Mathematics Competitions Vol 27 No 2 2014

28,800 E x tre m e ly M agic 5 x 5 S quares

A rthur Holshouser & Harold Reiter

Arthur Holshouser has had a lifelong

passion for doing mathematics. He
works on problems at least 5 hours each
day and often much more. His favorites
are plane geometry, polynomial alge­
bra, combinatorial games, and num­
ber theory. His current passion is Bro-
card geometry and Poincelet polygons.
His partnership with Harold Reiter has
produced more than 50 articles.

As Harold Reiter completes his 51st

year of teaching, his passion for light­
ing mathematical Gres has continued to
grow. In the past few years, he has
enjoyed working with Indian children
in Bangalore, Indonesian children and
teachers, Navajo children, Saudi teach­
ers and very young gifted students at
Epsilon Camp in the United States. He
spends time creating challenges includ­
ing puzzles like KenKen (with modulo
6 or 7 arithmetic and with prime num­
ber entries). He continues to work with
longtime coauthor Arthur Holshouser,
on papers in number theory, geometry,
and polynomial algebra.

1 Abstract
Using an algorithm from the paper “Creating Semi-Magic, Magic and
Extra Magic n x n Squares when n is Odd”, we define 28,800 extra
magic (or panmagic) 5 x 5 squares. We show that these 5 x 5 extra
magic squares are also extremely magic. This means that all of these

50
 Mathematics Competitions Vol 27 No 2 2014

28,800 extra magic 5 x 5 squares have the same 120 magic 5 element
subsets whose sum is 65. Of course, these subsets include the 5 rows and
5 columns and the 10 generalized diagonals. We also discuss additional
magic properties. We also define 28,800 similarity mappings on these
5 x 5 extremely magic squares that map these 120 magic 5 element
subsets onto each other. As always, these 28,800 similarity mappings
form a group. If we use any one of the 5 x 5 extremely magic squares
as the domain and use any other as the range then the function that
we create will be one of these 28,800 similarity mappings. Thus, we
can loosely say that the 28,800 magic squares themselves form a group.
However, the graph of a group is a more accurate term.
At the end, we state a variation of Hall’s Marriage Theorem that can
greatly enhance this paper. Much of this paper can be generalized, but
the 5 x 5 magic squares have properties that are unique.

2 In tr o d u ctio n
Using an algorithm developed in the paper “Creating Semi-Magic, Magic
and Extra Magic n x n Squares when n is Odd”, we define the following
5 x 5 Matrix. All the results of this paper are derived from this matrix.
Throughout this paper, we use the term line to mean any of the 20 rows,
columns, and generalized diagonals.

(ao, A)) (ai, A]) (a2, A2) (a.3, A3) A4 )

(tt4 ,
(a3, A2) (a4, A3) (ao, A4) (a i, A)) (a2, Ax)
(ai, A4) (a2, A0) (a3, Ai) (o4, A2) (ao, A3)
(a4, A{) (a-o, A2) (a i> a 3) (a2, A4) (a3, A0)
(a2, A3) (a3, A4) (a4, A0) (ao, Ax) (ai, A2)

Figure 1: A 5 x 5 Matrix

Note that this matrix contains all the ordered pairs (aj,A 7), i =
0 ,1 ,2 ,..., 4, j = 0 ,1 ,2 ,..., 4. We can also define the dual 5 x 5 matrix
where we change each (a i,A j) in Fig. 1 to (a^A;). Thus, the second
row of the dual matrix would read (a2, A3), (a3, A4), (a4, A0), (a0, A x),
(ai,A 2). In Fig. 1 we agree that a0, a4, a2, a3, a4 is any arbitrary but
fixed permutation of the integers 0,1,2,3,4. Also, A0, A u A2, A3, A 4 is
any arbitrary but fixed permutation of 0,1,2,3,4.

51
 Mathematics Competitions Vol 27 No 2 2014

In Fig. 1, if we permute ao, ax, 02,03, 0 ,4 in all possible ways and permute
A0, A i, A2, A3, A4 in all possible ways we can define 5! • 5! = 14,400
different 5 x 5 matrices.

For each ordered pair (a*, A,) in Fig. 1, let us assign the number
(a , i , A j = a* + 5 (Aj — 1) where a*, Aj are the numerical values that
have been assigned to at,Aj.

Now 1 < ai < (Oj, A j )# = at + 5 (Aj —1) < 5 + 5 (5 —1) = 52. That is,
1 < (Oi,Aj)* < 25.

We now show that (a*, A,-)* = (a k ,A e)* if and only if ai = ak and

Aj — Ae.

First, suppose Aj ^ Ae and by symmetry suppose Aj < A e. We

show that (ai,A j)# < (aj,,Ae)# . That is, we show (ai, A j)* = a* +
5 (Aj — 1) < aj, + 5 (Ae —1) = (oj,, A e)# . This is equivalent to —ak <
5(A e-A j).
Since a i ,A j, a k ,A e € {0 ,1 ,2 ,3 ,4} and Aj < Ae we see that at - ak <
5 < 5 (A e — Aj). Therefore, ai — ak < 5 (Ae — Aj).

If Aj = A e then obviously (a*, A ,-)* = (ak,Ae)# if and only if ai = ak-

Therefore, (ai, A j = (ak, A e)* if and only if ai = ak and Aj = Ae.

From this we see that

{(ai, Aj ) * :a i,Aj £ {0,1,2,3,4}} = {1,2,3,4, ■• ■,25}

since the matrix contains all the ordered pairs (a*, A,), i = 0 ,1 ,2 ,..., 4,
j = 0 ,1 ,2 ,. .. , 4. Thus, we can place 1,2,3, • • • ,25 in the 5 x 5 matrix.
Let us now observe the following about Fig. 1, namely that each row
contains each of an, a%, a2, a3, <14 and contains each of An Aj, A 2, A n A 4 .
Also, each column contains each of an, ai, a2, a3, 04 and contains each
of A), Ai, A2, A3, A4. Also, each of the two main diagonals and each of
the 8 generalized diagonals contains each of ao, ai, 02,03 , 0 4 and contains
each of A0, Aj, A 2, A3, A4.
Note that there axe 10 generalized diagonals in Fig. 1 when we count
the two main diagonals. One example of a generalized diagonal is the

52
 M athem atics C om petitions Vol 27 N o 2 2014

set {(ai, A-]) , (a0, A4) , (a4, A 2) , (a3, A0) , (a2,A 3)}. Another example
is {(oi, A \ ) , (a3, A 2) , (a0, A3) , (a2, A 4) , (<24, A0)}. If we use the code
n
(a,, A j ) w = a, + 5 (Aj - 1) we now show th a t the sum of the 5 numbers
in each row, in each column and in each generalized diagonal equals to
| (1 + 2 4-------1-25) = | | (26) = 65. This means th a t the 5 x 5 M atrix
of Fig. 1 becomes an extram agic (or panmagic) 5 x 5 square.

For example, the sum of the 5 numbers in the 1st main diagonal equals

(ao,Ao)^ + (a4, A 3)* + (o3, A i)^ + (<22, A4)^ + (ai,j4.2)^

= [ao + 5 (Ao —1)] + • • ■+ [ai + 5 (A2 —1)]
4 4

= ^ 2 ai + 5 ^ 2 Ai — 5-5
2=0 i—0
5 4

= ^ 2 ^ + -5-5
i=i i= 0

= 15 + 5 - 1 5 - 2 5
= 65,

since {a0, a i, a 2, a 3, a4} = { A o , A 1, A 2, A 3, A 4} = {0,1 ,2 ,3 ,4 }. This

reasoning is the same for all lines.

3 A S p ecific E x a m p le
In Fig. 1, we now let (00,04,0:2,03,04) = (3 ,5 ,1 ,2 ,4 ) and (A0,A i,
A2, A3, A4) = (2 ,1 ,5 ,4 ,3 ). Using the code (a*, A j ) * = o* + 5 (Aj - 1)
we have the extra magic 5 x 5 square of Fig. 2.

8 5 21 17 14
22 19 13 10 1
15 6 2 24 18
4 23 20 11 7
16 12 9 3 25

Figure 2: An E xtra Magic 5 x 5 square

53
 Mathematics Competitions Vol 27 No 2 2014

The magic 5x5 square of Fig. 2 is actually extremely magic and we show
later how to use this extremely magic 5 x 5 square to generate 28,800
different, extremely magic 5 x 5 squares.
Note 1. Suppose that instead of using the integers 1 though 25, we are
interested in putting any numbers a\, a?, . . . , 025 in the 25 positions of
the 5 x 5 matrix so that the sum of the 5 numbers in each row, column,
and generalized diagonal is the same. This leads to a system of linear
equations in the variables ax, 02, , a 25- Professor Ben Klein has shown
that the rank of this system of linear equations is 16. Thus, it is possible
to find a subset of nine letters a ^ , aj2, . . . , aj9, and arbitrarily specify the
values of these nine letters. Then we can compute the unique values of
the other 16 letters in terms of the 9 given letters. In the 5 x 5 matrix of
Fig. 1, let us now use the code = a, + Aj and then place these
values (a,i,Aj)& in the 5 x 5 matrix. We note from Fig. 1 that the sum
of the 5 numbers on each line is always the same value
4 4

y a‘+y
2— 0 2— 0
A*-

Now (ai,Aj)# = a* + Aj = (aj + x, Aj —x )#. Therefore, if we replace

each aj by aj + x and replace each Aj by Aj —x, we see that (a*, Aj)# =
on + Aj = (a* + x, Aj —x )*. Thus, there is no loss in generality to
always let ao = 0. Therefore there are nine degrees of freedom among
the ten letters ao, ax,. . . , 04, Ao, A i , . . . , A4, since ao = 0. Given that
the rank of the 5 x 5 matrix in Fig. 1 is 16, it is straightforward to
show that the 5 x 5 matrix of Fig. 1 with the code (aj, Aj)# = aj + Aj
where ao = 0 and where a 1( 02, 03, a4, A0, . . . , A4 vary over the set of
numbers, will give all of the 5 x 5 matrices with a common sum for
all lines. The panmagic square below is one example of the attractive
mathematics we can create from this theory. If we add A to each of the
25 squares, we will have 9 independent variables and we can easily find
ao = 0 , ax, a2, . . . , a4, A0, A i , . . . , A4 that will realize these 9 independent
variables ,______ .______ .______ ,______ ,______ ,
0 B+I C+H D+G E+F
D+H E+G F B C+I
B+F C D+I E+H G
E+I H B+G C+F D
C+G D+F E I B+H
54
 Mathematics Competitions Vol 27 No 2 2014

is an extra magic 5 x 5 square, where B, C, D, E, F, G, II, I are arbitrary,

and the common sum is B + C + D + E + F + G + H + I.

4 T h e E x tr em e M a g icn ess o f F igu res 1 and 2

In this section, we derive the extreme magicness of Fig. 1 and 2. From
Note 1 it will then be self-evident that any panmagic 5 x 5 square will
also be extremely magic. This is stated in [3], but it is not derived in
[3]. Also the proof in [3] is far from self-evident.

In the rest of this paper, we assume that (a0, ai, a2, 03, o4) and (A0, A\,
A2,A 3,A 4) have been arbitrarily fixed by Fig. 2. In Fig. 1, let us
now permute A0, A\, A 2, A3, A4 in all possible ways to create 120 dif­
ferent 5 element sets of ordered pairs {(a0, Ai0) , (ai, A n ) , (a2, A i2) ,
(<23, Ai3) , (a4, An)}.

Thus, one example would be the 5 element set {(a0, A i) , (ai, A2) ,
(a2, A4) , (a3, A0) , (a4, A3)}. These 120 5-element sets of ordered pairs
will define 120 different 5 element subsets of the Fig. 1 matrix. When
we use the code (a,, A j)w = ai + 5 (Aj - 1), each of these 120 differ­
ent 5 element subsets of the Fig. 1 matrix will be a magic 5 element
subset of the Fig. 1 matrix since the sum of the 5 elements in the set
will be equal to 65. This follows since each of these 120 different 5 el­
ement subsets of Fig. 1 will contain each of the letters ag, a4, a2, 0.3, a4
and contain each of the letters A0, A\, A2, A3, A4 and since the code is
(<2i, Aj)* = ai + 5 (Aj —1 ) .

Also, of course, {a0, a4, a2, a3, a4} = {0,1,2,3,4} and {A0, A 1,A 2,A 3,
A 4 } = (0 ,1 ,2 ,3 ,4 } .

After we go through all of the 120 sets {(a0, Ai0) , (a4, AiX) , (a2, Ai2) ,
(a3, A i3) , (a4, Ai4)} we have the classification of the 120 magic 5 element
subsets of Fig. 1 that is shown below. In these drawings, we also state
the number of times that each magic 5 element configuration appears in
Fig. 1. We note that there are 16 different configurations th at are listed
below and most of these 16 configurations can appear many times in the
5 x 5 matrix of Fig. 1. For example, the configuration of drawing G can
appear 16 times in the 5 x 5 matrix of Fig. 1.

55
Mathematics Competitions Vol 27 No 2 2014

56
 Mathematics Competitions Vol 27 No 2 2014

We note that the sum of the numbers in these 16 drawings equals 120
which means that these configurations appear 120 times in Figs. 1, 2.
We also note that every time that one of the 16 configurations appears
in Fig. 1 this set of 5 elements will contain all of ao, ai, a2, a3, a4 and
contain all of A0, A \, A2, As, A4. These 120 subsets of Fig. 1 are also the
only 5 elements subsets of Fig. 1 that contain all of ao, ai, a2, a3 , a4 and
contain all of Ao, Ai, A2, A3, A4. Also, these 120 subsets of Fig. 1 are
the only five-element subsets of Fig. 1 such that each pair of points in
the set lie on a line.
We note that drawing A represents the horizontal rows and vertical
columns for a total of 10. Also, drawing N represents the two main
diagonals for a total of 2. Also, drawings O, P together represent the 8
generalized diagonals for a total of 8.
Let us now focus our attention on drawing E. This configuration appears
4 times in Figs. 1, 2. These are
{(ao, A0) , (ai, A i), (a3, A2) , (a4, A4) , (a2, A3)} = {8,5,22,14,16} ,
{(a2, A3) , (a4, A i), (a3, A4) , (a0, A0) , (a4, A2)j = {16,4,12,8, 25} ,
{(°i> A2) , (ao, A i), (a3, A0) , (a2, A3) , (a4, A4)} = {25,3,7,16,14} ,
{(a4, A4) , (a2, A i), (a3, A3) , (aj, A2) , (ao, Ao)} = {14,1,17,25,8} .
Note 1. Each of these four 5 element subsets contain all of ao, a 4, a2, a3,
a4 and contain all of Ao, A4, A2, A3, A4 and the sum of the 5 numbers
in each of these four 5 element subsets is always 65. The sum of the 5
elements in each of the 120 magic subsets will always equal 65.
In Section 10, we state that a variation of Hall’s Marriage Theorem can
be used to show that these 5 x 5 extremely magic squares are actually
much more magic than what we have studied in this section. However,
we consider Hall’s Marriage Theorem to be outside of the scope of this
paper. So in Section 10 we state these results without proof. In Section
11, we state a variation of Hall’s Marriage Theorem.

57
 Mathematics Competitions Vol 27 No 2 2014

Note 2. There is a large number of other properties embedded in the

panmagic 5 x 5 square of Fig. 1 when we use the code (a i , A j )# =
a*+ 5 (Aj —1) or (a,, A j ) * = a,+A,-. Consider for example, the two sets
of doubleton entries {(oi, A2), (<23, A4)} and {(ai, A4), (03, A2)}. Now
(■a.i, A 2 + (a3, A i)^ = (ai, + (<13,^2)^ = (ai + 0,3) + (A2 + A4).
Therefore, in any panmagic 5 x 5 square, these two doubleton sums must
always be equal. Thus, in Fig.2 we see that 12 + 25 = 22 + 15 = 37. In
Fig. 1, consider the 5 element sets {(ai, A2), (ao, -A), (a2, A3), (01, A0),
(a3,A i)} and {(au A4), (a0, A 2), (a2, A x), (ai, A 3 ) , (a3, A0)}. It is obvi­
ous that the sum of the 5 elements in each of these two sets is the same.
Thus, in Fig. 2, we see that 25+3+16+10+12 = 15+ 23+ 1+ 20+ 7 = 66.
Reference [3] does not mention these properties.

5 S im ila rity M a p p in g s on th e F ig . 1 M a tr ix
Remember that 00,01,02,03,04 and A),A , A2, An Ai can be any ar­
bitrary permutations of the integers 0,1,2,3,4. In this paper we have
agreed that 00,01,02,03,04 and A0,Ai , A2,A 3,A 4 have been fixed by
Fig. 2.

In the Fig. 1 matrix, we note that 00,01,02,03,04 can be permuted

in 5! = 120 different ways and A0, A , A2, A3,A 4 can be permuted in
5! = 120 different ways. Thus, we can use Fig. 1 to define 120 • 120 =
14,400 different arrangements (or permutations) of the 25 ordered pairs
{(a i ,A j ) : i , j € {0,1,2,3,4}} .

Each of these arrangements (or permutations) will define an extremely

magic 5 x 5 square and each of these magic 5 x 5 squares will have the
common 120 magic 5 element subsets that are listed in Section 4.

We observe that the arrangement of the five (a*, A'Vs in the t°P row of
Fig. 1 will completely determine the arrangement (or permutation) of
all 25 (oi, AjYs in Fig. 1. For example, if the top row of Fig. 1 reads
( a i, A0) , (a3, A t) , (ao, A ) , («4, A ) , (a2, A ) we know that we are using
the permutations and A0,A, A, A, A
These
^ A0, A4, A3, A i , A2
15 ^3 5 CLO 5 ^4 > ^2
permutations determine all 25 entries in the Fig. 1 matrix and these 25
entries are shown in Fig. 3.

58
 Mathematics Competitions Vol 27 No 2 2014

(a i, A0) (a3, A 4) (a 0 , A3) (a4, A j) (0 2 , A2)

(<24, A 3) (0 2 , A i) (a i, A 2 ) (03, A0) (ao, A4)
(a3, A 2) (ao, Ao) (0 4 , A 4) (fl2 , A3) (® ii A i )
(a2, A 4 ) (a i, A3) (a3, A i ) (a0, A2) (a4, A0)
( o q , Ai) (a4, A 2) (a2, A0) (a i, A4) (a3, A3)

Figure 3: A Similarity Mapping of Fig. 1

These 14,400 perm utations form a group of similarity mappings on the

5 x 5 magic square of Fig. 1 and we explain this in a moment. The
group structure of the 14,400 perm utations is S 5 x S 5 where S 5 is the
symmetric group on {0,1,2,3,4}. We go deeper into this in Section 6.
We will also soon show why these 14,400 perm utations of Fig. 1 are
sim ilarity mappings on the magic squares.

Suppose, x 3, x 2, x 3, x 4, Xr> and X\, x 2, S 3 , £ 4 , £ 5 are any arbitrary magic

5 element subsets of the Fig. 1 m atrix whose elements X\ , x2, x 3, X4, x3
and x i , X 2 , x z ,X 4 ,X 5 are listed in any arbitrary order. It is easy to see
th at we can choose perm utations
ao, d\, a 2, a 3, o4 A0, Ax, A 2, A3, A 4
a,;01 O n , d i2 , d i 3, O il Ajo, A j i , A j 2, A j 3, A j 4

so th a t the corresponding similarity m apping / will m ap f ( { x i , x 2, x 3,

x4, x 5}) = { xi , X2, x3, X4, x5} with / (Xi) = Xi for all i = 0 ,1 ,2 ,3 ,4 . We
also note th a t th e perm utations
ao, a i, <22 , a 3 , 04 Ao, A i, A 2 , A-j, A 4
_04,a2,ao,a3,a i A 4 , A i, A3, A 0, A 2
will define a sim ilarity m apping / th a t rotates the Fig. 1 m atrix 90°
count erclo ckwise.
a 0, a i, 0 2 , a 3, a4
We can also show th at there are no perm utations
^iOt &,i2’i 0*131
Ao, A j, A 2, A 3, A 4
and
_Ajo,A j 1 , A j 2, A 3 ,A j 4_ th at
j
will flip the m atrix of Fig. 1 over.
The dual m atrix of Fig. 1 is needed to flip Fig. 1 over.

In Fig. 4 we have num bered the squares of Fig. 1 1,2,3, • • • , 25. Thus,
1 = (a0,A 0) ,2 = ( a i,A i ) ,3 = (o2,A 2), etc.

59
 Mathematics Competitions Vol 27 No 2 2014

1 2 3 4 5
6 7 8 9 10
11 12 13 14 15
16 17 18 19 20
21 22 23 24 25

Figure 4: Numbering the 25 squares

Of course, each of the 14,400 permutations that we have defined will

define a permutation on the Fig. 4 matrix. Each permutation will be
a similarity mapping on the 5 x 5 matrix in the sense that it maps the
120 magic 5 element subsets of the matrix onto each other. This fact
by itself would imply that each mapping of Figs. 1, 2, 4 would produce
an extremely magic 5 x 5 square from any extremely magic 5 x 5 square
and all such magic squares would have the same 120 magic 5 element
subsets. The similarity mappings on any structure always form a group.
From Note 1 it is obvious that any panmagic 5x5 square is also extremely
magic. Thus, these 14400 permutations would produce a panmagic 5 x 5
square from any panmagic 5 x 5 square. Reference [3] does not mention
these similarity mappings.

6 A B asis for th e Sim ilarity M appings

As stated in Section 5, the arrangement of the five (a,, Aj)’s in the top
row of Fig. 1 will completely determine the arrangement (or permu­
tation) of all 25(a;,A j)’s in Fig. 1. Now the arrangement of the five
(a*, Aj)’s in the top row of Fig. 1 will be determined by the two permu­
do, d i , d2, d3, d4 A0, Ai, A 2 , A3, A4
tations. and
djo, di 1 , dj2 , dj3, dv4 A j o, A j i , A j 2 , A j s , A j 4

A basis for the 14,400 member group of similarity mappings that we

are dealing with would be / o i ,/ 1 2 , / 2 3 , / 3 4 and £0 1 , 5 1 2 , 5 2 3 , 5 3 4 where
/ i j i + 1 is the permutation of Figs. 1, 4 that is defined by interchanging
dj,dj-|-i and 5 1 ,1 + 1 is the permutation of Figs. 1, 4 that is defined by
interchanging A , A,+ 1 .
We now list these basic permutations.

60
 M athem atics C om petitions Vol 27 No 2 2014

9 24 3 4 5 1 10 25 4 5
6 7 11 1 10 6 7 8 12 2
8 12 13 14 18 19 9 13 14 15
16 25 15 19 20 16 17 21 11 20
21 22 23 2 17 18 22 23 24 3

/oil O.Q — di / i 2 i a i —a 2

1 2 6 21 5 1 2 3 7 22
3 7 8 9 13 14 4 8 9 10
11 20 10 14 15 11 12 16 6 15
16 17 18 22 12 13 17 18 19 23
4 19 23 24 25 21 5 20 24 25

/ 2 3 i 0-2 ~ 0-3 /34, a 3 —a 4

24 9 3 4 5 1 25 10 4 5
6 7 8 2 12 13 7 8 9 3
11 10 20 14 15 11 12 6 16 15
23 17 18 19 13 14 24 18 19 20
21 22 16 1 25 21 22 23 17 2

goi, Ao - A\ 912, — A2

1 2 21 6 5 1 2 3 22 7
4 14 8 9 10 6 5 15 9 10
11 12 13 7 17 18 12 13 14 8
16 15 25 19 20 16 17 11 21 20
3 22 23 24 18 19 4 23 24 25

9 2 3 , A-2 — A 3 934, A3 — A4

1 2 3 4 5
21 22 23 24 25
16 17 18 19 20
11 12 13 14 15
6 7 8 9 10

61
 M athem atics C om petitions Vol 27 No 2 2014

We discuss the permutation h in a moment. It is obvious that /? i+1 =

fi,i +1 0 fi,i+i = I and (_i = 0 9i,i+1 ^i ^ where I is
the identity permutation.
Also, f i ,7+i o g j j+i = 9 j,j+ i ° j 6 {0,1,2,3} .
This is obvious from the definitions of fi:i+i and gj,j+i and this can
also easily be verified formally by direct calculation. These basic per­
mutations /o i ,/ i 2 , / 23 , / 34 ,5 oi,S,i 2 , 5 2 3 ,ff34 will generate all of the 14,400
permutations that we defined in Section 5.
If we start with the extremely magic square of Fig. 2 and operate on it
in any way by using the above basic permutations f i ,i+ i and g j j + i then
the image will always be an extremely magic 5 x 5 square and we can
generate from Fig. 2 14,400 extremely magic 5 x 5 squares by doing this.
For example,
' 8 5 21 17 14' ' 8 5 21 19 12'
22 19 13 10 1 24 17 13 10 1
15 6 2 24 18 _ 15 6 4 22 18
4 23 20 11 7 2 23 20 11 9
16 12 9 3 25 16 14 7 3 25
and this 5 x 5 matrix is an extremely magic 5 x 5 square since the sum
of the 5 numbers in each of the 120 magic, subsets stated in Section 4
is always 65. Of course, this includes the 10 rows and columns and 10
diagonals. We now deal with the permutation h. In Section 2, we defined
the dual matrix of Fig. 1 as the matrix where each (cti,Aj ) is changed
to (ctj,Ai ). This dual matrix will lead to 14,400 extremely magic 5 x 5
squares in exactly the same way as the original matrix. It can be shown
that these 14,400 dual magic squares can be obtained by flipping over
the original 14,400 magic squares.
The permutation h is the permutation that defines the dual matrix of
Fig. 1. That is,
h(Fig. 1) = dual matrix.
It is obvious that h2 = h o h = I.
By the symmetry between the a,’s and At’s, it is fairly easy to see
that the extremely magic 5 x 5 squares that are generated by the dual

62
 Mathematics Competitions Vol 27 No 2 2014

matrix will have the same 120 magic 5 element subsets as the origi­
nal matrix. Also, the 14,400 similarity mappings of Section 5 for the
original magic squares will be exactly the same as the 14,400 similarity
mappings for the dual magic squares. Therefore, we can use the same
basis /oi,/i2>/23,/34,5oi,512,523,534 for the similarity mappings of the
dual magic squares. By direct computation, it is very easy to show that
/*,»+1 ° h = ho gi<i+1 and 5i,i+i o h = ho / M+1 for i = 0, 1, 2,3.

Therefore, the number of permutations generated by / 0i, / i 2, / 23, / 34,

5oi, 5 i 2, 523, 534, h will be 2-14,400 = 28,800 since we also know that we
cannot flip Fig. 1 over if we do not use h. The group structure of these
28,800 permutations will be a dihedral type group that is generated from
/o i>/ 12, /23, / 34, 5oi, 5 i 2,523, 934,h subject to the restrictions of following
laws.
f • f i , i + 1° f i , i + 1 = 9 j , j + 1 ° 9 j , j + l — h O h = I .
2- f i , i + 1° 9 j , j + 1 = 9 j , j + 1 ° f i , i + 1-

f i , i - f-1 O ft — h O 1
4- 5 i,i+ i °h = ho / i>i+1.

Since g (/oi, / 12, /23, /34) — S5 and 5 (501, 512, 523, 534) — S5 where
S5 is the symmetric group on {0,1, 2,3,4}, we see from law 2 that
5 (/oi,/i2,/23,/34,501,512,523,534) — S5 x S5. This fact was also ob­
served in Section 5. When we add h to /oi, / 12, / 23, / 34, 5oi, 512, 523,534
we see that we generate a dihedral type group that has 28,800 permu­
tations.

7 A P u zz le
Suppose we start with the 5 x 5 extremely magic square of Fig. 2.
However, suppose that we are not aware that Fig. 2 is extremely
magic, but we only know that Fig. 2 is extra magic (or panmagic).
That is, we are only aware of the fact that the sums of the 5 num­
bers in each of the 5 rows, 5 columns and 10 generalized diagonals
equal 65. Also, suppose that we are given the 8 basic permutations
/oi,/i2,/23,/34,501,512,523,534 of Section 6 . If we operate on the
Fig. 2 magic square in any arbitrary way by using these permutations
/oi, / 12, • • • , 534, we will always have an extra magic 5 x 5 square, and we

63
 Mathematics Competitions Vol 27 No 2 2014

could generate thousands of these extra magic 5 x 5 squares. However,

we would not have the slightest idea why this is true since we are not
aware of the big picture concerning extremely magic squares. Since any
panmagic 5 x 5 square is also extremely magic, this puzzle would also
work with any panmagic 5 x 5 square.

8 A G e n era liza tio n o f th e F ig. 1 M a tr ix

In Fig. 1, suppose we now place the 25 ordered pairs
{(ai,Aj) : i , j 6 {0 , 1, 2,3,4}}

in any arbitrary way in the 5x5 matrix. As always, we let ao, a i, 02, 03, <24
and A0, Ai, A2, A3, A be any arbitrary permutation of 0 , 1,2,3,4. As
4

always, we use the code (a*, Aj')* = Oj + 5 (Aj —1) to place the numbers
1,2,3,4, • • • , 25 in the 5x5 matrix. Analogous to this paper, we can deal
with all permutations of ao, ai, a2, a3, and Ao, A \ , A 2, A3, A4 which we
0 ,4

ao ai a a 3 04
2 Ao Ai A2 A3 A4
call and
1 Qj,i 2 i4 Aj 0Aj 1Aj 2Aj 3Aj 4
As in this paper, we can find 120 magic 5 element subsets whose sum is
always 65 and all 14,400 5 x 5 matrixes will have these same 120 magic
5 element subsets. We can also define the dual matrix and we leave it
to the reader to study whether the entire theory in this paper holds in
general for the dual matrix.

9 A P r o je c t for th e R ead er
We invite the reader to use the permutations
a o a ia 2 0 3 a 4 A0A1A2A3A 4
0004030204 A0A3A1A4A2

and observe that the permutation / that we define maps rows and
columns of the Fig. 1 matrix onto generalized diagonals and maps gen­
eralized diagonals onto rows and columns. We then invite the reader to
find other such permutations / of the Fig. 1 matrix that map the 20
rows, columns and generalized diagonals onto the 20 rows, columns and
generalized diagonals in various ways.

64
 Mathematics Competitions Vol 27 No 2 2014

10 A n A p p lica tio n o f th e M arriage T h eo rem

The extremely magic 5 x 5 squares that we have studied are actually
much more magic than what we have explained in this paper. We need
the marriage theorems to explain the total magicness of these 5 x 5
squares. We state a variation in Section 11.
Suppose that we choose any one of the 16 configurations of Section 4 (call
it a) and place a on the 5 x 5 extremely magic square in any arbitrary
way. Of course, the sum of the 5 numbers in the configuration a is 65.
No matter how we choose a and place a on the 5 x 5 square we can then
choose one of the 16 configurations (call it b) and place b on the 5 x 5
extremely magic square so that a and b are disjoint. Of course, the sum
of the 5 numbers in configuration b is also 65. No matter how a and b
are chosen and placed on the 5 x 5 magic square, we can then choose one
of the 16 configurations (call it c) and place c on the 5 x 5 magic square
so that c is pairwise disjoint from each of a and b. Of course, the sum
of the 5 numbers in configuration c is also 65. No matter how a, b and
c are chosen and placed on the 5 x 5 magic square, we can then choose
one of the 16 configurations (call it d) and place d on the 5 x 5 magic
square so that d is pairwise disjoint from each of a, b, c. Of course, the
sum of the 5 numbers in d is 65. No matter how a, b, c and d are chosen
and placed on the 5 x 5 magic square, the 5 remaining squares of the
5 x 5 matrix will be one of the 16 configuration (call it e). Of course,
the sum of the 5 numbers in configuration e will be 65. Thus, we have
partitioned the 25 squares of the 5 x 5 magic square into 5 sets a, b, c, d, e
and the sum of the 5 numbers in each of a, b, c, d, e is 65. This fact is
much more powerful than what we have studied. This can be proved by
using the version of Hall’s Marriage Theorem stated in section 11, and
we consider this to be beyond the scope of this paper.
If a person did not know about the Fig. 1 drawing, the proof of this would
be almost hopeless even if one knows the variation of Hall’s Marriage
Theorem we mention below.

11 A U se fu l M arriage T h eo rem
Suppose dots are placed in the squares of an n x n checkerboard arbi­
trarily but so that each row and each column has exactly k dots. We

65
 Mathematics Competitions Vol 27 No 2 2014

allow more than one dot to be placed in a square. Then it is possible

to find a subset of n dots so that each row and each column contains
exactly one dot.

We are grateful to Prof. Ben Klein of Davidson College for his very
useful comments and for computing the rank of the matrix of Note 1.

R eferen ces
[1] http://en.Wikipedia.org/wiki/Hall

[2] http://www.cut-the-knot.org/arithmetic/elegant.shtml

[3] http://en.Wikipedia.org/wiki/Pandiagonal_magic_square

Arthur Holshouser Harold Reiter

3600 Bullard St. Department of Mathematics,
Charlotte, NC, University of North Carolina Charlotte,
USA Charlotte, NC 28223, USA
[email protected]

66
Copyright of Mathematics Competitions is the property of AMTT Limited, trading as
Australian Mathematics Trust and its content may not be copied or emailed to multiple sites
or posted to a listserv without the copyright holder's express written permission. However,
users may print, download, or email articles for individual use.