Quantum Gates and Circuits in Biquaternionic Form

Introduction

A quantum gate is a reversible operation on a quantum state. In the biquaternion framework the state of a qubit is an element of the Hermitian subspace $\mathbb{B}_{\Psi}$ of the biquaternion algebra $\mathbb{B} = \mathbb{C}\otimes_\mathbb{R}\mathbb{H}$, namely a positive element normalised by $\langle\tilde{Ϛ}_{\Psi},e_t\rangle_{*}=\tfrac12$ (trace one in the $2\times2$ image) with Bloch vector $\vec s_{\Psi}$.

$$ \tilde{Ϛ}_{\Psi} = \tfrac{1}{2}\left(e_t + i\,\vec s_{\Psi}\right), \qquad \vec s_{\Psi} \in \mathbb{R}^3, $$

and a gate is a single element of the algebra acting on it by conjugation:

$$ \tilde{Ϛ}_{\Psi} \;\longmapsto\; \tilde Q\,\tilde{Ϛ}_{\Psi}\,\tilde Q^{*}, \qquad \tilde Q\tilde Q^{*} = e_t . $$

The acting element $\tilde Q$ is a $*$-unitary biquaternion; the unitaries form the group $\mathrm{U}(2) \subset \mathbb{B}$. A circuit is then a product of such elements. Composing two gates is multiplying two unitaries, and the effect of a whole circuit is a single conjugation by that product. The algebraic product in $\mathbb{B}$ (or in the tensor product $\mathbb{B}^{\otimes n}$ for $n$ qubits) is what the wiring of a circuit expresses.

This formulation is the one already used in the companion articles: states and observables in $\mathbb{B}_{\Psi}$, reversible evolution by rotor conjugation, irreversible measurement by the two-sided operation. The companion article Quantum Channels and the Reversible/Irreversible Dichotomy established that the reversible state maps are exactly the conjugations by $*$-unitary biquaternions — the channels of Kraus rank one — and that everything else is a sum of conjugations. The present article reads the standard gate and circuit notation through that result: a gate is a Kraus-rank-one channel, a circuit of gates is again a gate, and the moment a circuit contains a measurement it is a channel and no longer a gate.

Two warnings are in order at the outset, both inherited from the companion articles.

First, the mathematics below is standard quantum information theory transcribed into biquaternion notation. Nothing here depends on the physical hypothesis that $\mathbb{B}_{\Psi}$ is a distinct sector of the world. The reformulation is structural: it says where the objects of quantum computation live in the algebra and which algebraic operation each circuit ingredient is.

Second, the word "rotor conjugation" is used for the gate action, and it must be distinguished sharply from the Lorentz rotor of the material sector. Both actions are conjugations by the acting element, but the condition on that element is different. The material-sector Lorentz rotor, written $\tilde U_{\mathrm{lor}}$, satisfies the unit-norm condition $\tilde U_{\mathrm{lor}}\tilde U_{\mathrm{lor}}{}^{\natural} = e_t$ and generates $SL(2,\mathbb{C})$; a gate satisfies the matrix-unitarity condition $\tilde Q\tilde Q^{*} = e_t$ and generates $\mathrm{U}(2)$. A Lorentz boost $\tilde U_{\Psi,\mathrm b}$ is Hermitian and therefore satisfies $\tilde U_{\Psi,\mathrm b}\tilde U_{\Psi,\mathrm b}^{*} = \tilde U_{\Psi,\mathrm b}^2 \neq e_t$: it is a perfectly good rotor on $\mathbb{B}_{\Phi}$ but not a gate on $\mathbb{B}_{\Psi}$, because it does not preserve the trace. The two groups sit in the same algebra, and keeping them apart is the main technical discipline of this article.

The article is organised as follows. The gate group and its structure are described first. Then the standard single-qubit gate set is exhibited in biquaternion form. Then composition is treated as the algebra product, including the Clifford group. Then multi-qubit gates are built in the tensor-product arena, with the controlled gate, CNOT, CZ, and SWAP. Then a simple circuit — the preparation of a Bell state — is worked through algebraically. Then the reversible gates are contrasted with the irreversible channels of the read list. The article closes with what the reformulation does and does not claim, and with open questions.

The conventions are those of the companion articles. The biquaternion algebra is $\mathbb{B} = \mathbb{C}\otimes_\mathbb{R}\mathbb{H}$; the quaternion basis is $e_t = 1, \vec e_x, \vec e_y, \vec e_z$ with $\vec e_\eta^2 = -e_t$ and $\vec e_\mu \vec e_\nu=\epsilon_{\mu\nu\lambda}\vec e_\lambda$ for distinct $\mu,\nu$, with $\lambda$ the remaining index; the scalar imaginary is $i$, central and with $i^2 = -1$; the fixed-point subspaces are $\mathbb{B}_{C}$ (the complex subspace, the centre), $\mathbb{B}_{H}$ (the real-quaternion subspace), $\mathbb{B}_{\Psi}$ (Hermitian), and $\mathbb{B}_{\Phi}$ (anti-Hermitian). The doubled scalar part is $2\,\mathrm{Sc}(\tilde{Q}) = 2\,\mathrm{Sc}(\tilde{Q})$, and the reduction formula $2\,\mathrm{Sc}(\tilde\Pi\tilde Q_{\mathrm{Had}}^{*})=2\,\mathrm{Sc}(\tilde\Pi\tilde Q_{\mathrm{Had}})=2\,\mathrm{Sc}(\tilde\Pi\tilde Q_{\mathrm{Had}}^{*}) = 2\langle\tilde\Pi,\tilde Q_{\mathrm{Had}}\rangle_{*}$ is the Born rule; the star is inert on $\mathbb{B}_{\Psi}$ and the unstarred trace is real, and non-negative when $\tilde Q_{\mathrm{Had}}\ge0$. The isomorphism $\mathbb{B}\cong M_2(\mathbb{C})$ is the one fixed by the companion article on quantum physics,

$$ e_t \mapsto I_2, \qquad \vec e_x \mapsto -i\sigma_1, \qquad \vec e_y \mapsto -i\sigma_2, \qquad \vec e_z \mapsto -i\sigma_3, \qquad i \mapsto i I_2 , $$

with $\sigma_1,\sigma_2,\sigma_3$ the Pauli matrices. Under it, $\sigma_\eta$ is the image of $i \vec e_\eta$, and the idempotent $\tfrac{1}{2}(e_t + i\vec u)$ is the spin-up pure state along $\vec u$.

The Gate Group

Gates as $*$-Unitary Elements

Definition. A gate is a $*$-unitary element of $\mathbb{B}$,

$$ \boxed{\;\mathcal{U} \;=\; \{\tilde Q \in \mathbb{B} : \tilde Q\tilde Q^{*} = e_t\} \;\cong\; \mathrm{U}(2)\;}, $$

acting on a state $\tilde{Ϛ}_{\Psi} = \tfrac{1}{2}(e_t + i\vec s_{\Psi})$ by rotor conjugation

$$ Ξ_{\tilde Q}(\tilde{Ϛ}_{\Psi}) = \tilde Q\,\tilde{Ϛ}_{\Psi}\,\tilde Q^{*} . $$

The image is again Hermitian, positive, and of trace one: since $\tilde{Ϛ}_{\Psi}^{*} = \tilde{Ϛ}_{\Psi}$ and $\tilde Q^{*}\tilde Q = e_t$,

$$ Ξ_{\tilde Q}(\tilde{Ϛ}_{\Psi})^{*} = \tilde Q\tilde{Ϛ}_{\Psi}\tilde Q^{*} = Ξ_{\tilde Q}(\tilde{Ϛ}_{\Psi}), \qquad 2\,\mathrm{Sc}\!\left(Ξ_{\tilde Q}(\tilde{Ϛ}_{\Psi})\right) = 2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}) = 1 . $$

So a gate maps states to states.

The gate is a channel of Kraus rank one. In the language of the companion article on quantum channels, $Ξ_{\tilde Q}$ has the single Kraus operator $\tilde M_{\mathrm{kraus}} = \tilde Q$ with $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}} = e_t$. It is therefore completely positive and trace preserving, and it is exactly the class of channels that are invertible within the channels, with inverse $Ξ^{-1}_{\tilde Q} = Ξ_{\tilde Q^{*}}$ (whose Kraus operator is $\tilde Q^{*}$); it is also exactly the class that preserves purity, $2\,\mathrm{Sc}(Ξ_{\tilde Q}(\tilde{Ϛ}_{\Psi})^2) = 2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}^2)$. A gate is a reversible process, and conversely every reversible process of the informational sector is a gate.

Why Matrix-Unitarity and Not Unit Norm

The condition $\tilde Q\tilde Q^{*} = e_t$ must not be confused with the unit-norm condition $\tilde U_{\mathrm{lor}}\tilde U_{\mathrm{lor}}{}^{\natural} = e_t$ that defines the Lorentz rotors $SL(2,\mathbb{C})$. The two conditions pick out two different subgroups of the same algebra, and only the first is a group of gates.

The distinction is visible in a single example. A boost biquaternion

$$ \tilde U_{\Psi,\mathrm b} = \cosh\frac{\theta_{\mathrm{rpd}}}{2} + i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b} $$

is Hermitian, $\tilde U_{\Psi,\mathrm b}^{*} = \tilde U_{\Psi,\mathrm b}$, and has unit norm, $\tilde U_{\Psi,\mathrm b}\tilde U_{\Psi,\mathrm b}^{\natural} = e_t$. Here $\theta_{\mathrm{rpd}}$ is the rapidity, $\tanh\theta_{\mathrm{rpd}}=v/c$, real at every speed below $c$. Hence $\tilde U_{\Psi,\mathrm b}\tilde U_{\Psi,\mathrm b}^{*} = \tilde U_{\Psi,\mathrm b}^2 \neq e_t$, and conjugation by $\tilde U_{\Psi,\mathrm b}$ does not preserve the trace:

$$ 2\,\mathrm{Sc}\!\left(\tilde U_{\Psi,\mathrm b}\tilde{Ϛ}_{\Psi}\tilde U_{\Psi,\mathrm b}^{*}\right) \neq 2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}) \quad \text{in general}. $$

A boost is a rotor on the material sector $\mathbb{B}_{\Phi}$ and not a gate on the informational sector $\mathbb{B}_{\Psi}$. The distinction is not a technicality: it is the algebraic expression of the fact that the Lorentz group and the gate group are different groups inside one algebra, related to the two complementary subspaces.

The two actions are collected here for contrast.

Material sector (Lorentz) Informational sector (gates)
Acting element $\tilde U_{\mathrm{lor}}\in\mathbb{B}$, $\tilde U_{\mathrm{lor}}\tilde U_{\mathrm{lor}}{}^{\natural} = e_t$ $\tilde Q\in\mathbb{B}$, $\tilde Q\tilde Q^{*} = e_t$
Group $SL(2,\mathbb{C})$ $\mathrm{U}(2)$
Acting on $\tilde{Q}\in\mathbb{B}_{\Phi}$ $\tilde{Ϛ}_{\Psi}\in\mathbb{B}_{\Psi}$
Action $\tilde{Q}\mapsto\tilde U_{\mathrm{lor}}\tilde{Q}\tilde U_{\mathrm{lor}}{}^{*}$ $\tilde{Ϛ}_{\Psi}\mapsto\tilde Q\tilde{Ϛ}_{\Psi}\tilde Q^{*}$
Pure boost Hermitian, in $\mathbb{B}_{\Psi}$, not $*$-unitary not a gate

The Structure of the Gate Group

Every $*$-unitary biquaternion factorises into a central phase and a unit real quaternion:

$$ \mathrm{U}(2) \;=\; \mathrm{U}(1)\cdot SU(2), \qquad \tilde Q = \exp(i\theta_{\mathrm{pha}})\,\tilde u_{\mathrm r}, \qquad \theta_{\mathrm{pha}}\in\mathbb{R}, \quad \tilde u_{\mathrm r}\in\mathbb{B}_{H}, \quad \tilde u_{\mathrm r}\tilde u_{\mathrm r}^{\natural} = e_t . $$

Under the isomorphism, the unit real quaternions are exactly the image of $SU(2) \subset M_2(\mathbb{C})$, since $\langle\tilde u_{\mathrm r}\rangle_{\natural} = \tilde u_{\mathrm r}\tilde u_{\mathrm r}^{\natural} = 1$. The phase $\exp(i\theta_{\mathrm{pha}})$ lies in the centre $\mathbb{B}_{C}$, and $i\mathbb{B}_{\Psi} = \mathbb{B}_{\Phi}$ for the anti-Hermitian generators.

The phase is not observable in the action. Conjugation by a central element is trivial:

$$ (\exp(i\theta_{\mathrm{pha}})\tilde u_{\mathrm r})\,\tilde{Ϛ}_{\Psi}\,(\exp(i\theta_{\mathrm{pha}})\tilde u_{\mathrm r})^{*} = \exp(i\theta_{\mathrm{pha}})\tilde u_{\mathrm r}\,\tilde{Ϛ}_{\Psi}\,\tilde u_{\mathrm r}^{*} \exp(-i\theta_{\mathrm{pha}}) = \tilde u_{\mathrm r}\,\tilde{Ϛ}_{\Psi}\,\tilde u_{\mathrm r}^{*} . $$

Hence the kernel of $\tilde Q\mapstoΞ_{\tilde Q}$ is the unit circle $\mathrm{U}(1)\subset\mathbb{B}_{C}$, and the effective group of gates is

$$ \boxed{\;\mathrm{U}(2)/\mathrm{U}(1) \;\cong\; PU(2) \;\cong\; SO(3)\;}, $$

acting on the Bloch ball by rotations. This is the algebraic form of the unobservability of the global phase: two unitaries that differ by a central phase define the same gate.

The action on the Bloch vector. Writing the state as $\tilde{Ϛ}_{\Psi} = \tfrac{1}{2}(e_t + i\vec s_{\Psi})$, a gate acts as

$$ \vec s_{\Psi} \;\longmapsto\; R(\tilde Q)\,\vec s_{\Psi}, \qquad R(\tilde Q) \in SO(3), $$

a proper rotation of $\mathbb{R}^3$. The scalar part of the state is fixed, the norm $|\vec s_{\Psi}|$ is preserved, and with it the purity $2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}^2) = \tfrac{1}{2}(1 + |\vec s_{\Psi}|^2)$ and the spectrum $\lambda_\pm = \tfrac{1}{2}(1\pm|\vec s_{\Psi}|)$. A gate rotates the Bloch ball rigidly; it never contracts it.

Rotation gates. The unit real quaternion with axis $\vec u_{\mathrm r}$ and angle $\theta_{\mathrm{axr}}$,

$$ \tilde u_{\mathrm r}(\theta_{\mathrm{axr}}) = \cos\frac{\theta_{\mathrm{axr}}}{2}\,e_t + \sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r}, \qquad \vec u_{\mathrm r}^2 = -e_t , $$

acts on the Bloch vector as the rotation by $\theta_{\mathrm{axr}}$ about $\vec u_{\mathrm r}$ (the Rodrigues formula). The half-angle is the double cover: $\tilde u_{\mathrm r}(\theta_{\mathrm{axr}} + 2\pi) = -\tilde u_{\mathrm r}(\theta_{\mathrm{axr}})$, which is the same rotation and, since $-e_t$ is a central phase, the same gate. The general gate is $\exp(i\theta_{\mathrm{pha}})\tilde u_{\mathrm r}(\theta_{\mathrm{axr}})$, and modulo the phase it is a rotation.

Involutions: The Hermitian Unitaries

A $*$-unitary can be Hermitian, $\tilde Q = \tilde Q^{*}$. Then $\tilde Q^2 = \tilde Q\tilde Q^{*} = e_t$, so a Hermitian $*$-unitary is an involution. Apart from $\pm e_t$, the Hermitian unitaries are exactly

$$ \tilde Q = \pm\, i\vec u_{\mathrm r}, \qquad \vec u_{\mathrm r} \text{ a unit pure real quaternion}, $$

since $(i\vec u_{\mathrm r})^{*} = -i\vec u_{\mathrm r}^{*} = i\vec u_{\mathrm r}$ and $(i\vec u_{\mathrm r})^2 = -\vec u_{\mathrm r}^2 = e_t$. They lie in $\mathbb{B}_{\Psi}$, and as rotations they are the $\pi$-rotations of the Bloch sphere: conjugation by $i\vec u_{\mathrm r}$ sends $\vec s_{\Psi}\mapsto 2(\vec u_{\mathrm r}\cdot\vec s_{\Psi})\vec u_{\mathrm r} - \vec s_{\Psi}$. The Pauli gates and the Hadamard gate below are of this type; the phase and $T$ gates are not.

A caution about representatives. A gate is a coset $\tilde Q\cdot \mathrm{U}(1)$; the representative can be chosen Hermitian only when the gate is an involution. Membership of the representative in $\mathbb{B}_{\Psi}$ or in $\mathbb{B}_{H}$ is therefore a property of the chosen representative, not an invariant of the gate. The one representative-independent statement is the factorisation $\tilde Q = \exp(i\theta_{\mathrm{pha}})\tilde u_{\mathrm r}$: every gate is a central phase times a rotation, and no gate is anything else.

The Standard Single-Qubit Gate Set

Under the isomorphism $\vec e_\eta \mapsto -i\sigma_\eta$, the image of $i \vec e_\eta$ is $\sigma_\eta$. The standard gate set therefore reads as follows. In each row the second column is a biquaternion representative; a representative differing by a central phase gives the same gate.

Gate Biquaternion representative Matrix image Action on $\vec s_{\Psi}=(r_1,r_2,r_3)$
$X$ (Pauli) $\tilde Q_x= i \vec e_x$ $\sigma_1$ $(r_1,-r_2,-r_3)$
$Y$ (Pauli) $\tilde Q_y= i \vec e_y$ $\sigma_2$ $(-r_1,r_2,-r_3)$
$Z$ (Pauli) $\tilde Q_z= i \vec e_z$ $\sigma_3$ $(-r_1,-r_2,r_3)$
$H$ (Hadamard) $\tilde Q_{\mathrm{Had}} = \dfrac{i}{\sqrt{2}}\left(\vec e_x+\vec e_z\right)$ $\dfrac{1}{\sqrt{2}}\begin{pmatrix}1&1\\1&-1\end{pmatrix}$ $(r_3,-r_2,r_1)$
$S$ (phase) $\tilde Q_{\mathrm{ph}} = \dfrac{1}{\sqrt{2}}\left(e_t+\vec e_z\right)$ $\operatorname{diag}(\exp(-i\pi/4),\exp(i\pi/4))$ $(-r_2,r_1,r_3)$
$T$ ($\pi/8$) $\tilde Q_{T} = \cos\dfrac{\pi}{8}e_t + \sin\dfrac{\pi}{8}\vec e_z$ $\operatorname{diag}(\exp(-i\pi/8),\exp(i\pi/8))$ rotation by $\pi/4$ about $\vec e_z$
$R_{\vec u_{\mathrm r}}(\theta_{\mathrm{axr}})$ $\cos\dfrac{\theta_{\mathrm{axr}}}{2}e_t + \sin\dfrac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ $\exp(-i\theta_{\mathrm{axr}}\,\vec u_{\mathrm r}\cdot\boldsymbol{\sigma}/2)$ rotation by $\theta_{\mathrm{axr}}$ about $\vec u_{\mathrm r}$

The matrix column is the image under $\mathbb{B}\cong M_2(\mathbb{C})$. For $S$ and $T$ the displayed matrices are $\operatorname{diag}(\exp(-i\pi/4),\exp(i\pi/4))$ and $\operatorname{diag}(\exp(-i\pi/8),\exp(i\pi/8))$, which are the standard $S$ and $T$ up to the central phases $\exp(i\pi/4)$ and $\exp(i\pi/8)$; the biquaternion representatives shown are exact, and their matrix images differ from the standard matrices by those phases. This is the representative-dependence noted above.

The Pauli Gates

The three Pauli gates are the Hermitian unitaries along the coordinate axes:

$$ \tilde Q_x= i \vec e_x, \qquad \tilde Q_y= i \vec e_y, \qquad \tilde Q_z= i \vec e_z . $$

Each satisfies $\tilde Q_x^2 = \tilde Q_y^2 = \tilde Q_z^2 = e_t$, so each is an involution and a $\pi$-rotation; the actions on the Bloch vector in the table are the corresponding reflections through the coordinate axes. Their products reproduce the Pauli algebra as an algebra identity:

$$ \tilde Q_x\tilde Q_y= i\tilde Q_z, \qquad \tilde Q_y\tilde Q_z= i\tilde Q_x, \qquad \tilde Q_z\tilde Q_x= i\tilde Q_y, \qquad \tilde Q_x\tilde Q_z= -\tilde Q_z\tilde Q_x, \qquad \tilde Q_x\tilde Q_y\tilde Q_z= i e_t . $$

The central factor $i e_t$ is the price of writing the Pauli matrices in a basis of real quaternions: the standard Pauli matrices obey $\sigma_1\sigma_2 = i\sigma_3$ with the same scalar imaginary, and the isomorphism carries that scalar imaginary to the central element $i e_t$ of $\mathbb{B}$. The sign of the product is what distinguishes the two orders, $\tilde Q_x\tilde Q_z= -\tilde Q_z\tilde Q_x$: this is the algebraic origin of the non-commutativity that makes gates into a group rather than a set.

The Hadamard Gate

The Hadamard gate is the Hermitian $*$-unitary along the diagonal axis:

$$ \tilde Q_{\mathrm{Had}} = \frac{i}{\sqrt{2}}\left(\vec e_x + \vec e_z\right), \qquad \tilde Q_{\mathrm{Had}}^2 = e_t, \qquad \tilde Q_{\mathrm{Had}}^{*} = \tilde Q_{\mathrm{Had}}. $$

It is the $\pi$-rotation about $\vec u_{\mathrm r} = (\vec e_x+\vec e_z)/\sqrt{2}$, so on the Bloch vector it exchanges the $\vec e_x$ and $\vec e_z$ axes and reverses $\vec e_y$:

$$ \tilde Q_{\mathrm{Had}}\,\tilde Q_z\,\tilde Q_{\mathrm{Had}} = \tilde Q_x, \qquad \tilde Q_{\mathrm{Had}}\,\tilde Q_x\,\tilde Q_{\mathrm{Had}} = \tilde Q_z, \qquad \tilde Q_{\mathrm{Had}}\,\tilde Q_y\,\tilde Q_{\mathrm{Had}} = -\tilde Q_y. $$

The first two are exact identities among elements of $\mathbb{B}$; the third carries a sign, and since $-e_t$ is central, $\tilde Q_{\mathrm{Had}}\tilde Q_y\tilde Q_{\mathrm{Had}} = -\tilde Q_y$ and $\tilde Q_y$ define the same gate. This is a small but representative point: an algebraic identity of representatives need not be an identity of gates, and the discrepancy is always a central phase.

A different complex Hadamard matrix. A second object shares the name but not the setting. In The Chiral Algebra of Biquaternions and the Cyclic Representation of the Dirac Equation, the cyclic conjugation of the biquaternion wave function, read on the Cartesian components $(f_x, f_y, f_z, f_t)$, is a $4\times4$ matrix with entries in $\{\pm1, \pm i\}$ and $H_4H_4^\dagger = 4I$. Its entries have modulus one, so $H_4/2$ is $*$-unitary on a four-dimensional complex space; it is a transvection on that space, not an element of the two-dimensional gate group treated here. The two Hadamards therefore share the modulus-one property and the name, and nothing else: $\tilde Q_{\mathrm{Had}}$ above is a qubit gate, the printed $H_4$ is a change of coordinates of the Dirac equation.

The Phase and $T$ Gates

The phase gate is the rotation by $\pi/2$ about $\vec e_z$:

$$ \tilde Q_{\mathrm{ph}} = \frac{1}{\sqrt{2}}\left(e_t + \vec e_z\right) = \tilde u_{e,3}\!\left(\frac{\pi}{2}\right), \qquad \tilde Q_{\mathrm{ph}}^2 = \vec e_z . $$

The element $\vec e_z$ is not $Z = i \vec e_z$; it differs from it by the central phase $-i$, so $\tilde Q_{\mathrm{ph}}^2$ and $\tilde Q_z$ define the same gate. In the standard normalisation, $S^2 = Z$ exactly; in the biquaternion representative, $S^2 = Z$ up to a central phase, which is the honest statement.

The $T$ gate is the rotation by $\pi/4$ about $\vec e_z$:

$$ \tilde Q_{T} = \cos\frac{\pi}{8}\,e_t + \sin\frac{\pi}{8}\,\vec e_z = \tilde u_{e,3}\!\left(\frac{\pi}{4}\right), \qquad \tilde Q_{T}^2 = \tilde Q_{\mathrm{ph}}, \qquad \tilde Q_{T}^{8} = -e_t, \qquad \tilde Q_{T}^{16} = e_t . $$

Thus $T$ has order eight as a gate and order sixteen as an element of $SU(2)$. The relation $T^2 = S$ and the order were recomputed. Both $S$ and $T$ are represented by unit real quaternions, i.e. by elements of the rotation group, and neither is Hermitian; this is why their biquaternion representatives lie in $\mathbb{B}_{H}$ rather than in $\mathbb{B}_{\Psi}$, in contrast with the Pauli and Hadamard gates.

Composition of Gates as Algebraic Product

The Product Law

Let $\tilde Q_{\mathrm{gate},1}$ and $\tilde Q_{\mathrm{gate},2}$ be gates, and consider the physical process that applies $\tilde Q_{\mathrm{gate},1}$ first and $\tilde Q_{\mathrm{gate},2}$ second. The state passes through

$$ \tilde{Ϛ}_{\Psi} \;\xrightarrow{\ \tilde Q_{\mathrm{gate},1}\ }\; \tilde Q_{\mathrm{gate},1}\tilde{Ϛ}_{\Psi}\tilde Q_{\mathrm{gate},1}^{*} \;\xrightarrow{\ \tilde Q_{\mathrm{gate},2}\ }\; \tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1}\tilde{Ϛ}_{\Psi}\tilde Q_{\mathrm{gate},1}^{*}\tilde Q_{\mathrm{gate},2}^{*} = (\tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1})\,\tilde{Ϛ}_{\Psi}\,(\tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1})^{*} . $$

So the composite gate is the product $\tilde Q_{\mathrm{gate},3} = \tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1}$, and the map on channels is

$$ \boxed{\;Ξ_{\tilde Q_{\mathrm{gate},2}} \circ Ξ_{\tilde Q_{\mathrm{gate},1}} = Ξ_{\tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1}}\;}. $$

The product is the biquaternion product (for several qubits, the product in $\mathbb{B}^{\otimes n}$). Because that product is associative, a circuit is well defined without any bracketing convention: it is one element, the ordered product of its gates, and its action is a single conjugation by that element. The identity gate is $e_t$; the inverse of $\tilde Q$ is $\tilde Q^{*} = \tilde Q^{-1}$; and the set of gates is closed under the product, which is the statement that the gate group is a group.

Two remarks make the product law precise.

First, the order is the reverse of the circuit interpretation. If a diagram is read left to right, the algebraic product is written right to left, as above. This is the standard convention for operator composition and carries no biquaternion-specific content.

Second, because the map $\tilde Q\mapstoΞ_{\tilde Q}$ has kernel $\mathrm{U}(1)$, the product law holds at the level of gates but the algebraic identity of representatives may fail by a central phase. If $\tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1} = \exp(i\theta_{\mathrm{pha}})\tilde Q_{\mathrm{gate},3}$, then $Ξ_{\tilde Q_{\mathrm{gate},2}}\circΞ_{\tilde Q_{\mathrm{gate},1}} = Ξ_{\tilde Q_{\mathrm{gate},3}}$ nevertheless. Circuit identities in the framework are therefore identities up to a central phase, and the phase is unobservable.

Worked Identities

The Hadamard conjugation of the Pauli gates is a compact illustration. The identity $\tilde Q_{\mathrm{Had}}\tilde Q_z\tilde Q_{\mathrm{Had}} = \tilde Q_x$ already displayed says that the circuit "Hadamard, then $Z$, then Hadamard" realises the $X$ gate:

$$ Ξ_{\tilde Q_{\mathrm{Had}}}\circΞ_{\tilde Q_z}\circΞ_{\tilde Q_{\mathrm{Had}}} = Ξ_{\tilde Q_{\mathrm{Had}}\tilde Q_z\tilde Q_{\mathrm{Had}}} = Ξ_{\tilde Q_x}. $$

The identity $\tilde Q_{\mathrm{Had}}\tilde Q_x\tilde Q_{\mathrm{Had}} = \tilde Q_z$ gives the reverse. These are exact as algebra identities, and they hold as gate identities. The identity $\tilde Q_{\mathrm{Had}}\tilde Q_y\tilde Q_{\mathrm{Had}} = -\tilde Q_y$ holds only up to the central phase $-e_t$; as gates, $Ξ_{\tilde Q_{\mathrm{Had}}}\circΞ_{\tilde Q_y}\circΞ_{\tilde Q_{\mathrm{Had}}} = Ξ_{\tilde Q_y}$.

Non-commuting gates give distinct circuits. For example, the two orderings of a Hadamard and a controlled gate differ, as they do in standard circuit notation; the framework expresses this as the failure of $\tilde Q_{\mathrm{Had}}\otimes e_t$ and the controlled gate to commute in $\mathbb{B}\otimes\mathbb{B}$. Composition is the algebra product, and the algebra is non-commutative.

The Clifford Group as an Algebraic Normaliser

The single-qubit Pauli group is the sixteen-element set

$$ \mathcal{P} = \{\pm e_t,\ \pm i e_t,\ \pm \vec e_\eta,\ \pm i \vec e_\eta : \eta = 1, 2, 3\}, $$

which is closed under the biquaternion product and is the image of the standard Pauli group $\langle X,Y,Z\rangle$ under the isomorphism. The Clifford group is its normaliser in the gate group,

$$ \mathcal{C} = \{\tilde Q\in \mathrm{U}(2) : \tilde Q\,\mathcal{P}\,\tilde Q^{*} = \mathcal{P}\}, $$

i.e. the set of gates that permute the Pauli group under conjugation, up to sign. The elementary gates $H$ and $S$ both lie in it, and in fact generate it; this is the biquaternion form of the standard statement that $H$ and $S$ generate the single-qubit Clifford group. Modulo the phase, the conjugation action of the Clifford group on the Bloch sphere is the rotation symmetry group of the octahedron, of order $24$; a direct enumeration of the group generated by the two rotation matrices $R(\tilde Q_{\mathrm{Had}})$ and $R(\tilde Q_{\mathrm{ph}})$ returns exactly $24$ distinct rotations. In the algebra, the octahedral symmetry is the symmetry of the Pauli axes $\{\pm i \vec e_x,\pm i \vec e_y,\pm i \vec e_z\}$ that the Clifford gates permute.

The $T$ gate is not Clifford. Its conjugation of $\tilde Q_x$ produces a rotation by $\pi/4$ about $\vec e_z$, which does not return the Pauli axes to themselves. The standard statement that $\{H,T\}$ generates a dense subgroup of the gate group — the basis of the Solovay–Kitaev universality of single-qubit computation — is inherited unchanged by the reformulation; it is a statement about the group generated by a Hadamard involution and a rotation of order eight, and the biquaternion representatives are $i(\vec e_x+\vec e_z)/\sqrt{2}$ and $\cos(\pi/8)e_t+\sin(\pi/8)\vec e_z$. This article does not re-derive universality; it records where the generators live.

Multi-Qubit Gates and the Tensor Arena

The $n$-Qubit Algebra

For $n$ qubits the arena is the tensor product

$$ \mathbb{B}^{\otimes n} \;\cong\; M_{2^n}(\mathbb{C}), $$

with the trace $\mathrm{Tr}\bigl(\mathsf M_2(x)\otimes\mathsf M_2(y)\bigr) = \mathrm{Tr}_\mathbb{B}\mathsf M_2(x)\cdot\mathrm{Tr}_\mathbb{B}\mathsf M_2(y)$, where $\mathrm{Tr}_{\mathbb{B}}\mathsf M_2(e_t)=2$ and $\mathrm{Tr}_{\mathbb{B}}\mathsf M_2(\vec e_\eta)=0$, and with the partial traces $\mathrm{Tr}_1,\mathrm{Tr}_2$ of the companion article on entangled subsystems. A gate is a $*$-unitary element of $\mathbb{B}^{\otimes n}$, acting on states (Hermitian, positive, unit-weight elements of $\mathbb{B}_{\Psi}^{\otimes n}$) by conjugation. The elementary gates of a circuit are of two kinds: local gates $\tilde Q_1\otimes\tilde Q_2\otimes\cdots$ on the factors, and entangling gates, which are not tensor products.

Local gates on different factors commute, $(\tilde Q_{\mathrm{gate},1}\otimes e_t)(e_t\otimes \tilde Q_{\mathrm{gate},2}) = (e_t\otimes \tilde Q_{\mathrm{gate},2})(\tilde Q_{\mathrm{gate},1}\otimes e_t) = \tilde Q_{\mathrm{gate},1}\otimes \tilde Q_{\mathrm{gate},2}$, so the order of operations on disjoint wires is immaterial — the algebraic statement of the fact that commuting gates may be drawn in either order.

The status of the tensor product is the open question inherited from the companion articles. Everything below is a statement inside $\mathbb{B}^{\otimes n}$ once that algebra is granted; whether it is canonical for the framework is left open, exactly as in the companion article on the Bell basis.

Controlled Gates

A controlled gate is built from the two fundamental objects of the framework: an idempotent on the control factor and a gate on the target factor. Along the control axis $\vec u_{\mathrm r}$, let $\tilde\Pi_{1,2}(\vec u_{\mathrm r}) = \tfrac{1}{2}(e_t \pm i\vec u_{\mathrm r})$ be the complementary idempotents, and let $\tilde Q$ be any single-qubit gate. Define

$$ \tilde{Q}_{\tilde Q} = \tilde\Pi_1(\vec u_{\mathrm r})\otimes e_t + \tilde\Pi_2(\vec u_{\mathrm r})\otimes \tilde Q \;\in\; \mathbb{B}\otimes\mathbb{B}. $$

This is a single element of the two-qubit algebra, and it is $*$-unitary:

$$ \tilde{Q}_{\tilde Q}\tilde{Q}_{\tilde Q}^{*} = \tilde\Pi_1^2\otimes e_t + \tilde\Pi_1\tilde\Pi_2\otimes\tilde Q^{*} + \tilde\Pi_2\tilde\Pi_1\otimes\tilde Q + \tilde\Pi_2^2\otimes\tilde Q\tilde Q^{*} = (\tilde\Pi_1 + \tilde\Pi_2)\otimes e_t = e_t\otimes e_t , $$

where the cross terms vanish because $\tilde\Pi_1\tilde\Pi_2 = 0$ and the last step uses the resolution of the identity $\tilde\Pi_1 + \tilde\Pi_2 = e_t$ together with $\tilde Q\tilde Q^{*} = e_t$. The same computation with a general single-qubit $\tilde Q$ shows that any $*$-unitary target operation is admitted, so the construction defines $\tilde{Q}_{\tilde Q}$ for every gate $\tilde Q$.

The unitarity of the controlled gate uses exactly two properties: the idempotents are complementary and orthogonal, and the target element is $*$-unitary. It is worth pausing on this, because the same idempotents, used differently, are the operators of an irreversible measurement. The controlled gate is the coherent use of an idempotent: the two outcomes are kept, paired with two different target operations, and added inside a single $*$-unitary element. The measurement channel, in contrast, adds the two conjugated states,

$$ \tilde{Ϛ}_{\Psi} \;\longmapsto\; \tilde\Pi_1\tilde{Ϛ}_{\Psi}\tilde\Pi_1 + \tilde\Pi_2\tilde{Ϛ}_{\Psi}\tilde\Pi_2 , $$

which is a sum of conjugations and not a single conjugation. This is the algebraic distinction between a reversible gate and an irreversible measurement, and it is taken up again in the section on channels.

CNOT, CZ, and SWAP

With the control axis $\vec u_z$ — the $|0\rangle/|1\rangle$ basis of the control — the standard gates are:

Gate Biquaternion representative in $\mathbb{B}\otimes\mathbb{B}$ Matrix image
$\mathrm{CNOT}$ $\tilde\Pi_1(\vec u_z)\otimes e_t + \tilde\Pi_2(\vec u_z)\otimes (i \vec e_x)$ $\lvert 0\rangle\langle 0\rvert\otimes I_2 + \lvert 1\rangle\langle 1\rvert\otimes\sigma_1$
$\mathrm{CZ}$ $\tilde\Pi_1(\vec u_z)\otimes e_t + \tilde\Pi_2(\vec u_z)\otimes (i \vec e_z)$ $\lvert 0\rangle_{*}\langle 0\rvert\otimes I_2 + \lvert 1\rangle\langle 1\rvert\otimes\sigma_3$
$\mathrm{SWAP}$ $\dfrac{1}{2}\left(e_t\otimes e_t - \vec e_x\otimes \vec e_x - \vec e_y\otimes \vec e_y - \vec e_z\otimes \vec e_z\right)$ $\tfrac{1}{2}\left(I_4 + \sum_\eta \sigma_\eta\otimes\sigma_\eta\right)$

All three were verified to be $*$-unitary, Hermitian, and involutive. The CNOT is the controlled-$X$; the CZ is the controlled-$Z$. The SWAP form follows from the standard identity $\mathrm{SWAP} = \tfrac{1}{2}(I\otimes I + \sum_\eta\sigma_\eta\otimes\sigma_\eta)$ together with $\vec e_\eta\otimes \vec e_\eta = -(i\vec e_\eta)\otimes(i\vec e_\eta) = -\sigma_\eta\otimes\sigma_\eta$ under the isomorphism. The SWAP is not a controlled gate but a permutation of the two factors, and it satisfies the standard factor-exchange relation

$$ (\tilde{Q}_{12})\,\mathrm{SWAP} = \mathrm{SWAP}\,(\tilde{Q}_{21}), $$

where $\tilde{Q}_{12}$ is the CNOT with control on the first qubit and $\tilde{Q}_{21}$ the CNOT with control on the second; this too was recomputed. It is the algebraic statement that swapping the qubits converts a control on one into a control on the other.

Two standard circuit identities express the relations among these gates, and both were verified in the tensor algebra:

$$ \mathrm{SWAP} = \tilde{Q}_{12}\,\tilde{Q}_{21}\,\tilde{Q}_{12}, \qquad \tilde{Q}_{12} = (I_2\otimes \tilde Q_{\mathrm{Had}})\,\mathrm{CZ}\,(I_2\otimes \tilde Q_{\mathrm{Had}}). $$

The first is the three-CNOT decomposition of the swap. The second is the conjugation of the CZ into the CNOT by a Hadamard on the target; equivalently, conjugating the CNOT by the same Hadamard returns the CZ. Each identity is a statement about the product of $*$-unitary elements of $\mathbb{B}\otimes\mathbb{B}$: the circuits are equal because the algebra products are equal.

A Simple Circuit: Preparing a Bell State

The composition law is best seen in a circuit that produces something the algebra names. Consider two qubits prepared in the pure state

$$ \tilde{Ϛ}_{\Psi,0} = \tilde\Pi_1(\vec u_z)\otimes\tilde\Pi_1(\vec u_z) , $$

the biquaternion idempotent corresponding to $|00\rangle_{*}$. Apply the Hadamard to the first qubit and then the CNOT with the first qubit as control. As a circuit, the operation is the product

$$ \tilde Q_{\mathrm{cir}} = \tilde{Q}_{\tilde Q_x}(\tilde Q_{\mathrm{Had}}\otimes e_t), \qquad \tilde{Q}_{\tilde Q_x} = \tilde\Pi_1(\vec u_z)\otimes e_t + \tilde\Pi_2(\vec u_z)\otimes (i \vec e_x), $$

and the output state is $\tilde Q_{\mathrm{cir}}\tilde{Ϛ}_{\Psi,0}\tilde Q_{\mathrm{cir}}^{*}$, a single conjugation because $\tilde Q_{\mathrm{cir}}$ is a single gate.

Step one: the Hadamard. Since $\tilde Q_{\mathrm{Had}}$ is Hermitian and $\tilde Q_{\mathrm{Had}}^2 = e_t$, the conjugation of the first factor gives

$$ \tilde Q_{\mathrm{Had}}\,\tilde\Pi_1(\vec u_z)\,\tilde Q_{\mathrm{Had}} = \tilde\Pi_1(\vec u_x), $$

computed directly from the quaternion product. So after the Hadamard the state is

$$ \left(\tilde Q_{\mathrm{Had}}\otimes e_t\right)\tilde{Ϛ}_{\Psi,0}\left(\tilde Q_{\mathrm{Had}}\otimes e_t\right) = \tilde\Pi_1(\vec u_x)\otimes\tilde\Pi_1(\vec u_z), $$

the product state $|{+}\rangle|0\rangle$.

Step two: the CNOT. Write $\tilde\Pi_{\mathrm x} = \tilde\Pi_1(\vec u_x)$ for the control state and $\tilde\Pi_{\mathrm z} = \tilde\Pi_1(\vec u_z)$ for the target state, with $\tilde\Pi_{1,2} = \tilde\Pi_{1,2}(\vec u_z)$ the control idempotents; set $\tilde{Q}_{12} = \tilde{Q}_{\tilde Q_x}$. Since $\tilde{Q}_{12} = \tilde{Q}_{12}^{*}$ and $\tilde{Q}_{12}^2 = e_t\otimes e_t$, conjugation by $\tilde{Q}_{12}$ expands into four terms:

$$ \tilde{Q}_{12}(\tilde\Pi_{\mathrm x}\otimes\tilde\Pi_{\mathrm z})\tilde{Q}_{12} = \tilde\Pi_1\tilde\Pi_{\mathrm x}\tilde\Pi_1 \otimes \tilde\Pi_{\mathrm z} + \tilde\Pi_1\tilde\Pi_{\mathrm x}\tilde\Pi_2 \otimes \tilde\Pi_{\mathrm z}\,(i\vec e_x) + \tilde\Pi_2\tilde\Pi_{\mathrm x}\tilde\Pi_1 \otimes (i\vec e_x)\,\tilde\Pi_{\mathrm z} + \tilde\Pi_2\tilde\Pi_{\mathrm x}\tilde\Pi_2 \otimes (i\vec e_x)\,\tilde\Pi_{\mathrm z}\,(i\vec e_x). $$

The four idempotents evaluate, using $\tilde\Pi_{1,2}\tilde\Pi_{\mathrm x}\tilde\Pi_{1,2} = \tfrac{1}{2}\tilde\Pi_{1,2}$ (the transition probability $\tfrac{1}{2}(1 + \vec u_z\cdot\vec u_x) = \tfrac12$) and the Peirce components

$$ \tilde\Pi_1\tilde\Pi_{\mathrm x}\tilde\Pi_2 = \tfrac{1}{4}\left(i\vec e_x - \vec e_y\right), \qquad \tilde\Pi_2\tilde\Pi_{\mathrm x}\tilde\Pi_1 = \tfrac{1}{4}\left(i\vec e_x + \vec e_y\right), $$

while the target products are $\tilde\Pi_{\mathrm z}(i\vec e_x) = \tfrac12(i\vec e_x - \vec e_y)$ and $(i\vec e_x)\tilde\Pi_{\mathrm z} = \tfrac12(i\vec e_x + \vec e_y)$, and $(i\vec e_x)\tilde\Pi_{\mathrm z}(i\vec e_x) = \tilde\Pi_2$. Substituting,

$$ \tilde{Q}_{12}(\tilde\Pi_{\mathrm r_x}\otimes\tilde\Pi_{\mathrm r_z})\tilde{Q}_{12} = \tfrac{1}{4}\left[e_t\otimes e_t - \vec e_x\otimes \vec e_x + \vec e_y\otimes \vec e_y - \vec e_z\otimes \vec e_z\right]. $$

The right-hand side is the Bell idempotent $\tilde\Pi_\epsilon$ of the companion article on the Bell basis, with sign pattern $\epsilon = (-1,+1,-1)$, which is the idempotent of $|Ξ^{+}\rangle$. The identification was verified independently by mapping both sides to $M_4(\mathbb{C})$: the biquaternion element equals $\tfrac{1}{2}(|00\rangle+|11\rangle)(\langle 00|+\langle 11|)$.

The whole circuit is thus the single element $\tilde Q_{\mathrm{cir}} = \tilde{Q}_{\tilde Q_x}(\tilde Q_{\mathrm{Had}}\otimes e_t)$ of $\mathbb{B}\otimes\mathbb{B}$, and its output is

$$ \boxed{\;\tilde Q_{\mathrm{cir}}\left[\tilde\Pi_1(\vec u_z)\otimes\tilde\Pi_1(\vec u_z)\right]\tilde Q_{\mathrm{cir}}^{*} = \tfrac{1}{4}\left[e_t\otimes e_t - \vec e_x\otimes \vec e_x + \vec e_y\otimes \vec e_y - \vec e_z\otimes \vec e_z\right]\;}. $$

The output is not a product of idempotents; its partial trace is $\tfrac12 e_t$, the maximally mixed qubit state, by the computation of the companion article. The circuit has converted a product state into a maximally entangled state by a single conjugation, and every step of the conversion is an algebraic product and conjugation in $\mathbb{B}\otimes\mathbb{B}$.

Gates and the Irreversible Channels

The read list for this article closes with Quantum Channels and the Reversible/Irreversible Dichotomy, and the contrast between the two is the conceptual content of the gate formalism.

A state map is a completely positive, trace-preserving map on $\mathbb{B}_{\Psi}$, extended complex-linearly to $\mathbb{B} = \mathbb{B}_{\Psi} \oplus \mathbb{B}_{\Phi}$ and written in Kraus form $Ξ(\tilde{Ϛ}_{\Psi}) = \sum_l \tilde M_{\mathrm{kraus},l}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},l}^{*}$ with $\sum_l\tilde M_{\mathrm{kraus},l}^{*}\tilde M_{\mathrm{kraus},l} = e_t$. The dichotomy is:

$$ \text{reversible} \quad\Longleftrightarrow\quad \text{Kraus rank one} \quad\Longleftrightarrow\quad Ξ = Ξ_{\tilde Q}\ \text{for a }{}^{*}\text{-unitary } \tilde Q \quad\Longleftrightarrow\quad Ξ \text{ is a gate.} $$

Equivalently, a channel is a gate if and only if it preserves purity, and if and only if it is an inner automorphism of the algebra. Three consequences organise the distinction.

Closure. The composition of two gates is a gate, because the product of two unitaries is $*$-unitary. Hence a circuit built only from gates is a gate: it is one element, it has Kraus rank one, and it is invertible. The inverse circuit is the product of the inverses in the reverse order, which in the algebra is just $\tilde Q_{\mathrm{cir}}^{*}$. The Bell-preparation circuit above is reversible for exactly this reason, even though it creates entanglement: entanglement is not irreversibility.

Irreversibility is a sum. A channel is irreversible precisely when its Kraus rank is at least two, i.e. when it is a sum of conjugations rather than one conjugation. The canonical example is dephasing along $\vec u_{\mathrm r}$,

$$ Ξ_{\mathrm{deph},p}(\tilde{Ϛ}_{\Psi}) = (1-p)\tilde{Ϛ}_{\Psi} + p\left(\tilde\Pi_1\tilde{Ϛ}_{\Psi}\tilde\Pi_1 + \tilde\Pi_2\tilde{Ϛ}_{\Psi}\tilde\Pi_2\right), $$

which acts on the Bloch vector as $\vec s_{\Psi}\mapsto(1-p)\vec s_{\Psi} + p(\vec u_{\mathrm r}\cdot\vec s_{\Psi})\vec u_{\mathrm r}$. Its Bloch matrix is $\operatorname{diag}(1-p,1-p,1)$ in the eigenbasis of $\vec u_{\mathrm r}$, of determinant $(1-p)^2$. A gate acts by a rotation, whose Bloch matrix has determinant $+1$ and is orthogonal; hence for $0is not a gate, and no representative of it is a single $*$-unitary conjugation. It is the canonical irreversible channel, and the companion article on decoherence analyses its effect on idempotency in detail.

The same idempotents, used coherently and incoherently. The sharpest contrast is between two objects built from the same complementary idempotents:

$$ \tilde{Q}_{\tilde Q} = \tilde\Pi_1\otimes e_t + \tilde\Pi_2\otimes\tilde Q \quad\text{(a gate)}, \qquad Ξ(\tilde{Ϛ}_{\Psi}) = \tilde\Pi_1\tilde{Ϛ}_{\Psi}\tilde\Pi_1 + \tilde\Pi_2\tilde{Ϛ}_{\Psi}\tilde\Pi_2 \quad\text{(a channel)}. $$

In the first, the idempotents sit inside a single $*$-unitary element and the sum is over the two control branches; both branches survive, the coherence between them is retained, and the result is reversible. In the second, the state itself is replaced by the sum of its two projected pieces; the coherence between the branches is discarded, the map contracts the Bloch ball, and the result is irreversible. This is the algebraic content of the familiar statement that a controlled gate is a measurement made coherent: the difference is not in the idempotents but in whether the sum is taken inside one element or between two conjugated states. It is the same reversible/irreversible dichotomy the framework already expresses as $*$-unitary versus idempotent acting elements.

The channel set is convex, with the gates among its extreme points; the companion article establishes this and the associated Stinespring picture, in which every irreversible channel is the reduction of a reversible evolution on a larger system after the environment is discarded. A circuit that contains a measurement, or a dephasing step, has left the gate group and is a general channel; it may still be drawn as a circuit, but it is no longer a single element of $\mathbb{B}^{\otimes n}$ and no longer invertible within the channels.

What the Reformulation Does and Does Not Claim

It does provide:

  • A single algebraic home for the objects of quantum computation: states are positive elements of $\mathbb{B}_{\Psi}^{\otimes n}$ with $\mathrm{Sc}=2^{-n}$ (trace one in the $2^n\times2^n$ image), gates are $*$-unitary elements of $\mathbb{B}^{\otimes n}$, and a circuit is their product.
  • The action of a gate as rotor conjugation, and reversibility as the property $\tilde Q\tilde Q^{*} = e_t$ that makes the acting element $*$-unitary.
  • The standard gate set in explicit biquaternion form: the Pauli and Hadamard gates as Hermitian unitaries in $\mathbb{B}_{\Psi}$, and the phase and $T$ gates as rotations in $\mathbb{B}_{H}$, with the global phase sitting in $\mathbb{B}_{C}$ and acting trivially.
  • The controlled gate as an algebraic construction from complementary idempotents and a target gate, and the interpretation of the controlled gate as the coherent use of an idempotent, in contrast with the measure-and-forget channel.
  • The Clifford group as the normaliser of the Pauli group in the gate group, and its octahedral rotation image.
  • The clean statement of the reversible/irreversible split as Kraus rank one versus higher rank, which is the algebra's own dichotomy rather than an additional postulate.

It does not claim:

  • A new theory of quantum computation. The formalism is standard quantum information theory in biquaternion notation, and it is empirically empty by itself.
  • A re-derivation of the universality theorems. The statement that $\{H,T\}$ generates a dense subgroup is inherited; the article records the generators but proves no approximation theorem.
  • Any claim about fault tolerance, error correction, or resource theories; none of these is developed here.
  • A resolution of the measurement problem. The controlled-gate-versus-channel contrast states the algebraic difference precisely; it does not say why a physical circuit contains one rather than the other.
  • A justification of the tensor-product arena, whose status is the open question inherited from the companion articles.

Open Questions

1. The two groups in one algebra. The gate group $\mathrm{U}(2)$ and the Lorentz group $SL(2,\mathbb{C})$ both act on the framework, on $\mathbb{B}_{\Psi}$ and $\mathbb{B}_{\Phi}$ respectively, by the same form of conjugation. A boost is a rotor on $\mathbb{B}_{\Phi}$ but not a gate on $\mathbb{B}_{\Psi}$; a gate is $*$-unitary on $\mathbb{B}_{\Psi}$ but has no biquaternion-norm-preserving interpretation on $\mathbb{B}_{\Phi}$. What is the precise sense in which a physical operation can be one and not the other, and is there any process that is naturally described by both?

2. The measurement as a gate. Every channel is a generalized measurement followed by a state transformation. The controlled gate shows that an idempotent can be used coherently, and dephasing shows that it can be used incoherently. Is there an algebraic criterion, beyond Kraus rank, that distinguishes the two uses in terms of the elements involved?

3. Universality in the algebra. The density of $\langle H, T\rangle$ is a statement about two specific elements of $\mathbb{B}$. Does the biquaternion framework give a natural reason for the existence of dense finitely generated subgroups of the gate group, or is this the same statement in different notation?

4. The tensor product. As in the companion article on the Bell basis, whether $\mathbb{B}^{\otimes n}$ is canonical for the framework or requires additional structure is unresolved. Multi-qubit gates inherit the contingency.

5. The $T$ gate and the non-compact directions. The gates generated by $H$ and $T$ lie, modulo phase, in the compact subgroup $SU(2)$ of $SL(2,\mathbb{C})$. The non-compact directions of the algebra — the boosts — are not reachable as gates. Is there any operation on the informational sector that involves them, or is the gate group necessarily compact?

6. Empirical content. As everywhere in the framework, a reformulation of gates and circuits is empirically empty. What, if anything, distinguishes it from standard quantum computation is the central open question.

Summary

A gate in the biquaternion framework is a $*$-unitary element $\tilde Q\in\mathbb{B}$ with $\tilde Q\tilde Q^{*} = e_t$, acting on a state $\tilde{Ϛ}_{\Psi}\in\mathbb{B}_{\Psi}$ by rotor conjugation $\tilde{Ϛ}_{\Psi}\mapsto\tilde Q\tilde{Ϛ}_{\Psi}\tilde Q^{*}$. This is exactly the class of channels of Kraus rank one, and exactly the reversible, purity-preserving state maps. The gate group is $\mathrm{U}(2) = \mathrm{U}(1)\cdot SU(2)$: every gate is a central phase times a rotation, and the phase is unobservable, so the effective group is $PU(2)\cong SO(3)$ acting on the Bloch ball. The condition $\tilde Q\tilde Q^{*} = e_t$ is not the unit-norm condition $\tilde U_{\mathrm{lor}}\tilde U_{\mathrm{lor}}{}^{\natural} = e_t$ of the Lorentz rotors: a boost is a rotor on $\mathbb{B}_{\Phi}$ but not a gate on $\mathbb{B}_{\Psi}$.

The standard single-qubit gate set has explicit representatives: the Pauli gates are $\tilde Q_x=i\vec e_x$, $\tilde Q_y=i\vec e_y$, $\tilde Q_z=i\vec e_z$; the Hadamard is $\tilde Q_{\mathrm{Had}}=\tfrac{i}{\sqrt2}(\vec e_x+\vec e_z)$; the phase and $T$ gates are the rotations $\tfrac{1}{\sqrt2}(e_t+\vec e_z)$ and $\cos\tfrac{\pi}{8}e_t+\sin\tfrac{\pi}{8}\vec e_z$; a general gate is $\exp(i\theta_{\mathrm{pha}})(\cos\tfrac{\theta_{\mathrm{axr}}}{2}e_t+\sin\tfrac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r})$. Apart from the identity gate, the Hermitian gates are the $\pi$-rotations $\pm i\vec u_{\mathrm r}$; the Pauli and Hadamard gates are of this type, the phase and $T$ gates are not.

Composition is the algebra product: applying $\tilde Q_{\mathrm{gate},1}$ then $\tilde Q_{\mathrm{gate},2}$ gives $\tilde Q_{\mathrm{gate},2}\tilde Q_{\mathrm{gate},1}$, and the whole circuit is a single element whose action is a single conjugation. The Pauli group is the sixteen-element set $\{\pm e_t,\pm ie_t,\pm \vec e_\eta,\pm i\vec e_\eta\}$, and the Clifford group is its normaliser in the gate group; $H$ and $S$ generate it, and its image in $PU(2)$ is the octahedral rotation group of order $24$.

In the multi-qubit arena $\mathbb{B}^{\otimes n}\cong M_{2^n}(\mathbb{C})$, the controlled gate is $\tilde{Q}_{\tilde Q} = \tilde\Pi_1(\vec u_{\mathrm r})\otimes e_t + \tilde\Pi_2(\vec u_{\mathrm r})\otimes\tilde Q$, $*$-unitary by the complementarity of the idempotents and the unitarity of $\tilde Q$. The CNOT, CZ, and SWAP gates have explicit representatives, and the standard circuit identities hold as algebra identities. A worked circuit — Hadamard then CNOT on two qubits — produces the Bell idempotent $\tfrac14(e_t\otimes e_t - \vec e_x\otimes \vec e_x + \vec e_y\otimes \vec e_y - \vec e_z\otimes \vec e_z)$ from a product state by a single conjugation.

The contrast with the irreversible channels is the Kraus-rank dichotomy. A circuit of gates is a gate, hence reversible; inserting a measurement or a dephasing step makes it a channel of Kraus rank at least two, represented by a sum of conjugations rather than one. The controlled gate and the measurement channel use the same idempotents; the difference is whether the sum over the outcomes is taken inside a single $*$-unitary element (coherent, reversible) or between conjugated states (incoherent, irreversible). The reformulation is standard quantum information theory in biquaternion notation: it makes the algebraic location of each circuit ingredient explicit without adding physical content.

Summary of Notation

Symbol Meaning
$\mathbb{B} = \mathbb{C}\otimes_\mathbb{R}\mathbb{H}$ Biquaternion algebra, $\cong M_2(\mathbb{C})$
$\mathbb{B}_{\Psi}$ Hermitian subspace (states, observables)
$\mathbb{B}_{\Phi}$ Anti-Hermitian subspace (generators of reversible evolution)
$\mathbb{B}_{C}$ Complex subspace (centre); home of the global phase
$\mathbb{B}_{H}$ Real-quaternion subspace (rotation rotors, $SU(2)$)
$e_t=1,\vec e_x,\vec e_y,\vec e_z$ Quaternion basis, $\vec e_\eta^2=-e_t$
$i$ Scalar imaginary, $i^2=-1$
$\tilde{Ϛ}_{\Psi}=\tfrac12(e_t+i\vec s_{\Psi})$ State of a qubit
$\tilde\Pi_{1,2}(\vec u_{\mathrm r})=\tfrac12(e_t\pm i\vec u_{\mathrm r})$ Complementary idempotents
$\tilde Q\tilde Q^{*}=e_t$ Gate (matrix-unitary element); $\mathrm{U}(2)$
$\tilde U_{\mathrm{lor}}\tilde U_{\mathrm{lor}}{}^{\natural}=e_t$ Lorentz rotor (unit norm); $SL(2,\mathbb{C})$, not a gate
$Ξ_{\tilde Q}(\tilde{Ϛ}_{\Psi})=\tilde Q\tilde{Ϛ}_{\Psi}\tilde Q^{*}$ Rotor conjugation (gate action)
$2\,\mathrm{Sc}(\tilde\Pi\tilde Q_{\mathrm{Had}}^{*})=2\,\mathrm{Sc}(\tilde\Pi\tilde Q_{\mathrm{Had}})=2\,\mathrm{Sc}(\tilde\Pi\tilde Q_{\mathrm{Had}}^{*}) = 2\langle\tilde\Pi,\tilde Q_{\mathrm{Had}}\rangle_{*}$ Scalar-part reading at normalisation (Born rule)
$\tilde Q_x=i\vec e_x,\ \tilde Q_y=i\vec e_y,\ \tilde Q_z=i\vec e_z$ Pauli gates
$\tilde Q_{\mathrm{Had}}=\tfrac{i}{\sqrt2}(\vec e_x+\vec e_z)$ Hadamard gate
$\tilde Q_{\mathrm{ph}}=\tfrac{1}{\sqrt2}(e_t+\vec e_z),\ \tilde Q_{T}=\cos\tfrac{\pi}{8}e_t+\sin\tfrac{\pi}{8}\vec e_z$ Phase and $T$ gates
$\tilde u_{\mathrm r}(\theta_{\mathrm{axr}})=\cos\tfrac{\theta_{\mathrm{axr}}}{2}e_t+\sin\tfrac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ General rotation gate
$\mathcal{P}=\{\pm e_t,\pm ie_t,\pm \vec e_\eta,\pm i\vec e_\eta\}$ Pauli group (16 elements)
$\tilde{Q}_{\tilde Q}=\tilde\Pi_1\otimes e_t+\tilde\Pi_2\otimes\tilde Q$ Controlled gate
$\mathrm{CNOT}=\tilde\Pi_1(\vec u_z)\otimes e_t+\tilde\Pi_2(\vec u_z)\otimes(i\vec e_x)$ Controlled-$X$
$\mathrm{SWAP}=\tfrac12(e_t\otimes e_t-\sum_\eta \vec e_\eta\otimes \vec e_\eta)$ Swap gate
$Ξ(\tilde{Ϛ}_{\Psi})=\sum_l\tilde M_{\mathrm{kraus},l}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},l}^{*}$ Kraus representation of a channel
$Ξ_{\mathrm{deph},p}$ Dephasing channel (irreversible, not a gate)
$\langle\tilde{Q}_1,\tilde{Q}_2\rangle$ the general plain bilinear form, the scalar part of the general plain bilinear product, $\mathrm{Sc}(\tilde{Q}_1\tilde{Q}_2)$
$\langle\tilde{Q}_1,\tilde{Q}_2\rangle_{\natural}$ the general quaternionic bilinear form, $\mathrm{Sc}(\tilde{Q}_1^{\natural}\tilde{Q}_2)$; on the diagonal $\langle\tilde{Q}_2,\tilde{Q}_2\rangle_{\natural}=\langle\tilde{Q}_2\rangle_{\natural}$
$\langle\tilde{Q}_1,\tilde{Q}_2\rangle_{*}$ the general plain sesquilinear form, $\mathrm{Sc}(\tilde{Q}_1\tilde{Q}_2^{*})$; on the diagonal $\langle\tilde{Q}_2,\tilde{Q}_2\rangle_{*}=\sum_\mu\lvert Q_{2\mu}\rvert^{2}$

Further Reading

  • Michael A. Nielsen and Isaac L. Chuang, Quantum Computation and Quantum Information (Cambridge, 2000), for the standard gate set, circuit identities, the Clifford group, and the Solovay–Kitaev theorem.
  • John Preskill, Quantum Computation (California Institute of Technology lecture notes), for the circuit model, universality, and the stabiliser formalism.
  • K. Kraus, States, Effects, and Operations (Springer, 1983), and M.-D. Choi, "Completely positive linear maps on complex matrices," Linear Algebra and its Applications 10 (1975) 285–290, for the Kraus representation and the rank criterion of reversibility.
  • M. B. Ruskai, S. Szarek, and E. Werner, "An analysis of completely-positive trace-preserving maps on $2\times2$ matrices," Linear Algebra and its Applications 347 (2002) 159–187, for the Bloch-ball picture of qubit channels and the geometry of the reversible maps among them.
  • Daniel Gottesman, "The Heisenberg representation of quantum computers" (1998), for the stabiliser formalism and the Clifford group in the Pauli normaliser picture.
  • P. A. M. Dirac, The Principles of Quantum Mechanics (Oxford, 1930), for the spin-1/2 formalism underlying the gate representatives.
  • Pertti Lounesto, Clifford Algebras and Spinors (Cambridge, 2001), and Chris Doran and Anthony Lasenby, Geometric Algebra for Physicists (Cambridge, 2003), for the rotor formulation of rotations used here.
  • The companion articles of this series: Introduction to the Biquaternion Universe, Four-Vectors and the Interval, States, Observables and the Born Weight, Quantum Physics in Biquaternionic Form, Spin-1/2 Quantum Physics in Biquaternionic Form, Quantum Channels and the Reversible/Irreversible Dichotomy, The Bell Basis as the Idempotent Basis of $\mathbb{B}\otimes\mathbb{B}$, and Decoherence as Idempotent Projection.