The Niven–Jacobson Theorem: Roots over Quaternion Algebras

Q: Why does the proof need to detour through left roots?

Because the root of ff may happen to kill f rather than f, and then the conjugation rule (16.3) gives no information about f. Conjugating the resulting identity turns f() = 0 into a statement that is a *left* root of f, which peels off a left linear factor and lets the induction proceed.

Q: Why does t^n = a behave so differently for central and noncentral a?

For noncentral a, every solution must commute with a by (16.19), which confines it to the maximal subfield R(a) — a commutative problem with exactly n answers. For central a the constraint is vacuous, the whole polynomial is central, and its root set is a union of full conjugacy classes, hence typically infinite.

Q: Is the root set of a quaternionic polynomial always a union of points and spheres?

Yes. By (16.4) it meets at most f conjugacy classes; by (16.17) a class contributing two roots contributes the whole class; and the noncentral classes of D are exactly the spheres tr = c_1, N = c_2. So each class contributes nothing, one point, or one sphere.

Q: Who proved what?

Niven solved quaternionic polynomial equations over R in 1941, and Eilenberg and Niven gave a topological proof of the fundamental theorem of algebra for quaternions in 1944. Jacobson's contribution is the algebraic formulation over an arbitrary real-closed base. Lam's proof of (16.14) is a simplification of Niven's original argument.

Executive Summary

Let $R$ be a real-closed field and $D = R \oplus R i \oplus R j \oplus R k$ the quaternion division ring over $R$ . The Niven–Jacobson theorem $(16.14)$ says $D$ is right and left algebraically closed: every nonconstant $f \in D [t]$ has a right root, hence splits completely into linear factors.

The proof is a two-line reduction to the commutative case. Quaternionic conjugation is an anti-automorphism, so $f \bar{f}$ has coefficients in $R$ ; a root of that real polynomial lives in the algebraically closed field $R (i)$ , and the conjugation rule $(16.3)$ converts it into a root of $f$ itself — or, in the bad case, into a left root, which an induction on degree disposes of.

The remaining results make the root set explicit. $(16.18)$ gives three equivalent tests for infinitely many roots, all reducible to the presence of an irreducible real quadratic right factor; $(16.19)$ – $(16.20)$ isolate the family with a guaranteed finite count, including the equation $t^{n} = a$ , which has exactly $n$ solutions whenever $a$ is not central.

$\deg f \geq 1$ Always has a root

$2 n$ Degree of

f \bar{f}

over

R

$n$ Solutions of

t^{n} = a

a \notin R

1941Niven

Overview

A field is algebraically closed if every nonconstant polynomial has a root. For a division ring the definition has to pick a side, because roots do: $D$ is right algebraically closed if every nonconstant $f \in D [t]$ has a right root. By the remainder theorem $(16.2)$ this is equivalent to every $f \in D [t]$ splitting into a product of linear factors.

The quaternions over $ℝ$ are the model case, and Niven's 1941 solution of quaternionic polynomial equations is the model proof. Jacobson's contribution places it in the general theory. The converse — that these are the only noncommutative centrally finite examples — is Baer's theorem, treated on Right Algebraically Closed Division Rings.

f \in D [t], \deg f = n \geq 1 ⟹ f (t) = a_{n} (t - c_{n}) (t - c_{n - 1}) \dots (t - c_{1}) for some c_{i} \in D,

(16.14)

$R$ real-closed, $D$ the quaternions over $R$ . The $c_{i}$ are not the roots of $f$ ; only $c_{1}$ is.

Following Lam, $R$ denotes the real-closed base field throughout this page, not a general ring. $F = Z (D) = R$ .

Learning Objectives

Verify that quaternionic conjugation is an anti-automorphism with fixed set $R$ , and that $\bar{f g} = \bar{g} \bar{f}$ .
Show $f \bar{f} \in R [t]$ and use it to find a root of $f$ by induction on degree.
Prove the quadratic-class lemma $(16.17)$ and use it in both directions of $(16.18)$ .
Decide from the coefficients whether a quaternionic polynomial has finitely or infinitely many roots.
Prove that $t^{n} = a$ has exactly $n$ solutions for $a \in D ∖ R$ and explain why $a \in R$ is different.
Run Niven's procedure for finding all roots once one root is known.

Definitions

Definition§16Right and left algebraically closed

A division ring $D$ is right algebraically closed if every nonconstant $f \in D [t]$ has a right root in $D$ ; equivalently, by $(16.2)$ , every $f \in D [t]$ is a product of linear factors in $D [t]$ . Left algebraically closed is defined by the mirror condition, evaluating with the variable on the left. For centrally finite division rings the two notions coincide.

$\bar{q}$: For $q = a + b i + c j + d k$ , the conjugate $a - b i - c j - d k$ . Then $\bar{q_{1} q_{2}} = {\bar{q}}_{2} {\bar{q}}_{1}$ and $\bar{q} = q$ exactly when $q \in R$ .
$N (q)$ and $tr (q)$: The norm $\bar{q} q = a^{2} + b^{2} + c^{2} + d^{2} \in R$ and the trace $q + \bar{q} = 2 a \in R$ . Every $q$ satisfies $q^{2} - tr (q) q + N (q) = 0$ .
$\bar{f}$: For $f = \sum_{r} a_{r} t^{r}$ , the polynomial $\sum_{r} {\bar{a}}_{r} t^{r}$ . Then $\bar{f g} = \bar{g} \bar{f}$ in $D [t]$ .
$R (i)$: The subfield $R \oplus R i \subseteq D$ ; it is the algebraic closure of $R$ by the Artin–Schreier theorem, since $R$ is real-closed.
Quadratic class: A conjugacy class whose minimal polynomial over the centre has degree $2$ . In $D$ every noncentral class is quadratic.

$R$ real-closed forces $char R = 0$ and makes $a^{2} + b^{2} + c^{2} + d^{2}$ nonzero for $(a, b, c, d) \neq 0$ , which is why $D$ is a division ring.

Core Concepts

Conjugation turns quaternionic polynomials into real ones

Quaternionic conjugation is $R$ -linear and reverses products, and its fixed set is exactly $R$ . Extending it coefficientwise to $D [t]$ gives $\bar{f g} = \bar{g} \bar{f}$ . Applying this to $f \bar{f}$ shows $\bar{f \bar{f}} = f \bar{f}$ , so all coefficients of $f \bar{f}$ are fixed by the bar:

f \bar{f} = \bar{f} f \in R [t], \deg (f \bar{f}) = 2 \deg f .

(16.14a)

The two products agree because both are central in $D [t]$ and $D [t]$ is a domain. The degree is exact because the leading coefficient is $N (a_{n}) \neq 0$ .

f \in D [t]

f \bar{f} \in R [t]

root

α \in R (i)

root of

f

D

Every noncentral class is quadratic

Each $q \in D$ satisfies $q^{2} - tr (q) q + N (q) = 0$ with $tr (q), N (q) \in R$ . So every element of $D$ is algebraic of degree at most $2$ over $R$ , and a noncentral $q$ has minimal polynomial exactly $λ (t) = t^{2} - tr (q) t + N (q)$ , irreducible over $R$ because its discriminant $tr (q)^{2} - 4 N (q)$ is negative in the ordering of $R$ .

Geometrically: the noncentral conjugacy classes of $D$ are the spheres ${q : tr (q) = 2 a, N (q) = a^{2} + b^{2}}$ , and the root set of any nonzero $f$ is a finite union of isolated points and complete spheres.

Key Results

Theorem(16.14)Niven–Jacobson

Let $R$ be a real-closed field and let $D$ be the division ring of quaternions over $R$ . Then $D$ is right algebraically closed, and also left algebraically closed.

Proof

Induct on $n = \deg f \geq 1$ . For $n = 1$ , $f = a_{1} t + a_{0}$ with $a_{1} \neq 0$ has the right root $- a_{1}^{- 1} a_{0}$ . Let $n \geq 2$ .

By $(16.14 a)$ , $f \bar{f} \in R [t]$ has degree $2 n \geq 2$ . Since $R$ is real-closed, $R (i)$ is algebraically closed, so $f \bar{f}$ has a root $α \in R (i) \subseteq D$ .

**Case 1: $\bar{f} (α) \neq 0$ .** Apply the conjugation rule $(16.3)$ to the factorisation $f \bar{f} = f \cdot \bar{f}$ at $d = α$ , with $a = \bar{f} (α)$ . It gives $0 = (f \bar{f}) (α) = f (a α a^{- 1}) a$ , so $a α a^{- 1}$ is a right root of $f$ and we are done.

**Case 2: $\bar{f} (α) = 0$ .** Write $f = \sum_{r} a_{r} t^{r}$ , so $\sum_{r} {\bar{a}}_{r} α^{r} = 0$ . Applying the anti-automorphism $q \mapsto \bar{q}$ to this identity reverses each product and gives $\sum_{r} {\bar{α}}^{r} a_{r} = 0$ ; that is, $β : = \bar{α}$ is a left root of $f$ . By the left-handed form of $(16.2)$ we may write $f (t) = (t - β) g (t)$ with $\deg g = n - 1$ .

By the inductive hypothesis $g$ has a right root $γ \in D$ . Writing $g = \sum_{i} b_{i} t^{i}$ , we get $f (γ) = \sum_{i} b_{i} γ^{i + 1} - \sum_{i} β b_{i} γ^{i} = g (γ) γ - β g (γ) = 0$ . So $γ$ is a right root of $f$ . The statement for left roots follows by the mirror argument, or by applying the result to $D^{op}$ , which is isomorphic to $D$ via quaternionic conjugation.

Lemma(16.17)Two roots in a quadratic class

Let $D$ be a division ring with centre $F$ and let $A$ be a conjugacy class of $D$ whose minimal polynomial $λ \in F [t]$ is quadratic. If $f \in D [t]$ has two distinct roots in $A$ , then $f \in D [t] \cdot λ$ and $f$ vanishes identically on $A$ .

Proof

Divide on the right by the monic $λ$ : $f = q λ + (a t + b)$ with $a, b \in D$ . By $(16.6)$ the product $q λ$ vanishes on all of $A$ , so $a t + b$ vanishes at the two given distinct elements of $A$ . A nonzero polynomial of degree at most $1$ over a division ring has at most one root, so $a = b = 0$ . Hence $f = q λ \in D [t] λ$ , and $f (A) = 0$ by $(16.6)$ again.

Proposition(16.18)Niven's criteria for infinitely many roots

Let $R$ be real-closed, $D$ the quaternions over $R$ , and $f \in D [t]$ nonzero. The following are equivalent:

$f$ has infinitely many roots in $D$ ;
there exist $a, b \in R$ with $b \neq 0$ such that $f (a + b i) = f (a - b i) = 0$ ;
$f$ has a right factor $λ \in R [t]$ which is an irreducible quadratic.

When these hold, $f$ vanishes identically on the conjugacy class of $a + b i$ .

Proof

**(1) $\Rightarrow$ (3).** By the class bound $(16.4)$ the roots of $f$ lie in at most $\deg f$ conjugacy classes, so infinitely many roots force some class $A$ to contain two of them. Then $A$ is not a singleton, so it is noncentral and its minimal polynomial $λ \in R [t]$ is an irreducible quadratic. By $(16.17)$ , $f \in D [t] λ$ and $f (A) = 0$ .

**(3) $\Rightarrow$ (1).** Let $ζ \in R (i)$ be a root of $λ$ ; since $λ$ is irreducible over $R$ , $ζ \notin R$ , so $ζ$ is noncentral in $D$ and by Herstein's theorem $(13.26)$ has infinitely many conjugates. All of them are roots of $λ$ , because $λ$ has central coefficients, and hence roots of $f = h λ$ by $(16.2)$ .

**(3) $\Rightarrow$ (2).** The roots of an irreducible quadratic $λ \in R [t]$ inside the algebraically closed field $R (i)$ are a pair $a \pm b i$ with $a, b \in R$ , $b \neq 0$ ; both are roots of $λ$ , hence of $f$ .

**(2) $\Rightarrow$ (3).** The elements $a + b i$ and $a - b i$ are distinct and conjugate in $D$ , since $j (a + b i) j^{- 1} = a - b i$ . They therefore lie in a common class whose minimal polynomial is $λ (t) = t^{2} - 2 a t + (a^{2} + b^{2})$ , irreducible over $R$ because $b \neq 0$ . Now $(16.17)$ gives $f \in D [t] λ$ .

Proposition(16.19)Central coefficients except the constant term

Let $R$ be real-closed, $D$ the quaternions over $R$ , and $f (t) = \sum_{i = 0}^{n} a_{i} t^{i} \in D [t]$ with $a_{n} \neq 0$ , $a_{1}, \dots, a_{n} \in R$ and $a_{0} \in D ∖ R$ . Then $f$ has at most $n$ roots in $D$ .

Proof

Let $α \in D$ be a root, so $\sum_{i = 1}^{n} a_{i} α^{i} = - a_{0}$ . Every $a_{i}$ with $i \geq 1$ is central, so the left-hand side is a polynomial expression in $α$ alone and therefore commutes with $α$ . Hence $α$ commutes with $a_{0}$ , i.e. $α \in C_{D} (a_{0})$ .

Since $a_{0} \notin R = Z (D)$ , the field $R (a_{0})$ is a maximal subfield of $D$ , of dimension $2$ over $R$ , and for the quaternion algebra $C_{D} (R (a_{0})) = R (a_{0})$ — the double centraliser property for maximal subfields of a centrally finite division ring. So every root of $f$ lies in the field $R (a_{0})$ . But all coefficients of $f$ lie in $R (a_{0})$ as well, so $f$ may be read as a polynomial of degree $n$ over that field, where it has at most $n$ roots.

Corollary(16.20)Niven:

n

-th roots of a noncentral quaternion

Let $R$ be real-closed, $D$ the quaternions over $R$ , and $a \in D ∖ R$ . Then the equation $t^{n} = a$ has exactly $n$ solutions in $D$ , all lying in the field $R (a)$ .

Proof

Apply $(16.19)$ to $f (t) = t^{n} - a$ : the coefficients $1, 0, \dots, 0$ are in $R$ and the constant term $- a$ lies in $D ∖ R$ . So there are at most $n$ roots and all lie in $R (- a) = R (a)$ . That field is $R$ -isomorphic to $R (i)$ , hence algebraically closed, and $char R = 0$ makes $t^{n} - a$ separable — its derivative $n t^{n - 1}$ shares no root with it since $a \neq 0$ . A separable polynomial of degree $n$ over an algebraically closed field has exactly $n$ distinct roots.

Proof Techniques and Method

How these proofs work, and which move to reuse.

Move 1

Norm down to the centre

Multiplying by the conjugate polynomial produces a central polynomial of twice the degree. This is the polynomial analogue of the reduced norm, and it is the only bridge between $D [t]$ and $R [t]$ available.

Move 2

Trade a left root for a right root

The bad case of the induction produces a left root. Peeling off $(t - β)$ on the left and inducting on the degree converts it into a right root of the original polynomial — a genuinely noncommutative manoeuvre with no field analogue.

Move 3

Centralise using central coefficients

In $(16.19)$ the hypothesis that $a_{1}, \dots, a_{n}$ are central makes every root commute with the constant term, collapsing the search space from $D$ to a single maximal subfield.

Move 3 is the reusable one. Whenever all but one coefficient is central, the equation forces its solutions into the centraliser of the exceptional coefficient, and for a centrally finite division ring that centraliser is a field. The problem then becomes commutative.

Worked Example

Square roots of $i$ in $ℍ$

Take $R = ℝ$ , $D = ℍ$ , $a = i \notin ℝ$ , and $n = 2$ . By $(16.20)$ the equation $t^{2} = i$ has exactly two solutions, both in $ℝ (i) = ℂ$ . Solving there:

t = \pm \frac{\sqrt{2}}{2} (1 + i), (\frac{1 + i}{\sqrt{2}})^{2} = \frac{1 + 2 i + i^{2}}{2} = i .

(E.1)

Exactly two solutions in the whole of $ℍ$ — no quaternionic solutions outside $ℂ$ .

Deciding finiteness from a factorisation

Consider $f_{1} (t) = t^{2} + 1$ and $f_{2} (t) = t^{2} - (1 + i) t + i$ over $ℍ$ .

$f_{1}$ is an irreducible real quadratic, so it is its own right factor and criterion (3) of $(16.18)$ holds: infinitely many roots — the sphere of unit imaginary quaternions. Criterion (2) is visible too: $f_{1} (\pm i) = 0$ with $a = 0$ , $b = 1$ .
$f_{2} = (t - 1) (t - i)$ has roots exactly $1$ and $i$ , a finite set, so by $(16.18)$ it has no irreducible real quadratic right factor. Indeed the only candidate would be $t^{2} + 1$ , and $f_{2} \neq t^{2} + 1$ while both are monic of degree $2$ .

Applying the theorem to a cubic

Let $f (t) = t^{3} - (1 + i) t^{2} + i t$ , the Bray–Whaples polynomial for the nodes $0, 1, i$ . Its roots are exactly $0, 1, i$ — three isolated points, no spheres — so by $(16.18)$ it has no irreducible real quadratic right factor. Consistently with $(16.14)$ it does split completely: $f (t) = (t - 1) (t - i) t$ , a product of three linear factors, even though only the rightmost supplies a root of $f$ directly.

Process and Workflow

Niven's procedure for finding all roots of $f \in D [t]$ once a single root $a$ is known.

A root $a$ of $f$ has been found. What next?

$a \in R$ (central)Write

f (t) = g (t) (t - a)

and continue with

g

, whose degree is one lower. No conjugates need to be considered: the class of

a

is the singleton

{a}

$a \notin R$ and some conjugate $a' \neq a$ also satisfies $f (a') = 0$ By

(16.17)

f (t) = h (t) λ (t)

with

λ

the minimal polynomial of

a

over

R

. The entire conjugacy class of

a

consists of roots; deflate by

λ

and continue with

h

$a \notin R$ and $f (a') \neq 0$ for a conjugate $a' \neq a$ The class of

a

contributes exactly one root, namely

a

. Deflate by

t - a

and look for roots in other classes.

Form the real polynomialCompute

f \bar{f} \in R [t]

of degree

2 \deg f

; its roots in

R (i)

enumerate the candidate conjugacy classes.

Test each candidate classFor a candidate

α = a + b i

with

b \neq 0

, decide whether

λ (t) = t^{2} - 2 a t + (a^{2} + b^{2})

right-divides

f

Sphere or pointIf

λ

divides, the whole class is a root set; if not,

(16.3)

recovers the single root of that class as

\bar{f} (α) α \bar{f} (α)^{- 1}

Deflate and repeatDivide out the linear or quadratic factor found and restart with the quotient, whose degree has dropped by one or two.

Comparison and Classification

Root behaviour of quaternionic polynomials by coefficient pattern
Polynomial $f \in D [t]$ , $\deg f = n$	Root set	Reference
All coefficients in $R$ , no repeated real irreducible quadratic factor	real roots as points, complex conjugate pairs as full spheres	$(16.18)$
Some irreducible quadratic $λ \in R [t]$ right-divides $f$	infinite: contains the whole class of the roots of $λ$	$(16.18)$
$a_{1}, \dots, a_{n} \in R$ and $a_{0} \in D ∖ R$	at most $n$ points, all inside the field $R (a_{0})$	$(16.19)$
$f (t) = t^{n} - a$ with $a \in D ∖ R$	exactly $n$ points, all in $R (a)$	$(16.20)$
$f$ a product of linear factors with pairwise nonconjugate roots	exactly $n$ isolated points	$(16.13)$
General nonzero $f$	finite union of points and spheres; nonempty if $n \geq 1$	$(16.14)$ , $(16.4)$

Which properties transfer from $ℂ$ to the quaternions over a real-closed field
	Algebraically closed field	Quaternions over real-closed $R$
Every nonconstant polynomial has a root	yes	yes
Every polynomial splits into linear factors	yes	yes
Factorisation into linear factors is unique	yes	no
At most $\deg f$ roots	yes	no
Root set is finite	yes	no
Left and right roots coincide	yes	no

Which properties transfer from $ℂ$ to the quaternions over a real-closed field

Relationship Map

Artin–Schreier:

R (i)

closed

(16.14)

Niven–Jacobson

(16.15)

Baer converse

(16.16)

classification

Right Algebraically Closed Division Rings supplies the converse: a noncommutative centrally finite division ring in which every central polynomial has a root must be of this form.
The Gordon–Motzkin Theorem supplies the class bound used at the start of $(16.18)$ and the dichotomy that makes finite or infinite the only options.
Vanishing Polynomials supplies $(16.6)$ , without which $(16.17)$ has no proof.
Generalised Quaternion Algebras gives the structure theory of $D$ itself, including the maximal subfields used in $(16.19)$ .
The Bray–Whaples Theorem supplies the extremal examples with exactly $n$ isolated roots.

Applications and Industry Use

Applications here means where this structure is used — inside mathematics and in the engineering and computing disciplines that consume it.

Quaternionic root finding. Attitude and orientation computations in robotics and aerospace occasionally require solving quaternionic polynomial equations; $(16.18)$ is the test that decides whether the solution set is a discrete set of orientations or a continuous family, which is a design-relevant distinction.
Roots and powers of rotations. $(16.20)$ says a noncentral unit quaternion has exactly $n$ distinct $n$ -th roots, all in the complex line it spans. This is the algebraic content of the fact that a rotation about a fixed axis has exactly $n$ $n$ -th roots about the same axis.
Slice-regular function theory. The structure of zero sets — isolated points together with isolated spheres — is inherited by quaternionic slice-regular functions, where the polynomial case treated here is the model.
**Numerical linear algebra over $ℍ$ .** Eigenvalue problems for quaternionic matrices have right eigenvalues occurring in full conjugacy classes; the same all-or-nothing phenomenon and the same $f \bar{f}$ reduction underlie the standard algorithms.

Honest summary: within mathematics the theorem is a classification tool. Outside it, the practical content is $(16.18)$ and $(16.20)$ , which are the statements an implementation actually needs.

Computational Notes

Computational notes cover algorithms, cost and library behaviour rather than manufacturing process.

The $f \bar{f}$ reduction costs $O (n^{2})$ quaternion multiplications and turns the problem into finding the $2 n$ complex roots of a real polynomial — a standard numerical task.
Each candidate class needs one right-division test by a real quadratic, at cost $O (n)$ ; the whole algorithm is $O (n^{2})$ after the complex root finding.
Deflation must be right division. Left division by $t - c$ does not preserve the remaining root set.
Conditioning: a class contributing a single root is recovered by a conjugation $\bar{f} (α) α \bar{f} (α)^{- 1}$ , which is ill-conditioned when $\bar{f} (α)$ is small — precisely when the polynomial is near the sphere case. Implementations should test divisibility with a tolerance rather than trusting the conjugation blindly.
Symbolically, Sage and Magma provide quaternion algebras over number fields and can perform the divisibility tests exactly; there is no general routine for arbitrary division rings, and there cannot be one.

Failure Modes and Common Mistakes

Do not apply $(16.14)$ to a general quaternion algebra $(\frac{a, b}{K})$ ; the base field must be real-closed and the algebra the standard $(- 1, - 1)$ one over it.
Do not assume $f \bar{f}$ and $\bar{f} f$ differ — they are equal and both lie in $R [t]$ — but do keep track of which factor you apply $(16.3)$ to.
Do not conclude from $f (a) = 0$ with $a \notin R$ that all conjugates of $a$ are roots; that requires a second root in the class, which is exactly the test in $(16.18)$ .
Do not expect the classification to extend to centrally infinite division rings: the algebraically closed ones there are not understood.

Quick Reference

Setting

R

real-closed,

D = R \oplus R i \oplus R j \oplus R k

Z (D) = R

Conjugation

\bar{q_{1} q_{2}} = {\bar{q}}_{2} {\bar{q}}_{1}

, fixed set

R

Norm polynomial

f \bar{f} = \bar{f} f \in R [t]

, degree

2 \deg f

Niven–Jacobsonevery nonconstant

f \in D [t]

has a right root

Quadratic class lemmatwo roots in a class

\Rightarrow f \in D [t] λ

Infinitely many roots

\Leftrightarrow

an irreducible

λ \in R [t]

right-divides

f

Finite family

a_{1}, \dots, a_{n} \in R

a_{0} \notin R

\Rightarrow

at most

n

roots

$n$ -th roots

t^{n} = a

a \notin R

: exactly

n

solutions, all in

R (a)

Statements and hypotheses
Reference	Hypotheses	Conclusion
$(16.14)$	$R$ real-closed, $D$ quaternions over $R$	$D$ is right and left algebraically closed
$(16.17)$	$D$ any division ring, class $A$ with quadratic minimal polynomial $λ$ , $f$ with two roots in $A$	$f \in D [t] λ$ and $f (A) = 0$
$(16.18)$	$R$ real-closed, $D$ quaternions over $R$ , $f \neq 0$	infinitely many roots $\Leftrightarrow$ two conjugate roots $\Leftrightarrow$ irreducible real quadratic right factor
$(16.19)$	$a_{1}, \dots, a_{n} \in R$ , $a_{0} \in D ∖ R$ , $a_{n} \neq 0$	at most $n$ roots, all in $R (a_{0})$
$(16.20)$	$a \in D ∖ R$	$t^{n} = a$ has exactly $n$ solutions, all in $R (a)$

Frequently Asked Questions

How can a division ring be algebraically closed and still have a polynomial with infinitely many roots?

Algebraic closure asserts existence of a root, nothing more. Over a field, existence plus the degree bound gives the familiar picture; over $D$ the degree bound fails because noncentral conjugacy classes are infinite. So $t^{2} + 1$ has a root — in fact a two-sphere of them — and both facts are consistent.

Why does the proof need to detour through left roots?

Because the root $α$ of $f \bar{f}$ may happen to kill $\bar{f}$ rather than $f$ , and then the conjugation rule $(16.3)$ gives no information about $f$ . Conjugating the resulting identity turns $\bar{f} (α) = 0$ into a statement that $\bar{α}$ is a left root of $f$ , which peels off a left linear factor and lets the induction proceed.

Does the theorem hold for a general quaternion algebra over a general field?

No. Two hypotheses are needed: the base field must be real-closed, so that adjoining a square root of $- 1$ gives an algebraically closed field; and the algebra must be the standard quaternion algebra, which is then a division ring because sums of squares are nonzero. Over $ℚ$ , for instance, the rational quaternions are very far from algebraically closed.

Why does

t^{n} = a

behave so differently for central and noncentral

a

For noncentral $a$ , every solution must commute with $a$ by $(16.19)$ , which confines it to the maximal subfield $R (a)$ — a commutative problem with exactly $n$ answers. For central $a$ the constraint is vacuous, the whole polynomial is central, and its root set is a union of full conjugacy classes, hence typically infinite.

Is the root set of a quaternionic polynomial always a union of points and spheres?

Yes. By $(16.4)$ it meets at most $\deg f$ conjugacy classes; by $(16.17)$ a class contributing two roots contributes the whole class; and the noncentral classes of $D$ are exactly the spheres $tr = c_{1}$ , $N = c_{2}$ . So each class contributes nothing, one point, or one sphere.

Who proved what?

Niven solved quaternionic polynomial equations over $ℝ$ in 1941, and Eilenberg and Niven gave a topological proof of the fundamental theorem of algebra for quaternions in 1944. Jacobson's contribution is the algebraic formulation over an arbitrary real-closed base. Lam's proof of $(16.14)$ is a simplification of Niven's original argument.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §16, (16.14) and (16.17)–(16.20) (pp. 269–274).
I. Niven, “Equations in quaternions”, American Mathematical Monthly 48 (1941), 654–661.
S. Eilenberg and I. Niven, “The fundamental theorem of algebra for quaternions”, Bulletin of the American Mathematical Society 50 (1944), 246–248.
N. Jacobson, The Theory of Rings, Mathematical Surveys 2, American Mathematical Society, 1943.
P. K. Draxl, Skew Fields, London Mathematical Society Lecture Note Series 81, Cambridge University Press, 1983.
N. Jacobson, Basic Algebra I, 2nd edition, W. H. Freeman, 1985, for the Artin–Schreier theory of real-closed fields.

AI Suggested Questions

Compare the algebraic proof of $(16.14)$ with the Eilenberg–Niven topological degree argument, and identify what each needs that the other does not.
Compute all roots of a specific quaternionic cubic with one real and one nonreal coefficient.
How does $(16.18)$ generalise to octonions, where associativity fails?
What is the analogue of $(16.20)$ for the exponential and logarithm of a quaternion?
Describe the zero set of a polynomial in two noncommuting quaternionic variables — is there any analogue of the point-or-sphere dichotomy?
Give an algorithm that computes the exact root set of $f \in ℍ [t]$ over a number field, with a complexity analysis.

01Executive Summary

02Overview

03Learning Objectives

04Definitions

05Core Concepts

Conjugation turns quaternionic polynomials into real ones

Every noncentral class is quadratic

06Key Results

07Proof Techniques and Method

Norm down to the centre

Trade a left root for a right root

Centralise using central coefficients

08Worked Example

Square roots of i in ℍ

Deciding finiteness from a factorisation

Applying the theorem to a cubic

09Process and Workflow

10Comparison and Classification

11Relationship Map

12Applications and Industry Use

13Computational Notes

14Failure Modes and Common Mistakes

15Quick Reference

16Frequently Asked Questions

17Related KEVOS Topics

18References

19AI Suggested Questions

Continue learning

Executive Summary

Overview

Learning Objectives

Definitions

Core Concepts

Key Results

Proof Techniques and Method

Worked Example

Square roots of $i$ in $ℍ$

Process and Workflow

Comparison and Classification

Relationship Map

Applications and Industry Use

Computational Notes

Failure Modes and Common Mistakes

Quick Reference

Frequently Asked Questions

Related KEVOS Topics

References

AI Suggested Questions