The Rotations, the Boosts and the Groups

Introduction

This entry reads the eighteenth section of the mathematical category: the groups of the norm. Six articles, one group and two actions. The group is the set of invertibles of the algebra, and the two actions are the two ways an invertible element can act on the algebra it belongs to. The mathematics states both actions and the groups each one generates: the adjoint, which preserves the product, and the congruence by the star, which preserves the two Hermitian sectors and the indefinite form. The entry reads the two actions against each other, which the mathematics does not do, and that comparison is the whole of its content. The one-sided actions by left and right multiplication are read with them, because they are what carries the element, and the reflections because they are the one family of isometries the two-sided action cannot reach.

The first claim is that there is one group and two actions, not two groups. Every unit $\tilde U$ acts on the algebra by

$$ \mathrm{Ad}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{-1}, \qquad \mathrm{Cg}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{*}, $$

and the two maps differ only in the factor on the right, the inverse against the dagger. The adjoint is multiplicative and carries the product to the product. The congruence is not multiplicative, and it carries each of the two Hermitian sectors into itself. The two maps coincide exactly on the $*$-unitary elements, that is on the compact slice $\mathrm{U}(\mathbb{B})$. Verified: on one hundred random $*$-unitary elements the two maps differ by at most $2.1\times10^{-15}$ in Euclidean norm, and on a boost of rapidity two they differ by $6.9$.

The second claim is that this one coincidence is what separates a rotation from a boost. A rotation lies in the coincidence, so it is an automorphism and an isometry at once, and there is no ambiguity in calling it a symmetry. A boost lies in one family each time: the congruence is an isometry of the form that preserves the sectors and is not multiplicative, and the adjoint is an automorphism that mixes the sectors. Verified: for a boost of rapidity two the adjoint fixes $e_t$ to $5.0\times10^{-16}$ while the congruence sends it to $\cosh 2\,e_t+i\sinh 2\,\vec u_{\mathrm b}$, whose scalar part is $3.762196$; the congruence preserves both sectors to $8.7\times10^{-15}$ and the adjoint leaves the Hermitian sector by $36.5$.

The third claim is that the algebra reaches one of the four components of $O(1,3)$ through its units and the other three through two operations that are not motions. The congruence of an invertible element is always proper and orthochronous, of determinant $+1$ and positive time orientation; on the modulus-one slice it is an isometry of the form as well. Verified on one hundred units: the determinant is $1$ to $4\times10^{-14}$, the time entry is at least $1$, and the interval form is preserved to $2.2\times10^{-16}$. The improper component is reached by conjugating the argument with the quaternion conjugation, the anti-orthochronous one by negating the result, and the fourth by both at once. The twofold cover is the same statement one level down: the operations of even length lie in the algebra, the odd ones do not, and $\mathrm{Pin}(1,3)$ is therefore not contained in $\mathbb{B}$.

The fourth claim is that the automorphism group over $\mathbb{R}$ is the complex one times a reflection, and the reflection is conjugate-linear. Over $\mathbb{C}$ every automorphism is inner, so $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})=\mathbb{B}^{\times}/\mathbb{C}^{\times}\cong PGL_2(\mathbb{C})\cong SO^{+}(1,3)$, of real dimension six. Over $\mathbb{R}$ the group gains one component, the coset of complex conjugation, which negates the central imaginary and fixes the quaternion units. That coset is not inner, because every inner automorphism fixes the centre pointwise. In the language of quantum mechanics it is an antiunitary operator, and its existence is the algebraic reason a reversal of time cannot be a motion.

The fifth claim is that the compact group is the gauge group and carries the whole topology. The $*$-unitary biquaternions are $\mathrm{U}(\mathbb{B})\cong\mathrm{U}(2)$, the maximal compact subgroup of the group of invertibles, and they split as a phase circle times the unit quaternions, $\mathrm{U}(\mathbb{B})=S^{1}\cdot S^{3}$ with intersection $\{\pm e_t\}$. The polar decomposition retracts the group of invertibles onto it and the norm-one group onto $S^{3}$, so the group of invertibles and the compact group have the same homotopy type: $\pi_1\cong\mathbb{Z}$, $\pi_2=0$, $\pi_3\cong\mathbb{Z}$. No homotopy class distinguishes a boost from a rotation; only the phase winds.

The sixth claim is that the indefinite form has an isometry group larger than the group the algebra's elements generate, and the gap is nine dimensions. The Krein isometry group is $U_{J}(\mathbb{B})\cong\mathrm{U}(1,3)$, of real dimension sixteen. The algebra's two-sided operators supply the norm-one congruences, which form $PSL_2(\mathbb{C})\cong SO^{+}(1,3)$ of dimension six, together with the central phases, of dimension one. Seven of the sixteen dimensions are reachable by elements of the algebra. The other nine are isometries of a form that the product does not see.

The one-sided actions complete the list without changing the thesis. Left and right multiplication by an invertible element are representations of the same group, of eight real dimensions each, and they are what carries the element: they preserve the definite form and not the product, so they act on the algebra as a module. The congruence is the two of them used in tandem, $L_{\tilde Q}R_{\tilde Q^{*}}$, and the group the algebra's actions generate on itself is seven-dimensional inside the sixteen of $\mathrm{U}(4)$.

The reflections sit outside the two-sided action, and the sign is the reason. On the vector subspace the mirror is $\rho_v=-v\tilde Xv^{\natural}$, which is minus a congruence, because on the real form the dagger is minus the element. Every isometry of that three-dimensional subspace is a product of at most three of them, and no continuous path of invertibles produces one.

The first objection is that this is a catalogue of standard groups. It is a catalogue, and the catalogue is short: five groups, four of them inside the algebra and one acting on it, and one table compares them. The content is not the list. It is the coincidence of the two actions on the compact slice, which makes "a rotation is a symmetry and a boost is not" an algebraic statement rather than a convention, and the kernels, the scales and the two images of one group, which differ by exactly the scale.

The second objection is that a statement about groups is not a statement about physics. It is not, and the entry does not make one. Each group is attached to the structure it preserves, the reach of each action is computed, and the eight readings at the end are the archive's, each with its warrant and its limit. The entry contributes the arrangement and the boundaries, not a physical claim.

The letters. The algebra is $\mathbb{B}$ with basis $e_t,\vec e_x,\vec e_y,\vec e_z$ and $\vec e_k^{2}=-e_t$; the element is $\tilde Q=Q_te_t+\vec Q$; the group of invertibles is $\mathbb{B}^{\times}=\{N\neq0\}$ and the norm-one group is $\mathbb{B}^{\times}_1=\{N=1\}$; the $*$-unitary group is $\mathrm{U}(\mathbb{B})=\{\tilde U^{*}\tilde U=e_t\}$; the adjoint action is $\mathrm{Ad}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{-1}$, the congruence is $\mathrm{Cg}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{*}$, and the bilinear two-sided operator of the mathematics is $\mathrm{H}_{\tilde Q}$, which is $\mathrm{Cg}_{\tilde Q}$ read as an operator; the Hermitian sectors are $\mathbb{B}_{\Psi}=\{\tilde X^{*}=\tilde X\}$ and $\mathbb{B}_{\Phi}=\{\tilde X^{*}=-\tilde X\}$; the derivations are $\mathrm{Der}(\mathbb{B})$ and the automorphisms are $\mathrm{Aut}(\mathbb{B})$, with the ground field named whenever it matters.

Boundaries. The double covers, the rotor, the two-sided operator and its action on the remarkable subspaces are Biquaternion Rotations and Lorentz Transformations; the automorphism group over each ground field, the derivations and their exponential are Biquaternion Automorphisms and Derivations; the Clifford envelope, the grading of the versors, the four components and the volume element are Biquaternion Versors and the Orthogonal Group; the compact slice, the two retractions and the homotopy are The Unitary Group of the Biquaternion Algebra; the indefinite isometry group, its two involutions and its Cartan decomposition are The Krein Isometry Group and Its Cartan Decomposition; the three level sets, the positive lines and the ball are The Krein Level Sets and the Hyperbolic Structure. The physical reading of the frames is the archived Frames, Rotations and the Lorentz Group, of the compact slice The Unitary Group: Gauge, Phase and the Double Cover, of the indefinite form The Krein Form and the Indefinite Structure.

The Source Table

The six articles fall into two pairs and two singletons. The first pair is the geometry of the motions: the double covers, the rotor, the two-sided operator. The second pair is the symmetry of the algebra itself: the automorphisms and their infinitesimal form. The two singletons are the two ends of the range: the Clifford envelope, which says which isometries the algebra can carry, and the indefinite form, which says which isometries exist at all.

mathematical content source article
The double covers $Sp(1)\to SO(3)$ and $\mathbb{B}^{\times}_1\to SO^{+}(1,3)$; the rotor and the half angle; the bilinear two-sided operator, its kernel $\mathrm{U}(1)$ and its scaling by $\lvert N(\tilde Q)\rvert^{2}$; the orbits, the rank invariant and the mass shell; the reflections and Cartan–Dieudonné; the two-sided action and $SO(4)$; the representations of the group of invertibles Biquaternion Rotations and Lorentz Transformations
The simplicity and the centre; $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})=PGL_2(\mathbb{C})$; $\mathrm{Aut}_{\mathbb{R}}(\mathbb{B})=\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})\rtimes\mathbb{Z}/2$; the inner derivations and $\mathrm{Der}(\mathbb{B})=\mathbb{B}/\mathbb{B}_{C}$; the bivector reading and the exponential Biquaternion Automorphisms and Derivations
The Clifford envelope $\mathrm{Cl}_{1,3}=\mathbb{B}\oplus\mathbb{B}\zeta$; the parity of the length as the reflection count; $\mathrm{Pin}(1,3)\not\subseteq\mathbb{B}$ and $\mathrm{Spin}(1,3)\cong\mathbb{B}^{\times}_1$; the four components and the two discrete operations; the volume element; the two involutions Biquaternion Versors and the Orthogonal Group
$\mathrm{U}(\mathbb{B})=\{\tilde U^{*}\tilde U=e_t\}\cong\mathrm{U}(2)$; the normal form; $S^{1}\cdot S^{3}$ and the flag variety; the two retractions; the homotopy groups and the universal cover; $\mathrm{U}(\mathbb{B})/U(1)\cong SO(3)$ The Unitary Group of the Biquaternion Algebra
$U_{J}(\mathbb{B})\cong\mathrm{U}(1,3)$; the criterion $T^{*}ET=E$; the isometries among the three families; the boosts and their spectrum; the Krein and Cartan involutions; the symmetric space and the contractions The Krein Isometry Group and Its Cartan Decomposition
The positive, negative and null level sets and their homotopy types; the positive region and the positive lines; $\mathbb{CH}^{3}$ and the Bergman metric; the isotropic boundary The Krein Level Sets and the Hyperbolic Structure

The Results in Brief

The group of invertibles. $\mathbb{B}^{\times}=\{N\neq0\}\cong GL_2(\mathbb{C})$, of real dimension eight, connected and non-compact, with centre $\mathbb{C}^{\times}$. Its subgroup of norm one is $\mathbb{B}^{\times}_1\cong SL_2(\mathbb{C})$, of real dimension six, with centre $\{\pm e_t\}$, and the quotient $\mathbb{B}^{\times}_1/\{\pm e_t\}\cong SO^{+}(1,3)$ is the proper orthochronous Lorentz group.

The adjoint is an automorphism and the congruence is not. $\mathrm{Ad}_{\tilde U}(\tilde X\tilde Y)=\mathrm{Ad}_{\tilde U}(\tilde X)\,\mathrm{Ad}_{\tilde U}(\tilde Y)$ for every invertible element, to $2.7\times10^{-13}$ on one hundred random triples. $\mathrm{Cg}_{\tilde U}$ fails the same identity by a relative error of order one, and preserves the sectors instead: $\mathrm{Cg}_{\tilde U}(\mathbb{B}_{\Psi})\subseteq\mathbb{B}_{\Psi}$ and $\mathrm{Cg}_{\tilde U}(\mathbb{B}_{\Phi})\subseteq\mathbb{B}_{\Phi}$, to $9\times10^{-15}$.

The coincidence. $\mathrm{Cg}_{\tilde U}=\mathrm{Ad}_{\tilde U}$ exactly when $\tilde U^{*}=\tilde U^{-1}$, that is on $\mathrm{U}(\mathbb{B})$. Verified at $2.1\times10^{-15}$ over one hundred $*$-unitary elements and violated by $6.9$ at a boost of rapidity two.

The scaling and the kernel. $\mathrm{Cg}_{\tilde U}$ multiplies the biquaternion norm by $\lvert N(\tilde U)\rvert^{2}$, so it is an isometry of the indefinite form exactly on the modulus-one slice; the adjoint preserves $N$ identically. The kernel of the congruence on the group of invertibles is the central circle $\mathrm{U}(1)$, since a central phase $\lambda$ gives $\mathrm{Cg}_{\lambda}=\lvert\lambda\rvert^{2}\mathrm{id}$; the kernel of the adjoint is $\mathbb{C}^{\times}$.

The rotor. $\tilde u=\cos\tfrac{\theta}{2}+\sin\tfrac{\theta}{2}\vec u$ conjugates the vector part by the rotation of angle $\theta$ about $\vec u$, and $\tilde u$ and $-\tilde u$ give the same rotation; at $\theta=2\pi$ the rotor is $-e_t$. Verified: the unit vector $\vec e_x$ is carried to $\cos\theta\,\vec e_x+\sin\theta\,\vec e_y$ at $\theta=0.5,\pi/2,\pi$.

The one-sided actions. $L_{\tilde Q}(\tilde X)=\tilde Q\tilde X$ and $R_{\tilde Q}(\tilde X)=\tilde X\tilde Q$ are representations of the same group, of eight real dimensions each. For $*$-unitary $\tilde Q$ the left action preserves $\langle\tilde X,\tilde Y\rangle_{*}$ to $4.4\times10^{-15}$ and fails multiplicativity by $18.7$, so it acts on the algebra as a module; the congruence is the pair in tandem, $\mathrm{Cg}_{\tilde Q}=L_{\tilde Q}R_{\tilde Q^{*}}$, to $9.2\times10^{-16}$. The group they generate is seven-dimensional inside the sixteen of $\mathrm{U}(4)$.

The reflections. On the real vector subspace the dagger is minus the element, so $\mathrm{Cg}_v$ is the half turn about $v$, of determinant $+1$ there, and the mirror is $\rho_v=-v\tilde Xv^{\natural}$: it negates $v$, fixes $v^{\perp}$, has determinant $-1$ on the subspace and $+1$ on the algebra, and squares to the identity. Cartan–Dieudonné gives $\mathrm{O}(3)$ from at most three of them.

The four components. $\mathrm{Cg}_{\tilde U}$ with $\lvert N(\tilde U)\rvert=1$ is proper orthochronous. Quaternion conjugation of the argument supplies parity, negation supplies the product of parity with time reversal, and the two together supply the fourth component. Verified on the $4\times4$ real matrices of the four operations: determinants $+1,-1,+1,-1$ and time entries $+1.337,+1.337,-1.337,-1.337$.

The automorphisms. $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})\cong PGL_2(\mathbb{C})$, of complex dimension three, all of it inner. $\mathrm{Aut}_{\mathbb{R}}(\mathbb{B})\cong\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})\rtimes\mathbb{Z}/2$, of real dimension six and two components, the second the coset of complex conjugation, which is conjugate-linear and not inner.

The derivations. $\mathrm{Der}_{\mathbb{C}}(\mathbb{B})\cong\mathbb{B}/\mathbb{B}_{C}$, the scalar-free part, of complex dimension three; $\mathrm{Der}_{\mathbb{R}}(\mathbb{B})=\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$, because the centre admits no nonzero real derivation. The three derivations $D_k=\tfrac12\mathrm{ad}_{\vec e_k}$ satisfy $[D_1,D_2]=D_3$ and the cyclic relations, and $\exp(t\,\mathrm{ad}_{\tilde Q})=\mathrm{Ad}_{\exp(t\tilde Q)}$.

The versors. $\mathrm{Cl}_{1,3}=\mathbb{B}\oplus\mathbb{B}\zeta$ with the algebra as the even slot; the parity of the length of a versor is the parity of its reflection count; $\mathrm{Pin}(1,3)$ is not contained in $\mathbb{B}$ and $\mathrm{Spin}(1,3)=\mathbb{B}^{\times}_1$ is its whole intersection with the algebra. The volume element is the central imaginary up to sign, acts as $-\mathrm{id}$ on the Lorentzian space through the signed conjugation, and as $+\mathrm{id}$ through the two-sided operator.

The compact group. $\mathrm{U}(\mathbb{B})\cong\mathrm{U}(2)$, of real dimension four, equal to $S^{1}\cdot S^{3}$ with intersection $\{\pm e_t\}$ and homeomorphic to $S^{1}\times S^{3}$; maximal torus $T^{2}$ and flag variety $S^{2}$; the two retractions; $\pi_1\cong\mathbb{Z}$ and $\pi_3\cong\mathbb{Z}$; $\mathrm{U}(\mathbb{B})/S^{3}\cong S^{1}$ and $\mathrm{U}(\mathbb{B})/U(1)\cong SO(3)$.

The indefinite group. $U_{J}(\mathbb{B})\cong\mathrm{U}(1,3)$, of real dimension sixteen, non-compact, with centre $\mathrm{U}(1)$ and maximal compact subgroup $\mathrm{U}(1)\times\mathrm{U}(3)$ of dimension ten. The central phases and the unitary two-sided operators generate $\mathrm{U}(1)\times PU(2)\cong\mathrm{U}(1)\times SO(3)$, of dimension four; the norm-one congruences form $PSL_2(\mathbb{C})\cong SO^{+}(1,3)$, of dimension six. The boosts $T_t$ have spectrum $\{e^{t},e^{-t},1,1\}$ and restrict to the Lorentz boost of rapidity $t$ on the quaternionic slice.

The level sets. The levels read here are those of the indefinite form: $\{\langle\tilde Q,\tilde Q\rangle_{\natural*}=1\}\cong S^{1}\times\mathbb{R}^{6}$, $\{\langle\tilde Q,\tilde Q\rangle_{\natural*}=-1\}\cong S^{5}\times\mathbb{R}^{2}$, and the null set a cone on $S^{1}\times S^{5}$, contractible. The positive lines form the open unit ball of $\mathbb{C}^{3}$, which is $\mathbb{CH}^{3}=\mathrm{U}(1,3)/(\mathrm{U}(1)\times\mathrm{U}(3))$, with boundary the isotropic sphere $S^{5}$.

The Chain of Groups

Five Groups and Five Invariants

The algebra carries four groups and one group acts on it. The four inside are the centre, the $*$-unitary group, the norm-one group and the group of invertibles; the one outside is the group of isometries of the indefinite form. Each is cut out by one equation, and the equation is a statement about what the group leaves alone.

$$ \boxed{\; \begin{aligned} S^{1}e_t&=\{\lambda e_t:\lvert\lambda\rvert=1\}, & \dim_{\mathbb{R}}&=1,\\ \mathrm{U}(\mathbb{B})&=\{\tilde U:\tilde U^{*}\tilde U=e_t\}\cong\mathrm{U}(2), & \dim_{\mathbb{R}}&=4,\\ \mathbb{B}^{\times}_1&=\{\tilde U:N(\tilde U)=1\}\cong SL_2(\mathbb{C}), & \dim_{\mathbb{R}}&=6,\\ \mathbb{B}^{\times}&=\{\tilde U:N(\tilde U)\neq0\}\cong GL_2(\mathbb{C}), & \dim_{\mathbb{R}}&=8,\\ U_{J}(\mathbb{B})&=\{T:T^{*}ET=E\}\cong\mathrm{U}(1,3), & \dim_{\mathbb{R}}&=16. \end{aligned}\;} $$

The four inside are nested only in part. The centre lies in both $\mathrm{U}(\mathbb{B})$ and $\mathbb{B}^{\times}_1$; each of those lies in $\mathbb{B}^{\times}$; and neither of the two middle groups contains the other. The intersection is the unit quaternions,

$$ \mathrm{U}(\mathbb{B})\cap\mathbb{B}^{\times}_1=S^{3}, $$

because a $*$-unitary element of norm one is a unit quaternion, and conversely. A pure phase $\lambda e_t$ has norm $\lambda^{2}$, which is one only for $\lambda=\pm1$, so the central circle is not inside the norm-one group; a phase times a unit quaternion has norm $\lambda^{2}$, so the compact group is not inside it either. The two middle groups are two different slices of the eight-dimensional group of invertibles, one cutting by the dagger and one by the norm, and they meet on the three-sphere.

This is the first structural remark of the entry. The compact slice and the norm-one slice are the two natural reductions of the group of invertibles, and they are not the same reduction. The dagger gives the maximal compact subgroup; the norm gives the double cover of the Lorentz group. They share the rotations and disagree on everything else: the phase belongs to the first and not the second, the boost to the second and not the first.

The Dimensions and the Containments

The dimension count reads the equations. The group of invertibles is open in an eight-dimensional real space, so it is eight. The norm is one complex equation, two real equations, so the norm-one group is six. The dagger condition $\tilde U^{*}\tilde U=e_t$ is a Hermitian matrix equation, four real equations, so the compact group is four. The centre is a circle. The indefinite isometry group is $\mathrm{U}(1,3)$, whose real dimension is $1^{2}+3^{2}+2\cdot1\cdot3=16$.

The difference between the last two rows is the point of the indefinite theory. The compact group is four-dimensional because the dagger is a definite form and its isometry group is compact and small. The indefinite form has a group four times larger, and the algebra's own elements reach a seven-dimensional part of it.

group equation real dimension compact what it preserves
$S^{1}e_t$ $\lvert\lambda\rvert=1$ $1$ yes every line, every sector, every product
$\mathrm{U}(\mathbb{B})$ $\tilde U^{*}\tilde U=e_t$ $4$ yes the dagger form, the two sectors, the norm
$\mathbb{B}^{\times}_1$ $N(\tilde U)=1$ $6$ no the norm, hence $\lvert N\rvert$ under the congruence
$\mathbb{B}^{\times}$ $N(\tilde U)\neq0$ $8$ no invertibility and the rank; $\lvert N\rvert^{2}$ under the congruence
$U_{J}(\mathbb{B})$ $T^{*}ET=E$ $16$ no the indefinite form, and nothing of the product

The table is read down the last column. Each group is the stabiliser of one object, and the objects are ordered by strength: a line, a form, a number, invertibility, an indefinite form. The two middle rows are the ones the physics uses, and they are the two rows whose preserved objects are a form and a number.

What Each One Preserves

The centre preserves everything, and that is why it acts trivially on the algebra and only by phase on the elements. It is the kernel of the projective actions below.

The compact group preserves the dagger form, the two Hermitian sectors and the norm. It acts on the algebra as an automorphism and as an isometry at the same time, because on this slice the two actions coincide. It is the only group of the five with that property.

The norm-one group preserves the norm under the adjoint, identically, and preserves the indefinite form under the congruence. It does not preserve the sectors under the adjoint and does not preserve the product under the congruence. It is the group of the motions of the Lorentzian slice, and the price of the motion is that no single action of it preserves everything the algebra has.

The group of invertibles preserves invertibility, hence the rank of the matrix image, and scales the norm by $\lvert N\rvert^{2}$ under the congruence. It is the group of all rescalings of the algebra, and its two actions are the two actions of the entry.

The indefinite group preserves the fourth form and nothing of the product. It is not generated by the algebra's elements, and the seven dimensions of overlap are the subject of §The Krein Group.

The Two Actions of One Group

The Adjoint and the Congruence

An invertible element $\tilde U$ acts on the algebra in two ways, and the two formulas differ in one symbol:

$$ \mathrm{Ad}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{-1}, \qquad \mathrm{Cg}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{*}. $$

Both are group homomorphisms from $\mathbb{B}^{\times}$ into the linear group of the eight-dimensional real space. That is, both are representations: $\mathrm{Ad}_{\tilde U\tilde V}=\mathrm{Ad}_{\tilde U}\mathrm{Ad}_{\tilde V}$ and $\mathrm{Cg}_{\tilde U\tilde V}=\mathrm{Cg}_{\tilde U}\mathrm{Cg}_{\tilde V}$, the second because $(\tilde U\tilde V)\tilde X(\tilde U\tilde V)^{*}=\tilde U(\tilde V\tilde X\tilde V^{*})\tilde U^{*}$.

What separates them is what they do to the product of two arguments. The adjoint is an algebra automorphism, $$ \mathrm{Ad}_{\tilde U}(\tilde X\tilde Y)=\mathrm{Ad}_{\tilde U}(\tilde X)\,\mathrm{Ad}_{\tilde U}(\tilde Y), $$ because the two inner factors cancel. The congruence is not, since $\tilde U\tilde X\tilde Y\tilde U^{*}$ and $\tilde U\tilde X\tilde U^{*}\tilde U\tilde Y\tilde U^{*}$ differ by a factor $\tilde U^{*}\tilde U$ in the middle. Verified on one hundred random triples: the adjoint identity holds to $2.7\times10^{-13}$, and the congruence identity fails with a relative error of $1.18$.

The reason the congruence is the right map for the form is that it is the matrix congruence $X\mapsto MXM^{\dagger}$, and a congruence preserves the inertia of a Hermitian form. The reason the adjoint is the right map for the product is that it is the inner automorphism. The mathematics uses both, and it names the difference: the two-sided operator of Biquaternion Rotations and Lorentz Transformations is the congruence, and the automorphism of Biquaternion Automorphisms and Derivations is the adjoint.

What the Congruence Preserves

The congruence preserves each Hermitian sector. The computation is one line, $$ (\tilde U\tilde X\tilde U^{*})^{*}=(\tilde U^{*})^{*}\tilde X^{*}\tilde U^{*}=\tilde U\tilde X^{*}\tilde U^{*}, $$ so $\tilde X^{*}=\pm\tilde X$ gives $(\tilde U\tilde X\tilde U^{*})^{*}=\pm\tilde U\tilde X\tilde U^{*}$. Verified: the Hermitian sector is preserved to $8.7\times10^{-15}$ and the anti-Hermitian sector to $6.0\times10^{-15}$, over one hundred random units.

The adjoint does not preserve them. The same computation with the inverse gives $(\tilde U\tilde X\tilde U^{-1})^{*}=\tilde U^{-*}\tilde X^{*}\tilde U^{*}$, which is not proportional to $\tilde U\tilde X\tilde U^{-1}$ unless $\tilde U^{*}\tilde U$ is central. Verified: an Hermitian element leaves the sector by $36.5$ in Euclidean norm after the adjoint of a random unit.

The congruence also scales the norm, by exactly the squared modulus of the norm of the acting element: $$ N\bigl(\mathrm{Cg}_{\tilde U}(\tilde X)\bigr)=\lvert N(\tilde U)\rvert^{2}N(\tilde X), $$ verified to $6.9\times10^{-13}$. The identity is the multiplicativity of the norm together with $N(\tilde U^{*})=\overline{N(\tilde U)}$. On the modulus-one slice the factor is one, and there the congruence is an isometry of the indefinite form; off it, it is a similarity.

The adjoint preserves the norm identically, since $\mathrm{Ad}$ is an automorphism and the norm is built from the product; verified to $2.9\times10^{-13}$. This is the sharpest way to state the difference in the norm: the adjoint cannot change the size of anything, and the congruence changes the size by the size of the actor.

The Coincidence on the Compact Slice

The two maps coincide exactly when the inverse and the dagger agree, that is when $\tilde U^{*}=\tilde U^{-1}$, that is when $\tilde U$ is $*$-unitary. On that slice both are the congruence, both preserve the product, both preserve the sectors, and the distinction of the previous two sections disappears.

Verified in both directions. On one hundred random $*$-unitary elements, built as $\exp(iH)$ of a random Hermitian matrix and pulled back to the algebra, the largest difference between the two maps is $2.1\times10^{-15}$ in Euclidean norm and the largest dagger defect $\lVert\tilde U^{*}\tilde U-e_t\rVert$ is $4.5\times10^{-16}$. At a boost of rapidity two, the dagger defect is $4.559$ and the difference between the two maps is $6.947$.

The coincidence is an if-and-only-if, and the proof is one line: if the two maps agree on every $\tilde X$ then they agree on $e_t$, so $\tilde U\tilde U^{-1}=\tilde U\tilde U^{*}$ and $\tilde U^{*}=\tilde U^{-1}$. The compact slice is therefore exactly the set of elements for which a single action is both a symmetry of the product and a symmetry of the form.

That is the sentence the entry exists to state. A rotation is an element of that slice; a boost is not. The rotation has one action because the two actions agree there. The boost has two, and the physics chooses between them.

The Kernel, the Scale and the Two Images

The two representations have different kernels, and the difference is the scale.

The adjoint kills the centre: $\mathrm{Ad}_{\lambda}=\mathrm{id}$ for every nonzero complex $\lambda$, so the kernel is $\mathbb{C}^{\times}$, of real dimension two, and the adjoint factors through $\mathbb{B}^{\times}/\mathbb{C}^{\times}\cong PGL_2(\mathbb{C})$, of dimension six. The adjoint therefore cannot see the scale at all.

In the projective space of the algebra that quotient is one phrase: the invertibles are the classes with $N\neq0$, so the group of invertibles modulo the scale is the complement of the null quadric of the norm, and that complement is $PGL_2(\mathbb{C})$. The null quadric is the set of classes of the zero divisors, so a degenerate element is a point of the projective space that the quotient leaves out.

The congruence kills only the phases: $\mathrm{Cg}_{\lambda}=\lvert\lambda\rvert^{2}\mathrm{id}$, so the kernel is the central circle $\mathrm{U}(1)$, of real dimension one, and the congruence factors through $\mathbb{B}^{\times}/\mathrm{U}(1)$, of dimension seven. Verified: $\mathrm{Cg}_{ie_t}$ is the identity to $0.0$ and $\mathrm{Cg}_{2e_t}$ is four times the identity to $0.0$. The congruence sees the scale, and reports it as a dilation of the whole algebra.

The images are therefore of dimensions six and seven, both inside the linear group of the eight-dimensional space, and they meet in the compact subgroup of dimension four: $$ \mathrm{Im}\,\mathrm{Ad}\cap\mathrm{Im}\,\mathrm{Cg}=\mathrm{U}(1)\cdot PU(2)\cong \mathrm{U}(1)\times SO(3). $$ The intersection is computed from the coincidence: the two maps agree exactly on $\mathrm{U}(\mathbb{B})$, whose image modulo the common kernel is $PU(2)\cong SO(3)$, and the central phases agree when their modulus is one. Four dimensions is the size of the set on which a motion of the algebra is also a symmetry of its multiplication.

adjoint $\mathrm{Ad}$ congruence $\mathrm{Cg}$
formula $\tilde U\tilde X\tilde U^{-1}$ $\tilde U\tilde X\tilde U^{*}$
kernel $\mathbb{C}^{\times}$, dimension $2$ $\mathrm{U}(1)$, dimension $1$
image $PGL_2(\mathbb{C})$, dimension $6$ $\mathbb{B}^{\times}/\mathrm{U}(1)$, dimension $7$
the product preserved not preserved
the sectors mixed preserved
the norm preserved identically scaled by $\lvert N(\tilde U)\rvert^{2}$
the indefinite form preserved on the compact part only preserved iff $\lvert N(\tilde U)\rvert=1$
agrees with the other on $\mathrm{U}(\mathbb{B})$ on $\mathrm{U}(\mathbb{B})$

The last row is not a symmetry of the table; it is the table's subject. Everything above it is a difference between two maps, and the last row is the set on which the difference vanishes.

The One-Sided Actions and the Group They Generate

The adjoint and the congruence are two-sided: one factor on the left and one on the right. The one-sided maps are representations of the same group,

$$ L_{\tilde Q}(\tilde X)=\tilde Q\tilde X, \qquad R_{\tilde Q}(\tilde X)=\tilde X\tilde Q, $$

with $L_{\tilde Q\tilde Q'}=L_{\tilde Q}L_{\tilde Q'}$ and $R_{\tilde Q\tilde Q'}=R_{\tilde Q'}R_{\tilde Q}$; the left one is a homomorphism, the right one an anti-homomorphism, and each is an eight-dimensional real space of operators.

A one-sided action preserves the form and not the algebra. For $\tilde Q$ $*$-unitary, $\langle\tilde Q\tilde X,\tilde Q\tilde Y\rangle_{*}=\langle\tilde X,\tilde Y\rangle_{*}$, verified to $4.4\times10^{-15}$ over one hundred random pairs, because $\tilde Q^{*}\tilde Q=e_t$ and the Euclidean form $\langle\tilde X,\tilde Y\rangle_{*}=\mathrm{Sc}(\tilde X\tilde Y^{*})$ is a Hermitian pairing. The same map fails multiplicativity on the same sample, with a defect of $18.7$ in Euclidean norm, and it does not preserve the Hermitian sectors: the left multiple of an Hermitian element has an Hermitian defect of $5.8$. So a one-sided action carries the element and neither the multiplication nor the involution: it acts on the algebra as a module and not as an algebra.

That is the action a gauge transformation performs on a spinor. The spinor is the element, the transformation is the left factor, and the four-vector is the same element read through the congruence on the Hermitian sector. The two readings are two representations of one group, and the difference between them is the difference between $L$ and $\mathrm{Cg}$, which is the right factor.

The congruence is the two one-sided actions used in tandem. On the compact slice $\mathrm{Cg}_{\tilde Q}=L_{\tilde Q}R_{\tilde Q^{*}}$, verified to $9.2\times10^{-16}$, and $R_{\tilde Q^{*}}=R_{\tilde Q}^{-1}$ there. The two-sided action therefore adds no ingredient beyond a left factor and a right factor; what it adds is the constraint that the two factors are one element and its inverse.

The actions generate seven dimensions inside sixteen. The left and the right images are each a copy of $\mathrm{U}(\mathbb{B})$, of dimension four, and they meet in the central circle, so the group $L_{\mathrm{U}(\mathbb{B})}R_{\mathrm{U}(\mathbb{B})}$ has dimension $4+4-1=7$. The Euclidean form on $\mathbb{B}\cong\mathbb{C}^{4}$ is preserved by $\mathrm{U}(4)$, of dimension sixteen, and that is its whole isometry group; the algebra's own actions supply seven of the sixteen. A generic element of $\mathrm{U}(4)$ is not of the shape $L_{\tilde Q}R_{\tilde Q'}$, and the witness is the image of $e_t$: such a pair sends $e_t$ to the product $\tilde Q\tilde Q'$, whose norm has modulus one, while the unit vectors of $\mathbb{C}^{4}$ carry no such constraint. Verified on two hundred random unit vectors: the moduli of their norms run from $0.081$ to $0.990$, and none is one.

The count is the one of the indefinite theory. Seven of sixteen there, seven of sixteen here, and in both cases the algebra's elements are the smaller part of the group that preserves the form.

The Rotation: An Automorphism

The Rotor and the Half Angle

A rotation is generated by a unit quaternion. For a real unit vector $\vec u$ and an angle $\theta$,

$$ \tilde u(\theta)=\cos\frac{\theta}{2}\,e_t+\sin\frac{\theta}{2}\,\vec u, \qquad \tilde u(\theta)\,\tilde X\,\tilde u(\theta)^{-1} $$

rotates the vector part of $\tilde X$ by $\theta$ about $\vec u$ and fixes the scalar part. Verified: with $\vec u=\vec e_z$ the unit vector $\vec e_x$ is carried to $\cos\theta\,\vec e_x+\sin\theta\,\vec e_y$, giving $(0.877583,0.479426,0)$ at $\theta=0.5$, $(0,1,0)$ at $\theta=\pi/2$, and $(-1,0,0)$ at $\theta=\pi$.

The angle in the rotor is half the angle of the rotation. The factor is not a convention of the parametrisation. The element acts twice, once on the left and once on the right, and each action contributes half the angle; the composition of the two gives the full turn.

The half angle is the rotor's coordinate and not the spin. The halved angle is what a square root does, and it is a coordinate on the two-sheeted cover. The spin of a carrier is a different number: it is half the highest weight, and on a half-integer carrier the weights are $\pm\tfrac12$ while on an integer one they are $-1,0,+1$. The rotor acts on a carrier by the exponential of the weight, so the halved angle follows from the weight and does not define it.

The rotor is a real quaternion of norm one, so it is $*$-unitary, so it lies in the compact slice, so the two actions of §The Two Actions of One Group coincide on it. A rotation is therefore an automorphism and an isometry of the indefinite form at the same time. That is what a symmetry is, and the rotation is the only kind of motion of the algebra of which it is true.

The Double Cover

At $\theta=2\pi$ the rotor is $\cos\pi\,e_t+\sin\pi\,\vec u=-e_t$, and the rotation is the identity, since conjugation by the centre is trivial. At $\theta=4\pi$ the rotor is $+e_t$. The map $$ Sp(1)\longrightarrow SO(3),\qquad \tilde u\longmapsto \bigl(\tilde X\mapsto\tilde u\tilde X\tilde u^{-1}\bigr), $$ is therefore two-to-one, with kernel $\{\pm e_t\}$, and it is the double cover $SU(2)\to SO(3)$.

The minus sign at $2\pi$ is arithmetic, not an extra rule. The rotor is a square root of the rotation: a half turn of the object is a quarter turn of the rotor, and a full turn of the object is a half turn of the rotor, which is minus the identity. The rotor returns to $+e_t$ only after two full turns of the object. The archived spin articles read the fermion sign this way, and the algebra's contribution is that the sign is forced by the parametrisation of a square root.

A frame is invertible up to sign, and a spinor is not. The congruence of $\tilde U$ and the congruence of $-\tilde U$ are the same map, exactly, since the two signs cancel between the left and the right factor; verified to the last bit on random units. So a frame and its negative are the same frame. A spinor is the element itself, and its negative is a different spinor: the algebra carries the frame as a class and the spinor as an element, and the double cover is the dictionary between the two. Interpreted physically, a frame is what a change of frame acts on and a spinor is what a representation acts on, and the algebra has both because its action has two kinds of object.

The sign acts by the parity of the weight. The element $-e_t$ is central, so its congruence is the identity, while its left multiplication is minus the identity. A rotation by $2\pi$ is therefore the identity on a carrier of integer weight and minus the identity on a carrier of half-integer weight. Verified: at $\theta=2\pi$ the rotor fixes every vector of the vector part to $1.1\times10^{-16}$, and left multiplication by the same element sends a spinor to its negative exactly. The two periods of $4\pi$ and $2\pi$ are the two columns of that comparison, and the carrier decides which one applies.

Complexifying, the same construction with a unit of $\mathbb{B}^{\times}_1$ gives the double cover of the Lorentz group, $$ \mathbb{B}^{\times}_1\longrightarrow SO^{+}(1,3),\qquad \tilde U\longmapsto \mathrm{Cg}_{\tilde U}\big|_{\mathbb{B}_{\Psi}}, $$ with kernel $\{\pm e_t\}$. The parameter count is why the complexification is needed: a Lorentz transformation has six parameters, the quaternion layer has real dimension four, so the group manifold does not fit in the algebra and the action on a four-dimensional space does.

The Rotation in Both Families

The rotation is the element of the algebra on which the two actions of the entry agree, and the agreement can be read factor by factor. With $\tilde u$ real and of norm one, $\tilde u^{*}=\tilde u^{\natural}=\tilde u^{-1}$, so the inverse and the dagger are the same element. The adjoint and the congruence are then literally the same formula, and every property of one is a property of the other.

The consequences are worth listing, because they are the content of "a rotation is a symmetry".

  • The product is preserved, so every algebraic relation written with the multiplication is preserved.
  • The two Hermitian sectors are preserved, so the split into an informational and a material sector is preserved.
  • The norm is preserved, and the indefinite form is preserved.
  • The centre is fixed pointwise, as it is by every automorphism.

A rotation therefore does nothing that the algebra can detect except move the vector directions. It is the identity on the centre, the identity on the scalar of every element, and a rotation on the three vector directions. In the language of the carriers, it acts on the spin-one carrier and on the two spin-half carriers, and trivially on the spin-zero one.

The rotation is complex-linear on the vector part. The vector part is $\mathbb{C}\{\vec e_x,\vec e_y,\vec e_z\}$, six real dimensions, and the central imaginary commutes with the rotor, so the rotor carries $i\tilde X$ to $i$ times the rotor of $\tilde X$; verified to $2.2\times10^{-16}$. The six real dimensions therefore carry one complex rotation and not two independent ones, and the imaginary half of the vector part is not a second space.

The Two-Sided Action and $SO(4)$

Two independent unit quaternions act by $$ \tilde X\longmapsto \tilde u_1\tilde X\tilde u_2, $$ and the pair preserves the Euclidean form of the algebra, since both factors preserve it. The kernel is $\{\pm(e_t,e_t)\}$, so the image is $$ (Sp(1)\times Sp(1))/\{\pm e_t\}\cong SO(4). $$ Verified: on one hundred random pairs and random elements the Euclidean norm is preserved to $8.9\times10^{-16}$.

The one-sided action is the diagonal restriction $\tilde u_1=\tilde u_2^{-1}$, and it is the rotation family of the previous sections. The two-sided action with independent factors is a four-dimensional rotation, and it is not a Lorentz transformation: it preserves a definite form, not an indefinite one. The distinction matters for the reading, because a frame change in the algebra is a Lorentz transformation and a frame change in the quaternion layer is a four-dimensional rotation, and the two are different objects that share the rotation subgroup.

The Boost: A Motion and Not an Automorphism

The Boost Element

A boost is generated by the hyperbolic analogue of the rotor. For a real unit vector $\vec u$ and a rapidity $\theta$,

$$ \tilde U(\theta)=\cosh\frac{\theta}{2}\,e_t+i\sinh\frac{\theta}{2}\,\vec u, \qquad N\bigl(\tilde U(\theta)\bigr)=\cosh^{2}\frac{\theta}{2}-\sinh^{2}\frac{\theta}{2}=1 . $$

The element is Hermitian, $\tilde U^{*}=\tilde U$, and its inverse is its quaternion conjugate, $\tilde U^{-1}=\tilde U^{\natural}=\cosh\tfrac{\theta}{2}-i\sinh\tfrac{\theta}{2}\vec u$. The dagger and the inverse are therefore different elements, and this is the whole of the difference between the two actions at a boost.

The element is not $*$-unitary: $\tilde U^{*}\tilde U=\tilde U^{2}=\cosh\theta+i\sinh\theta\,\vec u$, which is not the identity. Verified at rapidity two: the dagger defect is $4.559$. The element is of norm one, so it lies in the norm-one group and in neither the compact group nor the coincidence.

The failure of multiplicativity is first order and the time dilation is second order. The defect of $\mathrm{Cg}_{\tilde U}$ on a product grows like $\theta$, while the dilation $\cosh\theta-1$ grows like $\tfrac{\theta^{2}}{2}$. Verified on two elements of the informational sector at rapidity $0.02$: the defect is $2.83\times10^{-2}$ and $\cosh\theta-1$ is $2.0\times10^{-4}$, a ratio of $141$. So the algebra's multiplication has already failed by two parts in a hundred before the clock has moved by a part in ten thousand, and the non-multiplicativity of a boost is a first-order effect of the motion and not a relativistic correction to it.

What the Adjoint Cannot See

Every automorphism fixes the centre pointwise, because the centre is the set of elements that commute with everything and an automorphism preserves commutation. The adjoint of a boost is therefore the identity on $\mathbb{C}e_t$. Verified: the adjoint of the boost of rapidity two fixes $e_t$ to $5.0\times10^{-16}$.

This is not a small point. The time axis of the algebra is the centre, and the adjoint action of a boost does nothing to it. If a boost were read only through the adjoint, it would be a transformation that leaves the time direction alone and moves the vector directions among themselves, which is not a boost at all. What the adjoint does instead is mix the sectors.

The mixing is exact and simple in an example. Take the boost of rapidity two about $\vec e_z$ and act on $i\vec e_y$, an Hermitian element of the informational sector. The result is $$ \mathrm{Ad}_{\tilde U}(i\vec e_y)=\sinh 2\;\vec e_x=3.6269\,\vec e_x, $$ a purely anti-Hermitian element of the material sector. Verified: the Hermitian defect of the image is $7.25$ and its anti-Hermitian part has norm $3.7622$. The adjoint of a boost carries an amplitude direction into a spatial direction, and the algebra says so without any convention being chosen.

So the adjoint of a boost is an automorphism, and it is not the boost. It is the boost read in the wrong family.

What the Congruence Does

The congruence of the same element does the opposite. It preserves the sectors and it moves the time axis: $$ \mathrm{Cg}_{\tilde U}(e_t)=\tilde U\tilde U^{*}=\tilde U^{2}=\cosh\theta\,e_t+\sinh\theta\,i\vec u . $$ Verified at rapidity two: the image is $3.762196\,e_t$ plus the vector term, and $\cosh 2=3.762196$.

The time entry of the image of $e_t$ is the Euclidean norm of the element, $\lVert\tilde U\rVert^{2}$, which is at least $\lvert N(\tilde U)\rvert=1$ by Cauchy–Schwarz. The congruence therefore never reverses the time direction, and this is the computation behind the proper-orthochronous theorem. Verified on one hundred norm-one units: the time entry is at least $1$, the determinant is $1$ to $4\times10^{-14}$, and the interval form is preserved to $2.2\times10^{-16}$.

The congruence also fixes the two null lines of the Hermitian sector. The elements $e_t\pm i\vec u$ are null, $N(e_t\pm i\vec u)=1-1=0$, and the boost scales them by $e^{\pm\theta}$: $$ \mathrm{Cg}_{\tilde U}(e_t\pm i\vec u)=e^{\pm\theta}(e_t\pm i\vec u). $$ Verified at rapidity two: both ratios are $3.762196=e^{2}$, one up and one down. The pair of null lines is fixed, the hyperboloid sheet between them is translated, and nothing leaves the sector. That is a Lorentz boost written in the algebra's own coordinates, with no matrix and no preferred basis.

The Rapidity and the Hyperbolic Angle

The parameter $\theta$ is the rapidity, and it is additive for collinear boosts: $\tilde U(\theta_1)\tilde U(\theta_2)=\tilde U(\theta_1+\theta_2)$ when the two axes agree, because the addition formula for the hyperbolic functions is then the multiplication of two commuting elements. The velocity is $\tanh\theta$, so the rapidity of a composition is the sum of the rapidities, and the velocity composition law is the hyperbolic tangent of that sum.

The two parameters, the angle of the rotation and the rapidity of the boost, are the two real coordinates of the same complex parameter. A rotation has $\theta$ real and the trigonometric functions; a boost has $\theta$ real and the hyperbolic functions; a general motion of the Lorentz group has a complex combination. The complexification of the algebra is what makes one parameter cover both, and it is the reason the compact and the hyperbolic families sit in the same group rather than in two groups glued along the identity.

The null lines of the previous section are the fixed points of the boost in the projective Hermitian sector, and the rapidity is the logarithm of the scale factor on them. That is the cleanest reading of the parameter: the boost is a dilation of the two null directions, one up and one down, by reciprocal factors.

Two Boosts Are Not a Boost

The set of boost elements is closed under the inverse and not under multiplication. The dagger is an anti-automorphism, so $\tilde U_1^{*}=\tilde U_1$ for each factor gives $$ (\tilde U_1\tilde U_2)^{*}=\tilde U_2^{*}\tilde U_1^{*}=\tilde U_2\tilde U_1 , $$ which differs from $\tilde U_1\tilde U_2$ whenever the two axes are not parallel, because two such elements do not commute. The product is then not Hermitian, so it fails the criterion of §The Boost Element. The inverse is not the exception: $\tilde U(\theta,\vec u)^{-1}=\tilde U(-\theta,\vec u)=\tilde U(\theta,-\vec u)$ is again a boost, and an antiparallel pair commutes and composes to $\tilde U(\theta_1-\theta_2,\vec u)$. The failure is of closure under multiplication alone, so the boosts are a subset of the norm-one group and not a subgroup.

The product is a boost and a rotation. The polar decomposition of §The Two Retractions writes it as $$ \tilde U_1\tilde U_2=\tilde U_{\mathrm{rot}}\,\tilde U_{\mathrm b}, $$ with $\tilde U_{\mathrm b}$ Hermitian of norm one, hence a boost, and $\tilde U_{\mathrm{rot}}$ $*$-unitary, acting on the Hermitian sector by congruence as a rotation — the central phase inside it acts trivially there. Since $\mathrm{Cg}$ is a homomorphism, $\mathrm{Cg}_{\tilde U_1}\mathrm{Cg}_{\tilde U_2}=\mathrm{Cg}_{\tilde U_{\mathrm{rot}}}\mathrm{Cg}_{\tilde U_{\mathrm b}}$: the boost is performed first, the rotation second.

Verified at rapidities $1.1$ and $0.9$ about $\vec e_x$ and $\vec e_y$: the dagger defect of the product is $2.5793$, against $4.5589$ for the single boost of rapidity $2$; the Hermitian factor is the boost of rapidity $1.518$ about $0.615\vec e_x+0.789\vec e_y$; and the compact factor is $0.978423-0.206613\vec e_z$, a rotation through $0.4162$ rad.

The rotation is the Wigner rotation. The angle is the one a single boost to the resultant velocity cannot account for, and it vanishes only for parallel and antiparallel axes. Interpreted physically, the composition of two changes of frame is not a third change of frame, and the excess is a rotation of the axes rather than an acceleration of the observer.

The boundary of the reading. The angle is a property of the composition of two motions, not a derived dynamical effect. The corpus does not integrate it along a world line, so the accumulated angle of an accelerated observer — the precession that carries the name — is not computed here. The algebra supplies the factor, not the history.

Why the Difference Is Not a Convention

Both actions are canonical. The adjoint comes from the product and nothing else; the congruence comes from the dagger and hence from the form. Neither is a choice of coordinates, and neither can be derived from the other. The algebra has two structures, a multiplication and an involution, and each structure has its own group of symmetries.

A rotation is the case where the two symmetries coincide. A boost is the case where they do not: it is a symmetry of the involution and not of the multiplication. That is the exact sense of the sentence carried by the archived Frames, Rotations and the Lorentz Group, that a rotation is an algebra automorphism and a boost is a motion which is not one.

The sentence needs its quantifier to be true. The congruence of a boost is not an automorphism; the adjoint of a boost is an automorphism but is not the motion. There is no single map of the algebra that is both. The physics takes the congruence for the four-vector and the left multiplication for the spinor, and the algebra's structure explains why those are two different representations and not two conventions for one.

The Automorphisms and the Derivations

Inner and Outer

Every automorphism of $\mathbb{B}$ over $\mathbb{C}$ is inner, by Skolem–Noether, because the algebra is central simple. The inner automorphisms are the adjoints, and two units give the same adjoint exactly when they differ by a central scalar, so

$$ \mathrm{Aut}_{\mathbb{C}}(\mathbb{B})=\mathrm{Inn}(\mathbb{B})=\mathbb{B}^{\times}/\mathbb{C}^{\times}\cong PGL_2(\mathbb{C})\cong SO^{+}(1,3), $$

of complex dimension three and real dimension six. The identification of the automorphism group with the Lorentz group is not an accident of dimension: conjugation by an invertible element is the Lorentz action on the four-dimensional Hermitian sector, and the kernel of that action is the central scalar, so the two groups are the same group read on two spaces.

Over $\mathbb{R}$ the group gains a component. An $\mathbb{R}$-linear automorphism need not be $\mathbb{C}$-linear, but it must carry the centre to the centre, because the centre is characterised by commutation; so it restricts to an automorphism of the centre $\mathbb{C}$, and $\mathbb{C}$ as a real algebra has exactly two automorphisms, the identity and complex conjugation. The kernel of the restriction is the $\mathbb{C}$-linear part, which is $\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})$, and the sequence splits because conjugation is an involution:

$$ \mathrm{Aut}_{\mathbb{R}}(\mathbb{B})\cong\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})\rtimes\mathbb{Z}/2 . $$

Concretely every real automorphism is

$$ \tilde X\longmapsto \tilde g\tilde X\tilde g^{-1} \qquad\text{or}\qquad \tilde X\longmapsto \tilde g\,\bar{\tilde X}\,\tilde g^{-1}, $$

with $\tilde g$ an invertible element determined up to a nonzero complex scalar, and the second family is the coset of complex conjugation.

The Conjugate-Linear Coset

Complex conjugation $c(\tilde X)=\bar{\tilde X}$ is an $\mathbb{R}$-algebra automorphism: it conjugates the four coefficients and leaves the quaternion units alone, so $c(\tilde X\tilde Y)=c(\tilde X)c(\tilde Y)$. Verified to $0.0$ on one hundred random pairs.

It is not inner. Every inner automorphism fixes the centre pointwise, and $c(i)=-i$, so the centre is moved. Verified: the adjoint of a random unit fixes $e_t$ to $9.2\times10^{-16}$, and $c$ sends $i e_t$ to $-i e_t$.

The coset is therefore a genuine outer part of the real automorphism group, of real dimension six like the identity component, and the full group has exactly two connected components. The dimension of the real group equals the dimension of the complex one, and the extra component doubles the group without enlarging its tangent space.

The physical reading of the coset is that a symmetry which conjugates the central imaginary is an antiunitary one. In the algebra the conjugation is the only automorphism that moves the centre, and a motion of the algebra cannot move it. The archived CPT Theorem in Biquaternionic Form and Kramers Degeneracy and Antiunitary Symmetry carry the reading; the entry records only the placement, that the discrete symmetries which reverse a phase live in the coset and not in the group.

The Derivations and the Bivectors

A derivation is a linear map with the Leibniz rule, $D(\tilde X\tilde Y)=D(\tilde X)\tilde Y+\tilde X D(\tilde Y)$, and the inner derivations are the commutators $\mathrm{ad}_{\tilde Q}(\tilde X)=[\tilde Q,\tilde X]$. For a central simple algebra every derivation is inner, so

$$ \mathrm{Der}_{\mathbb{C}}(\mathbb{B})=\mathrm{ad}(\mathbb{B})\cong\mathbb{B}/\mathbb{B}_{C}, $$

the quotient by the centre, which is the scalar-free part $\mathrm{span}_{\mathbb{C}}\{\vec e_x,\vec e_y,\vec e_z\}$, of complex dimension three and real dimension six. Verified: the Leibniz rule holds on one hundred random triples, $\mathrm{ad}_{\tilde Q}$ vanishes on $e_t$ identically, and $\mathrm{ad}_{\tilde Q}=0$ exactly for central $\tilde Q$.

Over $\mathbb{R}$ the derivation space does not grow, and the reason is one line. A real derivation must carry the centre into itself, and $\mathbb{C}$ has no nonzero real derivation, since $0=D(-1)=D(i^{2})=2iD(i)$ forces $D(i)=0$. So $D$ vanishes on the centre and, the centre being central, $D(i\tilde X)=iD(\tilde X)$: every real derivation is already $\mathbb{C}$-linear, and $\mathrm{Der}_{\mathbb{R}}(\mathbb{B})=\mathrm{Der}_{\mathbb{C}}(\mathbb{B})$.

The asymmetry with the automorphisms is the ground-field fact of the section. Complex conjugation supplies a second coset of automorphisms but no derivation, because a derivation cannot conjugate a coefficient. The automorphism group doubles over $\mathbb{R}$ and the derivation algebra does not.

A basis is given by $D_k=\tfrac12\mathrm{ad}_{\vec e_k}$, with $$ D_1(\vec e_y)=\vec e_z,\quad D_1(\vec e_z)=-\vec e_y, \qquad [D_1,D_2]=D_3, $$ and the cyclic permutations. Verified: $D_1(\vec e_y)=\vec e_z$, $D_1(\vec e_z)=-\vec e_y$, $D_1(\vec e_x)=0$, and $[D_1,D_2](\vec e_x)=D_3(\vec e_x)=\vec e_y$ to $10^{-12}$.

The real span of the six elements $\vec e_x,\vec e_y,\vec e_z,i\vec e_x,i\vec e_y,i\vec e_z$ is the bivector subspace of the Clifford envelope, and its bracket is the bracket of $\mathrm{so}(1,3)$. The three derivations $D_k$ are the rotation generators and the three $iD_k$ are the boost generators; the derivation algebra is the Lie algebra of the Lorentz group, and the three descriptions — traceless part, complex pure-vector part, bivector part — are one space.

The Exponential from Derivation to Automorphism

The link between the two structures is the exponential, $$ \exp\bigl(t\,\mathrm{ad}_{\tilde Q}\bigr)(\tilde X)=\exp(t\tilde Q)\,\tilde X\,\exp(-t\tilde Q)=\mathrm{Ad}_{\exp(t\tilde Q)}(\tilde X), $$ so the Lie algebra of the automorphism group is the derivation algebra and the exponential of an inner derivation is an inner automorphism. The identity is the reason a one-parameter group of automorphisms is the flow of a derivation, and it is the entry's point of contact with the exponential article.

The generators are inner, so every one-parameter group of automorphisms of the algebra is generated by an element of the algebra. There is no outer one-parameter group, only the discrete outer coset of the previous section. A continuous symmetry of the algebra is a flow; a discrete symmetry is a coset; and the two kinds are separated by the centre.

The Versors and the Reach

The Envelope and the Two Slots

The algebra is the even part of the Clifford algebra of a four-dimensional space of inertia $(1,3)$:

$$ \mathrm{Cl}(V)=\mathbb{B}\oplus\mathbb{B}\zeta,\qquad \mathrm{Cl}^{+}=\mathbb{B}, $$

for any non-isotropic $\zeta\in V$, with the odd slot a second copy of the algebra of real dimension eight. The Lorentzian space $V$ on which the orthogonal group acts is odd: no element of the algebra is a Lorentzian vector. The algebra does contain a second copy of the same quadratic space, the Hermitian sector $\mathbb{B}_{\Psi}$, and the dictionary between the two copies is what lets a statement about one be read as a statement about the other.

This is the structural reason the reach of the algebra is limited. Every operator the algebra supplies is built from one element on each side, so it is an even element of the envelope, and an even element composes an even number of reflections. The isometries it realises on its own Hermitian copy by the congruence are the proper orthochronous ones and no others; on the Lorentzian space of the envelope the same elements act by the inner automorphism and reach one component more, which is the subject of §The Dagger and the Inverse Give Different Reach.

The Parity of the Length

A versor is a product of non-isotropic vectors, and its length is its number of factors. The reflection in the hyperplane orthogonal to a vector $\zeta$ is $\rho_\zeta(\eta)=-\zeta\eta\zeta^{-1}$, of determinant $-1$; a product of $k$ reflections has determinant $(-1)^{k}$; and Cartan–Dieudonné writes every isometry of the space as such a product. Hence the parity of the length of a versor is the parity of the determinant of the isometry it induces.

The products of even length lie in the algebra and the products of odd length lie in the odd slot. Since a non-isotropic vector can be rescaled to a vector of square $\pm1$, such a vector is an element of $\mathrm{Pin}$ and is odd, so it is not an element of $\mathbb{B}$. Therefore

$$ \mathrm{Pin}(1,3)\not\subseteq\mathbb{B}, \qquad \mathrm{Pin}(1,3)\cap\mathrm{Cl}^{+}=\mathrm{Spin}(1,3)\cong\mathbb{B}^{\times}_1 . $$

The only part of the Pin group that is an object of the algebra is the spin group, which is its even part. A single reflection cannot be written as an element of the algebra acting two-sidedly, and no amount of rescaling repairs it: the obstruction is the grading, not the normalisation.

The Reflections of the Vector Subspace

On the real vector subspace $\mathbb{R}\{\vec e_x,\vec e_y,\vec e_z\}$ the multiplication is Clifford, $v\tilde X+\tilde Xv=-2\langle v,\tilde X\rangle e_t$ for a real $v$, so orthogonality there is anticommutation and a mirror is available. The mirror in the hyperplane $v^{\perp}$ is

$$ \rho_v(\tilde X)=-v\,\tilde X\,v^{\natural}, \qquad \langle v\rangle_{\natural}=1, $$

with the quaternion conjugate on the right, and the leading minus is essential.

On the real form the dagger is minus the element. The dagger negates the three vector units, so for real $v$ one has $v^{*}=-v$ and not $v^{*}=v$. Three consequences, each verified to $10^{-16}$ on a random unit $v$:

  • The two-sided operator alone, $\mathrm{Cg}_v(\tilde X)=v\tilde Xv^{*}=v\tilde X(-v)$, is the half turn about $v$: it fixes $v$, negates every $w\perp v$, and has determinant $+1$ on $\mathbb{R}\{\vec e_x,\vec e_y,\vec e_z\}$. It is the same map as $\mathrm{Ad}_v$, since $v^{*}=v^{-1}$ on the norm-one slice.
  • The mirror is that half turn composed with the negation of the whole algebra, $\rho_v=-\mathrm{H}_v=-\mathrm{Cg}_v$. It negates $v$, fixes $v^{\perp}$ pointwise, has determinant $-1$ on the vector subspace and $+1$ on the eight-dimensional algebra, and its square is the identity.
  • So the reflection is minus a congruence and not a congruence, and its determinant on the vector subspace is the minus sign. A formula that writes $v\tilde Xv^{*}$ for the mirror writes a rotation by $\pi$; the minus of the source's prose is right and the sign belongs in the display.

Cartan–Dieudonné then gives the whole orthogonal group of the vector subspace. Every isometry of $\mathbb{R}\{\vec e_x,\vec e_y,\vec e_z\}$ is a product of at most three mirrors $\rho_v$, so $\mathrm{O}(3)$ is generated by them and a reflection is an odd word. The mirrors are not motions of the algebra: a $\rho_v$ is minus a congruence of an invertible element, and no continuous path of invertibles supplies the minus, because the congruence of a real unit quaternion has determinant $+1$ on the vector subspace. Interpreted physically, a mirror operation is not a rotation, and the algebra says so by putting the sign outside the two-sided operator.

The Four Components and the Two Discrete Operations

On the Hermitian sector the congruence of a modulus-one unit is proper and orthochronous. The other three components are reached by composing it with two operations that are not of that form:

operation on $\mathbb{B}_{\Psi}$ the determinant of the image time orientation component
$\tilde X\mapsto\tilde U\tilde X\tilde U^{*}$, $\lvert N(\tilde U)\rvert=1$ $+1$ $+$ the identity component $SO^{+}(1,3)$
$\tilde X\mapsto\tilde U\tilde X^{\natural}\tilde U^{*}$, the argument conjugated $-1$ $+$ the improper orthochronous coset
$\tilde X\mapsto-\tilde U\tilde X\tilde U^{*}$, the result negated $+1$ $-$ the proper anti-orthochronous coset
$\tilde X\mapsto-\tilde U\tilde X^{\natural}\tilde U^{*}$, both $-1$ $-$ the improper anti-orthochronous coset

Verified on the $4\times4$ real matrices of the four operations: the determinants are $+1,-1,+1,-1$ and the time entries are $+1.337,+1.337,-1.337,-1.337$, at a rapidity of $0.8$.

The two extra operations are quaternion conjugation of the argument and negation of the result. Quaternion conjugation fixes $e_t$ and negates the three vector units, so on the Hermitian sector it is the parity $\mathrm{diag}(1,-1,-1,-1)$; negation is the product of parity with time reversal. Neither is a congruence of an invertible element, and neither is a motion. The four signs are distinct, and the table is complete because $O(1,3)$ has exactly four components.

The placement is the content. A parity is an operation on the argument, not a change of frame. A time reversal is the same operation composed with a negation. Neither can be reached by moving in the group of invertibles, because the group moves continuously and the components are open and closed.

The Volume Element

The volume element of the envelope is the product of an orthonormal frame, $\Omega=\eta_t\eta_x\eta_y\eta_z$, with $\Omega^{2}=-1$; it commutes with the even slot and anticommutes with every vector. In the biquaternion dictionary the central scalar imaginary is $-\Omega$ up to the labelling.

On the Lorentzian space the signed conjugation of the volume element negates every vector, $$ \mathrm{Ad}^{\alpha}_{\Omega}(\eta)=\Omega\eta\Omega^{-1}=-\eta, $$ because $\eta_j\Omega=-\Omega\eta_j$ for every vector of the frame. That is $-\mathrm{id}$ on $V$, of determinant $+1$ and anti-orthochronous, and it is the product of parity with time reversal.

On the Hermitian copy the corresponding element is the central imaginary, and its two-sided operator is the identity, since $i$ is central and $i^{*}=-i$. The same element, read through the two actions, is a reflection on one copy and the identity on the other. That is the sharpest single example of the entry's thesis, and it is the reason the reach of the two actions differs.

The Dagger and the Inverse Give Different Reach

The two actions reach different parts of the orthogonal group, and the difference is the replacement of the dagger by the inverse.

Through the congruence, the algebra's units reach exactly the proper orthochronous component on the Hermitian copy. Verified on one hundred norm-one units: determinant $+1$ and time entry positive, always, with no exception.

Through the adjoint on the Lorentzian space of the envelope, the even elements reach the whole of $SO(1,3)$, the anti-orthochronous half included, because the volume element is an even element and its signed conjugation is $-\mathrm{id}$ there. The inner automorphism of the algebra is what the signed conjugation restricts to on the even part, and it is the inverse that appears in it.

So the algebra's elements supply more of the orthogonal group than the two-sided operator does, and the surplus is exactly the anti-orthochronous half. A statement about "what the algebra can do" is therefore ambiguous until the action is named: the same element does one thing through the inverse and another through the dagger, and the two things differ by a component of the orthogonal group.

The Unitary Group: The Gauge and Its Double Cover

The Compact Slice

The $*$-unitary biquaternions are the elements with $\tilde U^{*}\tilde U=e_t$. Under the matrix model they are the unitary matrices, so

$$ \mathrm{U}(\mathbb{B})=\{\tilde U:\tilde U^{*}\tilde U=e_t\}\cong \mathrm{U}(2), $$

of real dimension four. Every element has a normal form $$ \tilde U=A\tilde q,\qquad A\in\mathbb{C},\ \lvert A\rvert=1,\quad \tilde q\in\mathbb{H},\ N(\tilde q)=1, $$ that is, a phase times a unit quaternion. Verified: on one hundred random unitary matrices pulled back to the algebra, the dagger defect is at most $1.1\times10^{-15}$ and, after the phase is removed, the imaginary part of the vector coefficients is at most $2.6\times10^{-14}$.

The group splits as $S^{1}\cdot S^{3}$ with intersection $\{\pm e_t\}$, so the multiplication map $S^{1}\times S^{3}\to\mathrm{U}(\mathbb{B})$ is two-to-one and $$ \mathrm{U}(\mathbb{B})\cong(S^{1}\times S^{3})/\{\pm e_t\}, $$ homeomorphic to $S^{1}\times S^{3}$ but not isomorphic to it as a group. The centre of the compact group is the connected circle, while the centre of $S^{1}\times S^{3}$ is not connected; the two-to-one covering is where the difference lives.

A maximal torus is $T^{2}$, the product of the central circle and a maximal circle of the three-sphere, and the quotient is the complete flag variety, $$ \mathrm{U}(\mathbb{B})/T^{2}\cong P^{1}\cong S^{2}, $$ so the compact group is a bundle over the two-sphere with a two-torus as fibre.

The fibre is the stabiliser of a pure state, and it is the two-torus. The Hermitian idempotent $\tfrac12(e_t+i\vec u)$ is fixed by conjugation exactly by the pair of diagonal phases, so the stability group of a pure state is $T^{2}$ of real dimension two, and the central circle alone is too small by a circle. Verified: the diagonal two-torus fixes the idempotent of a real unit vector to $5.6\times10^{-17}$, and a generic unitary moves it by $0.575$. The sphere and the ball of §The Level Sets are then the same quotient read on two sectors, and the flag variety is the set of measurement bases because a basis of $\mathbb{C}^{2}$ is a pair of orthogonal pure states, each carrying its own phase.

In the matrix model the norm is the determinant. Under $\mathbb{B}\cong M_2(\mathbb{C})$ an element is a $2\times2$ matrix and $N(\tilde Q)=\det$, verified symbolically and numerically: the determinant of the image $\mathsf{M}_2(\tilde Q)$ of $Q_te_t+Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z$, with $\mathsf{M}_2(\vec e_\eta)=-i\sigma_\eta$, is $Q_t^{2}+Q_x^{2}+Q_y^{2}+Q_z^{2}$. The compact slice is therefore the unitary matrices with the norm a phase, and that is why the quotient by $S^{3}$ is the norm's circle rather than something larger.

A point of the flag variety is a measurement basis. A complete flag of $\mathbb{C}^{2}$ is a line inside a plane; the plane is the whole space, so the flag is one line, and the line together with its orthogonal complement is a pair of orthogonal pure states. The two-torus is the pair of their phases, so the quotient is the pair itself, and the sphere carries two readings: the pure states of Idempotents, Ideals and Observables, where they are the Hermitian idempotents $\tfrac12(e_t+i\vec u)$ over the real unit vectors, and the bases in which a complete measurement is stated. The two coincide because a pure state is a line and a pair of outcomes is a pair of lines, and the space of each is $P^{1}$ in two dimensions.

The Two Retractions

The dagger supplies a polar decomposition on the group of invertibles. Every unit is $$ \tilde X=\tilde U\tilde H,\qquad \tilde U\in\mathrm{U}(\mathbb{B}),\quad \tilde H\ \text{Hermitian and positive definite}, $$ uniquely, with $\tilde H=(\tilde X^{*}\tilde X)^{1/2}$. The map is a homeomorphism of the group of invertibles onto the product of the compact group and the positive cone, and it retracts the group of invertibles onto the compact group by collapsing the second factor.

On the norm-one group the same decomposition does more. If $N(\tilde X)=1$ then both factors have norm of modulus one, the compact factor is a unit quaternion and the Hermitian factor has norm one; interpolating the Hermitian factor to the identity by $$ \tilde H_t=\frac{(1-t)\tilde H+t e_t}{N\bigl((1-t)\tilde H+t e_t\bigr)^{1/2}} $$ gives a strong deformation retraction of $\mathbb{B}^{\times}_1$ onto $S^{3}$.

Two retractions, one onto each of the two middle groups of §The Chain of Groups. The group of invertibles retracts onto the compact slice and the norm-one slice retracts onto the rotations. The two retractions are the reason the topology of the whole group is the topology of the compact slice.

The Homotopy and the Cover

The consequences are collected in one box: $$ \mathbb{B}^{\times}\simeq\mathrm{U}(\mathbb{B})\simeq S^{1}\times S^{3}, \qquad \mathbb{B}^{\times}_1\simeq S^{3}, $$ $$ \pi_1(\mathbb{B}^{\times})\cong\mathbb{Z},\quad \pi_2(\mathbb{B}^{\times})=0,\quad \pi_3(\mathbb{B}^{\times})\cong\mathbb{Z}, \qquad \pi_1(\mathbb{B}^{\times}_1)=0,\quad \pi_2(\mathbb{B}^{\times}_1)=0,\quad \pi_3(\mathbb{B}^{\times}_1)\cong\mathbb{Z}, $$ with universal covers $(\mathbb{B}^{\times})_{\mathrm{cov}}\cong\mathbb{R}\times S^{3}$ and $S^{3}$ its own cover. The Hurewicz theorem then gives $H_1(\mathbb{B}^{\times})\cong\mathbb{Z}$ and $H_1(\mathbb{B}^{\times}_1)=0$.

The generator of the circle is the phase loop, and the norm sees it twice. The loop $\gamma(t)=\exp(2\pi i t)e_t$ has norm $\exp(4\pi i t)$, so the induced map on the fundamental group is multiplication by two, injective with image of index two. A loop in the compact group cannot wind the norm by an odd number. The double cover of the rotation group appears here as a degree, and the degree is a property of the norm, not of the parametrisation.

The Quotients and the Two Covers

Three quotients of the compact group are used in the corpus: $$ \mathrm{U}(\mathbb{B})/S^{3}\cong S^{1}, \qquad \mathrm{U}(\mathbb{B})/U(1)\cong PU(2)\cong SO(3), \qquad \mathrm{U}(\mathbb{B})/T^{2}\cong S^{2}. $$

The first is the norm: the compact group modulo the rotations is the phase, and the quotient map is the norm with kernel the unit quaternions. The second is the rotation: the compact group modulo the phase acts on the Hermitian sector by rotations, and the kernel of that action is the central circle, so the quotient is the adjoint form of $\mathrm{U}(2)$ and is $SO(3)$. The third is the flag variety.

The two covers of the corpus sit on either side of the compact group. Below it, $S^{3}\to SO(3)$ is the double cover of the rotations. Above it, $\mathbb{B}^{\times}_1\to SO^{+}(1,3)$ is the double cover of the Lorentz group. The compact group is the intersection of the two covering groups, and the two covers are the compact and the hyperbolic halves of one complex group.

The Krein Group: The Indefinite Limit

The Group and the Criterion

The fourth form of the algebra is Hermitian and indefinite, and its isometry group is the Krein group. For a $\mathbb{C}$-linear operator $T$ on the algebra, $$ U_{J}(\mathbb{B})=\{T:T^{\dagger}T=\mathrm{id}\},\qquad T^{\dagger}=JT^{*}J, $$ and in the coefficient basis the condition is $T^{*}ET=E$ with $E=\operatorname{diag}(1,-1,-1,-1)$, so

$$ U_{J}(\mathbb{B})\cong \mathrm{U}(1,3), $$

of real dimension $16=1^{2}+3^{2}+2\cdot1\cdot3$, non-compact, with centre the scalars of modulus one and maximal compact subgroup $\mathrm{U}(1)\times\mathrm{U}(3)$ of dimension $10$. Three equivalent forms of the criterion are used: $T^{*}JT=J$, $T^{\dagger}T=\mathrm{id}$, and $T^{-1}=T^{\dagger}$. In finite dimension the isometry monoid is the group, so the three conditions cut the same set.

The group is not the group of invertibles of the algebra and is not contained in it. It is a group of operators on the algebra, of dimension sixteen, against the eight of the group of invertibles.

The Algebra's Seven of Sixteen

Which operators of the group are supplied by the algebra's elements? The mathematics answers by families.

Left and right multiplication are isometries exactly for the central phases: $L_{\tilde Q}$ and $R_{\tilde Q}$ are $J$-unitary if and only if $\tilde Q\in S^{1}e_t$. The two-sided operator is an isometry exactly on the modulus-one slice: $\mathrm{Cg}_{\tilde Q}$ is $J$-unitary if and only if $\lvert N(\tilde Q)\rvert=1$. Verified on forty norm-one units: the congruence is a Krein isometry to $2.1\times10^{-14}$, and the adjoint of the same units fails by $168$.

So the algebra's contribution is $$ \mathrm{U}(1)\cdot PSL_2(\mathbb{C})\cong \mathrm{U}(1)\times SO^{+}(1,3), $$ of dimension $1+6=7$, inside a group of dimension $16$. The compact part of the contribution is $\mathrm{U}(1)\times PU(2)\cong\mathrm{U}(1)\times SO(3)$, of dimension four: the phases and the rotations. The non-compact part is the boosts, of dimension three more.

Nine dimensions of the group are not reachable by any element of the algebra. Those nine are isometries of the fourth form that no product, no conjugate and no scalar of the algebra can produce. The gap is a fact about the algebra, not about the form.

The Boost Inside the Group

The group contains one-parameter subgroups with real spectrum off the unit circle, and they are the boosts. On the Krein-orthogonal sum of a positive and a negative line, $$ T_te_t=\cosh t\,e_t+\sinh t\,\vec e_x, \qquad T_t\vec e_x=\sinh t\,e_t+\cosh t\,\vec e_x, $$ with the identity on the other two directions, and $$ \operatorname{spec}(T_t)=\{e^{t},e^{-t},1,1\}. $$ Verified at $t=0.3,1.0,2.5$: the spectra are $\{1.349859,0.740818,1,1\}$, $\{2.718282,0.367879,1,1\}$ and $\{12.182494,0.082085,1,1\}$, and the isometry condition $T^{T}ET=E$ holds to $7.3\times10^{-15}$.

The spectrum is the mark of the indefinite theory. A compact isometry has its spectrum on the unit circle; a Krein isometry has it symmetric under $\lambda\mapsto1/\bar\lambda$ and not confined to the circle. A real eigenvalue off the circle is a boost, and the pair of reciprocal eigenvalues is the pair of scale factors on the two null directions.

The restriction of $T_t$ to the quaternionic slice is the Lorentz boost of rapidity $t$ along $\vec e_x$. The group therefore contains the boosts twice: once as operators on the coefficient space, and once as congruences of the algebra's elements, and the second is the seven-dimensional part of the previous section.

The Two Involutions

The group carries two involutions, and they are the reason the Cartan theory of the indefinite group is not the theory of the compact one.

The Krein involution is $\mathrm{J}_{J}(T)=(T^{\dagger})^{-1}$. Its fixed set is the group itself, since $T=T^{\dagger}$-inverse is the defining equation, so its fixed group is non-compact and it is not a Cartan involution.

The Cartan involution is $\mathrm{J}_{0}(T)=(T^{*})^{-1}$, built from the definite adjoint. On the group it is the inner automorphism $T\mapsto JTJ$, and its fixed group is the maximal compact subgroup, $$ \mathrm{Fix}(\mathrm{J}_{0})=\mathrm{U}(1)\times\mathrm{U}(3), $$ of dimension ten, which is compact, so this one is a Cartan involution. Its differential is $-M^{*}$ and the split is $$ \mathfrak{u}(1,3)=\mathfrak{k}_{0}\oplus\mathfrak{p}_{0}, \qquad \dim_{\mathbb{R}}\mathfrak{k}_{0}=10,\quad \dim_{\mathbb{R}}\mathfrak{p}_{0}=6 . $$

The dimensions add to the sixteen of the group, and the six of the hyperbolic part is the dimension of the symmetric space: $$ \mathrm{U}(1,3)/\bigl(\mathrm{U}(1)\times\mathrm{U}(3)\bigr)\cong\mathbb{CH}^{3}, $$ the complex hyperbolic space of §The Level Sets. The group decomposes as the compact part times the exponential of the hyperbolic part, which is the polar decomposition of the indefinite theory.

The two involutions are the two ways to make the group act on a space: the Krein one keeps the indefinite form and does not compactify; the Cartan one keeps a definite form and produces the symmetric space. The compact group has only the second, and that is why the indefinite group needs two.

The Krein involution also acts on the operator algebra, and there it splits the whole of $\mathfrak{gl}_4(\mathbb{C})$, not $\mathfrak{u}(1,3)$. Its derivative at the identity is $\theta(X)=-X^{\dagger}$, conjugate-linear because the dagger conjugates the scalars, with $\theta^{2}=\mathrm{id}$. The two eigenspaces are real subspaces of a complex algebra of real dimension $32$: $$ \mathfrak{gl}_4(\mathbb{C})=\mathfrak{u}(1,3)\oplus J\cdot\mathrm{Herm}_4, \qquad \dim_{\mathbb{R}}\mathfrak{u}(1,3)=16,\quad \dim_{\mathbb{R}}J\cdot\mathrm{Herm}_4=16 , $$ the $+1$ eigenspace $\{X:X^{\dagger}=-X\}$ being the Lie algebra of the group and the $-1$ eigenspace $\{X:X^{\dagger}=X\}$ being $J$ times the Hermitian matrices, since $(JX)^{\dagger}=JX$ for Hermitian $X$. Verified on a random complex $X$: the parts $P=\tfrac12(X+\theta(X))$ and $Q=\tfrac12(X-\theta(X))$ satisfy $P^{\dagger}+P=0$ and $Q^{\dagger}-Q=0$ to the last bit of double precision.

The two splits are of different algebras, and that is the whole distinction. The Cartan split is $10+6$ inside $\mathfrak{u}(1,3)$ and compacts the group; the Krein split is $16+16$ across the whole operator algebra and does not, because its fixed group is the non-compact group itself. So the involutions are not two conventions for one decomposition: one compacts a group and yields a symmetric space, the other decomposes the ambient algebra and yields none.

The Contractions

The operator theory adds a third layer, which the entry records and does not use. A $J$-contraction is an operator whose defect $\mathrm{id}-T^{\dagger}T$ is $J$-positive; the $J$-positive cone is $\mathcal{P}_{J}=J\{S\ge0\}$, the image of the definite positive cone under $J$; and the model contraction is the Krein-orthogonal projection $\pi_y=\tfrac12(\mathrm{id}-J)$ onto the vector subspace, whose defect is the projection $\pi_x=\tfrac12(\mathrm{id}+J)$ onto the centre. The set of contractions is invariant under conjugation by Krein isometries.

The cone contains $J$ and does not contain the identity, and it is not the cone of squares of the Jordan algebra. The point for the entry is that the indefinite theory has a positivity of its own, and it is not the positivity of the definite form: the identity fails to be positive, and the operator $J$, which is a reflection, is positive. An indefinite metric reverses which operators count as contractions, and the algebra's fourth form is where that reversal is written.

The Level Sets: The Shells and the State Space

The Three Levels

The indefinite form is real on the diagonal, $\langle\tilde Q,\tilde Q\rangle_{\natural*}=\lvert Q_t\rvert^{2}-\lvert Q_x\rvert^{2}-\lvert Q_y\rvert^{2}-\lvert Q_z\rvert^{2}$, and its level sets are hyperboloids: $$ \{\langle\tilde Q,\tilde Q\rangle_{\natural*}=1\}\cong S^{1}\times\mathbb{R}^{6}, \qquad \{\langle\tilde Q,\tilde Q\rangle_{\natural*}=-1\}\cong S^{5}\times\mathbb{R}^{2}, $$ $$ \{\langle\tilde Q,\tilde Q\rangle_{\natural*}=0\}=C\bigl(S^{1}\times S^{5}\bigr), $$ the cone on $S^{1}\times S^{5}$, with apex at the origin. All three are of real dimension seven. The first is homotopy equivalent to a circle, the second to a five-sphere, and the third is contractible; the complement of the apex in the null set is $S^{1}\times S^{5}\times\mathbb{R}_{>0}$.

Verified on the base points: $\langle e_t,e_t\rangle_{\natural*}=1$, $\langle\vec e_x,\vec e_x\rangle_{\natural*}=-1$, $\langle e_t+\vec e_x,e_t+\vec e_x\rangle_{\natural*}=0$, $\langle\vec e_x+\vec e_y,\vec e_x+\vec e_y\rangle_{\natural*}=-2$, and $\langle e_t+\tfrac12 i\vec e_x,e_t+\tfrac12 i\vec e_x\rangle_{\natural*}=0.75$.

The positive region $\{\lVert c\rVert>\lVert\tilde v\rVert\}$ is a cone and retracts onto the circle of phases, because the vector part ranges over an open ball for each fixed scalar. The negative region retracts onto the five-sphere of vector directions. The two regions and the null set make up the Krein cone, with the positive region as interior and the null set as boundary.

The Positive Lines and the Ball

A positive line is a complex line consisting of positive elements. Every such line meets the affine hyperplane $\{Q_t=1\}$ exactly once, and the representative is $e_t+\tilde Q_1$ with $$ \langle e_t+\tilde Q_1,e_t+\tilde Q_1\rangle_{\natural*}=1-\lVert\tilde Q_1\rVert^{2}, $$ so the line is positive exactly when $\lVert\tilde Q_1\rVert<1$. The assignment $\tilde Q_1\mapsto\mathbb{C}(e_t+\tilde Q_1)$ is therefore a bijection from the open unit ball of $\mathbb{C}^{3}$ onto the set of positive lines, and that set is

$$ \mathbb{CH}^{3}=\mathrm{U}(1,3)/\bigl(\mathrm{U}(1)\times\mathrm{U}(3)\bigr), $$

a real manifold of dimension six with the Bergman metric, whose geodesics are the complex lines.

The stabiliser of the positive line $\mathbb{C}e_t$ is exactly the maximal compact subgroup, so the ball is the symmetric space of the Cartan involution of the previous section. The compact group is the isotropy of a point; the hyperbolic part moves the point.

The Phase Bundle and the Boundary

The phase circle acts on the algebra by $\tilde Q\mapsto\lambda\tilde Q$ with $\lvert\lambda\rvert=1$, preserving the form and every level set, freely off the origin. Its quotient on the positive level set is the set of positive lines, so the positive level set is a principal circle bundle over the ball: $$ \{\langle\tilde Q,\tilde Q\rangle_{\natural*}=1\}\longrightarrow \mathbb{CH}^{3}\cong\mathbb{R}^{6}, $$ and a principal bundle over a contractible base is trivial, which is the diffeomorphism $S^{1}\times\mathbb{R}^{6}$ of the first theorem.

The boundary of the ball is the set of isotropic lines, the representatives with $\lVert\tilde Q_1\rVert=1$, a copy of $S^{5}$. The closure of the ball adds the boundary, and the boundary is exactly the set of null lines. Real scaling by $s>0$ relates the level $t$ to the level $s^{2}t$, so the positive region is the union of the positive levels, and the cone is the union over $t\ge0$ including the apex.

object dimension homotopy type reading
positive level set $7$ $S^{1}$ a phase circle over a contractible fibre
negative level set $7$ $S^{5}$ a direction sphere over a contractible fibre
null set $7$ a point the cone, contractible
positive region $8$ $S^{1}$ the cone over the positive level
positive lines $6$ a point the ball, $\mathbb{CH}^{3}$
isotropic boundary $5$ $S^{5}$ the unit sphere of $\mathbb{C}^{3}$

The table is the geometry of the fourth form in one page. The definite form gives spheres; the indefinite form gives a circle, a five-sphere and a ball, and the three are the three ways a nonzero element can sit with respect to the cone.

The Four Groups Compared

The entry can now be stated as one table. Four groups act, and each is characterised by the object it preserves and by what it cannot see.

$\mathrm{U}(\mathbb{B})$ $\mathbb{B}^{\times}_1$ $\mathbb{B}^{\times}$ $U_{J}(\mathbb{B})$
dimension $4$ $6$ $8$ $16$
compact yes no no no
generated by elements of the algebra yes yes yes no
the product preserved preserved by the adjoint preserved by the adjoint not preserved
the sectors preserved preserved by the congruence preserved by the congruence preserved
the indefinite form preserved preserved by the congruence preserved up to $\lvert N\rvert^{2}$ preserved
the time axis fixed moved by the congruence moved by the congruence moved
the double cover $S^{3}\to SO(3)$ $\mathbb{B}^{\times}_1\to SO^{+}(1,3)$ — —

Three readings follow from the table and nothing else.

The columns are ordered by what they give up. The compact group preserves everything the algebra has. The norm-one group gives up the product, keeping the form and the sectors. The group of invertibles gives up the exact preservation of the form, keeping it up to scale. The indefinite group gives up the product entirely. Each column to the right is a weaker invariance, and the weakening is what makes the group larger.

Only one column is compact. Compactness is the property that makes a group a gauge group, and it belongs to the dagger. Every other group of the table contains a boost and is non-compact.

Only one column is not generated by the algebra. The indefinite group is the group of the form, not of the algebra. Its extra nine dimensions are the price of asking for isometries of a form that the product does not determine.

The table also explains the naming of the entry. The rotations are the intersection of the first two columns; the boosts are the difference between the second and the first; the gauge is the first column; and the indefinite limit is the fourth.

The Interpretations

A rotation is a symmetry of the algebra and a boost is a motion of the space. The adjoint preserves the product and the congruence preserves the form; the two coincide exactly on the $*$-unitary slice. A rotation lies in that slice, so there is a single map of the algebra that is both an automorphism and an isometry, and the phrase "a rotation is a symmetry" needs no qualification. A boost does not lie in the slice, so each action shows one half of it: the congruence shows the motion and hides the algebra, the adjoint shows the algebra and hides the motion. Interpreted physically, a change of velocity is not a relabelling of the algebra's operations; it is a change of the element. The boundary: this is a statement about the action on the algebra, not about a field, and a field transforms through a representation chosen separately.

The half angle is the double cover and the minus sign is the fermion. The rotor is a square root of the rotation, so a full turn of the object is a half turn of the rotor, and the rotor returns with a minus sign. The sign is arithmetic: it is what a square root does after one turn. Interpreted physically, the spinor needs two turns to come back, and the algebra produces the two turns without an extra hypothesis. The boundary: this is a statement about the covering and not about statistics; exchange symmetry needs the permutation of two particles, which the group action alone does not supply.

The algebra reaches four components through two operations that are not motions. The congruence of a norm-one unit is proper and orthochronous, and the other three components need the quaternion conjugation of the argument, the negation of the result, or both. Interpreted physically, parity and time reversal are not changes of frame: they are operations on the argument, and the algebra says so by putting them outside the two-sided action. The boundary: discrete symmetries of a field theory are operators on a Hilbert space, and the entry supplies the placement of the algebraic operations, not the operators.

The antiunitary coset is why time reversal cannot be a motion. The real automorphism group has a second component, the coset of complex conjugation, which negates the central imaginary. Every inner automorphism fixes the centre pointwise, so no motion can do what a time reversal must do. Interpreted physically, an antiunitary operator is the algebraic shape of a symmetry that reverses every phase, and Kramers degeneracy is the consequence of its square being the $2\pi$ rotor. The boundary: the algebra supplies the involution and its square; that a physical time reversal is antiunitary is an input from quantum mechanics.

A single reflection is not in the algebra. The algebra is the even part of the Clifford envelope, the reflection count is the length of a versor, and the parity of the length is the determinant. A reflection is odd, so it lies in the odd slot; a product of two reflections is even, so it lies in the algebra. Interpreted physically, a mirror operation is not a rotation, and no rearrangement of the algebra's elements produces one. The boundary: the odd slot is itself a copy of the algebra, so the obstruction is the grading and not the size of the algebra.

The compact group is the gauge group and carries the topology. The shape of the group of invertibles is the shape of the compact slice, by two retractions: the group of invertibles onto the phases and the rotations, the norm-one group onto the rotations. The only obstruction is the phase winding, and the norm winds it twice. Interpreted physically, a gauge transformation is the compact part, and every unit deformation is homotopic to a gauge transformation: no homotopy class separates a boost from a rotation. The boundary: this is a statement about the group, not about a gauge theory; no connection, no field strength and no dynamics enter.

The indefinite group is larger than the algebra. The isometry group of the fourth form has dimension sixteen, and the algebra's elements reach seven. Interpreted physically, a theory built on the indefinite form has a symmetry group larger than its own algebra, and the nine extra dimensions are isometries that no product or conjugate of the algebra can generate. The gap is where an indefinite sector's extra symmetry lives, and the reciprocal eigenvalues of the boost are the mark of the indefinite metric. The boundary: that the extra isometries are physically realised is not claimed; the entry states only the dimension of the gap.

The level sets are the shells and the state space at once. The positive level set is a phase circle over a contractible base, the negative one is a direction sphere, the null one is contractible, and the positive lines form the complex hyperbolic ball. Interpreted physically, a massive state carries a phase and a massless one does not, and the projective state space is the ball with the isotropic sphere as its boundary. The boundary: the level sets of the fourth form are not the level sets of the algebraic norm; the algebraic norm's zero set is a different cone, and the two must not be identified.

The boundary of the interpretations. The two actions act on the algebra and on its Hermitian copy, not on a Hilbert space of states. The rotations act on the finite-dimensional sectors and the boosts act on the same sectors, but a boost is not $*$-unitary, so it does not preserve the dagger form and therefore does not act unitarily on a state space built from that form. A unitary representation of the full Lorentz group on a finite-dimensional space does not exist, and the algebra does not supply one: it supplies a congruence on the Hermitian forms, which is a representation on the forms and not on the states. The gap between the two is the gap between the covariant action and the unitary representation, and the entry does not close it.

Limits

The entry is an arrangement, and the arrangement has edges.

It does not classify the representations. The groups are described by their actions on the algebra and on one copy of the Lorentzian space. Their irreducible representations are not enumerated, and the spin classification appears only in the carrier language: two half-integer carriers and one integer carrier. A representation-theoretic account would be a different entry.

It does not treat the infinite-dimensional theory. The retractions, the homotopy groups and the covers are computed for the finite-dimensional group. Whether the same statements hold for the group of invertibles of a completed algebra, or for a group of operators on an infinite-dimensional module, is not addressed.

It does not identify the physical group with the algebraic one. That the compact slice is the gauge group, and that the indefinite group is the group of the fourth form, are identifications of structure. A physical gauge group is a group of transformations of fields with a connection, and the entry supplies no connection.

It does not resolve the fourth form's status. The algebra has four forms, and the entry uses the definite one for the compact group and the indefinite one for the Krein group. Which of them a physical theory should use is not a group-theoretic question, and the entry does not answer it.

It does not treat the discrete symmetries as operators. The four components are reached by two operations on the argument, and the placement is algebraic. The passage from the placement to an operator on a Hilbert space, with its square and its commutation with the Hamiltonian, is the archived articles' work.

It does not derive a gauge group. The compact slice is read as the gauge group of the state space, and its quotients $S^{1}$, $SO(3)$ and $S^{2}$ are computed, but no group of a field theory follows: what the algebra supplies is $\mathrm{U}(2)$ and its quotients, and the identification of one of them with an internal gauge group is an input. In particular the entry does not produce $\mathrm{SU}(3)\times\mathrm{SU}(2)\times\mathrm{U}(1)$, and it makes no claim to.

It does not give the metric on the symmetric space. The ball, the Bergman metric and the geodesics are recorded as the shape of the positive lines. The curvature computations, the comparison with the real hyperbolic case and the boundary theory remain with the geometry article.

Summary

The algebra has one group of invertibles and two actions on itself. The adjoint $\mathrm{Ad}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{-1}$ preserves the product, has kernel the centre and image $PGL_2(\mathbb{C})\cong SO^{+}(1,3)$. The congruence $\mathrm{Cg}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{*}$ preserves the two Hermitian sectors, scales the norm by $\lvert N(\tilde U)\rvert^{2}$, has kernel the central circle and image of dimension seven. The two coincide exactly on the $*$-unitary slice, and that coincidence is the sense in which a rotation is a symmetry and a boost is not: a rotation is $*$-unitary, a boost is not.

The two one-sided actions complete the picture. Left and right multiplication are representations of the same group on the algebra as a module: they preserve the definite form, do not preserve the product and do not preserve the sectors, and they carry the element rather than the frame. The congruence of the compact slice factors through them, $\mathrm{Cg}_{\tilde Q}=L_{\tilde Q}R_{\tilde Q^{*}}$, and the group the algebra's actions generate on itself has dimension seven inside the sixteen of $\mathrm{U}(4)$, the same seven of sixteen as in the indefinite case.

The reflections of the vector subspace are the negated congruences of real unit vectors. On the real form the dagger negates a vector, so $\mathrm{Cg}_v$ is a half turn of determinant $+1$ and the mirror is $\rho_v=-v\tilde Xv^{\natural}$, of determinant $-1$ on the vector subspace and $+1$ on the algebra; Cartan–Dieudonné generates $\mathrm{O}(3)$ from at most three of them, and no path of invertibles supplies the minus sign.

The group of invertibles is $\mathbb{B}^{\times}\cong GL_2(\mathbb{C})$ of dimension eight, with norm-one part $\mathbb{B}^{\times}_1\cong SL_2(\mathbb{C})$ of dimension six and maximal compact part $\mathrm{U}(\mathbb{B})\cong\mathrm{U}(2)$ of dimension four. The two middle groups intersect in $S^{3}$. Over $\mathbb{C}$ every automorphism is inner and $\mathrm{Aut}_{\mathbb{C}}\cong PGL_2(\mathbb{C})$; over $\mathbb{R}$ the group gains exactly one component, the coset of complex conjugation, which is conjugate-linear, not inner, and the algebraic shape of an antiunitary operator. The derivations are the traceless part, of dimension six, and they do not grow over $\mathbb{R}$.

The algebra is the even part of the Clifford envelope, so its elements realise only even numbers of reflections: $\mathrm{Pin}(1,3)$ is not contained in $\mathbb{B}$, and $\mathrm{Spin}(1,3)$ is its whole intersection with $\mathbb{B}$. On the Hermitian copy the congruence reaches the proper orthochronous component, and the other three are reached by conjugating the argument, negating the result, or both. On the Lorentzian copy the inner automorphism reaches one component more, by the volume element, and the surplus is exactly the replacement of the dagger by the inverse.

The compact group is $\mathrm{U}(2)=S^{1}\cdot S^{3}$ with intersection $\{\pm e_t\}$, with maximal torus $T^{2}$, flag variety $S^{2}$, and quotients $S^{1}$, $SO(3)$ and $S^{2}$. The group of invertibles retracts onto it and the norm-one group onto $S^{3}$, so $\pi_1\cong\mathbb{Z}$, $\pi_2=0$, $\pi_3\cong\mathbb{Z}$; the norm maps the phase circle to itself with degree two.

The isometry group of the fourth form is $\mathrm{U}(1,3)$ of dimension sixteen, with maximal compact $\mathrm{U}(1)\times\mathrm{U}(3)$ of dimension ten, symmetric space $\mathbb{CH}^{3}$ of dimension six, and the algebra's elements reach seven of the sixteen dimensions. Its boosts have reciprocal real spectrum and restrict to Lorentz boosts. Its level sets are a circle times a cell, a five-sphere times a cell, and a cone; the positive lines are the ball.

Summary of Notation

Symbol Meaning
$\mathbb{B}^{\times}\cong GL_2(\mathbb{C})$ group of invertibles; real dimension $8$
$\mathbb{B}^{\times}_1\cong SL_2(\mathbb{C})$ norm-one group; real dimension $6$
$\mathrm{U}(\mathbb{B})\cong\mathrm{U}(2)$ $*$-unitary group; the compact slice; real dimension $4$
$S^{3}=Sp(1)$ unit quaternions; the rotations
$\mathrm{Ad}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{-1}$ adjoint action; the automorphism
$\mathrm{Cg}_{\tilde U}(\tilde X)=\tilde U\tilde X\tilde U^{*}$ congruence; the two-sided operator; the motion
$L_{\tilde Q}(\tilde X)=\tilde Q\tilde X$, $R_{\tilde Q}(\tilde X)=\tilde X\tilde Q$ one-sided actions; the module action; the spinor
$\rho_v(\tilde X)=-v\tilde Xv^{\natural}$ reflection in $v^{\perp}$ on the vector subspace; minus a congruence
$\mathrm{U}(4)$ isometry group of the Euclidean form on $\mathbb{B}\cong\mathbb{C}^{4}$; real dimension $16$
$\langle\tilde X,\tilde Y\rangle_{*}=\mathrm{Sc}(\tilde X\tilde Y^{*})$ the Euclidean form; the eight basis elements are orthonormal
$\mathrm{Aut}_{\mathbb{C}}(\mathbb{B})\cong PGL_2(\mathbb{C})$ automorphism group over $\mathbb{C}$
$\mathrm{Aut}_{\mathbb{R}}(\mathbb{B})\cong\mathrm{Aut}_{\mathbb{C}}\rtimes\mathbb{Z}/2$ automorphism group over $\mathbb{R}$; two components
$\mathrm{Der}(\mathbb{B})\cong\mathbb{B}/\mathbb{B}_{C}$ derivations; the scalar-free part
$\mathrm{Cl}(V)=\mathbb{B}\oplus\mathbb{B}\zeta$ Clifford envelope; the algebra is the even slot
$\Omega=\eta_t\eta_x\eta_y\eta_z$ volume element; $\Omega^{2}=-1$
$U_{J}(\mathbb{B})\cong\mathrm{U}(1,3)$ Krein isometry group; real dimension $16$
$\mathbb{CH}^{3}\cong\mathrm{U}(1,3)/(\mathrm{U}(1)\times\mathrm{U}(3))$ complex hyperbolic space; the positive lines
$\tilde u=\cos\tfrac{\theta}{2}+\sin\tfrac{\theta}{2}\vec u$ rotation rotor; the angle is halved
$\tilde U=\cosh\tfrac{\theta}{2}+i\sinh\tfrac{\theta}{2}\vec u$ boost element; Hermitian and of norm one

Further Reading