Spring 2015

Chris Christensen
MAT/CSC 483

Polygraphic Substitution Ciphers: The Hill cipher

The Playfair cipher is a polygraphic cipher; it enciphers more than one letter
at a time. Recall that the Playfair cipher enciphers digraphs – two-letter
blocks. An attack by frequency analysis would involve analyzing the
frequencies of the 26 × 26 = 676 digraphs of plaintext. Complications also
occur when digraph frequencies are considered because sometimes common
plaintext digraphs are split between blocks. For example, when encrypting
the phrase another type the digraphs are an ot he rt yp e_, and the common
digraph th is split between blocks.

Frequency analysis would be even more complicated if we had a

cryptosystem that enciphered trigraphs – three-letter blocks. Frequency
analysis would involve knowing the frequencies of the 26 × 26 × 26 =17576
trigraphs.

Systems that enciphered even larger blocks would make cryptanalysis even
more difficult. There are 26 × 26 × 26 × 26 =456976 blocks of length 4,
26 × 26 × 26 × 26 × 26 = 11881376 blocks of length 5,
26 × 26 × 26 × 26 × 26 × 26 =308915776 blocks of length 6 … .

If we had a cryptosystem that encipher blocks of length 100, there would be

3142930641582938830174357788501626427282669988762475256374173
1753989959842010402346543259906970228933096407508161171919783
5869803511992549376

plaintext blocks.

But, there is no obvious way to extend the Playfair key square to three or
more dimensions.

… cryptographers … tried to extend [Wheatstone’s Playfair cipher’s]

geometrical technique to trigraphic substitutions. Nearly all have
failed. Perhaps the best known effort was that of Count Luigi Gioppi
di Tükheim, who in 1897 produced a pseudo-trigraphic system in

1
 which two letters were monoalphabetically enciphered and the third
depended only on the second. Finally, about 1929, a young American
mathematician. Jack Levine, used six 5 × 5 squares to encipher
trigraphs in an ingenious extension of the Playfair. But he did not
disclose his method.

This was the situation when a 38-year-old assistant professor of

mathematics at Hunter College in New York published a seven-page
paper entitled “Cryptography in an Algebraic Alphabet” in The
American Mathematical Monthly for June-July 1929. He was Lester
S. Hill [1890 – 1961]. … Later in the summer in which his paper on
algebraic cryptography appeared, he expanded the topic before the
American Mathematical Society in Boulder, Colorado. This lecture
was later published in The American Mathematical Monthly [March
1931] as “Concerning Certain Linear Transformation Apparatus of
Cryptography.” The Codebreakers by David Kahn

The Hill cipher is a cryptosystem that enciphers blocks. Any block size may
be selected, but it might be difficult to find keys for enciphering large
blocks.

2
 Matrices

The Hill cipher is usually taught by means of matrices.

A matrix is just a rectangular array of numbers. For example,

 −1 2 17   0 6 −8 23 65 
 1 9   ,  3  , and  −9 76 1 98 −10 
 −3 7  
, 43 0 9  
   7 −23 9   2   
 11 7 34 72 1 
 

are all matrices.

The dimension of a matrix is given as

number of rows × number of columns.

For the four matrices given above, the dimensions are 2 × 2 , 3 × 3 , 2 × 1 , and
3 × 5 , respectively. The dimensions are read as “2 by 2, 3 by 3, 2 by 1, and 3
by 5.”

If the number of rows equals the number of columns, the matrix is said to be
a square matrix. The first two matrices above are square matrices.

If the matrix has only one column, the matrix is said to be a column matrix.
The third matrix above is a column matrix.

We will deal exclusively with square and column matrices.

Matrices can be added, subtracted, multiplied, and in some cases divided just
like numbers. The fact that an array of numbers can be treated as a single
number is what permits the theory of matrices to extend cryptographic
techniques to higher dimensions.

3
 Multiplication by Square Matrices

One origin of matrices is the solution of systems of linear equations, and the
multiplication of matrices reflects that use.

For example, consider this system of two linear equations in two variables x
and y. a, b, c, d, u, and v represent constants (i.e., numbers).

ax + by =
u
cx + dy =
v

This system of equations can be represented by one matrix equation

 a b   x  u 
c d   y  = v  .
    

The square matrix is called the coefficient matrix (a, b, c, and d are the
coefficients of the variables x and y). There are two column matrices – one
consisting of the two variables x and y and the other of the two constants that
appear on the right hand side of the system u and v.

For the moment, we will only consider matrix multiplication of the form

square matrix × column matrix

Such a multiplication is only defined if the number of columns of the square

matrix equals the number of rows of the column matrix.

The system of equations gives us the pattern for multiplication.

 a b   x   ax + by 
 c d   y  =  cx + dy 
    

The top entry of the product is calculated by, first, taking the entries of the
first row of the square matrix and multiplying them "term-by-term" with the
entries of the column matrix and then adding those products.

4
The lower entry of the product is calculated by, first, taking the entries of the
second row of the square matrix and multiplying them "term-by-term" with
the entries of the column matrix and then adding those products.

3 7  8  59 
For example,    =  
5 12  5 100 

3 × 8 + 7 × 5 =59
5 × 8 + 12 × 5 =100

3 7  18  68 
and    =  
5 12   2  114 

3 ×18 + 7 × 2 = 68
5 ×18 + 12 × 2 =114

Multiplication is defined similarly for higher dimension square and column

matrices. For example, for 3 × 3 matrices

a b c   x  ax + by + cz 
d e f   y =  dx + ey + fz  .
    
 g h i   z   gx + hy + iz 

 1 0 7   5   89 
For example,  −3 4 9   −3 =  81 
    
12 −7 5  12  141

1× 5 + 0 × −3 + 7 ×12 = 89
−3 × 5 + 4 × −3 + 9 ×12 = 81
12 × 5 + −7 × −3 + 5 ×12 = 141

Etc.

5
To multiply two square matrices of the same dimension, we just do the
multiplication one column at a time.

3 7  8 18  59 68 
For example,   = .
5 12  5 2  100 114 

3 7  8  59  3 7  18  68 
   =   and    =  .
5 12 5 100 5 12  2  114

6
 Hill's Cipher

What is usually referred to as the Hill cipher is only one of the methods that
Hill discussed in his papers, and even then it is a weakened version. We will
comment more about this later, but first we will consider what is usually
called the Hill cipher.

The Hill cipher is a multiplicative cipher. It uses matrix multiplication to

transform blocks of plaintext letters into blocks of ciphertext.

Here is an example that encrypts digraphs.

Consider the following message:

Herbert Yardley wrote The American Black Chamber.

Break the message into digraphs:

he rb er ty ar dl ey wr ot et he am er ic an bl ac kc
ha mb er

(If the message did not consist of an even number of letters, we would place
a null at the end.)

Now convert each pair of letters to its number-pair equivalent. We will use
our usual a = 01, …, z = 26.

8 5 18 2 5 18 20 25 1 18 4 12 5 25 23 18 15 20 5 20 8 5
1 13 5 18 9 3 1 14 2 12 1 3 11 3 8 1 13 2 5 18

3 7 
Now we encrypt each pair using the key which is the matrix  .
5 12 

Make the first pair of numbers into a column vector (h (8) e (5)), and
multiply that matrix by the key.

3 7  8  59 
5 12  5 = 100 
    

7
Of course, we need our result to be mod 26

 59   7 
100  ≡  22  mod 26
   

The ciphertext is G (7) V (22).

For the next pair r (18) b (2),

3 7  18 16 
5 12   2  ≡ 10  mod 26 ,
    

and 16 corresponds to P and 10 corresponds to J. Etc.

Doing this for every pair we obtain

GVPJKGAJYMRHHMMSCCYEGVPEKGVCWQLXXOBMEZAKKG

What makes the Hill cipher a block cipher is that each plaintext string is
encrypted “at once:” the change of one letter in a plaintext block is likely to
change all of the letters in the ciphertext block. Notice, for example, that
changing the plaintext block he to ie changes the ciphertext block from GV
to JA.

Decryption

Of course, we need a procedure for decrypting this. Just like for the
multiplicative ciphers, we cannot use all matrices as keys because we cannot
undo the multiplication for all matrices.

To go from plaintext to ciphertext in the first example above we did

 3 7  8   7 
5 12  5 ≡  22  mod 26
    

Now we want to undo this; we want to find a matrix so that

8
  ? ?   7  8 
? ?  22  ≡ 5 mod 26
    

? ? 
i.e, we want to find a matrix   so that
? ? 

 ? ?   3 7  8  8 
? ? 5 12  5 ≡ 5 mod 26
     

? ?   3 7  8 
We want   5 12  to leave 5 unchanged.
 ? ?    

Matrix Inverse

3 7 
The matrix we are looking for is called the inverse of   and is
5 12 
−1
3 7 
denoted   .
 5 12 

 d −b 
−1
a b   ad − bc ad − bc 
It is easy to verify that   =  .
c d   −c a 
 ad − bc ad − bc 

 d −b 
−1
 a b   a b   ad − bc ad − bc   a b  1 0 
The=
product     = 
a   c d  0 1 
which
 c d   c d   −c
 ad − bc ad − bc 

is called the identity matrix because the effect of multiplying a matrix by it

is to leave the other matrix unchanged. (It is like multiplying a number by
1.)

9
 a b
Notice that to calculate the inverse of the matrix   we must be able to
 c d 
divide by ad – bc; i.e., we must have a multiplicative inverse for ad – bc.
Because we are working modulo 26, that means that ad – bc must be one of
1, 3, 5, 7, 9, 11, 15, 17, 19, 21, 23, or 25. Otherwise, the multiplication
cannot be undone; therefore, encryption cannot be undone.

10
 Determinant

a b 
ad – bc is called the determinant of   . Notice that the determinant of
c d 
a 2 × 2 is just the product down the upper left to lower right diagonal minus
the product down the upper right to lower left diagonal. For a matrix to have
an inverse modulo 26, the determinant of the matrix must be 1, 3, 5, 7, 9, 11,
15, 17, 19, 21, 23, or 25 modulo 26. To be able to undo multiplication by a
matrix mod 26, the determinant of the matrix must be 1, 3, 5, 7, 9, 11, 15,
17, 19, 21, 23, or 25 modulo 26. For a matrix to be a key for a Hill cipher,
the determinant of the matrix must be 1, 3, 5, 7, 9, 11, 15, 17, 19, 21, 23, or
25 modulo 26.

3 7 
The determinant of   is 3 × 12 − 7 × 5 = 1 ≡ 1mod 26 . So, the inverse
5 12 
−1
3 7  3 7  12 −7  12 19 
of  =
 5 12 
is  −5 3  ≡  21 3  mod 26 . This is a special
 5 12       
case because the determinant is 1.

Here is an example of finding the inverse of a 2 × 2 matrix when the

determinant is not 1.

9 4 
The determinant of   is 9 × 7 − 4 × 5 = 63 − 20 = 43 ≡ 17 mod 26 .
5 7 
Because 17 has a multiplicative inverse modulo 26, this matrix has an
inverse. The inverse of the matrix is

7 −4 
 17 17 
  mod 26 .
 −5 9 
 17 17 

Dividing by 17 modulo 26 is the same as multiplying by the multiplicative

inverse of 17 modulo 26. Recall that the multiplicative inverse of 17 is 23
modulo 26. So, the inverse of the matrix is

11
 7 −4 
 17 17   7 × 23 −4 × 23
  mod 26 ≡   mod 26
 −5 9   −5 × 23 9 × 23 
 17 17 

 161 −92   5 12 
≡  mod 26 ≡   mod 26
 −115 207  15 25 

Calculating the determinant of an n × n matrix with n > 2 is more difficult.

The pattern used for a 2 × 2 matrix is a very special case. Usually
calculators and computer algebra systems are able to calculate determinants.

Similarly, calculating the inverse of an n × n matrix with n > 2 differs from

calculating the inverse of a 2 × 2 matrix. Again, usually calculators and
computer algebra systems are able to calculate inverses.

12
 Decryption

We return to the earlier example. Encrypting

Herbert Yardley wrote The American Black Chamber.

3 7 
using the key   resulted in the ciphertext
5 12 

GVPJKGAJYMRHHMMSCCYEGVPEKGVCWQLXXOBMEZAKKG

12 19
We use the inverse of the key   to decrypt GV, which is the first
 21 3 
digraph of the ciphertext.

G orresponds to 7, and V corresponds to 22.

12 19   7  8 
 21 3   22  ≡ 5 mod 26
    
h (8) e (5).

In a similar manner, we can decrypt the remainder of the ciphertext.

13
 Hill ciphers that encipher larger blocks

Notice that the multiplicative cipher is just the 1×1 case of the Hill cipher;
individual letters are enciphered one at a time.

2 × 2 invertible matrices modulo 26 (an invertible matrix is a matrix that has

an inverse) can be used to encipher digraphs. 3 × 3 invertible matrices
modulo 26 can be used to encipher trigaphs. 4 × 4 invertible matrices
modulo 26 can be used to encipher blocks of 4 letters. Etc.

Finding keys is pretty much a trial and error process. That means that it can
be very difficult to find a key for encrypting large blocks.

14
 Known plaintext attack

It can be difficult to cryptanalyze a Hill cipher using a ciphertext only attack

(we will do one later), but it is easy to break using a known plaintext
attack. A known plaintext attack means that we know a bit of ciphertext
and the corresponding plaintext – a crib. This is not an unusual situation.
Often messages have stereotypical beginnings (e.g., to …, dear …) or
stereotypical endings (e.g, stop) or sometimes it is possible (knowing the
sender and receiver or knowing what is likely to be the content of the
message) to guess a portion of a message.

A known plaintext attack works well against a Hill cipher because there is
an algebraic relationship between plaintext and ciphertext (as there is for
Caesar ciphers, multiplicative ciphers, and affine ciphers) and the
relationship is linear (as it was for Caesar ciphers, multiplicative ciphers, and
affine ciphers). The algebraic relationship between plaintext and ciphertext
combines with the relationship between some plaintext and ciphertext letters
given by the crib and results in a system of algebraic equations that usually
can be solved because it is linear (as we did for Caesar ciphers,
multiplicative ciphers, and affine ciphers).

For a 2 × 2 Hill cipher, if we know two ciphertext digraphs and the

corresponding plaintext digraphs, we can easily determine the key or the key
inverse. Assume that we know that the plaintext of our ciphertext message
that begins WBVE is inma. Because WB corresponds to in
 e f   23  9   e f   22 13
    =   , and because VE corresponds to ma    =   .
 g h   2  14   g h  5   1 
This results in two sets of linear congruences modulo 26:

23e + 2 f =9
22e + 5 f =
13

and

23g + 2h =14
22 g + 5h =1

We solve the systems modulo 26 using Mathematica.

15
Solve[23e + 2f == 9 && 22e + 5f == 13, {e, f}, Modulus -> 26]

{{e->1,f->19}}

Solve[23g + 2h == 14 && 22g + 5h ==1, {g, h}, Modulus -> 26]

{{g->20,h->11}}

 1 19
Again (with a lot less assuming) we find that the key inverse is  .
 20 11

One thing that should be learned from this example is that a cryptosystem
should have some nonlinear piece.

16
 Two More Examples of a Known Plaintext Attack

Here are two more examples of cryptanalyzing a Hill cipher with a known
plaintext attack. Each example is done by hand – without using
Mathematica. In example one, there is no need to reduce the modulus; in
example two the modulus must be reduced.

Example one:

Ciphertext: FAGQQ ILABQ VLJCY QULAU STYTO JSDJJ

PODFS ZNLUH KMOW

We are assuming that this message was encrypted using a 2 × 2 Hill cipher
and that we have a crib. We believe that the message begins “a crib.”

ac ri
[1, 3] [18, 9]
[6, 1] [7, 17]
FA GQ

17
We could either solve for the key or the key inverse. To solve for the key,
we would solve

 a b  1 6 
 c d   3 =  1 
    

and

 a b  18  7 
 c d   9  = 17 
    

To solve for the key inverse, we would solve

e f  6  1
g =
 h  1  3

and

e f   7  18
g =
 h  17   9 

We will solve for the key.

 a b  1 6 
 c d  3 = 1  represents two linear equations:
    

a +3b =6
c +3d =1

a b  18 7
and     =   represents
 c d   9  17 

18a +9b = 7
18c +9d =17

Now we solve the following linear congruences mod 26.

18
  a +3b = 6  c +3d = 1
 and 
18a +9b =7 18c +9d =17

 a +3b = 6
We will solve the pair of congruences  first.
18a +9b =7

To eliminate an unknown, multiply congruence 1 by 3

 3a +9b =18

18a +9b = 7

and subtract congruence 2 from congruence 1.

−15a =
11

Modulo 26, -15 is 11.

11a = 11

Divide by 11 to obtain a.

a =1

Now substitute this in congruence 1.

1 + 3b =
6

3b = 5

The multiplicative inverse of 3 is 9 modulo 26.

b = 9 × 3b = 9 × 5 = 45 = 19 mod 26

So, the key looks like

1 19 
c d 
 

19
  c +3d = 1
Now solve the system 
18c +9d =17

 3c +9d = 3

18c +9d =17

15c = 14

c =7 ×15c =7 ×14 =98 =20 mod 26

20 + 3d =
1

3d =−19 =
7 mod 26

d = 9 × 3d = 9 × 7 = 63 = 11mod 26

1 19 
The key is  .
 20 11

Example two:

We are assuming that we have a ciphertext message was that encrypted

using a 2 × 2 Hill cipher and that we have a crib. We believe that ciphertext
UKJN corresponds to plaintext word.

wo rd
[23, 15] [18, 4]
[21, 11] [10, 14]
UK JN

The two systems of congruences are:

23a +15b =21 23c +15d =11

 and 
18a +4b =10 18c +4d =
14

20
We will solve the system on the left.

To eliminate an unknown, multiply congruence number 1 by 4 and

congruence number 2 by 15 both modulo 26.

14a +8b =6

10a +8b =20

Subtract the second congruence from the first.

4a =−14 =
12 mod 26

This congruence corresponds to the equation 4a= 12 + 26k , 4a is 12 plus a

multiple of 26. Notice that 2 divides the coefficient of a, the constant 12,
and the modulus 26. We reduce the modulus by dividing by 2.

2a= 6 + 13k

and we have a congruence modulo 13.

2a = 6 mod13

This congruence does not have a common factor among the coefficient, the
constant, and the modulus.

Here are the multiplicative inverses of the integers modulo 13:

Number 1 2 3 4 5 6 7 8 9 10 11 12
Multiplicative inverse 1 7 9 10 8 11 2 5 3 4 6 12

To find a, multiply 2a = 6 mod13 by the multiplicative inverse of 2, which is

7.

a = 7 × 2a = 7 × 6 = 42 = 3mod13

So, a is 3 modulo 13. But, there are two integers mod 26 that are 3 mod 13,
namely, 3 and 3 + 13 = 16. So, there are two possible values for a.

If a = 3,

21
 18 × 3 = 4b = 10

54 + 4b =
10

2 + 4b =
10

4b = 8 mod 26

2b = 4 mod13

b = 7 × 2b = 7 × 4 = 26 = 2 mod13

So, b=2 or b = 2+13 = 15 modulo 16.

If a = 16,

18 ×16 + 4b =
10

288 + 4b =
10

2 + 4b =
10

which yields the same solutions for b.

Here are the 4 possible solutions for a and b.

=a 3=b 2
=a 3=b 15
=a =
16 b 2
=a =
16 b 15

23c +15d =11

Now solve  .
18c +4d =
14

14a +8b =18


10a +8b = 2

4c = 16 mod 26

22
 2c = 8 mod13

c = 7 × 2c = 14c = 7 × 8 = 56 = 4 mod13

So, c = 4 or c = 4 + 13 = 17 modulo 26.

If c = 4,

18 × 4 + 4d =
14

20 + 4d =
14

4d =−6 =20 mod 26

2d = 10 mod13

d =7 × 2d =7 ×10 =5 mod13

So, d = 5 or d = 5 + 13 = 18 modulo 26.

If c = 17,

18 ×17 + 4d =
14

20 + 4d =
14

and we are led to the same solutions for d.

=c 4=d 5
=c 4=d 18
=c =
17 d 5
=c =
17 d 18

There are 16 possible 2 × 2 matrices that could be the key.

23
 3 2 3 2   3 2 3 2
4 5   4 18 17 5  17 18
      
3 15  3 15  3 15  3 15
4 5   4 18 17 5  17 18
  
16 2  16 2  16 2  16 2 
4 5   4 18 17 5  17 18
  
16 15 16 15 16 15 16 15
4 5   4 18 17 5  17 18
  

First, calculate the determinant of each. Any matrix that does not have an
invertible determinant modulo 26 (i.e., the determinant is not one of 1, 3, 5,
7, 9, 11, 15, 17, 19, 21, 23, 25 modulo 26) can be eliminated. Then try to
decipher the messages with each of the remaining messages. The matrix that
yields plaintext is the key.

To break a Hill cipher with a 2 × 2 key requires determining four entries –

the four entries of the key or the four entries of the key inverse. We can do
that if we know the correspondence between plaintext and ciphertext for two
digraphs because the correspondences will permit us to set up two systems
of congruences – each system has two congruences of two unknowns.

To break a Hill cipher with a n × n key requires determining n 2 entries – the

n 2 entries of the key or the n 2 entries of the key inverse. We can do that if
we know the correspondence between plaintext and ciphertext for n
independent n-graphs because the correspondences will permit us to set up n
systems of congruences – each system has n congruences of n unknowns.

The reason that we can solve these systems of congruences is because they
are linear. The solutions of linear systems of equations of congruences is
well-understood.

24
 Ciphertext Attack

Here is a ciphertext that is known to be enciphered with a Hill cipher.

wbvec itxwb mphsr hytyw gmqdq egxyf yncta zdkyi eenin zkygh
yntgb pbpkl azfgy ikkru drzcp aaaci fueqg ywbuu urozm vfgmy
vkwwo zbpyn ezsbg jfynz yvmeo zctiu ghfgu aekds ayicc tkrus
xgbpz cufve lvsjg lklls vefyt onmdk

The first thing to be determined would be the size of the blocks. If the key
were an n × n matrix, then n must divide the number of letters in the
ciphertext. This ciphertext has 180 letters. There are many possibilities for
n, but let us assume that it was encrypted using a 2 × 2 key. (That’s a really
good assumption.)

Because such a key encrypts digraphs, we might begin by looking at digraph

frequencies.

25
Here are the digraphs that appear more than once and their frequencies:
Digraph Frequency

bp 4
yn 4
ct 3
fg 3
fy 3
oz 3
ve 3
wb 3
yi 3
zc 3
aa 2
az 2
ci 2
dk 2
gb 2
gh 2
gm 2
gy 2
hy 2
kl 2
kr 2
ky 2
nz 2
ru 2
uu 2
yt 2
yv 2
yw 2

If we’re lucky the most common plaintext digraph th will correspond to

(one of) the most common ciphertext digraph(s). BP and YN each appear 4
times in the ciphertext. Let’s assume that ciphertext BP corresponds to
plaintext th. (Another really good assumption.)

We could try to determine the key or the key inverse. Because we are trying
to determine the plaintext, let’s try to directly determine the key inverse. We
e f 
want to find a 2 × 2 matrix   that is the inverse of the key. If we are
g h 
correct that B(2)P(16) corresponds to t(20)h(8), then

e f   2   20
   =  
g h  16  8 

26
This corresponds to two linear equations:

2e + 16 f =
20
2 g + 16h =
8

Because this Hill cipher (we assume) encrypts digraphs, the key inverse is a
2 × 2 matrix. The key inverse has 22 = 4 entries e, f, g, and h that must be
determined. We would like to have four equations – two involving e and f
and two involving g and h.

If we knew another plaintext/ciphertext digraph correspondence, we would

have the other two equations that we need. Perhaps, the next most common
ciphertext digraph YN corresponds to the next most common plaintext
digraph he. (But, it doesn’t.)

We could try assuming that YN corresponds to another common digraph, but

here is another technique.

The most common letter that follows plaintext th is e. We might examine

the digraphs that follow BP in the ciphertext and assume that the next
ciphertext digraph corresponds to plaintext e_. We notice that we have
BP KL, BP BP, BP YN, and BP ZC. If we are correct that BP
corresponds to th, the second pair of digraphs corresponds to plaintext th
th. In each of the other cases, we will assume that the two ciphertext
digraphs correspond to th e_. Making this assumption, we should be
correct more than half the time.

e f  11 5
So, if KL corresponds to e_,     =   which yields the equation
 g h  12 *
 e f   25 5
11e + 12 f =
5 . If YN corresponds to e_,     =   which yields the
 g h  14  *
e f   26 5
equation 25e + 14 f =
5 . If ZC corresponds to e_,     =   which
 g h   3  * 
yields the equation 26e + 3 f =
5.

Each of these can be solved simultaneously with,12e + 16 f =

20 which was
obtained by assuming that BP corresponds to th. All of the solving,

27
however, is to be done modulo 26. We may use whatever techniques we
know for solving systems of linear equations provided that we divide only
when division is possible – we can divide by only 1, 3, 5, 7, 9, 11, 15, 17,
19, 21, 23, and 25. We will use Mathematica to solve the equations.

Solve[2e+16f==20&&11e+12f==5,{e,f},Modulus ->
26]

{{e->1,f->6+13 C[1]}}

Solve[2e+16f==20&&25e+14f==5,{e,f},Modulus -> 26]

{{e->1,f->6+13 C[1]}}

Solve[2e+16f==20&&26e+3f==5,{e,f},Modulus -> 26]

{{e->1+13 C[1],f->19}}

Each system of congruences has two solutions modulo 26. e = 1 and f = 19

is common to all of the pairs of solutions. That would happen if in each of
these three cases th were followed by e_. Let us assume that is the case.
(That’s another really good assumption.) We could later try the other
possibilities if needed.

 1 19
So, we believe that the key inverse is  .
g h 

28
We have one more congruence: 2 g + 16h = 8mod 26 . It is possible to solve a
congruence of the form ax + by = c mod n provided that the greatest common
divisor of a, b, and n also divides c. In our case, the greatest common
divisor of a=2, b=16, and n=26 is 2 which does divide c=8. It is necessary
to reduce the modulus; remove the factor of 2 to get g + 8h = 4mod13 . Then
rearrange the terms to get g= 4 − 8h mod13 . Modulo 13, the possible values
of h are 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, and 12.

For example, if h = 3, g = 6 mod 13. But, we are ultimately interested in

what happens modulo 26. 6 and 6 + 13 = 19 are congruent mod 13, but they
are not congruent mod 26. So, each solution mod 13 becomes two solutions
mod 26.

h g mod 13 g mod 26
0 4 4, 17
1 9 9, 22
2 1 1, 10
3 6 6, 19
4 11 11, 24
5 3 3, 22
6 8 8, 21
7 0 0, 13
8 5 5, 18
9 10 10, 23
10 2 2, 15
11 7 7, 20
12 12 12, 25

The determinant of the key inverse must be one of 1, 3, 5, 7, 9, 11, 15, 17,
 1 19
19, 21, 23, or 25 mod 26. So, try each of these pairs of g and h in  
g h 
and calculate the determinant mod 26. Again, we use Mathematica.

In[8]:= Mod[Det[{{1, 19}, {4, 0}}], 26]

Out[8]= 2

29
Mod[Det[{{1, 19}, {17, 0}}], 26]

15

Mod[Det[{{1, 19}, {9, 1}}], 26]

12

Mod[Det[{{1, 19}, {22, 1}}], 26]

25

Mod[Det[{{1, 19}, {1, 2}}], 26]

Mod[Det[{{1, 19}, {10, 2}}], 26]

20

Mod[Det[{{1, 19}, {6, 3}}], 26]

19

Mod[Det[{{1, 19}, {19, 3}}], 26]

Mod[Det[{{1, 19}, {11, 4}}], 26]

Mod[Det[{{1, 19}, {24, 4}}], 26]

16

Mod[Det[{{1, 19}, {3, 5}}], 26]

Mod[Det[{{1, 19}, {22, 5}}], 26]

30
Mod[Det[{{1, 19}, {8, 6}}], 26]

10

Mod[Det[{{1, 19}, {21, 6}}], 26]

23

Mod[Det[{{1, 19}, {0, 7}}], 26]

Mod[Det[{{1, 19}, {13, 7}}], 26]

20

Mod[Det[{{1, 19}, {5, 8}}], 26]

17

Mod[Det[{{1, 19}, {18, 8}}], 26]

Mod[Det[{{1, 19}, {10, 9}}], 26]

Mod[Det[{{1, 19}, {23, 9}}], 26]

14

Mod[Det[{{1, 19}, {2, 10}}], 26]

24

Mod[Det[{{1, 19}, {15, 10}}], 26]

11

Mod[Det[{{1, 19}, {7, 11}}], 26]

31
Mod[Det[{{1, 19}, {20, 11}}], 26]

21

Mod[Det[{{1, 19}, {12, 12}}], 26]

18

Mod[Det[{{1, 19}, {25, 12}}], 26]

 1 19  1 19 1 19 1 19

The possible key inverses are  ,  ,  ,  ,
17 0   22 1  1 2 6 3
 1 19   1 19  1 19 1 19 1 19  1 19  1 19
 ,  ,  ,  ,  ,  , ,
11 54  22 5   21 6  0 7  5 8  10 9  15 10
 1 19  1 19
  , and  .
 20 11  25 12

We have reduced the problem to checking 13 possible key inverses. We try

 1 19
to decrypt the ciphertext with each possible inverse.   is the correct
 20 11
key inverse.

32
 Other Hill Ciphers

Hill’s papers contain techniques that are much more secure than the
technique that we have called the Hill cipher. Hill’s papers include ciphers
that are nonlinear.

One technique used by Hill is to do a (nonlinear) simple substitution cipher –

a permutation -- prior to the matrix multiplication. Hill uses the following
substitutions:
a b c d e f g h i j k l m n o p q r s t u v w x y z
5 23 2 20 10 15 8 4 18 25 0 16 13 7 3 1 19 6 12 24 21 17 14 22 11 9

For example, th becomes 24 4 and then

3 7   24   22 
5 12   4  = 12 
    

V (22) L (12).

Another technique used by Hill is similar to what we did when we went

from the multiplicative cipher (C = mp) to the affine cipher (C = mp + b) by
adding a shift. Multiplicative ciphers are linear ciphers; affine ciphers are
not linear ciphers. Hill adds a shift to what we have called the Hill cipher.
For example, (using a = 1, …, z = 26) to encrypt h (8) e (5)

3 7  8  6  13
5 12  5 +  20  =
 
      16 

M (13) P (16).

Hill’s 1929 and 1931 papers include other generalizations of the Hill cipher.

33
 Size of the Keyspace

Multiplicative ciphers have a very small keyspace; the key must be one of 1,
3, 5, 7, 9, 11, 15, 17, 19, 21, 23, 25. How large is the keyspace for a Hill
cipher?

There are 264 = 456976 2 × 2 matrices having entries modulo 26; i.e., each
entry must be 0, 1, …, 25. But, recall that for a matrix to be usable as a key
for a Hill cipher the matrix must have an inverse. How many of these 2 × 2
matrices are invertible? This is answered for n × n matrices in “On the
Keyspace of the Hill Cipher” by Overby, Traves, and Wojdylo in
Cryptologia, January 2005; there are 157248 possible 2 × 2 keys.

Hill ciphers use two keys: one key is used for encryption and a second key
(the key inverse) is used for decryption. Of course, anyone who knows
some elementary linear algebra can construct the key inverse from the key,
but the encryption and decryption keys are not the same – except in certain
cases. Hill, in his second paper, discusses using involutory matrices
(matrices that are self-inverse) as keys.

 0 1 25
 4 22 4  is involutory.
 
 3 22 4 

Using involutory keys would make encryption and decryption completely

symmetric, but this significantly restricts the number of keys (see the
previously cited article in Cryptologia.)

34
 Hill ciphers with n × n keys form a group

Encrypting with a Hill cipher and re-encrypting with another key of the
same size does not improve security because the Hill ciphers with n × n keys
form a group.

For example, if we encrypted digraphs with a Hill cipher using the key
3 7 
  (which has determinant 1 modulo 26) and then encrypted that
 5 12 
9 4 
ciphertext using a Hill cipher with key   (which has determinant 17
 5 7 
modulo 26), the result would be the same as encrypting once with the
9 4 3 7   47 111  21 7 
key  =  =    mod ulo 26 . Because “the
 5 7  5 12   50 119   24 15 
determinant of a product is the product of the determinants” (even modulo
 21 7 
26), the determinant of   is17 ×1 = 17 mod ulo 26 ; so, it is a valid Hill
 24 15 
cipher key.

The point is, you have only one shot at using a Hill cipher – re-encrypting
does not improve security.

35
 Was the Hill cipher ever used?

Hill’s papers contributed two important ideas to cryptology. First, they freed
cryptology from encrypting just single letters and digraphs – they showed
that encryption of blocks of more than two letters was possible. And, Hill’s
papers showed the close connection between cryptology and mathematics.
This connection was emphasized by A.A. Albert in an address to the
American Mathematical Society in 1941.

But, was the Hill cipher ever used?

Hill’s cipher is a nice application of matrices, but matrix multiplication is

probably not easily done by soldiers who are in the trenches and watching
artillery shells flying overhead.

As a cipher to encrypt digraphs, Hill’s cipher is harder to use and weaker

than the Playfair cipher.

… Hill’s cipher system itself saw almost no practical use …

… the Hill system has served as a U.S. governmental cryptosystem in

only one minor capacity – to encipher the three-letter groups of radio
call-signs. The Codebreakers by David Kahn

36
Exercises

3 7 
1. Use a Hill cipher with key   to encrypt the following message.
5 12 

Agnes Driscoll worked for NSA.

2. The following message was encrypted with a Hill cipher with key
3 7 
5 12  . Decrypt the message.
 

ZKYZR QHBDM JMPVX WLCGF MIXGM PKBUZ

FHPCI XZTIW

3. Which of the following matrices can be used as keys for a Hill cipher?

5 2
3a.  .
7 3 

 5 12 
3b.  .
15 25 

 20 2 
3c.  .
 5 4 

 5 8
3d.  .
12 3

 21 13
3e.  
 7 16 

37
  2 7 19 
4. Use a Hill cipher with key  0 5 8  to encrypt the following
 
1 3 7 
message: enigma.

5 6 4 1
2 1 0 3
5. Use a Hill cipher with key   to encyrpt the following
1 8 9 2
 
2 4 6 7
message: united states.

6. Find a 2 × 2 matrix that can be used as a key for a Hill cipher.

5 8
7. What is special about the Hill cipher key   ? Does this make this
10 21
matrix a good or bad choice for a key?

 6 3
8. A message is first encrypted with a Hill cipher with key   and then
 7 8
3 2 
encrypted again with key   . What is the resulting cipher?
8 5 

9. Known plaintext attack on the Hill cipher. Find the key for the following
ciphertext message that was enciphered with a Hill cipher.

VRAAU OTNLK NJWVJ QJXXY BEOLW CVRYK FOYPQ

TWVMP ALUEA ACWWE GB

The plaintext message begins “The Riddle.”

38
10. For the very (very) brave – a ciphertext attack on the Hill cipher.

TZZOK HMOTZ MOINY FTWCO UWINH CAGZH AZXME

ICZVH ZOTWE IUGGG AETWE ICZVH ZOYKX ZFAMW
PGWQQ JTZGT YRFAI KTWEI HIQTN ZAGVM YKXZP
GWQQJ ZCHNE ITZXG IKTWE IVUOZ XUXBQ DPTPS
WQZHA ZXMEI CZVHZ OYKXZ FAMWP GWQQJ ELTZQ
POZRF

11. Another ciphertext attack on the Hill cipher.

xrqsx ibkfy lawcc jrohm ouyyl mrqgi ucscc ahakc

zwuhg axroc bipwe zatqd eqrmh zgtmv ygoyq qlmqd
kbpyd dqqcj glhka pbkae clxru cuhbg wtetx riymt
ezdyd dhksj opgia piwzw fmzwa egpgi jpqxn oaady

12. You know that you are intercepting messages that are encrypted with a
2 × 2 Hill cipher. You are able to trick one of the parties of the
communications to send the plaintext digraphs az and za. You are able to
determine that az is enciphered as OJ and za is enciphered as YI.
Determine the encryption matrix and the decryption matrix.

13. Consider the affine version of the Hill cipher. First encrypt a message
3 7 
by multiplying by the matrix   and then adding to the result the
5 12 
6 9 4 
matrix   . Encrypt again by multiplying by the matrix 5 7  and
 20   
3
adding the matrix   . What is the resulting cipher?
17 

14. Consider the affine version of the Hill cipher. Would re-encrypting
using this cryptosystem increase security?

39
15. Would using an involutory key reduce the security of a Hill cipher?

40