Structure of The NTH Roots of A Matrix
Structure of The NTH Roots of A Matrix
Structure of The NTH Roots of A Matrix
ABSTRACT
0. INTRODUCTION
solutions is given. One of the results, mentioned in [I, Corollaries 4.7 and
4.81, includes the corollary stated in Section 2.
/
11
12
\ Is
Let (Y~ be the eigenvalue corresponding to the m, X mi Jordan block I,, and
z E M,,C.
For x E C let g(z) = x1/n denote any value of the nth root of x, and let
g(Ji) denote any of the nth roots of Ji (see for instance [2, p. 2321).
When A is non-singular, all solutions X of X” = A are given by the
matrices
g(A) := Z-‘U’g(J,)UZ
km \
= z-Iu-I d12) UZ
\ g(l,T),
for all nonsingular U E M,C satisfying UJA = JAU [2, p. 2321. When all
eigenvalues oi of A are distinct, the matrices commuting with JA are
polynomials in JA [5, p. 419, Proposition l] and thus polynomials in g(JA).
These matrices commuting with JA must consequently commute with g(JA),
too. Therefore we find exactly ns solutions of X” = A. These solutions can
be written as polynomials in A [2, p. 2331.
According to Lancaster and Tismenetsky [5, p. 418, Theorem 11, when not
all of the eigenvalues are distinct, there is at least one g(JA) for which an
infinity of nonsingular matrices U exists such that U commutes with JA but
not with the Jordan canonical form J, of g(JA). Since U is a block matrix,
partitioned as J, and g(JA) and consisting of upper-triangular Toeplitz
matrices [5, p. 4191, u commutes with g(JA) if and only if it commutes with
1,. Thus U does not commute with g(JA) either. In Theorem 2 we will prove
STRUCTURE OF THE nTH ROOTS 61
2. ROOTS IN A FIELD K
\ 0 0 ..: ; 0 ,
which has h(x) as its characteristic polynomial and as its minimum (annihilat-
ing) polynomial. Suppose h(x) is irreducible over K with factorization over C
given by h(x) = (x - al> ... (x - a,). By definition we have, for any posi-
tive integer k, that hk is the minimum polynomial of C(hk>. It follows that
C(hk) is similar to the Jordan canonical form J((Y~) @ ... @ J(a,,,>, where
each J(oi) is a k X k Jordan block.
Let @ dA stand for the block-diagonal matrix A @ A @ ... @ A where A
appears d times. Then a matrix A E M,, K has a K-normal form C, =
@;= i djC(fik+, w h ere the f/g(x) are the elementary divisors in K[ x] of A [2,
pp. 149-1501. By the definition of elementary divisor the f,< x) are irreducible,
and by usin the aforementioned multiplicity di we can assume that no two
2
of the fj(x> 1 are the same. Note that, when mi is the degree of fi(x>, the
order rn of the matrix A is equal to C:=, d,kim,. Now, for every i, take an
(Y~= ‘Y!‘) which satisfies fi( (wi) = 0, and for j = 1, . . . , n, take the n distinct
62 GABRIELLE TEN HAVE
values yij such that r,; = (Y~. Let njj be given by nij = [K(yij): K((Y~)] for
i = l,..., t and j = 1,. . . , n.
For the construction, find the Jordan canonical form of an nth root P of
A as indicated in the proof. From this form a K-normal form C, can
be deduced. Since C: is similar to A, a matrix Y E M, K exists such
that (Y-lC,Y)” = Y-‘Cp”Y = A. This Y can be found from the equation
C;Y = YA.
Further, let
6 = 0 for n odd,
” 1 for n even
and
for n=2
n-l
I
h/21
c z+j -1
for n 2 3odd,
j=o
_i
64 GABRIELLE TEN HAVE
AV = VA,
Take
I PV + VP,
det V # 0,
all V- ’ PV are distinct.
A,, E K
depending only on V such that V, is nonsingular for IAl > A,; thus we take
]A( > A, in the rest of the proof. Clearly V, fulfills the first three conditions
of (1) for all A if V satisfies these conditions. Now assume that VA,, . , V,
satisfy (1) for certain j > 1. We shall show that we can find a matrix VAj+:
such that VA,, . . . , VA]+, satisfy (1). Note that V-lPV = W- ‘PW exactly when
(W-‘)-lP(vW-l) = P, i.e., when VW-l commutes with P. Thus consider
VAj+,VAi’ = (V + Aj+lZ)VAi’ for Aj+i e (A,,. .., Aj} and for k = l,..., j.
Suppose there is a k such that V, Vi;’ commutes with P. We know that
A’V*; ’ does not commute with P f&‘any A’ # 0. Hence VA,+,+ hjVAy’ = (V +
A.+ i Z>V*L’ + A’V*; i does not commute with P for A’ # 0. Thus, we can
c 1:oose V, such that V,,, . . , VA,+, satisfy (l), since only finitely many values
A >, A,, for Aj + i are excluded. n
4. INTEGRAL SOLUTIONS
(t -Y)and(ii 2)
are similar over Q, but they are not similar over Z.
66 GABRIELLE TEN HAVE
For m = 2, [3] gi ves all solutions in case A # al for a E Z \ {0}, and (an
upper bound for) the number of similarity classes in case A = al.
For m 2 2, Latimer and MacDuffee [4] connect the number of similarity
classes over Z of the set of nonsingular matrices in M,Z having m distinct
characteristic roots with the number of ideal classes in a certain order. This
result can very well be used in case a matrix root P of A has m distinct
characteristic roots, although even then in some cases only a lower bound of
the number of similarity classes can be given. In case P has less than m
distinct characteristic roots, either because some Jordan blocks have order
greater than one or because some Jordan blocks occur more than once, no
results are known.
REFERENCES
1 J.-C. Evard and F. Uhlig, on the matrix equation j(X) = A, Linear Algebra
Appl. 162-164:447-519 (1992).
2 F. R. Gantmacher, Matrix Theory, Vol. 1, Chelsea, 1959.
3 G. N. ten Have, Matrix solutions of the equation X” = A, Zndag. Math. (N.S.)
2:57-64 (1991).
4 C. G. Latimer and C. C. MacDuffee, A correspondence between classes of ideals
and classes of matrices, Ann. Math. 34:313-316 (1933).
5 P. Lancaster and M. Tismenetsky, The Theory of Matrices, Academic, 1985.
6 D. E. Otero, Extraction of mth roots in matrix rings over fields, Linear Algebra
Appl. 128:1-26 (1990).
7 J. Riordan, An Introduction to CombinutoriuZ Analysis, Wiley, 1958.