Franklin Squares - a Chapter in the Scienti…c

Studies of Magical Squares

Peter Loly
Department of Physics, The University of Manitoba,
Winnipeg, Manitoba Canada R3T 2N2
(loly@cc.umanitoba.ca)
July 6, 2006

Abstract
Several aspects of magic(al) square studies fall within the computa-
tional universe. Experimental computation has revealed patterns, some
of which have lead to analytic insights, theorems or combinatorial results.
Other numerical experiments have provided statistical results for some
very di¢ cult problems.
Schindel, Rempel and Loly have recently enumerated the 8th order
Franklin squares. While classical nth order magic squares with the en-
tries 1::n2 must have the magic sum for each row, column and the main
diagonals, there are some interesting relatives for which these restrictions
are increased or relaxed. These include serial squares of all orders with
sequential …lling of rows which are always pandiagonal (having all parallel
diagonals to the main ones on tiling with the same magic sum, also called
broken diagonals), pandiagonal logic squares of orders 2n derived from
Karnaugh maps, with an application to Chinese patterns, and Franklin
squares of orders 8n which are not required to have any diagonal prop-
erties, but have equal half row and column sums and 2-by-2 quartets, as
well as stes of parallel magical bent diagonals.
We modi…ed Walter Trump’s backtracking strategy for other magic
square enumerations from GB32 to C++ to perform the Franklin count
[a data…le of the 1; 105; 920 distinct squares is available], and also have a
simpli…ed demonstration of counting the 880 fourth order magic squares
using Mathematica [a draft Notebook]. Our early explorations of magic
squares considered as square matrices used Mathematica to study their
eigenproperties. We have also studied the moment of inertia and multi-
pole moments of magic squares and cubes (treating the numerical entries
as masses or charges), …nding some elegant theorems, and have shown how
to easily compound smaller squares into very high order ones, e.g. order
12; 544 = 28 72 . At least two groups have patents on using relatives
of Franklin squares for cryptography, while a group at Siemens in Munich
using pandiagonal logic squares has another pending. Other possible ap-
plications include dither matrices for image processing and providing tests

1
 for developing CSP (constraint satisfaction problem) solvers for di¢ cult
problems.

1 Introduction
This article is based on a poster and allied commentary given at NKS2006.
The present contribution comes on the heels of the landmark count of Franklin
squares on a chessboard by Schindel, Rempel and Loly [1]. During the conference
I received a message from Miguel Amela [2] enlarging on this work. It also
incorporates a response to some of the questions posed by visitors to the poster.
Pasles [3] has provided a beautiful historical context for the resurgence of interest
in Franklin squares, and Maya Ahmed [4][5] posed the question which lead to
our e¤ort to count the number of Franklin squares which used the full set of
elements 1::64. This work has already been reviewed by Ivars Peterson [6].
Small magic squares are often encountered in early grades as an arithmetic
game on square arrays (patterns or motifs). Classical magic squares of the whole
numbers, 1:::n2 , have the same line sum (magic constant) for each row, column,
and main diagonals:

Cn = n(n2 + 1)=2 (1)

This line sum invariance depends only on the order, n, of the magic square.
The ancient 3 3 Chinese Lo-shu square of the …rst nine consecutive integers
is the smallest magic square and apart from rotations and re‡ections there are
no others this size or smaller:

4 9 2
3 5 7 (2)
8 1 6
where the magic sum is 15. For n = 3, C3 = 15;as expected. The best statement
that can be made about its age is 2500 1500 years! While the middle …gure
may be most appropriate, the hardest evidence gives just the lesser [7], while
legends [8] claim the older.
We brie‡y review the recreational aspect of magic squares before drawing
attention to the simpler semi-magic squares, and to pandiagonal non-magic
squares. These we lump together under the rubric of magical squares. Then
we examine the scienti…c aspects of all these squares through applications in
classical physics and matrix analysis. In fact it is partly through the coupled
oscillator problem that the mathematics of matrices was developed. Through
an elementary example in matrix-vector multiplication, which can be done at
the high school level, we demonstrate a simple eigenvalue-eigenvector problem.
Magic squares can then play a valuable role in modern courses in linear algebra.

2
2 Recreational Mathematics
At the recreational level magic squares are fun for all ages, as I found when
introducing them to visitors during the summer of 2000 whilst volunteering
at the "Arithmetricks" travelling exhibit at the Museum of Man and Nature in
Winnipeg, Manitoba. Various types of magic squares have become a recreational
pastime of amateurs, often very gifted individuals e.g. Albrecht Dürer, and Ben
Franklin. In 2006 I gave talks at various grade levels on aspects of magical
squares.
While there are several journals which publish results in this area of recre-
ational mathematics, the rise of the World Wide Web now a¤ords many of the
actors a place to publish extensive work without recourse to the oft tedious
rigours of peer review, the selectivity of editors, and a considerable time delay.
Fortunately there are some real gems from these e¤orts, but there are obvious
drawbacks for the longer term ‡owing from absence of refereeing and lack of
permanent archiving.
However magic squares present di¢ cult challenges for mathematicians and
over the past few hundred years many famous mathematicians have contributed
to our knowledge of them, including Euler.
Consider the number of possible arrangements of 1..9 in a 3x3 square after
removing an 8-fold redundancy factor from rotations and re‡ections, 9!=8 =
45; 360, so that the current fad of Sudoku has plenty of scope. Constraints on
row, column, or diagonals line sums sharply reduce the number of squares, and
there is plenty of room for other types of alternative constraints to produce
interesting squares.

3 Art and Design: Line paths in Magic Squares

Stephen Wolfram posed several questions at my poster, one concerning the line
path which can be drawn through successive numbers of a Franklin square.
This probably goes back at least a century to a time when their were many
fewer magic squares than are available today. Some of these are shown in
Cli¤ord Pickover’s recent book [9], and he includes one from Ben Franklin. If
one considers the 880 fourth order magic squares enumerated in 1693 by Frenicle
de Bessy [10][11], which have been classi…ed by Dudeney into 13 types, then the
symmetry of the line path may distinguish squares with more than the minimal
constraints. Stephen had in mind an underlying group theory. With the advent
of larger complete sets of magic squares more study in this direction would be
timely.
Early in the 20th century the famed architect Claude Fayette Bragdon [12]
used line paths from some order 3; 4 and 5 magic squares as the basis of ornament
for the interior and exterior of buildings, especially in Rochester, New York.

3
4 Where is the Science?
Why did a theoretical physicist get involved with magic squares? This was not a
linear process. Having been blissfully unaware of them for my …rst …ve decades,
an encounter with the Myers-Briggs Type Indicator c [13] scheme of person-
alities resonated with my background in mathematical structures, partly from
research in solid state physics. Coordinate rotation matrices in classical and
relativistic mechanics, combined with periodic boundary conditions for …nite
crystals soon lead to links with magic squares, while the psychological nature
of the MBTI eventually made connections with early work of Carl Jung, which
…nally connected with Jung’s [14] interest in Chinese patterns, especially the
dichotomous yin-yang schemes of Feng Shui (a.k.a. The Golden Flower ) and
the I Ching.
The science begins whenever we go beyond a recipe approach for constructing
a single square, e.g. counting or estimating the populations of various classes
of squares, proving that none are possible in a given case, interpreting them as
arrays of point masses or electric charges. There are more examples, but another
early aspect of our own work focusses on the remarkable results associated with
treating magic squares as matrices, in the context of linear algebra, i.e. solving
sets of simultaneous equations, as well as some topics in classical physics.

5 Some Special Varieties of Squares

For a number of purposes it is important to recognize that there are several
specially important variations on the theme of magic squares.
Firstly, semi-magic squares, which do not necessarily have the diagonals
summing to the row-column line sum, some of which may be obtained simply
by moving an edge row and/or column to the opposite side, e.g.

9 2 4
5 7 3 (3)
1 6 8

The removal of the diagonal constraints means more squares due to the smaller
number of constraints, in this case there are eight more. Secondly, pandiagonal
non-magic squares, which have the same magic line sum for all the split lines
parallel to the main diagonals. We can illustrate pandiagonals by taking a non-
magic serial square (having the consecutive integers …ll row-by-row) and tiling
a copy to its right (or left, or top or bottom):

1 2 3 1 2 3
4 5 6 4 5 6 (4)
7 8 9 7 8 9
The pandiagonals are (1; 5; 9), (2; 6; 7), (3; 4; 8), (3; 5; 7), (1; 6; 8), (2; 4; 9), to-
gether with the main diagonals. Observe that for a given order the number

4
of row and column constaints is the same as the number of pandiagonal con-
straints. Serial squares exist for all orders, unlike magic squares which do not
exist for n = 2:
A pandiagonal magic square is the combination of this pandiagonal property
with the requirements of a magic square. These …rst occur in order 4, and of
these 48 are found amongst the 4th order squares, with none existing (possible)
for singly even orders (6; 10; :::):

16 2 3 13
5 11 10 8
(5)
9 7 6 12
4 14 15 1
A third interesting type, namely associative (or regular) squares have the an-
tipodal property:

aij + an i+1;n j+1 = n2 + 1; i; j = 1::n (6)

Finally, our recent Franklin result [1], and the most-perfect pandiagonal magic
squares of McClintock [15], for which Ollerenshaw and Brée’s combinatorial
count ranks as a major achievement, draw attention to a fourth type which
have the same sum for all 2 2 subsquares (or quartets). Franklin Squares also
have constant half row and column sums, and constant parallel bent diagonals
[1].
Ollerenshaw and Brée [16] have a patent for using most-perfect magic squares
for cryptography, and Besslich [17][18] has proposed using pandiagonal magic
squares as dither matrices for image processing.

6 Counting Magic Squares - the role of Back-

tracking Computations
There are 880 distinct 4 4 magic squares of the …rst 16 integers, and the
275; 305; 224 distinct 5 5 of the …rst 25 integers, the latter were …rst counted by
computer in 1973 [19]. Already by order six they have become uncountable, and
as a result only statistical estimates are then possible. Pinn and Wieczerkowski
[20] performed a Monte Carlo simulated annealing computation for an estimate
of (0:17745 0:00016) 1020 for the 6 6, and a less accurate estimate of the
7 7.
Walter Trump [21] developed a more e¢ cient hybrid backtracking Monte
Carlo method and improved the accuracy of these estimates. He has also made
good estimates of the number of various types of magic squares up to 10 10,
a remarkable achievement. During summer 2003 some of my undergraduate
students at Manitoba took Trump’s 6 6 GB32 code, which uses 13 random
cells, and converted it to C++. Matt Rempel …nds that the C++ code runs
about 10% faster than GB32 on the same PC in producing a sample of some
700; 000 squares, and has begun to remove random cells, …nding longer run

5
times, but more accurate results, and a larger sample of magic squares. Dan
Schindel has taken the ideas in the 6 6 code and constructed pure backtracking
codes without random cells to count the known numbers of magic squares for
the 4 4 and 5 5 cases. While we have previously been able to analyse the
complete set of 4 4’s, we can now begin to analyse the 5 5’s. This gives us the
ability to study a variety of interesting questions, e.g. their eigenproperties. Dan
Schindel has also amended the 4 4 code to count the number of pandiagonal
non-magic 4 4 squares, …nding some three million.
A number of others have polished backtracking codes which can be run from
their web pages. Meyer [22] has one of the best and it will …nd a stream of
di¤erent squares for n = 4; 5 and 6.
A simpli…ed demonstration of counting the 880 fourth order magic squares
using Mathematica [a draft Notebook][23]. Our Mathematica code can be re-
…ned and we hope to harmonize it with Eric Weisstein’s Mathematica tools for
magic squares [24].

7 Integer Points in Polyhedral Cones

In 2004 Ahmed [4] asked “How Many Squares Are There, Mr. Franklin?...” in
a paper which exploited Hilbert bases of polyhedral cones (PHC) to construct
several new natural Franklin squares. Ahmed [4][5] was also able to use PHC
techniques to count the total number of 8th order Franklin squares as a func-
tion of a variable magic sum (s), obtaining a count of 228; 881; 701; 845; 346
for s = 260. However this includes many squares with degenerate elements, so
that this count is an upper bound to the population of natural squares. PHC
techniques do not at present permit the elements to be distinct so that they do
not give the smaller counts expected for natural squares, where the elements
1::n2 are distinct. These PHC techniques, and their upper bound statements,
have received considerable attention recently [25], in publications [Ahmed [4]
[5]; Ahmed, De Loera and Hemmecke [26]; Beck et al [27]; and have recently
been the subject of advanced courses ([28][29]).

7.1 Comparison of Various Counts

We construct a table after the fashion of Trump [21] to collect some extant
results for relevant counts and to summarize some of the major results:

6
 order of square, n 4 5 8
natural magic sum, Cn 34 65 260
natural magic 880 275; 305; 224 (a) 5:2210(70) 1054 (b)
associative natural 48 48; 544 (b) 2:5228(14) 1027 (b)
natural panmagic 48 3; 600 (b) (c)
‘complete’ 48 - 368; 640 (d)
natural Franklin 0 (e) - 1; 105; 920 (here)
PHC UPPER BOUNDS
pan Franklin (f,g) - - 10; 308; 923; 109; 408
Franklin (f,g) - - 228; 881; 701; 845; 346
magic (h,i) 163; 890; 864
panmagic (g) 35; 208 53; 852; 072; 626

Table 1: Comparison of Various Counts: footnotes: (a) Schroeppel [19];

(b) Trump [21] with statistical errors for n = 8; (c) lies between preceding
numbers in this column (b); (d) Ollerenshaw and Brée [7]; (e) Pasles [3], see his
footnote 22 on page 506; (f) Ahmed [4]; (g) Ahmed [5]; (h) Ahmed, De Loera
and Hemmecke [26]; (i) Beck [27].

8 Magic Squares and CSP’s (Constraint Satis-

faction Problems)
CSP’s fall into several classes: genetic algorithms, evolutionary computing, ....
Magic squares are frequently used as test targets in benchmarking improved
CSP strategies. Sometimes it is the time to …nd the …rst square [30], or the
time to …nd a complete set such as the 880 fourth order squares. The highly
constrained eighth order most-perfect squares have also been used.

9 Compound Squares
An undergraduate project with Wayne Chan took an old Chinese idea for com-
pounding a 3 3 magic square with itself to construct a 9 9 magic square, or a
3 3 with a 4 4 to make a pair of 12 12 magic squares, and devised a computer
program [31] to extend this to very large squares. One of the squares is used as
a frame and the other is incremented on placement in the appropriate position
in the frame. Compounding preserves the row, column and pandiagonal char-
acteristics which are common to both squares, and even for the smallest case
of 9th order there are very large numbers of distinct squares [37]. As a result
we were able to set a new world record sized magic square at 12; 544 12; 544.
Since it is di¢ cult to grasp a square of this size with numbers running from
1::157; 351; 936 we used a colour scale to make an image which might pass for
a piece of art [32]. In 2006 the world record for magic squares is still held by a
smaller square of order 3001, because of rules which require writing or printing
out the square on paper. We stopped at order 12; 544 simply because it was the

7
largest which we could write onto a CD. It is clear that larger magic squares
could be stored on a DVD, or some higher capacity disks, but there is little
point since potential applications will need fast access.
Let us note that Sudoku is a special type of Latin square which bears some
similarity to compound squares [33].

9.1 Multimagic Squares

Christian Boyer [34] has drawn attention to magic squares which remain magic
when their elements are raised to integer powers (2 bimagic; 3: trimagic; etc.).
An interesting mathematical analysis of multimagic squares has been given re-
cently by Derksen, Eggermont, and van den Essen [35], which includes a con-
tribution to compound squares. Rempel, Chan and Loly have a related project
underway [36].

10 A Mechanical Application - Moment of In-

ertia
Another success growing out of teaching undergraduate classical mechanics for
many years was the discovery of a new invariance, or universal property, for
magic squares through calculations of their "moments of inertia" (essentially
the inertia of the square through an axis perpendicular to its centre), which
eventually turned out to depend only on the order of the square, i.e. the number
of rows or columns [38]. The numbers in the magic square are replaced by
corresponding multiples of a unit mass placed on a square unit lattice. In fact
this was truly a "Eureka" moment for quite spontaneously I had the idea which
fused long activity in teaching moment of inertia in introductory courses with
a more recent activity in magic squares.
The moment of inertia, In , of a magic square of order n about an axis
perpendicular to its centre is obtained by summing mr2 for each cell, where
m is the number centred in a cell and r is the distance of the centre of that
cell from the centre of the square measured in units of the nearest neighbour
distance. For
p the Lo-shu square the corner cells then have their centres at a
distance of 2 from the axis. We can now calculate the sum for the 3 3:

2
p 2
I3 = [1 + 3 + 7 + 9] (1) + [2 + 4 + 6 + 8] 2 = 60 (7)

The moments of inertia about the horizontal and vertical axes through the centre
are each 30, reminding us of the perpendicular axis theorem, which says that
their sum gives the moment of inertia about the axis through the centre and
perpendicular to the plane.
When I used a data …le for the complete set of the 4 4’s (by courtesy of
Harvey Heinz [39]) it was a surprise to …nd that they all gave I4 = 340. I was
then motivated to attempt a derivation, which was easy since the calculations
only depended on the semi-magic property so that the parallel axis theorem

8
and the perpendicular axis theorem could be used. In retrospect this could
have been set as an examination problem in a sophomore course on classical
mechanics!
1 2 4
In = n (n 1) (8)
12
This remarkably simple formula recovers the results for n = 3; 4 above and is
valid for arbitrary order. The derivation of the formula only depends on the
row and column properties, and not on the diagonals of magic squares, so that
it actually applies to the larger class of semi-magic squares which lack one or
both diagonal magic sums of magic squares.
Of related interest, Abiyev et al [40] have studied the centre of mass of
certain magic squares and suggest applications to robotics.

10.1 RC (RowColumn) symmetry.

In the above one can factor out the magic linesum (Cn ):

1 2 4 1 1 1
In = n (n 1) = n n2 + 1 n n2 1 = Cn n n2 1 (9)
12 2 6 6
for a result which applies to any square of equally spaced rows and columns
with the same mass, Cn . The limit of a uniform continuous sheet agrees with
the standard result using the calculus. It is clear that large random semimagic
squares tend towards the limitting value of this expression.

11 Magical Cubes and Hypercubes

Along with magic squares, there has long been an interest in magic cubes,
going back at least to Leibniz [41], and then later in 4D hypercubes following
Riemann’s n-dimensional geometry in the second half of the 19th century. There
are now studies of higher dimensional hypercubes. For magic cubes there are
some fascinating applications. We give a link to an image of a magic cube with
spheres sized according to mass [43], as well as a paper [44] on perfect magic
cubes, which also has an image.
Adam Rogers has recently helped extend my inertia ideas ideas to the cal-
culation of the full inertia tensor of magic cubes [42]. RCP (RowColumnPillar)
symmetry means that any cube of equally spaced RCP’s with the same mass in
each will have the same form of inertia tensor. This work has also recently been
reviewed by Ivars Peterson [46].

12 Electric Quadrupoles
A new magic square topic has just emerged from my renewed involvement with
our honours electromagnetism course.

9
 The idea is to treat the numerical value of each element as an electric charge.
It is soon clear for small squares that the dipole moment vanishes, so we then
proceed to study the quadrupole moment. As a …rst thought I neutralized
magic squares so that their elements ran from n2 1 =2 to + n2 1 =2,
but later Rogers and Loly [45] analyzed the multipole expansion for a normal
magic square, …nding that it takes care of many of the details. The full story
involves calculating the quadrupole tensor, something beyond the scope of the
present article. The short story is that the quadrupole tensor vanishes, so one
then proceeds to the octupole!

13 Pandiagonal Non-Magic Squares (order 2n =

2; 4; 8; 16; :::)
Here the highlight is the discovery [51] of a small new class of purely pandi-
agonal non-magic, number squares having dimensions of the powers of 2, with
the additional property that the binary representation changes by just 1-bit
in horizontal and vertical moves (they are counted from 0::15 in the example
below):

0 1 3 2
4 5 7 6
(10)
12 13 15 14
8 9 11 10
These squares derive from the Myers-Briggs dichotomous scheme of personality
types. Then work with summer undergraduate Marcus Steeds (Loly and Steeds
2003), who devised a number of useful test tools, explored how generalizations
of these squares are related to the Gray code and square Karnaugh maps of
digital logic design, a connection made by some of my third year engineering
students a few years ago. I have also applied these ideas to ancient Chinese
patterns based on the yin-yang duality [52].
Recently Dan Schindel found that of the 3; 465; 216 pandiagonal non-magic
squares, 48 have the 1-bit property. This agrees with a symmetry argument
made recently by Ian Cameron [47].
Meine and Schuett [48] of Siemens have suggested applications of these
squares to cryptography and image processing [49].

14 From Coupled Oscillators to Modern Linear

Algebra
Undergraduate physics provides several examples in mechanics and wave motion
(from coupled oscillators) where semi-magic matrices arise with algebraic or non-
integer elements. The investigation of the mechanical problems had a vibrant
interplay with mathematics for two centuries from the time of Huyghens and

10
Newton. Huyghens, of course, is well known for his study of the isochrony of
pendulum motion. An excellent chronology is found in Brillouin [53], whose
study of waves in periodic systems is a tour-de-force.
A central mathematical theme in physical science and engineering concerns
what are known as eigenvalue problems. These involve homogeneous linear
equations which only have non-trivial solutions if the determinant of the coe¢ -
cients vanishes, with as many solutions as the number of equations. These issues
can be clari…ed by using a speci…c example for which the coupled oscillator is
ideal. At the same time we can prepare the ground for studying magic square
matrices in their own right.

14.1 Homogeneous Simultaneous Equations for the Cou-

pled Oscillator
In the usual description of this one-dimensional problem [54] the equations of
motion for masses M displaced along the x-direction from their equilibrium
positions by x1 and x2 , a coupling spring of force constant ; and with each tied
to …xed posts at opposite ends by springs of force constant are:

Mx
•1 + ( + )x1 x2 = 0 (11)

Mx
•2 + ( + )x2 x1 = 0 (12)
These are simpli…ed by taking out a simple time-dependence: x(t) = B exp (it!)
for:
( + M ! 2 )B1 B2 = 0 (13)
B1 + ( + M ! 2 )B2 = 0 (14)
Instead of the general approach of setting the determinant of the coe¢ cients of
these simultaneous equations to zero, this simple problem may be solved simply
by forming ratios of the variables:

B1 ( + ) M !2
= = (15)
B2 ( + ) M !2

Cross multiplication of the second equality results in a quadratic equation in

! 2 , with two solutions, one, ! 2 = M , just the frequency of each oscillator
without coupling, and the other higher, ! 2 = +2
M . Extended to a chain of two
or more alternating masses and springs, we have the origin of the gaps in the
spectrum which are a characteristic feature of solid state physics. We continue
this example after introducing some essential matrix operations, which are easy
enough to cover in high school.

11
15 Determinants and Matrices
Cayley initiated matrix theory in 1846, followed by contributions from Peirce,
Hamilton, Poincaré, and Sylvester. We highlight the issues of interest with
respect to magic squares, the semi-magic matrices and tensors arising in me-
chanics, and the related interest for pandiagonal non-magic squares with a brief
discussion using 2 2 matrices.

15.1 Matrix Multiplication

Consider the usual matrix-vector multiplication:

a b P aP + bQ
= (16)
c d Q cP + dQ
Clearly if P = Q = 1, this sums the rows:

a b 1 a+b
= (17)
c d 1 c+d
This [1; 1] vector will be referred to as a diagonal (or 2-agonal) vector, and
generalizes to higher orders. However if one takes the original matrix operator
to act from the right onto a row vector on the left, as a ”left-hand” problem,
then one …nds the column sums of the original matrix:
a b
P Q = P a + Qc P b + Qd (18)
c d
if P = Q = 1, this sums the columns.
a b
1 1 = a+c b+d (19)
c d
This may also be achieved by left multiplication of the transposed matrix with
a row vector as illustrated next.

15.2 Eigenproblems - Eigenvectors and Eigenvalues

My interest in these issues began with an observation from experimental com-
puting of a few di¤erent order magic squares with Mathematica during a sopho-
more course which I was teaching. My colleague Frank Hruska had published a
relevant and stimulating paper [55], David Lavis drew my attention to left and
right eigenvectors, whilst another, Joe Williams, knew from teaching linear al-
gebra that this eigenvector adds row elements, etc. The utility of the "n-agonal"
eigenvector [1; 1; 1; :::] of the n-cube is seen by showing how the rows sum in the
Lo-shu magic square :
2 32 3 2 3 2 3
3 5 7 1 15 1
4 4 9 2 5 4 1 5 = 4 15 5 = 15 4 1 5 (20)
8 1 6 1 15 1

12
where we have factored out the eigenvalue, 15, to show the action of the matrix
operator in leaving the eigenvector unchanged. The columns have the same
eigenvalue as follows from the transposed matrix:
2 32 3 2 3 2 3
3 4 8 1 15 1
4 5 9 1 5 4 1 5 = 4 15 5 = 15 4 1 5 (21)
7 2 6 1 15 1

Alternatively we can work with a row eigenvector on the left with the original
matrix:

22 33
3 5 7
1 1 1 44 4 9 2 55 = 15 15 15 = 15 1 1 1 (22)
8 1 6

We must note that the n-agonal eigenvector is both a left and a right eigen-
vector, and that this property depends only on the semi-magic property. An
immediate application is now a¤orded by the coupled oscillator.

16 Coupled Oscillator Eigenvectors

When written in modern matrix notation, the semi-magic nature of the charac-
teristic equation (13; 14) is apparent:

+ B1 B1
= M !2 (23)
+ B2 B2
It is immediately clear that a solution is the [1; 1] eigenvector, both as a right,
and as a left eigenvector. It has the lower eigenvalue of ! 2 = M . The other
eigenvector is [1; 1], which corresponds to the higher solution of ! 2 = +2 M
A similar semi-magic property is also found for the full moment of inertia
tensor of magic cubes [42].

16.1 Left and right eigenvectors

Of signi…cant interest for our studies of magic squares, this topic is important for
teaching linear algebra beyond the introductory course. The use of magic square
examples already occurs in such courses, but it can be used even more seriously.
The existence of identical left and right eigenvectors implies deeper properties
giving rise to the theorem of biorthogonality, and the theorem of Perron [56].
These theorems develop deep links between the left and right eigenvectors and
eigenvalues. As such, magic squares are insightful examples for advanced linear
algebra courses.
Software such as Mathematica, Maple and MATLAB can be pro…tably em-
ployed in such studies. Indeed, MATLAB has initiated some of this already by
including a function, magic(n), which returns a magic square from one of three

13
algorithms, one each for odd, even and doubly-even cases. A drawback with
magic(n) is that the single squares which result are not representative of the
richness of the spectrum of magic squares of a given order, save for n > 3.

16.2 Eigenvalues of Integer Squares

In another study we have nearly …nished a study of the eigenvalues of magic
square matrices [58]. For the 880 distinct 4 4’s in the 12 Dudeney groups, we
…nd that members of the …rst six (singular) groups have three distinct eigenvalue
patterns, with a subset of the …rst three groups having three zero eigenvalues,
while the last six (non-singular) groups have two further eigenvalue patterns.
Also if the 1-bit pandiagonal non-magic squares discussed earlier are treated
as matrices they possess examples with just two non-zero eigenvalues for any
order (Loly and Steeds, 2003).
I can add here that 8th order Franklin squares have 3 non-zero eigenvalues,
as do also the corresponding most-perfect squares.
Stephen Wolfram was interested in these large nullspaces, and in what one
might conclude about random matrices.

17 Conclusion
Further information on the history may be found in recent books (Swetz, 2000)
by Frank Swetz, a mathematics educator, by René Descombes (2001), as well as
in Cli¤ord Pickover’s recent book (Pickover, 2002). Those sources also enlarge
on the philosophical aspects, which began in China as a cosmology, or organizing
scheme.
There are opportunities to enrich teaching in classical physics, and likely
in quantum physics as well. Certainly more can be done in the context of
teaching linear algebra, which can begin in high school. I have found wonderful
opportunities for students to cooperate in some group work as summer research
assistants, indeed their enthusiasm and initiative in tackling problems has been
gratifying.

18 Acknowledgments
This article is an elaboration of a poster talk given at NKS2006, which has been
augmented as a result of feedback from Stephen Wolfram and others. David
Lavis, Frank Hruska, Joe Williams, Walter Trump, John Hendricks, and Harvey
Heinz have helped with advice on various aspects of this study, which engaged
the following undergraduates: Marcus Steeds, Wayne Chan, Adam Rogers, Matt
Rempel and Dan Schindel.
Critical funding for this research was provided in 2003 by the Winnipeg
Foundation Post Secondary Grants Program, especially for support to Adam
Rogers and Matt Rempel. My colleagues Professors John Vail and Tom Osborn

14
have generously let their summer student Daniel Schindel continue working part
time on some of these projects during 2003 and 2004.

References
[1] Schindel, D., M. Rempel and Loly, P.D., 2006, Enumerating the bent di-
agonal squares of Dr Benjamin Franklin FRS, Proc. R. Soc. A: Physical,
Mathematical and Engineering, 462, 2271-2279, August 2006.
[2] Amela, M., Structured 8 x 8 Franklin Squares,
http://www.region.com.ar/amela/franklinsquares/
[3] Pasles, P. C., The lost squares of Dr. Franklin: Ben Franklin’s missing
squares and the secret of the magic circle, Am. Math. Monthly, 108 (2001),
pp. 489-511.
[4] Ahmed, M. M., How Many Squares Are There, Mr. Franklin?: Constructing
and Enumerating Franklin Squares, Am. Math. Monthly, 111 (2004), pp.
394-410.
[5] Ahmed, M.M. 2004 Algebraic Combinatorics of Magic Squares, Ph.D. dis-
sertation, University of California, Davis..
[6] Peterson, I., Counting Franklin’s Magic Squares, Science News Online, 24
June 2006, http://www.sciencenews.org/articles/20060624/mathtrek.asp
[7] Ollerenshaw, K. and Brée, D. S.: 1998, Most-perfect pandiagonal magic
squares: their construction and enumeration. Southend-on-Sea, U.K.: The
Institute of Mathematics and its Applications.
[8] Swetz, F. J.: 2002, Legacy of the Luoshu - The 4000 Year Search for the
Meaning of the Magic Square of Order Three. Chicago: Open Court.
[9] Pickover, C.: 2002, The Zen of Magic Squares, Circles, and Stars - An Ex-
hibition of Surprising Structures across Dimensions. Princeton University
Press.
[10] Frénicle de Bessy, M., Des Quarrez ou Tables Magiques, Histoire de
l’Académie royale des sciences avec les mémoires de mathématique et de
physique tirés des registres de cette Académie, 1729 (T.4), p.209-304;
http://gallica.bnf.fr/ark:/12148/bpt6k34948
[11] Frénicle de Bessy, M., Table Generale des Quarrez Magiques de quatre, His-
toire de l’Académie royale des sciences avec les mémoires de mathématique
et de physique tirés des registres de cette Académie, 1729 (T.4), p.305-354;
http://gallica.bnf.fr/ark:/12148/bpt6k34948
[12] Bragdon, C.,The Frozen Fountain: Being Essays on Architecture and the
Art of Design in Space, New York, 1932, Alfred A. Knopf, Kessinger Pub-
lishing’s Rare Mystical Reprints, www.kessinger.net

15
[13] Myers, I. B. with P. B. Myers, 1980, 1993. Gifts Di¤ering: Understanding
Personality Type. CPP Books, Palo Alto, California.
[14] von Franz, M. L.: 1974, Number and Time: Re‡ections Leading Toward
a Uni…cation of Depth Psychology and Physics, Northwestern University
Press.
[15] McClintock, E., On the Most Perfect Forms of Magic Squares, with Meth-
ods for Their Production, Am. J. Math. 19 (1897), pp. 99-120.
[16] WO 99/66671 Patent: Ollerenshaw, Kathleen and Brée, David S., Trans-
position Cipher, pp. 30, 17 June 1999.
[17] Besslich, Ph. W., Comments on electronic techniques for pictorial image
reproduction, IEEE Trans. on Comm. 31, 846-847, 1983
[18] Besslich, Ph. W., A Method for the Generation and Processing of Dyadic
Indexed Data, IEEE Trans. on Computers, C-32, No. 5, 487-4494, 1983

[19] Schroeppel, R.: 1973, see M. Gardner in his Scienti…c American column,
Jan. 1976.
[20] Pinn, K. and Wieczerkowski, C.: 1998, Number of Magic Squares from
Parallel Tempering Monte Carlo. International Journal of Modern Physics
C, 9, No. 4, 541-546.
[21] Trump, W.: 2003, Notes on Magic Squares and Cubes,
http://www.trump.de/magic-squares/
[22] Meyer, H. B., magic squares of order 4,
www.faust.fr.bw.schule.de/mhb/backtrack/mag4en.htm
[23] Loly, P. D. and Schindel, D. G., A simpli…ed demonstration of counting the
880 fourth order magic squares using Mathematica [a draft Notebook]
[24] Weisstein, E., Magic Square, Wolfram MathWorld,
http://mathworld.wolfram.com/MagicSquare.html
[25] D. Mackenzie, Quadramagicology, New Scientist, 180 (2003), pp. 50-53.
[26] M. M. Ahmed, J. De Loera, and R. Hemmeke, Polyhedral cones of magic
cubes and squares, in New Directions in Computational Geometry, The
Goodman-Pollack Festschrift volume, Aronov et al., eds., Springer-Verlag,
2003, pp. 25-51.
[27] M. Beck, M. Cohen, J. Cuomo, and P. Gribelyuk, The number of "magic"
squares and hypercubes, Amer. Math. Monthly 110 (2003), pp. 707-717.

16
[28] Joint AMS-IMS-SIAM Summer Research Conference on Integer Points in
Polyhedra, Geometry, Number Theory, Algebra, Optimization, 2003; pro-
ceedings to appear in Cont. Math. (Proceedings of the Summer Research
Conference on Integer Points in Polyhedra, July 13 - July 17, 2003 in Snow-
bird, Utah).: ams.org/meetings/src-barvin.html

[29] Ban¤ International Research Station summer school, 2005:

pims.math.ca/birs/workshops/2005/05ss027/
[30] Jacobs, C. and Hadler, H., TITLE, http://ls11-www.cs.uni-
dortmund.de/people/kursawe/Demos/EvoNet/index.html
[31] Chan, W. and Loly, P. D.: 2002, Iterative Compounding of Square Matrices
to Generate Large-Order Magic Squares, Mathematics Today (IMA), 38(4),
113-118.
[32] Chan, W. S. and Loly, P. D., Colour image of 12,544th order compound
magic square, http://home.cc.umanitoba.ca/~loly/smallmath.jpg

[33] Pegg, E. Jr., Sudoku, Wolfram MathWorld,

http://mathworld.wolfram.com/Sudoku.html
[34] Boyer,C., Multimagic squares, http://www.multimagie.com/
[35] Derksen, H., Eggermont, C. and van den Essen, A., Multimagic Squares,
arXiv:math.CO/0504083 v2 5 Apr 2005, to appear in Am. Math. Mnthly.
[36] Rempel, M., Chan, W.S. and Loly, P.D., Compounding Regular and Mul-
timagic Squares, and a Second Method, preprint.
[37] Bellew, James (1997). Counting the Number of Compound and Nasik Magic
Squares, Mathematics Today, August 1997, 111–118.
[38] Loly, P. D.: 2004, The Invariance of the Moment of Inertia of Magic
Squares. The Mathematical Gazette 88 (511), 151-3, March 2004.
[39] Heinz, H., Order-4 Magic Squares, http://www.geocities.com/~harveyh/order4list.htm
[40] Abiyev, A-A, et al, Investigation of center of mass by using magic squares
and its possible engineering applications, Robotics and Autonomous Sys-
tems, 49 (2004) 219-226.
[41] Heinz, H., Magic Cubes - Introduction,
http://members.shaw.ca/hdhcubes/

[42] Rogers, A. and Loly, P. D.: 2004 The Inertia Tensor of Magic Cubes,
American Journal of Physics, 72(6), 786-9, June 2004
[43] Loly, P. D., Magic Cube with Massive Balls using Visual Python,
http://home.cc.umanitoba.ca/~loly/squareBalls6.bmp

17
[44] Boyer, C., Perfect magic cubes, http://cboyer.club.fr/multimagie//English/Perfectcubes.htm
[45] Rogers, A. and Loly, P. D.: 2005 The electric multipole expansion for a
magic cube, European Journal of Physics 26 (2005) 809-813.
[46] Peterson, I., Magic Square Physics, Science News Online, 1 July 2006,
http://www.sciencenews.org/articles/20060701/mathtrek.asp
[47] Cameron, I.: 2003, private communication.
[48] Meine, S., and Schuett, D., On Karnaugh Maps and Magic Squares,
Informatik-Spektrum, 28(2), 120-123 (2005).
[49] Schuett, D., Eckhardt, U. and Suda, P., Boot Algebras, TR-92-039, Jun
1992, International Computer Science Institute, Berkeley, CA.
[50] Loly, P. D.: 2002-2003, A purely pandiagonal 4*4 square and the Myers-
Briggs type Table, J. Rec. Math., 31 (1), 29-31.
[51] Loly, P. D. and Steeds, M.: 2002, A New Class of Pandiagonal Non-Magic
Squares, Int. J. Math. Ed. Sci. Tech. 36 (4), 2005, 375-388,
[52] Loly, P. D.: 2002, A Logical Way of Ordering the Trigrams and Hexagrams
of the Yijing. The Oracle: The Journal of Yijing Studies, 2 (12), 2-13.
[53] Brillouin, L.: 1946, Wave Propagation in Periodic Structures. New York:
McGraw-Hill.
[54] Marion, J. B. and Thornton, R. K.: 1995, Classical Dynamics of Particles
and Systems, 4th edition, Harcourt, Brace, Jovanovich.
[55] Hruska, F.: 1991, Magic Squares, Matrices, Planes and Angles. J. Rec.
Math. 23(3), 183-189.
[56] Mattingly, R. B.: 2000, Even Order Regular Magic Squares are Singular.
American Mathematical Monthly, 107, 777.
[57] Moler, C., MATLAB’s Magical Mystery Tour, The MathWorks Newsletter,
7 (1), 8-9, 1993.

[58] Schindel, D.S., Trump, W. and Loly, P. D., Singular Regular Magic Squares:
a special set of skew-centrosymmetric matrices, preprint 2006
[59] Descombes, R., 2001: Les Carrés Magiques, Histoire, théorie et technique
du carré magique, de l’Antiquité aux recherches actuelles. Vuibert, Paris.

18