Sie sind auf Seite 1von 13

Journal of Mathematical Sciences, Vol. 160, No.

6, 2009

ON ROOTS OF QUATERNION POLYNOMIALS

N. Topuridze UDC 512.6


Abstract. The topological structure of the zero-sets of quaternion polynomials is discussed. As was
earlier proved by the author, such a zero-set consists of several points and two-dimensional spheres
with centers on the real line. We also show that one can define multiplicities of components of each
type in such way that their sum is equal to the algebraic degree of the polynomial considered.

CONTENTS
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 843
1. Preliminary Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 844
2. Some Classes of Quaternion Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 846
3. Zero-Sets of Canonical Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 847
4. Multiplicities of the Components of the Zero-Set . . . . . . . . . . . . . . . . . . . . . . . 853
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854

Introduction
The aim of this paper is to elaborate and develop the results on the roots of quaternion polynomials
obtained in [17, 31]. As was established in [31] (cf. [17]), the zero-set of a canonical quaternion
polynomial consists of p points and s two-dimensional spheres, where p + 2s does not exceed the
algebraic degree of a given polynomial. However, such a formulation is not completely satisfactory
since in order to have a complete analogy with the “fundamental theorem of algebra” for the field of
complex numbers [19] one has to indicate a way of assigning multiplicity to each connected component
of the zero-set.
In order to count multiplicities correctly, one needs a detailed understanding of the algebraic struc-
ture of the so-called quasi-norm of a canonical quaternion polynomial introduced in [31] (cf. [27]).
This done, it is easy to give an ad hoc definition of multiplicities which leads to a quaternionic version
of the “fundamental theorem of algebra.” We present in some detail all necessary technical results as
well as the definitions of multiplicity for components of both types. We also indicate a connection of
our definition with the general definition of multiplicity for a connected component of the zero-set of
a proper polynomial mapping suggested in [2].
In the first two sections, we present necessary concepts and auxiliary results concerned with quater-
nions and quaternion polynomials. In the third section, we give a detailed proof of the main result
of [31] which yields that the zero-set of a canonical quaternion polynomial consists of points and two-
dimensional spheres. To this end we carefully investigate factorization of the quasi-norm and relate
it to the structure of the zero-set. In the last section we define multiplicities and present the desired
version of the ”Fundamental Theorem of Algebra” over the skew-field of quaternions.

Translated from Sovremennaya Matematika i Ee Prilozheniya (Contemporary Mathematics and Its Applica-
tions), Vol. 59, Algebra and Geometry, 2008.

1072–3374/09/1606–0843 
c 2009 Springer Science+Business Media, Inc. 843
1. Preliminary Remarks
We consider the quaternion skew-field H as a four-dimensional vector space R4 over the field of real
numbers R
H = {(a, b, c, d) | a, b, c, d ∈ R}. (1.1)
The generators (unit vectors) we denote by
e = (1, 0, 0, 0), i = (0, 1, 0, 0), j = (0, 0, 1, 0), k = (0, 0, 0, 1). (1.2)
Here e is a neutral element for multiplication. We denote it by 1, and instead of a · 1 we write simply
a; then any quaternion can be written in the form a + bi + cj + dk, where a, b, c, d ∈ C. The unit
vectors i, j, and k are sometimes called imaginary units. The multiplication in H is defined by the
famous rules found by R. Hamilton:
ij = −ji = k, jk = −kj = i, ki = −ik = j. (1.3)
Note that quaternions of the type (a, 0, 0, 0), a ∈ R, are of special interest. Multiplication by them
coincides with multiplication by a scalar a, and the mapping a → (a, 0, 0, 0) is the monomorphism
of the field R in H. Therefore, it is natural to identify such quaternions with the corresponding real
numbers and assume that the field R is contained in H.
An arbitrary quaternion α can be considered as a sum of two quaternions:
α = a + (bi + cj + dk). (1.4)
Here a is called the real part of the quaternion; we denote it by Re(α); bi + cj + dk is the “imaginary”
part, which is denoted by Im(α). For every quaternion α = a+bi+cj+dk, the quaternion a−bi−cj−dk
is called conjugate, and we denote it by α.
It is not difficult to see that the mapping α → α is an involutory anti-automorphism and that,
multiplying α by α according to (1.3), we obtain
αα = (a + bi + cj + dk)(a − bi − cj − dk) = a2 + b2 + c2 + d2 . (1.5)
The latter expression is denoted by Nr(α) and is called the norm of quaternion α:
Nr(α) = a2 + b2 + c2 + d2 . (1.6)
As (1.5) shows, the norm of the quaternion
 is a real, nonnegative number which is equal to zero only
for zero quaternion. The number Nr(α) is called the modulus of the quaternion α, and we denote
it by |α|. Then Eq. (1.5) can be rewritten as αα = |α|2 .
The pure imaginary quaternions bi + cj + dk are characterized by the condition that their squares
are nonpositive real numbers. For any nonzero quaternion α, there exists an inverse element α−1
which can be expressed as
α
α−1 = .
Nr(α)
Thus, quaternions form a system with division, i.e., the equations
αq = β (1.7)
and
qα = β, (1.8)
−1 −1
where q is an unknown quaternion, possess the solutions ql = α β and qr = βα , respectively.
Moreover, the quaternion norm is multiplicative, i.e.,
Nr(α · β) = Nr(α) Nr(β).
In the sequel, we also need the following fact.
Proposition 1.1 (see [33]). Every quaternion satisfies a polynomial equation with real coefficients.

844
More precisely, it can be verified directly that the quaternion α = a + bi + cj + dk satisfies the
quadratic equation with real coefficients:

q 2 − 2aq + a2 + b2 + c2 + d2 = q 2 − 2 Re(α)q + Nr(α) = 0. (1.9)

The polynomial
fα (q) = q 2 − 2 Re(α)q + Nr(α) (1.10)
is called the characteristic polynomial of the quaternion α and is an irreducible quadratic trinomial
from the ring of polynomials R[q]. The converse statement is also valid: if g(q) = q 2 + 2tq + s is a
quadratic trinomial with a negative discriminant, then any quaternion β = a + b i + c j + d k, for
which a = Re(β) = −t, and Nr(β) = s, is a root of the polynomial fα (q).
Thus there are infinitely many quaternions that are roots of such a quadratic trinomial, and it will
be shown below that the roots of polynomial (1.10) form a two-dimensional sphere S 2 . Thus, unlike
the well-known situation in a field, where an nth degree polynomial may have not more than n roots
in virtue of the Bezout theorem, a polynomial over H may have infinitely many roots. The following
proposition collects the most important properties of quaternions, which are easy to verify directly.

Proposition 1.2 (see [33]). Let α, β, and γ be quaternions from H. Then the following relations
hold :
(1) αα = αα, i.e. |α| = |α|;
(2) |α + β| ≤ |α| +|β| and |αβ| = |βα|= |α| · |β|;
(3) |α|2 + |β|2 = 12 |α + β|2 + |α − β|2 ;
(4) ∃u ∈ H such that Nr(u) = 1 and α = |α|u;
(5) jc = cj or jcj = c for any complex number c;
(6) αiα = (a2 + b2 − c2 − d2 )i + 2(−2ad + bc)j + 2(ac + bd)k if α = a + bi + cj + dk;
(7) α = 12 (α + α) + 12 (α + iαi) + 12 (α + jαj) + 12 (α + kαk), α = − 12 (α + iαi + jαj + kαk);
(8) α2 = | Re(α)|2 − | Im(α)|2 + 2| Re(α)|| Im(α)|;
(9) in general, (α + β)2 = α2 + 2αβ + β 2 ;
(10) α = α if and only if α ∈ R;
(11) any quaternion α can be represented in the form α = c1 + c2 j, where c1 , c2 ∈ C. In the general
case, for α = a + bi + cj + dk we have the representation α = (a + bi) + (c + di)j.

Definition 1.1 (see [33]). We say that α is similar to β (α ∼ β), where α, β ∈ H, if there exists
u ∈ H, u = 0, such that α = uβu−1 .

Obviously, α and β are similar if and only if, for some quaternion v with Nr(v) = 1, the condition
v −1 αv = β is fulfilled, and the both similar quaternions have the same norms. The similarity is the
equivalence relation in H and, consequently, we can speak about a class of equivalence [α] which is
formed of quaternions similar to α. The next lemma follows from the above remarks.

Lemma √ 1.1 (see [33]). If α = a + bi + cj + dk, then α and a + b2 + c2 + d2 i are similar, i.e.,
α ∈ [a + b2 + c2 + d2 i].

From the latter lemma we immediately get a convenient criterion of similarity.

Proposition 1.3 (see [33]). Let q = a + bi + cj + dk and q  = a + b i + c j + d k be quaternions. Then


q ∼ q  if and only if a = a and b2 + c2 + d2 = b 2 + c 2 + d 2 .

It is now clear that if q = bi + cj + dk and q  = b i + c j + d k are pure imaginary quaternions, and


b2 + c2 + d2 = b2 + c2 + d2 = 1, then q ∼ q  . In particular, all elements on the unit sphere S02 of pure
imaginary quaternions are similar to each other.

845
2. Some Classes of Quaternion Polynomials
Because of noncommutativity, over the quaternion body one can consider various natural classes of
polynomials in one variable depending on whether a variable commutes with polynomial coefficients or
not. General polynomials are defined as finite sums of noncommutative monomials of type αqβq . . . qγ.
The degree of such a monomial is called the number of appearances of the symbol (variable) q in it.
The algebraic degree of the polynomial is defined as the maximal degree of monomials contained in
it. If we have only one monomial of maximal degree, then the polynomial is called monic.
It is known that a general quaternion polynomial may have no roots (see, e.g., [8]). Hence, for a set
of roots to be nonempty, it is necessary to restrict ourselves to the class of polynomials for which the
“fundamental theorem of algebra” is fulfilled. From this point of view, the class of monic polynomials
is more suitable because they always have roots [8]. However, the set of roots of such a polynomial
may be infinite, and it is a priori unclear how to give a useful description to it.
Therefore, it makes sense to restrict ourselves to the consideration of a still more narrow class of
polynomials, the so-called canonical polynomials. In the sequel, we will be basically concerned with
that class of polynomials. In should be added that besides their naturality, polynomials of that class
are used in mathematical physics and in the theory of quaternion differential equations [6]. Below we
will present the definition and description of the basic algebraic properties of such polynomials.
Let us consider the polynomial of one variable with coefficients from H
n

P (q) = ξm q m , where ξi ∈ H and ξn = 1. (2.1)
m=0

In the sequel, we only work with such polynomials and call them canonical polynomials.
According to S. Eilenberg and I. Niven, the “fundamental theorem of algebra” holds for canonical
quaternion polynomials [8]. Their proof was based on the properties of canonical polynomials con-
sidered as mappings of H into itself. The following statement is crucial for their approach and will
also be necessary in our considerations. Recall that a mapping is proper if the full pre-image of any
compact set is compact.

Lemma 2.1 (see [8]). A canonical quaternion polynomial defines a proper mapping of H into H.

Since for a proper self-mapping of affine space the notion of topological degree (mapping degree) is
well defined, the above lemma enables one to use degree theory and establish the existence of roots
for canonical polynomials.

Theorem 2.1 (see [8]). For every ξi ∈ H, i = 0, n − 1, there exists q0 ∈ H such that P (q0 ) = 0, where
the polynomial P is given by formula (2.1).

Given this result, we are guaranteed that the zero-set X(P ) is not empty and we can start investi-
gating its geometric and topological structure. To this end, it appears useful to consider the algebraic
properties of coefficients of the polynomial P .

Definition 2.1. The centralizer of a quaternion q is defined as the set


 
Z(q) = q  ∈ H : qq  = q  q .

A useful description of the centralizer is given by the following lemma.

Lemma 2.2. For every quaternion q = a + bi + cj + dk,


 b c d
Z(a + bi + cj + dk) = q  = a + b i + c j + d k :  =  =  .
b c d

846
Proof. Let q = a + bi + cj + dk, q  = a + b i + c j + d k and qq  = q  q. This equality results in the
following system of equations:
⎧ 

⎪ aa − bb − cc − dd = a a − b b − c c − d d,

⎨ba + ab − dc + cd = a b + b a + c d − d c,

⎪ ca + db − bd + ac = a c − b d + c a + d b,

⎩ 
da − cb + bc + ad = a d + b c − c b + d a,

from which it follows that


b c d

=  = .
b c d
The lemma is proved.

To obtain results on the structure of the zero-set X(P ), we consider centralizers of coefficients of
polynomial (2.1) Z(ξ) ⊆ H, i = 0, n and introduce the important notion of the polynomial centralizer.

Definition 2.2 (see [31]). The centralizer of a canonical polynomial P is defined as the set
n−1

ZP = Z(ξi ).
i=0

Lemma 2.3. For every η ∈ ZP and q0 ∈ P −1 (0), the quaternion ηq0 η −1 is a root of P .

Proof. Let q0 ∈ P −1 (0) and η ∈ ZP , then

n
 n

ξm (ηq0 η −1 )m = ξm (ηq0 η −1 )(ηq0 η −1 ) · · · (ηq0 η −1 )
  
m=0 m=0
m times
n
 n 
n 
= ξm (ηq0m η −1 ) = η(ξm q0m )η −1 = η ξm q0m η −1 = ηP (q0 )η −1 = 0,
m=0 m=0 m=0

i.e., ηq0 η −1 is a root of the polynomial P .


If q0 ∈ P −1 (0) and q0 ∈ Z(ZP ), then ηq0 η −1 = q0 ηη −1 = q0 , i.e., in this case we do not obtain new
roots. If q0 ∈ X(P ) \ Z(ZP ), then every η ∈ ZP performs “cloning” of the root q0 as in Lemma 2.3.
Obviously, dim(ZP ) may have values 1, 2, or 4. It is easy to indicate examples in which all possible
values of dim(ZP ) are realized. As we will see below, the dimension of the centralizer gives certain
information on the structure of zeroes of the polynomial in question.

3. Zero-Sets of Canonical Polynomials


It is well known that the zero-set of a canonical quaternion polynomial can be infinite, and so the
first task in its topological investigation is to find out how many connected components it may have
and what are the homeomorphy types of those components. Elementary examples like P1 (q) = q 2 − 1
and P2 (q) = q 2 + 1 show that the components of a zero-set can be homeomorphic to a point (as for P1 )
or to a two-dimensional sphere (as for P2 ). It turns out that these are the only possible topological
types of the components, and to show this is our next goal.
When studying the structure of the zero-set of a canonical polynomial, it is reasonable to consider
separately three cases according to the centralizer dimension of the polynomial under consideration.

847
3.1. Polynomials with real coefficients (dim(ZP ) = 4). In this section, we will consider quater-
nion polynomials with real coefficients, i.e., polynomials of the type
P (q) = q n + rn−1 q n−1 + · · · + r0 , (3.1)
where ri ∈ R, i = 0, n − 1, q ∈ H.
As is known, any polynomial with real coefficients can be decomposed into a product of the first-
and second-degree polynomials, and every second degree multiplier is clearly not decomposable (into a
product of two first-degree multipliers), i.e., the discriminant of the corresponding quadratic equation
is negative. According to Proposition 1.1, there exist infinitely many quaternions satisfying a quadratic
equation with negative discriminant, and all of them lie in a two-dimensional sphere (cf. [33]).
Proposition 3.1. If a quaternion polynomial with real coefficients has at least one nonreal root, then
there exist infinitely many quaternion roots.
Proof. By Lemma 2.3, if q0 ∈ P −1 (0), then for every η ∈ ZP one has ηq0 η −1 ∈ P −1 (0). If q0 ∈ R,
then it is easy to show that every η ∈ ZP R gives a new root q0 of the polynomial P . Hence there
exist infinitely many roots.
Theorem 3.1. The zero-set of a quaternion polynomial
P (q) = q n + rn−1 q n−1 + · · · + r0 , (3.2)
where ri ∈ R, i = 0, n − 1, consists of t points, where t is the number of real roots of P and of s
two-dimensional spheres, and the inequality t + 2s ≤ n is valid.
Proof. As was said above, the polynomial P (q) can be decomposed as follows:
P (q) = P1 (q) · P2 (q) · · · Pk (q), (3.3)
where each of the multipliers on the right-hand side is either a first-degree polynomial or a second
degree indecomposable polynomial. Let decomposition (3.3) of the polynomial P (q) have t first-degree
factors, or in other words let the polynomial P (q) have t real roots. Then the number of different
quadratic polynomials among Pi (q) is not more than n−t
2 = s. Since every quadratic trinomial provides
us with a whole sphere of quaternion roots, we can conclude that the set of zeros of quaternion
polynomial (3.3) consists of t points and s two-dimensional spheres, where t + 2s ≤ n.
Actually, one can assign multiplicity to each component of the zero-set in such a way that their
sum is equal to the algebraic degree of polynomial P . We do not discuss this here because in order
to define multiplicities we need some further considerations. An appropriate definition of multiplicity
will be given in the last section.
Corollary 3.1. If dim(ZP ) = 4, then the set P −1 (0) is finite if and only if all roots of the polynomial
P are real.
Proof. Let P be a polynomial with real coefficients and let the set P −1 (0) be finite. Then from
Proposition 3.1 it follows that all its roots are real. Now, on the contrary, if all roots of the polynomial
are real, then from Theorem 3.1 it follows that P −1 (0) consists of n points, i.e., it is finite.
We illustrate the above results by considering an equation of the type
q n = b, (3.4)
where b ∈ R. It is clear that in this case dim(ZP ) = 4. Hence according to the discussion above, if
the equation q n = b has at least one nonreal root, then there are infinitely many quaternion roots.
The structure of quaternion roots of unity is completely known [5], and so we arrive at the following
conclusion.
Proposition 3.2. The solution of the equation q n = b, where b ∈ R, consists of:
(1) m − 1 spheres for n = 2m and b > 0;

848
(2) m spheres for n = 2m and b < 0;
(3) one point and m − 1 spheres for n = 2m − 1.
3.2. Polynomials with commuting coefficients (dim(ZP ) = 2). Now let us consider polyno-
mials for which dim(ZP ) = 2. There naturally arises the question: under which coefficients of the
quaternion polynomial is there dim(ZP ) = 2. To answer this question, we consider subalgebras in H
isomorphic to C.
Proposition 3.3 (see [5]). A set Kα of elements of the type m·1+nα, where α ∈ H\R and m, n ∈ R,
forms a subalgebra which is isomorphic to the algebra of complex numbers.
Proof. As is known, every quaternion α satisfies certain quadratic equation with negative discriminant.
Let this equation be of the type
q 2 + sq + t = 0. (3.5)
2
This implies that α = −sα − t · 1, and hence the set of elements of the type m · 1 + nα is closed
with respect to the multiplication. Thus the above-mentioned set Kα is a hyper-complex system of
2
dimension 2, while for s4 − t < 0 (the negativeness of the discriminant) such system is isomorphic to
the system of complex numbers [5].
Let there be given a canonical quaternion polynomial of the form
n
P (q) = αm q m , where αn = 1, αi ∈ H, i = 0, n − 1.
m=0

Proposition 3.4. The dimension of the centralizer ZP is equal to 2 if and only if αt αs = αs αt , where
t, s = 0, n − 1.

n−1
Proof. Let dim(ZP ) = 2. By definition, ZP = Z(αi ), i.e., ∀ξ ∈ ZP , and we have
m=0

ξαt = αt ξ, i = 0, n − 1. (3.6)
From Lemma 2.2 it follows that if ξ = a + bi + cj + dk, then
αt = at + bt i + ct j + dt k = at + λt bi + λt cj + λt dk, where t = 0, n − 1.
Thus, we have that for any αs and αt ,
bs cs ds λs
= = = ,
bt ct dt λt
the imaginary parts of quaternions αs and αt are proportional, and this implies that they commute.
Now conversely, let αt αs = αs αt , where t, s = 0, n − 1. Suppose that dim(ZP ) = 2. Indeed, let
αs = as + bs i + cs j + ds k, where s = 0, n − 1. Then since the coefficient αt , where t = s, t = 0, n − 1,
commutes with αs , we have
αt = at + λt bs i + λt cs j + λt ds k = (at − λt as ) + λt (as + bs i + cs j + ds k) = λt · 1 + λt αs , (3.7)
where by λtwe denote the expression at − λt as . It follows from (3.7) that αt ∈ Kα∼ , t = 0, n − 1,
i.e., by Proposition 3.3, all coefficients of the polynomial P lie in a two-dimensional plane, isomorphic
to C.
Corollary 3.2. The centralizer ZP of the polynomial P is of dimension 2 if and only if it is isomorphic
to the algebra of complex numbers.
By means of the above corollary we can determine the structure of the zero-set of quaternion
polynomial P for dim(ZP ) = 2. Indeed, a polynomial P with subalgebra ZP ∼ = C can be decomposed
into linear factors:
P (q) = (q − ξ1 )(q − ξ2 ) · . . . · (q − ξn ), (3.8)

849
where ξi ∈ P −1 (0), i = 1, n.
Among ξi , i = 1, n, let there exist complex-conjugate pairs. Then we have
(q − ξ)(q − ξ) = q 2 − (ξ + ξ)q + ξξ = q 2 − 2 Re(ξ) + Nr(ξ),
i.e., we obtain the characteristic trinomial of the quaternion ξ. Thus
P (q) = (q − ξ1 )(q − ξ2 ) · . . . · (q − ξn−2 )(q 2 − 2 Re(ξ) + Nr(ξ)).
Since the quaternion zero-set of the equation q 2 − 2 Re(ξ) + Nr(ξ) = 0 is the whole sphere S 2 , we
find that the set of roots of polynomial P consists of n − 2 points and one sphere. In a general case,
the number of conjugate pairs in decomposition (3.8) of the polynomial P coincides with the number
of spheres in the set P −1 (0). Thus, we have determined the possible structure of roots in this case.
Theorem 3.2. If dim(ZP ) of the quaternion polynomial P is equal to 2, then the set P −1 (0) consists
of t points and s two-dimensional spheres, where t + 2s ≤ n.
Some words may be added about quaternion roots. Consider the equation q n = b, where b ∈ / R
(the case b ∈ R has been discussed above). As is known, the number of roots of the above equation
is equal to n (see [5]). Moreover, we can show that all these roots lie in the plane R{1, b}. Indeed,
if in a plane different from R{1, b}, there exists c such that cn = b, then this c must be real because
these two planes intersect with respect to the real line, which is a contradiction. Thus the structure
of quaternion roots is clear in this case as well.
3.3. Polynomials with one-dimensional centralizer (dim(ZP ) = 1). Consider the general case
of canonical quaternion polynomial, i.e., the case in which the coefficients of the polynomial are not
pairwise commuting. Thus we are given a polynomial
n

P (q) = αm q m , where αi ∈ H, i = 0, n, αn = 1. (3.9)
m=0
Following our reasoning, division of the given polynomial by the characteristic polynomials of the
quaternions plays an important role. Let Pξ be a characteristic polynomial of the quaternion ξ.
Lemma 3.1. For any quaternion polynomial P (q) with deg(P (q)) ≥ 2 and any ξ ∈ H, there exist
polynomials Q(q) and L(q) such that
P (q) = Q(q)Pξ + L(q), (3.10)
and either deg(L(q)) ≤ 1 or L(q) ≡ 0. The polynomials Q(q) and L(q) satisfying that equation are
defined uniquely.
Since coefficients of each characteristic polynomial are real, there is no need to prove this lemma
because its proof is identical to that of the theorem on the divisibility of two polynomials with real
coefficients (see [19]). We also need to introduce the notion of conjugate polynomial for (3.9). Denote
by P a polynomial of the type
n
P (q) = αm q m ,
m=0
i.e., the polynomial which is obtained from P by replacing the coefficients αi , i = 0, n, by their
conjugates αi .
Determine now an auxiliary polynomial with real coefficients that will allow us to investigate the
roots of the given polynomial. Namely, we put that N (P ) = P · P , where the polynomial N (P ) is
obtained by multiplying P by P according to the rule that the unknown q commutes with coefficients
αi and αi , i = 0, n.
Definition 3.1. The polynomial N (P ) is called the quasi-norm of P .

850
Lemma 3.2. The quasi-norm N (P ) of an arbitrary canonical polynomial P is a polynomial with real
coefficients of degree 2 deg(P ).
Proof. Consider coefficients of the product P · P :
2n
 2n

m
P (q) · P (q) = βm q , βm = αk αm−k .
m=0 k=0
In more detail:
β0 = α0 α0 ,
β1 = α0 α1 + α1 α0 ,
β2 = α0 α2 + α1 α1 + α2 α0 ,
..
.
β2n−1 = α0 α2n−1 + α1 α2n−2 + · · · + αn−1 αn + αn αn−1 + · · · + α2n−2 α1 + α2n−1 α0 ,
β2n = α0 α2n + α1 α2n−1 + · · · + αn αn + · · · + α2n−1 α1 + α2n α0 .
Thus, for every coefficient with even index we have a sum of summands of the type αα = Nr(α) ∈ R
and αβ + βα = αβ + αβ ∈ R, while coefficients with odd index consist of a sum of summands of the
type (αβ + βα ∈ R. It is now obvious that N (P ) is a polynomial with real coefficients.
For example, let P (q) = q 2 + iq + j; then P (q) = q 2 − iq − j and P P = (q 2 + iq + j)(q 2 − iq − j) =
q4 + q 2 + 1, i.e., we obtain a fourth-degree polynomial with real coefficients.
Lemma 3.3. Let a polynomial P and a quaternion ξ be given. Then we have a dichotomy: either
Pξ divides P and then the whole [ξ] consists of roots of P , or there is no more than one root of P
in [ξ], where Pξ is the characteristic polynomial of the quaternion ξ, where [ξ] is the class of similar
quaternions ξ.
Proof. If Pξ divides the polynomial P , then we have P (q) = Q(p)Pξ , i.e., all roots of the polynomial
Pξ are the roots of P as well, i.e., [ξ] ⊂ P −1 (0).
Now let P be indivisible by Pξ . Then we divide P by Pξ with a remainder. According to Lemma 3.1,
this is always possible, and we have
P (q) = Q(q)Pξ (q) + L(q), (3.11)
where L(q) is a first-degree polynomial, i.e., L(q) = αq + β, where α, β ∈ H, and at least one of α, β
is different from zero. Moreover, it is clear that if α = 0, β = 0, then the class [ξ] has no one root.
Therefore, in the sequel it will be assumed that α = 0.
Now let ξ0 ∼ ξ and ξ0 be a root of the polynomial P . Then we have
P (ξ0 ) = Q(ξ0 )Pξ (ξ0 ) + αξ0 + β.
Since ξ0 ∈ [ξ], Pξ (ξ0 ) = 0, i.e., 0 = 0 + αξ0 + β, we find that ξ0 = −α−1 β.
Thus, if the polynomial P is indivisible by Pξ , then from the class [ξ] it may have only one root
ξ0 = −α−1 β (by Lemma 3.1, representation (3.11) is unique).
Corollary 3.3. The set of roots of a polynomial P is infinite if and only if there exists ξ ∈ H such
that P is divisible by Pξ .
For example, consider P = q 2 + iq + j and ξ = 12 − 12 i − 12 j + 12 k. Then Pξ = q 2 − q + 1. We divide
the polynomial P by Pξ , and find that P (q) = q 2 − q + 1 + (i + 1)q + j − 1, i.e., P (q) = Pξ + αq + β,
where α = i + 1 and β = j − 1. By Lemma 3.3, [ξ] does not consist entirely of the roots of P . Consider
1−i 1 1 1 1
ξ0 = −α−1 β = − (j − 1) = − i − j + k.
2 2 2 2 2

851
Substituting the obtained quantity ξ0 in P , we obtain P (ξ0 ) = 0, i.e., in the class [ξ] we have exactly
one real root ξ0 = ξ.
Lemma 3.4. If ξ is a root of the polynomial P , then the characteristic polynomial P (ξ) of the quater-
nion ξ divides N (P ).
Proof. Let ξ ∈ P −1 (0). We divide the polynomial P by P (ξ), and obtain
P (q) = Q(q)Pξ + L(q). (3.12)
Applying to the above equation the conjugation, we have
P (q) = Q(q)Pξ + L(q). (3.13)
Multiplying P (q) by P (q), we find that
N (q) = P · P = Q(q)Q(q)Pξ2 + Q(q)L(q)Pξ + L(q)Q(q)Pξ + L(q)L(q),
whence
 
N (q) = Q(q)Q(q)Pξ2 + Q(q)L(q) + L(q)Q(q) Pξ + L(q)L(q). (3.14)
Now let us show that the summand L(q)L(q) is likewise divisible by P (ξ). In Eq. (3.12), deg L(P ) ≤
1, or L(P ) ≡ 0. The case L(P ) ≡ 0 is trivial, and in this case, we even have P (ξ)2 /N (P ). Let
L(P ) = αq + β, then L(P ) = αq + β. Consider
 
2 2 αβ + βα ββ
L(P )L(P ) = (αq + β)(αq + β) = ααq + (αβ + βα)q + ββ = αα q + + ,
αα αα
i.e.,
 
2 αβ + βα ββ
L(P )L(q) = αα q + q+ . (3.15)
αα αα
By the condition of the lemma, ξ ∈ P −1 (0). From Lemma 3.3 it follows that ξ is of the type
ξ = −α−1 β. Let us show that
αβ + βα ββ
Pξ = q 2 + + (3.16)
αα αα
Indeed,

− 2 Re(ξ) = −2 Re(−α−1 β) = 2 Re(α−1 β) = α−1 β + α−1 β


αβ αβ αβ + βα
= + =
Nr(α) Nr(α) αα
and

Nr(−α−1 β) = Nr(α−1 β) = (α−1 β)(α−1 β) = α−1 ββ α−1


α α Nr(β) ββ
= Nr(β) = = .
Nr(α) Nr(α) Nr(α) αα
Thus, L(P )L(P ) is divisible by Pξ .
Theorem 3.3. The zero-set of a canonical quaternion polynomial
n

P (q) = αm q m , αi ∈ H, i = 0, n, αn = 1,
m=0
n−t
consists of t isolated points and s ≤ 2 two-dimensional spheres, i.e., the inequality t + 2s ≤ n is
valid.

852
Proof. From the results obtained in this section it follows that for the given theorem to be proven
completely, it is sufficient to show that the inequality t + 2 ≤ n is valid. Thus let ξ ∈ H. If Pξ divides
the polynomial P , then [ξ] ∼= S 2 ⊂ P −1 (0) and N (P ) is divisible by Pξ2 . However, if Pξ does not
divide P , then the polynomial P from [ξ] may have no more than one root, and if it exists in [ξ], then
N (P ) is divisible by Pξ .
Next, if the number of such ξ that are isolated roots of the polynomial P is equal to t, and the
number of such ξ  for which [ξ  ] ⊂ P −1 (0) is equal to s, then applying the method of induction to the
polynomial N (P ), we easily get 2t + 4s ≤ 2n, and thus t + 2s ≤ n, which completes the proof.

4. Multiplicities of the Components of the Zero-Set


As is shown, the set of roots of a canonical polynomial always consists of a finite number of points
and two-dimensional spheres. Thus the set of roots possesses connected components of two types: zero-
dimensional components (points) and two-dimensional ones (spheres). It is seen from the above proof
that the number of components of each type does not exceed the algebraic degree of the polynomial. In
fact, a more precise statement will be established in this section. As a pattern we take the well-known
statement that the sum of multiplicities of all roots of a complex polynomial is equal to its algebraic
degree. Our ultimate goal is to get an analogous statement for quaternion polynomials.
To this end we have to introduce the notion of multiplicity of a component of the set of roots.
Notice that for a component consisting of one point one can apply the well-known definition of the
multiplicity of polynomial mapping at a point as the dimension of the local algebra of mapping at
that point [25, 26]. The geometrical meaning of such a definition for complex polynomial mappings
is well known from algebraic geometry: if it equals m, then under small perturbations of the initial
mapping this root decomposes into m simple (regular) roots where the Jacobian is nonzero [26]. Thus
one can consider a canonical quaternion polynomial as a pair of complex polynomials in two variables
and apply this definition. However, such an approach causes problems since a quaternion polynomial
can be “(de)complexified” in different ways, and some effort is needed to obtain a well-defined notion.
For continual components of roots, an analogous definition in terms of the structural sheaf of the
zero-set requires still more preparatory work involving a number of nontrivial notions of algebraic
geometry, and so it is rather difficult to describe and use it. Therefore we prefer a simple ad hoc
definition of algebraic nature, in which use will be made of the results obtained by us on the divisibility
of the polynomial quasi-norm by the characteristic polynomial of a root.
This definition takes into account the specific properties of canonical quaternion polynomials and
allows us to get the desired statement about the sum of multiplicities of the components. With
such an approach the final result may seem tautological, but in fact the following definition becomes
meaningful and useful only due to the preceding results on the structure of quasi-norm and roots of a
quaternion polynomial.
Definition 4.1. The algebraic multiplicity of an isolated root of the polynomial P is defined as the
exponent with which the characteristic polynomial of the given root is involved in the factorization of
the quasi-norm N (P ). The algebraic multiplicity of a two-dimensional component of the zero-set is
defined as the half of the exponent with which the characteristic polynomial of the given component
is involved in the factorization of the quasi-norm N (P ).
The preceding results show that this definition is meaningful. The number obtained in such a way
may be called the algebraic multiplicity of the component in question, in order to distinguish it from
the general definition of the geometric multiplicity of a connected component of a proper polynomial
mapping suggested in [2]. With our definition we immediately arrive at a version of the “fundamental
theorem of algebra” for quaternions. Indeed, from our definition of multiplicity it follows that the
sum of multiplicities is equal to one-half of the algebraic degree of the quasi-norm N (P ). Since the
algebraic degree of N (P ) is twice the algebraic degree of P we obtain the desired result.

853
Theorem 4.1. For any canonical quaternion polynomial P , the sum of algebraic multiplicities of all
components of its zero-set ZP is equal to the algebraic degree of P .
Sounding as nice as it does, this result may not seem satisfactory because the definition of multi-
plicities is rather narrow and technical. However, this circumstance can be remedied by making use
of the general definition of multiplicity for a component of a zero-set of a proper polynomial mapping
suggested in [2]. It can be shown that an analog of Theorem 4.1 remains true for the geometric
multiplicities of the components of zero-set, but such developments require a separate discussion.
In conclusion we would like to add that in concrete situations it is often important to know how many
geometrically different components of both types does the set of roots of a given quaternion polynomial
have (see, e.g., [6]). In particular, special interest lies in counting continual components (spheres) in
the set of roots. For example, this kind of information is necessary for a series of problems of the
qualitative theory of quaternionic ordinary differential equations [7]. Such problems can be effectively
treated using the above results and the signature method for counting points in semi-algebraic sets
developed in [15]. The author plans to work out this issue in forthcoming publications.

REFERENCES
1. V. Arnol’d, A. Varchenko, and S. Gusein-zade, Singularities of Differentiable Mappings [in Rus-
sian], Vols. 1, 2, Nauka, Moscow (1982, 1984).
2. L. Blum, F. Cucker, M. Shub, and S. Smale, Complexity and Real Computation, Springer-Verlag
(1997).
3. J. Bochnak, J. Cost, and M.-F. Roy, Géométrie Algébrique Réelle, Ergeb. Math., 12, Springer-
Verlag (1987).
4. J. Bruce, “Euler characteristics of real varieties,” Bull. Lond. Math. Soc., 22, 213–219 (1990).
5. I. L. Cantor and A. S. Solodovnikov, Hypercomplex Numbers [in Russian], Moscow (1973).
6. S. De Leo and G. C. Ducati, “Quaternionic differential operators,” J. Math. Phys., 42, 2236–2265
(2001).
7. S. De Leo and G. C. Ducati, “Solving simple quaternionic differential equations,” J. Math. Phys.,
44, 2224–2233 (2003).
8. S. Eilenberg and I. Niven, “Fundamental theorem of algebra for quaternions,” Bull. Am. Math.
Soc., 50, 246–248 (1944).
9. D. Eisenbud and H. Levine, “An algebraic formula for the degree of a C ∞ map germ,” Ann.
Math., 106, 19–44 (1977).
10. F. Gantmakher, Theory of Matrices [in Russian], Nauka, Moscow (1966).
11. P. Griffiths and J. Harris, Principles of Algebraic Geometry, Wiley (1978).
12. M. Hirsh, Differential Topology, Springer-Verlag (1978).
13. G. Khimshiashvili, “On the local degree of a smooth mapping,” Bull. Acad. Sci. Georgian SSR,
85, 309–312 (1977).
14. G. Khimshiashvili, “The Euler characteristic of manifold and critical points of smooth functions,”
Proc. A. Razmadze Math. Inst., 85, 123–141 (1982).
15. G. Khimshiashvili, “On the cardinality of a semi-algebraic subset,” Georgian Math. J., 1, 511–521
(1994).
16. G. Khimshiashvili, “Signature formulae for topological invariants,” Proc. A. Razmadze Math.
Inst., 125, 1–121 (2001).
17. G. Khimshiashvili, “Counting roots of quaternionic polynomials,” Bull. Georgian Acad. Sci., 165,
465–468 (2002).
18. A. Khovansky, “The index of a polynomial vector field,” Funkts. Anal. Prilozh., 13, 49–58 (1979).
19. S. Lang, Algebra, Addison-Wesley, New York (1965).
20. A. Lecki and Z. Szafraniec, “An algebraic method for calculating the topological degree,” Banach
Center Publ., 35, 73–83 (1996).

854
21. J. Milnor, Topology from the Differentiable Viewpoint, Virginia Univ. Press (1965).
22. J. Milnor, Singular Points of Complex Hypersurfaces, Ann. Math. Stud., 61, Princeton (1968).
23. J. Milnor and A. Wallace, Differential Topology [Russian translation], Mir, Moscow (1972).
24. I. Niven, “The roots of a quaternion,” Am. Math. Mon., 49, 386–388 (1942).
25. P. Orlik, “The multiplicity of a holomorphic mapping at an isolated critical point,” in: Real and
Complex Singularities. Proc. Nordic Summer School, Oslo (1977), pp. 409–460.
26. V. Palamodov, “On the multiplicity of a holomorphic mapping,” Funkts. Anal. Prilozh., 1, 54–65
(1967).
27. A. Serodio and L.-S. Siu, “Zeros of quaternion polynomials,” Appl. Math. Lett., 14, 237–239
(2001).
28. A. Serodio, E. Pereira, and J. Vitoria, “Computing the zeros of quaternion polynomials,” Comp.
Math. Appl., 42, 1229–1237 (2001).
29. E. Spanier, Algebraic Topology, McGraw-Hill, New York (1965).
30. Z. Szafraniec, “On the Euler characteristic of analytic and algebraic sets,” Topology, 25, 411–414
(1986).
31. N. Topuridze, “On the structure of the zero-set of a quaternionic polynomials,” Bull. Georgian
Acad. Sci., 164, 228–231 (2001).
32. N. Topuridze, “On the roots of polynomials over division algebras,” Georgian Math. J., 10, 745–
762 (2003).
33. F. Zhang, “Quaternions and matrices of quaternions,” Linear Alg. Appl., 251, 21–57 (1997).

N. Topuridze
Sukhumi Branch of I. Javakhishvili Tbilisi State University, Sukhumi, Georgia

855

Das könnte Ihnen auch gefallen