The Tensor Product of Two Biquaternion Algebras

Introduction

The tensor product of two copies of $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ is not a biquaternion algebra, and there are two tensor products to compute. This article computes both: the complex one, $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\cong M_4(\mathbb{C})$, of complex dimension sixteen, and the real one, $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}\cong M_4(\mathbb{C})\oplus M_4(\mathbb{C})$, of real dimension sixty-four. It also computes what happens to the two structures the algebra carries: the star, which survives only in its total form, and the sector splitting, which does not survive at all but leaves a parity rule behind.

The article is the algebra behind the multi-qubit extension in the physics corpus. What is computed here is used there: Composite Systems and the Multi-Qubit Extension for the two routes, The Multi-Qubit Split, I for the obstruction and the collapse, and The Multi-Qubit Split, II for the central bit. The algebra is in The Biquaternion Algebra; the star is in The Group of Involutions; the decompositions are in Subspaces and Decompositions of the Biquaternion Algebra; the modules are in The Category of Modules in the Biquaternion Algebra.

The four results. First, the complex tensor product is a full matrix algebra of degree four, central simple with no non-trivial two-sided ideal, and not isomorphic to $\mathbb{B}$. Second, the partial star does not descend to it, so there is no per-factor sector label; the total star does descend, so one global label survives. Third, in the complex tensor product the four sector corners collapse pairwise, because a material unit of one factor is $i$ times an informational unit and $i\otimes i=-1$. Fourth, over $\mathbb{R}$ the algebra doubles and a central involution splits it into two matrix algebras, which is where the central bit of the physical description comes from.

The Two Tensor Products

Over $\mathbb{R}$. $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ is a real algebra of dimension eight, with centre $\mathbb{C}e_t$ of real dimension two. The real tensor product $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ has real dimension sixty-four and centre of real dimension four.

Over $\mathbb{C}$. The complex tensor product is the quotient

$$ \mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}=\bigl(\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}\bigr)\big/\bigl(i\otimes e_t-e_t\otimes i\bigr), $$

that is, the real tensor product with the two central scalars identified. Its dimension over $\mathbb{C}$ is sixteen, and over $\mathbb{R}$ thirty-two. The single relation is the whole difference between the two objects: every other product is untouched, so $e_t\otimes\vec e_x$ and $\vec e_x\otimes e_t$ remain distinct, and the quaternion units of the two factors do not commute by fiat.

The universal property, stated to fix the reading. For any complex algebra $A$ and any pair of $\mathbb{C}$-algebra maps $f_1,f_2:\mathbb{B}\to A$, there is a unique $\mathbb{C}$-algebra map $f:\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\to A$ with $f(x\otimes e_t)=f_1(x)$ and $f(e_t\otimes y)=f_2(y)$. So the complex tensor product is the algebra of pairs of commuting copies of $\mathbb{B}$, and the real tensor product is the algebra of pairs of copies with independent centres.

A dimension table.

object field dimension centre
$\mathbb{B}$ $\mathbb{R}$ $8$ $2$
$\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ $\mathbb{R}$ $64$ $4$
$\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$ $\mathbb{C}$ $16$ $1$

The Complex Tensor Product

The theorem. $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\cong M_4(\mathbb{C})$.

Proof. $\mathbb{B}\cong M_2(\mathbb{C})$ as a complex algebra, and the tensor product of matrix algebras is the matrix algebra of the product of the degrees: $M_m(\mathbb{C})\otimes_{\mathbb{C}}M_n(\mathbb{C})\cong M_{mn}(\mathbb{C})$ through the Kronecker product. With $m=n=2$ the degree is four.

The consequences of being a full matrix algebra. $M_4(\mathbb{C})$ is simple, so it has no two-sided ideal other than $0$ and the whole algebra; it is central, so its centre is $\mathbb{C}\mathrm I_4$, of complex dimension one; it is the endomorphism algebra of a four-dimensional complex space, unique up to isomorphism; and every automorphism is inner, so $\operatorname{Aut}(M_4)=\mathrm{PGL}_4(\mathbb{C})$.

The consequence that matters for the sectors. A decomposition of an algebra into a direct sum of two-sided ideals is the same thing as a family of central idempotents summing to one. Since the centre of $M_4(\mathbb{C})$ is one-dimensional, its only idempotents are $0$ and $1$, so there is no such decomposition. The complex tensor product admits no sector decomposition into two-sided ideals, at any level.

The explicit isomorphism. In the Pauli model $e_t=\mathrm I_2$ and $\vec e_k=i\sigma_k$, so

$$ e_\mu\otimes e_\nu\;\longmapsto\;\mathsf M_2(e_\mu)\otimes\mathsf M_2(e_\nu)\in M_2(\mathbb{C})\otimes M_2(\mathbb{C})=M_4(\mathbb{C}), $$

and the sixteen products $e_\mu\otimes e_\nu$ are a complex basis of the image. The identification of the two central scalars is the identification of $i\mathrm I_2\otimes\mathrm I_2$ with $\mathrm I_2\otimes i\mathrm I_2$, which is $i\mathrm I_4$ in both cases.

The reading of the degree. The degree of the tensor product is the product of the degrees, so $2\cdot2=4$. This is the whole of the obstruction to a biquaternion reading of two qubits: a biquaternion algebra has degree two, and no relabelling changes a degree. Four amplitudes per subsystem give sixteen amplitudes, which is the square of four and not four.

The Star on the Complex Tensor Product

The total star descends. Let $\sigma_{\mathrm{tot}}={}^{*}\otimes{}^{*}$ on $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$. It preserves the quotient relation, since

$$ \sigma_{\mathrm{tot}}\bigl(i\otimes e_t-e_t\otimes i\bigr)=(-i)\otimes e_t-e_t\otimes(-i)=-\bigl(i\otimes e_t-e_t\otimes i\bigr), $$

so it induces an involution $\bar\sigma_{\mathrm{tot}}$ of $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$. Under the isomorphism with $M_4(\mathbb{C})$ it is the Hermitian adjoint $X\mapsto X^{\dagger}$, and its fixed space is $\mathrm{Herm}(M_4(\mathbb{C}))$, of real dimension sixteen, half of thirty-two.

The partial star does not descend. Let $\sigma_1={}^{*}\otimes\mathrm{id}$. Then

$$ \sigma_1\bigl(i\otimes e_t-e_t\otimes i\bigr)=(-i)\otimes e_t-e_t\otimes i , $$

whose class in the quotient is $-2\,(i\otimes e_t)$, that is $-2i\mathrm I_4$, which is nonzero. Since a map descends to a quotient exactly when it carries the ideal into itself, no induced map exists.

Why the partial star had no chance, in algebraic terms. $\sigma_1$ is not an algebra map, because the star reverses the order of a product; and it is not an anti-algebra map either, since the reversal would have to be applied in both slots while $\sigma_1$ applies it in one. So the candidate for a per-factor label reverses the order in one factor and preserves it in the other, and no such map is an involution of an algebra of pairs. Its natural home is $\mathbb{B}^{\mathrm{op}}\otimes\mathbb{B}$, and the identification of $\mathbb{B}$ with its opposite does not carry the star to the identity.

The count of the involutions. Of the four maps built from the two stars — the identity, the two partial stars, the total star — the two partial ones fail to descend and the total one descends. So the tensor product carries exactly one sector involution, and its eigenspaces are the Hermitian and anti-Hermitian parts. The label that survives labels the whole and not a factor.

The Real Tensor Product

The theorem. $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}\cong M_4(\mathbb{C})\oplus M_4(\mathbb{C})$.

Proof. Factor the algebra: $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}=(\mathbb{C}\otimes_{\mathbb{R}}\mathbb{C})\otimes_{\mathbb{R}}(\mathbb{H}\otimes_{\mathbb{R}}\mathbb{H})$. The scalar extension is $\mathbb{C}\otimes_{\mathbb{R}}\mathbb{C}\cong\mathbb{C}\oplus\mathbb{C}$ through $a\otimes b\mapsto(ab,\bar a b)$, and the quaternion tensor square is $\mathbb{H}\otimes_{\mathbb{R}}\mathbb{H}\cong M_4(\mathbb{R})$. Hence the algebra is $(\mathbb{C}\oplus\mathbb{C})\otimes_{\mathbb{R}}M_4(\mathbb{R})\cong M_4(\mathbb{C})\oplus M_4(\mathbb{C})$.

The dimension check. Each summand is sixteen over $\mathbb{C}$, that is thirty-two over $\mathbb{R}$, and there are two, for sixty-four, which is $8\cdot8$.

The isomorphism made explicit. $\mathbb{H}\otimes_{\mathbb{R}}\mathbb{H}\cong M_4(\mathbb{R})$ is the regular representation of one factor on the other, and the two copies of $\mathbb{C}$ are the two characters of $\mathbb{C}\otimes_{\mathbb{R}}\mathbb{C}$: the pairs $(a\otimes b)\mapsto ab$ and $(a\otimes b)\mapsto\bar ab$. So the two summands are the complex tensor product and the tensor product against the conjugate algebra.

The Centre and the Central Involution

The centre. The centre of $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ is $\mathbb{C}\otimes_{\mathbb{R}}\mathbb{C}$, of real dimension four, with real basis $e_t\otimes e_t$, $i\otimes e_t$, $e_t\otimes i$ and $i\otimes i$.

The central involution. Put $\mathsf J=i\otimes i$. Then

$$ \mathsf J^{2}=i^{2}\otimes i^{2}=(-e_t)\otimes(-e_t)=e_t\otimes e_t , $$

so $\mathsf J$ is an involution, and it is central because $i$ is central in each factor.

The idempotents.

$$ u=\tfrac12\bigl(e_t\otimes e_t+\mathsf J\bigr),\qquad v=\tfrac12\bigl(e_t\otimes e_t-\mathsf J\bigr), $$

are central, idempotent, orthogonal and sum to the identity:

$$ u^{2}=u,\quad v^{2}=v,\quad uv=0,\quad u+v=e_t\otimes e_t . $$

The evaluation of $\mathsf J$. Under $\mathbb{C}\otimes_{\mathbb{R}}\mathbb{C}\to\mathbb{C}\oplus\mathbb{C}$, $a\otimes b\mapsto(ab,\bar a b)$, the element $i\otimes i$ maps to $\bigl(i\cdot i,\ (-i)\cdot i\bigr)=(-1,+1)$. So $\mathsf J$ is $-1$ on the summand cut by $v$ and $+1$ on the summand cut by $u$; and since $\mathsf J=u-v$ on the centre, $u\mathsf J=u$ and $v\mathsf J=-v$.

The two ideals. The summands are $u(\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B})$ and $v(\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B})$, each of real dimension thirty-two, each isomorphic to $M_4(\mathbb{C})$.

The Two Summands and the Physical One

The summand cut by $v$ is the complex tensor product. There $\mathsf J=-1$, so

$$ \mathsf J=(i\otimes e_t)(e_t\otimes i) $$

is the product of the two central scalars, and each of them squares to $i^{2}=-1$. Two equal scalars therefore multiply to $-1$, and a scalar and its conjugate multiply to $-i\,i=+1$. So the value of $\mathsf J$ is exactly the verdict on whether the two scalars agree, and $\mathsf J=-1$ says they do. Agreement is the relation that defines $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$, so this summand is the quotient, and it is $M_4(\mathbb{C})$.

The summand cut by $u$ is the tensor product against the conjugate algebra. There $\mathsf J=+1$, so $i\otimes e_t=-(e_t\otimes i)$: the second scalar is the negative of the first, that is the conjugate of the first. So the algebra is $\mathbb{B}\otimes_{\mathbb{C}}\overline{\mathbb{B}}$, of the same size and with the opposite complex structure.

The naming is easy to invert, so the sign is checked against the explicit map. With $a\otimes b\mapsto(ab,\bar a b)$ and $i\otimes i\mapsto\bigl(i^{2},\bar i\,i\bigr)=(-1,+1)$, the first entry corresponds to the product $ab$, in which the two scalars are taken as they are, and the second to $\bar a b$, in which the first is conjugated. Since $u=\tfrac12(e_t\otimes e_t+\mathsf J)$ maps to $(0,1)$ and $v=\tfrac12(e_t\otimes e_t-\mathsf J)$ to $(1,0)$, the first entry — and therefore the summand with $\mathsf J=-1$, cut by $v$ — is the one in which the scalars agree. That is the complex tensor product.

Stated once, unambiguously. The physical summand, the complex tensor product $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$, is the one on which $\mathsf J=-1$, cut by $v=\tfrac12(e_t\otimes e_t-\mathsf J)$. The other summand, on which $\mathsf J=+1$ and cut by $u$, is $\mathbb{B}\otimes_{\mathbb{C}}\overline{\mathbb{B}}$.

The Corners and the Parity Rule

The four corners. For $a,b\in\mathbb{B}_\Psi$ and $a',b'\in\mathbb{B}_\Phi$, the four real subspaces

$$ C_{\Psi\Psi}=\operatorname{span}_{\mathbb{R}}\{a\otimes b\},\quad C_{\Phi\Phi}=\operatorname{span}_{\mathbb{R}}\{a'\otimes b'\},\quad C_{\Psi\Phi},\quad C_{\Phi\Psi}, $$

each of real dimension sixteen, are the corners that a per-factor sector decomposition would separate.

The corner identity. For a quaternion unit $\vec e_\mu$ the material element is $\vec e_\mu$ itself and the informational element it generates is $i\vec e_\mu$; in the complex tensor product $i\otimes i=-e_t\otimes e_t$, so

$$ (i\vec e_\mu)\otimes(i\vec e_\mu)=(i\otimes i)\bigl(\vec e_\mu\otimes\vec e_\mu\bigr)=-\bigl(\vec e_\mu\otimes\vec e_\mu\bigr). $$

Hence $C_{\Psi\Psi}=C_{\Phi\Phi}$ and, by the same computation with one slot moved, $C_{\Psi\Phi}=C_{\Phi\Psi}$. The four corners are two spaces, each of real dimension sixteen, and together they are the whole algebra of thirty-two.

The parity rule. For $n$ factors and a decomposable tensor,

$$ \bigl(x_1\otimes\cdots\otimes x_n\bigr)^{*\otimes n}=(-1)^{k}\,x_1\otimes\cdots\otimes x_n , $$

where $k$ is the number of material factors. So the real span indexed by $k$ is the eigenspace of the total star for $(-1)^{k}$, and the $2^{n}$ corners collapse onto

$$ \bigoplus_{k\ \mathrm{even}}C_k=\mathrm{Herm}\bigl(M_{2^{n}}(\mathbb{C})\bigr),\qquad \bigoplus_{k\ \mathrm{odd}}C_k=\operatorname{anti\text{-}Herm}\bigl(M_{2^{n}}(\mathbb{C})\bigr), $$

of real dimension $2^{2n-1}$ each.

A basis check at $n=2$. The sixteen tensors $e_\mu\otimes e_\nu$ with $e_\mu\in\{e_t,i\vec e_x,i\vec e_y,i\vec e_z\}$ are Hermitian and independent, since in the Pauli model they are the products of two real bases of $\mathrm{Herm}(2)$, which form a real basis of $\mathrm{Herm}(2)\otimes_{\mathbb{R}}\mathrm{Herm}(2)\cong\mathrm{Herm}(4)$. So the corner has the dimension the count gives, and the equality of the two aligned corners is an equality of sixteen-dimensional spaces and not a containment.

The Star on Each Summand

The star preserves the summands. Since $\mathsf J^{*}=\mathsf J$ — each $i$ changes sign and there are two of them — the idempotents satisfy $u^{*}=u$ and $v^{*}=v$, so the total star restricts to each summand.

On either summand it is an involution of the second kind. The summand is a complex algebra and the star multiplies the complex scalar by $-1$; hence it is conjugate-linear, an involution, and its fixed space has real dimension sixteen, half of thirty-two. In the matrix model it is the Hermitian adjoint, and its fixed space is the Hermitian part.

What this does not give. An involution of the second kind is not a sector decomposition: its two eigenspaces are the Hermitian and anti-Hermitian parts, and the eigenspaces of a conjugate-linear map are real subspaces exchanged by multiplication by $i$, not ideals. So even on the summand where the labels would be expected, the surviving structure is one real grading and not a family of sectors.

What the Tensor Product Is Not

It is not a biquaternion algebra. Its degree is four and a biquaternion algebra's degree is two. Consequently no element of it is a two-qubit state in the sense in which an element of $\mathbb{B}_\Psi$ is a one-qubit state: the Hermitian part has real dimension sixteen, and the state space has real dimension fifteen, so the states are a positivity cone inside it and not the whole of a slice that dimension alone would call the state space.

It has no sector decomposition. By simplicity there is no non-trivial central idempotent, hence no splitting into two-sided ideals. The corners exist as real subspaces and collapse; they are not ideals, and they are not stable under multiplication by $i$.

It has no per-factor star. By the non-descent theorem there is no involution of the quotient that acts as the star on one factor. So the object that a two-system sector structure would need does not exist, and the failure is a statement about the quotient and not about a choice of basis.

What it does have, and this is not nothing. It has the total star, hence one global label; it has the corners, hence a parity rule; and over $\mathbb{R}$ it has the central involution, hence a bit. The three are the exact residues of the sector structure, and each is computed above.

Summary

The complex tensor product is $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\cong M_4(\mathbb{C})$, central simple of degree four, with centre $\mathbb{C}\mathrm I_4$; it is not a biquaternion algebra, and it has no non-trivial two-sided ideal, hence no sector decomposition. The partial star $\sigma_1={}^{*}\otimes\mathrm{id}$ does not descend, since it sends the relation $i\otimes e_t-e_t\otimes i$ to $-2i\mathrm I_4\neq0$; the total star ${}^{*}\otimes{}^{*}$ does descend, and is the Hermitian adjoint with fixed space of real dimension sixteen. In the quotient the corner identity $(i\vec e_\mu)\otimes(i\vec e_\mu)=-\vec e_\mu\otimes\vec e_\mu$ collapses the four corners onto two spaces of real dimension sixteen, and the parity rule generalises this to $n$ factors, where the $2^{n}$ corners collapse onto the Hermitian and anti-Hermitian parts of $M_{2^{n}}(\mathbb{C})$, of real dimension $2^{2n-1}$ each. Over $\mathbb{R}$ the algebra is $M_4(\mathbb{C})\oplus M_4(\mathbb{C})$, split by the central involution $\mathsf J=i\otimes i$, which is $-1$ on the summand where the two scalars agree and $+1$ on the summand where they are conjugate; the complex tensor product is the summand with $\mathsf J=-1$, cut by $v=\tfrac12(e_t\otimes e_t-\mathsf J)$. For $n$ factors there are $2^{n-1}$ summands.

Summary of Notation

Symbol Meaning
$\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ the real tensor product, real dimension sixty-four, centre of dimension four
$\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\cong M_4(\mathbb{C})$ the quotient by $i\otimes e_t-e_t\otimes i$, complex dimension sixteen
$\sigma_1={}^{*}\otimes\mathrm{id}$ the partial star; it does not descend
$\sigma_{\mathrm{tot}}={}^{*}\otimes{}^{*}$ the total star; it descends, and is the Hermitian adjoint
$\mathsf J=i\otimes i$ the central involution of the real tensor product
$u=\tfrac12(e_t\otimes e_t+\mathsf J)$ the idempotent of the summand where the scalars are conjugate
$v=\tfrac12(e_t\otimes e_t-\mathsf J)$ the idempotent of the summand where the scalars agree, that is the complex tensor product
$C_{\Psi\Psi}=C_{\Phi\Phi}=\mathrm{Herm}(M_4)$ the corner collapse
$(-1)^{k}$ the eigenvalue of ${}^{*\otimes n}$ on a tensor with $k$ material factors
$2^{2n-1}$ the real dimension of each surviving space at $n$ factors
$2^{n-1}$ the number of summands of $\mathbb{B}^{\otimes_{\mathbb{R}}n}$

Further Reading