Polygonal Designs: Existence and Construction

John Hegeman
Department of Mathematics, Stanford University, Stanford, CA 94305
Jeff Langford
Department of Mathematics, Drake University, Des Moines, IA 50311
G. Bhattacharyya, J. Kim and S. Y. Song ∗
Department of Mathematics, Iowa State University, Ames, Iowa, 50011, U. S. A.

September 5, 2004

Abstract

Polygonal designs form a special class of partially balanced incomplete block designs.
We resolve the existence problem for polygonal designs with various parameter sets and
present several construction methods with blocks of small sizes.

1 Introduction and preliminaries

Some results and notation that are used throughout the paper are collected in this section
for convenience.
For positive integers v, b, k, and r with 2 < k < v , 1 ≤ r < b, a (doubly regular)
incomplete block design with parameters (v, b, k, r), is a pair (V, B), where V is a set of v
elements, called points or varieties and B is a collection of b k-element subsets of V called
blocks, satisfying the condition: each point appears in exactly r blocks. An incomplete block
design is said to be balanced if any two points are contained in exactly λ blocks together.
It is called partially balanced if every pair of points occurs in a certain number of blocks
depending on an association relation between the points.
A special class of partially balanced incomplete block designs called polygonal designs
can be defined on a regular polygon. The set V forms a v-gon with vertex (point) set

V = Zv = {0, 1, 2, . . . , v − 1}.
∗
This research was supported by NSF grant DMS-0353880

1
We define the distance δ(x, y) between points x and y to be the length of the shortest path
connecting x and y on V . That is, for x, y ∈ V
jv k
δ(x, y) := min{|x − y|, v − |x − y|}, and thus 0 ≤ δ(x, y) ≤ .
2

Definition 1.1 Let V = {0, 1, . . . , v − 1} be the point set of a regular v-gon, and let
m < (v − 1)/2. An incomplete block design (V, B) with parameters (v, b, k, r) is called a
polygonal design with minimum interval m, if any two points of V that are at distance
m + 1 or greater appear together in λ blocks while other pairs do not occur in the blocks
at all. This design is denoted by PD(v, k, λ; m). We note that a PD(v, k, λ; 0) is a balanced
incomplete block design which is also known as a 2-(v, k, λ) design.

A number of authors have provided solutions to the existence and construction prob-
lems of polygonal designs for various combinations of v, b, k and λ. Some of the relevant
references, almost all of which deal primarily with the case m = 1, are as follows. Hedayat,
Rao, and Stufken ([3]) introduced polygonal designs for the first time in 1988 as balanced
sampling plans excluding contiguous units pertaining to finite population sampling. They
showed that v ≥ 3k is a necessary condition for the existence of PD(v, k, λ; 1). They also
provided an iterative construction method by showing that if a PD(v, k, λ; 1) exists, then a
PD(v + 3α, k, λ0 ; 1) exists for any positive integer α. Stufken, Song, See and Driessel ([7])
showed that if a PD(v, k, λ; m) exists, then b ≥ v and v ≥ k(2m + 1). They also showed
that a PD(3k, k, λ; 1) does not exist for any λ if k ≥ 5. In regard to the construction of
designs, Colbourn and Ling ([1], [2]) constructed all PD(v, k, λ; 1) for k = 3 and k = 4.
Stufken and Wright ([8]) constructed all possible PD(v, k, λ; 1) with k = 5, 6 and 7, except
possibly one, and several designs with block size 9 and 10.
In this paper, we study the polygonal designs with v = k(2m + 1) for an arbitrary m.
We resolve the existence of polygonal designs with v = k(2m + 1) and k = 3 completely.
We present this result in Section 2. In Section 3, we show that if a PD(k(2m + 1), k, λ; m)
exists, then so does PD((k − 1)(2m + 1), k − 1, λ0 ; m) for some λ0 . We also show that given
a PD(v, k, λ; m) we can construct PD(v + (2m + 1)α, k, λ0 ; m) for any positive integer α
with some λ0 . These results are a generalization of some of the results provided in [3] and
[7]. In Section 4, we introduce a new construction method for PD(v, k, λ; m) by using a
‘perfect (k, m)-grouping’. We also show that the inequality k(k − 1) ≤ 4(2m + 1) holds in a
PD(k(2m + 1), k, λ; m). This result confirms the non-existence of PD(3k, k, λ; 1) for k ≥ 5
which was proved in [7].
Throughout the paper, we shall use V = {0, 1, . . . , v − 1} and B = {B1 , B2 , . . . , Bb }
for the point set and block set of a PD(v, k, λ; m) unless otherwise specified. Whenever we
consider a block Bi = {bi1 , bi2 , . . . , bik } ∈ B, we shall assume that the points are ordered as
0 ≤ bi1 < bi2 < · · · < bik ≤ v − 1. Given an integer a and a block Bi , by Bi + a we shall
denote the set {bi1 + a, bi2 + a, . . . , bik + a} where elements are reduced modulo v if needed.
We shall also consider the differences x − y (computed modulo v) as well as the distances

2
δ(x, y) between the points x and y. Let Bi = {bi1 , bi2 , ..., bik }, i = 1, 2, . . . , g, be g blocks of
size k based on V = {0, 1, . . . , v − 1}. Consider the gk(k − 1) differences
bij − bil , j, l ∈ {1, 2, .., k}, j 6= l, 1 ≤ i ≤ g,
and call the multi-set
D = {bij − bil : j, l ∈ {1, 2, . . . , k}, j 6= l, 1 ≤ i ≤ g}
the difference collection of G = {B1 , B2 , . . . , Bg }. Notice that we use the set notation, the
curly bracket, for any collection of objects whether it is a set or a multi-set. We also often
allow a design to have repeated blocks. We note that in a polygonal design PD(v, k, λ; m),
the difference collection of the entire block set B consists of each integer m − 1 through
v − m − 1 equally |B|k(k−1)
v−2m−1 times.

Definition 1.2 A collection G = {B1 , B2 , ......, Bg } of k-subsets Bi of a v-set V is called a

generating set of a PD(v, k, λ; m) (V, B) if
B = {Bi + a : 0 ≤ a ≤ ti − 1, 1 ≤ i ≤ g}
where ti is the smallest positive integer such that Bi + ti ≡ Bi (mod v). A polygonal design
is cyclically generated if and only if it has a generating set.

Note that for most blocks in the generating set ti will simply be v. However, for blocks
that are ‘rotationally symmetric’, ti will be a proper divisor of v. When the differences are
computed modulo v, each distance represented in a block, Bi , will occur a multiple of v/ti
times in that block. Suppose all ti are equally v for a set G of blocks. Then it follows that G
generates a polygonal design if and only if each of the differences m+1, m+2, ..., v −(m+1)
are represented in the difference collection of G exactly λ times. This statement can be
modified slightly when G has a rotationally symmetric block, that is, there is a ti that is a
proper divisor of v.
We shall restrict our attention to cyclically generated designs. We note that there are
no known polygonal designs that are not cyclically generated. It is obvious that whenever
we have a polygonal design we can derive a cyclically generated polygonal design, perhaps,
with a larger block set while keeping v, k and m constant.
For given m and k the polygonal designs PD(v, k, λ; m) can possibly exist only when
v ≥ (2m + 1)k and r = λ(v − 2m − 1)/(k − 1). When we look at the class of polygonal
designs with v = (2m + 1)k (in this case, we must have b = (2m + 1)2 λ and r = (2m + 1)λ),
the following lemma, which is Corollary 3.2 in [7] is very useful. (For the proof of this
lemma we refer the reader to [7]).

Lemma 1.1 Let B = {b1 , b2 , . . . , bk } be a block in a PD((2m + 1)k, k, λ; m), where b1 <
b2 < · · · < bk . If 1 ≤ i < j ≤ k, then the difference bj − bi lies in the set
{(j − i)(2m + 1) − m, (j − i)(2m + 1) − (m − 1), . . . , (j − i)(2m + 1) + m}.

3
2 Existence and construction of PD(6m + 3, 3, λ; m)

In this section, we present a series of existence theorems, whose proofs provide construction
methods for polygonal designs of block size k = 3 on 3(2m + 1) points with the exception
of the case when m ≡ 2 (mod 3) and λ = 1. For this exceptional case a cyclically generated
polygonal design does not exist.
For these constructions, we will be concerned with the distances (rather than differ-
ences), as by obtaining the distances of m + 1, m + 2, ..., 3m + 1, we get all of the differences
m + 1, m + 2, ..., 5m + 2. For polygonal designs PD(6m + 3, 3, λ; m) we can write the possible
sets of distances, contributed to the difference collection D of a generating set, by a single
generating block in the form (d1 , d2 , d3 ). For this to be a possible triple of distances we
must have either d1 + d2 + d3 = 6m + 3 or d1 + d2 = d3 . In the case that we are examining;
i.e, with k = 3 and v = 6m + 3, d1 + d2 = d3 is excluded as a possibility because d1 + d2
must be in {3m + 2, 3m + 3, . . . , 5m + 2} due to Lemma 1.1. The following technical lemma
will be used repeatedly.
 3n+1 
Lemma 2.1 The integers 1, 2, 3, ..., 3n + 1 excluding 2 can be partitioned into triples
(Y1 , Y2 , Y3 ) such that Y1 + Y2 = Y3 .

Proof: In the following we present two side-by-side tables having headers Y1 , Y2 , and Y3 .
Each row from each table presents a valid Y1 , Y2 , Y3 triple. These tables give the required
partitions where each integer appears in exactly one triple.
Case 1: n is even
Y1 Y2 Y3 Y1 Y2 Y3
n n + 1 2n + 1 n − 1 2n + 2 3n + 1
n−2 n+2 2n n − 3 2n + 3 3n
n − 4 n + 3 2n − 1 n − 5 2n + 4 3n − 1
.. .. .. .. .. ..
. . . . . .
3n 3n 5n 5n
2 2 2 +2 1 2 + 1 2 +2

Case 2: n is odd
Y1 Y2 Y3 Y1 Y2 Y3
n 2n + 1 3n + 1 n−1 n+1 2n
n − 2 2n + 2 3n n − 3 n + 2 2n − 1
n − 4 2n + 3 3n − 1 n − 5 n + 3 2n − 2
.. .. .. .. .. ..
. . . . . .
.. .. .. 2 3n 3n
. . . 2 2 +2
5n+1 5n+1 .. .. ..
1 2 2 +1 . . .

4
Theorem 2.2 PD(6m + 3, 3, 1; m) exists for m ≡ 1 (mod 3).

Proof: To demonstrate this we partition the distances m + 1, m + 2, ..., 3m + 1 into triples

(d1 , d2 , d3 ) such that d1 + d2 + d3 = 6m + 3 and each distance is in exactly one triple. First,
rewrite the distances m+1, m+2, ..., 3m+1 as (2m+1)−m, (2m+1)−(m−1), ..., (2m+1)+m.
We use Lemma 2.1 in order to partition the integers 1, 2, . . . , m with m = 3n + 1 into
triples (Y1 , Y2 , Y3 ) such that Y1 +Y2 = Y3 . Let (Y1 , Y2 , Y3 ) be one such triple in the partition
of 1, 2, ..., m. We then see that the triples ((2m + 1) − Y1 , (2m + 1) − Y2 , (2m + 1) + Y3 )
and ((2m + 1) + Y1 , (2m + 1) + Y2 , (2m + 1) − Y3 ) are both valid distance triples, as both
sum to 6m + 3. Therefore, for each triple (Y1 , Y2 , Y3 ) of elements from {1, 2, ..., m} we have
two valid distance triples. Now we observe that every distance m + 1 through 3m + 1 except
for lmm lmm
2m + 1 − , 2m + 1, and 2m + 1 +
2 2
m
appears in exactly  m one of the distance triples. However, the distances 2m + 1 − 2 , 2m + 1
and 2m + 1 + 2 clearly form a final valid distance triple as they sum to 6m + 3. As all
distances now appear in one distance triple we can form a generating set for the PD(6m +
3, 3, 1; m) by forming a generating block {1, d1 + 1, d1 + d2 + 1} for each distance triple
(d1 , d2 , d3 ), showing that a PD(6m + 3, 3, 1; m) exists for all m ≡ 1 (mod 3).

Theorem 2.3 PD(6m + 3, 3, 1; m) exists for m ≡ 0 (mod 3).

Proof: In the following tables, triples (X1 , X2 , X3 ) are presented such that X1 +X2 +X3 = 0.
From each triple (X1 , X2 , X3 ) we simply form the distance triple ((2m + 1) + X1 , (2m +
1) + X2 , (2m + 1) + X3 ). Each of the necessary distances will appear once in the resulting
triples and thus we can use them to form blocks which will generate PD(6m + 3, 3, 1; m).

5
 Case 1: m is odd
X1 X2 X3 X1 X2 X3
−m 3 −( m
3 + 1)
2m
3 + 1 −( m
3 − 1) −( 2m
3 + 1) m
m m 2m m 2m
−( 3 − 2) −( 3 + 2) 3 −( 3 − 3) −( 3 + 2) m−1
−( m m
3 − 4) −( 3 + 3)
2m
3 − 1 −( m
3 − 5) −( 2m
3 + 3) m−2
.. .. .. .. .. ..
. . . . . .
−1 − m+1 2
m+1
2 +1 −2 −( 5m+3
6 − 1) 5m+3
6 +1

m m+1
3 2 − 5m+36
m
3 −1 m
3 +1 − 2m 3
m 2m m m
3 −2 3 +2 −m 3 −3 3 +2 −( 2m
3 − 1)
m 2m m m
3 −4 3 +3 −(m − 1) 3 −5 3 +3 −( 2m
3 − 2)
.. .. .. .. .. ..
. . . . . .
5m+3
1 6 −( 5m+3
6 + 1) 2 m+1
2 −1 −( m+1
2 + 1)

Case 2: m is even
X1 X2 X3 X1 X2 X3
m m 2m
−3 −( 3 + 1) 3 + 1 −( m
3
2m
− 1) −( 3 + 1) m
−( m
3 − 2) −( m
3 + 2)
2m
3 −( m
3 − 3) −( 2m
3 + 2) m−1
−( 3 − 4) −( m
m
3 + 3)
2m
3 − 1 −( m
3 − 5) −( 2m
3 + 3) m−2
.. .. .. .. .. ..
. . . . . .
−2 −m 2
m
2 +2 −1 5m
− 6 5m
6 +1

m m
3 2 +1 −( 5m
6 + 1)
m
3 −1 m
3 +1 − 2m 3
m 2m m m
3 −2 3 +2 −m 3 −3 3 +2 −( 2m
3 − 1)
m 2m m m
3 −4 3 +3 −(m − 1) 3 −5 3 +3 −( 2m
3 − 2)
.. .. .. .. .. ..
. . . . . .
5m
2 6 −1 −( 5m
6 + 1) 1 m
2 −( m
2 + 1)

This completes the proof.

Theorem 2.4 PD(6m + 3, 3, 2; m) exists for m ≡ 2 (mod 3).

Proof: Case 1: m is even

Use Lemma 2.1 to partition the integers 1, 2, ..., (m − 1) and then form two distance
triples as before from each triple in this partition. By repeating this a second time we

6
will have every distance appearing twice in the resulting triples with the exception of the
distances
m m
(2m + 1) − m, (2m + 1) − , (2m + 1) + 0, (2m + 1) + , (2m + 1) + m
2 2
which do not appear at all. Now form the distance triples
m m
((2m + 1) + , (2m + 1) + , (2m + 1) − m),
2 2
m m
((2m + 1) − , (2m + 1) − , (2m + 1) + m), and
2 2
((2m + 1) − m, (2m + 1) + 0, (2m + 1) + m).
This leaves us with every distance appearing twice except for 2m + 1 which appears once.
However, 3(2m + 1) = 6m + 3 and so we can form the rotationally symmetric block that
only contributes a single distance of 2m + 1 to the difference collection. We now have
each of the necessary distances represented twice and so we can form generating blocks for
PD(6m + 3, 3, 2; m) when m ≡ 2(mod 3).
Case 2: m is odd
Observe that by increasing the second and third values of each triple in Lemma 2.1
by one we get a partition of the integers 1, 2, ...., m excluding m+1
3 and m+1
2 into triples
(X1 , X2 , X3 ) such that X1 + X2 = X3 . Using the previous partition of 1, 2, ..., m − 1
excluding m−1 2 and this new partition each one to form distance triples as before, we are
left missing one instance of
m+1 m−1 m+1
(2m + 1) − m, (2m + 1) − , (2m + 1) − , (2m + 1) − ,
2 2 3
m+1 m−1 m+1
(2m + 1) + , (2m + 1) + , (2m + 1) + , (2m + 1) + m
3 2 2
and two instances of 2m + 1. We now form the distance triples
m−1 m+1
((2m + 1) − m, (2m + 1) + , (2m + 1) + ),
2 2
m−1 m+1
((2m + 1) + m, (2m + 1) − , (2m + 1) − ),
2 2
m+1 m−1
((2m + 1) − , (2m + 1) + 0, (2m + 1) + ) and
3 3
((2m + 1) + 0, (2m + 1) + 0, (2m + 1) + 0).
Each distance is now represented exactly twice in this group of valid distance triples and
so we can generate PD(6m + 3, 3, 2; m) for all m ≡ 2(mod 3).

7
Theorem 2.5 PD(6m + 3, 3, 3; m) exists for all m.

Proof: From the following table we can form the distance triples ((2m + 1) + X1 , (2m +
1) + X2 , (2m + 1) + X3 ). The resulting distance triples contain each of the distances
m + 1, m + 2, . . . , 3m + 1 inclusive exactly 3 times and so can be used to construct a
PD(6m + 3, 3, 3; m).
X1 X2 X3
−m 0 m
−(m − 1) 1 m−2
−(m − 2) 2 m−4
.. .. ..
. . .
0 m −m
1 −m m−1
2 −(m − 1) m−3
.. .. ..
. . .
m −1 −(m − 1)

Remark 2.1 A PD(v, k, λ; m) can be formed from PD(v, k, 1; m) by simply repeating each
of the blocks in PD(v, k, 1; m) λ times. Similarly, PD(v, k, λ; m) for λ ≥ 2 can be formed
from a linear combination of the blocks from PD(v, k, 2; m) and PD(v, k, 3; m). We thus
have that PD(6m + 3, 3, λ; m) exists for all combinations of λ and m except for when
m ≡ 2 (mod 3) and λ = 1. Here, a cyclically generated PD(v, k, 1; m) cannot exist for
m ≡ 2 (mod 3).

Next, we conclude the current section by introducing a way to construct polygonal

designs of block size 3 using the λ-fold triple systems discussed in Chapter 2 of [4]. A
λ-fold triple system is a pair (V, B), where V is a finite set and B is a collection of 3-element
subsets of V called triples such that each pair of distinct elements of V belongs to exactly
λ triples of B.

Theorem 2.6 If there exists a cyclically generated λ-fold triple system with N (N ≥ 3)
points, then PD(N (2m + 1), 3, λ; m) exists.

Proof: First take the distance triples (n1 , n2 , n3 ) from the generating blocks for the λ-fold
triple system. Observe that either n1 + n2 + n3 = N or n1 + n2 = n3 must hold. From each
triple where n1 + n2 + n3 = N form the 2m + 1 distance triples of the form
(n1 (2m + 1) + X1 , n2 (2m + 1) + X2 , n3 (2m + 1) + X3 )

8
where the values of X1 , X2 , and X3 are taken from the rows of the following table. Observe
that as

(n1 (2m + 1) + X1 ) + (n2 (2m + 1) + X2 ) + (n3 (2m + 1) + X3 ) = N (2m + 1)

these will be valid distance triples.

X1 X2 X3
−m 0 m
−(m − 1) 1 m−2
−(m − 2) 2 m−4
.. .. ..
. . .
0 m −m
1 −m m−1
2 −(m − 1) m−3
.. .. ..
. . .
m −1 −(m − 1)

Similarly, from the triples where n1 + n2 = n3 form the 2m + 1 distance triples of the
form
(n1 (2m + 1) + Y1 , n2 (2m + 1) + Y2 , n3 (2m + 1) + Y3 )
where the values of Y1 , Y2 , and Y3 are taken from the rows of the following table. These
will also be valid distance triples as

(n1 (2m + 1) + Y1 ) + (n2 (2m + 1) + Y2 ) = (n3 (2m + 1) + Y3 ).

Y1 Y2 Y3
m 0 m
m−1 −1 m−2
m−2 −2 m−4
.. .. ..
. . .
0 −m −m
−1 m m−1
−2 m−1 m−3
.. .. ..
. . .
−m 1 −(m − 1)

9
3 Recursive construction of new designs from old

In this section we present two ways to construct another polygonal designs from given
polygonal designs. Both are iterative construction methods and they produce designs over
different size of point sets.

Theorem 3.1 If a cyclic PD(k(2m+1), k, λ; m) exists then PD((k−1)(2m+1), k−1, λ0 ; m)

exists where λ0 = (k − 1)λ.

Proof: Take a block from the generating set for PD(k(2m+1), k, λ; m) and form the k-tuple
of first distances that appear in the block:

{b2 − b1 , b3 − b2 , ..., v + b1 − bk } = {d1 , d2 , ..., dk }.

Replace all of the k possible pairs of adjacent distances, di and di+1 (or dk and d1 ) by
d0i = di +di+1 −(2m+1) to form the k-tuple {d01 , d02 , . . . , d0k }. Now consider the (k−1)-tuples
of ordered distances that can be obtained from the k-tuple {d01 , d02 , . . . , d0k } by removing a
single distance d0i for each i = 1, 2, . . . , k. (There are k such (k − 1)-tuples.) From each
(k − 1)-tuple, say {r1 , r2 , . . . , rk−1 }, form the block {0, r1 , r1 + r2 , . . . , r1 + r2 + · · · + rk−1 }
of size k − 1 on (k − 1)(2m + 1) points. Repeat this process for each generating block. The
resulting blocks will form a generating set for PD((k − 1)(2m + 1), k − 1, λ0 ; m).
This can be verified by counting the differences in the resulting blocks. First, though,
we use Lemma 1.1. From this lemma we know that when v = k(2m + 1) the first differences
(differences between adjacent points in a block) will be elements of the set

{(2m + 1) − m, (2m + 1) − (m − 1), ..., (2m + 1) + m},

the second differences will be elements of the set

{2(2m + 1) − m, 2(2m + 1) − (m − 1), ..., 2(2m + 1) + m},

and more generally that nth differences will be elements of the set

{n(2m + 1) − m, n(2m + 1) − (m − 1), ..., n(2m + 1) + m}.

Observe that for each time a first difference appeared in a generating block for the original
design, it will appear k − 2 times in the generating blocks for the resulting design. For
each time a second difference appeared in the original generating blocks, it will appear
unchanged k − 3 times in the new generating blocks and will be reduced by 2m + 1 to
the corresponding first difference once. For each time a third difference appeared in the
original design, it will appear unchanged k − 4 times and will be reduced by 2m + 1 to
the corresponding second difference twice. This pattern continues on up to the (k − 1)th

10
differences which will be reduced all k − 2 times by 2m + 1 in the resulting blocks to
the corresponding (k − 2)th difference. The result is that for every time the differences
m + 1, m + 2, ..., (k − 1)(2m + 1) + m appeared in the generating blocks for the original
design, the differences m + 1, m + 2, ..., (k − 2)(2m + 1) + m will appear k − 1 times in
the resulting generating blocks.

Theorem 3.2 If PD(v, k, λ; m) exists then PD(v + (2m + 1), k, λ0 ; m) exists where λ0 =
(v − 2m − 1)λ.

Proof: The new design can be constructed by taking each of the blocks from PD(v, k, λ; m)
with points labeled 0, 1, 2, ..., v − 1 and replacing each instance of v − 1 with v + 2m − 2,
each instance of v − 2 with v + 2m − 5, each instance of v − 3 with v + 2m − 8,..., and
each instance of v − m with v − (m − 1). These new blocks form a generating set for
PD(v + (2m + 1), k, λ0 ; m).
This can be seen by observing that each of the differences m + 1, m + 2, ..., v − (m + 1)
appears vλ times in the difference collection of the blocks of PD(v, k, λ; m). The replacement
described above will increase the difference, d, by 1, 2, ..., 2m each λ times. The distance
will be increased by 2m + 1 exactly λ(d − m) times and will be unchanged the remaining
λ(v − d − m) times. It is straightforward to check that each of the distances m + 1, m +
2, ..., (v + 2m + 1) − (m + 1) appears λ(v − 2m − 1) times in these new blocks. These blocks
thus can be used as a generating set to form PD(v + 2m + 1, k, λ0 ; m).

Remark 3.1 It follows that if there is a PD(v, k, λ; m), then there is a polygonal design
with v + (2m + 1)α points in blocks of size k for any positive α. This iterative method has
been provided by John Stufken [6] as a generalization of the result of Hedayat, Rao and
Stufken [3] for the case of m = 1.

4 Construction of PD((2m + 1)k, k, λ; m) from perfect (k, m)-

grouping

In this section we introduce another technique that may be useful in construction and in
the study of polygonal designs.

Definition 4.1 A perfect (k, m)-grouping is a collection of k-element multi-sets having

elements from {0, 1, ..., m} such that each of {1, 2, . . . , m} appears as the distance between

11
two elements of a multi-set precisely N times and 0 appears as the distance between two
elements of a multi-set no more than N/2 times. The k-element multi-sets that form a
perfect (k, m)-grouping will be called groups.

Example 4.1 (1) The collection {{0, 1, 3}} of the single group {0, 1, 3} is a perfect (3, 3)-
grouping with N = 1, while {{0, 1, 4, 6}} is a perfect (4, 6)-grouping with N = 1.

(2) {{0, 0, 2}, {0, 1, 2}, {0, 1, 2}} is a perfect (3, 2)-grouping with N = 3.

(3) The collections {{0, 0, 1, 1}} and {{0, 0, 1, 2, 2}} are perfect (4, 1)- and (5, 2)-groupings
respectively with N = 4.

(4) {{0, 2, 4}, {0, 1, 4}, {0, 1, 4}, {0, 1, 3}} is a perfect (4, 4)-grouping with N = 3.

Theorem 4.1 PD((2m + 1)k, k, λ; m) exists for some λ if and only if there exists a perfect
(k, m)-grouping.

Proof: Suppose that PD((2m + 1)k, k, λ; m) exists for some λ. According to Lemma 1.1,
given a point bi in an arbitrary block, the next point bi+1 must be an element of the set
{bi + (m + 1), bi + (m + 2) , ..., bi + (3m + 1)}. More generally, point bj must belong to

{bi + (2m + 1)(j − i) − m, bi + (2m + 1)(j − i) − (m − 1), ..., bi + (2m + 1)(j − i) + m}.

Now select the point br in the block such that br − bi − (2m + 1)(r − i) is minimized and
apply Lemma 1.1 again. We thus have that an arbitrary point bj must belong to the set

{br + (2m + 1)(j − r) − m, br + (2m + 1)(j − r) − (m − 1), ..., br + (2m + 1)(j − r) + m}.

However, as the value of br − bi − (2m + 1)(r − i) was minimized by our selection of br , bj

cannot fall in the first half of these values with respect to br and thus bj must belong to
the set

{br + (2m + 1)(j − r), br + (2m + 1)(j − r) + 1, ..., br + (2m + 1)(j − r) + m}.

This allows us to write each point, bj , in the block in the form br + (2m + 1)(j − r) + aj
where aj takes on one of the values 0, 1, ..., m. By writing the points of each block in this
form it follows that if a multi-set {a1 , a2 , ..., ak } is formed for each block in the polygonal
design then by collecting all such multi-sets we can form a perfect (k, m)-grouping.
Conversely, suppose we have a perfect (k, m)-grouping. Take a particular group {a1 , a2 , ..., ak }
and form the (k − 1)! blocks of the form

{f (a1 ), (2m + 1) + f (a2 ), ..., (k − 1)(2m + 1) + f (ak )}

12
where f is taken over all of the (k − 1)! permutations on the group {a1 , a2 , ..., ak } such
that f (a1 ) = a1 . When this is done for each group in the perfect (k, m)-grouping, all of
the necessary differences m + 1, m + 2, ..., v − (m + 1) with possible exception of (k − 1)
multiples of 2m + 1 will each be represented N (k − 2)! times in the difference collections
from these (k − 1)! blocks. All of the differences that are multiples of 2m + 1 can be added
to the difference collections a single time by the symmetric block

{0, 2m + 1, 2(2m + 1), ..., (k − 1)(2m + 1)}.

Thus, we can add some number of instances of this block to the previous blocks to make all
necessary differences m+1, m+2, ..., v−(m+1) be represented in the difference collections
for these blocks precisely N (k − 2)! times. These blocks will thus form a generating set for
PD((2m + 1)k, k, λ; m) where λ = N (k − 2)!.

Theorem 4.2 k(k − 1) ≤ 4(2m + 1) is a necessary condition for the existence of a perfect
(k, m)-grouping.

Proof: Suppose that there exists a perfect (k, m)-grouping where k(k − 1) > 4(2m + 1).
If this is the case, then the average number of times that each of the distances 1, 2, . . . , m
appears per group of the grouping must be greater than 4. We will develop a contradiction
by demonstrating that if the distance m appears on average more than 4 times per group,
then the distance 0 will appear more than half as many times as the distance m. This is
not allowed by the definition of a perfect (k, m)-grouping.
We start by examining the number of times the distance m appears in a given group.
Observe that m can only appear as the distance between the integers 0 and m. We let the
number of instances of 0 in a particular group be denoted x and the number of instances of
m in that group be denoted y. The number of times that m appears as a distance in this
group will then be simply xy. Observe that the distance 0 will appear between two points
in this group a minimum of 21 (x(x − 1) + y(y − 1)) times. We now find for what values of x
and y the distance m will appear in the group at least twice as many times as the distance
0 by solving the inequality
xy ≥ x(x − 1) + y(y − 1),
or equivalently
x + y − xy ≥ (x − y)2 .
It is straightforward to show that the only positive integral solutions (x, y) to this inequality
are (2, 2), (1, 2), (2, 1), and (1, 1). With the exception of (x, y) = (1, 1) all of these solutions
yield strict equality. Thus, if the average number of times the distance m appears per group
is greater than 4, there must be a group in which xy > 4 and thus in which the values of

13
x and y are not solutions to our inequality. This means that the distance m will appear in
this group fewer than twice as many times as the distance 0. As the only type of group in
which the distance m can possibly appear more than twice as many times as the distance
0 has (x, y) = (1, 1), we must have (x − y)2 + xy − x − y blocks of this form to compensate
for the group with xy > 4. The average number of times that the distance m appears per
group in these groups is thus
xy + (x − y)2 + xy − x − y
.
1 + (x − y)2 + xy − x − y
This must still be greater than 4 for it to be possible for the average number of times that
m appears as a distance to be greater than 4. However it is straightforward to show that
the expression is never greater than 4 when x and y are positive integers with xy > 4. This
contradicts the original assumption thus proving that a perfect (k, m)-grouping does not
exist for k(k − 1) > 4(2m + 1).

Example 4.2 This bound of the above result is sharp for small k and m as perfect group-
ings exist for (k, m) = (4, 1) and (5, 2) both of which yield k(k − 1) = 4(2m + 1). As we
have seen in (3) of Example 4.1, each of these groupings consists of a single group of size k
with N = 4, namely {{0, 0, 1, 1}} and {{0, 0, 1, 2, 2}}. We note that the polygonal designs
PD(12, 4, 4; 1) and PD(25, 5, 12; 2) generated by

G = {{0, 3, 7, 10}, {0, 4, 6, 10}, {0, 4, 7, 9}}

and
G = {{0, 5, 11, 17, 22}, {0, 5, 12, 16, 22}, {0, 5, 12, 17, 21}, {0, 6, 10, 17, 22},
{0, 6, 12, 15, 22}, {0, 6, 12, 17, 20}, {0, 7, 10, 16, 22}, {0, 7, 10, 17, 21},
{0, 7, 11, 15, 22}, {0, 7, 11, 17, 20}, {0, 7, 12, 15, 21}, {0, 7, 12, 16, 20}},

respectively are obtained from these perfect groupings. Notice that the examples provided
here uses the method presented in the proof of Theorem 4.1 for converting between perfect
groupings and polygonal designs, but we have removed the redundant generating blocks;
so the number of blocks and λ for each designs have been reduced by a half.

Remark 4.3 While the bound of the previous theorem is sharp for small k and m as we
have just seen in the above example, for larger values of k, this bound for m is not the
3
strongest possible bound. We have a better bound that approaches the relationship m ≥ k48
for large k. The proof of this is given in the appendix.

Remark 4.4 As a consequence of the previous two theorems, PD((2m + 1)k, k, λ; m) does
not exists if k(k −1) > 4(2m+1). That is, for m = 1, there exists no polygonal designs with

14
v = 3k if k ≥ 5 as it was proved in [7]. For m = 2, there exist no polygonal designs with
v = 5k if k ≥ 6. Also it is an immediate consequence of the theorems that PD(9, 3, 1; 1)
and PD(25, 5, 1; 2) are the only symmetric polygonal designs with v = b = (2m + 1)k.

Definition 4.2 A natural perfect (k, m)-grouping is any perfect (k, m)-grouping in which
every group is a set, having no redundant elements; i.e., 0 does not appear as the distance
between any two elements of any one group in the grouping.

For example, the perfect groupings listed in (1) and (4) of Example 4.1 are natural
groupings. now the following theorem shows that if we have a perfect natural grouping, we
can have another perfect grouping.

Theorem 4.3 If there exists a natural perfect (k, m)-grouping then there exists a natural
perfect (k, m0 )-grouping of size m0 for all m0 ≥ m.

Proof: We prove this by demonstrating that a natural perfect (k, m + 1)-grouping can be
formed from a natural perfect (k, m)-grouping. This is done by taking every group in the
natural perfect (k, m)-grouping and forming new groups by adding one to every element of
the group that is greater than or equal to n. Do this for n = 1, 2, ..., m + 1 to form m + 1
new groups for each group in the natural perfect (k, m)-grouping. Each of the distances
1, 2, ..., m will be represented an equal number of times in the resulting groups and 0 will
not appear at all.

For example, the perfect (4, 4)-grouping in (4) of Example 4.1 can be obtained from the
perfect (3, 3)-grouping given in (1) of Example 4.1 in the manner described in the above
proof.

Remark 4.5 For k = 3 or 4 there exist natural perfect (k, m)-groupings formed by a single
k-element set. These are the perfect (3,3)-grouping and (4,6)-grouping illustrated in (1) of
Example 4.1. Unfortunately, there are no natural perfect (k, m)-groupings consisting of a
single (multi-)set for any other combinations of k and m. Natural perfect groupings with
a single (multi-)set are very difficult to construct and as a result we have not found any
natural perfect (k, m)-groupings for k ≥ 5.

Remark 4.6 As we have found a natural perfect (3, 3)-grouping and (4, 6)-grouping, it thus
follows from Theorems 4.1 and 4.2 that PD((2m+1)k, k, λ; m) exists for some λ when k = 3
and m ≥ 3 or k = 4 and m ≥ 6. Additionally, for k = 3, m = 1, 2 and k = 4, m = 1, 2, ..., 5
we have manually confirmed the existence of such designs. Thus, PD((2m + 1)3, 3, λ; m)
and PD((2m + 1)4, 4, λ; m) exist for some λ given any m.

15
 Acknowledgment

This research was performed by the first two authors when they visited Iowa State Uni-
versity in 2004. The authors would like to thank the Department of Mathematics at Iowa
State University for its support of the authors during the period that this research was
performed.

References

[1] Colbourn, C. J. and Ling, A. C.(1998) A class of partial triple systems with applications
in survey sampling. Commun. Statist. – Theory Meth., 27(1), 1009-1018.

[2] Colbourn, C. J. and Ling, A. C.(1999) Balanced sampling plans with block size four
excluding contiguous units. Australasian J. of Combin., 20, 37-46.

[3] Hedayat, A. S., Rao, C. R., and Stufken, J. (1988), A sampling plan excluding con-
tiguous units., J. Stat. Plann. Infer. 19, 159-170.

[4] Lindner C. C. and Rodger, C. A. (1997) Design Theory. CRC Press, Boca Raton,
Florida.

[5] See, K., Song, S. Y., and Stufken, J. (1997). On a class of partially balanced incomplete
block designs with applications in survey sampling. Commun. Statist. – Theory Meth.,
26(1), 1-13.

[6] Stufken, J. (1993) Combinatorial and statistical aspects of sampling plans toavoid the
selection of adjacent units. J. Combin. Inform. System Sci. 18, 81-92.

[7] Stufken, J., Song, S. Y., See, K, and Driessel, K. (1999) Polygonal designs: Some
existence and non-existence results. J. Stat. Plann. Infer. 77, 155-166.

[8] Stufken, J. and Wright, J. (2001) Polygonal designs with blocks of size K ≤ 10. Metrika
54, 179-184.

16
Appendix
Theorem For given k, let the minimum value of m for which a perfect (k, m)-grouping
exists be denoted mk . Then lim mk
k3
1
≥ 48 .
k→∞

Proof: We shall prove the statement only for the case when m ≡ 3(mod 4), as m belongs
to the other congruence classes modulo 4 can be treated similarly. Suppose there exists a
perfect (k, m)-grouping that consists of R groups of the form a1 , a2 , . . . , ak with a1 ≤ a2 ≤
· · · ≤ ak .
Observe that the sum of the distances between points in a group a1 , a2 , . . . , ak will be

(k − 1)(a1 + ak−1 ) + (2k − 4)(a2 + ak−2 ) + · · · + (nk − n2 )(an + ak−n ) + · · ·

 !
k−1 2 
  
k−1 
+ k− a( k−1 ) + a( k+1 )
2 2 2 2

where n = 1, 2, . . . , k−1
2 . We can rewrite this as

k−1
k−1
2 ! X !
2
k−1 2
      
k−1 k−1 X k−1
k− ai − k− − (nk − n2 ) (an + ak−n ).
2 2 2 2
i=1 n=1

k−1
P k−1
P
Summing up this over all groups and replacing ai by m (as ai ≤ m), we obtain
i=1 i=1

k−1
2 ! !
2
k−1 2
      
k−1 k−1 X X k−1 2
k− Rm − k− − (nk − n ) (an + ak−n )
2 2 groups 2 2
n=1

for an upper bound on the sum of all distances in all groups. Observe that when this
expression achieves its maximum value it must be the case that an , ak−n ≥ [ n+1
2 ] for n =
1, 2, . . . , k−1
2 . We thus have another upper bound RU for the sum where
k−1
2 ! !
2
k−1 2
      
k−1 k−1 X k−1 2
hni
U := k− m − k− − (nk − n ) · 2 .
2 2 2 2 2
n=1

Note that U is an upper bound for the average sum of the distances per group. This must
be greater than or equal to the number of distances per group multiplied by the minimum
average distance which is k(k−1)
2 · m(m+1)
2m+1 . We thus have the inequality

k(k − 1) m(m + 1)
U≥ · .
2 2m + 1

17
Multiplying both sides of this inequality by 2m + 1, expanding, simplifying, dividing by k 3
and taking the limit as k approaches infinity yields
mk 1
lim 3
≥ .
k→∞ k 48

18