Biquaternion Automorphisms and Derivations
Introduction
The biquaternion algebra is the complexification of the quaternion algebra,
$$ \mathbb{B} = \mathbb{C} \otimes_{\mathbb{R}} \mathbb{H}. $$
It carries two structures, distinguished by the ground field. Over $\mathbb{C}$ it is a four-dimensional central simple algebra; over $\mathbb{R}$ the same set is an eight-dimensional algebra, simple but not central. This article describes two standard invariants of $\mathbb{B}$: the group of algebra automorphisms and the Lie algebra of derivations. Both depend on the ground field, so the two views are kept separate and the field is named at each step.
We use the conventions of the article on the biquaternion algebra throughout: the basis $\{e_t, \vec e_x, \vec e_y, \vec e_z\}$ with $\vec e_k^2 = -e_t$; the central scalar imaginary $i$ with $i^2 = -1$; the conjugations ${}^{\natural}$, $\bar{\cdot}$, ${}^{*} = {}^{\natural} \circ \bar{\cdot}$, ${}^{\flat} = -{}^{*}$; and the remarkable subspaces $\mathbb{B}_{C}$, $\mathbb{B}_{V}$, $\mathbb{B}_{H}$, $\mathbb{B}_{iH}$, $\mathbb{B}_{\Psi}$, $\mathbb{B}_{\Phi}$. A general element is $\tilde{Q} = Q_t e_t + Q_x \vec e_x + Q_y \vec e_y + Q_z \vec e_z$ with $Q_\mu \in \mathbb{C}$.
The inverse action and the dagger action. The automorphisms of this article are the maps $\tilde{Q}_2\mapsto\tilde{Q}\tilde{Q}_2\tilde{Q}^{-1}$ built from the inverse. They use the product and the biquaternion norm alone, and for an invertible element $\tilde{Q}$ they are the inner automorphisms below. A second two-sided action of an invertible element replaces the inverse by the Hermitian conjugate, $$ \mathrm{H}_{\tilde{Q}}(\tilde{Q}_2)=\tilde{Q}\tilde{Q}_2\tilde{Q}^{*}, $$ and it is not an automorphism of the algebra: it is multiplicative exactly on the $*$-unitary elements, and it preserves the two Hermitian sectors rather than the product. That action is the isometry of the algebra treated in Biquaternion Rotations and Lorentz Transformations, and the two maps coincide exactly on the $*$-unitary slice.
No physics is invoked and no new results are claimed. Everything below is standard structure theory of the algebra over each of the two ground fields.
Standing Facts: Simplicity and the Centre
The automorphism group and the derivation algebra of $\mathbb{B}$ are governed by two structural facts, recorded here and used throughout. Both are proved elsewhere and are cited, not reproved.
Simplicity. Over $\mathbb{C}$, the only two-sided ideals of $\mathbb{B}$ are $0$ and $\mathbb{B}$: the algebra is simple. The proof is in The Regular Module in the Biquaternion Algebra, §The Two-Sided Ideals, where the ideal structure of the algebra is developed; the lattice is the same over $\mathbb{R}$ and over $\mathbb{C}$, because the condition $\tilde{Q}_2\tilde{Q}_4,\tilde{Q}_4\tilde{Q}_2\in I$ is field-independent.
The centre. The centre of $\mathbb{B}$ is $Z(\mathbb{B})=\mathbb{B}_{C}=\mathbb{C}e_t$, a complex vector space of dimension $1$, the subspace $\mathbb{B}_{C}$ of Introduction to the Algebra of Biquaternions. Hence $\mathbb{B}$ is central simple over $\mathbb{C}$ — simple, of dimension $4$, with centre exactly $\mathbb{C}$ — while over $\mathbb{R}$ the same algebra is simple but not central, its centre $\mathbb{B}_{C}\cong\mathbb{C}$ being strictly larger than $\mathbb{R}e_t$. Stating that "$\mathbb{B}$ is central simple" without naming the field is false over $\mathbb{R}$, and the sections below respect the distinction.
Automorphisms over $\mathbb{C}$
Throughout this section the ground field is $\mathbb{C}$.
Definition. A $\mathbb{C}$-algebra automorphism of $\mathbb{B}$ is a bijective $\mathbb{C}$-linear map $\sigma : \mathbb{B} \to \mathbb{B}$ with $\sigma(\tilde{Q}_2\tilde{Q}_4) = \sigma(\tilde{Q}_2)\sigma(\tilde{Q}_4)$ and $\sigma(e_t) = e_t$. These maps form a group under composition, written $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$.
Inner automorphisms. For any invertible $g \in \mathbb{B}$ the map $\operatorname{Ad}_g(\tilde{Q}_2) = g \tilde{Q}_2 g^{-1}$ is a $\mathbb{C}$-algebra automorphism, the inner automorphism determined by $g$. Since $\mathbb{C}$ is central, $\operatorname{Ad}_g = \operatorname{Ad}_{\lambda g}$ for every nonzero $\lambda \in \mathbb{C}$, so $\operatorname{Ad}_g$ depends only on the class of $g$ modulo the scalars.
Theorem (Skolem–Noether). For any field $k$ and $n \geq 1$, every $k$-algebra automorphism of $M_n(k)$ is inner: for each $\sigma$ there exists $g \in GL_n(k)$ with $\sigma(\tilde{Q}_2) = g\tilde{Q}_2g^{-1}$ for all $\tilde{Q}_2$.
Applied to $\mathbb{B}$, which is central simple over $\mathbb{C}$, the theorem says that every $\mathbb{C}$-linear automorphism of $\mathbb{B}$ is inner: each has the form $\operatorname{Ad}_g$ for an invertible $g \in \mathbb{B}^{\times}$. Since $\operatorname{Ad}_g$ depends only on $g$ modulo the central scalars, there is a surjection
$$ \mathbb{B}^{\times} \longrightarrow \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}), \qquad g \longmapsto \operatorname{Ad}_g, $$
whose kernel is the group of nonzero central scalars $\mathbb{C}^{\times} e_t$. Hence
$$ \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \cong \mathbb{B}^{\times} / \mathbb{C}^{\times}. $$
Dimension. Since $\mathbb{B}^{\times}$ has complex dimension $4$ and the central scalars have complex dimension $1$, $\dim_{\mathbb{C}} \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) = 4 - 1 = 3$. Equivalently, the group has real dimension $6$, and it is connected.
Corollary (the projective linear group). Under the matrix model $\mathbb{B}\cong M_2(\mathbb{C})$ the invertibles are $GL(2,\mathbb{C})$ and the nonzero central scalars are the scalar matrices, so $$ \boxed{\;\operatorname{Aut}_{\mathbb{C}}(\mathbb{B})\cong\mathbb{B}^{\times}/\mathbb{C}^{\times}\cong PGL(2,\mathbb{C})\;}, $$ the projective general linear group, of complex dimension three. This is the biquaternion form of the classical isomorphism $\operatorname{Aut}(M_n(k))\cong PGL(n,k)$.
Corollary (the orthogonal form of the group). The centre of $SL(2,\mathbb{C})$ is $\{\pm e_t\}$, and the low-dimensional classical isomorphisms give $$ \operatorname{Aut}_{\mathbb{C}}(\mathbb{B}) \cong PGL(2,\mathbb{C}) \cong PSL(2,\mathbb{C}) \cong SO(3,\mathbb{C}), $$ the complex orthogonal group in three variables. The reason is visible in the action. Conjugation fixes the scalar part, so the scalar coordinate is preserved and the action falls on the three-dimensional scalar-free part $\tilde{S}=\tilde{Q}-Q_t\,e_t$ as $\tilde{S}\mapsto g\tilde{S}g^{-1}$, which fixes the quadratic form $\langle\tilde S\rangle_{\natural}$; that is the defining representation of $SO(3,\mathbb{C})$. The group is complex of complex dimension three, hence of real dimension six, and as a real group it is also $SO^+(1,3)$: the two are both $PSL(2,\mathbb{C})$, connected and six-real-dimensional, and are therefore the same real group. So the group that appears here as the abstract automorphism group of the algebra is the restricted Lorentz group. It sits inside the automorphism group of the biquaternion norm, $SO(4,\mathbb{C})\cong(SL(2,\mathbb{C})\times SL(2,\mathbb{C}))/\{\pm(e_t,e_t)\}$ with $\tilde{Q}\mapsto\tilde Q_3\tilde{Q}\tilde{Q}_1^{-1}$, as the diagonal $\tilde Q_3=\tilde{Q}_1=g$; that placement, and the physical construction built on it in the complexified-coordinates setting, are the subject of the section The Algebra Automorphisms: the Diagonal, and an Induced Causality of the companion article The Lorentz Group as Biquaternion Norm Automorphisms.
Corollary (the norm). Every $\mathbb{C}$-linear automorphism preserves the biquaternion norm, and every $\mathbb{R}$-linear automorphism preserves the real norm $r=\sqrt{|N|}$. For an inner automorphism the complex statement is multiplicativity, $\langle g\tilde{Q}g^{-1}\rangle_{\natural}=\langle g\rangle_{\natural}\langle\tilde{Q}\rangle_{\natural}\langle g\rangle_{\natural}^{-1}=\langle\tilde{Q}\rangle_{\natural}$; complex conjugation reverses the sign of the imaginary part, $\langle\bar{\tilde{Q}}\rangle_{\natural}=\langle\tilde{Q}\rangle_{\natural}^{*}$, and so preserves $|N|$ without preserving $\langle\cdot\rangle_{\natural}$ itself.
Remark (automorphisms against isometries). The automorphism group is a proper subgroup of the full isometry group $O(N)$ of the real norm on $\mathbb{B}\cong\mathbb{R}^8$: $PGL(2,\mathbb{C})$ has real dimension $6$, whereas the isometry group of a non-degenerate form of inertia $(4,4)$ on $\mathbb{R}^8$ has real dimension $\tfrac{8\cdot7}{2}=28$. The automorphisms are the isometries that also preserve the algebra; the further isometries are not algebra maps.
Example. For $g = \vec e_x$, with $\vec e_x^{-1} = -\vec e_x$, conjugation fixes $\vec e_x$, $e_t$, $i$ and reverses the signs of $\vec e_y$ and $\vec e_z$: $\operatorname{Ad}_{\vec e_x}(\vec e_x) = \vec e_x$, $\operatorname{Ad}_{\vec e_x}(\vec e_y) = -\vec e_y$, $\operatorname{Ad}_{\vec e_x}(\vec e_z) = -\vec e_z$. Indeed $\vec e_x \vec e_y \vec e_x^{-1} = -(\vec e_x \vec e_y)\vec e_x = -\vec e_z \vec e_x = -\vec e_y$, using $\vec e_x \vec e_y = \vec e_z$ and $\vec e_z \vec e_x = \vec e_y$.
Remark. Quaternion conjugation satisfies $(\tilde{Q}_2\tilde{Q}_4)^{\natural} = \tilde{Q}_4^{\natural}\,\tilde{Q}_2^{\natural}$ and is an anti-automorphism, not an automorphism, so it is not in $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$; complex conjugation is an automorphism but is not $\mathbb{C}$-linear, so it is not in $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$ either. It reappears over $\mathbb{R}$ below.
Automorphisms over $\mathbb{R}$
Now the ground field is $\mathbb{R}$: $\mathbb{B}$ is eight-dimensional, and automorphisms need only be $\mathbb{R}$-linear, not $\mathbb{C}$-linear, which makes the group strictly larger.
Automorphisms preserve the centre. If $\sigma$ is an $\mathbb{R}$-algebra automorphism and $A$ is central, then $\sigma(A)\sigma(\tilde{Q}_2) = \sigma(A\tilde{Q}_2) = \sigma(\tilde{Q}_2A) = \sigma(\tilde{Q}_2)\sigma(A)$ for every $\tilde{Q}_2$, so $\sigma(A)$ is central. Thus $\sigma$ restricts to an $\mathbb{R}$-algebra automorphism of $\mathbb{B}_{C} \cong \mathbb{C}$; since $\mathbb{C}$ as a real algebra has exactly two automorphisms, the identity and $\kappa(A) = \bar{A}$, restriction gives a homomorphism $\rho : \mathrm{Aut}_{\mathbb{R}}(\mathbb{B}) \to \mathrm{Aut}_{\mathbb{R}}(\mathbb{B}_{C}) = \{\mathrm{id}, \kappa\} \cong \mathbb{Z}/2$.
The kernel is the $\mathbb{C}$-linear part. An automorphism lies in $\ker \rho$ exactly when it fixes the centre pointwise, and an $\mathbb{R}$-linear map fixing $\mathbb{B}_{C}$ pointwise is automatically $\mathbb{C}$-linear, since it commutes with multiplication by the central element $i$. Hence $\ker \rho = \mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$, the group computed above; these are precisely the inner automorphisms, by Skolem–Noether.
The conjugation coset. The map $\rho$ is surjective: $c(\tilde{Q}) = \bar{\tilde{Q}}$ is an $\mathbb{R}$-algebra automorphism, since $(\tilde{Q}_2\tilde{Q}_4)^{*} = \bar{\tilde{Q}_2}\bar{\tilde{Q}_4}$, and it induces $\kappa$ on the centre, since $c(i) = -i$. It is not $\mathbb{C}$-linear, and it is not inner, because inner automorphisms fix the centre pointwise while $c(i) = -i \neq i$. So the extension is nontrivial.
The full real automorphism group. There is a short exact sequence $1 \to \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \to \mathrm{Aut}_{\mathbb{R}}(\mathbb{B}) \xrightarrow{\rho} \mathbb{Z}/2 \to 1$, split by $c$ because $c^{2} = \mathrm{id}$. Hence
$$ \boxed{\;\mathrm{Aut}_{\mathbb{R}}(\mathbb{B}) \cong \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \rtimes \mathbb{Z}/2\;}, $$
the generator acting on $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$ by $\operatorname{Ad}_g \mapsto \operatorname{Ad}_{g^{*}}$. Concretely, every real automorphism of $\mathbb{B}$ has exactly one of the two forms $\sigma(\tilde{Q}_2) = g \tilde{Q}_2 g^{-1}$ or $\sigma(\tilde{Q}_2) = g\, \bar{\tilde{Q}_2}\, g^{-1}$, with $g \in \mathbb{B}^{\times}$ determined up to a nonzero complex scalar. The first family is the identity coset of $\mathbb{C}$-linear inner automorphisms; the second is the coset of $c$, consisting of conjugate-linear automorphisms.
Dimension and scope. As a real Lie group, $\mathrm{Aut}_{\mathbb{R}}(\mathbb{B})$ has real dimension $6$ and exactly two connected components, each a copy of $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$; it is larger and is a real group rather than a complex one. Skolem–Noether describes the identity component $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$; the conjugate-linear coset exists because $\mathbb{C}/\mathbb{R}$ has a nontrivial Galois automorphism, and it is not inner.
Derivations over $\mathbb{C}$
Throughout this section the ground field is $\mathbb{C}$.
Definition. A $\mathbb{C}$-linear derivation of $\mathbb{B}$ is a $\mathbb{C}$-linear map $D : \mathbb{B} \to \mathbb{B}$ with $D(\tilde{Q}_2\tilde{Q}_4) = D(\tilde{Q}_2)\,\tilde{Q}_4 + \tilde{Q}_2\,D(\tilde{Q}_4)$ for all $\tilde{Q}_2, \tilde{Q}_4 \in \mathbb{B}$. The set of all such maps is a complex vector space, written $\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$, and it is a Lie algebra under the commutator bracket $[D_1, D_2] = D_1 \circ D_2 - D_2 \circ D_1$.
Every derivation satisfies $D(e_t) = 0$, since $D(e_t) = D(e_t e_t) = 2D(e_t)$.
Inner derivations. For each $\tilde Q_3 \in \mathbb{B}$ the map $\mathrm{ad}_{\tilde Q_3}(\tilde{Q}_2) = \tilde Q_3 \tilde{Q}_2 - \tilde{Q}_2 \tilde Q_3 = [\tilde Q_3, \tilde{Q}_2]$ is a $\mathbb{C}$-linear derivation, the inner derivation determined by $\tilde Q_3$; the Jacobi identity in the form $[[\tilde Q_3,\tilde{Q}_2],\tilde{Q}_4] + [\tilde{Q}_2,[\tilde Q_3,\tilde{Q}_4]] = [\tilde Q_3,[\tilde{Q}_2,\tilde{Q}_4]]$ is exactly the Leibniz rule for $\mathrm{ad}_{\tilde Q_3}$. The map $\mathrm{ad} : \mathbb{B} \to \mathrm{Der}_{\mathbb{C}}(\mathbb{B})$, $\tilde Q_3 \mapsto \mathrm{ad}_{\tilde Q_3}$, is $\mathbb{C}$-linear with kernel the centre, since $\mathrm{ad}_{\tilde Q_3} = 0$ says exactly that $\tilde Q_3$ commutes with everything: $\ker(\mathrm{ad}) = Z(\mathbb{B}) = \mathbb{B}_{C}$.
Every derivation is inner. For a central simple algebra over a field every derivation is inner. Hence $\mathrm{ad}$ is surjective and induces a $\mathbb{C}$-linear isomorphism
$$ \boxed{\;\mathrm{Der}_{\mathbb{C}}(\mathbb{B}) \cong \mathbb{B} / \mathbb{B}_{C}\;}. $$
Identification with the scalar-free part. The quotient $\mathbb{B}/\mathbb{B}_{C}$ is the scalar-free part of $\mathbb{B}$, the subspace of vanishing scalar part $\{\tilde{Q} : Q_t = 0\} = \mathrm{span}_{\mathbb{C}}\{\vec e_x, \vec e_y, \vec e_z\}$, the complex pure-vector part of $\mathbb{B}$, a complex vector space of dimension $3$. The isomorphism is one of Lie algebras, because $[\mathrm{ad}_{\tilde Q_3}, \mathrm{ad}_{\tilde Q_1}] = \mathrm{ad}_{[\tilde Q_3,\tilde Q_1]}$.
Dimension. The derivation space has $\dim_{\mathbb{C}} \mathrm{Der}_{\mathbb{C}}(\mathbb{B}) = 3$, hence real dimension $6$. This matches the automorphism group, which also has complex dimension $3$; its Lie algebra is $\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$.
The bivector reading. Under the identification $\mathbb{B} \cong \mathrm{Cl}^{+}_{1,3}$, the real span $\mathrm{span}_{\mathbb{R}}\{\vec e_x, \vec e_y, \vec e_z, i \vec e_x, i \vec e_y, i \vec e_z\}$ is the six-dimensional bivector subspace, the real form of the complex pure-vector part, and its elements bracket into themselves. Thus the derivation algebra, viewed as a real Lie algebra, is the bivector part of $\mathrm{Cl}^{+}_{1,3}$, equivalently, as a real Lie algebra, the realification of the complex pure-vector part, of real dimension $6$, which is the Lie algebra of the Lorentz group. The traceless part, the complex pure-vector part, and the bivector part are three descriptions of one object, with $\dim_{\mathbb{R}} = 2 \dim_{\mathbb{C}}$.
An explicit basis. The three derivations $D_1 = \tfrac{1}{2}\mathrm{ad}_{\vec e_x}$, $D_2 = \tfrac{1}{2}\mathrm{ad}_{\vec e_y}$, $D_3 = \tfrac{1}{2}\mathrm{ad}_{\vec e_z}$ are $\mathbb{C}$-linear, so each vanishes on $e_t$ and on $i$, and they act on $\vec e_x, \vec e_y, \vec e_z$ by
$$ D_1(\vec e_y) = \vec e_z, \quad D_1(\vec e_z) = -\vec e_y; \qquad D_2(\vec e_z) = \vec e_x, \quad D_2(\vec e_x) = -\vec e_z; \qquad D_3(\vec e_x) = \vec e_y, \quad D_3(\vec e_y) = -\vec e_x, $$
with $D_1(\vec e_x) = D_2(\vec e_y) = D_3(\vec e_z) = 0$. They satisfy the standard commutation relations,
$$ [D_1, D_2] = D_3, \qquad [D_2, D_3] = D_1, \qquad [D_3, D_1] = D_2, $$
so $\{D_1, D_2, D_3\}$ is a complex basis of $\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$. For instance $D_1(\vec e_y) = \tfrac{1}{2}(\vec e_x \vec e_y - \vec e_y \vec e_x) = \tfrac{1}{2}(\vec e_z + \vec e_z) = \vec e_z$, and $[D_1, D_2] = \tfrac{1}{4}\mathrm{ad}_{[\vec e_x,\vec e_y]} = \tfrac{1}{4}\mathrm{ad}_{2\vec e_z} = D_3$.
Derivations and automorphisms. The two structures are linked by the exponential: $\exp(t\,\mathrm{ad}_{\tilde Q_3})(\tilde{Q}_2) = \exp(t\tilde Q_3)\, \tilde{Q}_2\, \exp(-t\tilde Q_3)$ for $\tilde Q_3 \in \mathbb{B}$, $t \in \mathbb{R}$. The right-hand side is the inner automorphism determined by $\exp(t\tilde Q_3)$, so the Lie algebra of the automorphism group is the derivation algebra, $\mathrm{Lie}\,\mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) = \mathrm{Der}_{\mathbb{C}}(\mathbb{B})$.
Derivations over $\mathbb{R}$
Now the ground field is $\mathbb{R}$. An $\mathbb{R}$-linear derivation is required to satisfy the Leibniz rule but need not be $\mathbb{C}$-linear. At first sight this seems to allow a larger space, but in fact it does not.
Every real derivation is automatically $\mathbb{C}$-linear. Let $D$ be an $\mathbb{R}$-linear derivation. As in the automorphism case, $D$ maps the centre into itself: if $A$ is central, then for every $\tilde{Q}_2$,
$$ D(A)\tilde{Q}_2 = D(A\tilde{Q}_2) - AD(\tilde{Q}_2) = D(\tilde{Q}_2A) - D(\tilde{Q}_2)A = \tilde{Q}_2D(A). $$
So $D$ restricts to a derivation $\mathbb{B}_{C} \to \mathbb{B}_{C}$. But $\mathbb{C}$ has no nonzero $\mathbb{R}$-linear derivations: a derivation of $\mathbb{C}$ is determined by $D(i)$, and $0 = D(-1) = D(i^{2}) = i\,D(i) + D(i)\,i = 2i\,D(i)$ forces $D(i) = 0$, hence $D$ vanishes on the centre. Since $i$ is central, the Leibniz rule then gives $D(i \tilde{Q}_2) = D(i)\,\tilde{Q}_2 + i\,D(\tilde{Q}_2) = i\,D(\tilde{Q}_2)$, so $D$ is $\mathbb{C}$-linear. Therefore $\mathrm{Der}_{\mathbb{R}}(\mathbb{B}) = \mathrm{Der}_{\mathbb{C}}(\mathbb{B})$.
There is no semilinear analogue for derivations: a derivation cannot conjugate a coefficient.
Dimension and structure. Consequently the real derivation space has real dimension $6$, that is, complex dimension $3$, and $\mathrm{Der}_{\mathbb{R}}(\mathbb{B}) = \mathrm{Der}_{\mathbb{C}}(\mathbb{B})$, the realification of the complex derivation algebra. As a real Lie algebra this is the orthogonal Lie algebra $\mathrm{SO}(1,3)$ of the Lorentz group, equivalently the bivector subspace of $\mathrm{Cl}^{+}_{1,3}$ under the commutator. The derivations $D_1, D_2, D_3$ of the previous section span it over $\mathbb{C}$, and together with $iD_1, iD_2, iD_3$ over $\mathbb{R}$.
Summary of the asymmetry. For automorphisms, the real group is strictly larger than the complex one, because complex conjugation supplies a second coset. For derivations, the real and complex spaces coincide, because the centre is étale over $\mathbb{R}$ and admits no nonzero derivation. This is a ground-field distinction and not a convention.
Worked Examples
The conjugation involution. Complex conjugation $c(\tilde{Q}) = \bar{\tilde{Q}}$ is an $\mathbb{R}$-algebra automorphism with $c(\vec e_k) = \vec e_k$ ($k = 0,1,2,3$) and $c(i) = -i$; it fixes the quaternion subspace $\mathbb{B}_{H}$ pointwise, negates the scalar imaginary, satisfies $c^{2} = \mathrm{id}$, preserves the product, and is conjugate-linear over $\mathbb{C}$. It is not inner, because inner automorphisms fix the centre pointwise whereas $c(i) = -i$, so it represents the nontrivial coset of $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$ in $\mathrm{Aut}_{\mathbb{R}}(\mathbb{B})$.
A rotation derivation. Take $D = D_3 = \tfrac{1}{2}\mathrm{ad}_{\vec e_z}$, so that $D(\vec e_x) = \vec e_y$ and $D(\vec e_y) = -\vec e_x$. Its exponential acts by
$$ \exp(tD)(\vec e_x) = \cos t\, \vec e_x + \sin t\, \vec e_y, $$
as one checks by summing the series. Equivalently, $\exp(t \vec e_z/2) = \cos(t/2) + \sin(t/2) \vec e_z$, and conjugation by this unit quaternion rotates the $\vec e_x$-$\vec e_y$ plane, matching $\exp(t D) = \mathrm{Ad}_{\exp(t \vec e_z/2)}$.
Every element of $\mathrm{span}_{\mathbb{C}}\{\vec e_x, \vec e_y, \vec e_z\}$ gives an inner derivation, and $\mathrm{ad}_{\tilde Q_3}$ depends only on $\tilde Q_3$ modulo the centre $\mathbb{B}_{C}$; for instance $\tilde Q_3 = \vec e_x + i \vec e_y$ gives a derivation that is not a scalar multiple of any $D_k$.
Summary
The two ground fields give the following table; the field is stated explicitly in every entry.
| Structure | Over $\mathbb{C}$ | Over $\mathbb{R}$ |
|---|---|---|
| Algebra | simple $\mathbb{C}$-algebra, complex dimension $4$ | simple $\mathbb{R}$-algebra, real dimension $8$ |
| Ideals | $\{0\}$ and $\mathbb{B}$ only | $\{0\}$ and $\mathbb{B}$ only |
| Centre | $\mathbb{B}_{C} = \mathbb{C} e_t$, dimension $1$ over $\mathbb{C}$ | $\mathbb{B}_{C} \cong \mathbb{C}$, dimension $2$ over $\mathbb{R}$ |
| Central simple? | yes, central simple over $\mathbb{C}$ | no: simple, but centre $\mathbb{C} \neq \mathbb{R}$ |
| Automorphism group | $\mathbb{B}^{\times}/\mathbb{C}^{\times}$, complex dimension $3$ ($6$ over $\mathbb{R}$), connected | $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \rtimes \mathbb{Z}/2$, real dimension $6$, two components |
| Derivation space (the Lie algebra of the automorphism group) | $\mathrm{Der}_{\mathbb{C}}(\mathbb{B}) \cong \mathbb{B}/\mathbb{B}_{C}$, complex dimension $3$ ($6$ over $\mathbb{R}$); scalar-free part, complex pure-vector part, bivectors | $\mathrm{Der}_{\mathbb{R}}(\mathbb{B}) = \mathrm{Der}_{\mathbb{C}}(\mathbb{B})$, real dimension $6$, isomorphic to the Lorentz algebra $\mathrm{SO}(1,3)$ |
In summary: over $\mathbb{C}$ the algebra is central simple, every automorphism is inner, and $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \cong \mathbb{B}^{\times}/\mathbb{C}^{\times}$; over $\mathbb{R}$ complex conjugation adds a second, conjugate-linear coset, so $\mathrm{Aut}_{\mathbb{R}}(\mathbb{B}) \cong \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \rtimes \mathbb{Z}/2$ is strictly larger; and over either field the derivations are the inner derivations $\tilde{Q}_2 \mapsto [\tilde Q_3,\tilde{Q}_2]$ with $\tilde Q_3$ scalar-free, forming $\mathbb{B}/\mathbb{B}_{C}$, of dimension $3$ over $\mathbb{C}$ and $6$ over $\mathbb{R}$, identified with the bivector part of $\mathrm{Cl}^{+}_{1,3}$.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\mathbb{B} = \mathbb{C} \otimes_{\mathbb{R}} \mathbb{H}$ | Biquaternion algebra; complex dimension $4$, real dimension $8$ |
| $e_t, \vec e_x, \vec e_y, \vec e_z$ | Algebra basis, $e_t = 1$, $\vec e_k^2 = -e_t$ |
| $i$ | Central scalar imaginary, $i^2 = -1$ |
| $\mathbb{B}_{C} = \mathrm{span}_{\mathbb{R}}\{e_t, ie_t\}$ | Centre of $\mathbb{B}$; the scalar subspace |
| $\mathbb{B}_{V}$ | Vector subspace, vanishing scalar part |
| $\mathbb{B}_{H}, \mathbb{B}_{iH}$ | Real-quaternion and anti-quaternion subspaces |
| $\mathbb{B}_{\Psi}, \mathbb{B}_{\Phi}$ | Hermitian and anti-Hermitian subspaces |
| ${}^{\natural}, \bar{\cdot}, {}^{*}, {}^{\flat}$ | Quaternion, complex, Hermitian and anti-Hermitian conjugations |
| $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$ | Algebra automorphisms over $\mathbb{C}$, $\cong \mathbb{B}^{\times}/\mathbb{C}^{\times}$ |
| $\mathrm{Aut}_{\mathbb{R}}(\mathbb{B})$ | Algebra automorphisms over $\mathbb{R}$, $\cong \mathrm{Aut}_{\mathbb{C}}(\mathbb{B}) \rtimes \mathbb{Z}/2$ |
| $\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$ | Derivations over $\mathbb{C}$, $\cong \mathbb{B}/\mathbb{B}_{C}$ |
| $\mathrm{Der}_{\mathbb{R}}(\mathbb{B})$ | Derivations over $\mathbb{R}$, $\cong \mathrm{SO}(1,3)$ as a real Lie algebra |
| $\mathrm{ad}_{\tilde Q_3}(\tilde{Q}_2) = [\tilde Q_3,\tilde{Q}_2]$ | Inner derivation by $\tilde Q_3$ |
| $D_k = \tfrac{1}{2}\mathrm{ad}_{\vec e_k}$ | Basis of $\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$, $[D_1,D_2] = D_3$ etc. |
| $\mathrm{Cl}_{1,3}^{+}$ | Even Clifford algebra; isomorphic to $\mathbb{B}$ |
Further Reading
- Richard S. Pierce, Associative Algebras, Graduate Texts in Mathematics 88, Springer, 1982.
- I. N. Herstein, Noncommutative Rings, Carus Mathematical Monographs 15, Mathematical Association of America, 1968.
- Benson Farb and R. Keith Dennis, Noncommutative Algebra, Graduate Texts in Mathematics 144, Springer, 1993.
- John Voight, Quaternion Algebras, Graduate Texts in Mathematics 288, Springer, 2021.
- Max-Albert Knus, Quadratic and Hermitian Forms over Rings, Grundlehren der mathematischen Wissenschaften 294, Springer, 1991.
- William Fulton and Joe Harris, Representation Theory: A First Course, Graduate Texts in Mathematics 129, Springer, 1991.
- Pertti Lounesto, Clifford Algebras and Spinors, 2nd edition, Cambridge University Press, 2001.
- Brian C. Hall, Lie Groups, Lie Algebras, and Representations, 2nd edition, Graduate Texts in Mathematics 222, Springer, 2015.