ORIGINAL ARTICLE

Roots of Matrices
Boonrod Yuttanan1 and Chaufah Nilrat2

Abstract
Yuttanan, B. and Nilrat, C.
Roots of Matrices
Songklanakarin J. Sci. Technol., 2005, 27(3) : 659-665

A matrix S is said to be an nth root of a matrix A if Sn = A, where n is a positive integer greater than or
equal to 2. If there is no such matrix for any integer n > 2, A is called a rootless matrix. After investigating
the properties of these matrices, we conclude that we always find an nth root of a non-singular matrix and
a diagonalizable matrix for any positive integer n. On the other hand, we find some matrix having an nth root
for some positive integer n. We call it p-nilpotent matrix.

Key words : roots of matrices, rootless matrix, nilpotent matrix, non-singular matrix,
diagonalizable matrix

1
Student in Mathematics, 2M.S.(Mathematics), Assoc. Prof., Department of Mathematics, Faculty of Science,
Prince of Songkla University, Hat Yai, Songkhla 90112 Thailand.
Corresponding e-mail: chaufah.n@psu.ac.th
Received, 18 May 2004 Accepted, 11 October 2004
Songklanakarin J. Sci. Technol. Roots of Matrices
Vol.27 No.3 May - Jun. 2005 660 Yuttanan, B. and Nilrat, C.

∫∑§—¥¬àÕ
∫ÿ≠√Õ¥ ¬ÿ∑∏“π—π∑å ·≈– ™àÕøÑ“ π‘≈√—µπå
√“°¢Õ߇¡∑√‘°´å
«.  ß¢≈“π§√‘π∑√å «∑∑. 2548 27(3) : 659-665
‡√“°≈à“««à“ S ‡ªìπ√“°∑’Ë n ¢Õ߇¡∑√‘°´å A ∂â“ Sn = A ‡¡◊ËÕ n ‡ªìπ®”π«π‡µÁ¡∫«°∑’Ë¡“°°«à“À√◊Õ‡∑à“°—∫ 2
∂Ⓣ¡à¡’‡¡∑√‘°´å S ·≈–®”π«π‡µÁ¡∫«° n ¥—ß°≈à“« ‡√“‡√’¬° A «à“‡¡∑√‘°´å∑’ˉ¡à¡’√“° À≈—ß®“°∑”°“√µ√«® Õ∫
 ¡∫—µ‘¢Õ߇¡∑√‘°´å‡À≈à“π’È æ∫«à“ ”À√—∫∑ÿ°®”π«π‡µÁ¡∫«° n ‡√“ “¡“√∂À“√“°∑’Ë n ¢Õ߇¡∑√‘°´å‰¡à‡Õ°∞“π·≈–
‡¡∑√‘°´å∑’˧≈⓬°—∫‡¡∑√‘°´å∑·¬ß¡ÿ¡‰¥â‡ ¡Õ πÕ°®“°π’ȇ√“¬—ßæ∫«à“¡’∫“߇¡∑√‘°´å∑’ˇ√“ “¡“√∂À“√“°∑’Ë n ‰¥â‡æ’¬ß
∫“ß®”π«π‡µÁ¡∫«° n ‡∑à“π—Èπ ‡¡∑√‘°´å¥—ß°≈à“«§◊Õ ‡¡∑√‘°´åæ’-π‘≈‚æ‡∑πµå
¿“§«‘™“§≥‘µ»“ µ√å §≥–«‘∑¬“»“ µ√å ¡À“«‘∑¬“≈—¬ ß¢≈“π§√‘π∑√å Õ”‡¿ÕÀ“¥„À≠à ®—ßÀ«—¥ ß¢≈“ 90110
r
An m×m matrix A is called nilpotent if A = 0 for some positive integer r > 2. Yood (2002) showed
m-1
that any nilpotent m×m matrix A such that A 0 is rootless. Such a matrix is called principal
nilpotent. After we finished reading this article, we raised the question of which matrices always have an
th th
n root for any positive integer n and which have an n root only for some positive integer n. In this
paper, we give the answer for these questions.

1. Roots of non-singular matrices

th
In this section, we prove that every non-singular matrix has an n root for any positive integer.
Before discussing on a non-singular matrix, we start with a property of upper triangular matrices.

Lemma 1.1 If A = [aij]m×m is an upper triangular matrix, then so is A = [αij]m×m and

n


 0 if i > j,

α ij =  if i = j,
n
aii
 ∑ a a ...a if i < j.
i≤k1≤...≤kn−1≤ j ik1 k1k2 kn−1 j
Proof. We give a proof by mathematical induction. For n = 2, we have

 a11 a12 K a1m   a11 a11a12 + a12 a22 K a11a1m + a12 a2m +...+a1m amm 
2 2

0 a K a  0 K a22 a2m + a23a3m +...+a2m amm 

[ ]
2
a22
A =
2 22 2m
 =  = :α .
 M M M   M M M  ij

0 0 L a  0 K
2 
 mm   0 amm 
2
Apparently, A is an upper triangular matrix such that for each i, j, 1 < i, j < m,


 0 if i > j,

α ij =  if i = j,
2
aii
 ∑ a a if i < j.
i≤k1≤ j ik1 k1 j
«.  ß¢≈“π§√‘π∑√å «∑∑. √“°¢Õ߇¡∑√‘°´å
ªï∑’Ë 27 ©∫—∫∑’Ë 3 æ.§. - ¡‘.¬. 2548 661 ∫ÿ≠√Õ¥ ¬ÿ∑∏“π—π∑å ·≈– ™àÕøÑ“ π‘≈√—µπå
Now, we assume that A = [αij] where
k


 0 if i > j,

α ij =  if i = j,
k
aii
 ∑ a a ...a if i < j.
i≤k1≤...≤kk −1≤ j ik1 k1k2 kk −1 j
Then
= Ak A
k+1
A

 α11 α12 K α1m   a11 a12 K a1m 

 0 α K α 2m   0 a22 K a2m 
= 22
 
 M M M  M M M 
 0 0 L α mm   0 0 L amm 


 α11a11 α11a12 + α12 a22 K α11a1m + α12 a2m +...+α1m amm 

 0 α 22 a22 K α 22 a2m + α 23a3m +...+α 2m amm 
= 
 M M M 
 0 K α 
 0 a
mm mm 

[ ]
= : α ′ij .

It is clear that α ′ij = 0 if i > j. For each integer i, 1 < i < m, α ′ii = α ii aii = aiik aii = aiik+1 . We also obtain
j j  
α ′ij = ∑ α ik akj = ∑  ∑ aik ak k ...ak k  akj = ∑ aik ak k ...ak j
k=i k=i  i≤k1 ≤...≤kk −1 ≤k 1 1 2 k −1  i≤k1 ≤...≤kk ≤ j 1 1 2 k

for all integers i and j, 1 < i < j < m.

Theorem 1.2 Let A be an m×m complex matrix. If A is non-singular, then A always has an n root for
th

any positive integer n.

Proof. Let A be non-singular. By Schur's theorem (Strang, 1988), there exists a non-singular matrix S
-1
such that A = SBS where B is upper triangular. Let B = bij [ ] m×m
. We have det(B) ≠ 0; that is, bii ≠ 0 for

i = 1, 2, ..., m. Let bii* be any n root of bii. If bii = bi′i′ , we let bii* = bi*′i′ . We define C = cij
th
[ ] m×m
as follows.
n−1
For each i, cii = bii* . For i > j, let cij = 0. For j = i+1, let cij = bij ∑ ciin−1− p c jjp . For j = i+k, where
p=0

n−1
2 < k < m-i, let cij = (bij − Rij ) ∑ ciin−1− pc jjp , and Rij be the sum of the products cik ck k Kck , where the
p=0 1 1 2 n−1 j

sum is taken over integers k1, k2, ..., kn-1 such that i < k1 < ... < kn-1 < j and none of the term in the
products contains cij.
Songklanakarin J. Sci. Technol. Roots of Matrices
Vol.27 No.3 May - Jun. 2005 662 Yuttanan, B. and Nilrat, C.

Since bii ≠ 0 for i = 1, 2, ..., m, we have cii ≠ 0 for each i and ci,in−1 + ci,in−2ci+k,i+k +K+ci+k,i+k
n−1
≠ 0 for
n
1 < k < m-i. This guarantees that ci,i+k is well-defined. We claim that C = B.
Let C n = γ ij[ ] m×m
. By Lemma 1.1, we have


 0 if i > j,

γ ij =  n
cii if i = j,
 ∑ c c Kck ,i+k if j = i + k,k = 1,2,K, m − i.
i≤k1≤K≤kn−1≤i+k i,k1 k1,k2 n−1

If i = j, γ ij = cii = (bii ) = bii .

n n *

If j = i + 1 , we have
γ i,i+1 = ∑ ci,k ck ,k Kck
i≤k1 ≤K≤kn−1 ≤i+1 1 1 2 n−1,i+1

= ci,i+1 (ci,in−1 + ci,in−2ci+1,i+1 +K+ci+1,i+1

n−1
)

bi,i+1 (ci,i + ci,i ci+1,i+1 +K+ci+1,i+1 )

n−1 n−2 n−1

=
ci,in−1 + ci,in−2ci+1,i+1 +K+ci+1,i+1
n−1

= bi,i+1 .
If j = i + k , when k = 2, 3, ..., m-i, we have
γ i,i+k = ∑ ci,k ck ,k Kck
n−1,i+k
i≤k1 ≤K≤kn−1 ≤i+k 1 1 2

= ci,i+k (ci,in−1 + ci,in−2ci+k,i+k +K+ci+k,i+k

n−1
) + Ri,i+k

(bi,i+k − Ri,i+k )(ci,i + ci,i ci+k,i+k +K+ci+k,i+k )

n−1 n−2 n−1

= + Ri,i+k
ci,i + ci,i ci+k,i+k +K+ci+k,i+k
n−1 n−2 n−1

= bi,i+k .

[ ] [ ]
Then we obtain γ ij = bij . Therefore A = (SCS ) .
-1 n

We illustrate the procedure in Theorem 1.2 by the following example. Let

 8 −12 7 −8 
 0 −1 0 6 
B= .
 0 0 1 −28
 
0 0 0 8 

[ ]
A third root of B is a matrix C = cij where cij = 0, if i > j and c11 = 2, c22 = -1, c33 = 1, c44 = 2,

[ ] [
c12 = b12 c11 + c11c22 + c22 = (−12) 22 + (2)(−1) + (−1)2 = −4,
2 2
]
c23 = b23 [c
22
2
+ c22c33 + c ] = (0) [(−1)
2
33
2
+ (−1)(1) + 12 = 0, ]
«.  ß¢≈“π§√‘π∑√å «∑∑. √“°¢Õ߇¡∑√‘°´å
ªï∑’Ë 27 ©∫—∫∑’Ë 3 æ.§. - ¡‘.¬. 2548 663 ∫ÿ≠√Õ¥ ¬ÿ∑∏“π—π∑å ·≈– ™àÕøÑ“ π‘≈√—µπå
[ ] [
c34 = b34 c33 + c33c44 + c44 = (−28) 12 + (1)(2) + 22 = −4,
2 2
]
[
c13 = b13 − {c11c12c23 + c12c22c23 + c12c23c33} ] [c 2
11
+ c11c33 + c33
2
]
= [ 7 − {(2)(−4)(0) + (−4)(−1)(0) + (−4)(0)(1)}] 22 + (2)(1) + 12 = 1, [ ]
[
c24 = b24 − {c22c23c34 + c23c33c34 + c23c34c44} ] [c 2
22
+ c22c44 + c44
2
]
= [6 − {(−1)(0)(−4) + (0)(1)(−4) + (0)(−4)(2)}] (−1) + (−1)(2) + 22 = 2,[ 2
]
[
c14 = b14 − {c11c12c24 + c11c13c34 + c12c22c24 + c12c23c34 + c12c24c44 + c13c33c34 + c13c34c44} ] [c 2
11
+ c11c44 + c44
2
]
= [ −8 − {(2)(−4)(2) + (2)(1)(−4) + (−4)(−1)(2) + (−4)(0)(−4) + (−4)(2)(2) + (1)(1)(−4) +
[
(1)(−4)(2)}] 22 + (2)(2) + 22 = 3. ]
 2 −4 1 3 
 0 −1 0 2 
 
That is  0 0 1 −4 is a third root of B.
 
0 0 0 2 

th
Some singular matrices also have an n root such as
n
 1 0  1 0
  =  .
 0 0  0 0

 0 1
Moreover, we have a singular matrix   as a rootless matrix (Yood, 2002).
 0 0

th
Corollary 1.3 If all eigenvalues of A are not zero, then A has an n root.
Proof. Since A has all non-zero eigenvalues, A is a non-singular matrix.

Note. If only one eigenvalue of A is zero, in Theorem 1.2, we have bii = 0 for only one value of i.
That means we still have ci,i + ci,i ci+k,i+k +K+ci+k,i+k ≠ 0. Then we can say that "A matrix with only one
n−1 n−2 n−1

th
zero eigenvalue always has an n root".

2. Roots of diagonalizable matrices

th
In this section, we consider an n root of a diagonalizable matrix.

Theorem 2.1 Let A be an m×m complex matrix. If A is diagonalizable, then A has an n root, for any
th

positive integer n.
-1
Proof. Let A be a diagonalizable matrix, i.e., there exists a non-singular matrix S such that A = SDS
where D = dij[ ] m×m
is a diagonal matrix.

Let D = dij
1/n
[ ] 1/n

m×m
th 1/n n -1
, where dij1/n is an n root of dij. So A = S(D ) S = (SD S ) . Therefore an n
1/n -1 n th

root of A exists.

Songklanakarin J. Sci. Technol. Roots of Matrices
Vol.27 No.3 May - Jun. 2005 664 Yuttanan, B. and Nilrat, C.
th
However, we have some non-diagonalizable matrices having an n root, for example,
 2 0 0
 0 2 0
 
 2 0 2
th
has an n root because it is a non-singular matrix. Moreover, we see that diagonalizable matrices and
th
non-singular matrices are not the only matrices which have an n root, since
n
 1 1
 1 1 1 
1
n n
 0 1 1 = 0 1 1  ,n ≥ 2.
   
 0 0 0 0 0 0
 
There are some further questions the reader might like to consider. For instance, what is a
th
necessary and sufficient condition for a matrix to have an n root?
As an immediate consequence of the above theorem, we can conclude that a matrix A with all
th th
distinct eigenvalues has an n root. On the other hand, a real symmetric matrix also has an n root for
any positive integer n, as well as a complex Hermitian matrix and a normal matrix.

3. Roots of p-nilpotent matrices

th
In the previous two sections, we considered matrices whose n root always exists for any positive
th
integer n. In this section, we consider some kind of matrices which has an n root for just some value
of n.
An m×m matrix A is called p-nilpotent if A is a nilpotent matrix but not principal nilpotent and p
p p-1
is the least positive integer such that A = 0 but A 0. Before discussing on p-nilpotent matrices,
we first give the following lemma.

Lemma 3.1 Let A be an m×m complex matrix. If A = 0 for some k > 2, then A = 0.
k m

Proof. If 2 < k < m, then we are done. Now we suppose k > m. By Schur's theorem (Strang, 1988),
-1 k
there exists a non-singular matrix S such that A = SBS where B is upper triangular. Since A = 0, we
k
have B = 0.

[ ]
Let B = bij
m×m
k
[ ]
and B = βij . For 1 < i < m, we have βii = biik , so bii = 0. Then B is strictly
m×m

m m
upper triangular. It was proved by Yood (2002) that B = 0. Therefore A = 0.

Theorem 3.2 Let A be an m×m p-nilpotent matrix. If an n root of A exists, then n < m - p + 1.
th

r rp p
Proof. The proof is by contradiction. Suppose that A = S , for r > m - p + 2. Then S = A = 0 so that
S is an m×m nilpotent matrix. By Lemma 3.1, the m power of S is zero. Therefore, S = 0 for all positive
th k

r(p-1) p-1
integers k > m. But we also have S = A 0. Now p > 2, hence, 2r - 2 < rp - p, so that
rp-r p-1
r + p - 2 < rp - r. Since r > m - p + 2, m < r + p - 2 < rp - r. Therefore S = 0 or A = 0, which is contrary


th
to the hypotheses on A. Hence, if an n root of p-nilpotent matrix exists, then n < m - p + 1.
2 -1
Let A be a 2-nilpotent matrix of size 3×3, i.e., A = 0. By Schur's theorem, A is of the form SBS
2 3
where S is non-singular and B is upper triangular. Hence B = 0. This implies B = 0. By Yood (2002),
B is a strictly upper triangular matrix. It is possible to classify B which is not principal nilpotent as five
different types:
«.  ß¢≈“π§√‘π∑√å «∑∑. √“°¢Õ߇¡∑√‘°´å
ªï∑’Ë 27 ©∫—∫∑’Ë 3 æ.§. - ¡‘.¬. 2548 665 ∫ÿ≠√Õ¥ ¬ÿ∑∏“π—π∑å ·≈– ™àÕøÑ“ π‘≈√—µπå
 0 a 0   0 0 0   0 0 a  0 a b   0 0 b 
 0 0 0  ,  0 0 a ,  0 0 0  ,  0 0 0  ,  0 0 a ,
         
 0 0 0  0 0 0  0 0 0  0 0 0  0 0 0

where a, b 0. Then we see that all five types of B has a square

2 -1 -1 2
We observe that root, say T. Therefore A = ST S = (STS ) . This
2 shows that a square root of any 2-nilpotent matrix
 0 a 0   0 0 a of size 3×3 always exists.
 0 0 0 =  0 0 0 ,
   
 0 0 0  0 1 0 Conclusion and Discussion
2
 0 0 0   0 0 a According to this article, we obtain a
th
 0 0 a =  1 0 0  , formula for calculating an n root of a matrix
    which is non-singular or diagonalizable.
 0 0 0  0 0 0
However, being non-singular or diagonaliz-
th
2 able are not necessary for matrices to have n
 0 0 a  0 a 0 
roots. The reader may try to find other properties
 0 0 0 =  0 0 1 , th

    of his own. In addition, a matrix having an n

 0 0 0  0 0 0 root for some positive integer n is not only a p-
nilpotent matrix.
2
 
 0 a b  0 −a 0
b
References
 0 0 0 =  0 −1 −  , Nicholson, W.K. 1990. Elementary Linear Algebra
   a with Applications, 2nd ed., PWS-KENT, Boston.
 0 0 0 0 a
 1  Strang, G. 1988. Linear Algebra and Its Applications,
b
3rd ed., Harcourt, New York.
2 Yood, B. 2002. Rootless Matrices, Math. Mag., 75:
 b 
1 − 0 219-223.
 0 0 b  a 
 0 0 a a 
= −1 −a .
  b
 0 0 0 0 0 0
 
 