible tiling. Then the P -type vertices
of the -tiling are called -constrained if at least one belongs to either an edge hub
or an interior hub, i.e., hP > 0: Otherwise the P -type vertices are called -free.
Similar denitions hold for Q-type and R-type vertices.
Remark 3.2. As the same triangle tiling can rene two dierent quadrilateral
tilings freeness is relative to though we rarely mention :
3.1. Free Vertices. The freeness of vertices is exemplied in the gures in Table 3.1. There we show the rst four quadrilaterals of the innite family of (2 3 5d)
tilings of (d 5d d 5d) quadrilaterals where d 2: All the quadrilaterals have the
same divided structure. The free vertices may be \freely" dragged to the boundary
of the hyperbolic plane through an innite discrete set of positions. (In contrast,
a constrained vertex cannot since it has a xed measure.) The angles at the free
vertices become smaller and smaller as we approach the boundary until we reach
a (2 3 1) tiling of an (1 1 1 1)-quadrilateral. The free vertices are on the
boundary and have measure 0 = 1 : As we prove below, every divisible tiling with
free vertices gives rise to such an innite family of tilings.
Remark 3.3. Free vertices, or rather the lack of them play a special role in the
interpretation and calculation of the \monodromy group" T =coreT (Q) of the
tiling pair . This group and its relation to tiling groups of surfaces with
divisible tilings are discussed in greater detail in 8] and in the forthcoming paper
4].
The search algorithm for quadrilaterals with free vertices has to be handled differently from those with constrained vertices only, since there are innitely many
possibilities. Also, it turns out that the quadrilaterals with small numbers of triangles have free vertices and those with a large number of triangles have only
constrained vertices, and for the midrange of values of K we have both types.
We dene special K to be the number such that for any tiling pair with
K > special K the divisible tiling has only constrained vertices. For values below
or equal to special K we will use one algorithm and for those values above special
K we will use another algorithm. The following proposition species special K:
Proposition 3.4. Let be an arbitrary kaleidoscopic tiling pair consisting of
an (l m n)-triangle and an (s t u v)-quadrilateral. Let
2 ; 1s ; 1t ; u1 ; v1
K=
1 ; 1l ; m1 ; n1
Divisible Tilings in the Hyperbolic Plane
Table 3.1.
247
A family of divisible quadrilaterals with free vertices
(2 3 10) tiling of (2 10 2 10)
(2 3 15) tiling of (3 15 3 150)
(2 3 20) tiling of (4 20 4 20)
(2 3 25) tiling of (5 25 5 25)
be the number of triangles covering : Then if K > 12 then there are no -free
vertices in the -tiling, i.e., special K equals 12. Furthermore if there are two types
of free vertices then K 4 and if there are 3 types of free vertices then K = 2.
Proof. The examples in Table 3.1 show that special K is 12 or greater. Now
suppose that is a tiling pair such that Q is a free vertex. We shall rst
construct an innite family of divisible quadrilaterals all with the same structure.
Assume that our tile has sides p and q meeting at the origin at R in an angle l , so
that q is a portion of the positive x-axis and p is a diameter in the rst quadrant.
The third side r of our triangle is a portion of a (euclidean) circle in the rst
quadrant meeting q at P in an angle m meeting p at Q in an angle n and meeting
the boundary at right angles. (see the gures in Table 3.1). Now each triangle in the
quadrilateral is of the form gi where g1 : : : gK 2 T : For the quadrilaterals in the
examples the 12 dierent g's are 1 q qr qrp qrpr qrprq pq p pr prp prpr prprq:
Move the side r so that its point of intersection moves along the x-axis, but the
angle remains m : The moving line, denoted rN will intersect the diameter p in an
248
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
angle N that varies continuously from (1 ; 1l ; m1 ) (when the point of intersection
is at the origin) down to 0 (when the side rN meets p on the boundary). Let N
denote the triangle bounded by p q rN and let p q rN also denote the reections
in the sides of N . Now let gi (N ) be the isometry constructed from gi by the
replacements p ! p q ! q and r ! rN : Because Q is a free vertex then
K
i=1
gi (N )N
is a quadrilateral whose subdivided combinatorial structure is independent of N:
The angles at Q-vertices have measure Ni where 1 : : : k are the numbers of
triangles meeting at Q-vertices. In our example k = 4 1 = 3 = 1 and 2 =
4 = 5: Let be the least common multiple of the i 's. There are now an innite
number of integer values of N = d d = d0 d0 +1 : : : 2 Z such that Ni = i d is an
integer for all Q vertices, and such that the resulting (l m N )-triple corresponds
to a hyperbolic triangle. The resulting quadrilaterals, of type (sN tN uN vN ) for
appropriately selected integers, will then form our innite family of kaleidoscopic
quadrilaterals. If the other types of vertices are free then a multiparameter family
may be created.
Finally let us get a bound on K: We know that
2 ; s1N ; t1N ; u1N ; v1N
2
K=
1
1
1
1
1; l ; m ; N
1 ; l ; m1 ; N1
Now we may chose (l m N ) of the form (l m d) so that
2
2
K dlim
=
1
1 :
!1 1 ; 1 ; 1 ; 1
l m d 1 ; l ; m
The largest possible side for the right hand side is 12 when l = 2 m = 3: The
next largest size is K = 8 for l = 2 m = 4: If there are two or three types of free
vertices then a two or three parameter family limit calculation shows that K 4
and K 2 respectively.
3.2. Constrained Vertices. We have established a limit on K when there are
free vertices, we may establish a limit on l m n when there are only constrained
vertices. We shall also show that l m and n must be chosen from the table in
Appendix A.
Proposition 3.5. Let be an arbitrary kaleidoscopic tiling pair consisting of
an (l m n)-triangle and an (s t u v)-quadrilateral. Let
2 ; 1s ; 1t ; u1 ; v1
K=
1 ; 1l ; m1 ; n1
be the number of triangles covering : Then if all the vertices of the -tiling are
constrained then
(12)
l m n K:
Proof. If the R-vertices are constrained then there is at least one R-hub and hence
at least l triangles. But then K l: The proofs of the other inequalities are
identical.
Divisible Tilings in the Hyperbolic Plane
249
The above proposition implies that the universal bound of 60 is also a bound for
l m n when we have constrained vertices. However we can do much better than
this. Assume that l m n: Then, s t u v n and
(s t u v) = 2 ; 1s ; 1t ; u1 ; v1 2 ; n1 ; n1 ; n1 ; n1 = (n n n n):
It follows that
(13)
l m n K
(n n n n) :
(l m n)
Thus, for instance, for a (2 3 n) triangle we have
2; 4
2:
n 1 n1 = 12 nn ;
;6
6 ; n
Solving this inequality for integer solutions, we obtain 3 n 16: Since the triangle
is hyperbolic then we further restrict 7 n 16: Analogously, for (2 4 n) triangles
we get 5 n 10 and for (3 3 n) triangles we have 4 n 7:
Now let us obtain a lower bound for K when there are constrained vertices.
Proposition 3.6. Let be a kaleidoscopic tiling pair with only constrained
vertices. Then the number of triangles, K is at least six.
Proof. Suppose that has at least one interior hub. Then the proposition is
satised unless the interior hub has four triangles. If there are only 4 triangles then
the two other vertices are free. Suppose there are ve triangles. Then, adjoining
a single triangle to an interior hub of 4 will force us to have at least one edge hub
with three triangles. In turn this will force an equivalent vertex to be a corner hub
with exactly two angles of measure 3 which is not allowed. In fact the hub of
four must be completed to a quadrilateral tiled with six triangles as in case F12 in
Table 6.6.
Thus we may assume that there are no interior hubs and that there is an edge
hub for each dierent type of vertex, say a P -hub HP centered at VP , a Q-hub
HQ centered at VQ and an R-hub HR centered at VR . The number of triangles in at least one of the hubs satises jHP HQ HR j K: We shall estimate
jHP HQ HR j by inclusion-exclusion and arrive at a contradiction. We have:
jHP HQ HR j = jHP j + jHQ j + jHR j
; jHP \ HQ j ; jHP \ HR j ; jHQ \ HR j
+ jHP \ HQ \ HR j
=l+m+n
; jHP \ HQ j ; jHP \ HR j ; jHQ \ HR j
+ jHP \ HQ \ HR j :
Now jHP \ HQ j = 0 unless VP and VQ are vertices of the same triangle. Furthermore, jHP \ HQ j = 1 if the edge joining VP and VQ is part of an edge of and
jHP \ HQ j = 2 otherwise. Also jHP \ HQ \ HR j = 1 if VP VQ and VR are the
vertices of the same triangle and it is zero otherwise. If jHP \ HQ j = jHP \ HR j =
jHQ \ HR j = 2 then jHP \ HQ \ HR j = 1 and jHP HQ HR j = l + m + n ; 5:
250
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
On the other hand if jHP \ HQ \ HR j = 0 then one of jHP \ HQ j jHP \ HR j and
jHQ \ HR j is zero so jHP HQ HR j l + m + n ; 4: Thus in all cases we have
l + m + n ; 5 jHP HQ HR j K 5 or
l + m + n 10:
Now as we go through hyperbolic (l m n)-triples in Appendix A we see that there
is only one triple that satises this namely (3 3 4): The quadrilateral must contain
one hub of order 4, but it is impossible to add a single triangle to make it into
quadrilateral. We have eliminated all cases.
The inequality (13) combined with K 6 shows that we get an area restriction
< 13 : We now prove a stronger restriction 41 which allows us to select all our
-data from the table in Appendix A.
Proposition 3.7. If an (l m n)-triangle subdivides an (s t u v)-quadrilateral, with
only constrained vertices then (l m n) 14 .
Proof. Suppose that = (l m n) > 41 : Let = (s t u v) we know that 2
and so thus
K = < 12=4 = 8:
Since K must be an integer, and K 6 by our last proposition, then 6 K 7:
Also, by Proposition 3.5, l m n K so we may assume, without loss of generality,
l m n 7: We consider four cases.
Case 1. n = 7 K = 7. There exists exactly one 7-hub and no other triangles.
However it is not possible to select the other angles to form a quadrilateral.
Case 2. n = 6 K = 6, 7. There is exactly one 6-hub. At most one other triangle
can be added and thus either 2 or possibly 3 triangles meet at the vertices. Since
there is at least one hub of each type, then the only possible hyperbolic triangle is
(3 3 6): However this satises 41 :
Case 3. n = 5 K = 6, 7. Again there is exactly one 5-hub and one or two triangles
must be added. Thus, except at the 5-hub, 2, 3 or 4 triangles meet at each vertex.
The only hyperbolic possibilities are (2 4 5) (3 3 5) (3 4 5) and (4 4 5) triangles.
)
The rst satisfy two satisfy 14 and for the last two the upper bound (nnnn
(lmn)
for K is smaller than 6:
Case 4. n = 4 K = 6, 7. The only possible triangles are (3 3 4) (3 4 4) and
(4 4 4) triangles, all of which satisfy the inequality.
Since there are no hyperbolic triangles with n 3 the proof is complete.
Remark
3.8. After constructing the tables we may actually verify that (l m n)
1
6
:
3.3. The Two Search Methods. The existence of free vertices and families of
divisible tilings forces us to split our search into two dierent approaches: the direct
construction method (K 12) and the boundary construction method (K > 12).
We describe the two approaches very briey here and then devote one section each to
the full description, implementation, and computational examples of both methods
Divisible Tilings in the Hyperbolic Plane
251
Direct construction method (K 12). Assume for the moment that there are no
interior hubs. Then each edge of a triangle must either be a part of the side of
the quadrilateral or reach from one side of the quadrilateral to another. Thus, as
a \combinatorial" object the quadrilateral may be viewed as a circle with a set
of non-intersecting chords or a better yet as a polygon with K + 2 vertices and
with a collection of K ; 1 diameters (see Figure 4.1 below). The diameters of the
vertices can then be labeled as P Q R vertices via the tiling structure. In order to
transform the polygon into a quadrilateral triangle K ; 2 corners of the polygon
must be attened to straight angles, leaving four corners to form a quadrilateral.
This imposes restrictions on l m and n and allows us to compute s t u and v
or conclude that no tiling pair exists. The algorithm can be extended to the case
where there are interior hubs. The algorithm depends only on the combinatorial
structure of the polygon and not the actual angles so it can handle the case of free
vertices. Unfortunately, the computational complexity of the computer search rises
very quickly with K and is not useful for large numbers of triangles.
Boundary construction method (K > 12). If K > 12 then there is only a nite
number of possibilities for (l m n) and hence a nite number of possibilities for
(s t u v) according to Remark 2.8. For each compatible pair of an (l m n) and
(s t u v) we try to construct an (s t u v)-quadrilateral in the (l m n)-tiling by
constructing the possible boundaries of a quadrilateral. The tile edges must occur
in certain sequences in the tiling, thus the boundary can be constructed \combinatorially". By making a geometric check we can tell whether the hypothesized
boundary closes up, and hence forms a quadrilateral. Again there is only a nite
number of cases to check.
4. Direct Construction Method (K
12)
We are rst going to show that all divisible quadrilaterals may be constructed
from a collection of Euclidean polygons, subdivided into K triangles and their
combinatorial analogs. By using the dual graph we will show the existence of an
algorithm that allows us to inductively construct all such polygons by attaching
triangles, interior hubs and conglomerations of interior hubs to a polygon with a
fewer number of triangles. Each such polygon may then be tested to see if it yields
a quadrilateral.
An associated divided polygon is constructed as follows. See Figures 4.1 and 4.3
below for examples without interior hubs. In Figure 4.2 an associated polygon
with interior hubs has been drawn along with its dual graph discussed below. Let
P1 : : : Pn be the vertices of the triangular tiling of as we move clockwise around
: Construct a convex Euclidean n-gon whose vertices are labeled P1 : : : Pn : Add
in all diagonals Pi Pj that correspond to edges of the tiling contained in . Next
we add points into the interior of the polygon corresponding to vertices of the
tiling in the interior of : We denote these vertices by Q1 : : : Qs (see Figure 4.2
below). We further add in all the segments Qi Pj and Qi Qj corresponding to
edges of tiling interior to : A combinatorial representation of the associated divided polygon or combinatorial divided polygon may then dened as the quintuple
(fPi g fPi Pj g fQi g fQiPj g fQiQj g) = (V@ E@ Vi Ei@ Ei ) where each component
is taken over the appropriate index set. Note that the set E@ = fPi Pj g the set of all
edges connecting boundary points, contains all the segments Pi Pi+1 1 i n ; 1
252
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Figure 4.1.
Polygon without interior hubs
and Pn P1 : The combinatorial representation is what we use for computer representation and calculation, the Euclidean polygon realization serves for visualization.
Note that the same set of associated polygons also arises from a pentagon, hexagon
etc., tiled by a triangle. The two critical features we need for the associated polygon
are:
it is a convex n-gon that is subdivided into triangles, and
an even number of triangles meet at each interior vertex.
In order to prove that we may inductively construct all the associated polygons
we work with a modication of the dual graph of an associated polygon. This is
constructed as follows. Place a node in the interior of each of the triangles. Connect
the nodes of neighbouring triangles by an arc crossing the common boundary. The
constructed graph has cycles if and only if there are interior hubs. See Figure 4.2
for a combined picture of an associated polygon and its dual graph. The graph has
the following properties, though we don't use them all.
It is planar.
Each node is connected to at most three other nodes.
The arcs of the graph may be coloured according to which type of edge they
cross.
The minimal cycles have an even number of nodes.
The region enclosed by a minimal cycle contains no other point of the graph.
We construct the modied dual graph as follows. For each interior hub, ll in the
portion of the dual graph bounded by the corresponding cycle. We could achieve
this by blackening in the two quadrilaterals in the dual graph in Figure 4.2. The
resulting object consists of nodes, arcs, isolated hubs and conglomerated hubs. An
Divisible Tilings in the Hyperbolic Plane
Figure 4.2.
253
Polygon and dual graph
isolated hub is one which is connected to another node or hubs by an arc only as
in Figure 4.2. A conglomerated hub is one or more hubs joined together, as in
cases F30, F31, and F33 in Table 6.6. There are some restrictions on the hubs
and conglomerated hubs. The modied dual graph has a tree-like structure and
therefore can be constructed by adding one component at a time. This leads to the
following proposition which is the basis of our combinatorial search.
Proposition 4.1. Let be the associated subdivided polygon constructed from a
tiled quadrilateral. Then there is a sequence of associated subdivided polygons 0 1 : : : s = such that 0 is empty and each i+1 is obtained from i by adjoining a triangle, hub or conglomerated hub along a single edge.
Proof. If we replace each hub and conglomerated hub by a node then we obtain a
tree. This follows from the fact that the modied dual graph is simply connected
because it a deformation retract of a quadrilateral. Trees may be constructed by
starting at any single node and then connecting additional nodes, one at a time,
by exactly one arc each. This method of tree construction directly translates into
the statement about adjoining the components of the associated polygons.
4.1. Polygon Construction and Elimination|No Interior Hubs. Let us
rst concentrate on those associated polygons with no interior hubs, as the development is simpler. The case for polygons with interior points requires a few
modications which are described below. Let PK be the set of polygons subdivided
into K triangles without any interior vertices. The dual graph of any one of these
polygons is a tree with K nodes and K ; 1 arcs. Now 3K sides are contributed by
the triangles, 2(K ; 1) of which are absorbed as interior edges corresponding to the
K ; 1 arcs. Thus there are K + 2 triangle sides on the boundary and that many
254
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Figure 4.3.
Associated polygons for K = 4
vertices as well. In this case we are going to geometrically represent the associated
polygon by a regular polygon for the following reasons. The rst is that the number
of such polygons is well known to be a Catalan number (see 7, p. 457])
;2K jPK j = c(K ) = K K+ 1 and hence we will call the associated divided polygons Catalan polygons. This
will help us in making sure that enumeration is complete. The second is that we
may use the dihedral symmetries of regular polygons to reduce the computational
complexity. Clearly, dihedrally equivalent associated
polygons will lead to the same
24
(
12)
divisible quadrilateral, if any. Since jP12 j = 13 = 208 012 some reduction in complexity is desirable. The following proposition demonstrates that the complexity is
reduced for low values of K:
Proposition 4.2. Let PK denote the set of subdivisions of regular (K +2)-polygons
into K triangles. Let OP K denote the set of dihedral equivalence classes of the
subdivided regular polygons. Let c(K ) = jPK j as above, oc(K ) = jOP K j. Then
oc(K ) is given by:
c(K ) + (K + 2)c( K2;1 )
oc(K ) =
K = 3 5 mod 6
2(K + 2)
c(K ) + 2(K3+2) c( K3;1 ) + (K + 2)c( K2;1 )
oc(K ) =
K = 1 mod 6
2(K + 2)
P 2;1
K
c(K ) + (K + 2)c( K2 ) + (K2+2) K=
s=0 c(s)c( 2 ; 1) K = 0 2 mod 6
oc(K ) =
2(K + 2)
and
K ;1
P 2;1
K
c(K ) + 2(K +2)3c( 3 ) + (K + 2)c( K2 ) + (K2+2) K=
s=0 c(s)c( 2 ; 1) oc(K ) =
2(K + 2)
for K = 4 mod 6: The growth is still exponential but at least is manageable for
K 12 since oc(12) = 7528 and oc(K ) v c(K )=(2K + 4): The formulas are
given in 5] and further references are listed in sequence M2375 of the Sloan integer
sequence reference 11].
The adjoining process leads to a combinatorial divided polygon (fPi g fPi Pj g
): The next step is to determine which (l m n) lead to divisible hyperbolic
quadrilaterals. To do this we need to determine the type of the vertices on the
Divisible Tilings in the Hyperbolic Plane
255
boundary. Suppose that Pi Pj Pj Pk and Pk Pi are the three edges of an arbitrarily
chosen triangle = Pi Pj Pk , contained in . Declare the types of Pi Pj and Pk
to be P Qand R respectively. The reection of in at least one of its sides must
lie in : Suppose for the sake of argument that this side is Pi Pj : Then there must
be a Pk0 such that Pj Pk0 and Pk0 Pi belong to the edge set of our combinatorial
polygon. Now Pk0 is the reected image of Pk and so they must have the same
type, namely R. Since the Pk0 is found by examining only the combinatorial data
(fPi g fPi Pj g) we say that Pk0 is the combinatorial reection of Pk : Combinatorial
reection may be continued until every vertex is assigned a unique type.
Once the vertices have been assigned a type then the vertices of type R P and
Q may be assigned an angle of l m n : Let mi be the integer so chosen for Pi .
Next we assign to each vertex Pi the number si of triangles that meet at the vertex.
This can be determined combinatorially by noting that there must be si + 1 edges
of the form Pi Pj in the edge set E@ : The angle measure at vertex Pi is msii . Finally
we must select four vertices in V@ to serve as the corners of the quadrilateral, or
actually K ; 2 corners to atten into a straight angle. At this stage many of the
congurations are eliminated.
Let us illustrate the process by discussing the case K = 4: Though jP4 j = 14
there are only 3 dihedrally inequivalent associated polygons. See Figure 4.3. The
list of vertices, in counter-clockwise order, starting at the \3 o'clock vertex", the
vertex types, and the angle measures divided by are as given in the following
table
Vertex Case 1 Case 2 Case 3
P1
P2
P3
P4
P5
P6
P m1
Q n4
P m1
R 2l
P m2
R 2l
P m1
Q n3
R 2l
Q n1
P m3
R 2l
P m1
Q n3
R 1l
P m3
Q n1
R 3l
Now from the six vertices we must select two to atten out to a straight angle,
the remaining vertices are corners of the quadrilateral. To be a straight angle we
must have msii = or si = mi . This automatically forces corners with a single
triangle to be a quadrilateral corner, as is geometrically obvious. In Case 1 we can
either choose fP4 P6 g or fP2 P5 g to atten. For, P1 and P3 cannot be chosen and
if we choose one of P4 or P6 we are forced to also choose the other since the angle
measure for both is 2l : If we choose fP4 P6 g then l = 2 and both n4 and m2 are
reciprocals of integers. It follows that (l m n) must have the form (2 2d 4e) and
(s t u v) must have the form (2d e 2d d): The values of d and e are d 2 e 1
except that d = 2 e = 1 is disallowed since would then be Euclidean. If we
choose fP2 P5 g then n = 4 m = 2 and (l m n) must have the form (2d 2 4)
and (s t u v) must have the form (2 2 d d) with d 3: The complete analysis of
256
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
K = 4 without interior hubs is given in the table below.
Case attened corners (l m n) (s t u v)
restrictions
1
P4 P6
(2 2d 4e) (2d e 2d d) d 2 e 1 (d e) 6= (2 1)
1
P2 P5
(2d 2 4) (2 2 d d) d 3
2
P3 P6
(2 3d 3e) (3d e 3e d) d e 2
2
P2 P5
(2d 3 3) (3 d 3 d) d 2
3
P2 P4
(3d 3 3) (3 3d 3 d) d 2
For larger values of K the \wheat to cha ratio" decreases drastically so we
need to identify some methods of quickly rejecting combinatorial polygons. Let
R1 : : : RL be the vertices of type R and let 1 : : : L be the angle multiplicities
at these points. Dene P1 : : : PM 1 : : : M and Q1 : : : QN 1 : : : N be
similarly dened. These quantities satisfy the following relations:
L+M +N = K +2
1 + : : : + L = K
1 + : : : + M = K
1 + : : : + N = K
Next let l = lcm(1 : : : L ) and write i = lsi : There only two ways that we can
assign an angle l to R-vertices so that i l is an integer submultiple of . Either
we set the angle to be l in which case the angle of the hub at Ri is i l = si or we
set it to l d d 2 and then the angle at hub at Ri is i l d = si d : In the rst case
we get an edge hub for each si = 1 otherwise we get a corner hub. In the second
case every hub is a corner hub and we get a free vertex. Similar considerations
apply to the vertices of type P and Q:
An example will help illustrate. Consider the associated polygon for K = 10 in
Figure 4.1. Proceeding counter-clockwise around the polygon from the three o'clock
position vertices may be labeled R P Q P R Q P R Q R P Q with multiplicities
3 3 4 1 3 1 3 3 4 1 3 1: The sequences of multiplicities are:
fi g = f3 3 3 1g fi g = f3 1 3 3g fi g = f4 1 4 1g:
Thus we can choose either 1 or 4 corner hubs of type R 1 or 4 corner hubs of type
P and 2 or 4 corner hubs of type Q: Of the eight possible choices only one yields
a quadrilateral, namely, l = 3 m = 3 and n = 4 yielding 1 1 and 2 corner hubs
of types R P and Q respectively. Thus we get a (3 3 4) (3 4 3 4) tiling pair
yielding case C7 in Table 6.7. Note, for instance, that this method also allows us
to construct a (d 3 4d d 3 4d) hexagon tiled by a (3 3 4d) triangle.
4.2. Computer Algorithm and Extension to Interior Hubs. Two Maple
worksheets catpolys.mws and hubpolys.mws 15] were used to implement the search
for divisible quadrilaterals with K 12. Maple was used so that the graphical
capabilities could be used to draw various polygons (such as the gures in this
section) to check the logic of the program during development. The algorithm
steps were the following.
1. Create a sequence of les FK containing the representatives of each dihedral
equivalence class of Catalan polygons with a given number of sides.
a) Start o with F1 consisting of a triangle.
Divisible Tilings in the Hyperbolic Plane
257
b) From FK create the le GK +1 consisting of all polygons that can be
created from polygons in FK by adding a triangle to each side of a polygon
from FK : The polygons are created as an ordered list.
c) Create an empty le FK +1 : Place the rst element of GK +1 in FK +1 .
Create the dihedral orbit of rst element of GK +1 and then remove this
list of polygons from GK +1 : Repeat the procedure until GK +1 is empty.
2. For each polygon in FK do the following:
a) Compute the multiplicity of all vertices.
b) Label P1 and P2 R and Q respectively. Find the unique point P = Pk
which completes the triangle. Now repeatedly use combinatorial reection
to label vertices of the polygon with P Q or R: This part can be written
so that it is guaranteed to terminate.
c) Construct the fi g fj g fk g and determine all possible assignments of
l m and n that lead to 4 vertices which are corners. This part can be
easily modied to nd tilings of pentagons, hexagons, etc.
d) Collect the valid assignments into a le QK :
The algorithm for the case of interior hubs requires just a few modications.
1. Construct by hand a list consisting of a triangle and all hubs and conglomerated hubs with twelve or less triangles.
2. Start o with F4 containing the hub of 4. Next create G5 by adding triangles
to the element in F4 at all possible locations.
3. Find representatives of the dihedral orbits and place them in F5 :
4. Construct further GK by attaching a triangle, hub or conglomerated hub to
polygons in sets with smaller numbers of triangles, so that the total number
of triangles is K: The rst non-triangle addition will be a hub of 4 to another
hub of 4. Create FK to be a set of representatives of dihedral orbits.
5. Create a labeling scheme for all vertices including interior vertices.
6. Create all valid assignments of angles to boundary vertices.
7. Check for compatibility with interior vertices, since the angles here are predetermined.
5. Boundary Construction Method (K > 12)
The boundary construction method was used for all divisible tilings for which
K > 12 since in this case we are guaranteed that the tiling has no free vertices and
hence there are only a nitely many (l m n)-triangles to consider. After describing
the methods in the next two subsections we will do some sample calculations to
illustrate the ideas.
According to the discussion in the previous sections we shall do the following.
1. Enumerate all of the (l m n)-triangles for which l m n (l m n) 41 l m n 60 and n (2 ; n4 )=(l m n): The table of -values in Appendix A
can be used here.
2. For each triangle in Step 1, enumerate all candidate quadruples (s t u v)
formed from l m n: The numbers s t u v must be selected from the divisors
of l m n which are greater than 1.
3. For each candidate pair of an (l m n) and an (s t u v) select those which
pass the K -test and the hub tests, i.e., the quotients calculated in (9), (10)
258
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
and (11) must be integers. In addition we eliminate those cases in which
)
K = ((stuv
lmn) 12 since these have already been determined by the direct
construction method.
4. For each candidate pair resulting from 3, enumerate all possible assignments
of vertex types of the triangle to the quadrilateral. This allows us to take
multiple corners into account.
5. For each candidate pair resulting from 4, perform the quadrilateral search
algorithm, as we describe next.
Quadrilateral Search. Having found an (l m n) and an (s t u v) such that all the
restrictions in 1{4 hold, quadrilaterals tiled by the triangle are sought out using
a Maple program tilequad.mws (see 15]). Two somewhat dierent programs were
developed, though they dier only in how the quadrilateral test is implemented.
The quadrilateral search program is given an (l m n)-triangle, and a quadruple
(s t u v) and tries to nd to a corresponding quadrilateral along the lines of the
triangle tiling.
Look at Figures 5.1{5.3 or the gures in the tables in Section 6 to help understand
how the algorithm works. In addition to the information (l m n) and (s t u v)
we need an assignment of vertex types and hub multiplicities to the corners of the
quadrilateral. We must consider all compatible type assignments. For example if
(l m n) = (2 4 5) and (s t u v) = (2 4 2 4) then the possible vertex assignments
are (R P R P ) (P P R P ), (R P P P ) and (P P P P ) with hub multiplicities
(1 1 1 1), (2 1 1 1), (1 1 2 1) and (2 1 2 1), respectively. Special attention needs
to be paid to isosceles triangles. For instance, in attempting to tile a (5 5 5 5) with
a (2 5 5) 16 = 24 type assignments should be considered.
Suppose that the type assignment is (S1 S2 S3 S4 ) where Si 2 fP Q Rg: The
algorithm starts by picking the vertex A0 of type S1 on the master tile. Pick an edge
of the angle at A0 and move out along the ray determined by this edge. We move
along the ray and observe the triangles we meet on the same side of the ray as our
original triangle. We stop at some (there are many choices) triangle whose second
vertex encountered is of type S2 : Call this vertex B 0 our rst side will be A0 B 0
(see Figures 5.2 and 5.3). To get the second side we turn the corner through ; t
radians toward the side of the original triangle and proceed along the next ray in
the tiling and stop at a vertex of type S3 say C 0 to produce the second side B 0 C 0 :
Because of the compatibility conditions, when we turn the corner we will move out
along a line of the tiling. Turn again through ; u radians to produce a third
side C 0 D0 and turn again to produce a fourth side D0 E 0 : When we have nished, if
A0 = E 0 and angle D0 A0 B 0 has measure s then we have a quadrilateral of the desired
type. Note that there are only a nite number of possibilities for each side of the
quadrilateral since the number of hubs on the edges is bounded by the hub numbers,
as illustrated in the attempted (2 3 7) tiling of a (7 7 7 7) quadrilateral later in
this section. Now the program keeps track of the four \combinatorial" sides of the
quadrilateral by writing down the sequence of vertex types that we pass through as
we move along an edge of the proposed boundary of the quadrilateral. This follows
from the following observation:
Remark 5.1. Let ` be a line of the tiling, and let : : : S;1 S0 S1 : : : be the biinnite sequence of types of vertices lying on ` in the order in which they occur.
Then, the entire sequence is determined by the types of two adjacent vertices. For
Divisible Tilings in the Hyperbolic Plane
259
example, in a (2 3 7) triangle only one sequence occurs f: : : R P Q R Q P : : : g :
In a (2 4 5) triangle RPQ two dierent sequences occur f: : : R P : : : g and
f: : : P Q R Q : : : g : For a (2 4 6) triangle three sequences occur f: : : R P : : : :g f: : : R Q : : : g and f: : : P Q : : : g. In each of the sequences above, the entire
sequence is the bi-innite cyclic repetition of the basic pattern shown. The pattern
types only depend on parity of the numbers l m and n.
Now that we have the four edges, how can we determine if it is a quadrilateral
with the correct angles. There are two methods that were used, which are explained
in the next two subsections.
Remark 5.2. Though one really only needs one type of test, two were developed
for the following reasons. The rst test is faster to calculate and less susceptible to
round o error. The second test produces \boundary words" which can be used to
determine a surface of minimum genus which supports both tilings. See 4].
5.1. Geometric Quadrilateral Test. The rst method uses the construction of
a quadrilateral given by the following proposition which is proved in 2].
Proposition 5.3. Let be four angles satisfying 0 < 2 and
+ + + < 2: Then for every value of h satisfying
+1
(14)
h > cossin cos
sin there is a quadrilateral ABCD unique up to congruence, such that
m(\DAB ) = m(\ABC ) = and
m(\BCD) = m(\CDA) = AB = h:
The algorithm so far has given us a set of points A0 B 0 C 0 D0 E 0 such that all
the lengths are known from the lists of points on the edges. For, the lengths of the
sides of the triangles are easily computed and the types of all edges comprising a side
of a quadrilateral are known. Let h be the length of the side A0 B 0 and construct
quadrilateral ABCD as in the proposition above with = s = t = u = v : The algorithm constructs the points A B C D in the unit disc and so the
lengths
BC ,
CD
calculated. If any of the dierences
and DA may
be numerically
BC ; B 0 C 0 CD ; C 0 D0 DA ; D0 E 0 exceed an appropriately chosen then
the candidate quadrilateral does not close up. Because Maple is used we can specify
that the calculations be carried out to a large number of decimal places to guarantee
good accuracy in the calculation of the lengths, and therefore that a small value
of may be chosen. If all the dierences are less than the tolerance then the
candidate tiling is directly constructed, using the Matlab drawing scripts.
5.2. Boundary Word Test. Our second method uses group theory. For each
of the possible quadrilateral boundaries the program creates four edge words in
p q and r by making reections through the various hubs along an edge of the
quadrilateral. The four edge words are concatenated to form a boundary word.
The four edges close up to form a quadrilateral if and only if the boundary word
260
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
reduces to the identity. The pattern of hubs along the boundary allow us to quickly
write down the boundary word. The details of determining the boundary word
and showing it is the identity is described in detail in some sample calculations in
Subsection 5.4.
By attempting to draw a quadrilateral in the triangle tiling, as above, we have a
word which we want to prove is the identity. There are two possible approaches. The
rst approach is see if the word can be reduced to the identity, using the relations
that we have in the group T . Though in any specic example it always seems
possible to nd a reduction to the identity (when it exists) the authors decided
against attempting to develop a computer algorithm to decide whether a word was
reducible to the identity because of apparent complexity of doing so suggested by
9, p. 672]. If the maximum length L of the boundary words to be reduced is known
beforehand the Knuth-Bendix algorithm 6, p. 116] produces, for each (l m n), a
nite list of substitution rules which, if applied recursively to a given word of length
at most L, will reduce to the identity if and only if the word is the identity. The
length of each edge word is (very crudely) bounded by the number of triangles in
the quadrilateral and so L 4 60 = 240. Thus it is possible (and interesting), in
principle, to devise a reduction algorithm, however more work is required than by
using the second approach that we describe next.
The second approach converts a product of an even number of reections into a
matrix by means of a homomorphism q : T ! PSL2(C ) which we describe shortly.
The matrix images are computed numerically using Maple, which introduces some
error into our calculations due to rounding o the entries in the matrices. This
can be remedied that noting that if the computations are carried out with great
accuracy, say to 50 decimal places as done in the study 8] (see Remarks 5.5 and 5.6),
then a product which is not computed to be within a prescribed distance of the
identity will not equal the identity.
We now describe the map q and state a theorem that guarantees that our conclusions based on approximate calculations are justied. The reections in the sides
of an (l m n)-triangle are inversions in the circle dening the side (or reection in
a diameter). Every such reection has the form TA " where
TA (z ) = az + b A = ab ab aa ; bb = 1
bz + a
(z ) = z:
The matrix is determined only up to a scalar multiple of 1. If the inversion is in
a circle centered at z0 and perpendicular to the unit disc then A has the form
i
z
;
1
0
A = pz z ; 1 1 ;z :
0
0 0
Otherwise, the line is a diameter meeting the positive x-axis at the angle and A
has the form.
i
A = e0 e;0i
If two of the edges of the (l m n)-triangle are diameters through the origin, then
the third side is easily determined using analytic geometry (see 2]). Also note
that the origin is a vertex of the tiling. Let P Q R be the matrices such that the
Divisible Tilings in the Hyperbolic Plane
261
reections p q r are associated to:
p $ TP " q $ TQ " r $ TR ":
Then the elements a = pq b = qr and c = rp are associated to:
a $ TPQ b $ TQR and c $ TRP since
" TA " = TA and
TAB = TA TB where A is obtained by conjugating the entries of A:
Now a word w in p q r in T representing the identity must have an even number
of factors since an odd word is orientation reversing. Thus w is a word in a = pq
b = qr c = rp a;1 = qp b;1 = rq and c;1 = pr and the matrix corresponding to
w is obtained by making substitutions based on:
a ! PQ
(15)
b ! QR
c ! RP:
The following theorem now guarantees when an element which is approximately
equal to the identity is actually equal to the identity. We rst dene the L1 norm
on matrices:
kAk = max
(jA(k l)j)
kl
Proposition 5.4. Let T be the group of orientation-preserving isometries generated by an (l m n)-triangle in the hyperbolic plane. Assume that a vertex of is at the origin. Let A 2 PSL2(C ) be a matrix representing an element of g 2 T
according to the substitution in (15). Then, there is an depending on and
not on A such that if min(kA ; I k kA + I k) then g is the identity in T:
The proposition gets used in the following way. Suppose that the matrix A
^ Suppose further that
in
has been computed approximately
as A:
the proposition
^
^
by controlling the accuracy
A ; I 2 by calculation and that A ; A
2
of the computation. Then, kA ; I k A ; A^ + A^ ; I : and so that g is
the identity.
0+b = b=a: Observe that is one of
Proof. Let A = ab ab and let = g 0 = ba0+
a
the vertices of the tiling. There is a small hyperbolic ball of radius h such that 0 is
the only vertex contained in that ball and hence ( 0) > h unless = 0. We may
take h to be any number smaller than all the sidelengths of the master tile. Let
= tanh(h=2) be the Euclidean radius of this ball. Thus j j > unless = 0.
But since min(kA ; I k kA + I k) then jbj : Also, as aa ; bb = 1 then
i
jaj 1 and hence j j = jjabjj : This forces = 0 and so A = e0 e;0i :
p
But then min(kA ; I k kA + I k) = ei ; 1
= 2 ; 2 cos(): The smallest value
262
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
p
of this expression is 2 ; 2 cos( b ) where b is the largest of l m and n. Thus we
need only to adjust to be smaller than this.
Remark 5.5. For a given (l m n) the coe$cients of PQ QR and RP all belong
to nite extension of Q . Therefore it is possible to exactly calculate words in the
matrices by symbolic means. However the \cure" of exact symbolic computation is
worse than the \disease" of approximation.
Remark 5.6. The 50 decimal places used in 8] are probably overkill. However,
the scripts can be run automatically, computing with a large number of decimal
places without an onerous time penalty. It is not to hard to show that an error
bound for q-image of a word of length L is (2M )L=2 where M is the maximum of
the entries of the matrices in Remark 5.5, and is the maximum error of individual
matrix entries. Thus the required number decimal places is linear in the word
length.
5.3. Example: Failure of (2 3 7) to Tile (7 7 7 7). The largest possible
value of K occurs for the attempt to tile a (7 7 7 7)-quadrilateral by a (2 3 7)triangle. The candidate pair passes the K -test and the hub tests since K = 60
h2 = 30, h3 = 20, and h7 = 8. Each of the four edges of any quadrilateral must
have two 7-hubs. For, an edge with only one 7-hub and terminating 7 angles must
close up to a (7 7 7) triangle as in Figure 5.1 below. Similarly there cannot be zero
7-hubs, as Figure 5.1 also shows. Thus every edge must have exactly two 7-hubs,
and hence there is a unique way to construct the quadrilateral. But Figure 5.2
shows how the attempt to construct such a quadrilateral clearly fails.
Fig. 5.1.
(2 3 7) tiling of (7 7 7)
Fig. 5.2.
A failed boundary search
5.4. Example: Successful Tiling of (5 5 5 5) by (2 4 5). Now let us examine a quadrilateral which can be subdivided by triangles. Let us subdivide the
(5 5 5 5)-quadrilateral with the (2 4 5)-triangle. See Case C25 in Table 6.7 for a
picture of the subdivided tiling. It is found that K = 24 h2 = 12 h4 = 6 and
h5 = 4: By looking at the tiling in Section 6 it is clear the any proposed edge of a
(5 5 5 5)-quadrilateral without a 5-hub closes up to a triangle. Since h5 = 4 then
there must be exactly one 5-hub on each side. This greatly restricts the boundary searches. Let us consider the lower boundary tiles on the lower edge of the
quadrilateral as pictured in Figure 5.3.
Divisible Tilings in the Hyperbolic Plane
263
The far left triangle as oriented is a (5 4 2)-triangle. Though the order of l m and
n is unimportant for rst three steps of the search, the ordering is very important
in the quadrilateral construction phase. Thus for this particular triangle we have:
(pq)5 = a5 = 1
(qr)4 = b4 = 1
(rp)2 = c2 = 1:
Now let us construct the edge word corresponding to this part of the boundary.
We label the triangles 0 1 : : : 8 as we move from left to right. It is obvious
that 1 = r0 since we reect over the r edge. The reections in sides of 1
are rpr, rqr and r = rrr: In fact if 0 = g0 for some g 2 T the reections
in the sides of 0 are gpg;1, gqg;1 and grg;1 : From the picture we see that 2
is the rqr image of 1 and hence 2 = (rqr)r0 = rq0 : Continuing one more
step, we see that the reections in the sides of 2 are rqpqr, rqr = rqqqr and
rqrqr. Now 3 is obtained by reecting 2 over the common rqrqr edge. Thus
3 = rqrqr2 = (rqrqr)rq0 = rqr0 : The pattern is now evident which we now
state as a proposition. We omit the easy induction proof.
5.3. Boundary tiles of lower edge
Proposition 5.7. Let 0 1 : : : s be a sequence of tiles such that i and i;1
meet in edge ei : Let ri the unique edge among fp q rg such that ei matches ri by
the unique transformation in T carrying 0 to i : Let ri also denote the reection
in ri : Then
i = r1 r2 ri 0 for i = 1 : : : s:
Applying the proposition we get
8 = rqpqprqr0 = w1 0
Continuing on with the next edges we get:
16 = w2 w1 0 24 = w3 w2 w1 0 32 = w4 w3 w2 w1 0 :
Figure
264
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
where w1 = w2 = w3 = w4 = rqpqprqr: The fact that all the edge words w1 w2 w3 ,
and w4 are equal is a consequence of the four-fold rotational symmetry of the
quadrilateral. Observe that the edgewords are completely determined by the initial
orientation of 0 and the number and type of hubs we pass through along an edge.
The process of nding the word can be sped up by observing that passing through
a hub along the edge concatenates a well-dened subword to the edge word. Thus
one multiplies together a sequence of \hub words".
The boundary word we have is (rqpqprqr)4 : The candidate boundary closes up
if and only if and only if 32 = 0 i.e., if and only if (rqpqprqr)4 0 = 0 : Since
T acts simply transitively on the triangles then the boundary closes up if and only
if
(rqpqprqr)4 = 1:
But we have
(rqpqprqr)4 = rqpqprq(rr)qpqprq(rr)qpqprq(rr)qpqprqr
= rqpqpr(qq)pqpr(qq)pqpr(qq)pqprqr
= rqpq(prp)q(prp)q(prp)qprqr
Now prpr = 1 so prp = r and likewise qrqrqrq = r: Thus we further obtain
(rqpqprqr)4 = rqp(qrqrqrq)prqr
= rq(prpr)qr
= rqqr
= 1:
We have exactly computed the expected reduction. Alternatively, we could translate the boundary word into a word in a b c and compute the matrix image in
PSL2(C ): One simply takes the letters in boundary word two at a time with the
replacements: p2 = q2 = r2 = 1, pq = a, qr = b, pr = c, qp = a;1 , rq = b;1 and
pr = c;1 : Our word and the replacement matrices are (to 5 decimal places):
(rqpqprqr)4 = (b;1 ac;1 b)4 :
a ;! A =: :80903 + :58778i
0
0
:80903 ; :58778i
:
70708
+
:
97325
i
;
:
66876
i
:
b ;! B =
:66876i
:70708 ; :97325i
;
1
:
2030
i
;
:
39308
+
:
54105
i
:
c ;! C = ;:39308 ; :54105i
1:2030i
The test matrix is seen to be approximately a scalar matrix:
;
:
9999
0
;
1
;
1
4 :
(B AC B ) =
0
;:9999 :
The matrix calculations are done to 5 decimal places here to ensure clarity. By
computing to a greater number of digits enough accuracy can be achieved to conclude that the boundary word is the identity. The matrix method is preferable
since the computer (i.e., programmer) need not get creative about how to reduce
words.
Divisible Tilings in the Hyperbolic Plane
6. Catalogue of Divisible Tilings
265
This section contains a complete catalogue of all (s t u v)-quadrilaterals which
can be subdivided by (l m n)-triangles. For completeness we have also included the
small number of tilings of triangles by triangles and quadrilaterals by quadrilaterals.
For each category there are two tables, a table of data and a table of gures.
Theorem 6.1. Let and denote a hyperbolic, kaleidoscopic (l1 m1 n1)-triangle
and (l2 m2 n2 )-triangle respectively. Then there is one 2-parameter family, and ve
1-parameter families of divisible tiling pairs in which there are free vertices.
There are two exceptional tiling pairs with constrained vertices only. These
families and exceptional cases are listed in Table 6.1 and pictured in Table 6.5.
Theorem 6.2. Let and denote a hyperbolic, kaleidoscopic (l m n)-triangle
and (s t u v)-quadrilateral respectively. Then there is one 3-parameter family, ve
2-parameter families, and 28 1-parameter families of divisible tiling pairs in which there are free vertices. These families are listed in Table 6.2 and pictured
in Table 6.6. In addition there are 27 divisible tiling pairs with constrained
vertices only. These are listed Table 6.3 and pictured in Table 6.7
Theorem 6.3. Let and denote a hyperbolic, kaleidoscopic (s1 t1 u1 v1)-quadrilateral and (s2 t2 u2 v2 )-quadrilateral respectively. Then there is one 2-parameter
family, and one 1-parameter family of divisible tiling pairs in which there
are free vertices. There are no examples with constrained vertices only. The two
families are listed in Table 6.4 and pictured in Table 6.8.
Some notes on the tables:
Although it is usual to arrange for l m n the geometry of a triangle
being included in a larger polygon may force a dierent ordering. For both
the triangle and the quadrilateral the ordering is obtained by moving about
the gure in a counter-clockwise sense. A dierent orientation is obtained
from the reected quadrilateral. These permutations of the ordering are in
the tables of pictures only, where comparison of the picture and the ordering
makes sense.
The groups T and Q denote the groups generated by reections in the sides
of and respectively. Then Aut() which is equal to the group taking
to itself and StabT () = T \ Aut() are also given in the tables. The
groups are computed by visual inspection of the pictures. These groups are
useful in determining the structure of tiling groups of surfaces of minimal
genus supporting both tilings. See 8] and the forthcoming paper 4] for more
details.
266
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Case
TF1
TF2
TF3
TF4
TF5
TF6
TC1
TC2
Table 6.1. Triangles subdivided by triangles
(2 d 2e)
(2 3 2d)
(2 3 3d)
(2 4 2d)
(2 3 2d)
(2 3 4d)
(2 3 8)
(2 3 7)
(d d e)
(2 d 2d)
(3 d 3d)
(d 2d 2d)
(d d d)
(d 4d 4d)
(4 8 8)
(7 7 7)
restrictions
2
1
d+e <1
d4
d4
d3
d4
d2
K StabT () Aut()
2 Z2
3 trivial
4 trivial
4 Z2
6 D3
6 Z2
12 Z2
21 Z3
Z2
trivial
trivial
Z2
D3
Z2
Z2
Z3
Divisible Tilings in the Hyperbolic Plane
267
Table 6.2. Quadrilaterals subdivided by triangles with free vertices
Case
F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14
F15
F16
F17
F18
F19
F20
F21
F22
F23
F24
F25
F26
F27
F28
F29
F30
F31
F32
F33
F34
(d 2e 2f )
(2 2d 3e)
(3 2d 3e)
(2 2d 2e)
(2 2d 4e)
(2 3d 3e)
(2 4 2d)
(3 3 2d)
(3 3 3d)
(2 4 6d)
(2 5 2d)
(2 3 3d)
(2 3 4d)
(2 4 3d)
(2 4 6d)
(2 5 6d)
(2 6 2d)
(2 3 10d)
(2 3 12d)
(2 3 4d)
(2 3 6d)
(2 4 2d)
(2 4 3d)
(2 4 4d)
(2 3 12d)
(2 3 15d)
(2 3 6d)
(2 3 14d)
(2 3 20d)
(2 3 30d)
(2 3 4d)
(2 3 5d)
(2 3 6d)
(2 3 8d)
(d e d f )
(2 d 2d e)
(d e 2d 2e)
(d e d e)
(d 2d e 2d)
(d 3d e 3e)
(2 2 d d))
(3 d 3 d)
(3 d 3 3d)
(2 4 2d 3d)
(2 d d 2d)
(2 d 2 d)
(2 2d 2 d)
(4 4 d d)
(2 3d 6d 2d)
(4 2d 3d 6d)
(d d 2d 2d)
(2 5d 3 2d)
(2 3d 3 4d)
(3 d 3 d)
(3 d 3 3d)
(d d d d)
(d 3d d 3d)
(d 4d 2d 4d)
(2 4d 6d 3d)
(2 3d 15d 5d)
(2 d 6d 3d)
(3 d 14d 7d)
(3 4d 20d 5d)
(3 10d 15d 6d)
(d 2d d 2d)
(d 5d d 5d)
(d 6d 2d 3d)
(d 8d 2d 8d)
restrictions
2
1
1
d+e+f <2
d 2 e 2
1
4
1
d+e<3
d 2 e 2
d 2 e 2
1
3
1
d+e<2
d3
d2
d2
d1
d2
d2
d2
d2
d1
d1
d2
d1
d1
d2
d2
d3
d1
d2
d1
d1
d2
d1
d1
d1
d2
d2
d2
d1
K StabT () Aut()
2
3
3
4
4
4
4
4
4
5
5
6
6
6
6
6
6
7
7
8
8
8
8
8
9
9
9
10
10
10
12
12
12
12
Z2
trivial
Z2
Z2 Z2
Z2
Z2if d = e
Z2
Z2
trivial
trivial
trivial
Z2
Z2
Z2
trivial
trivial
Z2
trivial
trivial
Z2
Z2
D4
Z2
Z2
trivial
trivial
trivial
trivial
trivial
trivial
Z2 Z2
Z2
trivial
Z2
Z2
trivial
Z2if d = e
D 4 if d = e
Z2
Z2if
Z2
Z2
Z2
d=e
trivial
trivial
Z2
Z2
Z2
trivial
trivial
Z2
trivial
trivial
Z2
Z2
D4
Z2
Z2
trivial
trivial
trivial
trivial
trivial
trivial
Z2 Z2
Z2
trivial
Z2
268
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.3. Quadrilaterals subdivided by triangles without free vertices
Case
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
C22
C23
C24
C25
C26
C27
(3 4 4)
(3 3 5)
(3 4 4)
(2 4 5)
(2 4 5)
(2 4 5)
(3 3 4)
(2 4 5)
(2 3 8)
(2 4 6)
(2 4 6)
(2 5 5)
(3 3 4)
(2 3 8)
(2 3 10)
(2 3 10)
(2 3 9)
(2 4 5)
(2 3 12)
(2 3 8)
(2 4 5)
(2 4 5)
(2 3 8)
(2 3 8)
(2 4 5)
(2 3 7)
(2 3 7)
(4 4 4 4)
(3 3 5 5)
(2 3 3 4)
(2 2 4 4)
(2 2 4 4)
(2 4 2 4)
(3 4 3 4)
(2 2 4 5)
(2 2 4 4)
(3 3 6 6)
(4 4 4 4)
(5 5 5 5)
(4 4 4 4)
(2 3 3 4)
(3 3 5 5)
(2 5 5 10)
(3 3 3 9)
(2 4 5 4)
(6 6 12 12)
(3 4 3 4)
(4 4 4 4)
(4 4 5 5)
(2 8 4 8)
(4 4 4 4)
(5 5 5 5)
(2 7 2 7)
(3 7 3 7)
K StabT () Aut()
6
7
7
10
10
10
10
11
12
12
12
12
12
12
14
15
16
16
18
20
20
22
24
24
24
30
44
Z2
trivial
trivial
Z2
Z2
Z2
trivial
trivial
Z2
Z2
Z2 Z2
Z2
Z4
trivial
Z2
trivial
trivial
trivial
Z2
Z2
Z2 Z2
Z2
Z2
D4
Z4
Z2
Z2
Z2 Z2
Z2
trivial
Z2
Z2
Z2
Z2
trivial
Z2
Z2
Z2 Z2
Z2
Z4
trivial
Z2
trivial
trivial
trivial
Z2
Z2
Z2 Z2
Z2
Z2
D4
Z4
Z2
Z2
Table 6.4. Quadrilaterals subdivided by quadrilaterals
Case restrictions K StabT () Aut()
QF1 (2 2 d e) (d d e e) 1d + 1e < 1 2 Z2
Z2
QF2 (2 2 2 d) (d d d d) d 3
4 Z2Z2
D4
Divisible Tilings in the Hyperbolic Plane
Table 6.5. Triangles subdivided by triangles
Case TF1: K = 2
(d 2 2e) (d d e)
Case TF2: K = 3
(2 3 2d) (2 2d d)
Case TF3: K = 4
(3 2 3d) (3 3d d)
Case TF4: K = 4
(2d 4 2) (2d 2d d)
Case TF5: K = 6
(2d 2 3) (d d d)
Case TF6: K = 6
(4d 3 2) (4d 4d d)
269
270
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.5, part 2
Case TC1: K = 12
(8 2 3) (8 8 4)
Case TC2: K = 24
(7 3 2) (7 7 7)
Divisible Tilings in the Hyperbolic Plane
Table 6.6. Divisible quadrilaterals with free vertices
Case F1: K = 2
(d 2e 2f ) (d e d f )
Case F2: K = 3
(2 2d 3e) (2 d 2d e)
Case F3: K = 3
(2d 3 2e) (2d 2e d e)
Case F4: K = 4
(2d 2e 2) (d e d e)
Case F5: K = 4
(2d 2 4e) (2d d 2d e)
Case F6: K = 4
(3d 3e 2) (3d e 3e d)
271
272
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.6, part 2
Case F7: K = 4
(2 4 2d) (2 2 d d)
Case F8: K = 4
(3 3 2d) (3 d 3 d)
Case F9: K = 4
(3 3 3d) (3 3d 3 d)
Case F10: K = 5
(2 4 6d) (2 4 2d 3d)
Case F11: K = 5
(2 2d 5) (2 d d 2d)
Case F12: K = 6
(2 3 3d) (2 d 2 d)
Divisible Tilings in the Hyperbolic Plane
Table 6.6, part 3
Case F13: K = 6
(2 3 4d) (2 2d 2 d)
Case F14: K = 6
(4 2 3d) (4 4 d d)
Case F15: K = 6
(6d 2 4) (6d 2d 2 3d)
Case F16: K = 6
(5 6d 2) (5 2d 3d 6d)
Case F17: K = 6
(2d 6 2) (2d 2d d d)
Case F18: K = 7
(2 3 10d) (2 5d 3 2d)
273
274
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.6, part 4
Case F19: K = 7
(2 3 12d) (2 3d 3 4d)
Case F20: K = 8
(3 4d 2) (3 d 3 d)
Case F21: K = 8
(3 6d 2) (3 d 3 3d)
Case F22: K = 8
(2d 2 4) (d d d d)
Case F23: K = 8
(3d 2 4) (3d d 3d d)
Case F24: K = 8
(4d 4 2) (4d 2d 4d d)
Divisible Tilings in the Hyperbolic Plane
Table 6.6, part 5
Case F25: K = 9
(2 3 12d) (2 4d 6d 3d)
Case F26: K = 9
(2 3 15d) (2 3d 15d 5d)
Case F27: K = 9
(2 6d 3) (2 d 6d 3d)
Case F28: K = 10
(3 14d 2) (3 2d 14d 7d)
Case F29: K = 10
(3 2 20d) (3 4d 20d 5d)
Case F30: K = 10
(3 2 30d) (3 10d 15d 6d)
275
276
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.6, part 6
Case F31: K = 12
(4d 2 3) (2d d 2d d)
Case F32: K = 12
(5d 3 2) (5d d 5d d)
Case F33: K = 12
(6d 3 2) (6d 2d 3d d)
Case F34: K = 12
(8d 3 2) (8d 4d 8d d)
Divisible Tilings in the Hyperbolic Plane
Table 6.7. Divisible quadrilaterals with constrained vertices
Case C1: K = 6
(4 4 3) (4 4 4 4)
Case C2: K = 7
(3 5 3) (3 3 5 5)
Case C3: K = 7
(4 3 3) (2 3 3 4)
Case C4: K = 10
(2 4 5) (2 2 4 4)
Case C5: K = 10
(4 2 5) (2 2 4 4)
Case C6: K = 10
(2 5 4) (2 4 2 4)
277
278
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.7, part 2
Case C7: K = 10
(3 3 4) (3 4 3 4)
Case C8: K = 11
(2 4 5) (2 2 4 5)
Case C9: K = 12
(2 8 3) (2 2 4 4)
Case C10: K = 12
(6 4 2) (3 3 6 6)
Case C11: K = 12
(4 2 6) (4 4 4 4)
Case C12: K = 12
(5 2 5) (5 5 5 5)
Divisible Tilings in the Hyperbolic Plane
Table 6.7, part 3
Case C13: K = 12
(4 3 3) (4 4 4 4)
Case C14: K = 14
(8 3 2) (2 3 3 4)
Case C15: K = 14
(3 10 2) (3 3 5 5)
Case C16: K = 15
(2 3 10) (2 5 5 10)
Case C17: K = 16
(9 3 2) (3 3 3 9)
Case C18: K = 16
(4 2 5) (2 4 5 4)
279
280
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Table 6.7, part 4
Case C19: K = 18
(12 3 2) (6 6 12 12)
Case C20: K = 20
(3 2 8) (3 4 3 4)
Case C21: K = 20
(4 2 5) (4 4 4 4)
Case C22: K = 20
(4 2 5) (4 4 5 5)
Case C23: K = 24
(8 3 2) (2 8 4 8)
Case C24: K = 24
(8 3 2) (4 4 4 4)
Divisible Tilings in the Hyperbolic Plane
Table 6.7, part 5
Case C25: K = 24
(5 4 2) (5 5 5 5)
Case C26: K = 30
(2 3 7) (2 7 2 7)
Case C27: K = 44
(3 7 2) (3 7 3 7)
Table 6.8. Quadrilaterals subdivided by quadrilaterals
Case QF1: K = 2
(d e 2 2) (d e e d)
Case QF2: K = 3
(d 2 2 2) (d d d d)
281
282
S. A. Broughton, D. Haney, L. McKeough, and B. Mayeld
Appendix A. Triangles with area
Angle Description
(2 3 7)
(2 3 8)
(2 4 5)
(2 3 9)
(2 3 d) d = 10 11
(2 3 12) (2 4 6) (3 3 4)
(2 3 d) d = 13 14
(2 3 15) (2 5 5)
(2 3 16)
(2 4 7)
(2 3 17)
(2 3 d) 18 d 23
(2 3 24) (2 4 8)
(2 3 d) 25 d 29
(2 3 30) (2 5 6) (3 3 5)
(2 3 d) 31 d 35
(2 3 36) (2 4 9)
(2 3 d) 37 d 59
(2 3 60) (2 4 10)
(2 3 d) 61 d 104
(2 3 105) (2 5 7)
(2 3 d) 106 d 131
(2 3 132) (2 4 11)
(2 3 d) d 133
(2 4 12) (2 6 6)
(3 3 6) (3 4 4)
(2 4 13)
(2 5 8)
(2 4 d) 14 d 16
(2 5 9)
(2 6 7) (3 3 7)
(2 4 d) 17 d 19
(2 4 20) (2 5 10)
(2 4 d) 21 d 23
1
42
1
24
1
20
1
18
d;6
6d
1
12
d;6
6d
1
10
5
48
3
28
11
102
d;6
6d
1
8
d;6
6d
2
15
d;6
6d
5
36
d;6
6d
3
20
d;6
6d
11
70
d;6
6d
7
44
d;6
6d
1
6
1
6
9
52
7
40
d;4
4d
17
90
4
21
d;4
4d
1
5
d;4
4d
4
Angle Description
(2 4 24) (2 6 8) (3 3 8)
(2 5 11)
(2 4 d) 25 d 27
(2 4 28) (2 7 7)
(2 4 29)
(2 4 30) (2 5 12) (3 4 5)
(2 4 d) 31 d 35
(2 4 36) (2 6 9) (3 3 9)
(2 4 37)
(2 5 13)
(2 4 d) 38 d 46
(2 5 14)
(2 4 d) 47 d 55
(2 4 56) (2 7 8)
(2 4 d) 57 d 59
(2 4 60) (2 5 15)
(2 6 10) (3 3 10)
(2 4 d) 61 d 79
(2 4 80) (2 5 16)
(2 4 d) 81 d 113
(2 5 17)
(2 4 d) 114 d 131
(2 4 132) (2 6 11) (3 3 11)
(2 4 d) 133 d 179
(2 4 180) (2 5 18)
(2 4 d) 181 d 251
(2 4 252) (2 7 9)
(2 4 d) 253 d 379
(2 4 380) (2 5 19)
(2 4 d) d 381
(2 5 20) (2 6 12)
(2 8 8) (3 3 12)
(3 4 6) (4 4 4)
5
24
23
110
d;4
4d
3
14
25
116
13
60
d;4
4d
2
9
33
148
29
130
d;4
4d
8
35
d;4
4d
13
56
d;4
4d
7
30
7
30
d;4
4d
19
80
d;4
4d
41
170
d;4
4d
8
33
d;4
4d
11
45
d;4
4d
31
126
d;4
4d
47
190
d;4
4d
1
4
1
4
1
4
Divisible Tilings in the Hyperbolic Plane
References
283
1] A. F. Beardon, The Geometry of Discrete Groups , Graduate Texts in Mathematics, no. 91,
Springer-Verlag, New York, 1995, MR 97d:22011, Zbl 528.30001.
2] S. A. Broughton, Constructing kaleidoscopic tiling polygons in the hyperbolic plane, Amer.
Math. Monthly, to appear.
3] S. A. Broughton, Kaleidoscopic tiling of surfaces , background notes for the Rose-Hulman
REU Tilings Project,
http://www.rose-hulman.edu/Class/ma/HTML/REU/Tilings/pubs.html#Kaleido.
4] S .A. Broughton, D. Haney, L. McKeough, B. Smith, Divisible Tilings on Hyperbolic Surfaces,
in preparation.
5] J. Dochkowa, H. Harboth, I. Mengerson, Cut set Catalan numbers , Proceedings of the
Twenty-ninth Southeastern International Conference on Combinatorics, Graph Theory and
Computing (Boca Raton, FL, 1998), Congr. Numer. 130 (1998), 133{139, CMP 1 676 458.
6] D. B. A. Epstein, J. W. Cannon, D. F. Holt, S. V. F. Levy, M. S. Paterson, and
W. P. Thurston, Word Processing in Groups, Jones and Bartlett, Boston, 1992, MR 93i:20036,
Zbl 764.20017.
7] R. P. Grimaldi, Discrete and Combinatorial Mathematics (fourth ed.), Addison Wesley, Reading, 1999, Zbl 787.05001.
8] D. Haney and L. McKeough Quadrilaterals Subdivided by Triangles in the Hyperbolic Plane ,
Rose-Hulman Math. Sci. Tech Rep. no. 98-04, 1998.
9] A. Seress, An Introduction to computational group theory , Notices of the AMS, 44 no. 6,
(June/July 1997), 671{679, MR 98e:20002, Zbl 929.20001.
10] D. Singerman, Finitely maximal Fucshian groups , J. London Math. Soc. 6 (1972), 29{38,
MR 48 #529, Zbl 251.20052.
11] N. J. A. Sloane & S. Ploue, The Encyclopedia of Integer Sequences , Acad. Press, San Diego,
1995, MR 96a:11001, Zbl 845.11001.
12] B. M. Smith, Triangle tilings of quadrilaterals in the hyperbolic plane , (in preparation).
13] MAPLE V, Waterloo Maple Inc., Waterloo, Canada.
14] MATLAB, The Mathworks, Natick, Massachusetts.
15] Rose-Hulman NSF-REU Tilings web site,
http://www.rose-hulman.edu/Class/ma/HTML/REU/Tilings/tilings.html
Rose-Hulman Institute of Technology, Terre Haute IN, 47803
allen.broughton@rose-hulman.edu http://www.rose-hulman.edu/~brought/
University of Georgia, Athens, GA 30602
haneydaw@arches.uga.edu
St. Paul's School, Concord NH
lmckeoug@sps.edu
3302 Cheyenne Court, Fairfield Twp, OH 45011
brandymayeld@hotmail.com
This paper is available via http://nyjm.albany.edu:8000/j/2000/6-12.html.
Download