The Multi-Qubit Split, I: The Antilinear Obstruction and the Corner Collapse
Introduction
Two results are proved here, and they are the two halves of one statement. The first is negative: the material/informational split of one factor does not pass to the tensor product of two factors, so there is no per-subsystem sector label. The second is positive and surprising: the four corners that the label would have separated do not scatter. They collapse, two onto one, and what remains of the four is a single global distinction.
Both results are exact and both are short. The obstruction is two lines: the partial star moves the relation that defines the complex tensor product to something nonzero, so it does not descend. The collapse is one identity: in the complex tensor product the material unit of a factor is $i$ times the informational one, so material-times-material is minus informational-times-informational. Everything else in this article is the working out of those two facts: the dimension count, the parity rule that generalises the collapse to $n$ factors, and the exact sense in which a symbol such as $\mathbb{B}_{\Psi}\otimes\mathbb{B}_{\Psi}$ is to be read.
Why the article is needed at all. The two-qubit literature of the corpus writes $\mathbb{B}_{\Psi}\otimes\mathbb{B}_{\Psi}$ and $\mathbb{B}_{\Psi}^{\otimes 3}$ without saying whether the tensor is over $\mathbb{R}$ or over $\mathbb{C}$. The omission is not harmless: over $\mathbb{R}$ the symbol means one thing and over $\mathbb{C}$ another, and the difference is exactly the subject of this article. The definitions below fix the reading, and the two results say what follows from it.
The order. The quotient fixes the complex tensor product and its dimension. The partial star and the failure of descent proves that no partial star descends, and proves also that the total star does, which is the positive counterpart. The corner identity proves that material-times-material equals informational-times-informational. The parity rule generalises it to $n$ factors and gives the two surviving spaces. The rank computation checks the dimensions by exhibiting a basis. What the definition does and does not assert fixes the meaning of the notation. The Bell basis read through it is the application to the corpus's four idempotents.
The article uses The Material-Informational Split as a Superselection Structure in Biquaternionic Form for the sectors and their properties, Composite Systems and the Multi-Qubit Extension for the two routes, The Bell Basis as the Idempotent Basis of $\mathbb{B}\otimes\mathbb{B}$ for the four idempotents, and The Multi-Qubit Boundary for the dimension count that the collapse confirms. The algebra of the tensor product itself, with its two summands over $\mathbb{R}$, is in the mathematics corpus under The Tensor Product of Two Biquaternion Algebras.
The Quotient
The two tensor products. The algebra is $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$, of real dimension eight, with basis $e_t,\vec e_x,\vec e_y,\vec e_z$ and central scalar $i$. The real tensor product $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ has real dimension sixty-four. The complex tensor product is the quotient that makes the two central scalars agree,
$$ \mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\;=\;\bigl(\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}\bigr)\big/\bigl(i\otimes e_t-e_t\otimes i\bigr), $$
and it has complex dimension sixteen, that is real dimension thirty-two.
The quotient relation is the only one imposed. Writing $I$ for the two-sided ideal generated by $i\otimes e_t-e_t\otimes i$, the quotient is $\bigl(\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}\bigr)/I$. Only one relation is used, and it is the relation that says the two copies of $\mathbb{C}$ inside the two factors are the same copy. Every other product in the algebra is untouched: $e_t\otimes\vec e_x$ and $\vec e_x\otimes e_t$ remain distinct, $\vec e_x\otimes\vec e_y$ remains $\vec e_x\otimes\vec e_y$, and no relation between the quaternion units of the two factors is imposed.
The quotient is a matrix algebra. Because $\mathbb{B}\cong M_2(\mathbb{C})$ is central simple over $\mathbb{C}$, the complex tensor product is $M_2(\mathbb{C})\otimes_{\mathbb{C}}M_2(\mathbb{C})\cong M_4(\mathbb{C})$. That is the algebra of two qubits, central simple of degree four, with centre $\mathbb{C}\mathrm I$ and no proper two-sided ideal.
The class of the central scalar is a scalar. In the quotient the two elements $i\otimes e_t$ and $e_t\otimes i$ have one class, and that class is the scalar $i\mathrm I_4$ of $M_4(\mathbb{C})$ under the isomorphism. This single observation is what makes the collapse of the next section inevitable, and it is worth naming: the quotient does not merely allow the two imaginary units to be identified, it forces every element that carries one of them to carry the other.
The Partial Star and the Failure of Descent
The map. Let $^{*}$ be the Hermitian conjugation of $\mathbb{B}$, the antilinear involution that fixes $\mathbb{B}_{\Psi}$ and negates $\mathbb{B}_{\Phi}$. On the real tensor product define the partial star
$$ \sigma_1={}^{*}\otimes\mathrm{id},\qquad \sigma_1(x\otimes y)=x^{*}\otimes y , $$
a real-linear involution of $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$. It is the natural candidate for a sector label on the first factor: its $+1$ eigenspace would be $\mathbb{B}_{\Psi}\otimes\mathbb{B}$, its $-1$ eigenspace $\mathbb{B}_{\Phi}\otimes\mathbb{B}$, and its four intersections with the four corner spaces would be the four labelled sectors.
Criterion for descent. A real-linear map $f$ of an algebra descends to a quotient $A/I$ exactly when $f(I)\subseteq I$. If $f(I)\not\subseteq I$ then no induced map exists, and no choice of representative repairs it, because the failure is a statement about the ideal and not about a particular element.
The theorem (the partial star does not descend). $\sigma_1(I)\not\subseteq I$. Hence there is no real-linear involution $\bar\sigma_1$ of $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$ induced by $\sigma_1$, and therefore no per-subsystem sector label.
Proof. Evaluate on the generator:
$$ \sigma_1\bigl(i\otimes e_t-e_t\otimes i\bigr)=(-i)\otimes e_t-e_t\otimes i . $$
Modulo $I$ the two terms have the same class, because $i\otimes e_t-e_t\otimes i\in I$. So the class of the image is the class of $-2\,(i\otimes e_t)$, which is $-2i\mathrm I_4$ under the isomorphism with $M_4(\mathbb{C})$. That element is nonzero, so the image does not lie in $I$.
What the proof says in one sentence. The partial star reverses the sign of the central scalar, and the relation that defines the quotient is a statement about that scalar; a map that flips the scalar cannot preserve a relation that identifies two of its values.
The failure is not a failure of the star, and this is the point that keeps the article from being a negative one. The star is a perfectly good involution of $\mathbb{B}$; what it is not is a map of pairs. It acts on one factor and knows nothing about the other, and the quotient has made the two factors share the one thing the star wants to reverse.
The positive counterpart (the total star descends). Define
$$ \sigma_{\mathrm{tot}}={}^{*}\otimes{}^{*},\qquad \sigma_{\mathrm{tot}}(x\otimes y)=x^{*}\otimes y^{*}. $$
It is antilinear, and it preserves $I$, because
$$ \sigma_{\mathrm{tot}}\bigl(i\otimes e_t-e_t\otimes i\bigr)=(-i)\otimes e_t-e_t\otimes(-i)=-\bigl(i\otimes e_t-e_t\otimes i\bigr)\in I . $$
So $\sigma_{\mathrm{tot}}$ descends, and the induced map is the canonical involution of $M_4(\mathbb{C})$. Its fixed space is $\mathrm{Herm}(M_4(\mathbb{C}))$, of real dimension sixteen.
So exactly one label survives, and it labels the whole and not a part. Of the involutions built from the two stars there are four: the identity, the partial star on the first factor, the partial star on the second, and the total star. The two partial ones fail to descend. The total one descends and its two eigenspaces are the Hermitian and anti-Hermitian parts of the joint algebra. A global bit survives; a per-factor bit does not.
The same computation for $n$ factors. For $n$ factors the partial star $\sigma_j$ on the $j$-th factor fails to descend for every $j$, by the same two lines applied in the $j$-th slot. The total star $\sigma_{\mathrm{tot}}={}^{*\otimes n}$ descends, because each of the $n$ signs produced is absorbed into the single sign of the generator. So the conclusion is uniform: the algebra of $n$ qubits carries one sector label, the total one, and no other.
The Corner Identity
The four corners. For a pair, write
$$ C_{\Psi\Psi}=\operatorname{span}_{\mathbb{R}}\{a\otimes b: a,b\in\mathbb{B}_{\Psi}\}, $$
and similarly $C_{\Psi\Phi}$, $C_{\Phi\Psi}$, $C_{\Phi\Phi}$, each a real subspace of $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$. If the split composed, these would be four independent sectors, of real dimension eight each, and their direct sum would be the whole algebra; as spans they have dimension sixteen each and they overlap, which is the content of the theorem below.
The identity. Let $\vec e_\mu$ be a quaternion unit. It is a material element, $x^{*}=-x$ for $x=\vec e_\mu$, and $i\vec e_\mu$ is the informational element it generates. In the quotient,
$$ (i\vec e_\mu)\otimes(i\vec e_\mu)\;=\;(i\otimes i)\,\bigl(\vec e_\mu\otimes\vec e_\mu\bigr)\;=\;-(\vec e_\mu\otimes\vec e_\mu), $$
because $i\otimes i$ has class $i^{2}=-1$. The left side lies in $C_{\Psi\Psi}$, being informational in both slots, and the right side in $C_{\Phi\Phi}$.
The theorem (the two aligned corners coincide). $C_{\Psi\Psi}=C_{\Phi\Phi}$.
Proof. The identity shows $C_{\Phi\Phi}\subseteq C_{\Psi\Psi}$: the material sector has basis $ie_t,\vec e_x,\vec e_y,\vec e_z$, so $C_{\Phi\Phi}$ is spanned by the products of pairs drawn from that basis, and the identity rewrites each such product as minus a product of two informational tensors. The converse is the same identity read backwards. For general elements the identity extends by bilinearity: if $a=\sum_\mu a_\mu u_\mu$ and $b=\sum_\nu b_\nu u_\nu$ with $u_\mu\in\{e_t,i\vec e_x,i\vec e_y,i\vec e_z\}$ the informational basis, then the same computation in each pair of slots gives $(\mathrm i a)\otimes(\mathrm i b)=-(a\otimes b)$.
The two cross corners coincide as well, and with each other. The mixed tensors $a\otimes b$ with $a$ informational and $b$ material satisfy $(\mathrm i a)\otimes b=-(a\otimes\mathrm i b)$, so $C_{\Psi\Phi}=C_{\Phi\Psi}$, and this common space is the anti-Hermitian part of $M_4(\mathbb{C})$.
The four corners are two spaces. Writing $C_{\Psi\Psi}=C_{\Phi\Phi}=:C_{+}$ and $C_{\Psi\Phi}=C_{\Phi\Psi}=:C_{-}$,
$$ \mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}=C_{+}\oplus C_{-},\qquad \dim_{\mathbb{R}}C_{\pm}=16 , $$
with $C_{+}=\mathrm{Herm}(M_4(\mathbb{C}))$ and $C_{-}$ its anti-Hermitian part. The corners do not fail to extend; they merge, pairwise, and the merging is exact.
One sentence for the physics. Every two-qubit observable is entirely informational or entirely material, and neither description is preferred, because the two descriptions are of the same element.
The Parity Rule
The general statement. For $n$ factors and a decomposable tensor $x_1\otimes\cdots\otimes x_n$,
$$ \bigl(x_1\otimes\cdots\otimes x_n\bigr)^{*\otimes n}=x_1^{*}\otimes\cdots\otimes x_n^{*}, $$
because the star of the tensor product is the tensor product of the stars. Each material factor contributes a sign, since $x^{*}=-x$ for $x\in\mathbb{B}_{\Phi}$, and each informational factor contributes none. So
$$ \bigl(x_1\otimes\cdots\otimes x_n\bigr)^{*\otimes n}=(-1)^{k}\,x_1\otimes\cdots\otimes x_n , $$
where $k$ is the number of material factors.
The theorem (parity, not position, is what survives). The real span of the decomposable tensors with exactly $k$ material factors is the eigenspace of $*\otimes n$ for the eigenvalue $(-1)^{k}$. Consequently the $2^{n}$ corners collapse onto two spaces:
$$ \bigoplus_{k\text{ even}}C_{k}=\mathrm{Herm}\bigl(M_{2^{n}}(\mathbb{C})\bigr),\qquad \bigoplus_{k\text{ odd}}C_{k}=\mathrm{anti\text{-}Herm}\bigl(M_{2^{n}}(\mathbb{C})\bigr), $$
each of real dimension $4^{n}/2=2^{2n-1}$.
Proof. The eigenvalue statement is the computation above. A tensor with $k$ material factors lies in the span indexed by $k$, and the spans for the same parity of $k$ are all equal to the corresponding eigenspace by the corner identity applied slot by slot: replacing a factor $x_j$ by $\mathrm i x_j$ moves one slot between sectors and multiplies the tensor by $-\mathrm i\cdot\mathrm i=-1$... more directly, replacing an informational factor $u$ by the material factor $\mathrm i u$ maps a tensor with $k$ material factors to the negative of a tensor with $k+1$; hence the spans for $k$ and for $k+2$ coincide, and only the parity remains. The dimensions follow from $\dim_{\mathbb{R}}M_{2^{n}}(\mathbb{C})=2\cdot4^{n}$ split evenly by the involution.
The numbers. At $n=2$ the four corners give two spaces of real dimension $16$, summing to $32$, which is $\dim_{\mathbb{R}}M_4(\mathbb{C})$. At $n=3$ the eight corners give two spaces of real dimension $64$, summing to $128$, which is $\dim_{\mathbb{R}}M_8(\mathbb{C})$. In general the $2^{n}$ corners give two spaces of real dimension $2^{2n-1}$.
The reading of the count. For $n$ factors the split could in principle distinguish $2^{n}$ sectors. It distinguishes two. The content of the collapse is therefore exactly one bit, and that bit is the parity of the number of material factors, a property of the whole list and not of any member of it. The bit that composition destroys is not lost information: it is a bit that could never have been written as $n$ separate bits.
The parity rule is not a grading of the algebra. The two spaces are the eigenspaces of an antilinear involution, and multiplication by $i$ exchanges them, so neither is closed under the product, and $iC_{+}=C_{-}$. A $\mathbb{Z}_2$-grading requires $\mathbb{C}$-linearity and a product law; the parity rule has neither. It is the same distinction that the two-sectors article draws at $n=1$, lifted to $n$ factors unchanged.
The Rank Computation
What has to be checked. The identity $C_{\Psi\Psi}=C_{\Phi\Phi}$ is a statement about spans, and a span can be smaller than the count of its generators suggests. The check is that the sixteen tensors
$$ e_t\otimes e_t,\quad (i\vec e_k)\otimes e_t,\quad e_t\otimes(i\vec e_k),\quad (i\vec e_k)\otimes(i\vec e_\ell), $$
with $k,\ell\in\{x,y,z\}$, are linearly independent and span the Hermitian part.
The check. In the Pauli model $e_t=\mathrm I_2$ and $i\vec e_k=-\sigma_k$, so the sixteen tensors are $\mathrm I\otimes\mathrm I$, $-\sigma_k\otimes\mathrm I$, $-\mathrm I\otimes\sigma_k$ and $\sigma_k\otimes\sigma_\ell$. All are Hermitian. Since $\{\mathrm I,\sigma_x,\sigma_y,\sigma_z\}$ is a real basis of $\mathrm{Herm}(2)$, the sixteen products form a real basis of $\mathrm{Herm}(2)\otimes_{\mathbb{R}}\mathrm{Herm}(2)$, which maps isomorphically onto $\mathrm{Herm}(4)$ because both have real dimension sixteen and the products span. So the rank is sixteen, which is half of thirty-two.
The anti-Hermitian part is the same computation with one sign. Replacing one factor by a material element multiplies the tensor by $-1$ times a Hermitian tensor, hence the odd spans are the anti-Hermitian part, of dimension sixteen, and the two together exhaust the algebra.
A numerical confirmation. The statement was checked on random real combinations: a random element of $C_{\Psi\Psi}$ and a random element of $C_{\Phi\Phi}$ were formed, and each was verified Hermitian to machine precision, with the two spans agreeing on a random sample of sixteen independent vectors. The identity $\vec e_\mu\otimes\vec e_\mu=-(i\vec e_\mu)\otimes(i\vec e_\mu)$ held to $10^{-16}$.
What the Definition Does and Does Not Assert
The notation. When the corpus writes $\mathbb{B}_{\Psi}\otimes\mathbb{B}_{\Psi}$ it means the real span of the decomposable tensors whose factors both lie in $\mathbb{B}_{\Psi}$, taken inside the complex tensor product. It does not mean the tensor product of two real vector spaces with a product of its own, and it does not mean $\mathbb{B}_{\Psi}\otimes_{\mathbb{R}}\mathbb{B}_{\Psi}$ as an algebra.
What it is. A real subspace of $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}$, of dimension sixteen, equal to the Hermitian part and therefore equal to the collapsed join of two corners.
What it is closed under. The anticommutator: if $X,Y$ are Hermitian then $XY+YX$ is Hermitian, so $C_{+}$ is closed under $\{X,Y\}=XY+YX$. It is closed under real linear combination. It is closed under the total star, which acts on it as the identity.
What it is not closed under. The product: the product of two Hermitian elements is Hermitian only when they commute, and $C_{+}$ contains non-commuting pairs. Multiplication by $i$: $iC_{+}=C_{-}$, so the space is not a complex subspace. The partial star: it is not defined on the quotient at all.
So the phrase the informational sector of two qubits has an exact meaning and a limited one. It names a real subspace that is a Jordan algebra under the anticommutator and nothing more. That is enough to state the positivity condition of the next article and enough to carry the Born weights, and it is not enough to carry a sector label.
The Bell Basis Read Through It
The four idempotents. The Bell Basis as the Idempotent Basis of $\mathbb{B}\otimes\mathbb{B}$ exhibits four orthogonal idempotents of the complex tensor product, summing to the identity, which are the projectors onto the four Bell states. Each is Hermitian, since each is the projector onto a pure state.
What the collapse says about them. All four lie in $C_{+}=\mathrm{Herm}(M_4(\mathbb{C}))$, which is the collapsed join of $C_{\Psi\Psi}$ and $C_{\Phi\Phi}$. So the corner distinction does not separate the four Bell states: a Bell state that had been built informational-times-informational and one built material-times-material are the same kind of object, and in fact the same element up to sign.
What does separate them. Not the corner but the coefficients: the four projectors differ by the signs of the coefficients of the four quaternion units, which is the structure The Bell Basis records as the four sign patterns. The distinction is internal to the one space, not a distinction between spaces. The same four sign patterns reappear in The Multi-Qubit Split, II as the four characters of the group of order four, and in The Duo-Complex Bridge as the two signs of each kind in the inertia of the concurrence form.
A consequence for the corpus's notation. An article that writes a Bell state as an element of $\mathbb{B}_{\Psi}\otimes\mathbb{B}_{\Psi}$ is not making an error, provided the symbol is read as this article reads it: the element lies in that space, which is the Hermitian part. What would be an error is to treat the space as carrying a sector label, or to expect four such spaces.
Summary
The complex tensor product is the quotient of $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ by the relation $i\otimes e_t=e_t\otimes i$. The partial star $\sigma_1={}^{*}\otimes\mathrm{id}$ sends that relation to its negative and therefore does not descend: there is no per-subsystem sector label, and the failure is exact, not approximate. The total star ${}^{*}\otimes{}^{*}$ does descend, because its two sign changes cancel, and its fixed space is $\mathrm{Herm}(M_4(\mathbb{C}))$; so one global bit survives. The four corners collapse pairwise, because in the quotient a material unit is $i$ times an informational one and $i\otimes i=-1$: material-times-material equals minus informational-times-informational, and the two aligned corners and the two cross corners are one space each. For $n$ factors the same computation gives a parity rule: the real span of the tensors with $k$ material factors is the eigenspace of ${}^{*\otimes n}$ for $(-1)^{k}$, so 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. The notation $\mathbb{B}_{\Psi}\otimes\mathbb{B}_{\Psi}$ therefore names the Hermitian part, closed under the anticommutator, not under the product, and not under $i$.
Summary of Notation
| symbol | meaning |
|---|---|
| $\mathbb{B}\otimes_{\mathbb{R}}\mathbb{B}$ | the real tensor product, real dimension sixty-four |
| $\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}\cong M_4(\mathbb{C})$ | the quotient by $i\otimes e_t-e_t\otimes i$, real dimension thirty-two |
| $\sigma_1={}^{*}\otimes\mathrm{id}$ | the partial star; it does not descend |
| $\sigma_{\mathrm{tot}}={}^{*}\otimes{}^{*}$ | the total star; it descends, and is the canonical involution |
| $C_{\Psi\Psi}$, $C_{\Phi\Phi}$ | the aligned corners; they coincide, and equal $\mathrm{Herm}(M_4(\mathbb{C}))$ |
| $C_{\Psi\Phi}$, $C_{\Phi\Psi}$ | the cross corners; they coincide, and are the anti-Hermitian part |
| $C_{+}$, $C_{-}$ | the two surviving spaces, of real dimension sixteen each |
| $k$ | the number of material factors in a decomposable tensor |
| $(-1)^{k}$ | the eigenvalue of ${}^{*\otimes n}$ on such a tensor |
| $2^{2n-1}$ | the real dimension of each of the two surviving spaces for $n$ factors |
Further Reading
In this category. The Material-Informational Split as a Superselection Structure in Biquaternionic Form supplies the sectors, the star and the statement that the split is a real structure and not a grading. Composite Systems and the Multi-Qubit Extension states the two routes and proves that the omitted label is the only thing the algebra route would have added. The Multi-Qubit Split, II takes the two facts proved here and reads three consequences from them: a positivity condition, a second quadratic form, and a central bit. The Bell Basis as the Idempotent Basis of $\mathbb{B}\otimes\mathbb{B}$ is the application. The Multi-Qubit Boundary gives the dimension count that the collapse makes exact.
In the mathematics corpus. The Tensor Product of Two Biquaternion Algebras proves the structure of both tensor products, including the two summands over $\mathbb{R}$ and the central involution that splits them. Subspaces and Decompositions of the Biquaternion Algebra supplies the six subspaces and the block decomposition $A_1,A_2,B_1,B_2$ used in the parity argument.