Exercise: Quantum Channels and Dephasing in M+

Introduction

This is one of a series of worked exercises on quantum physics in the biquaternion algebra $\mathbb{B} = \mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$. Earlier exercises treated the kinematics of a qubit and of a pair of qubits; each was a single problem solved twice, once in the standard Hilbert-space formulation and once in the biquaternion formulation of the Hermitian subspace $\mathbb{B}_{\Psi}$. This exercise is different in structure. Its parent is the article Quantum Channels and the Reversible/Irreversible Dichotomy, which develops the general state map on $\mathbb{B}_{\Psi}$ — the quantum channel, its Kraus representation, complete positivity, the Choi matrix, and the reversible/irreversible dichotomy. The reader is assumed to have that formalism and is asked to apply it.

The article is therefore a sequence of independent problems, each in its own section. We work throughout in the notation of the parent, returning to the standard matrix formulation only where the parent itself does (the transpose map, the Choi matrix, and the standard Kraus operators of amplitude damping). The problems are:

  1. Writing the dephasing channel as a Kraus sum and computing its action on a general Bloch vector.
  2. Computing the Kraus rank of dephasing and finding a minimal Kraus representation.
  3. Showing that the transpose map is positive but not completely positive.
  4. The amplitude-damping channel: Kraus operators, fixed states, and purity.
  5. Composition of channels, and the growth of the Kraus rank.
  6. The Bloch-ball contraction and the increase of the von Neumann entropy under dephasing.

The conventions are those of the companion articles. The quaternion units are $e_t = 1, \vec e_x, \vec e_y, \vec e_z$ with $\vec e_\eta^2 = -e_t$; $i$ is the central scalar imaginary. The Hermitian subspace $\mathbb{B}_{\Psi}$ consists of the elements $\tilde H_\Psi = h_t e_t + i\vec h_\Psi$ with real $h_t$ and $\vec h_\Psi \in \mathbb{R}^3$. A state is $\tilde{Ϛ}_{\Psi} = \tfrac{1}{2}(e_t + i\vec s_{\Psi})$ with $|\vec s_{\Psi}| \leq 1$, and $\vec s_{\Psi}$ is its Bloch vector. The doubled scalar part is $2\,\mathrm{Sc}(\tilde{Q}) = 2\,\mathrm{Sc}(\tilde{Q})$, so the Born rule is $2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}\tilde H_\Psi^{*}) = h_t + \vec s_{\Psi}\cdot\vec h_\Psi$. A channel is a completely positive, trace-preserving map with a Kraus representation $Ξ(\tilde{Q}) = \sum_l \tilde M_{\mathrm{kraus},l}\tilde{Q}\tilde M_{\mathrm{kraus},l}^{*}$ and normalization $\sum_l \tilde M_{\mathrm{kraus},l}^{*}\tilde M_{\mathrm{kraus},l} = e_t$, and the Choi matrix is $J(Ξ) = \sum_{j\eta}\tilde Q_{j\eta}\otimesΞ(\tilde Q_{j\eta})$ as in the parent. The dephasing channel along the unit pure real quaternion $\vec u_x$ is $$ Ξ_{\mathrm{deph},p}(\tilde{Ϛ}_{\Psi}) := (1-p)\,\tilde{Ϛ}_{\Psi} + p\left(\tilde\Pi_1\,\tilde{Ϛ}_{\Psi}\,\tilde\Pi_1 + \tilde\Pi_2\,\tilde{Ϛ}_{\Psi}\,\tilde\Pi_2\right), \qquad \tilde\Pi_{1,2}(\vec u_x) = \tfrac{1}{2}\left(e_t \pm i\vec u_x\right). $$ Throughout, $\log$ denotes the natural logarithm.

Problem 1: Dephasing as a Kraus Sum

Problem. Verify that the dephasing channel $Ξ_{\mathrm{deph},p}$ is trace preserving with the three Kraus operators $$ \tilde M_{\mathrm{kraus},0} = \sqrt{1-p}\,e_t, \qquad \tilde M_{\mathrm{kraus},1} = \sqrt{p}\,\tilde\Pi_1(\vec u_x), \qquad \tilde M_{\mathrm{kraus},2} = \sqrt{p}\,\tilde\Pi_2(\vec u_x), $$ and compute $Ξ_{\mathrm{deph},p}(\tilde{Ϛ}_{\Psi})$ for the general state $\tilde{Ϛ}_{\Psi} = \tfrac{1}{2}(e_t + i\vec s_{\Psi})$.

Solution. Put $\alpha = i\vec u_x$, so that $\tilde\Pi_{1,2} = \tfrac{1}{2}(e_t \pm \alpha)$. Since $\vec u_x^2 = -e_t$, $$ \alpha^2 = (i\vec u_x)^2 = i^2\vec u_x^2 = (-1)(-e_t) = e_t, \qquad \alpha^{*} = (i\vec u_x)^{*} = -i\,\vec u_x^{*} = -i(-\vec u_x) = \alpha, $$ so $\alpha^2 = e_t$ and $\alpha$ is Hermitian, as the idempotent structure requires. Trace preservation is immediate: because $\tilde\Pi_{1,2}$ are Hermitian idempotents, $\tilde M_{\mathrm{kraus},1}^{*}\tilde M_{\mathrm{kraus},1} = p\,\tilde\Pi_1$ and $\tilde M_{\mathrm{kraus},2}^{*}\tilde M_{\mathrm{kraus},2} = p\,\tilde\Pi_2$, whence $$ \tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0} + \tilde M_{\mathrm{kraus},1}^{*}\tilde M_{\mathrm{kraus},1} + \tilde M_{\mathrm{kraus},2}^{*}\tilde M_{\mathrm{kraus},2} = (1-p)\,e_t + p\,(\tilde\Pi_1 + \tilde\Pi_2) = e_t . $$

For the action on the state, use the elementary identity $$ \tilde\Pi_1\,\tilde{Q}\,\tilde\Pi_1 + \tilde\Pi_2\,\tilde{Q}\,\tilde\Pi_2 = \tfrac{1}{2}\left(\tilde{Q} + \alpha\,\tilde{Q}\,\alpha\right), $$ valid for every $\tilde{Q} \in \mathbb{B}$: the cross terms come with opposite signs, $$ \tfrac{1}{4}\left(e_t \pm \alpha\right)\tilde{Q}\left(e_t \pm \alpha\right) = \tfrac{1}{4}\left(\tilde{Q} \pm \alpha\tilde{Q} \pm \tilde{Q}\alpha + \alpha\tilde{Q}\alpha\right), $$ and cancel in the sum. The two products needed are $$ \alpha\,e_t\,\alpha = \alpha^2 = e_t, \qquad \alpha\,(i\vec s_{\Psi})\,\alpha = i^3\,\vec u_x\,\vec s_{\Psi}\,\vec u_x = -i\,\vec u_x\,\vec s_{\Psi}\,\vec u_x . $$ For the second we use the quaternion identity $$ \vec u_x\,\vec s_{\Psi}\,\vec u_x = \vec s_{\Psi} - 2(\vec u_x\cdot\vec s_{\Psi})\,\vec u_x . $$ Proof: decompose $\vec s_{\Psi} = \vec s_{\Psi,\parallel} + \vec s_{\Psi,\perp}$ with $\vec s_{\Psi,\parallel} = (\vec u_x\cdot\vec s_{\Psi})\vec u_x$ and $\vec s_{\Psi,\perp} \perp \vec u_x$. Then $\vec u_x\,\vec s_{\Psi,\parallel}\,\vec u_x = (\vec u_x\cdot\vec s_{\Psi})\,\vec u_x^3 = -(\vec u_x\cdot\vec s_{\Psi})\vec u_x = -\vec s_{\Psi,\parallel}$, because $\vec u_x^3 = \vec u_x\vec u_x^2 = -\vec u_x$; and for the orthogonal part $\vec u_x\,\vec s_{\Psi,\perp} = \vec u_x\times\vec s_{\Psi,\perp}$, so $\vec u_x\,\vec s_{\Psi,\perp}\,\vec u_x = (\vec u_x\times\vec s_{\Psi,\perp})\times\vec u_x = \vec s_{\Psi,\perp}$. Summing gives the identity. Therefore $$ \alpha\,\tilde{Ϛ}_{\Psi}\,\alpha = \tfrac{1}{2}\left(e_t - i\,\vec u_x\,\vec s_{\Psi}\,\vec u_x\right) = \tfrac{1}{2}\left(e_t - i\vec s_{\Psi} + 2i(\vec u_x\cdot\vec s_{\Psi})\,\vec u_x\right), $$ and hence $$ \tilde\Pi_1\tilde{Ϛ}_{\Psi}\tilde\Pi_1 + \tilde\Pi_2\tilde{Ϛ}_{\Psi}\tilde\Pi_2 = \tfrac{1}{2}\left(\tilde{Ϛ}_{\Psi} + \alpha\tilde{Ϛ}_{\Psi}\alpha\right) = \tfrac{1}{2}\left(e_t + i(\vec u_x\cdot\vec s_{\Psi})\,\vec u_x\right). $$ Combining with the mixture, $$ \boxed{\;Ξ_{\mathrm{deph},p}(\tilde{Ϛ}_{\Psi}) = \tfrac{1}{2}\left(e_t + i\,\vec s_{\mathrm{out}}\right), \qquad \vec s_{\Psi,\mathrm{out}} = (1-p)\,\vec s_{\Psi} + p\,(\vec u_x\cdot\vec s_{\Psi})\,\vec u_x\;}. $$ The transverse components of the Bloch vector are multiplied by $1-p$, while the longitudinal component is untouched. At $p=0$ the channel is the identity; at $p=1$ it is the projection $\vec s_{\Psi} \mapsto (\vec u_x\cdot\vec s_{\Psi})\vec u_x$, the "measure-and-forget" channel of the parent.

Special case $\vec u_x = \vec e_z$. The Kraus operators are $\tilde M_{\mathrm{kraus},0} = \sqrt{1-p}\,e_t$, $\tilde M_{\mathrm{kraus},1} = \tfrac{1}{2}\sqrt{p}\,(e_t + i\vec e_z)$, $\tilde M_{\mathrm{kraus},2} = \tfrac{1}{2}\sqrt{p}\,(e_t - i\vec e_z)$, and the Bloch map is $\vec s_{\Psi,\mathrm{out}} = (1-p)(r_1 \vec e_x + r_2 \vec e_y) + r_3 \vec e_z$.

Problem 2: The Kraus Rank of Dephasing

Problem. Compute the Choi matrix of $Ξ_{\mathrm{deph},p}$ in the computational basis and determine its rank. Deduce the minimal Kraus rank, exhibit a minimal Kraus representation, and decide whether the three-operator representation of Problem 1 is minimal.

Solution. The rank of the Choi matrix is invariant under $*$-unitary conjugation, and dephasing along any unit axis is unitarily equivalent to dephasing along $\vec e_z$; we therefore fix $\vec u_x = \vec e_z$. Let $E_{j\eta} = |j\rangle\langle k|$ be the matrix units of the computational basis, expressed in the algebra as $$ E_{00} = \tfrac{1}{2}(e_t + i\vec e_z), \quad E_{11} = \tfrac{1}{2}(e_t - i\vec e_z), \quad E_{01} = \tfrac{1}{2}(i\vec e_x - \vec e_y), \quad E_{10} = \tfrac{1}{2}(i\vec e_x + \vec e_y). $$ These are the elements the parent denotes $\tilde Q_{j\eta}$; they satisfy $2\,\mathrm{Sc}(E_{j\eta}E_{l\,n_{\mathrm{idx}}}) = \delta_{\eta l}\delta_{j\,n_{\mathrm{idx}}}$. The idempotents $\tilde\Pi_+ = \tfrac{1}{2}(e_t + i\vec e_z)$ and $\tilde\Pi_- = \tfrac{1}{2}(e_t - i\vec e_z)$ are exactly $E_{00}$ and $E_{11}$; the element $i\vec e_z$ is a root of $+1$, $(i\vec e_z)^2 = e_t$, and the two are its associated idempotents. Since dephasing fixes the populations and scales the coherences by $1-p$, $$ Ξ(E_{00}) = E_{00}, \qquad Ξ(E_{11}) = E_{11}, \qquad Ξ(E_{01}) = (1-p)\,E_{01}, \qquad Ξ(E_{10}) = (1-p)\,E_{10}, $$ the off-diagonal statement following because $\tilde\Pi_+ E_{01}\tilde\Pi_+ = \tilde\Pi_- E_{01}\tilde\Pi_- = 0$. With the parent's convention $J(Ξ) = \sum_{j\eta}E_{j\eta}\otimesΞ(E_{j\eta})$ and the product basis $(|00\rangle_{*},|01\rangle,|10\rangle,|11\rangle)$, $$ E_{01}\otimes E_{01} = |00\rangle\langle 11|, \qquad E_{10}\otimes E_{10} = |11\rangle\langle 00|, $$ so $$ J = \begin{pmatrix} 1 & 0 & 0 & 1-p\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ 1-p & 0 & 0 & 1 \end{pmatrix}. $$ The only nonzero block is $\begin{pmatrix}1 & 1-p\ 1-p & 1\end{pmatrix}$, whose eigenvalues are $1 \pm (1-p)$; hence $$ \mathrm{spec}\,J = \{\,2-p,\; p,\; 0,\; 0\,\}. $$ Thus $J \succeq 0$ for $0 \leq p \leq 2$, confirming complete positivity (the parent restricts to the sub-range $0 \leq p \leq 1$, where $Ξ_{\mathrm{deph},p}$ is a convex mixture of the identity and the full-dephasing measurement channel), and $$ \boxed{\;\mathrm{rank}\,J = 2 \text{ for } 0 < p < 2, \qquad \mathrm{rank}\,J = 1 \text{ at } p=0 \text{ and at } p=2\;} . $$ By the parent's rank theorem, the minimal Kraus rank is two for $0

Trace preservation holds because $(i\vec u_x)^{*}(i\vec u_x) = (i\vec u_x)^2 = i^2\vec u_x^2 = e_t$, so $\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0} + \tilde M_{\mathrm{kraus},1}^{*}\tilde M_{\mathrm{kraus},1} = (1-\tfrac{p}{2})e_t + \tfrac{p}{2}e_t = e_t$. The action is $$ Ξ(\tilde{Ϛ}_{\Psi}) = \left(1-\tfrac{p}{2}\right)\tilde{Ϛ}_{\Psi} + \tfrac{p}{2}\,(i\vec u_x)\,\tilde{Ϛ}_{\Psi}\,(i\vec u_x). $$ Since $(i\vec u_x)\tilde{Ϛ}_{\Psi}(i\vec u_x) = -\vec u_x\tilde{Ϛ}_{\Psi}\vec u_x = \tfrac{1}{2}\left(e_t - i\vec s_{\Psi} + 2i(\vec u_x\cdot\vec s_{\Psi})\vec u_x\right)$, this reproduces $\vec s_{\Psi,\mathrm{out}} = (1-p)\vec s_{\Psi} + p(\vec u_x\cdot\vec s_{\Psi})\vec u_x$, as required.

Conclusion. The three-operator representation of Problem 1 is valid but not minimal: for $0

Problem 3: A Positive Map That Is Not Completely Positive

Problem. In the matrix representative of $\mathbb{B}$, let $Ξ_T$ be the transpose, $Ξ_T(\tilde{Q}) = \tilde{Q}^{\mathsf{T}}$. Show that $Ξ_T$ is positive and trace preserving, but not completely positive.

Solution. On the matrix units $Ξ_T$ acts by $E_{j\eta} \mapsto E_{\eta j}$; equivalently, on a state, $$ Ξ_T\!\left(\tfrac{1}{2}(e_t + i\vec s_{\Psi})\right) = \tfrac{1}{2}\left(e_t + i(r_1 \vec e_x - r_2 \vec e_y + r_3 \vec e_z)\right), $$ because $\sigma_1$ and $\sigma_3$ are real while $\sigma_2$ is imaginary. So $Ξ_T$ acts on the Bloch ball by the reflection $$ (r_1, r_2, r_3) \longmapsto (r_1, -r_2, r_3). $$ This map is trace preserving (the transpose fixes $e_t$ and preserves the trace) and positive: the reflection is an isometry of $\mathbb{R}^3$ that maps the ball $|\vec s_{\Psi}|\leq 1$ onto itself, so it maps states to states.

To test complete positivity, compute the Choi matrix. Transposition swaps the off-diagonal units, $Ξ_T(E_{01}) = E_{10}$ and $Ξ_T(E_{10}) = E_{01}$, while fixing $E_{00}$ and $E_{11}$. Hence, using $E_{01}\otimes E_{10} = |01\rangle\langle 10|$ and $E_{10}\otimes E_{01} = |10\rangle\langle 01|$, $$ J_T = E_{00}\otimes E_{00} + E_{11}\otimes E_{11} + E_{01}\otimes E_{10} + E_{10}\otimes E_{01} = \begin{pmatrix} 1 & 0 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 0 & 1 \end{pmatrix}. $$ This is the swap operator, $J_T|jk\rangle = |kj\rangle$. Its eigenvalues are $+1$ on the symmetric subspace $\{|00\rangle, |11\rangle, (|01\rangle+|10\rangle)/\sqrt{2}\}$ and $-1$ on the antisymmetric (singlet) direction $(|01\rangle-|10\rangle)/\sqrt{2}$, so $$ \boxed{\;\mathrm{spec}\,J_T = \{\,1,\;1,\;1,\;-1\,\}\;}. $$ The Choi matrix is therefore not positive semidefinite, and by the parent's criterion $Ξ_T$ is not completely positive. Equivalently, transposing one qubit of a maximally entangled state can produce an operator with a negative eigenvalue: the antisymmetric direction of $J_T$ is exactly the singlet, which acquires the eigenvalue $-1$. No physical process can act that way on half of an entangled pair, so $Ξ_T$ is not a channel, despite being positive and trace preserving. This is the parent's point that positivity must be strengthened to complete positivity, made explicit on the simplest example.

The conclusion does not depend on the choice of matrix representative: transposes in two orthonormal bases are related by $*$-unitary conjugation, and the spectrum of the Choi matrix is invariant under $*$-unitary conjugation of the channel.

Problem 4: The Amplitude-Damping Channel

Problem. The amplitude-damping channel with damping parameter $\gamma \in [0,1]$ acts on the Bloch vector as $$ (r_1, r_2, r_3) \longmapsto \left(\sqrt{1-\gamma}\,r_1,\ \sqrt{1-\gamma}\,r_2,\ \gamma + (1-\gamma)r_3\right). $$ Find its Kraus operators in $\mathbb{B}$, verify trace preservation, and determine its fixed states, its action on the ground and excited states, and its effect on purity.

Solution. In the matrix representative the standard Kraus operators are $$ M_{\mathrm{kraus},0} = \begin{pmatrix}1 & 0\ 0 & \sqrt{1-\gamma}\end{pmatrix}, \qquad M_{\mathrm{kraus},1} = \begin{pmatrix}0 & \sqrt{\gamma}\ 0 & 0\end{pmatrix} = \sqrt{\gamma}\,|0\rangle\langle 1| . $$ Writing $s = \sqrt{1-\gamma} \in [0,1]$ and using $\sigma_3 = i\vec e_z$ and $|0\rangle\langle 1| = \tfrac{1}{2}(i\vec e_x - \vec e_y)$, these translate to $$ \boxed{\;\tilde M_{\mathrm{kraus},0} = \frac{1+s}{2}\,e_t + i\,\frac{1-s}{2}\,\vec e_z \;\in \mathbb{B}_{\Psi}, \qquad \tilde M_{\mathrm{kraus},1} = \frac{\sqrt{\gamma}}{2}\left(i \vec e_x - \vec e_y\right) \in \mathbb{B}\;}. $$ Here $\tilde M_{\mathrm{kraus},0}$ is Hermitian (it is an element of $\mathbb{B}_{\Psi}$), while $\tilde M_{\mathrm{kraus},1}$ is not. Trace preservation follows from $$ \tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0} = \frac{1+s^2}{2}\,e_t + i\,\frac{1-s^2}{2}\,\vec e_z, \qquad \tilde M_{\mathrm{kraus},1}^{*}\tilde M_{\mathrm{kraus},1} = \frac{\gamma}{2}\left(e_t - i\vec e_z\right), $$ whose sum is $e_t$, since $1 + s^2 + \gamma = 2$ and $1 - s^2 - \gamma = 0$. A direct quaternion expansion of $\tilde M_{\mathrm{kraus},0}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},0}^{*} + \tilde M_{\mathrm{kraus},1}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},1}^{*}$ for $\tilde{Ϛ}_{\Psi} = \tfrac{1}{2}(e_t + i\vec s_{\Psi})$ gives $\tfrac{1}{2}(e_t + i\vec s_{\mathrm{out}})$ with exactly the Bloch map stated above; the $\tilde M_{\mathrm{kraus},1}$ term contributes the population $\tfrac{\gamma}{4}(1-r_3)(e_t + i\vec e_z)$ that is transferred to the ground state.

Fixed states. Set $r_1 = r_2 = 0$ and $z = \gamma + (1-\gamma)z$, giving $z = 1$. For every $\gamma>0$ the channel has the unique fixed state $$ \tilde{Ϛ}_{\Psi,*} = \tfrac{1}{2}\left(e_t + i\vec e_z\right) = \tilde\Pi_1(\vec u_z), $$ the pure ground state. For $\gamma = 0$ the channel is the identity and every state is fixed.

Non-unitality. The channel does not fix the maximally mixed state: $$ Ξ\left(\tfrac{1}{2}e_t\right) = \tfrac{1}{2}\left(e_t + i\gamma \vec e_z\right) \neq \tfrac{1}{2}e_t . $$ The centre of the Bloch ball is moved to the Bloch vector $(0,0,\gamma)$. This is the structural difference from dephasing, which is unital.

Ground and excited states. The ground state $\vec s_{\Psi} = (0,0,1)$ is fixed, as noted. The excited state $\vec s_{\Psi} = (0,0,-1)$ is mapped to $$ \vec s_{\Psi,\mathrm{out}} = \left(0,\,0,\,\gamma - (1-\gamma)\right) = \left(0,\,0,\,2\gamma - 1\right). $$ At $\gamma = \tfrac{1}{2}$ the excited state is driven to the maximally mixed state $(0,0,0)$; at $\gamma = 1$ it reaches the ground state. This is the standard incoherent decay of the excited state.

Purity. For a general input, $$ 2\,\mathrm{Sc}\!\left(\tilde{Ϛ}_{\Psi,\mathrm{out}}^2\right) = \tfrac{1}{2}\left(1 + (1-\gamma)\left(r_1^2 + r_2^2\right) + \left(\gamma + (1-\gamma)r_3\right)^2\right). $$ For the maximally mixed input ($\vec s_{\Psi} = 0$) the purity rises from $\tfrac{1}{2}$ to $\tfrac{1}{2}(1+\gamma^2) \geq \tfrac{1}{2}$: amplitude damping purifies the maximally mixed state by driving it toward the pure ground state. This is possible precisely because the channel is non-unital, and it is the opposite of the behaviour of dephasing. For the excited input $\vec s_{\Psi} = (0,0,-1)$ the output purity is $\tfrac{1}{2}(1 + (2\gamma-1)^2)$, which is $\tfrac{1}{2}$ at $\gamma = \tfrac{1}{2}$ — the maximally mixed value — and rises to $1$ at $\gamma = 1$. Finally, the Choi rank of the channel is two for $\gamma>0$ (one at $\gamma=0$), so it is irreversible for every $\gamma>0$.

Problem 5: Composition of Channels and the Growth of Kraus Rank

Problem. Let $Ξ_1$ have Kraus operators $\{A_{n_{\mathrm{idx}}}\}_{n_{\mathrm{idx}}=1}^{n_1}$ and $Ξ_2$ have Kraus operators $\{B_l\}_{l=1}^{n_2}$. Show that $Ξ_2\circΞ_1$ has Kraus operators $\{B_l A_{n_{\mathrm{idx}}}\}$ and deduce that $\mathrm{rank}(Ξ_2\circΞ_1) \leq \mathrm{rank}(Ξ_1)\,\mathrm{rank}(Ξ_2)$. Then compose two dephasing channels and determine the resulting Kraus rank.

Solution. Composing, $$ Ξ_2\!\left(Ξ_1(\tilde{Q})\right) = \sum_l B_l\left(\sum_{n_{\mathrm{idx}}} A_{n_{\mathrm{idx}}}\tilde{Q}A_{n_{\mathrm{idx}}}^{*}\right)B_l^{*} = \sum_{l,n_{\mathrm{idx}}}\left(B_l A_{n_{\mathrm{idx}}}\right)\tilde{Q}\left(B_l A_{n_{\mathrm{idx}}}\right)^\dagger , $$ so $\{B_l A_{n_{\mathrm{idx}}}\}$ is a Kraus set for the composite, and it is trace preserving whenever both factors are. The Kraus rank of the composite is the dimension of the span of the products $B_l A_m$, which cannot exceed the product of the dimensions of the individual spans; hence $\mathrm{rank}(Ξ_2\circΞ_1) \leq \mathrm{rank}(Ξ_1)\,\mathrm{rank}(Ξ_2)$. For a qubit the composite has rank at most $4$, since its Choi matrix is $4\times4$.

Composition of dephasings. The Bloch map of $Ξ_{\mathrm{deph},p}$ along $\vec u_x$ is the linear map $D(\vec u_x,p) = (1-p)I + p\,\vec u_x\vec u_x^{\mathsf{T}}$ acting on $\mathbb{R}^3$. Composing two of them, $$ D(\vec u_y,q)\,D(\vec u_x,p) = (1-p)(1-q)\,I + p(1-q)\,\vec u_x\vec u_x^{\mathsf{T}} + q(1-p)\,\vec u_y\vec u_y^{\mathsf{T}} + pq\,(\vec u_y\cdot\vec u_x)\,\vec u_y\vec u_x^{\mathsf{T}} . $$ For parallel axes, $\vec u_y = \vec u_x$, the projector identity $\vec u_x\vec u_x^{\mathsf{T}}\vec u_x\vec u_x^{\mathsf{T}} = \vec u_x\vec u_x^{\mathsf{T}}$ collapses the sum to $$ \boxed{\;D(\vec u_x,q)\,D(\vec u_x,p) = D(\vec u_x,\,p+q-pq)\;}, $$ so the composite is again a dephasing channel, with effective parameter $p_{\mathrm{out}} = p+q-pq$; its Kraus rank is at most two.

For orthogonal axes and full dephasing, take $\vec u_x = \vec e_z$, $\vec u_y = \vec e_x$ and $p = q = 1$. Then $D(\vec e_z,1) = \vec e_z\vec e_z^{\mathsf{T}}$, $D(\vec e_x,1) = \vec e_x\vec e_x^{\mathsf{T}}$, and $$ D(\vec e_x,1)\,D(\vec e_z,1) = \vec e_x\left(\vec e_x^{\mathsf{T}}\vec e_z\right)\vec e_z^{\mathsf{T}} = 0 . $$ The composite is the completely depolarizing channel, $\tilde{Ϛ}_{\Psi} \mapsto \tfrac{1}{2}e_t$ for every state: every Bloch vector is sent to the origin. Its Choi matrix is $$ \boxed{\;J = \sum_{j\eta} E_{j\eta}\otimesΞ(E_{j\eta}) = \sum_{\eta} E_{\eta\eta}\otimes \tfrac{1}{2}e_t = \tfrac{1}{2}I_4\;}, $$ of rank four. Thus composing two rank-two channels produces a rank-four channel, saturating the qubit bound $4 = 2\times 2$. More generally, for two distinct axes and $0

Problem 6: Bloch-Ball Contraction and the Increase of Entropy

Problem. Show that dephasing contracts the Bloch ball, $|\vec s_{\Psi,\mathrm{out}}| \leq |\vec s_{\Psi}|$, that it does not increase the purity, and that it does not decrease the von Neumann entropy. Evaluate a concrete case.

Solution. Decompose $\vec s_{\Psi} = \vec s_{\Psi,\parallel} + \vec s_{\Psi,\perp}$ with $\vec s_{\Psi,\parallel} = (\vec u_x\cdot\vec s_{\Psi})\vec u_x$ and $\vec s_{\Psi,\perp} \perp \vec u_x$. From the Bloch map, $$ \vec s_{\Psi,\mathrm{out}} = (1-p)\vec s_{\Psi} + p(\vec u_x\cdot\vec s_{\Psi})\vec u_x = (1-p)\,\vec s_{\Psi,\perp} + \vec s_{\Psi,\parallel}, $$ so $$ |\vec s_{\Psi,\mathrm{out}}|^2 = (1-p)^2\,s_\perp^2 + s_\parallel^2 \leq s_\perp^2 + s_\parallel^2 = |\vec s_{\Psi}|^2 , $$ with equality if and only if $p=0$, or $s_\perp = 0$ (the state already lies on the dephasing axis). The purity, $$ 2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}^2) = \tfrac{1}{2}\left(1 + |\vec s_{\Psi}|^2\right), $$ therefore does not increase, and strictly decreases for any off-axis state when $p>0$.

Entropy. The eigenvalues of $\tilde{Ϛ}_{\Psi}$ are $\lambda_\pm = \tfrac{1}{2}(1 \pm x)$ with $r_x = |\vec s_{\Psi}|$, and $$ S(x) = -\lambda_+\log\lambda_+ - \lambda_-\log\lambda_- = -\frac{1+x}{2}\log\frac{1+x}{2} - \frac{1-x}{2}\log\frac{1-x}{2}. $$ Differentiating, $$ \frac{dS}{dx} = -\frac{1}{2}\left(\log\frac{1+x}{2} + 1\right) + \frac{1}{2}\left(\log\frac{1-x}{2} + 1\right) = \frac{1}{2}\log\frac{1-x}{1+x} \leq 0 \qquad (0 \leq x < 1), $$ so $S$ is a decreasing function of $x$. Since $|\vec s_{\Psi,\mathrm{out}}| \leq |\vec s_{\Psi}|$, $$ S(\tilde{Ϛ}_{\Psi,\mathrm{out}}) \geq S(\tilde{Ϛ}_{\Psi}), $$ with equality if and only if $p=0$ or the state is on the dephasing axis. Dephasing strictly increases the entropy of every off-axis state, for every $0 < p \leq 1$. This is consistent with the channel being unital: $Ξ(\tfrac{1}{2}e_t) = \tfrac{1}{2}e_t$, and unital channels are mixedness-increasing.

Concrete case. Take the pure state $\tilde{Ϛ}_{\Psi} = \tilde\Pi_1(\vec e_x) = \tfrac{1}{2}(e_t + i\vec e_x)$, with $\vec s_{\Psi} = (1,0,0)$, purity $1$, and entropy $0$. Dephase along $\vec e_z$ with $p = \tfrac{1}{2}$. Then $\vec s_{\Psi,\mathrm{out}} = (\tfrac{1}{2},0,0)$, so $|\vec s_{\Psi,\mathrm{out}}| = \tfrac{1}{2}$, the eigenvalues are $\lambda_\pm = \tfrac{3}{4}, \tfrac{1}{4}$, and $$ \boxed{\;S = -\tfrac{3}{4}\log\tfrac{3}{4} - \tfrac{1}{4}\log\tfrac{1}{4} = \tfrac{3}{4}\log\tfrac{4}{3} + \tfrac{1}{4}\log 4 \approx 0.562335 \ \text{nats}\;}, $$ while the purity falls to $\tfrac{1}{2}(1 + \tfrac{1}{4}) = \tfrac{5}{8} = 0.625$. At $p=1$ the Bloch vector collapses to the origin, the state becomes maximally mixed, and $S = \log 2 \approx 0.693147$ nats. The entropy rises monotonically from $0$ to $\log 2$ as $p$ runs from $0$ to $1$.

Summary

We have worked six problems that exercise the channel formalism of the parent article in $\mathbb{B}_{\Psi}$.

  1. Dephasing as a Kraus sum. The channel $Ξ_{\mathrm{deph},p}$ has Kraus operators $\sqrt{1-p}\,e_t$, $\sqrt{p}\,\tilde\Pi_1$, $\sqrt{p}\,\tilde\Pi_2$ and acts on the Bloch vector as $\vec s_{\Psi,\mathrm{out}} = (1-p)\vec s_{\Psi} + p(\vec u_x\cdot\vec s_{\Psi})\vec u_x$, scaling the transverse coherence by $1-p$ and leaving the populations fixed.
  2. Kraus rank. The Choi matrix of dephasing has spectrum $\{2-p,\,p,\,0,\,0\}$; its rank is two for $00$. The minimal Kraus rank is two, not three: the three-operator representation of the parent is valid but non-minimal, and $\sqrt{1-\tfrac{p}{2}}\,e_t$, $\sqrt{\tfrac{p}{2}}\,i\vec u_x$ is a minimal set.
  3. Positive but not completely positive. The transpose map acts on the Bloch ball as $(r_1,r_2,r_3)\mapsto(r_1,-r_2,r_3)$, is positive and trace preserving, and has Choi matrix equal to the swap operator, with spectrum $\{1,1,1,-1\}$. It is not a channel.
  4. Amplitude damping. With $s=\sqrt{1-\gamma}$, the Kraus operators are $\tilde M_{\mathrm{kraus},0} = \tfrac{1+s}{2}e_t + i\tfrac{1-s}{2}\vec e_z$ and $\tilde M_{\mathrm{kraus},1} = \tfrac{\sqrt{\gamma}}{2}(i\vec e_x - \vec e_y)$; the channel is non-unital, has the unique fixed state $\tilde\Pi_1(\vec u_z)$, sends the excited state to $(0,0,2\gamma-1)$, and raises the purity of the maximally mixed state to $\tfrac{1}{2}(1+\gamma^2)$.
  5. Composition and rank. Composing channels multiplies Kraus operators, $B_lA_m$, and the Kraus rank of a composition is at most the product of the ranks. Two dephasings compose to a dephasing along the common axis with $p_{\mathrm{out}} = p+q-pq$ when the axes are parallel, but to the rank-four completely depolarizing channel when they are orthogonal and full.
  6. Contraction and entropy. Dephasing contracts the Bloch ball, $|\vec s_{\Psi,\mathrm{out}}|^2 = (1-p)^2 s_\perp^2 + s_\parallel^2 \leq |\vec s_{\Psi}|^2$, lowers the purity, and raises the entropy, since $dS/d|\vec s_{\Psi}| = \tfrac{1}{2}\log\frac{1-|\vec s_{\Psi}|}{1+|\vec s_{\Psi}|} \leq 0$. For $\tilde\Pi_1(\vec e_x)$ dephased along $\vec e_z$ with $p=\tfrac{1}{2}$, the entropy rises from $0$ to $\approx 0.562335$ nats and the purity falls to $0.625$.

The two canonical channels illustrate the two faces of irreversibility. Dephasing is unital and destroys coherence without changing populations; amplitude damping is non-unital and drives the state toward a pure ground state. Both have Kraus rank two on their physical ranges ($0

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)
$e_t = 1, \vec e_x, \vec e_y, \vec e_z$ Quaternion basis, $\vec e_\eta^2 = -e_t$
$i$ Scalar imaginary, $i^2 = -1$
$\tilde{Ϛ}_{\Psi} = \tfrac{1}{2}(e_t + i\vec s_{\Psi})$ State; $\vec s_{\Psi}$ the Bloch vector
$\tilde\Pi_{1,2}(\vec u_x) = \tfrac{1}{2}(e_t \pm i\vec u_x)$ Idempotents along $\vec u_x$
$Ξ(\tilde{Q}) = \sum_l \tilde M_{\mathrm{kraus},l}\tilde{Q}\tilde M_{\mathrm{kraus},l}^{*}$ Kraus representation
$\sum_l \tilde M_{\mathrm{kraus},l}^{*}\tilde M_{\mathrm{kraus},l} = e_t$ Trace-preservation condition
$\langle\tilde{Q}_1,\tilde{Q}_2\rangle_{*}$ the general plain sesquilinear form, $\mathrm{Sc}(\tilde{Q}_1\tilde{Q}_2^{*})$; on the diagonal $\langle\tilde{Q}_2,\tilde{Q}_2\rangle_{*}=\sum_\mu\lvert Q_{2\mu}\rvert^{2}$
$J(Ξ) = \sum_{j\eta}\tilde Q_{j\eta}\otimesΞ(\tilde Q_{j\eta})$ Choi matrix; Kraus rank $=\mathrm{rank}\,J$
$Ξ_{\mathrm{deph},p}$ Dephasing channel along $\vec u_x$
$\vec s_{\Psi} \mapsto (1-p)\vec s_{\Psi} + p(\vec u_x\cdot\vec s_{\Psi})\vec u_x$ Dephasing Bloch map
$E_{j\eta} = \lvert j\rangle\langle k\rvert$ Matrix units of the computational basis
$\gamma$ Amplitude-damping parameter
$2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}^2) = \tfrac{1}{2}(1+\lvert\vec s_{\Psi}\rvert^2)$ Purity
$S = -\lambda_+\log\lambda_+ - \lambda_-\log\lambda_-$ von Neumann entropy

Further Reading

  • Michael A. Nielsen and Isaac L. Chuang, Quantum Computation and Quantum Information (Cambridge, 2000), for quantum operations, the Kraus representation, the Choi matrix, and the depolarizing, dephasing, and amplitude-damping channels.
  • K. Kraus, States, Effects, and Operations (Springer, 1983), for the original formulation of completely positive maps.
  • M.-D. Choi, "Completely positive linear maps on complex matrices," Linear Algebra and its Applications 10 (1975) 285–290, for the complete-positivity criterion used in Problem 3.
  • 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 affine Bloch-ball picture of qubit channels.
  • The companion articles of this series: Quantum Physics in Biquaternionic Form, Quantum Channels and the Reversible/Irreversible Dichotomy, and the earlier exercises of this series.