See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/305257580

Dürer's magic tesseract

Article  in  The Mathematical Gazette · March 2013

DOI: 10.1017/S0025557200005374

CITATIONS READS

0 1,692

1 author:

Anthony Sudbery
The University of York
104 PUBLICATIONS   3,744 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Anthony Sudbery on 06 December 2016.

The user has requested enhancement of the downloaded file.

Dürer’s magic tesseract
Anthony Sudbery
Department of Mathematics, University of York,
Heslington, York, England YO10 5DD
tony.sudbery@york.ac.uk

4 January 2012, revised 21 March 2012

Figure 1: Melencolia I (Albrecht Dürer, 1514)

1
 Albrecht Dürer’s mysterious engraving Melencolia I (Fig. 1) has always
intrigued both art critics and mathematicians. Among art critics, according
to Campbell Dodgson [1], “The literature on Melancholia is more extensive
than on any other engraving by Dürer” (he adds “that statement would
probably remain true if the last two words were omitted”). Mathematicians,
if disconcerted by the association between mathematics and melancholy, have
been fascinated by the objects appearing in the print, such as the polyhedron
occurring on the left of the engraving and—the subject of this note—the
magic square in the upper right-hand corner (Fig. 2).
16 3 2 13
5 10 11 8
9 6 7 12
4 15 14 1

Figure 2: Dürer’s magic square

This 4 × 4 square contains the numbers 1 to 16 arranged so that not only

do the rows, columns, and diagonals sum to the same number 34, but so
do many other regular sets of four numbers: Pickover ([2], p. 21) describes
26 such additional patterns. In this note I will explain these patterns, and
find 16 more, by considering the array as a four-dimensional 2 × 2 × 2 × 2
hypercube rather than a two-dimensional 4 × 4 square.
If we subtract 1 from each number in the square, we have the numbers 0
to 15, which all have binary expansions of four digits: thus the entry in row
r and column s can be written
nrs = 8ars + 4brs + 2crs + drs + 1 (r, s = 1, .., 4)
with ars , brs , crs , drs = 0 or 1. This can be seen as a 4 × 4 matrix equation
N = 8A + 4B + 2C + D + J
where N is the magic square, J is the matrix whose every entry is 1, and
   
1 0 0 1 1 0 0 1
0 1 1 0  1 0 0 1
A= 1 0 0
, B =  0 1 1 0 ,
  (1)
1
0 1 1 0 0 1 1 0
   
1 1 0 0 1 0 1 0
0 0 1 1 0 1 0 1 
C= 0 0 1
, D= 0 1 0 1  .

1 
1 1 0 0 1 0 1 0

2
 Moreover, the row and column numbers r and s both have two-digit
binary expansions: r = 2i + j, s = 2k + l with i, j, k, l = 0 or 1. Thus the
matrices (1) can be seen as 2 × 2 × 2 × 2 hypermatrices aijkl , bijkl , cijkl and
dijkl . Their entries can be simply expressed in terms of i, j, k and l:

aijkl = j ⊕ k ⊕ l ⊕ 1,
bijkl = i ⊕ k ⊕ l ⊕ 1,
cijkl = i ⊕ j ⊕ k ⊕ 1,
dijkl = i ⊕ j ⊕ l ⊕ 1.

where ⊕ denotes addition modulo 2.

We can now prove the magic square properties analytically. The sum of
the numbers in row ij of the square is
X
Rij = nijkl
kl
X
= [8(j ⊕ k ⊕ l ⊕ 1) + 4(i ⊕ k ⊕ l ⊕ 1)
kl
+ 2(i ⊕ j ⊕ k ⊕ 1) + (i ⊕ j ⊕ l ⊕ 1) + 1]

Now for any x ∈ Z2 ,

X X X X
(k ⊕ l ⊕ x) = 2 and (k ⊕ x) = 1, so (k ⊕ x) = (l ⊕ x) = 2
kl k kl kl
P
(the signs here denote addition in the integers, not in Z2 ); hence

Rij = 2(8 + 4 + 2 + 1) + 4 = 34.

Similarly the sum of the numbers in column kl is

X
Ckl = nijkl = 34.
ij

But there is complete symmetry between i, j, k and l, so we also have

X X
Σ(Sik ) = nijkl = 34, Σ(Til ) = nijkl = 34,
jl jk
X X
Σ(S jl ) = nijkl = 34 and Σ( T jk ) = nijkl = 34.
ik il

Here Σ(S) denotes the sum of the numbers in the set S, and the above
defines 16 more sets of four numbers, each summing to 34 (“quads” in the

3
terminology of Ollerenshaw and Bondi [3]). In the 4 × 4 square, the four sets
Sik occupy the four 2 × 2 squares forming the four quarters of the square
(Fig. 3C); the four sets S il occupy the corners of the four 3 × 3 subsquares
(Fig. 3D); and the eight sets Til and T jk occupy the corners of the rectangles
shown in Figs. 3E and 3F.
What about the diagonals of the 4 × 4 square, which also sum to 34?
Well, in each of the sets that we have seen so far, the four numbers occupy
a square in the four-dimensional space of points with coordinates (i, j, k, l).
(Since we are only considering coordinates with the values 0 and 1, we can
consider our points as the vertices of a hypercube in R4 , or (more profitably)
as the whole of the space Z42 , a four-dimensional vector space over the field
with two elements). The set S00 , for example, consists of the points on the
hypercube with i = 0, k = 0, i.e. lying in the jl coordinate plane, which
intersects the hypercube in a square. The other sets Sik lie in planes parallel
to this coordinate plane. In the same way, each of the above 24 sets forms a
square parallel to one of the six coordinate planes. But there are other planes,
not parallel to the coordinate planes, which also intersect the hypercube in
a square. Such a plane is given by two linear equations over the binary field
Z2 . For example, the plane

i ⊕ j = 0, k⊕l =0

contains the four points (i, i, k, k) with i, k = 0, 1. The sum of the corre-
sponding numbers is
X X
U00 = niikk = [8(i ⊕ 1) + 4(i ⊕ 1) + 2(k ⊕ 1) + (k ⊕ 1) + 1] = 34.
ik ik

The same argument applies to the parallel planes i ⊕ j = p, k ⊕ l = q for

p.q = 0, 1; and to the two other sets of four parallel planes

i ⊕ k = p, j ⊕ l = q and i ⊕ l = p, j ⊕ k = q.

Together, these give another 12 quads. They are shown in Fig. 3G-I.
Finally, there are four more sets of four parallel planes:

i ⊕ j = p, i ⊕ k = q; i ⊕ j = p, i ⊕ l = q;
i ⊕ k = p, i ⊕ l = q; j ⊕ k = p, j ⊕ l = q.

The resulting 16 quads are shown in Fig. 3J-M.

4
 • • • • • N H  • • N N
N N N N • N H  • • N N
H H H H • N H  H H  
    • N H  H H  
A. i = p, j = q B. k = p, l = q C. i = p, k = q

• N • N • N • N • • N N
H  H  • N • N H H  
• N • N H  H  • • N N
H  H  H  H  H H  
D. j = p, l = q E. i = p, l = q F. j = p, k = q

• N N • • N H  • H N 
H   H N •  H N  • H
H   H H  • N H •  N
• N N •  H N •  N H •
G. i ⊕ j = p, k ⊕ l = q H. i ⊕ k = p, j ⊕ l = q I. i ⊕ l = p, j ⊕ k = q

• N • N • N  H
 H  H H  N •
H  H  • N  H
N • N • H  N •
J. i ⊕ j = p, j ⊕ l = q K. j ⊕ k = p, k ⊕ l = q

• N  H • • N N
• N  H   H H
H  N • H H  
H  N • N N • •
L. j ⊕ k = p, k ⊕ l = q M. i ⊕ j = p, j ⊕ k = q
Fig. 3. Each of these figures shows a partition of the 4 × 4 square into four
subsets Xpq (p, q = 0, 1), such that in Dürer’s magic square the numbers in
each subset sum to 34. The cells in X00 are marked with •, those in X01 with
N, those in X10 with H, and those in X11 with .

5
 Altogether there are 86 quads [3]. Ollerenshaw and Bondi classify these
using the following terminology. If q = {a, b, c, d} is a quad (i.e. a, b, c, d are
distinct integers between 1 and 16 such that a + b + c + d = 34), then so is
q ∗ = {17 − a, 17 − b, 17 − c, 17 − d}, the complement of q. A quad is self-
complementary if q = q ∗ . A quadset is a set of four quads whose union is the
complete set of numbers {1, . . . , 16}. Thus Fig. 3 shows thirteen quadsets.
Ollerenshaw and Bondi’s first class contains all self-complementary quads.
There are 28 of these, and they can be arranged into seven quadsets. The
second class contains quads q for which q 6= q ∗ , and there is a quad p such
that {q, q ∗ , p, p∗ } is a quadset. This class contains 24 quads. The third class,
with 34 quads, contains quads q which are not self-complementary, and which
do not belong to any quadset consisting of two complementary pairs.
In this paper we have seen that 52 of the 86 quads correspond to two-
dimensional subspaces of the four-dimensional space Z42 . These are precisely
the quads in Ollerenshaw and Bondi’s first two classes. In Dürer’s magic
square complementary numbers occupy cells which are related by reflection
in the centre of the square. From this it is clear that the 28 quads in the
first class are those shown in Fig. 3G–L, while the second class consists of
the quads shown in Fig. 3 A–F.
The 34 quads in the third class do not have a simple four-dimensional
description. Four of them ({3,5,11,15}, {2,10,8,14}, {3,9,7,15}, {2,6,12,14})
are mentioned in the Wikipedia article [4] as forming kite shapes; the other
quads in this class do not form regular shapes in the magic square.

References
[1] C. Dodgson. Albrecht Dürer. Medici Society, London (1926). Quoted in
[5].

[2] Clifford A. Pickover. The Zen of Magic Squares, Circles and Stars,
Princeton University Press (2003).

[3] K. Ollerenshaw and H. Bondi. Magic squares of order 4. Phil. Trans.

Roy. Soc. Lond. A, 302 (1986), pp. 443-53.

[4] http://en.wikipedia.org/wiki/Magic square, accessed 4 January 2012.

[5] http://en.wikipedia.org/wiki/Melencolia I, accessed 4 January 2012.

View publication stats