Peculiarities of nilpotent matrices

School of Mathematics, Statistics and Applied Mathematics

Research Day - April 30 2015

Rachel Quinlan

April 30, 2015

Nilpotent Matrices

A n × n matrix with entries in a field F is called nilpotent if it has

the following equivalent properties.

Ak = 0 for some positive integer k.

An = 0.
All eigenvalues of A are equal to zero.
The characteristic polynomial of A is xn .

Example

Every strictly upper triangular matrix in Mn (F) is nilpotent.

Strictly upper triangular matrices

Every nilpotent matrix is similar to

a strictly upper triangular matrix.
 
A subspace of Mn (F) in which 0 ? ? ?
every element is nilpotent can have

 0 0 ? ? 

dimension at most n(n−1)
 
2 . A  0 0 0 ? 
nilpotent space of this maximum 0 0 0 0
dimension is similar to SUTn (F)
(Gerstenhaber, 1958).

It is not true that every space of nilpotent matrices is

triangulable. The problem of characterizing irreducible spaces
of nilpotent matrices over fields remains open.
What about a (skew-)symmetric nilpotent space?

The trace of the square of a symmetric or skew-symmetric matrix

is ± the sum of the squares of the entries. Over a real field, the
only nilpotent symmetric or skew-symmetric matrix is the zero
matrix (same for complex Hermitian or skew-Hermitian matrices).
What about a (skew-)symmetric nilpotent space?

The trace of the square of a symmetric or skew-symmetric matrix

is ± the sum of the squares of the entries. Over a real field, the
only nilpotent symmetric or skew-symmetric matrix is the zero
matrix (same for complex Hermitian or skew-Hermitian matrices).

Theorem 1 (Meshulam and Radwan, 1998)

The maximum possible dimension of a space of symmetric

matrices in Mn (C) is

m2 if n = 2m
m2 + m if n = 2m + 1.

For skew-symmetric matrices, the maximum dimension is

m2 − m if n = 2m
m2 if n = 2m + 1.
Adapted from Meshulam and Radwan . . .

The arguments of Meshulam and Radwan can be adapted to give

the following theorem, in which the stated dimension bounds can
be attained if char F 6= 2 and −1 is a square in F.

Theorem 2

Let F be a field with char F 6= 2. Let NS and NA respectively

denote nilpotent spaces of symmetric and skew-symmetric matrices
in Mn (F). Then

dim NS ≤ m2 if n = 2m
dim NS ≤ m2 + m if n = 2m + 1
dim NA ≤ m2 − m if n = 2m
dim NA ≤ m2 if n = 2m + 1
Spaces achieving these bounds

Suppose charF 6= 2 and write i for a square root of −1 in F. For

n = 2m + 1, write
 
Im×m Im×m 0m×1
Z =  iIm×m −iIm×m 0m×1  .
 
01×m 01×m 1

Symmetric nilpotent space of dimension m2 + m:

  

 A B E(m×1) A ∈ Mm (F)  
Z −1 Sn (F)Z =  C AT F(m×1)  : B, C ∈ Sm (F) .
 
 −2F T −2E T

D D∈F



The subspace SN with A strictly upper triangular, C = 0 and

F = 0 is nilpotent of dimension m2 + m, and is similar to a
symmetric space.
Spaces achieving these bounds

Suppose charF 6= 2 and write i for a square root of −1 in F. For

n = 2m + 1, write
 
Im×m Im×m 0m×1
Z =  iIm×m −iIm×m 0m×1  .
 
01×m 01×m 1

Skew-symmetric nilpotent space of dimension m2 :

  
 A B E(m×1) 
 A ∈ Mm (F)
 
−1 T
Z An (F)Z =  C −A F(m×1)  : .

 2F T 2E T
 B, C ∈ Am (F) 
0 

The subspace AN with A strictly upper triangular, C = 0 and

F = 0 is nilpotent of dimension m2 , and is similar to a
skew-symmetric space.
Decomposition of a nilpotent space of maximum dimension

In the above construction we have SN ∩ AN = 0 and

( !)
Um×m Xm×(m+1)
SN + AN = ,
0(m+1)×m L(m+1)×(m+1)

where U is strictly upper triangular, L is strictly lower triangular

(and X can be anything). Thus SN ⊕ AN is a space of nilpotent
matrices in Mn (F) of maximum possible dimension n(n−1) 2 .

Theorem 3

Let F be a field of characteristic different from 2, in which −1 is a

square. Then for every positive integer n, Mn (F) contains a
subspace of nilpotent matrices of dimension n(n−1)2 which is the
direct sum of a symmetric nilpotent space and a skew-symmetric
nilpotent space.
The strange case of characteristic 2

The story so far: for a field F whose characteristic is not 2, the

dimension of a nilpotent subspace of skew-symmetric elements
cannot exceed m2 − m if n = 2m is even, or m2 if n = 2m + 1 is
odd. These bounds are attainable if −1 is a square in F.
The corresponding numbers for fields of characteristic 2, which are
different at least for n odd:

at least equal to m2 − m in the even case n = 2m;

at least m2 + m in the odd case n = 2m + 1.
A special nilpotent space in A5 (F)

Now suppose that char F = 2. For t = 1, . . . , 5, let Bt be the

element of A5 (F) whose entries for i 6= j are given by
(
1 if t ∈ {i, j}
(Bt )ij =
0 if t 6∈ {i, j}

So for example
   
0 1 1 1 1 0 1 0 0 0
 1 0 0 0 0   1 0 1 1 1 
   
B1 =  1 0 0 0 0  , B2 =  0 1 0 0 0 .
   
   
 1 0 0 0 0   0 1 0 0 0 
1 0 0 0 0 0 1 0 0 0
A special nilpotent space in A5 (F)

Now suppose that char F = 2. For t = 1, . . . , 5, let Bt be the

element of A5 (F) whose entries for i 6= j are given by
(
1 if t ∈ {i, j}
(Bt )ij =
0 if t 6∈ {i, j}

Let R5 = hB1 , B2 , B3 , B4 , B5 i, a subspace of A5 (F) of dimension

4.
  

 0 a+b a+c a+d a 

 
 a + b 0 b+c b+d b

   

 
R5 =  a + c b + c 0 c+d c  : a, b, c, d, e ∈ F .
 
   




 a + d b + d c + d 0 c  



a b c d 0
 
Peculiar properties of R5

  

 0 a+b a+c a+d a 

 
 a + b 0 b+c b+d b

  

 
R5 =  a + c b + c 0 c+d c  : a, b, c, d ∈ F .
 
  




 a+d b+d c+d 0 d  



a b c d 0
 

Every element of R5 is nilpotent.

For every element A of A5 and every B ∈ R5 , the matrices A
and A + B have the same characteristic polynomial.
In particular if N is any nilpotent subspace of A5 (F), then
N + R5 is also a nilpotent subspace.
A nilpotent space of dimension m2 + m in A2m+1 (F)

For any odd n = 2m + 1 we can define Rn ⊆ An (F) as above.

For odd k ≤ n think of Ak (F) as the subspace of An (F)
consisting of those elements with zeros outside the upper left
k × k region, so we can view Rk (F) as a subspace of Mn (F).
Then ANn := R3 + R5 + · · · + Rn is an irreducible nilpotent
subspace of An (F) of dimension 2 + 4 + · · · + 2m = m2 + m.
Example: n = 5, m = 2

AN5 (F) consists of all matrices of the form

 
0 a+b+c+d a+c+e c+f c

 a+b+c+d 0 b+d+e d+f d 

a+c+e b+d+e 0 e+f e .
 

 
 c+f d+f e+f 0 f 
c d e f 0