The Electronic Journal of Linear Algebra ISSN 1081-3810

A publication of the International Linear Algebra Society

Volume 9, pp. 32-41, April 2002
ELA

ON THE mth ROOTS OF A COMPLEX MATRIX∗

PANAYIOTIS J. PSARRAKOS†

Abstract. If an n × n complex matrix A is nonsingular, then for every integer m > 1, A has
an mth root B, i.e., B m = A. In this paper, we present a new simple proof for the Jordan canonical
form of the root B. Moreover, a necessary and sufficient condition for the existence of mth roots of
a singular complex matrix A is obtained. This condition is in terms of the dimensions of the null
spaces of the powers Ak (k = 0, 1, 2, . . .).

Key words. Ascent sequence, eigenvalue, eigenvector, Jordan matrix, matrix root.

AMS subject classifications. 15A18, 15A21, 15A22, 47A56

1. Introduction and preliminaries. Let Mn be the algebra of all n × n

complex matrices and let A ∈ Mn . For an integer m > 1, a matrix B ∈ Mn is
called an mth root of A if B m = A. If the matrix A is nonsingular, then it always
has an mth root B. This root is not unique and its Jordan structure is related to
the Jordan structure of A [2, pp. 231-234]. In particular, (λ − µ0 )k is an elementary
divisor of B if and only if (λ − µm k
0 ) is an elementary divisor of A. If A is a singular
complex matrix, then
it may  have no mth roots. For example, there is no matrix
2 0 1
B such that B = . As a consequence, the problem of characterizing the
0 0
singular matrices, which have mth roots, is of interest [1], [2].
Consider the (associated) matrix polynomial P (λ) = In λm − A, where In is the
identity matrix of order n and λ is a complex variable. A matrix B ∈ Mn is an
mth root of A if and only if P (B) = B m − A = 0. As a consequence, the problem
of computation of mth roots of A is strongly connected with the spectral analysis of
P (λ). The suggested references for matrix polynomials are [3] and [7].
A set of vectors {x0 , x1 , . . . , xk }, which satisfies the equations

P (ω0 )x0 = 0
1 (1)
P (ω0 )x1 + P (ω0 )x0 = 0
1!
.. .. ..
. . .
1 (1) 1 (k)
P (ω0 )xk + P (ω0 )xk−1 + · · · + P (ω0 )x0 = 0,
1! k!
where the indices on P (λ) denote derivatives with respect to the variable λ, is called
a Jordan chain of length k + 1 of P (λ) corresponding to the eigenvalue ω0 ∈ C and
the eigenvector x0 ∈ Cn . The vectors in a Jordan chain are not uniquely defined and
for m > 1, they need not be linearly independent [3], [6]. If we set m = 1, then the
∗ Received by the editors on 19 February 2002. Accepted for publication on 9 April 2002. Handling

Editor: Peter Lancaster.

† Department of Mathematics, National Technical University, Zografou Campus, 15780 Athens,

Greece ([email protected]).
32
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
On the mth Roots of a Complex Matrix 33

Jordan structure of the linear pencil In λ − A coincides with the Jordan structure of
A, and the vectors of each Jordan chain are chosen to be linearly independent [2], [6].
Moreover, there exist a matrix
 
(1.1) JA = ⊕ξj=1 Ikj ωj + Nkj (k1 + k2 + . . . + kξ = n),

where Nk is the nilpotent matrix of order k having ones on the super diagonal and
zeros elsewhere, and an n × n nonsingular matrix
 
(1.2) XA = x1,1 . . . x1,k1 x2,1 . . . x2,k2 . . . xξ,1 . . . xξ,kξ ,

where for every j = 1, 2, . . . , ξ, {xj,1 , xj,2 , . . . , xj,kj } is a Jordan chain of A corre-

sponding to ωj ∈ σ(A), such that (see [2], [4], [6])
−1
A = XA JA XA .

The matrix JA is called the Jordan matrix of A, and it is unique up to permutations

of the diagonal Jordan blocks Ikj ωj + Nkj (j = 1, 2, . . . , ξ) [2], [4].
The set of all eigenvalues of P (λ), that is, σ(P ) = {µ ∈ C : detP (µ) = 0},
is called the spectrum of P (λ). Denoting by σ(A) = σ(In λ − A) the spectrum
of the matrix A, it is clear that σ(P ) = {µ ∈ C : µm ∈ σ(A)}. If JA is the
Jordan matrix of A in (1.1), then it will be convenient to define the J-spectrum of A,
σJ (A) = {ω1 , ω2 , . . . , ωξ }, where the eigenvalues of A follow exactly the order of their
appearance in JA (obviously,

 repetitions
are allowed).
 For example, the J-spectrum
0 1   1 1
of the matrix M = ⊕ 0 ⊕ is σJ (M ) = { 0, 0, 1}.
0 0 0 1
In this article, we study the Jordan structure of the mth roots (m > 1) of a
complex matrix. In Section 2, we consider a nonsingular matrix and present a new
constructive proof for the Jordan canonical form of its mth roots. This proof is
simple and based on spectral analysis of matrix polynomials [2], [3], [7]. Furthermore,
it yields directly the Jordan chains of the mth roots. We also generalize a known
uniqueness statement [5]. In Section 3, using a methodology of Cross and Lancaster
[1], we obtain a necessary and sufficient condition for the existence of mth roots of a
singular matrix.
2. The nonsingular case. Consider a nonsingular matrix A ∈ Mn and an
integer m > 1. If A is diagonalizable and S ∈ Mn is a nonsingular matrix such that

A = S diag{r1 eiφ1 , r2 eiφ2 , . . . , rn eiφn } S −1 ,

where rj > 0 , φj ∈ [0, 2π) (j = 1, 2, . . . , n), then for every n-tuple (s1 , s2 , . . . , sn ),
sj ∈ {1, 2, . . . , m} (j = 1, 2, . . . , n), the matrix
1 φ1 +2(s1 −1)π 1 φ2 +2(s2 −1)π 1 φn +2(sn −1)π
B = S diag{r1m ei m , r2m ei m , . . . , rnm ei m } S −1

is an mth root of A. Hence, the investigation of the mth roots of a nonsingular (and
not diagonalizable) matrix A via the Jordan canonical form of A arises in a natural
way [2]. The following lemma is necessary and of independent interest.
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
34 P.J. Psarrakos

Lemma 2.1. Let {x0 , x1 , . . . , xk } be a Jordan chain of A ∈ Mn (with linearly

independent terms) corresponding to a nonzero eigenvalue ω0 = r0 ei φ0 ∈ σ(A) (r0 >
0, φ0 ∈ [0, 2π)), and let P (λ) = In λm − A. Then for every eigenvalue
1 φ0 +2(t−1)π
r0m ei m ∈ σ(P ) ; t = 1, 2, . . . , m,
the matrix polynomial P (λ) has a Jordan chain of the form

y0 = x0
y1 = a1,1 x1
(2.1) y2 = a2,1 x1 + a2,2 x2
.. .. ..
. . .
yk = ak,1 x1 + ak,2 x2 + · · · + ak,k xk ,

where the coefficients ai,j (1 ≤ j ≤ i ≤ k) depend on the integer t and for ev-
m−1 φ0 +2(t−1)π
ery i = 1, 2, . . . , k, ai,i = (m r0 m ei(m−1) m )i
= 0. Moreover, the vectors
y0 , y1 , . . . , yk are linearly independent.
Proof. Since {x0 , x1 , . . . , xk } is a Jordan chain of the matrix A corresponding to
the eigenvalue ω0
= 0, we have

(A − In ω0 )x0 = 0
and
(A − In ω0 )xi = xi−1 ; i = 1, 2, . . . , k.

Let µ0 be an eigenvalue of P (λ) such that µm

0 = ω0 . By the equation

(In ω0 − A)x0 = (In µm

0 − A)x0 = 0,

it is obvious that y0 = x0 is an eigenvector of P (λ) corresponding to µ0 ∈ σ(P ).

Assume now that there exists a vector y1 ∈ Cn such that
P (1) (µ0 )
P (µ0 )y1 + y0 = 0.
1!
Then
m−1
(In µm
0 − A)y1 = − mµ0 y0 ,
or equivalently,
(In ω0 − A)y1 = mµ0m−1 (In ω0 − A)x1 .

Hence, we can choose y1 = a1,1 x1 , where a1,1 = mµm−1

= 0. Similarly, if we consider
the equation
P (1) (µ0 ) P (2) (µ0 )
P (µ0 )y2 + y1 + y0 = 0,
1! 2!
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
On the mth Roots of a Complex Matrix 35

then it follows

m−1 m(m − 1) m−2

(In µm
0 − A)y2 = − mµ0 y1 − µ0 y0 ,
2
or equivalently,
 
m−1 2 m(m − 1) m−2
(In ω0 − A)y2 = (In ω0 − A) (mµ0 ) x2 + µ0 x1 .
2

Thus, we can choose y2 = a2,1 x1 + a2,2 x2 , where a2,1 = m(m−1)

2 µ0m−2 and a2,2 =
 m−1 2
mµ
= 0. Repeating the same steps implies that the matrix polynomial P (λ)
has a Jordan chain {y0 , y1 , . . . , yk } as in (2.1).
Define the n × (k + 1) matrices
   
X0 = x0 x1 · · · xk and Y0 = y0 y1 · · · yk .

Since the vectors x0 , x1 , . . . , xk ∈ Cn are linearly independent, rank(X0 ) = k + 1.

Moreover, Y0 = X0 T0 , where the (k + 1) × (k + 1) upper triangular matrix
 
1 0 0 ··· 0
 0 a1,1 a2,1 · · · ak,1 
 
 a2,2 · · · ak,2 
T0 =  0 0 
 .. .. .. .. .. 
 . . . . . 
0 0 0 · · · ak,k

is nonsingular. As a consequence, rank(Y0 ) = k + 1, and the proof is complete.

The Jordan chain {y0 , y1 , . . . , yk } of P (λ) in the above lemma, is said to be
associated to the Jordan chain {x0 , x1 , . . . xk } of A, and clearly depends on the
choice of t ∈ {1, 2, . . . , m}. Consider now the nonsingular matrix XA ∈ Mn in (1.2),
and for any (s1 , s2 , . . . , sξ ), sj ∈ {1, 2, . . . , m} (j = 1, 2, . . . , ξ), define the matrix
 
YA (s1 , s2 , . . . , sξ ) = y1,1 . . . y1,k1 y2,1 . . . yξ,1 . . . yξ,kξ ,

where for every j = 1, 2, . . . , ξ, the set {yj,1 , yj,2 , . . . , yj,kj } is the associated Jordan
chain of P (λ) corresponding to the Jordan chain {xj,1 , xj,2 , . . . , xj,kj } of A and the
integer sj .
Corollary 2.2. For every (s1 , s2 , . . . , sξ ), sj ∈ {1, 2, . . . , m} (j = 1, 2, . . . , ξ),
the associated matrix YA (s1 , s2 , . . . , sξ ) is nonsingular.
Proof. By Lemma 2.1, there exist upper triangular matrices T1 , T2 , . . . , Tξ , which
depend on the choice of the ξ-tuple (s1 , s2 , . . . , sξ ) and have nonzero diagonal ele-
ments, such that
 
YA (s1 , s2 , . . . , sξ ) = XA ⊕ξj=1 Tj .

Since XA is also nonsingular the proof is complete.

The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
36 P.J. Psarrakos

Theorem 2.3. Let A ∈ Mn be a nonsingular complex matrix with Jordan

matrix JA = ⊕ξj=1 Ikj ωj + Nkj as in (1.1) and J-spectrum

σJ (A) = { ω1 = r1 eiφ1 , ω2 = r2 eiφ2 , . . . , ωξ = rξ eiφξ }.

Consider an integer m > 1, the nonsingular matrix XA ∈ Mn in (1.2) such that
−1
A = XA JA XA , a ξ-tuple (s1 , s2 , . . . , sξ ), sj ∈ {1, 2, . . . , m} (j = 1, 2, . . . , ξ) and
the associated matrix YA (s1 , s2 , . . . , sξ ). Then the matrix
  
1 φj +2(sj −1)π
ξ m i
B = YA (s1 , s2 , . . . , sξ ) ⊕j=1 Ikj rj e m + Nkj YA (s1 , s2 , . . . , sξ )−1
(2.2)
is an mth root of A.
Proof. Since the associated matrix YA (s1 , s2 , . . . , sξ ) is nonsingular, by Corollary
7.11 in [3], the linear pencil In λ − B is a right divisor of P (λ) = In λm − A, i.e., there
exists an n × n matrix polynomial Q(λ) of degree m − 1 such that
P (λ) = Q(λ) (In λ − B).
Consequently, by [2, pp. 81-82] (see also Lemma 22.9 in [7]), P (B) = B m − A = 0,
and hence B is an mth root of the matrix A.
At this point, we remark that the associated matrix YA (s1 , s2 , . . . , sn ) can be
computed directly by the method described in the proof of Lemma 2.1. Moreover,
it is clear that a nonsingular matrix may have mth roots with common eigenvalues.
Motivated by [5], we obtain a spectral condition that implies the uniqueness of an
mth root.
Theorem 2.4. Suppose A ∈ Mn is a nonsingular complex matrix and its
spectrum σ(A) lies in a cone
K0 = {z ∈ C : θ1 ≤ Arg z ≤ θ2 , 0 < θ2 − θ1 ≤ ϑ0 < 2π}.
Then for every k = 1, 2, . . . , m, A has a unique mth root Bk such that
 
θ1 + 2(k − 1)π θ2 + 2(k − 1)π
σ(Bk ) ⊂ z ∈ C : ≤ Arg z ≤ .
m m
2(k−1)π
In particular, for every k = 2, 3, . . . , m, Bk = ei m B1 .
Proof. Observe that the spectrum of P (λ), σ(P ) = {µ ∈ C : µm ∈ σ(A)}, lies
in the union
m  
θ1 + 2(k − 1)π θ2 + 2(k − 1)π
z∈C: ≤ Arg z ≤ ,
m m
k=1

and for every k = 1, 2, . . . , m, denote

 
θ1 + 2(k − 1)π θ2 + 2(k − 1)π
Σk = σ(P ) ∩ z ∈ C : ≤ Arg z ≤
m m
1 φ+2(k−1)π
= {rme i m iφ
: re ∈ σ(A), r > 0, φ ∈ [θ , θ ]}. 1 2
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
On the mth Roots of a Complex Matrix 37

Then by Theorem 2.3, for every k = 1, 2, . . . , m, the matrix A has an mth root Bk
such that σ(Bk ) = Σk . Since the sets Σ1 , Σ2 , . . . , Σm are mutually disjoint, Lemma
22.8 in [7] completes the proof.
 
It is worth noting that if JA = ⊕ξj=1 Ikj ωj + Nkj is the Jordan matrix of A in
(1.1), and σ (A) = { ω = r eiφ1 , ω = r eiφ2 , . . . , ω = r eiφξ } is the J-spectrum
J 1 1 2 2 ξ ξ
of A, then for every k = 1, 2, . . . , m, the mth root Bk in the above theorem is given
by (2.2) for s1 = s2 = · · · = sξ = k. Furthermore, if we allow θ1 −→ − π + and
θ2 −→ π − , then for m = 2, we have the following corollary.
Corollary 2.5. (Theorem 5 in [5]) Let A ∈ Mn be a complex matrix with
σ(A) ∩ (−∞, 0] = ∅. Then A has a unique square root B such that σ(B) ⊂ {z ∈ C :
Re z > 0}.
−1
Example 2.6. Consider a 5 × 5 complex matrix A = XA JA XA , where XA ∈
M5 is nonsingular and
 
i 1 0 0 0
 0 i 1 0 0 
 
JA =   0 0 i 0 0 .

 0 0 0 1 1 
0 0 0 0 1
Suppose m = 3 and for a pair (s1 , s2 ), sj ∈ {1, 2, 3} (j = 1, 2), denote
π/2+2(s1 −1)π 2(s2 −1)π
α = ei 3 and β = ei 3 .
Then the associated matrix of XA , corresponding to (s1 , s2 ), is
  
1 0 0

1 0
YA (s1 , s2 ) = XA  0 3α2 3α  ⊕ .
4 0 3β 2
0 0 9α
One can verify that the matrix
 
α 1 0 0 0
 0 α 1 0 0 
 
B = YA (s1 , s2 ) 
 0 0 α 0 0  YA (s1 , s2 )−1

 0 0 0 β 1 
0 0 0 0 β
 1 −2

α 3α − 19 α−5 0 0
 0 α 1 −2 
 3α 0 0  −1
= XA 
 0 0 α 0 0 X
 A
 0 0 0 β 1 −2
β 
3
0 0 0 0 β
is a 3rd root of A (see also the equation (58) in [2, p. 232]). Moreover, if we choose
s1 = s2 = 1, then
  
1 0 0

1 0 
YA (1, 1) = XA  0 3ei 3 3ei 6  ⊕
π π
,
i 2π 0 3
0 0 9e 3
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
38 P.J. Psarrakos

and the matrix

 π

ei 6 1 0 0 0
 π 
 0 ei 6 1 0 0 
 
B1 = YA (1, 1)  0 0 i
e 6
π
0 0  YA (1, 1)−1
 
 0 0 0 1 1 
0 0 0 0 1
 π

1 −i π 5π
ei 6 3e
3 − 91 e−i 6 0 0
 π 1 −i π 
 0 ei 6 3e
3 0 0 
  −1
= XA  0 0
π
ei 6 0 0  XA
 
 0 0 0 1 1 
3
0 0 0 0 1

is the unique 3rd root of A with σ(B1 ) ⊂ {z ∈ C : 0 ≤ Arg z ≤ π/6}.

3. The singular

 case. Let A ∈ Mn be a singular matrix, and let JA =
⊕ξj=1 Ikj ωj + Nkj be its Jordan matrix in (1.1). For the remainder and without
loss of generality, we assume that ωj = 0 for j = 1, 2, . . . , ψ (1 ≤ ψ ≤ ξ), with
ωj
= 0 otherwise, and k1 ≥ k2 ≥ · · · ≥ kψ [1], [2]. We also denote by
 
J0 = ⊕ψ ψ
j=1 Ikj ωj + Nkj = ⊕j=1 Nkj

the diagonal block of JA corresponding to the zero eigenvalue. Then by [2, pp. 234-
239], we have the following lemma.
Lemma 3.1. The matrix A ∈ Mn has an mth root if and only if J0 has an mth
root.
The ascent sequence of A is said to be the sequence

di = dim Null Ai − dim Null Ai−1 ; i = 1, 2, . . .

By [1], we have the following properties:

(P1) The ascent sequences of A and J0 are equal.
(P2) For every i = 1, 2, . . ., di is the number of the diagonal blocks of J0 of order at
ψ
least i. Thus, if d0 = j=1 kj is the order of J0 , then d0 ≥ d1 ≥ d2 ≥ · · · ≥
dk1 ≥ 0 and dk1 +1 = dk1 +2 = · · · = 0.
Theorem 3.2. The complex matrix A ∈ Mn has an mth root if and only if for
every integer ν ≥ 0, the ascent sequence of A has no more than one element between
m ν and m (ν + 1).
(Note that the result is obvious when the matrix A is nonsingular.)
Proof. By Lemma 3.1 and Property (P1), it is enough to prove the result for J0 .
First assume that J0 has an mth root Z, and that there exist a nonnegative integer
ν and two terms of the ascent sequence of J0 , say dt and dt+1 , such that

m ν < dt+1 ≤ dt < m (ν + 1).

The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
On the mth Roots of a Complex Matrix 39

For every i = 1, 2, . . . , k1 , Null Z mi = Null J0i , and consequently, if c1 , c2 , . . . is the

ascent sequence of Z, then
mi
 i

cj = dj .
j=1 j=1

Thus, we have

dt = cmt + cmt−1 + · · · + cmt−(m−1)

and

dt+1 = cm(t+1) + cm(t+1)−1 + · · · + cmt+1 ,

where

cmt−(m−1) ≥ · · · ≥ cmt ≥ cmt+1 ≥ · · · ≥ cm(t+1) .

If cmt ≥ ν + 1, then dt ≥ m cmt ≥ m (ν + 1), a contradiction. On the other hand,

if cmt ≤ ν, then dt+1 ≤ m cmt ≤ m ν, which is also a contradiction. Hence, we
conclude that if the matrix A has an mth root, then for every integer ν ≥ 0, the
ascent sequence of A has no more than one element between m ν and m (ν + 1).
Conversely, we indicate a constructive proof for the existence of an mth root of
J0 given that between two successive nonnegative multiplies of m there is at most one
i
term of the ascent sequence of J0 . Denote ni = j=1 kj for i = 1, 2, . . . , ψ, and let
nψ
{e1 , e2 , . . . , enψ } be the standard basis of C . By [1], for the vectors xi = eni (i =
1, 2, . . . , ψ), we can write the standard basis of Cnψ in the following scheme:

J0k1 −1 x1 J0k1 −2 x1 J0k1 −3 x1 J0k1 −4 x1 · · · J0 x1 x1

J0k2 −1 x2 J0k2 −2 x2 J0k2 −3 x2 ··· ··· x2
(3.1) J0k3 −1 x3 J0k3 −2 x3 ··· ··· x3
.. .. ..
. . .
k −1 k −2
J0 ψ xψ J0 ψ xψ ··· xψ .

In this scheme, there are ψ rows of vectors, such that the jth row contains kj vectors
(j = 1, 2, . . . , ψ). Recall that k1 ≥ k2 ≥ · · · ≥ kψ , and hence the rows are of non-
increasing length. Moreover, the above scheme has k1 columns and by Property (P2),
for t = 1, 2, . . . , k1 , the length of the tth column is equal to the tth term of the ascent
sequence of J0 , dt .
With respect to the above scheme (and the order of its elements), we define a
linear transformation F on Cnψ by
(j − 1, t)th element if j
= 1 (mod m),

(j, t)th element → (j + m − 1, t − 1)th element if j = 1 (mod m) and t
= 1,

0 if j = 1 (mod m) and t = 1.
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
40 P.J. Psarrakos

Separating the rows of the scheme in m-tuples, one can see that our assumption that
for every nonnegative integer ν, the ascent sequence has no more than one element
between m ν and m (ν + 1) ensures the existence of F . If B is the nψ × nψ matrix
whose jth column is F (ej ), then B m = J0 and the proof is complete.
Since between two successive even integers there is exactly one odd integer, for
m = 2, Theorem 3.2 yields the main result of [1].
Corollary 3.3. (Theorem 2 in [1]) The matrix A ∈ Mn has a square root if
and only if no two terms of its ascent sequence are the same odd integer.
The ascent sequence of the k × k nilpotent matrix Nk is 1, 1, . . . , 1, 0, . . . with
its first k terms equal to 1. Thus, for every integer m > 1, it has k terms (i.e., the k
ones) between 0 and m. Hence, it is verified that there is no matrix M ∈ Mn such
that M m = Nk (see also [2]).
Corollary 3.4. Let d1 , d2 , d3 , . . . be the ascent sequence of a singular complex
matrix A ∈ Mn .
(i) If d2 = 0 (i.e., J0 = 0), then for every integer m > 1, A has an mth root.
(ii) If d2 > 0, then for every integer m > d1 , A has no mth roots.
Our methodology is illustrated in the following example.
Example 3.5. Consider the Jordan matrix
 
0 1 0


0 1 0 1
J0 =  0 0 1  ⊕ ⊕ ,
0 0 0 0
0 0 0
and let m = 3. The ascent sequence of J0 is 3, 3, 1, 0, . . . , and if {e1 , e2 , . . . , e7 } is
the standard basis of C7 , then the scheme in (3.1) is
J02 x1 = e1 J0 x1 = e2 x1 = e3
J0 x2 = e4 x2 = e5
J0 x3 = e6 x3 = e7 .

As in the proof of Theorem 3.2, we define the linear transformation F on C7 by

F (e1 ) = 0, F (e2 ) = e6 , F (e3 ) = e7 , F (e4 ) = e1 ,

F (e5 ) = e2 , F (e6 ) = e4 and F (e7 ) = e5 .

One can see that the 7 × 7 matrix
 
0 0 0 1 0 0 0
 0 0 0 0 1 0 0 
 
 0 0 0 0 0 0 0 
   
B = 0 e6 e7 e1 e2 e4 e5 = 
 0 0 0 0 0 1 0 

 0 0 0 0 0 0 1 
 
 0 1 0 0 0 0 0 
0 0 1 0 0 0 0
is a 3rd root of J0 . Finally, observe that for every integer m > 1 different than 3,
the matrix J0 has no mth roots.
The Electronic Journal of Linear Algebra ISSN 1081-3810
A publication of the International Linear Algebra Society
Volume 9, pp. 32-41, April 2002
ELA
On the mth Roots of a Complex Matrix 41

REFERENCES

[1] G.W. Cross and P. Lancaster. Square roots of complex matrices. Linear and Multilinear Algebra,
1:289-293, 1974.
[2] F.R. Gantmacher. The Theory of Matrices, Vol. I. Chelsea Publishing Company, New York,
1959.
[3] I. Gohberg, P. Lancaster, and L. Rodman. Matrix Polynomials. Academic Press, New York,
1982.
[4] R.A. Horn and C.R. Johnson. Matrix Analysis. Cambridge University Press, Cambridge, 1985.
[5] C.R. Johnson, K. Okubo, and R. Reams. Uniqueness of matrix square roots and an application.
Linear Algebra and Its Applications, 323:51-60, 2001.
[6] P. Lancaster and M. Tismenetsky. The Theory of Matrices. Academic Press, Orlando, 1985.
[7] A.S. Markus. Introduction to the Spectral Theory of Polynomial Operator Pencils. Amer. Math.
Society, Providence, Translations of Math. Monographs, Vol. 71, 1988.