Operator Chirality on the Biquaternion Algebra
Introduction
Every operator on the biquaternion algebra multiplies on a side, and the side is given by the multiplication itself: $L(a)$ multiplies an element on the left, $R(b)$ multiplies it on the right. Operator chirality is the name of that side. The definition is one line long, it names no frame, and everything else in this article follows from it.
The obvious route to the notion goes through a single basis operator, and it is worth saying why that route misleads. Pick an element $u$ with $u^2=-e_t$ and no scalar part, and form the right multiplier $-iR_u$. Its square is the identity, its eigenvalues are $\pm1$ with two of each, and its eigenspaces are two halves of the algebra. The construction is correct, and the physics uses it. But its order is wrong: it produces an operator and then reads a chirality off it. The error shows in the product. The basis operator so obtained is a pure right multiplier, so by any definition of chirality worth the name it has a chirality of its own; and an object that has a chirality cannot be what chirality means. Here the sides come first, and that basis operator becomes one example of one of them.
Scope of the word. This article is about operator chirality, the side on which an operator multiplies. The bare word chirality in the physics reads the spinor chirality, the value an element takes under the chirality operator, and that notion belongs to the carrier entries. The two are different objects: a side is a membership in $L(\mathbb{B})\cup R(\mathbb{B})$ and needs no axis, a value is an eigenspace and needs one. Nothing below uses the second.
The article is purely algebraic. The algebra is $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$, central simple over $\mathbb{C}$, of complex dimension four, with unit $e_t$ and reduced trace $\operatorname{Sc}$; the identity operator is $\mathrm I$. A basis of $\mathbb{B}$ is chosen twice and only twice: in the section on the trace form, where an array is the convenience being discussed, and in one example about dimension, where the example says so. Nowhere else is a basis chosen. The proof of the structure isomorphism is quantified over all bases and picks none, so it counts as no choice. The basis operators and the two families belong to Families of Matrix Representations. The carriers and their splittings belong to The Unpolarized Module of the Biquaternion Algebra. The ambient operator space belongs to The Endomorphisms of the Biquaternion Algebra.
The Two Sides
The multiplication gives two maps into the endomorphism algebra, one for each end of a product,
$$ L\colon\mathbb{B}\to\operatorname{End}_{\mathbb{C}}(\mathbb{B}),\quad L(a)(\tilde X)=a\tilde X, \qquad R\colon\mathbb{B}\to\operatorname{End}_{\mathbb{C}}(\mathbb{B}),\quad R(b)(\tilde X)=\tilde Xb . $$
Both are complex-linear, both are unital, and both are injective, because a multiplier that vanishes on the unit vanishes everywhere: $L(a)(e_t)=a$. The two families are written $L(\mathbb{B})$ and $R(\mathbb{B})$.
They commute. That is the first structural fact and the reason the whole subject exists:
$$ L(a)R(b)(\tilde X)=a\tilde Xb=R(b)L(a)(\tilde X). $$
The left factor and the right factor act on the two ends of the argument, so they cannot meet. Products of basis operators multiply accordingly,
$$ \bigl(L_aR_b\bigr)\bigl(L_cR_d\bigr)=L_{ac}R_{db}, $$
with the reversal on the right, because the two right parameters meet in the middle and the one that was farther from the argument is now applied first.
The two images are four-dimensional, since $L$ and $R$ are injective on a four-dimensional algebra. Their intersection is the line of scalars and nothing else. If $L(a)=R(b)$ then $a\tilde X=\tilde Xb$ for every $\tilde X$; at $\tilde X=e_t$ this gives $a=b$, and then $a$ commutes with everything, so $a$ is central. A central element gives the equality back, so the intersection is exactly $\mathbb{C}\,\mathrm I$. Their sum is therefore seven-dimensional, inside the sixteen-dimensional endomorphism algebra:
$$ \dim L(\mathbb{B})=\dim R(\mathbb{B})=4,\qquad \dim\bigl(L(\mathbb{B})\cap R(\mathbb{B})\bigr)=1,\qquad \dim\bigl(L(\mathbb{B})+R(\mathbb{B})\bigr)=7 . $$
These numbers are worth reading twice. The second says that the two sides overlap in a line, and the fourth says that nine of the sixteen dimensions lie outside the sum: those are the operators that multiply on neither side alone.
The two families are not two constructions of one object. They are the two factors of a tensor square. The multiplication defines
$$ \Xi\colon\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\to\operatorname{End}_{\mathbb{C}}(\mathbb{B}),\qquad \Xi(a\otimes b)=L_aR_b , $$
and $\Xi$ is an isomorphism, because for any basis $e_1,\dots,e_4$ the sixteen basis operators $L_{e_\mu}R_{e_\nu}$ are independent: an operator on a four-dimensional space is an arbitrary array of sixteen numbers, and the basis operators supply the sixteen coordinate directions. Under $\Xi$ the two sides become the two factors, and their overlap becomes the tensor of the two units:
$$ L(\mathbb{B})=\Xi(\mathbb{B}\otimes e_t),\qquad R(\mathbb{B})=\Xi(e_t\otimes\mathbb{B}),\qquad L(\mathbb{B})\cap R(\mathbb{B})=\Xi(e_t\otimes e_t). $$
This is the technical core of the article. A basis operator $L_aR_b$ is a coordinate of $\Xi$ and not a primitive object, and a statement written in basis operators is a statement about $\Xi$ which becomes intrinsic once it is written in the two images.
Definition. An operator has the left slot when it lies in $L(\mathbb{B})$, the right slot when it lies in $R(\mathbb{B})$, both slots when it lies in the intersection, and no single side otherwise:
$$ X\in L(\mathbb{B}),\qquad X\in R(\mathbb{B}),\qquad X\in L(\mathbb{B})\cap R(\mathbb{B})=\mathbb{C}\,\mathrm I,\qquad X\notin L(\mathbb{B})\cup R(\mathbb{B}). $$
Operator chirality is that label. It is a property of an operator, it takes four values, and it is fixed by the multiplication before anything else is chosen.
The third case is the one that needs care. A basis operator such as $L_aR_b$ with neither parameter central occupies both slots in the plain sense of the words, and it has not both slots in the sense of the definition: it has no single side. The word both names two different things in the two sentences, a label and a family, and the corpus keeps them apart by grammar. The basis operator $L_aR_b$ is a member of the family $LR$; its label is no single side. The third case is reserved for the operators that genuinely multiply on both ends at once, and those are the scalars.
The asymmetry appears the moment one multiplies. The product of two left multipliers is a left multiplier, and the product of two right multipliers is a right multiplier, so each side is a four-dimensional subalgebra and the two subalgebras meet in the scalars. A product of one of each is a basis operator, and a general basis operator is on neither side. The label is therefore not respected by the product, except on the scalars, and this is the first hint that it is not a grading; the last section makes that a theorem.
The label is also a name and not a quantity. It is not additive: $L_a+L_c$ has the left slot while $L_a+R_b$ has none, so the set of operators with no single side is the complement of a union and not a subspace. It carries no sign: $cL_a$ and $-cL_a$ are labelled alike. No linear functional on the endomorphism algebra assigns the four labels, for the reason given at the end.
The Swap
The two sides are exchanged by one canonical involution, and that involution does the work the older reading asked of a single multiplier built from a chosen element. It is defined on the tensor square and transported by $\Xi$. Let $\tau$ be the flip of the two factors, $\tau(a\otimes b)=b\otimes a$, and put
$$ \sigma=\Xi\circ\tau\circ\Xi^{-1},\qquad\text{so that}\qquad \sigma(L_aR_b)=L_bR_a . $$
The rule exchanges the two entries of an elementary tensor; it refers to no basis. It is well defined because the basis operators are a basis of the endomorphism algebra, so every operator has one coefficient array and the rule permutes its two indices.
Theorem. The map $\sigma$ is an involution and an anti-automorphism, and it exchanges the two images:
$$ \sigma^2=\mathrm{id},\qquad \sigma(XY)=\sigma(Y)\sigma(X),\qquad \sigma\bigl(L(\mathbb{B})\bigr)=R(\mathbb{B}),\qquad \sigma\bigl(R(\mathbb{B})\bigr)=L(\mathbb{B}). $$
Proof. The flip is its own inverse, so $\sigma$ is. For the reversal, take two basis operators:
$$ \sigma\bigl((L_aR_b)(L_cR_d)\bigr)=\sigma(L_{ac}R_{db})=L_{db}R_{ac}=(L_dR_c)(L_bR_a)=\sigma(L_cR_d)\,\sigma(L_aR_b), $$
and linearity extends it. The exchange of the images is the case of one factor, $\sigma(L_a)=R_a$ and $\sigma(R_b)=L_b$, and linearity extends that too. On the intersection the two statements agree, so the scalars are fixed.
The one case to remember is the case of a single factor:
$$ \sigma(L_a)=R_a\qquad\text{for every }a\in\mathbb{B}. $$
The left multiplier of an element and the right multiplier of the same element are the same operator written on the two ends, and the swap is what changes the writing. This is why the two families have arrays that agree up to a sign, and it is why every statement about a left multiplier has a twin statement about a right one.
Two further properties are used later. The swap permutes the basis operators of any basis, so it preserves the number of one-sided factors in a decomposition, which is the length $\ell$ of the operator articles. It also preserves the rank, since it transposes the coefficient matrix:
$$ \ell\bigl(\sigma(M)\bigr)=\ell(M),\qquad \operatorname{rank}\sigma(M)=\operatorname{rank}M . $$
And it is not inner. An inner automorphism $X\mapsto \mathsf SX\mathsf S^{-1}$ preserves products, while $\sigma$ reverses them, so no matrix conjugates the swap. The exchange of the sides cannot be produced by choosing a frame, changing a display or picking a basis. It is an operation on operators.
The Trace Form
An exchange is easy to define and hard to compute. The trace form supplies both a computation and a second description of $\sigma$ that is manifestly independent of every choice:
$$ \beta(\tilde X,\tilde Y)=\operatorname{Sc}(\tilde X\tilde Y). $$
The form is complex-bilinear, symmetric and nondegenerate, since the reduced trace is nondegenerate on a central simple algebra, and it is cyclic in the sense that $\operatorname{Sc}(\tilde X\tilde Y\tilde Z)=\operatorname{Sc}(\tilde Y\tilde Z\tilde X)$. Nothing is chosen here: the form is built from the multiplication and the trace.
Theorem. For every operator $X$ and all elements $\tilde Y,\tilde Z$,
$$ \beta\bigl(X\tilde Y,\tilde Z\bigr)=\beta\bigl(\tilde Y,\sigma(X)\tilde Z\bigr). $$
The swap is the adjoint of $X$ with respect to $\beta$.
Proof. For a basis operator, the left side is $\operatorname{Sc}(a\tilde Yb\tilde Z)$ and the right side is $\operatorname{Sc}(\tilde Yb\tilde Za)$, and cyclicity makes them equal. Linearity extends the identity to all operators.
The identity is worth pausing on. Adjointness is a basis-free notion, so the theorem describes $\sigma$ using the multiplication and the trace and nothing else, and it says what the exchange of sides is in the language a bilinear form provides. The reading of the same fact in a display is an array, and an array has to be written in a basis. Here it is. In a basis of the carrier where $\beta$ has Gram matrix $\mathsf W$ the adjoint of a matrix is $\mathsf W^{-1}[X]^{\mathsf T}\mathsf W$, so
$$ [\sigma(X)]=\mathsf W^{-1}[X]^{\mathsf T}\mathsf W, \qquad [R(b)]=\mathsf W^{-1}[L(b)]^{\mathsf T}\mathsf W . $$
The second display is the one-factor case, and it is the reason the two families appear as transposes of one another up to a change of basis. The matrix $\mathsf W$ is the one the corpus calls the switch. In a general carrier basis it is the Gram matrix of $\beta$, and there $\mathsf W^2=\mathrm I$ fails, so the symmetric form $\mathsf W(\cdot)^{\mathsf T}\mathsf W$ is not the general rule. In the reference frame $\mathsf W$ is $\mathsf D=\operatorname{diag}(1,-1,-1,-1)$, where the two forms agree because $\mathsf D^2=\mathrm I$, and there the rule reads $[R(b)]=\mathsf D\,[L(b)]^{\mathsf T}\mathsf D$, which is the shape quoted most often.
The array belongs to the basis and not to the algebra, and the dependence cancels. A change of carrier basis replaces $\mathsf W$ by $\mathsf P^{\mathsf T}\mathsf W\mathsf P$ and $[X]$ by $\mathsf P^{-1}[X]\mathsf P$, and the two changes undo each other; that cancellation is the whole reason a computation may be done in a convenient basis.
On the sixteen coefficients of the basis operators the statement is shorter still. The array of $\sigma(X)$ is the transpose of the array of $X$, in every frame, because $\sigma$ matches the basis operator $L_aR_b$ with the basis operator $L_bR_a$ and does nothing else. The transpose is the intrinsic content, and the matrix $\mathsf W^{-1}(\cdot)^{\mathsf T}\mathsf W$ is what the transpose becomes once a carrier basis is imposed.
The Two Eigenspaces
An involution splits the algebra it acts on, and this one splits it into a fixed part and a negated part. The two parts have dimensions ten and six, and each is closed under an operation of its own.
The fixed operators are the self-adjoint ones, $\sigma(X)=X$, and they are the image under $\Xi$ of the symmetric square:
$$ \operatorname{Fix}(\sigma)=\Xi\bigl(\operatorname{Sym}^2\mathbb{B}\bigr),\qquad \dim=10 . $$
The negated operators are the anti-self-adjoint ones, $\sigma(X)=-X$, and they are the image of the exterior square:
$$ \operatorname{Neg}(\sigma)=\Xi\bigl(\Lambda^2\mathbb{B}\bigr),\qquad \dim=6 . $$
The second space is spanned by the six differences $L_aR_b-L_bR_a$, which are the commutators $[L_a,R_b]$ because the two regular maps commute. With $10+6=16$ the two spaces account for the algebra.
Neither is closed under composition. For two fixed operators, $\sigma(XY)=YX$, which equals $XY$ only when the two commute. The fixed space closes under the anti-commutator instead, since
$$ \sigma\bigl(\tfrac12(XY+YX)\bigr)=\tfrac12(YX+XY)=\tfrac12(XY+YX), $$
so it is a Jordan algebra of dimension ten. The negated space closes under the commutator, since
$$ \sigma(XY-YX)=\sigma(Y)\sigma(X)-\sigma(X)\sigma(Y)=YX-XY=-(XY-YX), $$
so it is a Lie algebra of dimension six. Both computations are one line and both use only the anti-automorphism property; the involution decides which operation survives, and that is the sense in which the exchange of sides and the Jordan–Lie split are one statement. The decomposition of an endomorphism algebra into a symmetric and an exterior square is standard for a nondegenerate symmetric form. What is specific here is that the form is the trace form of the algebra and the reflection is the exchange of the two sides, so the split is read off the two-sided structure rather than imposed on it.
A caution belongs here, because a dimension alone says nothing. The example needs a basis and uses the coefficient basis. Take the six basis operators $L_{e_\mu}R_{e_\nu}$ over the pairs $\mu<\nu$, and let $\mathsf S$ be their span. The six are independent, so $\mathsf S$ is six-dimensional like $\operatorname{Neg}(\sigma)$. The swap does not preserve it, and it is worth seeing how badly: it sends each of the six basis operators to a basis operator outside the span, the six images span the six complementary basis operators, and $\mathsf S\cap\sigma(\mathsf S)=\{0\}$ while $\mathsf S+\sigma(\mathsf S)$ is twelve-dimensional. Two spaces of the same dimension, one fixed by the multiplication and one by a choice of six basis operators, and the second is not merely displaced by the swap, it is carried off it entirely. Equal dimensions never identify two subspaces, and a count of one-sided factors is not a membership.
Conjugation Exchanges the Sides
The swap exchanges the sides and leaves the elements alone. A second map exchanges the sides and moves the elements, and it is the one the physics reaches for.
The quaternion conjugation $\natural$ fixes the unit and negates the vector part, $e_t^{\natural}=e_t$ and $u^{\natural}=-u$ for $u$ in the three-dimensional vector subspace. It fixes the coefficients, so it is complex-linear, and the sign on the three vectors makes it an anti-automorphism:
$$ (\tilde X\tilde Y)^{\natural}=\tilde Y^{\natural}\tilde X^{\natural}. $$
Its fixed space is the line of scalars and its anti-fixed space is the vector subspace; the dimensions $1$ and $3$ add to four.
Theorem. For every $a\in\mathbb{B}$,
$$ \natural\,L(a)\,\natural=R\bigl(a^{\natural}\bigr). $$
Proof. Apply $\natural$ to the two ends of $a\natural(\tilde Y)$ and use that it reverses products:
$$ \natural\bigl(L(a)\natural(\tilde Y)\bigr)=\natural\bigl(a\tilde Y^{\natural}\bigr)=\tilde Y a^{\natural}, $$
which is right multiplication by $a^{\natural}$. No basis is used.
The two side-exchangers are therefore different maps, and the difference is the interesting part. The swap exchanges the slots and keeps the parameters: $\sigma(L_aR_b)=L_bR_a$. The conjugation exchanges the slots after conjugating the parameters and reversing their order: $L_aR_b\mapsto R_{a^{\natural}}L_{b^{\natural}}$. On a pure left multiplier the two agree only when $a^{\natural}=a$, that is only on the scalars. The agreement is as small as it can be. Away from the scalars they differ by the sign of the three vector units, and that sign is where every phase in the corpus that comes from a change of side is generated. It is one identity, not a family of coincidences.
The division of labour is then clear. The swap tells you which slot an operator occupies, and it is the map that belongs to the definition. The conjugation tells you when two operators of opposite slots have arrays that agree up to a sign, and it is the map that belongs to the display. Neither is inner, and neither replaces the other.
What Operator Chirality Is Not
Eleven objects are routinely offered as the home of chirality. Each is a genuine object with a genuine use, and none is needed here. The reason repeats: the label is a membership in $L(\mathbb{B})$, in $R(\mathbb{B})$, or in neither, with the scalar line for the fourth case, so it requires the two images and nothing else. Every object in the list lives in a space other than the one the label lives in, or in the same space at a later stage.
| Object | What it is | Why it is not the label |
|---|---|---|
| a root | an element $\tilde Z$ with $\tilde Z^2=e_t$ | the label uses no element |
| an idempotent | an element with $\tilde\Pi^2=\tilde\Pi$ | a condition on an element, not on an operator |
| a Peirce decomposition | four lines cut by two idempotents | the label's four cases are not lines of the algebra |
| a Peirce basis | a basis of the algebra | a basis is a display; the images exist before one is declared |
| a spectrum | the list $+1,+1,-1,-1$ of one operator | a list of numbers is a consequence of a chosen basis operator |
| a spinor | an element of a minimal left ideal | a subspace of $\mathbb{B}$; the label is a membership in $\operatorname{End}(\mathbb{B})$ |
| the module $S$ | one minimal left ideal, a spinor module | needs a point of a projective line to be chosen |
| an ideal | a subspace $I$ with $\mathbb{B}I\subseteq I$ | a subspace of the algebra, and the label names no subspace |
| a Clifford algebra | the algebra with $\{e_i,e_j\}=2g_{ij}$ | fixes a dimension and an inertia; the images do not |
| a matrix unit of $M_4(\mathbb{C})$ | a basis element $E_{ij}$ of a display | the basis operator is a sum of four of them, and the display is not the algebra |
| the basis operator itself | one right multiplier $-iR_u$ | an example of the right slot, and an example is not the definition |
A root is an element $\tilde Z$ with $\tilde Z^2=e_t$, and it is what a Peirce decomposition is built from. It enters after the label, not before: the label depends on $L$ and $R$, which the multiplication supplies, while a root is a further choice made when a splitting of the algebra is wanted. An idempotent is a condition on an element, $\tilde\Pi^2=\tilde\Pi$; the label is a condition on an operator, a membership, and the two conditions have different shapes and different carriers.
A spectrum is offered in the same way, and it is the offer a physicist is likeliest to accept: chirality is said to be the pair of eigenvalues $+1$ and $-1$. The pair is a theorem about the basis operator, and the theorem needs two hypotheses, $u^2=-e_t$ and $\operatorname{Sc}u=0$. A theorem with hypotheses about a chosen element cannot be the definition of a label that needs no element. The comparison of shapes settles it: the label has four cases and a spectrum is a finite list of numbers in $\mathbb{C}$; a list cannot carry four cases, and two of the four have no spectrum at all. What the spectrum does supply is the balance, two of each sign, and that balance is why the basis operator is usable as a projector once it has been chosen.
A Peirce decomposition takes two idempotents and returns four lines of $\mathbb{B}$. The label returns four cases on the operator space, and not one of those cases is a line of the algebra. A Peirce basis is a basis, and the label is read without one: the images of $L$ and $R$ exist before any basis is declared, and they do not move when a basis changes.
A spinor is an element of a minimal left ideal. The objection here is one of space, and it is the objection that matters most, because the temptation to turn chirality into a state space is strong. A spinor is an element of $\mathbb{B}$; the label is a membership in $\operatorname{End}_{\mathbb{C}}(\mathbb{B})$, whose elements are operators. The two spaces are not the same space and not even of the same dimension, $4$ against $16$, and no canonical map carries one into the other. A membership in a subspace of the operators cannot be read off a vector.
The module $S$ is that same objection with a name on it: $S$ is one minimal left ideal, the spinor module, and $S^{\mathrm{c}}$ is its conjugate. Two spinor modules are labelled by two points of a projective line, and no point is canonical (The Unpolarized Module of the Biquaternion Algebra). A definition that required $S$ would inherit that choice before saying anything about an operator. The label has no such inheritance: $L$ and $R$ are defined by the multiplication of the algebra, they exist for every algebra, and no point of any line is chosen to state them.
An ideal is a subspace $I$ of the algebra with $\mathbb{B}I\subseteq I$, that is, a subspace closed under the multiplication. It is unrelated to the label, and the relation is not merely unnecessary but impossible. An ideal is a subset of a four-dimensional complex space; the label is a property of an element of a sixteen-dimensional one. Nothing in the definition of the label mentions a subspace of $\mathbb{B}$, no statement about the label is proved by exhibiting one, and the four cases of the label do not partition any ideal. Ideals remain the right tool for other questions, which is why the corpus has articles about them; what they are not is a home for chirality, and a proof that some ideal is chiral would be a confusion of the two spaces rather than a theorem.
A Clifford algebra is the algebra generated by vectors with $\{e_i,e_j\}=2g_{ij}$, and it is offered as the natural home of chirality with a product of generators as the chirality element. Two reasons settle it. The relation fixes squares and anticommutators and says nothing about the side on which an operator multiplies, which is the only datum the label uses. And a Clifford algebra fixes a dimension and an inertia, while the two images exist for every algebra in every dimension: $L$ and $R$ are defined by the multiplication, and their intersection is the centre, always. The Clifford reading is a display of the algebra, and it is a correct display; it is not a foundation.
A matrix unit of $M_4(\mathbb{C})$ is the last of the display artefacts, and it is the one that looks most like a definition, because in a display the basis operator is a diagonal of signs and a matrix unit is the smallest thing a display has. The basis operator is not invertible of that basis: written in any matrix display it is a sum of four matrix units, and it is one operator, not a combination of four. The number $4$ belongs to the display and not to the algebra, while the two images belong to the algebra and to every algebra. A definition that named a matrix unit would be a definition of one display, and it would not survive the change of display that leaves every operator label where it was.
The basis operator the older reading took for the definition is the case that matters, and it repays a careful look. That object is a right multiplier $-iR_u$ for an element $u$ with $u^2=-e_t$ and $\operatorname{Sc}u=0$. Three properties follow in a line each. Since $R_u^2=R_{u^2}=-\mathrm I$, the basis operator squares to $\mathrm I$. Its trace is zero, because $u$ is traceless. Its spectrum is $+1,+1,-1,-1$, so the projectors $\tfrac12(\mathrm I\pm iR_u)$ have rank two apiece and the split is balanced. A basis operator with square one, trace zero and a balanced spectrum is exactly what a chirality element should be, and it is a pure right multiplier. That is the whole point. Under the definition of this article the basis operator has the right slot; it is an example of one of the four labels, and an example cannot be the definition of the label. What it also is, and this is its limitation rather than its virtue, is an object built from a chosen element: choose another $u$ and the basis operator changes, while every label of every operator stays where it was.
One hypothesis must not be dropped in that computation, because it is false without it. The statement that $R_u$ has spectrum $+i,+i,-i,-i$ and trace zero for any $u$ with $u^2=-e_t$ fails at the central element $ie_t$, which satisfies the square condition and gives $R_{ie_t}=i\mathrm I$, a spectrum of $+i$ four times and a trace of $4i$. Tracelessness is the missing hypothesis, and it is what makes $R_u$ behave as a chirality element rather than as a scalar in disguise.
The basis operator acts on the algebra as well as on the operator space, and that action is where the two component letters come from. For a scalar-free $u$ with $u^2=-e_t$ every element splits,
$$ \tilde X=\tilde X_L+\tilde X_R,\qquad R_u\tilde X_L=i\tilde X_L,\qquad R_u\tilde X_R=-i\tilde X_R, $$
with each part of complex dimension two and no element but zero in both. A product takes the letter of its right factor, since $R_u(\tilde X\tilde Y)=\tilde X R_u(\tilde Y)$, and a left multiplication therefore keeps the letter. The letters belong to this basis operator, and they belong to the element rather than to the operator: a different $u$ gives a different pair of halves. That is why the split is a property of the chosen element, and why the operator label, which needs no element at all, is the more primitive notion.
Which operators keep a letter, and which change it. The rule above settles the transport of the letters for the whole of the algebra, and the answer separates the two slots cleanly. Every left multiplier keeps the letter, so $P_R\,L(a)\,P_L=0$ for every $a$ and every traceless $u$ with $u^2=-e_t$. A central scalar is a left multiplier, so it keeps it too, and a term in the centre multiplies each half by itself and exchanges nothing. A right multiplier keeps the letter when $v$ commutes with $u$ and exchanges the two letters when $v$ anticommutes with it, $$ u v+v u=0\quad\Longrightarrow\quad P_R\,R_v\,P_L=R_v\,P_L\quad\text{and}\quad P_L\,R_v\,P_R=R_v\,P_R , $$ since $R_vR_u=R_{uv}$ and $R_uR_v=R_{vu}$. So the transport of a letter is a right-slot operation and never a left-slot one, and the two questions which side is the operator and does the operator move a letter are answered by different memberships of the operator. An operator of the left slot keeps every letter; an operator of the right slot keeps or exchanges according to one anticommutator; a central scalar keeps them all.
The one genuine contact between the two theories is an identity, and it should be stated as an identity rather than as a coincidence of arrays. For a traceless $u$ with $u^2=-e_t$ the projector $\tfrac12(\mathrm I+iR_u)$ is idempotent of rank two, and its image is a minimal left ideal. It is at once the Peirce projector $R(\tilde\Pi)$ of the idempotent $\tilde\Pi=\tfrac12(e_t+iu)$ and the spectral projector of $-iR_u$ onto its eigenvalue $-1$. It holds for every such $u$, because both readings are built from the same element, and that is the reason it holds. What must not be inferred from it is that the two notions are one. The Peirce projector is a subspace chosen by an idempotent. The spectral projector is one of the two spectral projectors of the single operator $-iR_u$. They agree here because $u$ was chosen to make them agree.
Why the Slot Label Is Not a Grading
A $\mathbb{Z}/2$ grading of an algebra is a decomposition $E_0\oplus E_1$ with $E_iE_j\subseteq E_{i+j}$, and a grading by a larger group is a decomposition into more subspaces. The label is not one, and three facts settle it.
A grading is a decomposition into subspaces, and the label is not. Its fourth case, no single side, is the complement of a union and not a subspace, so no grading returns it, whatever the rank of the group. The count says the same thing in one line: two parts against four cases.
The algebra does carry a grading, and it is not the label. In one display, chosen for this example and used for nothing else, split $\operatorname{End}_{\mathbb{C}}(\mathbb{B})$ into $2+2$ blocks and let $E_0$ be the block-diagonal arrays and $E_1$ the block-off-diagonal ones. Then $E_iE_j\subseteq E_{i+j}$, so the split is a grading, and each part is eight-dimensional. It is not the label, for two reasons. A part of a grading is a subspace, and the case no single side is not one. And the scalar line $\mathbb{C}\mathrm I$ lies in $E_0$ and carries the label both slots, so one part holds two different labels, and the grading is not a relabelling of the four cases.
A sign is stronger than a grading, and there is none. A sign would be a multiplicative map $\chi$ to $\{0,+1,-1\}$, with $\chi(XY)=\chi(X)\chi(Y)$, taking $+1$ on a left multiplier and $-1$ on a right multiplier of a noncentral element. The scalar $\lambda\mathrm I$ is the left multiplier of $\lambda e_t$ and the right multiplier of the same element, so it would carry both values at once. No $\chi$ separates the two sides, and the count of four cases against two values says the same thing without a computation.
What the swap gives instead is an exchange, and an exchange is not a sign. It carries one side to the other, answers the question "which side" by moving the operator rather than by labelling it, and splits the algebra into a self-adjoint and an anti-self-adjoint part of dimensions ten and six, with two different closing operations.
A sign, when the physics wants one, has to be chosen as an operator. Any $R_u$ with $u$ traceless and $u^2=-e_t$ supplies it, and its two projectors are the two halves. The sign is therefore an object of the algebra and not a functional on it, which is why no sign is ever exhibited as a functional: there is none, and the sign is a choice of $u$. The contrast is the moral of the article. The label is fixed before anything is chosen, and the sign moves when $u$ moves; confusing the two is what produces the impression that chirality is a matrix.
Summary
Operator chirality is the side on which an operator multiplies. The left multipliers form a four-dimensional image, the right multipliers another, the two meet in the scalars and span seven of the sixteen dimensions, and the remaining nine directions multiply on neither side alone. An operator is labelled left, right, both, or neither; the scalar line is the third case, and a two-sided basis operator is the fourth.
Under the structure isomorphism $\Xi(a\otimes b)=L_aR_b$ the two sides are the two tensor factors and their overlap is the tensor of the two units. The exchange of the sides is the flip transported to the endomorphism algebra, $\sigma=\Xi\tau\Xi^{-1}$. It is intrinsic. It squares to the identity, reverses products, exchanges the two images and fixes the scalars. It carries every $L_a$ to $R_a$, preserves rank and length, and is not inner. It is the adjoint of the trace form, and in a basis where that form is diagonal it is the transpose through a sign matrix. Its fixed part is a ten-dimensional Jordan algebra and its negated part a six-dimensional Lie algebra, and the two are the symmetric and the exterior square.
The quaternion conjugate is the second side-exchanger, $\natural L(a)\natural=R(a^{\natural})$. It differs from the swap in that it conjugates the parameters and reverses their order, and the sign in which the two differ is the origin of every phase the corpus obtains from a change of side.
None of this needs a root, an idempotent, a Peirce decomposition, a Peirce basis, a spectrum, a spinor, the module $S$, an ideal, a Clifford algebra, a matrix unit of a display, or the basis operator the older reading used. That basis operator is the instructive case: a pure right multiplier with square one, trace zero and a balanced spectrum, hence an example of the right side and not the definition of chirality, and built from a chosen element besides. The one contact with the Peirce theory is an identity forced by a shared element, and it is an identity of two constructions and not of two notions.
$$ \boxed{\ \text{chirality}=\text{the side};\qquad X\in L(\mathbb{B}),\ \ X\in R(\mathbb{B}),\ \ X\in L(\mathbb{B})\cap R(\mathbb{B}),\ \ \text{or neither};\qquad \sigma=\Xi\tau\Xi^{-1},\quad \sigma(L_a)=R_a .\ } $$
Summary of Notation
| Symbol | Meaning |
|---|---|
| $L(a)$, $L_a$ | the left multiplier, $\tilde X\mapsto a\tilde X$ |
| $R(b)$, $R_b$ | the right multiplier, $\tilde X\mapsto\tilde Xb$ |
| $L_aR_b$ | the basis operator, with a left and a right slot |
| $L(\mathbb{B})$, $R(\mathbb{B})$ | the two images, four-dimensional, meeting in the centre |
| left slot, right slot, both slots, no single side | operator chirality, a membership in the two images |
| $\Xi(a\otimes b)=L_aR_b$ | the structure isomorphism $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\to\operatorname{End}_{\mathbb{C}}(\mathbb{B})$ |
| $\tau(a\otimes b)=b\otimes a$ | the flip of the two tensor factors |
| $\sigma=\Xi\tau\Xi^{-1}$ | the swap, $\sigma(L_aR_b)=L_bR_a$ |
| $\beta(\tilde X,\tilde Y)=\operatorname{Sc}(\tilde X\tilde Y)$ | the trace form, symmetric and nondegenerate |
| $\mathsf W$, $\mathsf D$ | the Gram matrix of $\beta$ in a carrier basis, called the switch; it equals $\operatorname{diag}(1,-1,-1,-1)$ in the reference frame |
| $\operatorname{Fix}(\sigma)=\Xi(\operatorname{Sym}^2\mathbb{B})$ | dimension $10$, a Jordan algebra under the anti-commutator |
| $\operatorname{Neg}(\sigma)=\Xi(\Lambda^2\mathbb{B})$ | dimension $6$, a Lie algebra under the commutator |
| $\natural$ | the quaternion conjugate, $\natural L(a)\natural=R(a^{\natural})$ |
| $\ell(M)$, $\operatorname{rank}M$ | the length and the rank, preserved by $\sigma$ |
| $\tfrac12(\mathrm I+iR_u)$, with $u$ traceless and $u^2=-e_t$ | the rank-two projector of a right multiplier, Peirce and spectral at once |
Further Reading
- Families of Matrix Representations, for the two sides, the switch and the change of frame.
- The Endomorphisms of the Biquaternion Algebra, for the ambient operator space and the two-sided operators.
- The Unpolarized Module of the Biquaternion Algebra, for the carrier, the projective line of splittings and the trace form.
- Operator Family Conventions, for the four families and the slot label.
- Matrix Representations Conventions, for the labels and the conventions of the category.