Expressing Lie Algebras via Biquaternions
Introduction
To express a Lie algebra in biquaternions is to do one of two different things, and the two give different lists.
- The algebra as a subspace. A real or complex subspace $\mathrm{G}\subset\mathbb{B}$ closed under the commutator, with the commutator as its own bracket.
- The algebra as a set of operators. A Lie algebra of operators on the carrier $\mathbb{B}$, generated by left multiplication $\rho_L(\tilde Q)$ and right multiplication $\rho_R(\tilde Q)$.
The first sense is short. The biquaternion algebra is $M_2(\mathbb{C})$, so its complex Lie subalgebras are the subalgebras of $\mathfrak{gl}(2,\mathbb{C})$, and there is nothing exotic in them. Its real subalgebras are another matter, and they are where the biquaternion language earns its place: the real algebra $M_2(\mathbb{C})_{\mathbb{R}}$ carries the compact, the split and the complex readings of $\mathfrak{sl}(2,\mathbb{C})$ as three different real subspaces of one algebra.
The second sense is where the algebras that do not fit inside $\mathbb{B}$ appear. Left and right multiplication commute, so a subalgebra of $\mathbb{B}$ yields two commuting copies of itself as operators, and the sum is the orthogonal algebra attached to the norm. That is how $\mathfrak{so}(4)$ and $\mathfrak{so}(2,2)$ are biquaternionic without being subspaces of $\mathbb{B}$.
The strongest objection, stated first. If $\mathbb{B}\cong M_2(\mathbb{C})$, why expect anything at all? The answer is that the isomorphism is one of complex algebras. Over $\mathbb{C}$ it settles the question and there is nothing left to say; over $\mathbb{R}$ it says only that the real algebra is $M_2(\mathbb{C})$ read over the reals, of real dimension eight, and that algebra has many real forms inside it. The whole content of the article is in the passage from the complex statement to the real one.
One formula decides the first sense. The bracket of the corpus is the antisymmetrisation of the plain product, and its value is the cross product of the vector parts,
$$ \tilde Q_1\wedge\tilde Q_2=\tfrac12\bigl(\tilde Q_1\tilde Q_2-\tilde Q_2\tilde Q_1\bigr)=\vec Q_1\times\vec Q_2 . $$
The scalar part never enters. Every statement below about subalgebras of $\mathbb{B}$ is a consequence of that one line, and §The One Formula That Decides the Question draws the consequences once and for all.
What is taken as known. The base, the subspaces and the conjugations are Algebra and Element Conventions. The $2\times2$ image $\mathsf{M}_2$ and the regular operators are Matrix Representations Conventions. The bracket on $\mathbb{B}$, its defect, its ideals and its Killing form are The Lie Structure of the Antisymmetric Plain Product. The vocabulary of Lie algebras, the Killing form, the adjoint map and the real forms are Lie Algebras, Structure of Lie Algebras, The Adjoint Action of a Lie Algebra and Real Forms of a Complex Lie Algebra. The exponential and the spin cover are Biquaternion Lie Group and Exponential Structure. The three real subspaces of $\mathbb{B}$ and their Killing signatures are also recorded in Coda: $n=2$ and the Biquaternion Algebra, and the internal note $\mathfrak{su}(2)$, $\mathfrak{su}(n)$ and the Anti-Hermitian Question in Biquaternion Language fixes the dictionary between $\mathfrak{su}(2)$ and the real pure quaternions.
Conventions. The element is $\tilde Q=Q_te_t+Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z$ with $Q_\mu\in\mathbb{C}$, the scalar part is $Q_t$ and the vector part is $\vec Q$. The scalar part is $\mathrm{Sc}(\tilde Q)=Q_t$, read off the plain form $\mathrm{Sc}(\tilde Q_1\tilde Q_2)$ at $\tilde Q_2=e_t$, so that the traceless condition means $Q_t=0$. The star is $\tilde Q^{*}=\bar Q_te_t-\bar Q_x\vec e_x-\bar Q_y\vec e_y-\bar Q_z\vec e_z$, and $\mathbb{B}_{\Phi}$ and $\mathbb{B}_{\Psi}$ are its anti-fixed and fixed subspaces, the anti-Hermitian and the Hermitian ones. The centre is $\mathbb{B}_C=\mathbb{C}e_t$ and the vector subspace is $\mathbb{B}_V$. The $2\times2$ image is $\mathsf{M}_2(\tilde Q)=Q_tI_2-i(Q_x\sigma_1+Q_y\sigma_2+Q_z\sigma_3)$, and the left and right operators are $\rho_L(\tilde Q)$ and $\rho_R(\tilde Q)$. The norm is $N(\tilde Q)=\sum_\mu Q_\mu^2$. The Killing form is $\kappa$. Lie algebras are written in Fraktur, $\mathfrak{su}(2)$, $\mathfrak{so}(3,1)$, $\mathfrak{gl}(2,\mathbb{R})$, and a generic Lie algebra is $\mathrm{G}$.
One factor is fixed here and never changes. The bracket of the corpus is the half-commutator. The matrix commutator of the images is twice it, $\mathsf{M}_2(\tilde Q_1\wedge\tilde Q_2)=\tfrac12[\mathsf{M}_2(\tilde Q_1),\mathsf{M}_2(\tilde Q_2)]$, because $\mathsf{M}_2$ is an algebra map. No isomorphism class below depends on the factor, the structure constants in the basis $\vec e_x,\vec e_y,\vec e_z$ are the Levi-Civita symbol, and the bracket of operators is taken with the same half, $[A,B]=\tfrac12(AB-BA)$, so that one convention holds throughout.
The Algebra and the Two Base Fields
The Isomorphism and the Two Dimensions
The algebra is $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$, with $\vec e_k^{\,2}=-e_t$ and $\vec e_x\vec e_y=\vec e_z$ cyclically. The map $\mathsf{M}_2$ is an isomorphism onto $M_2(\mathbb{C})$, so the algebra carries two dimensions:
| reading | algebra | dimension |
|---|---|---|
| complex | $\mathbb{B}\cong M_2(\mathbb{C})$ | $4$ complex |
| real | $\mathbb{B}\cong M_2(\mathbb{C})_{\mathbb{R}}$ | $8$ real |
Both entries are the same algebra and neither is a new object. What differs is the supply of subalgebras, and the supply differs because a real subspace is a weaker condition than a complex one.
What the Real Reading Adds
A complex subspace of $\mathbb{B}$ that is closed under the bracket is closed under multiplication by $i$ as well, so it is a complex Lie algebra of complex dimension $1$, $2$, $3$ or $4$. A real subspace need not be closed under $i$, and it may sit in $\mathbb{B}$ at an angle to the complex structure. That is the whole source of the extra real forms: the four complex dimensions become eight real ones, and a three-complex-dimensional subspace can be cut by a real form in several ways.
The reason is visible in one example. The traceless part $\mathbb{B}_V$ is three complex dimensions; the real span of $\vec e_x,\vec e_y,\vec e_z$ is three real dimensions inside it, and so is the real span of $i\vec e_x,\vec e_y,i\vec e_z$. Both are closed, both have real dimension three, and they are not isomorphic: the first is compact, the second split. One complex algebra, two real forms, one carrier.
The One Formula That Decides the Question
The Bracket Ignores the Scalar Part
Proposition. For all $\tilde Q_1,\tilde Q_2$ the bracket is the cross product of the vector parts, $\tilde Q_1\wedge\tilde Q_2=\vec Q_1\times\vec Q_2$, with zero scalar part and no mixed term.
Proof. Expand the plain product. The scalar part $Q_{1t}Q_{2t}-\vec Q_1\cdot\vec Q_2$ is symmetric in the two arguments and cancels. The two terms $Q_{1t}\vec Q_2$ and $Q_{2t}\vec Q_1$ of the vector part are symmetric and cancel. What is left is twice the cross product, and the half removes the two. Computed in the basis, $[\vec e_x,\vec e_y]=\vec e_z$ and its cyclic companions; checked on one hundred random pairs, the defect is $9\times10^{-16}$.
Three consequences follow at once, and they are used everywhere below.
- The bracket lands in $\mathbb{B}_V$, and $\mathbb{B}_V=[\mathbb{B},\mathbb{B}]$. The bracket of any two elements is traceless, and the cross product on $\mathbb{B}_V$ is onto, so the derived algebra is exactly the traceless part.
- The centre is $\mathbb{B}_C$, of real dimension two. An element is central exactly when its vector part vanishes; $\mathbb{B}_C=\mathbb{C}e_t$ is the centre and the radical at once.
- The mixed part is invisible to the bracket. The distinction between a biquaternion and its "vector part plus scalar part" does not show in any bracket, which is why the Hermitian and anti-Hermitian statements below have to be checked by hand and do not follow from closure alone.
The Consequence for Subalgebras
Theorem. Let $\mathrm{G}\subset\mathbb{B}$ be a real subspace, and let $\pi_V$ be the projection onto the vector part along $\mathbb{B}_C$. Then $\mathrm{G}$ is a Lie subalgebra if and only if $\pi_V(\mathrm{G})$ is closed under the cross product. Its bracket is then the cross product of the vector parts of its elements, and its centre contains $\mathrm{G}\cap\mathbb{B}_C$.
Proof. The bracket of two elements of $\mathrm{G}$ is $\vec Q_1\times\vec Q_2$ by the proposition, and this lies in $\mathrm{G}$ exactly when it lies in $\pi_V(\mathrm{G})$, because the bracket has no scalar part. The last clause holds because $\mathbb{B}_C$ is central in $\mathbb{B}$.
So the classification of real subalgebras of $\mathbb{B}$ reduces to two independent choices: a real subspace $\mathfrak{s}\subset\mathbb{B}_C$ of real dimension $0$, $1$ or $2$, and a real subspace $W\subset\mathbb{B}_V$ closed under the cross product, with $\mathfrak{s}=\mathrm{G}\cap\mathbb{B}_C$ and $W=\pi_V(\mathrm{G})$. The lift of $W$ inside $\mathrm{G}$ is free: since the bracket ignores the scalar part, any lift gives the same bracket.
This reduction is the reason the tables below are short. Everything that is genuinely a statement about Lie algebras is a statement about $W\subset\mathbb{B}_V$; the scalar directions only add an abelian factor.
The Complex Subalgebras
The Scalar Part
The centre $\mathbb{C}e_t$ is a complex Lie algebra of dimension one, abelian, isomorphic to $\mathfrak{gl}(1,\mathbb{C})$. Any complex subspace of it is a complex Lie subalgebra, so the possible scalar parts are $0$ and $\mathbb{C}e_t$.
The Vector Part
By the theorem, a complex subspace $W\subset\mathbb{B}_V$ is a complex Lie subalgebra exactly when it is closed under the cross product on $\mathbb{C}^3$. There are four possibilities.
- Dimension $0$: $W=0$.
- Dimension $1$: any complex line. A line is closed, because the cross product of two proportional vectors vanishes. This is the abelian algebra $\mathbb{C}$.
- Dimension $2$: a plane is closed exactly when it contains its own normal. For two independent $\vec u,\vec v$ in a plane $P$ the cross product $\vec u\times\vec v$ is orthogonal to both, so it lies along the normal $\vec n$ of $P$; the bracket is in $P$ exactly when $\vec n\in P$. Such a plane is isotropic, $\vec n\cdot\vec n=0$. Computed on one hundred random complex planes, every one fails closure, with a smallest defect of $0.71$; the plane spanned by $\vec e_x+i\vec e_y$ and $\vec e_z$ is closed, with $[\vec e_x+i\vec e_y,\vec e_z]=i(\vec e_x+i\vec e_y)$ and defect $9\times10^{-16}$.
- Dimension $3$: all of $\mathbb{B}_V$, which is $\mathfrak{so}(3,\mathbb{C})$.
The isotropic plane is the Borel subalgebra. Under $\mathsf{M}_2$ it is the set of upper-triangular traceless matrices: the condition is $Q_y=iQ_x$ with $Q_t=0$, and the two free complex parameters are $Q_x$ and $Q_z$.
The Eight Classes
Combining the two parts gives eight complex subalgebras, and the table is the whole list.
| scalar part | vector part | complex dimension | name |
|---|---|---|---|
| $0$ | $0$ | $0$ | the zero algebra |
| $0$ | a line | $1$ | $\mathfrak{gl}(1,\mathbb{C})$ |
| $\mathbb{C}e_t$ | $0$ | $1$ | the centre, $\mathfrak{gl}(1,\mathbb{C})$ |
| $0$ | an isotropic plane | $2$ | a Borel of $\mathfrak{sl}(2,\mathbb{C})$ |
| $\mathbb{C}e_t$ | a line | $2$ | abelian |
| $0$ | $\mathbb{B}_V$ | $3$ | $\mathfrak{sl}(2,\mathbb{C})$ |
| $\mathbb{C}e_t$ | an isotropic plane | $3$ | a Borel of $\mathfrak{gl}(2,\mathbb{C})$ |
| $\mathbb{C}e_t$ | $\mathbb{B}_V$ | $4$ | $\mathfrak{gl}(2,\mathbb{C})$ |
The verdict on the complex side. There is no new complex Lie algebra here, and the reason is structural rather than accidental: $\mathbb{B}$ is $M_2(\mathbb{C})$, and the classification above is the classification of the subalgebras of $\mathfrak{gl}(2,\mathbb{C})$ written in the coefficient basis. The count of eight agrees with that classification: the traceless part has the zero algebra, the lines, the Borels and the whole, and the scalar line can be adjoined or not.
The Real Subalgebras
The named real subalgebras are listed here with a representative basis, its real dimension, and the invariant that fixes its isomorphism class. The invariant used throughout is the signature of the Killing form, because on the algebras of dimension at most six that occur here it separates the compact, the split and the complex readings, and separates the reductive cases as well.
The Compact Form: Real Pure Quaternions
Take $W=\operatorname{span}_{\mathbb{R}}\{\vec e_x,\vec e_y,\vec e_z\}$, with no scalar part. The bracket is $[\vec e_x,\vec e_y]=\vec e_z$ and its cyclic companions, which are the structure constants of $\mathfrak{so}(3)$; the Killing form is negative definite, of signature $(0,3)$; the real dimension is $3$. The algebra is $\mathfrak{su}(2)\cong\mathfrak{so}(3)$.
Two readings make the identification concrete. Under $\mathsf{M}_2$ the three units become $-i\sigma_1,-i\sigma_2,-i\sigma_3$, anti-Hermitian and traceless, so the subspace is $\mathfrak{su}(2)$ in its usual matrix description; and the element $-\tfrac12\vec q$ for real $\vec q$ maps to $\tfrac{i}{2}(q_x\sigma_1+q_y\sigma_2+q_z\sigma_3)$, the basis $T_a=i\sigma_a$ of the corpus. The subspace is also $\mathbb{B}_{\Phi}\cap\mathbb{B}_V$, the vector part of the anti-Hermitian sector: $\vec e_k^{*}=-\vec e_k$, so the real pure quaternions are anti-Hermitian.
Why the real units and not the imaginary ones. The imaginary vector units $i\vec e_x,i\vec e_y,i\vec e_z$ are Hermitian, and they do not close: $[i\vec e_x,i\vec e_y]=-\vec e_z$ is anti-Hermitian. The Hermitian half of the algebra is not a Lie subalgebra, its closure defect is $6.8$ on one hundred random pairs, and this is the sharpest single fact of the whole article. The compact form lives in the anti-Hermitian half, and only there.
The Split Form: Real Traceless
Take $W=\operatorname{span}_{\mathbb{R}}\{i\vec e_x,\vec e_y,i\vec e_z\}$, with no scalar part. Closure is checked on the three brackets: $[i\vec e_x,i\vec e_z]=\vec e_y$, $[i\vec e_x,\vec e_y]=i\vec e_z$, $[\vec e_y,i\vec e_z]=i\vec e_x$, and each value is in the span. The Killing form has signature $(2,1)$, so the algebra is the split real form $\mathfrak{sl}(2,\mathbb{R})\cong\mathfrak{so}(2,1)$, of real dimension $3$.
Under $\mathsf{M}_2$ the basis becomes $\sigma_1$, $-i\sigma_2$ and $\sigma_3$, three real matrices of trace zero, so the subspace is exactly $\mathfrak{sl}(2,\mathbb{R})\subset M_2(\mathbb{R})$. The same computation shows the surrounding four-dimensional statement: adjoining $e_t$ gives $\operatorname{span}_{\mathbb{R}}\{e_t,i\vec e_x,\vec e_y,i\vec e_z\}$, whose image is all of $M_2(\mathbb{R})$, with $\kappa$ of signature $(2,1,1)$ and a one-dimensional radical, the scalar line. That is $\mathfrak{gl}(2,\mathbb{R})$.
The Complex Form Read over the Reals
Take $W=\mathbb{B}_V$ with no scalar part: the traceless biquaternions, real dimension $6$, complex dimension $3$. The algebra is $\mathfrak{sl}(2,\mathbb{C})$ read over $\mathbb{R}$, and its Killing form has signature $(3,3)$. Since $\mathfrak{sl}(2,\mathbb{C})_{\mathbb{R}}\cong\mathfrak{so}(3,1)$, this is the Lorentz algebra.
The identification is again by signature. The real forms of $\mathfrak{so}(4,\mathbb{C})$ are $\mathfrak{so}(4)$, $\mathfrak{so}(3,1)$ and $\mathfrak{so}(2,2)$, with Killing signatures $(0,6)$, $(3,3)$ and $(4,2)$; $\mathbb{B}_V$ has $(3,3)$ and is therefore $\mathfrak{so}(3,1)$.
The phrase "the three real forms of $\mathfrak{sl}(2,\mathbb{C})$" is a slip, and it is worth correcting once. $\mathfrak{sl}(2,\mathbb{C})$ has exactly two real forms, the compact $\mathfrak{su}(2)$ and the split $\mathfrak{sl}(2,\mathbb{R})$, in agreement with Real Forms of a Complex Lie Algebra, §Worked Case. The third object, $\mathfrak{sl}(2,\mathbb{C})$ read over $\mathbb{R}$, is not a real form of $\mathfrak{sl}(2,\mathbb{C})$: its complexification is $\mathfrak{sl}(2,\mathbb{C})\oplus\mathfrak{sl}(2,\mathbb{C})$, of complex dimension six, and not $\mathfrak{sl}(2,\mathbb{C})$. What is true, and what the biquaternion picture shows, is that one carrier holds three algebras of rank one:
| algebra | basis | real dimension | $\kappa$ signature | complexification |
|---|---|---|---|---|
| $\mathfrak{su}(2)$, compact form | $\vec e_x,\vec e_y,\vec e_z$ | $3$ | $(0,3)$ | $\mathfrak{sl}(2,\mathbb{C})$ |
| $\mathfrak{sl}(2,\mathbb{R})$, split form | $i\vec e_x,\vec e_y,i\vec e_z$ | $3$ | $(2,1)$ | $\mathfrak{sl}(2,\mathbb{C})$ |
| $\mathfrak{sl}(2,\mathbb{C})$ read over $\mathbb{R}$ | $\vec e_x,\vec e_y,\vec e_z,i\vec e_x,i\vec e_y,i\vec e_z$ | $6$ | $(3,3)$ | $\mathfrak{sl}(2,\mathbb{C})\oplus\mathfrak{sl}(2,\mathbb{C})$ |
The column of complexifications is the warrant: the first two double into $\mathfrak{sl}(2,\mathbb{C})$ and the third doubles into two copies, so the third is of another kind. The three are nonetheless three real subspaces of one algebra, and that is the honest content of the pun.
The Anti-Hermitian Subspace: $\mathfrak{u}(2)$
Take $\mathrm{G}=\mathbb{B}_{\Phi}$, the anti-Hermitian biquaternions. In coefficients the condition $\tilde Q^{*}=-\tilde Q$ reads $Q_t=i\alpha$ with $\alpha$ real and $Q_x,Q_y,Q_z$ real, so the subspace has basis $ie_t,\vec e_x,\vec e_y,\vec e_z$ and real dimension $4$. It is closed, and its Killing form has signature $(0,3,1)$: negative semidefinite with a one-dimensional radical, the line $\mathbb{R}(ie_t)$.
A semidefinite Killing form with a radical of dimension one says that the algebra is compact and reductive with a one-dimensional centre, so it is $\mathfrak{su}(2)\oplus\mathfrak{u}(1)$, that is $\mathfrak{u}(2)$. Under $\mathsf{M}_2$ the subspace is the set of anti-Hermitian matrices, $M^{\dagger}=-M$, which is $\mathfrak{u}(2)$ by definition.
This subspace is the material sector of the corpus, and the identification is worth stating as a reading rather than a coincidence: $\mathbb{B}_{\Phi}=(\mathbb{B}_{\Phi}\cap\mathbb{B}_C)\oplus(\mathbb{B}_{\Phi}\cap\mathbb{B}_V)$ is the splitting $\mathfrak{u}(1)\oplus\mathfrak{su}(2)$ of the material sector into its central phase and its rotation algebra. The Hermitian sector $\mathbb{B}_{\Psi}$ is not a Lie algebra under the bracket, as was said above.
The Named Table
The real subalgebras that have names, with the invariant that fixes each.
| algebra | basis | real dimension | $\kappa$ signature | centre |
|---|---|---|---|---|
| $\mathbb{R}$ | $\mathbb{R}(ie_t)$ | $1$ | $(0,0,1)$ | $1$ |
| $\mathbb{R}^2$ | $\mathbb{C}e_t$ | $2$ | $(0,0,2)$ | $2$ |
| $\mathfrak{su}(2)\cong\mathfrak{so}(3)$ | $\vec e_x,\vec e_y,\vec e_z$ | $3$ | $(0,3)$ | $0$ |
| $\mathfrak{sl}(2,\mathbb{R})\cong\mathfrak{so}(2,1)$ | $i\vec e_x,\vec e_y,i\vec e_z$ | $3$ | $(2,1)$ | $0$ |
| a real Borel | $i\vec e_x,\vec e_y+i\vec e_z$ | $2$ | $(1,0,1)$ | $0$ |
| $\mathfrak{u}(2)$ | $ie_t,\vec e_x,\vec e_y,\vec e_z$ | $4$ | $(0,3,1)$ | $1$ |
| $\mathfrak{gl}(2,\mathbb{R})$ | $e_t,i\vec e_x,\vec e_y,i\vec e_z$ | $4$ | $(2,1,1)$ | $1$ |
| $\mathfrak{so}(3,1)\cong\mathfrak{sl}(2,\mathbb{C})_{\mathbb{R}}$ | $\mathbb{B}_V$ | $6$ | $(3,3)$ | $0$ |
| $\mathfrak{gl}(2,\mathbb{C})_{\mathbb{R}}$ | $\mathbb{B}$ | $8$ | $(3,3,2)$ | $2$ |
The last row is the algebra itself: the whole of $\mathbb{B}$, of real dimension eight, with the two-dimensional centre and the Killing form of the corpus. The row before it is its traceless part. The two rows of dimension two are there for contrast. A real plane inside a complex line is closed and abelian. A real plane that is not contained in any complex line is closed only when the bracket of two of its spanning vectors stays inside it, and the Borel row is one that does, with $[i\vec e_x,\vec e_y+i\vec e_z]=\vec e_y+i\vec e_z$ equal to the second spanning vector, and with $\kappa$ of signature $(1,0,1)$.
What the table is not. It is not exhaustive. By the theorem of §The Consequence for Subalgebras, the real subalgebras are the pairs (a real subspace of $\mathbb{B}_C$, a closed real subspace of $\mathbb{B}_V$) together with a free lift, and the closed real subspaces of $\mathbb{B}_V$ include members of real dimension $0,1,2,3,4$ and $6$ — never $5$, since a codimension-one subalgebra of a simple algebra would be an ideal. The table names the ones that carry standard names; the rest are the unnamed planes of dimension two and four, and they are of little use.
Why the Larger Algebras Do Not Fit
The Dimension Theorem
By the reduction, the only obstruction is the size of the vector part.
Theorem. Let $\mathrm{G}\subset\mathbb{B}$ be a real Lie subalgebra with $\mathrm{G}\cap\mathbb{B}_C=0$, that is, with trivial centre. Then $\dim_{\mathbb{R}}\mathrm{G}\le6$, and equality holds exactly for $\mathrm{G}=\mathbb{B}_V$.
Proof. The kernel of $\pi_V$ on $\mathrm{G}$ is $\mathrm{G}\cap\mathbb{B}_C$, which vanishes, so $\pi_V$ is injective and $\dim\mathrm{G}\le\dim\mathbb{B}_V=6$. If the dimension is six then $\pi_V$ is an isomorphism onto $\mathbb{B}_V$, and every element of $\mathrm{G}$ is $\vec v+\phi(\vec v)$ for one linear map $\phi:\mathbb{B}_V\to\mathbb{B}_C$. Closure requires the bracket of $\vec u+\phi(\vec u)$ and $\vec v+\phi(\vec v)$, which is $\vec u\times\vec v$, to be the element of $\mathrm{G}$ with that vector part, namely $\vec u\times\vec v+\phi(\vec u\times\vec v)$; so $\phi(\vec u\times\vec v)=0$ for all $\vec u,\vec v$. The cross product is onto $\mathbb{B}_V$, so $\phi=0$ and $\mathrm{G}=\mathbb{B}_V$.
The contrapositive is the working form: every subalgebra of real dimension at least seven has a nonzero centre. Computed on one hundred random six-dimensional subspaces transverse to $\mathbb{B}_C$, every one fails closure, with a smallest defect of $0.33$; $\mathbb{B}_V$ itself closes exactly.
The Exclusions
The theorem disposes of the larger algebras in one stroke, and the compactness settles the two six-dimensional ones.
| algebra | real dimension | centre | verdict |
|---|---|---|---|
| $\mathfrak{su}(n)$, $n\ge3$ | $n^2-1\ge8$ | $0$ | excluded: centreless of dimension above six |
| $\mathfrak{so}(n)$, $n\ge5$ | $n(n-1)/2\ge10$ | $0$ | excluded: the same |
| $\mathfrak{su}(4)\cong\mathfrak{so}(6)$ | $15$ | $0$ | excluded: the same |
| $\mathfrak{so}(4,2)\cong\mathfrak{su}(2,2)$ | $15$ | $0$ | excluded: the same |
| $\mathfrak{so}(4)$ | $6$ | $0$ | excluded: $6$ and centreless forces $\mathbb{B}_V$, and this algebra is compact while $\mathbb{B}_V$ has $\kappa$ of signature $(3,3)$ |
| $\mathfrak{so}(2,2)$ | $6$ | $0$ | excluded: the same, with $\kappa$ of signature $(4,2)$ |
The two six-dimensional exclusions need one further remark, because the theorem alone does not decide them: a centreless subalgebra of dimension six must be $\mathbb{B}_V$ as a subspace, so its isomorphism type is fixed in advance, and neither $\mathfrak{so}(4)$ nor $\mathfrak{so}(2,2)$ is $\mathfrak{so}(3,1)$.
The Centraliser Argument
There is a second and independent proof of the two six-dimensional exclusions, and it is the one worth remembering because it explains why $\mathfrak{so}(4)$ is not merely excluded but impossible.
The centraliser of $\mathbb{B}_V$ in $\mathbb{B}$ is $\mathbb{B}_C$, of real dimension two and abelian; computed as the joint kernel of the six maps $X\mapsto[\vec e_\mu,X]$, its dimension is two, spanned by $e_t$ and $ie_t$. So if a copy of $\mathfrak{su}(2)$ sat inside $\mathbb{B}$, the only elements of $\mathbb{B}$ commuting with it would form a two-dimensional abelian algebra, and there would be no room for a second commuting copy of $\mathfrak{su}(2)$. Since $\mathfrak{so}(4)=\mathfrak{su}(2)\oplus\mathfrak{su}(2)$ has two commuting three-dimensional ideals, and the same holds for $\mathfrak{so}(2,2)=\mathfrak{sl}(2,\mathbb{R})\oplus\mathfrak{sl}(2,\mathbb{R})$, neither can be a subalgebra of $\mathbb{B}$.
The argument is the one to keep, because it generalises: no algebra with two commuting copies of a three-dimensional simple algebra fits inside $M_2(\mathbb{C})$, whatever its dimension.
Lie Algebras of Operators
Left and Right Multiplication
The operators are $\rho_L(\tilde Q):\tilde X\mapsto\tilde Q\tilde X$ and $\rho_R(\tilde Q):\tilde X\mapsto\tilde X\tilde Q$, and they have three properties.
- Each is an algebra map, hence a Lie map. $\rho_L(\tilde Q_1\tilde Q_2)=\rho_L(\tilde Q_1)\rho_L(\tilde Q_2)$ and the same on the right, so $\rho_L(\tilde Q_1\wedge\tilde Q_2)=[\rho_L(\tilde Q_1),\rho_L(\tilde Q_2)]$ and the image of a subalgebra is a subalgebra. Checked on one hundred random pairs, the defect is $2\times10^{-15}$.
- The two sides commute. $\rho_L(\tilde Q_1)\rho_R(\tilde Q_2)=\rho_R(\tilde Q_2)\rho_L(\tilde Q_1)$, because the product is associative. Defect $2\times10^{-15}$ on one hundred pairs.
- The images lie in the regular copies. $\rho_L(\mathbb{B})=\mathbb{B}_L$ and $\rho_R(\mathbb{B})=\mathbb{B}_R$ are the two commuting copies of $M_2(\mathbb{C})$ inside $\mathrm{End}_{\mathbb{C}}(\mathbb{B})$, as Matrix Representations Conventions, §The Left Family and §The Right Family, fixes them.
So for a Lie subalgebra $\mathrm{G}\subset\mathbb{B}$ the sum $\rho_L(\mathrm{G})+\rho_R(\mathrm{G})$ is a Lie algebra of operators, and the two summands are ideals in it. This is the second sense of the expression, and it reaches the algebras of the first sense only when $\mathrm{G}$ is one of the small ones.
The Kernel and the Correction for $\mathfrak{u}(2)$
Proposition. The map $\Phi:\mathrm{G}\oplus\mathrm{G}\to\mathrm{End}_{\mathbb{C}}(\mathbb{B})$, $(X,Y)\mapsto\rho_L(X)+\rho_R(Y)$, is a Lie homomorphism whose kernel is the diagonal copy of $\mathrm{G}\cap\mathbb{B}_C$. Hence it is injective exactly when $\mathrm{G}$ has trivial centre, and in general the image has dimension $2\dim\mathrm{G}-\dim(\mathrm{G}\cap\mathbb{B}_C)$.
Proof. If $\Phi(X,Y)=0$ then applying to $e_t$ gives $X+Y=0$, so $Y=-X$ and $\rho_L(X)-\rho_R(X)=0$, which says that $X$ commutes with every element, that is $X\in\mathbb{B}_C$. Conversely a central $X$ gives $\rho_L(X)=\rho_R(X)$ and $\Phi(X,-X)=0$.
The correction. The working notes list $\mathfrak{u}(2)_L\oplus\mathfrak{u}(2)_R$ of dimension eight beside the other rows of the pattern. The proposition gives seven, not eight, and the reason is that $\mathfrak{u}(2)$ contains the central line $\mathbb{R}(ie_t)$ on which left and right multiplication coincide. Computed as the rank of the eight operators, the image is seven-dimensional; for $\mathfrak{su}(2)$, $\mathfrak{sl}(2,\mathbb{R})$ and $\mathbb{B}_V$ the kernel is zero and the dimensions are six, six and twelve. The correct statement is that $\mathfrak{u}(2)_L+\mathfrak{u}(2)_R$ is the quotient of $\mathfrak{u}(2)\oplus\mathfrak{u}(2)$ by the diagonal central line. This is the one place where the sum has a kernel, and it is worth stating because the quotient is not a direct sum and does not split canonically.
$\mathfrak{so}(4)$ from the Real Pure Quaternions
Take $\mathrm{G}$ the real pure quaternions, with trivial centre, so the sum is direct and six-dimensional. The operators act on the real quaternions $\mathbb{H}\subset\mathbb{B}$ as real $4\times4$ matrices, because left and right multiplication by a real quaternion preserve the real coefficients.
- The six operators are antisymmetric for the Euclidean form of $\mathbb{H}$: left multiplication by a unit pure quaternion is a rotation, and the defect is $0$.
- They are independent, of rank $6$, so they span a six-dimensional subalgebra of $\mathfrak{so}(4)$.
- $\mathfrak{so}(4)$ has dimension six, so the span is all of it.
The Killing form of the span is negative definite, of signature $(0,6)$, as a compact algebra must be. In letters, $\mathfrak{su}(2)_L\oplus\mathfrak{su}(2)_R\cong\mathfrak{so}(4)$, the two factors being the left and the right multiplication, and the algebra acts on $\mathbb{H}\cong\mathbb{R}^4$; extended complex-linearly it acts on $\mathbb{B}\cong\mathbb{C}^4$.
This is the standard quaternionic realisation of $\mathfrak{so}(4)$: the two $\mathrm{SU}(2)$ of the spin cover act by left and right multiplication, and the fact that they commute is the fact that left and right multiplications commute. It is a subalgebra of $\mathfrak{gl}(4,\mathbb{C})_{\mathbb{R}}$ and not of $\mathfrak{gl}(2,\mathbb{C})_{\mathbb{R}}$, which is exactly the content of the previous section.
$\mathfrak{so}(2,2)$ from the Real Traceless
Take $\mathrm{G}=\operatorname{span}_{\mathbb{R}}\{i\vec e_x,\vec e_y,i\vec e_z\}$, again centreless, so the sum is direct and six-dimensional. The operators preserve the real matrices $M_2(\mathbb{R})=\operatorname{span}_{\mathbb{R}}\{e_t,i\vec e_x,\vec e_y,i\vec e_z\}$, on which the determinant is a quadratic form of signature $(2,2)$.
- The six operators are antisymmetric for that form, with defect $0$.
- They are independent, of rank $6$, so they span a six-dimensional subalgebra of $\mathfrak{so}(2,2)$.
- $\mathfrak{so}(2,2)$ has dimension six, so the span is all of it.
The Killing form of the span has signature $(4,2)$, the signature of $\mathfrak{so}(2,2)$, and $\mathfrak{sl}(2,\mathbb{R})_L\oplus\mathfrak{sl}(2,\mathbb{R})_R\cong\mathfrak{so}(2,2)$. The two cases are the same computation with the compact form replaced by the split one, and with $M_2(\mathbb{R})$ in place of $\mathbb{H}$.
The Orthogonal Algebra of the Norm
Take $\mathrm{G}=\mathbb{B}_V$, the traceless biquaternions, of real dimension six and trivial centre, so the sum is direct of dimension twelve. Two facts fix it.
The left copy is complex orthogonal. For traceless $X$ the operator $\rho_L(X)$ is anti-self-adjoint for the norm, $\rho_L(X)^{T}+\rho_L(X)=0$, checked on one hundred random traceless elements with defect $0$; with a nonzero scalar part the defect is $2$, so the condition is exactly tracelessness. The reason is the multiplicativity of the norm: $N((e_t+tX)\tilde Q)=N(e_t+tX)N(\tilde Q)=(1+t^2N(X))N(\tilde Q)$, whose linear term vanishes exactly when $\operatorname{Tr}X=0$. Hence $\rho_L(\mathbb{B}_V)\subset\mathfrak{so}(4,\mathbb{C})$ and $\rho_R(\mathbb{B}_V)\subset\mathfrak{so}(4,\mathbb{C})$, the complex orthogonal algebra of the norm, of real dimension twelve.
The two copies fill it. The twelve operators are independent, of rank $12$, so the sum is a twelve-dimensional subalgebra of a twelve-dimensional algebra, hence the whole: $\rho_L(\mathbb{B}_V)+\rho_R(\mathbb{B}_V)=\mathfrak{so}(4,\mathbb{C})$. Its Killing form has signature $(6,6)$, and each of the two summands alone has signature $(3,3)$.
As an abstract real algebra the sum is therefore $\mathfrak{so}(3,1)\oplus\mathfrak{so}(3,1)$, since each summand is $\mathbb{B}_V$ and $\mathbb{B}_V$ is $\mathfrak{so}(3,1)$; as a subalgebra of $\mathrm{End}_{\mathbb{C}}(\mathbb{B})$ it is the complex orthogonal algebra. The two descriptions agree because $\mathfrak{so}(3,1)\oplus\mathfrak{so}(3,1)$ and $\mathfrak{so}(4,\mathbb{C})$ read over $\mathbb{R}$ are the same real algebra, both being $\mathfrak{sl}(2,\mathbb{C})\oplus\mathfrak{sl}(2,\mathbb{C})$ read over $\mathbb{R}$.
The two factors are the two chiral halves. Complexified, $\mathfrak{so}(3,1)_{\mathbb{C}}=\mathfrak{sl}(2,\mathbb{C})\oplus\mathfrak{sl}(2,\mathbb{C})$ splits into the self-dual and the anti-self-dual part, and left and right multiplication are exactly the two summands. That is the structural reason a two-sided operator $H_A(T)=ATA^{\dagger}$ appears in the corpus wherever a rotation and its mirror are to be handled at once.
Matrices over the Biquaternions
The Two by Two Algebra
Nothing above reaches $\mathfrak{su}(4)$ or $\mathfrak{so}(4,2)$, and the reason is the dimension theorem: the ambient algebra has real dimension eight and these have fifteen. The way out is one size up. A $2\times2$ array over $\mathbb{B}$ is a biquaternion in each of four slots, so
$$ \mathbb{B}^{2\times2}\cong\mathbb{B}\otimes_{\mathbb{C}}M_2(\mathbb{C})\cong M_2(\mathbb{C})\otimes M_2(\mathbb{C})\cong M_4(\mathbb{C}), $$
of real dimension $32$. The isomorphism is the entrywise $\mathsf{M}_2$: the array $Y=(Y_{\lambda\rho})$ maps to the $4\times4$ matrix whose blocks are $\mathsf{M}_2(Y_{\lambda\rho})$.
Two operations come with the array. The star-transpose is $(Y^{*})_{\lambda\rho}=(Y_{\rho\lambda})^{*}$, the star applied entrywise after transposition, and its image is the conjugate transpose: computed on one hundred random arrays, $\mathsf{M}_4(Y^{*})=\mathsf{M}_4(Y)^{\dagger}$ with defect $0$. The trace is the trace of the image, and it reads in coefficients as twice the scalar coefficient of the two diagonal slots,
$$ \operatorname{Tr}Y=\operatorname{tr}\mathsf{M}_4(Y)=2\bigl(Y_{00}+Y_{11}\bigr)_t , $$
because the $2\times2$ image of one slot has trace $2\,\mathrm{Sc}(\tilde Q)=2Q_t$, and the two diagonal slots add. Computed on one hundred random arrays, the defect is $8.9\times10^{-16}$. Note that $\operatorname{Tr}Y$ is complex, so the trace condition is two real conditions on the scalar coefficient, and not the vanishing of the biquaternion $Y_{00}+Y_{11}$.
$\mathfrak{su}(4)\cong\mathfrak{so}(6)$
The anti-Hermitian traceless part of $M_4(\mathbb{C})$ is $\mathfrak{su}(4)$, of real dimension $15$. Transcribed to arrays it is
$$ \mathfrak{su}(4)=\{Y\in\mathbb{B}^{2\times2}:Y^{*}=-Y,\ \operatorname{Tr}Y=0\}, $$
that is: anti-Hermitian for the star-transpose, and with vanishing trace. Computed as the null space of the two conditions, the dimension is $15$ and the conditions hold on the basis to $4\times10^{-16}$; the Killing form is negative definite, of signature $(0,15)$, as a compact algebra must be.
A precision the word "traceless" hides. The condition is on the trace of the image, that is on the scalar part of the sum of the two diagonal slots. Computed on the same basis, $\operatorname{Sc}(Y_{00}+Y_{11})$ vanishes to $2\times10^{-16}$ while the vector part of $Y_{00}+Y_{11}$ does not, reaching $0.82$. So "traceless" here is the trace of the assembly and not the vanishing of the biquaternion $Y_{00}+Y_{11}$; the two differ, and only the first gives fifteen.
The isomorphism $\mathfrak{su}(4)\cong\mathfrak{so}(6)$ of the two real algebras is the exceptional isomorphism of $A_3$ and $D_3$; both are simple of dimension fifteen, and the statement is quoted from Root Systems and Classification.
$\mathfrak{su}(2,2)$ and the Paper's $\mathfrak{so}(4,2)$
Take the form with matrix $G=\operatorname{diag}(e_t,-e_t)$ in the two slots, of image $\operatorname{diag}(1,1,-1,-1)$, and put
$$ \mathfrak{su}(2,2)=\{Y\in\mathbb{B}^{2\times2}:Y^{*}G+GY=0,\ \operatorname{Tr}Y=0\}. $$
The image of the first condition is $X^{\dagger}G+GX=0$ with $G=\operatorname{diag}(1,1,-1,-1)$, because the star-transpose is the dagger and $G$ is carried by the slots. Computed as a null space, the dimension is $15$; the first condition holds on the basis to $2.4\times10^{-16}$; the trace condition holds to $4.3\times10^{-16}$; and the Killing form has signature $(8,7)$, which is the signature of the real form $\mathfrak{su}(2,2)$ of $\mathfrak{sl}(4,\mathbb{C})$, with eight non-compact and seven compact directions.
Without the trace condition the same construction gives $\mathfrak{u}(2,2)$, of real dimension $16$. The pattern is the pattern of the whole section: the involution gives the anti-Hermitian half, and one trace condition removes the centre.
The algebra of the paper. $\mathfrak{so}(4,2)$ is the real form of $\mathfrak{sl}(4,\mathbb{C})$ of dimension $15$ and of signature $(8,7)$, and $\mathfrak{so}(4,2)\cong\mathfrak{su}(2,2)$. The paper's "group algebra $\mathrm{so}(4,2)$" is the universal enveloping algebra of this real Lie algebra, and its twofold cover is $\mathrm{SU}(2,2)$. In biquaternion letters the group is the group of $2\times2$ arrays over $\mathbb{B}$ preserving the form $G$ with the star-transpose as adjoint, and the algebra is the traceless anti-Hermitian part of it.
The one thing this section does not give, and should not pretend to, is a reading of the indefinite form inside the algebra itself. A form of signature $(2,2)$ on $\mathbb{C}^4$ has no biquaternion version other than the pair of slot signs of $G$; the two signs are the two slots, and there is no further structure to read.
The Ladder
The three ambient sizes, and what each hosts, form the ladder of the whole subject.
| ambient algebra | real dimension | algebras hosted |
|---|---|---|
| $\mathbb{B}\cong M_2(\mathbb{C})$ | $8$ | $\mathfrak{su}(2)$, $\mathfrak{sl}(2,\mathbb{R})$, $\mathfrak{u}(2)$, $\mathfrak{gl}(2,\mathbb{R})$, $\mathfrak{so}(3,1)$, and the abelian subspaces |
| $\mathrm{End}_{\mathbb{C}}(\mathbb{B})$ | $32$ | $\mathfrak{so}(4)$, $\mathfrak{so}(2,2)$, $\mathfrak{so}(4,\mathbb{C})_{L+R}$, through left and right multiplication |
| $\mathbb{B}^{2\times2}\cong M_4(\mathbb{C})$ | $32$ | $\mathfrak{su}(4)\cong\mathfrak{so}(6)$, $\mathfrak{u}(2,2)$, $\mathfrak{su}(2,2)\cong\mathfrak{so}(4,2)$ |
| $\mathbb{B}^{n\times n}\cong M_{2n}(\mathbb{C})$ | $8n^2$ | $\mathfrak{su}(2n)$, $\mathfrak{so}(2n)$, and their real forms |
The fourth row is the general statement: a biquaternion matrix of size $n$ is a complex matrix of size $2n$, so every classical algebra of even size has a biquaternion description of this kind, and the description costs nothing and adds nothing beyond the notation.
The Two Senses Compared
The same algebra appears on the two sides in different ways, and the table collects the answers in one place.
| algebra | sense A: inside $\mathbb{B}$ | sense B: operators | real dimension |
|---|---|---|---|
| $\mathfrak{u}(1)$ | $\mathbb{R}(ie_t)$, or any real line in $\mathbb{B}_V$ | — | $1$ |
| $\mathfrak{su}(2)\cong\mathfrak{so}(3)$ | real pure quaternions | left and right give $\mathfrak{so}(4)$ on $\mathbb{H}$ | $3$ |
| $\mathfrak{sl}(2,\mathbb{R})\cong\mathfrak{so}(2,1)$ | $i\vec e_x,\vec e_y,i\vec e_z$ | left and right give $\mathfrak{so}(2,2)$ on $M_2(\mathbb{R})$ | $3$ |
| $\mathfrak{u}(2)$ | $\mathbb{B}_{\Phi}$ | left and right give a seven-dimensional quotient | $4$ |
| $\mathfrak{gl}(2,\mathbb{R})$ | $e_t,i\vec e_x,\vec e_y,i\vec e_z$ | — | $4$ |
| $\mathfrak{so}(3,1)\cong\mathfrak{sl}(2,\mathbb{C})_{\mathbb{R}}$ | $\mathbb{B}_V$, the traceless part | left and right give $\mathfrak{so}(4,\mathbb{C})$ | $6$ |
| $\mathfrak{so}(4)$ | no | $\mathfrak{su}(2)_L\oplus\mathfrak{su}(2)_R$ | $6$ |
| $\mathfrak{so}(2,2)$ | no | $\mathfrak{sl}(2,\mathbb{R})_L\oplus\mathfrak{sl}(2,\mathbb{R})_R$ | $6$ |
| $\mathfrak{gl}(2,\mathbb{C})_{\mathbb{R}}$ | all of $\mathbb{B}$ | $\mathbb{B}_L+\mathbb{B}_R$ modulo the diagonal centre, $14$ | $8$ |
| $\mathfrak{su}(4)\cong\mathfrak{so}(6)$ | no | no: needs $\mathbb{B}^{2\times2}$ | $15$ |
| $\mathfrak{so}(4,2)\cong\mathfrak{su}(2,2)$ | no | no: needs $\mathbb{B}^{2\times2}$ | $15$ |
The pattern is one sentence. The biquaternion algebra is $M_2(\mathbb{C})$, so it hosts the algebras of rank one and their real forms; the algebras of higher rank need a larger ambient algebra, and there are exactly two larger ones that the corpus uses, the operator algebra for the orthogonal algebras and the matrices over $\mathbb{B}$ for the unitary ones.
The Reading inside the Corpus
Three of the identifications above are already working statements in the corpus, and it is worth naming them so that the same object is not read twice.
- The material sector is $\mathfrak{u}(2)$. $\mathbb{B}_{\Phi}$ is the anti-Hermitian subspace, the material sector, and it is $\mathfrak{u}(1)\oplus\mathfrak{su}(2)$: a central phase and a rotation algebra. The rotation algebra of the corpus is the compact form, and it is the real pure quaternions, not the imaginary ones.
- The Lorentz algebra is the traceless part. $\mathbb{B}_V$ is $\mathfrak{so}(3,1)$, the algebra of the boosts and rotations of Biquaternion Rotations and Lorentz Transformations, and it is also the derived algebra and the quotient by the centre. Three descriptions, one subspace.
- The two-sided operators are the two chiral halves. $\rho_L(\mathbb{B}_V)$ and $\rho_R(\mathbb{B}_V)$ are the two $\mathfrak{sl}(2,\mathbb{C})$ of the complexified Lorentz algebra, so a two-sided operator carries a left-handed and a right-handed action at once.
The fourth reading, and the one the paper needs, is the last row of the ladder: $\mathfrak{so}(4,2)$ is a subalgebra of $\mathbb{B}^{2\times2}$ and of nothing smaller. A paper that writes conformal transformations in biquaternions is writing them in $2\times2$ arrays over $\mathbb{B}$, whatever the notation of the moment says.
Verification
A scratch Python script, run outside the repository and deleted after the pass, recomputes each claim on samples of one hundred elements. Every entry is a printed residual, a rank or a count.
| claim | computation | result |
|---|---|---|
| $\vec e_k^{\,2}=-e_t$ and $\vec e_x\vec e_y=\vec e_z$ cyclically | the nine products of the units | exact |
| the bracket is the cross product of the vector parts | $100$ random pairs | $9\times10^{-16}$ |
| $\mathsf{M}_2$ is an algebra map | $100$ random pairs | $3.7\times10^{-15}$ |
| $\mathsf{M}_2(\tilde Q^{*})=\mathsf{M}_2(\tilde Q)^{\dagger}$ | $100$ random elements | $0$ |
| $2\,\mathrm{Sc}(\tilde Q)=2Q_t$ | $100$ random elements | $6.3\times10^{-16}$ |
| $\det\mathsf{M}_2(\tilde Q)=N(\tilde Q)$ | $100$ random elements | $5.6\times10^{-15}$ |
| $\mathfrak{su}(2)$ is closed, $\kappa$ of signature $(0,3)$ | $100$ random pairs, adjoint matrices | defect $0$ |
| $\mathfrak{sl}(2,\mathbb{R})$ is closed, $\kappa$ of signature $(2,1)$ | the same | defect $0$ |
| $\mathbb{B}_V$ is closed, $\kappa$ of signature $(3,3)$ | the same | defect $0$ |
| $\mathbb{B}_{\Phi}$ is closed, $\kappa$ of signature $(0,3,1)$ | the same | defect $0$ |
| $\mathfrak{gl}(2,\mathbb{R})$ is closed, $\kappa$ of signature $(2,1,1)$ | the same | defect $0$ |
| $\mathbb{B}_{\Psi}$ is not closed | $100$ random pairs | defect $6.8$ |
| $\mathsf{M}_2(\mathbb{B}_{\Phi})$ is the anti-Hermitian set | $100$ random elements, $\lVert M^{\dagger}+M\rVert$ | $0$ |
| the centraliser of $\mathbb{B}_V$ is $\mathbb{B}_C$ | joint kernel of six maps, rank | dimension $2$ |
| random $6$-dimensional subspaces transverse to $\mathbb{B}_C$ are not closed | $100$ random graphs over $\mathbb{B}_V$ | $100$ of $100$, smallest defect $0.33$ |
| a complex plane is closed exactly when it contains its normal | $100$ random complex planes; then the Borel | $100$ of $100$ fail, smallest defect $0.71$; the Borel closes to $9\times10^{-16}$ |
| $\rho_L$ is a Lie map and the two sides commute | $100$ random pairs | $2\times10^{-15}$ |
| the kernel of $(X,Y)\mapsto\rho_L(X)+\rho_R(Y)$ | ranks of the operator spans | $0,0,0$ for $\mathfrak{su}(2)$, $\mathfrak{sl}(2,\mathbb{R})$, $\mathbb{B}_V$; $1$ for $\mathfrak{u}(2)$, whose image is $7$ |
| $\rho_L(X)$ is anti-self-adjoint for $N$ exactly when $X$ is traceless | $100$ random traceless $X$; then $Q_t=1$ | $0$; then $2$ |
| the six operators of $\mathfrak{su}(2)_L\oplus\mathfrak{su}(2)_R$ on $\mathbb{H}$ | rank, antisymmetry, Killing form | rank $6$, defect $0$, signature $(0,6)$ |
| the six operators of $\mathfrak{sl}(2,\mathbb{R})_L\oplus\mathfrak{sl}(2,\mathbb{R})_R$ on $M_2(\mathbb{R})$ | the same, for the determinant form | rank $6$, defect $0$, signature $(4,2)$; the form has signature $(2,2)$ |
| the twelve operators of $\mathbb{B}_{V,L}+\mathbb{B}_{V,R}$ | rank, Killing form | rank $12$, signature $(6,6)$; the left copy alone has $(3,3)$ |
| $\mathsf{M}_4(Y^{*})=\mathsf{M}_4(Y)^{\dagger}$ | $100$ random arrays | $0$ |
| $\operatorname{tr}\mathsf{M}_4(Y)=2(Y_{00}+Y_{11})_t$ | $100$ random arrays | $8.9\times10^{-16}$ |
| $\mathfrak{su}(4)$ has dimension $15$ and $\kappa$ of signature $(0,15)$ | null space of the two conditions | dimension $15$, residual $4\times10^{-16}$ |
| $\mathfrak{su}(2,2)$ has dimension $15$ and $\kappa$ of signature $(8,7)$ | the same for the form $G$ | dimension $15$, residual $2\times10^{-16}$ |
| $\mathfrak{u}(2,2)$ has dimension $16$ | the same without the trace condition | dimension $16$ |
| the trace condition is on the scalar part | $\operatorname{Sc}(Y_{00}+Y_{11})$ and its vector part on the basis | $2\times10^{-16}$; the vector part reaches $0.82$ |
Summary
A Lie algebra is expressed in biquaternions in one of two senses, and the two senses answer differently.
In the first sense the algebra is a subspace of $\mathbb{B}$ under the bracket. Since that bracket is the cross product of the vector parts, the whole question reduces to a real subspace of $\mathbb{B}_V$ closed under the cross product, with a free scalar part adjoined. Over $\mathbb{C}$ this gives eight subalgebras and no new algebra: $\mathbb{B}$ is $M_2(\mathbb{C})$. Over $\mathbb{R}$ it gives the compact, the split and the complex readings of $\mathfrak{sl}(2,\mathbb{C})$ as three subspaces of one carrier, together with $\mathfrak{u}(2)$ on the anti-Hermitian sector, $\mathfrak{gl}(2,\mathbb{R})$, $\mathfrak{gl}(2,\mathbb{C})_{\mathbb{R}}$, and the abelian subspaces. The compact form is the real pure quaternions and the Hermitian sector is not a Lie algebra at all.
The dimension theorem closes the first sense: a centreless subalgebra has real dimension at most six, and the only one of dimension six is the traceless part. Every algebra of dimension seven or more therefore has a centre, and $\mathfrak{so}(4)$ and $\mathfrak{so}(2,2)$, which are centreless of dimension six, would have to be $\mathfrak{so}(3,1)$, which they are not. The centraliser of the traceless part, which is the centre, gives the same two exclusions in one line.
In the second sense the algebra is a set of operators. Left and right multiplication commute and are Lie maps, so a subalgebra yields two commuting copies; the sum is direct exactly when the subalgebra has trivial centre, which is why $\mathfrak{u}(2)$ gives seven and not eight. On the real quaternions the two copies of $\mathfrak{su}(2)$ give $\mathfrak{so}(4)$; on the real matrices the two copies of $\mathfrak{sl}(2,\mathbb{R})$ give $\mathfrak{so}(2,2)$; and on the whole carrier the two copies of the traceless part give the complex orthogonal algebra $\mathfrak{so}(4,\mathbb{C})$, which is $\mathfrak{so}(3,1)\oplus\mathfrak{so}(3,1)$ read over $\mathbb{R}$ and is the two chiral halves.
The algebras of dimension fifteen need one size up. Two-by-two arrays over $\mathbb{B}$ are $M_4(\mathbb{C})$, and in them the anti-Hermitian traceless part is $\mathfrak{su}(4)\cong\mathfrak{so}(6)$ and the anti-Hermitian traceless part of the form $\operatorname{diag}(e_t,-e_t)$ is $\mathfrak{su}(2,2)\cong\mathfrak{so}(4,2)$, the algebra of the paper. The trace in that description is the trace of the assembly, which is the scalar part of the sum of the diagonal slots, and not the vanishing of the biquaternion.
Summary of Notation
| symbol | meaning |
|---|---|
| $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ | the biquaternion algebra, $\cong M_2(\mathbb{C})$ |
| $e_t,\vec e_x,\vec e_y,\vec e_z$ | the base, with $\vec e_k^{\,2}=-e_t$ and $\vec e_x\vec e_y=\vec e_z$ |
| $\tilde Q=Q_te_t+\vec Q$ | the element, its scalar part, its vector part |
| $\tilde Q_1\wedge\tilde Q_2$ | the bracket $\tfrac12(\tilde Q_1\tilde Q_2-\tilde Q_2\tilde Q_1)$, the cross product of the vector parts |
| $[\ ,\ ]$ | the same bracket, and for operators $\tfrac12(AB-BA)$ |
| $\mathrm{Sc}(\tilde Q)=Q_t$ | the scalar part, so that traceless means $Q_t=0$; on arrays, the trace of the image |
| $\tilde Q^{*}$ | the star, $\bar{\cdot}\circ{}^{\natural}$ |
| $\mathbb{B}_C$, $\mathbb{B}_V$ | the centre $\mathbb{C}e_t$ and the vector subspace, the fixed and anti-fixed spaces of ${}^{\natural}$ |
| $\mathbb{B}_{\Phi}$, $\mathbb{B}_{\Psi}$ | the anti-Hermitian and the Hermitian subspaces |
| $\mathsf{M}_2(\tilde Q)$ | the $2\times2$ complex image |
| $\rho_L$, $\rho_R$ | left and right multiplication on the carrier |
| $\mathbb{B}_L$, $\mathbb{B}_R$ | the two commuting copies inside $\mathrm{End}_{\mathbb{C}}(\mathbb{B})$ |
| $N(\tilde Q)=\sum_\mu Q_\mu^2$ | the norm, equal to $\det\mathsf{M}_2(\tilde Q)$ |
| $\kappa$ | the Killing form |
| $\mathrm{G}$ | a generic Lie algebra |
| $\pi_V$ | the projection onto the vector part along the centre |
| $\mathbb{B}^{2\times2}\cong M_4(\mathbb{C})$ | the arrays, with the star-transpose and the trace of the assembly |
| $G=\operatorname{diag}(e_t,-e_t)$ | the form of $\mathfrak{su}(2,2)$ in the two slots |
Further Reading
- Algebra and Element Conventions, for the base, the coefficients, the centre, the vector subspace and the four conjugations.
- Matrix Representations Conventions, for $\mathsf{M}_2$, the Pauli frame, the two regular copies and the rule on Pauli and gamma matrices.
- The Lie Structure of the Antisymmetric Plain Product, for the bracket on $\mathbb{B}$, its ideals, its radical and its Killing form.
- Lie Algebras, for the vocabulary, the Jacobi identity and the bracket as the second-order term of the group law.
- Structure of Lie Algebras, for the Killing form, Cartan's criteria and the decomposition into simple ideals.
- Real Forms of a Complex Lie Algebra, for the conjugations, the compact and split forms, and the classification by Satake diagrams.
- The Adjoint Action of a Lie Algebra, for $\operatorname{ad}$, its kernel and its image.
- Biquaternion Lie Group and Exponential Structure, for the exponential, the subgroups and the spin cover.
- Biquaternion Rotations and Lorentz Transformations, for the Lorentz algebra in the corpus's own coordinates.
- Coda: $n=2$ and the Biquaternion Algebra, for the dimensions and signatures of the four sectors.
- The Exponential and the Lie Group of $M_4(\mathbb{C})$, for the algebras of size four that the arrays over $\mathbb{B}$ become.
- The Clifford Algebra as a Lie Algebra, for the same brackets read in the Clifford language.