Cauchy-Riemann equations and

arXiv:1609.04193v1 [math.AG] 14 Sep 2016

Jacobians of quaternion polynomials

Takis Sakkalis and Sofia Douka

September 17, 2016

Abstract

A map f from the quaternion skew field H to itself, can also be thought as a
transformation f : R4 → R4 . In this manuscript, the Jacobian J(f ) of f is
computed, in the case where f is a quaternion polynomial. As a consequence,
the Cauchy-Riemman equations for f are derived. It is also shown that the
Jacobian determinant of f is non negative over H. The above commensu-
rates well with the theory of analytic functions of one complex variable.

Keywords: quaternion polynomials; Jacobians; CauchyRiemann equations

MCS 2010: 26B10, 12E15, 11R52

Takis Sakkalis, Mathematics Laboratory, Agricultural University of Athens, 75 Iera Odos,

Athens 11855, GREECE, corresponding author: e-mail stp@aua.gr
Sofia Douka, 37 Strymonos St., Athens 11855, GREECE, email sofia.nefeli@gmail.com

1
1 Introduction

Let H denote the skew field of quaternions. Its elements are of the form c =
c0 + ic1 + jc2 + kc3 , where cm ∈ R and i, j, k are such that i2 = j2 = k2 =
−1 and ij = −ji = k, jk = −kj = i, ki = −ik = j. The real part of c is
Re(c) = c0 while the imaginary part Im(c) = ic1 + jc2 + kc3 . The norm of c,
p
|c| = c20 + c21 + c22 + c23 , its conjugate c∗ = c0 − ic1 − jc2 − kc3 while its inverse is
c−1 = c∗ · |c|−2 , provided that |c| 6= 0. c is called an imaginary unit if Re(c) = 0
and |c| = 1, and it has the property c2 = −1. In that regard, H is a real normed
division (non commutative) algebra. An element c = c0 + ic1 + jc2 + kc3 of H can
also be represented via a real 4 × 4 matrix C as follows:

 
c0 −c1 −c2 −c3
 c1 c0 −c3 c2 
C=
 c2

c3 c0 −c1 
c3 −c2 c1 c0
Notice that |C| = |c|4 . The following notation will be frequently used in the
sequel:

Definition 1.1 For any 4 × 4 real matrix B and a quaternion c, we define:

cB ≡ CB and Bc ≡ BC.

We may identify H with R4 via the map (x + iy + jz + kw) → (x, y, z, w). Let f :
H → H. In view of this identification, we can also think of f as a map from R4 →
R4 . Indeed, if f (x+ iy + jz + kw) = f1 (x, y, z, w)+ if2 (x, y, z, w)+ jf3 (x, y, z, w)+
kf4 (x, y, z, w) we define f : R4 → R4 by f (x, y, z, w) = (f1 , f2 , f3 , f4 ). In that
case, we have
h i
∂fi
Definition 1.2 The Jacobian J(f )(c), c ∈ H is the matrix ∂xj , i = 1, 2, 3, 4,
x1 = x, x2 = y, x3 = z, x4 = w evaluated at c. The determinant of J(f ) will be
denoted by |J(f )|.

In this manuscript, we are concerned with the computation of J(f ) and its de-
terminant at a point c ∈ H, in the case where f is a polynomial with coefficients

2
in H. More specifically, we will consider polynomials of the form

f (t) = an tn + an−1 tn−1 + · · · + a0 , where ak ∈ H (1)

We will first show that |J(tn )| ≥ 0 at all points c ∈ H. By extending the above,
we will find the form of J(f ) at a complex number r +is (the complex number i is
to be identified with the quaternion i throughout this note). Using this we deduce
the form of J(f )(c), c ∈ H. From the above, the Cauchy-Riemman equations for
f are derived. Finally, we prove that |J(f )| ≥ 0 over H, a fact similar to the one
in the theory of analytic functions of a complex variable. The latter can be used
to compute the zeros ζk of f as well as its local degree at ζk .

2 A brief overview of quaternion polynomials

In this paragraph we will recall some facts, needed for the rest of the paper,
concerning quaternion polynomials. For more details, the reader is referred to
[2, 3, 4].

Due to the non commutative nature of H, polynomials over H are usually dis-
tinguished into the following three types: left, right and general, [3]. A left
polynomial is an expression of the form (1). If an 6= 0, n is called the degree
of f . Here we shall consider, unless otherwise stated, left polynomials only and
called them simply polynomials. If g(t) = bm tm + bm−1 tm−1 + · · · + b0 is another
polynomial, their product f g(t) is defined in the usual way:

m+n
X k
X
k
f g(t) = ck t , where ck = ai bk−i
k=0 i=0

Note that in the above setting the multiplication is performed as if the coefficients
were chosen in a commutative field.

An equivalent representation of f is f (t) = a(t)+ib(t)+jc(t)+kd(t) = a+ib+j(c−

id), where a, b, c, d ∈ R[t]. If c ∈ H, we define f (c) = an cn + an−1 cn−1 + · · · + a0 ;
if f (c) = 0, c is called a zero or a root of f . According to Theorem 1 of [3]

3
an element c ∈ H is a zero of f if and only if there exists a (left) polynomial
g(t) such that f (t) = g(t)(t − c). In their nominal paper [1], Eilenberg and
Niven, using a degree argument, proved the fundamental theorem of algebra for
quaternions, namely that any quaternion polynomial of positive degree n of the
form f (t) = a0 ta1 t · · · tan + φ(t), ai ∈ H, ai 6= 0 and φ(t) is a sum of finite number
of similar monomials b0 tb1 t · · · tbk , k < n, has a root in H.

Roots of f are distinguished into two types: (i) isolated and (ii) spherical. A
root c of f is called spherical if and only if its characteristic polynomial qc (t) =
t2 − 2t Re(c) + |c|2 divides f ; for any such polynomial, call αc ± iβ c its complex
roots. In that case any quaternion γ similar to c, is also a root of f ; (c1 , c2 ∈ H
are called similar, and denoted by c1 ∼ c2 , if c1 η = η c2 for a non zero η ∈ H).
For example, if f (t) = t2 + 1, any imaginary unit quaternion c = α1 i + α2 j + α3 k
is a root of f .

Remark 2.1 The polynomial f (t) has a spherical root if and only if it has roots
α + iβ, α − iβ, α, β ∈ R, β 6= 0.

The above facts, as well as the representation of f as in (1), allows us to factor

f into a product of linear factors (t − ci ), ci ∈ H. Indeed, since f (t) = g(t)(t − c)
and g(t) has a root, simple induction shows that

f (t) = an (t − cn )(t − cn−1 ) · · · (t − c1 ), cj ∈ H (2)

A word of caution: In the above factorization, while c1 is necessarily a root

of f , cj , j = 2, · · · n, might not be roots of f . For example, the polynomial
f (t) = (t + k)(t + j)(t + i) = t3 + (i + j + k)t2 + (−i + j − k)t + 1 has only one
root, namely t = −i. Theorem 2.1 of [2] provides a more detailed version of the
above factorization.

If we write f in the form f (t) = a(t) + ib(t) + jc(t) + kd(t) we see that f has no
spherical roots if and only if gcd(a, b, c, d) = 1. Such an f will be called primitive.
Then, it is known (Corollary 3.3, [4]) that a primitive f (t) of degree n, has at
most n distinct roots in H.

4
The conjugate f ∗ of f is defined as f ∗ = a(t) − ib(t) − jc(t) − kd(t). Note that
f ∗ f = a2 + b2 + c2 + d2 , which is a real positive polynomial. Observe that if α + iβ
is a root of f ∗ f , then there exists a c ∈ H, similar to α + iβ so that f (c) = 0.

Definition 2.1 Let φ(t) ∈ C[t] and ζ ∈ C be a root of φ. We denote by µ(φ)(ζ)

the multiplicity of ζ. Now let c ∈ H be a root of f and let m = µ(f ∗ f )(αc + iβ c ).
Then, if (1) c is isolated, we define its multiplicity µ(f )(c), as a root of f , to be
m; (2) if c is spherical, its multiplicity is set to be 2m.

Note that the above notion of multiplicity agrees with the one given in Definition
2.6 of [2], page 23. From the above we have:

Criterion 1 Let c be a root of the primitive polynomial f . Then c has multplicity

k if and only if gcd(qck , f ∗ f ) = qck .

3 The Jacobian determinant of tn

Eilenberg and Niven in [1], proved the fundamental theorem of algebra for quater-
nions, using a degree argument. The key ingredient was Lemma 2 whose proof was
depended on the positiveness of |J(tn )| at the roots of the equation tn = i. In this
section, we will show that |J(tn )| is non negative at any point (x, y, z, w) ∈ R4 .

First, we need a lemma.

Lemma 3.1 Let u, v, p, q ∈ R[x, y, z, w] be homogeneous polynomials of the same

degree m ≥ 1 so that yp = zv and yq = wv. Let

F : R4 → R4 , F (x, y, z, w) = (u, v, p, q)

Then, |J(F )| is equal to y −3 mv 2 (vux − uvx ).

Proof: Since yp = zv and yq = wv, we see that

z −z z v z z
p x = vx , p y = 2 v + vy , p z = + vz , p w = vw
y y y y y y
w −w w w v w
qx = vx , q y = 2 v + vy , q z = vz , q w = + vw
y y y y y y

5
Then, the Jacobian of F is
 
ux uy uz uw
 vx v y v z vw

 
 
 z −z z v z z 
J(F ) = 
 y vx y 2 v + y vy + vz vw  (3)
 y y y 

 
 w −w w w v w 
vx 2
v + vy vz + vw
y y y y y y

Now using the fact that u, v are homogeneous of degree m, an easy calculation
shows that |J(F )| = y −3 mv 2 (vux − uvx ).

For u, v, p, q a’s in Lemma 3.1, define new polynomials U, V, P, Q by the formula

U +iV +jP +kQ = (x+iy+jz+kw)(u+iv+jp+kq) = (u+iv+jp+kq)(x+iy+jz+kw)

Notice that U, V, P, Q are homogeneous of degree m + 1, yP = zV, yQ = wV and

U = xu − yv − zp − wq, V = yu + xv. Using the fact that yP = zV, yQ = wV we
observe that U takes the form
y 2 + z 2 + w2
 
U = xu − v
y

Corollary 3.1 Let U, V, P, Q be as above and define Φ = (U, V, P, Q). Then,

(m + 1)V 2 [y(x2 + y 2 + z 2 + w2 )(vux − uvx ) + y 2 u2 + v 2 (y 2 + z 2 + w2 )]
|J(Φ)| =
y4

Proof: We have
y 2 + z 2 + w2
 
Ux = u + xux − vx , Vx = yux + v + xvx
y
Now, a calculation shows that
y(x2 + y 2 + z 2 + w2 )(vux − uvx ) + y 2 u2 + v 2 (y 2 + z 2 + w2 )
V Ux − U Vx =
y
Thus, the result follows from Lemma 3.1.

Corollary 3.2 Let f (t) = tn = (x + iy + jz + kw)n = f1 + if2 + jf3 + kf4 =

(f1 , f2 , f3 , f4 ). Then, |J(f )| ≥ 0.

6
Proof: For induction purposes, we will rename f = tn = f [n] = (f [n]r ), r =
1, 2, 3, 4. Observe first that f [n]r are homogeneous of degree n and satisfy the
conditions yf [n]3 = zf [n]2 , yf [n]4 = wf [n]2 . If n = 1 we get |J(t)| = 1, while if
n = 2 we see that |J(t2 )| = 16x2 (x2 + y 2 + z 2 + w2 ) ≥ 0. Let now n ≥ 2. Then,
Corollary 3.1 shows that |J(tn+1 )| is equal to
[n]1 [n]
(n + 1)(f [n]2 )2 [y|f [1] |2 (f [n]2 fx − f [n]1 fx 2 ) + y 2 u2 + v 2 (y 2 + z 2 + w2 )]
y4
[n]1 [n]
From Lemma 3.1 we see that y(x2 + y 2 + z 2 + w2 )(f [n]2 fx − f [n]1 fx 2 ) ≥ 0 and
thus by induction the result follows.

4 The Jacobian of f (t)

In this section we will compute the Jacobian of a quaternion polynomial f (t).

First, we compute the Jacobian of a complex polynomial a(t) + ib(t), a, b ∈ R[t]
at the complex number t0 = r + is (in the sequel, for a complex polynomial φ,
we will denote its Jacobian over C by JC (φ) to differentiate it from the Jacobian
of φ over H). Using this, we will calculate J(f )(t0 ). Finally, utilizing the above
we compute the Jacobian of f at a point c0 ∈ H.

We begin with the following:

Lemma 4.1 For n ∈ N, define an = 2n − 1, bn = 2n and cn = (−1)n+1 . Then,

   
cn an 0 0 0 0 −cn bn 0 0
0 cn an 0 0 
 and J(t2n )(i) =  cn bn 0 0 0
J(t2n−1 )(i) = 
 
 (4)
 0 0 cn 0   0 0 0 0
0 0 0 cn 0 0 0 0

[n]
Proof: We will use the notation of Corollary 3.2. From (3) we see that fxj 3 (i) =
[n]
0 for xj = x, y, w and fxj 4 (i) = 0 for xj = x, y, z. Recall that

y 2 + z 2 + w2
 
[k+1]1 [k] [k]2
f = xf −f and f [k+1]2 = yf [k]1 + xf [k]2 (5)
y

7
By using induction and taking into account (5) we get

f [2n−1]1 (i) = 0, f [2n]1 (i) = (−1)n , f [2n−1]2 (i) = (−1)n+1 , f [2n]2 (i) = 0
 2 2 2
[2n+1]1 [2n] [2n] [2n]
Now fx = f [2n]1 + xfx 1 − fx 2 y +zy +w . By induction fx 2 (i) =
[2n+1]1
(−1)n+1 (2n) and thus fx (i) = (−1)n − (−1)n+1 (2n) = (−1)n+1+1 (2n + 1)
[2n−1]2 [2n]1
as required. Similarly, we get fy (i) = (−1)n+1 (2n − 1), fy (i) = −cn bn
[2n] [k]
and fx 2 (i) = cn bn . Finally, it is easy to see that fxj r (i) = 0 when r = 3, 4 and
xj = z, w. This finishes the proof.

Note that if s ∈ R, s 6= 0, J(tk )(is) = sk−1 J(tk )(i).

Corollary 4.1 Let g(t) = a(t) + ib(t), a(t), b(t) ∈ R[t] be a complex polynomial
and r + is ∈ C − R. Then,
 
α −β 0 0
 β α 0 0 
J(g)(r + is) = 
 0
, for α, β, γ, δ ∈ R (6)
0 γ −δ 
0 0 δ γ

Proof: Since J(g)(r + is) = J(g(t + r)(is), we may assume that r = 0. We write

g(t) = an t2n−1 + an−1 t2n−3 + · · · + a1 t + bm t2m + bm−1 t2m−2 + · · · + b1 t2 + b0 (7)

ai , bj ∈ C. Then,
Pn Pm
J(g)(is) = k=1 ak J(t2k−1 )(is) + l=1 bl J(t2l )(is)
Pn Pm
= k=1 ak s2k−2 J(t2k−1 )(i) + l=1 = bl s2m−1 J(t2l )(i)

Now use Lemma 4.1 to get

n
X m
X
α= (−1)k+1 s2k−2 (2k − 1)a1k − (−1)l s2l−1 (2l)b2l ,
k=1 l=1
n
X m
X
β= (−1)k+1 s2k−2 (2k − 1)a2k + (−1)l s2l−1 (2l)b1l ,
k=1 l=1
n
X n
X
γ= (−1)k+1 s2k−2 a1k and δ = (−1)k+1 s2k−2 a2k
k=1 k=1

8
where ak = a1k + ia2k , bl = b1l + ib2l .

Note
 that |J(g)(t) ≥ 0 for all t ∈ C. On the other hand, JC (g)(r + is) =
α −β
. Thus, |J(g)(r + is)| = |JC (g)(r + is)| · (γ 2 + δ2 ). In view of this
β α
it may happen that |JC (g)(r + is)| > 0 while |J(g)(r + is)| = 0. For example, if
g(t) = 2t5 + t3 − t + t2 + i(3t5 + t3 − 2t + 5) and r = 0, s = 1, α = −6, β = −12
but γ = δ = 0. Things, however, become more interesting when r + is is a root
of g. Indeed, we have:

Remark 4.1 For g, r +is as in Corollary 4.1, if |JC (g)(r +is)| > 0 and |J(g)(r +
is)| = 0, g is non primitive.

Proof: Without loss of generality, suppose that r + is = i. Write g as in (7).

Then, a(i) = iγ + m
P l 1 1
Pm l 2 2
l=1 (−1) bl + b0 and b(i) = iδ + l=1 (−1) bl + b0 . Since
0 = g(i) = a(i) + ib(i) we get a(i) = b(i) = 0. Thus, a(t), b(t) are both divisible
by t2 + 1 which makes g non primitive.

Now let f (t) be as in (1). We write f (t) = a(t) + ib(t) + j(c(t) − id(t)) =
g(t) + jh(t) a, b, c, d ∈ R[t].

Proposition 4.1 For r + is ∈ C − R, f, g, h as above,

 
α1 −β 1 −α2 β2
 β1 α1 β2 α2 
J(f )(r + is) = 
 α3 −β 3
, for αk , β k ∈ R (8)
α4 −β 4 
−β 3 −α3 β4 α4

Proof: We have J(f )(t) = J(g)(t) + jJ(h)(t). Note that jJ(h)(r + is) is equal
to
    
0 0 −1 0 α3 −β 3 0 0 0 0 −α2 β 2
 0 0 0 1   β3
  α3 0 0   0 0 β 2 α2 
 = 
 1 0 0 0  0 0 α2 −β 2   α3 −β 3 0 0 
0 −1 0 0 0 0 β2 α2 −β 3 −α3 0 0

Now the result follows from Corollary 4.1.

9
We finally proceed with the computation of J(f ) at t0 = x0 + iy0 + jz0 + kw0 ∈
H − C. Assuming, as usual, that x0 = 0, we will “push” t0 to is, with is ∼ t0 and
use Proposition 4.1 to find the said Jacobian.

To achieve the above, for a c ∈ H, |c| = 1, consider the transformation hc (t) =

ctc∗. This is a rotation in R4 , realized by an orthogonal matrix Ac in the sense
that hc (x + iy + jz + kw) = c(x + iy + jz + kw)c∗ = (h1 + ih2 + jh3 + kh4 ), then
h(x, y, z, w) = (h1 , h2 , h3 , h4 ) = Ac (x, y, z, w)T . On the other hand, observe that
ctc∗ commutes with itself and preserves the root structure of f (t). Indeed, if f (t)
is as in (1), consider a new polynomial φ(u),

φ(u) = cf (u)c∗ = can c ∗ un + can−1 c∗ un−1 + · · · + ca1 c∗ u + ca0 c∗ (9)

Note now that if c0 is a root of f , cc0 c∗ is a root of φ of the same structure.

Divide f by (t − t0 ) to get f (t) = g(t)(t − t0 ) + f (t0 ) and let γ ∈ H be so
that γt0 γ ∗ = is. Consider the map F (x, y, z, w) = (f ◦ hγ )(x, y, z, w). Then,
JF = J(f )(hγ ) · J(hγ ) = J(f )(hγ ) · Aγ . Therefore,

J(f )(t0 ) = J(f )(is)Aγ (10)

This along with formula (8) finishes the task of computing J(f )(t0 ).

4.1 Cauchy-Riemann equations

With the help from the previous section, we are now able to write Cauchy-
Riemann (CR) equations for quaternion polynomials. However, the results con-
cerning the CR equations over R can also be proven independently of the previous
ones. The motivation for this is an elementary proof of CR equations for complex
polynomials:

Let f (t) ∈ C[t]. We write f (x + iy) = a(x, y) + ib(x, y). Then, ax = by and
ay = −bx .

Proof: Let t1 = (x1 , y1 ) ∈ C. Then, f (t) = f 1 (t)(t − t1 ) + f (t1 ). In that case, if

f 1 (x+iy) = a1 (x, y)+ib1 (x, y) we get a(x, y)+ib(x, y) = (a1 (x, y)+ib1 (x, y))(x−
x1 + i(y − y1 )) + f (t1 ) = (x − x1 )a1 − (y − y1 )b1 + i((y − y1 )a1 + (x − x1 )b1 ) + f (t1 ).
Then, ax (t1 ) = a1 (t1 ) and by (t1 ) = a1 (t1 ). Similarly, ay (t1 ) = −bx (t1 ).

10
The same technique can be applied to any monic quaternion polynomial f (t) for
any t0 ∈ R. Indeed, divide f by t − t0 to get f (t) = g(t)(t − t0 ) + f (t0 ) =
(tn−1 + an−2 tn−2 + · · · a1 t + a0 )(t − t0 ) + f (t0 ). Since t0 is real, it commutes
with any quaternion, and thus we also have f (t) = (t − t0 )g(t) + f (t0 ). Thus, if
t = x + iy + jz + kw and g(x + iy + jz + kw) = g1 + ig2 + jg3 + kg4 , we see that
f (x + iy + jz + kw) = f1 + if2 + jf3 + kf4 = (g1 + ig2 + jg3 + kg4 )(x − t0 + iy +
jz + kw) + f (t0 ) = (x − t0 + iy + jz + kw)(g1 + ig2 + jg3 + kg4 ) + f (t0 ). Now an
easy calculation shows that
g1 (t0 ) −g2 (t0 ) 0 0
 

 g (t ) g1 (t0 ) 0 0
 
J(f )(t0 ) =  2 0 (11)

 0 0 g1 (t0 ) 0


0 0 0 g1 (t0 )
with all the other partials equal to zero. This provides the CR equations at a
real point t0 :
∂f1 ∂f2 ∂f1 ∂f2 ∂f3 ∂f4 ∂f1
= , =− , = = (12)
∂x ∂y ∂y ∂x ∂z ∂w ∂x

The above method, however, fails when t0 ∈ H − R. The reason is that t–

being a quaternion variable now–does not commute with t0 and thus f1 + if2 +
jf3 + kf4 6= (g1 + ig2 + jg3 + kg4 )(x − t0 + iy + jz + kw) + f (t0 ). For example, if
f (t) = t2 +(i+j)t−k = (t+j)(t+i), there is no linear polynomial g1 +ig2 +jg3 +kg4
so that f (x + iy + jz + kw) = (g1 + ig2 + jg3 + kg4 )(x + i(y + 1) + jz + kw).

Next, formula (6) provides the CR equations for an f (t) ∈ C[t] and t0 = r + is:
∂f1 ∂f2 ∂f1 ∂f2 ∂f3 ∂f4 ∂f3 ∂f4
= , =− , = , =− (13)
∂x ∂y ∂y ∂x ∂z ∂w ∂w ∂z
Observe that |J(f )(t0 )| ≥ 0.

Finally, for the general case, pick γ ∈ H so that γt0 γ ∗ = r + is. With the aid of
(8) and (10) we get the CR equations for f at t0 :
 
α1 −β 1 −α2 β2
 β1 α1 β2 α2 
J(f )(t0 ) = 
 α3 −β 3
 · Aγ , where αk , β k ∈ R (14)
α4 −β 4 
−β 3 −α3 β4 α4

11
4.2 |J(f )| at a root of f

In this section we will show that |J(f )| is non negative over H. In particular,
we will prove that if t0 is a root of f , t0 is simple if and only if |J(f )(t0 )| > 0.
Thus, at a multiple root |J(f )| vanishes. Furthermore, we will briefly talk about
the local degree of a primitive f at a root t0 and show how this relates to its
multiplicity µ(f )(t0 ).

Let t0 = x0 + iy0 + jz0 + kw0 ∈ H. Divide f (t) by t − t0 to get f (t) = g(t)(t −

t0 ) + f (t0 ). After making a suitable transformation, we will assume that t0 = i.
Let g(t) = bm tm + · · · + b1 t + b0 , bk ∈ H. We write f (t) = b0 (t − i) + b1 t(t − i) +
· · · + bm tm (t − i) + f (t0 ). Let A be the matrix
 
0 −1 0 0
 1 0 0 0 
A=  0

0 0 1 
0 0 −1 0

Notice that A2 = −I. Furthermore, from (4) we get J(tk (t − i))(i) = Ak for
k ≥ 0. Therefore,
m
X m
X
J(f )(i) = b0 I + b1 A + b2 A2 + · · · + bm Am = (−1)k b2k I + (−1)l b2l+1 A.
k=0 l=0
Pm k
Pm l
Set k=0 (−1) b2k = Be , l=0 (−1) b2l+1 = Bo . We claim that |Be I + Bo A| ≥ 0.
Indeed, if either of Be , Bo is zero there is nothing to prove. Suppose then Be Bo 6=
0. Then it is enough to show |I + CA| ≥ 0 for C = Bo /Be . If C = a+ ib+ jc+ kd,
I + CA takes the form
 
−b + 1 −a d −c

 a −b + 1 −c −d  
 d −c b+1 a 
−c −d −a b + 1

whose determinant |I + CA| = q(b2 ) = b4 + (−2 + 2a2 + 2c2 + 2d2 )b2 + (1 − 2c2 −
2d2 + 2a2 + 2c2 a2 + 2d2 a2 + a4 + d4 + c4 + 2d2 c2 ). The discriminant of q(b2 )
is equal to −16a2 and that proves the claim. Moreover, q(b2 ) vanishes precisely
when a = 0, b2 + c2 + d2 = 1. In addition, notice that if δ is an imaginary unit

12
quaternion, 1 + (a + ib + jc + kd)δ = 0 if and only if a = 0 and b2 + c2 + d2 = 1;
that is, when q(b2 ) = 0. In short, |Be I + Bo A| = 0 if and only Be + Bo δ = 0, for
any imaginary unit quaternion δ.

Let γ be any imaginary unit quaternion. Then, g(γ) = Be + Bo γ, since γ 2 = −1.

Thus, if g(γ) 6= 0, which in turn says that t = i has multiplicity 1, |Be I+Bo A| > 0.
On the other hand, if g(γ) = 0, which means that µ(f )(i) ≥ 2, then J(f )(γ)|
vanishes, as required.

Finally, we will consider the relationship between, the local degree of (a primitive)
f at its root ci and its multiplicity µ(f )(ci ). First, let us recall some facts about
degrees of maps. If we compactify R4 by adding the point ∞ to it, then R4 ∪
{∞} ≡ S 4 . In that regard, since f (t) → ∞ as t → ∞, we get a continuous map
f : S 4 → S 4 . It is then known that deg f = n, [1]. Let now c1 , · · · , cm be the
distinct roots of f and define the local degree ldf (ci ) of f at ci as follows: Choose
a ball Bi with center at ci so that it does not contain any other root of f . Let Si
be the boundary of Bi . We now can define the Gauss map

f (t)
G : Si → S 3 , G(t) =
kf (t)k
Pm
Then, ldf (ci ) = deg G. Since the degree of f is n, we note that n = i=1 ldf (ci ).

If ci is a simple root, then ldf (ci ) = 1 since |J(f )(ci )| > 0. On the other hand,
if µ(f )(cj ) = k ≥ 2, we see that µ(f ∗ f )(α + iβ) = k, where α + iβ ∼ cj .
Since the local degree (over C) of f ∗ f at α + iβ is k, for a suitable point
γ + iδ, δ 6= 0 near zero, the equation f ∗ f (t) = γ + iδ has k distinct solutions
γ j + iδj , j = 1, 2, · · · , k in C, not conjugate to one another. Indeed, if γ 1 + iδ1 =
γ 2 − iδ2 , then f ∗ f (γ 1 + iδ1 ) = f ∗ f (γ 2 − iδ2 ) = γ + iδ. But since f ∗ f is real we
should have γ + iδ = γ − iδ, a contradiction to δ 6= 0. Now there exist distinct
ζj ∼ γ j + iδj which are solutions of f (t) = α1 + ib1 . Indeed, for if ζi = ζj
we have η1 (γ i + iδi )η1∗ = η2 (γ j + iδj )η2∗ or η2∗ η1 (γ i + iδi )η1∗ η2 = γ j + iδj . Thus
η2∗ η1 (γ i + iδi )η1∗ η2 = γ i + iδi which shows γ i + iδi = γ j + iδj . Therefore, ldf (cj )
must also be k.

13
5 Closure

In this paper Jacobians of (left) quaternion polynomials were computed along

with their determinants. As a result, Cauchy-Riemman equations were obtained
for this type of functions. In addition, local degrees were considered and it was
shown that the above commensurates well with the corresponding theory of a
(complex) polynomial of a complex variable. Apparently, similar results can be
gotten for right quaternion polynomials.

It is hoped that the fact, of the Jacobian determinant of f at any point in H

being non negative, would help in finding the zeros of f as well local degrees at
its zeros. Finally, it would be interesting to investigate whether results of the
same nature hold true for general quaternion polynomials of the form f (t) =
a0 ta1 t · · · tan + φ(t), ai ∈ H, ai 6= 0 and φ(t) being a sum of finite number of
similar monomials b0 tb1 t · · · tbk , k < n.

References

[1] S. Eilenberg and I. Niven (1944), The“fundamental theorem of algebra” for

quaternions, Bull. Amer. Math. Soc. 50, No. 4, 246–248.

[2] G. Gentili and D. C. Struppa (2008), On the multiplicity of zeros of

polynomials with quaternionic coefficients, Milan J. Math. 76, 15–25.

[3] B. Gordon and T. S. Motzkin (1965), On the zeros of polynomials over

division rings, Trans. Amer. Math. Soc. 116, 218226.

[4] N. Topuridze (2009), On roots of quaternion polynomials, J. Math.

Sciences. Vol. 160, 6, 843–855.

14