Quantum Error Model and Decoherence Channels

A qubit is never truly alone. The stray electromagnetic modes, phonons, and defects around it are always listening, and the moment they learn anything about the qubit's state, the delicate superpositions and entanglement that computation runs on begin to fray. To build machines that survive this, we first need a language precise enough to say exactly what "noise" is — not a vague fuzzing of the wavefunction, but a definite mathematical object. That object is the quantum channel: the most general transformation a density operator can physically undergo, pinned down equivalently by complete positivity and trace preservation, by a Kraus (operator-sum) decomposition, or by a unitary interaction with an environment that is subsequently discarded. This article assembles that framework and then puts it to work on the three canonical single-qubit noise models. The depolarizing channel replaces a state with the maximally mixed state with some probability and otherwise applies a uniformly random Pauli; its perfect isotropy makes it the standard yardstick. The phase-damping (dephasing) channel is pure decoherence — coherences die while energy is untouched — and it is usually the dominant error on real hardware. The amplitude-damping channel is spontaneous emission and T₁ relaxation, an excited qubit decaying toward its ground state. Threaded through all three is the Stinespring picture: every channel is unitary quantum mechanics on a larger Hilbert space, and its irreversibility is just information leaking into places we cannot follow. The payoff arrives at the end — because single-qubit errors expand in the Pauli basis, a code that corrects the discrete set {X, Y, Z} corrects a whole continuum of physical noise, the discretization theorem that every downstream error-correcting code is built upon.

Key takeaways

Noise on qubits is a quantum channel: a completely positive, trace-preserving (CPTP) map $\mathcal{E}(\rho)=\sum_k E_k\rho E_k^\dagger$ with $\sum_k E_k^\dagger E_k=I$ — the Kraus (operator-sum) form.
Three canonical single-qubit channels: depolarizing (isotropic Pauli noise, shrink factor $1-\tfrac{4p}{3}$ ), dephasing (kills $X,Y$ coherence, leaves $Z$ ), and amplitude damping (energy decay toward the ground state).
Decoherence is entanglement with an inaccessible environment: Stinespring dilation makes every channel a unitary on system + environment with the environment traced out, and the Kraus operators are the environment matrix elements of that unitary.
Amplitude damping alone gives $T_2=2T_1$ because coherence decays as $e^{-t/2T_1}$ while population decays as $e^{-t/T_1}$ ; extra pure dephasing only pushes $T_2$ below $2T_1$ .
Because single-qubit Kraus operators expand in $\{I,X,Y,Z\}$ , correcting the three Paulis corrects arbitrary continuous noise — the discretization of errors, and the reason finite codes work.
Key idea: Every physical noise process on qubits is a CPTP map $\mathcal{E}(\rho)=\sum_k E_k\rho E_k^\dagger$ (equivalently a unitary on system + environment with the environment traced out); because single-qubit Kraus operators expand in the Pauli basis, correcting the discrete set $\{X,Y,Z\}$ suffices to correct continuous noise.

Where this lives today

The channel formalism is the daily working language of quantum-hardware characterization and fault-tolerance engineering. Every superconducting or trapped-ion processor is specified by its $T_1$ and $T_2$ — literally amplitude-damping and dephasing rates — quoted per qubit alongside gate fidelities extracted by randomized benchmarking, which deliberately twirls arbitrary noise into an effective depolarizing channel so a single number summarizes it. Pauli twirling and randomized compiling are standard software passes that convert coherent, hard-to-model errors into stochastic Pauli channels, tailoring real hardware to the depolarizing assumptions that threshold theorems rely on. Because dephasing usually dominates bit-flip errors, biased-noise codes such as the XZZX surface code and cat-qubit architectures allocate more protection to $Z$ errors — a choice that only makes sense once the channel model gives the relative Kraus weights of $X$ , $Y$ , and $Z$ . Every density-matrix or stabilizer noise simulator estimates a logical error rate by sampling Kraus operators as quantum trajectories, applying one $E_k$ per shot with probability $\|E_k|\psi\rangle\|^2$ and averaging — the Monte Carlo wavefunction method that turns equation (1) into a running experiment. The open frontier is the noise these idealized channels miss: temporally correlated (non-Markovian) errors, crosstalk, and leakage out of the computational subspace, all of which break the single-qubit CPTP picture.

The physical picture

Start from a single fact: decoherence is what happens when the environment learns something about your qubit. Write a relative phase between $|0\rangle$ and $|1\rangle$ , then let a nearby degree of freedom couple to $Z$ so that it records "which way did the qubit point?". You never read that record — it escapes into modes you cannot access — but the mere existence of a copy elsewhere is enough to destroy interference. No energy changes hands; only the phase relationship is scrambled. This is the operational heart of the theory, and it is why a channel is most honestly described as a unitary on a bigger space followed by throwing part of it away. The first figure makes that structure concrete.

Figure

A system qubit couples to a fresh environment through a joint unitary U, then the environment is discarded — the partial trace over the environment turns reversible unitary dynamics into the irreversible channel that maps ρ to the sum of Eₖ ρ Eₖ†, with each Kraus operator read off as an environment matrix element of U

Amplitude damping is a different flavor of the same story: here the environment does not merely watch, it accepts a quantum of energy. The excited state $|1\rangle$ slides toward $|0\rangle$ as a photon leaks away, so the map is asymmetric — it drags the Bloch ball toward the north pole rather than symmetrically shrinking it. Depolarizing noise is the maximally ignorant limit, forgetting the state entirely with some probability and returning the center of the ball. The unifying geometric picture is single: a channel is an affine map of the Bloch ball, and the three canonical channels deform it three characteristic ways — contract to the center, collapse toward an axis, or drag toward a pole. The second figure shows all three, with the exact contraction laws.

Figure

The three canonical channels as deformations of the Bloch ball: depolarizing shrinks it uniformly by the factor 1 − 4p/3 (a single point at p = 3/4), dephasing collapses it toward the z axis as coherences fade by √(1−λ), and amplitude damping drags it toward the north ground-state pole via r_z → (1−γ)r_z + γ

The core equations

Everything begins with the operator-sum form. A channel acts on a density operator by conjugating it with a set of Kraus operators and summing; the completeness relation is exactly what keeps probability (the trace) equal to one. This single equation encodes complete positivity and trace preservation at once.

\mathcal{E}(\rho)=\sum_k E_k\,\rho\,E_k^{\dagger},\qquad \sum_k E_k^{\dagger}E_k = I \qquad (1)

where: $\mathcal{E}$ — the quantum channel, a CPTP linear map on density operators; $E_k$ — Kraus (noise) operators, not necessarily unitary or Hermitian; $\rho$ — input density operator; $I$ — identity, enforcing trace preservation. Reading: each $E_k$ is one way the environment could have acted, and the sum averages coherently over the unread outcomes. The depolarizing channel is the isotropic benchmark. With probability $1-p$ nothing happens; with probability $p$ a uniformly random one of the three non-trivial Paulis is applied. Because it treats $X$ , $Y$ , $Z$ symmetrically, a single number $p$ specifies the whole channel.

\mathcal{E}_{\mathrm{dep}}(\rho)=(1-p)\,\rho+\frac{p}{3}\,(X\rho X+Y\rho Y+Z\rho Z) \qquad (2)

where: $p$ — total error probability, $0\le p\le \tfrac34$ for a valid map; $X,Y,Z$ — single-qubit Pauli operators. Reading: the equivalent form $\mathcal{E}_{\mathrm{dep}}(\rho)=(1-\tfrac{4p}{3})\rho+\tfrac{4p}{3}\tfrac{I}{2}$ shows the Bloch vector shrinking by $1-\tfrac{4p}{3}$ toward the center. The dephasing (phase-flip) channel damps coherence while leaving populations exactly alone. It is a mixture of the identity and a $Z$ conjugation, and its Kraus operators are proportional to $I$ and $Z$ .

\mathcal{E}_{\mathrm{pd}}(\rho)=(1-p_Z)\,\rho+p_Z\,Z\rho Z,\qquad E_0=\sqrt{1-p_Z}\,I,\quad E_1=\sqrt{p_Z}\,Z \qquad (3)

where: $p_Z$ — probability of a $Z$ flip. Reading: off-diagonal coherences are multiplied by $1-2p_Z$ ; the equivalent "phase-damping" parametrization uses $E_0=\mathrm{diag}(1,\sqrt{1-\lambda})$ , $E_1=\mathrm{diag}(0,\sqrt{\lambda})$ with $p_Z=\tfrac{1}{2}(1-\sqrt{1-\lambda})$ , so the coherence factor is $\sqrt{1-\lambda}$ . Amplitude damping is the one channel that is genuinely non-unital and non-Pauli. Its two Kraus operators are a "no-jump" reweighting and a literal quantum jump that emits energy.

E_0=\begin{pmatrix}1&0\\0&\sqrt{1-\gamma}\end{pmatrix},\qquad E_1=\begin{pmatrix}0&\sqrt{\gamma}\\0&0\end{pmatrix} \qquad (4)

where: $\gamma$ — decay probability $1-e^{-t/T_1}$ over time $t$ ; $E_1=\sqrt{\gamma}\,|0\rangle\langle1|=\sqrt{\gamma}\,\sigma_-$ is the emission jump; $E_0$ is the no-jump operator, which nonetheless reweights $|1\rangle$ . Reading: one checks $E_0^\dagger E_0+E_1^\dagger E_1=\mathrm{diag}(1,1)=I$ , so the pair is a valid channel. The Stinespring dilation is the structural backbone: any channel is a unitary on system plus a fresh environment, followed by a partial trace. The Kraus operators are simply the matrix elements of that unitary in the environment basis.

\mathcal{E}(\rho)=\mathrm{Tr}_E\!\left[U\,(\rho\otimes|e_0\rangle\langle e_0|)\,U^{\dagger}\right],\qquad E_k=\langle e_k|U|e_0\rangle \qquad (5)

where: $U$ — joint unitary on system $\otimes$ environment; $|e_0\rangle$ — initial pure environment state; $\{|e_k\rangle\}$ — orthonormal environment basis used in the trace; $\mathrm{Tr}_E$ — partial trace over the environment. Reading: irreversibility is a bookkeeping artifact of discarding $E$ , not a new law of physics. Any single-qubit operator, Kraus operators included, is a complex combination of the four Paulis. This trivial linear-algebra fact is the seed of the entire discretization theorem.

E=\alpha I+\beta X+\gamma Y+\delta Z,\qquad \alpha,\beta,\gamma,\delta\in\mathbb{C} \qquad (6)

where: $E$ — an arbitrary single-qubit operator; the coefficients are $c_P=\tfrac12\mathrm{Tr}(P^\dagger E)$ for $P\in\{I,X,Y,Z\}$ . Reading: the Paulis form an orthogonal basis of $2\times2$ matrices under the Hilbert–Schmidt inner product, so this expansion always exists and is unique. Applying the amplitude-damping Kraus pair to a general single-qubit state gives its explicit action, which makes the $T_1$ / $T_2$ story quantitative.

\rho'=\begin{pmatrix}\rho_{00}+\gamma\rho_{11} & \sqrt{1-\gamma}\,\rho_{01}\\ \sqrt{1-\gamma}\,\rho_{10} & (1-\gamma)\rho_{11}\end{pmatrix} \qquad (7)

where: $\rho_{ij}$ — matrix elements of the input; $\gamma$ — decay probability. Reading: population $\rho_{11}$ scales by $1-\gamma$ while coherence $\rho_{01}$ scales by $\sqrt{1-\gamma}$ — the square-root is precisely why coherence lives twice as long as population. Finally, the phenomenological decay laws hardware reports. Relaxation empties the excited state on the timescale $T_1$ ; transverse coherence dies on $T_2$ , bounded by $2T_1$ .

\langle Z\rangle(t)=\big[\langle Z\rangle_0+1\big]e^{-t/T_1}-1,\qquad \langle X\rangle(t)=\langle X\rangle_0\,e^{-t/T_2},\qquad \frac{1}{T_2}=\frac{1}{2T_1}+\frac{1}{T_\phi} \qquad (8)

where: $T_1$ — energy-relaxation time (amplitude damping); $T_2$ — transverse coherence time; $T_\phi$ — pure-dephasing time. Reading: with no pure dephasing ( $T_\phi\to\infty$ ) one gets $T_2=2T_1$ , the ceiling; any extra dephasing only lowers $T_2$ .

Derivation AFrom environment monitoring to the phase-damping Kraus operators

The goal here is to watch decoherence emerge from nothing but a unitary. We let the system qubit interact with a single fresh environment qubit through a controlled coupling that copies the system's $Z$ value into the environment, then trace the environment out and read off the resulting Kraus operators. The payoff is a complete, honest account of dephasing: populations survive untouched, coherences shrink, and the "loss" is nothing more than information that has walked next door — exactly the einselection picture that opens this article, now made into two matrices.

Step 2. Extract the Kraus operators by sandwiching $U$ between environment basis states, $E_k=\langle e_k|U|e_0\rangle$ , giving $E_0=\mathrm{diag}(1,\cos\theta)$ and $E_1=\mathrm{diag}(0,\sin\theta)$ . Why: this is the Stinespring formula (5) in action; the two environment outcomes "no click" and "click" become the two Kraus branches, and $E_0^\dagger E_0+E_1^\dagger E_1=\mathrm{diag}(1,\cos^2\theta+\sin^2\theta)=I$ confirms completeness.

Step 3. Apply the pair to a general state to find $\mathcal{E}(\rho)=E_0\rho E_0^\dagger+E_1\rho E_1^\dagger=\begin{pmatrix}\rho_{00}&\cos\theta\,\rho_{01}\\ \cos\theta\,\rho_{10}&\rho_{11}\end{pmatrix}$ . Why: the diagonal entries — the populations — are completely unchanged, while the off-diagonal coherences are multiplied by $\cos\theta$ ; this is the signature of dephasing, decoherence with no energy exchange.

Step 4. Rewrite the coherence-damped state as a Pauli mixture, $\begin{pmatrix}\rho_{00}&c\rho_{01}\\ c\rho_{10}&\rho_{11}\end{pmatrix}=\tfrac{1+c}{2}\rho+\tfrac{1-c}{2}Z\rho Z$ with $c=\cos\theta$ , so $p_Z=\tfrac{1-\cos\theta}{2}$ . Why: since $Z\rho Z$ flips the sign of the off-diagonals, a convex mix of $\rho$ and $Z\rho Z$ reproduces any real coherence-scaling — the "continuous" monitoring is identical to a discrete random $Z$ flip with probability $p_Z$ , the fact that makes dephasing correctable by a phase-flip code. *Verified: $\theta=0$ gives $\mathcal{E}=\mathrm{id}$ ; $\theta=\pi/2$ gives full dephasing $\rho\mapsto\mathrm{diag}(\rho)$ ; identifying $\lambda=\sin^2\theta$ recovers $E_0=\mathrm{diag}(1,\sqrt{1-\lambda})$ . Matches Nielsen and Chuang section 8.3.5.*

Go deeper · Derivation B · 1 minDepolarizing noise as uniform Bloch-vector contraction

Derivation BDepolarizing noise as uniform Bloch-vector contraction

The aim is to show that the depolarizing channel does exactly one geometric thing — it shrinks the Bloch vector by a single scalar factor while leaving the center fixed — and to pin down that factor and the range of $p$ for which the map is physical. We parametrize an arbitrary state by its Bloch vector, push it through the channel using the Pauli conjugation identities, and read off the contraction, confirming that the maximally mixed state is the unique fixed point.

Step 1. Write $\rho=\tfrac12(I+\vec r\cdot\vec\sigma)$ and use $X(\vec r\cdot\vec\sigma)X=r_x\sigma_x-r_y\sigma_y-r_z\sigma_z$ together with its cyclic partners. Why: each Pauli conjugation preserves its own axis and flips the other two, so it acts as a $180^\circ$ rotation of the Bloch vector about that axis — the building block of the whole computation.

Step 2. Sum the three conjugations: $X\rho X+Y\rho Y+Z\rho Z=\tfrac32 I-\tfrac12\vec r\cdot\vec\sigma$ . Why: the identity parts add to $\tfrac32 I$ , while each Bloch component is flipped twice and kept once, giving a net coefficient $(-1-1+1)=-1$ ; the three Paulis together nearly erase the vector, leaving $-\tfrac12\vec r\cdot\vec\sigma$ .

Step 3. Insert into the definition, $\mathcal{E}_{\mathrm{dep}}(\rho)=(1-p)\tfrac12(I+\vec r\cdot\vec\sigma)+\tfrac{p}{3}\big(\tfrac32 I-\tfrac12\vec r\cdot\vec\sigma\big)$ , and collect terms to get $\mathcal{E}_{\mathrm{dep}}(\rho)=\tfrac12\big(I+(1-\tfrac{4p}{3})\vec r\cdot\vec\sigma\big)$ . Why: the identity coefficients cancel to leave a clean $\tfrac12 I$ , confirming the map is unital, while the vector coefficient collapses to the single contraction factor $1-\tfrac{4p}{3}$ .

Step 4. Demand that the output is a valid state for every input: $|1-\tfrac{4p}{3}|\le1$ requires $0\le p\le\tfrac34$ , with the factor hitting zero exactly at $p=\tfrac34$ . Why: at $p=\tfrac34$ the channel sends every state to the center $I/2$ regardless of input — total depolarization, consistent with placing weight $\tfrac14$ on each of $I,X,Y,Z$ — and beyond that $p$ the map would over-invert the Bloch ball and stop being positive. *Verified: unital ( $\mathcal{E}(I/2)=I/2$ ), contraction $1-\tfrac{4p}{3}$ , valid range $0\le p\le\tfrac34$ , fully mixing at $p=\tfrac34$ . Matches Preskill Chapter 3.*

Go deeper · Derivation C · 1 minWhy correcting the Pauli set defeats a continuum of errors

Derivation CWhy correcting the Pauli set defeats a continuum of errors

This is the theorem that makes quantum error correction possible at all. Physical noise is a continuum — tiny over-rotations, partial energy loss, drifting phases — and it would seem hopeless to correct each of infinitely many possibilities. The goal is to prove that a code able to correct just the finite set $\{I,X,Y,Z\}$ on the affected qubit automatically corrects any single-qubit channel whatsoever, by combining the Pauli expansion (6) with the collapse induced by syndrome measurement.

Step 1. Expand every Kraus operator of the noise in the Pauli basis, $E_k=e_{kI}I+e_{kX}X+e_{kY}Y+e_{kZ}Z$ . Why: by (6) this is always possible and unique, so an "arbitrary" error is really just a complex-weighted combination of the four discrete Paulis — the continuum has been packaged into finitely many operators.

Step 2. Act on an encoded logical state, $E_k|\psi_L\rangle=\sum_a e_{ka}P_a|\psi_L\rangle$ , a coherent superposition of the discretely corrupted codewords $P_a|\psi_L\rangle$ . Why: the corrupted state is not some new vector outside our reach — it lives in the span of the four Pauli-error versions of the codeword, a finite-dimensional trap we have built the code to handle.

Step 3. Measure the syndrome. If the correctable Paulis carry distinct syndromes (the Knill–Laflamme conditions), the measurement projects the superposition onto a single branch $P_a|\psi_L\rangle$ with its label $s_a$ . Why: measurement is what discretizes — a continuous superposition of error amplitudes collapses to one definite Pauli error, chosen with the Born-rule probability $|e_{ka}|^2$ , converting analog damage into a digital diagnosis.

Step 4. Apply the inverse Pauli, $P_a^{-1}(P_a|\psi_L\rangle)=|\psi_L\rangle$ , recovering the codeword exactly. Why: Paulis are their own inverses up to a phase, so a finite recovery gate set restores the state perfectly — a continuum of physical errors has been beaten by correcting only three of them, the central miracle behind the three-qubit, Shor, Steane, and surface codes. *Verified: this is the rigorous content behind the discretization theorem; see Nielsen and Chuang Theorem 10.1 and the Knill–Laflamme conditions.*

Limiting cases

Case	Result	Meaning
No-noise limit	$p\to0,\ \gamma\to0\implies\mathcal{E}\to\mathrm{id}$	Every channel reduces to the identity (Kraus set $\{I\}$ ) when the error probability or damping rate vanishes.
Maximal depolarization	$p=\tfrac34\implies\mathcal{E}(\rho)=\tfrac{I}{2}$	The Bloch vector is fully contracted; all input information is erased and the classical/quantum capacity is zero.
Full amplitude damping	$\gamma=1\implies\mathcal{E}(\rho)=\|0\rangle\langle0\|$	Every state decays to the ground state; the channel is a constant map (a reset), not merely noisy.
Strong dephasing	$p_Z\to\tfrac12\implies\rho\mapsto\mathrm{diag}(\rho)$	Off-diagonal coherences are fully killed; the channel becomes "measure in the computational basis and forget the result".
Pure relaxation, no dephasing	$T_\phi\to\infty\implies T_2=2T_1$	Amplitude damping alone sets the coherence ceiling; coherence decays as $e^{-t/2T_1}$ because it scales by $\sqrt{1-\gamma}$ .

Common trapAmplitude damping is just a special dephasing or Pauli channel

It is neither, and conflating them causes real errors in noise modeling. Amplitude damping has a Kraus operator $E_1=\sqrt{\gamma}\,|0\rangle\langle1|$ that is non-unitary and irreversible, and the channel is non-unital: it does not fix $I/2$ , because it actively pushes every state toward $|0\rangle$ . Pauli channels (depolarizing and dephasing included) are mixtures of unitaries and are always unital, so amplitude damping is fundamentally outside that family — it has a preferred direction in Hilbert space where the Pauli channels are directionless. A second, related trap: the Kraus decomposition is not unique. Two sets $\{E_k\}$ and $\{F_j\}$ describe the identical channel whenever $F_j=\sum_k u_{jk}E_k$ for an isometry $u$ , so no physical meaning attaches to a particular set of Kraus operators — only the map $\mathcal{E}$ is observable. Do not read "the environment emitted a photon" off one Kraus operator as if it were the true history; it is one of infinitely many equivalent unravelings.

Worked examples

1 · Amplitude damping of a superposition, and why T₂ = 2T₁ (fully worked)

A transmon with $T_1=100\ \mu\mathrm{s}$ and no extra pure dephasing is prepared in $|+\rangle$ and left idle for $t=10\ \mu\mathrm{s}$ . Find the decay probability, the output state, and confirm the coherence lifetime.

Step 1. Compute the damping parameter from the relaxation time: $\gamma=1-e^{-t/T_1}=1-e^{-10/100}=1-e^{-0.1}=1-0.9048=\mathbf{0.0952}$ . This is the probability that the excited amplitude has emitted a photon during the idle window.

Step 2. Write the input. $|+\rangle\langle+|=\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$ , so $\rho_{00}=\rho_{11}=\rho_{01}=\rho_{10}=\tfrac12$ . Apply the amplitude-damping action (7): the excited population becomes $(1-\gamma)\rho_{11}=0.9048\times0.5=\mathbf{0.4524}$ , and the ground population picks up the lost weight, $\rho_{00}+\gamma\rho_{11}=0.5+0.0952\times0.5=0.5476$ .

Step 3. The coherence scales by the square root: $\rho_{01}'=\sqrt{1-\gamma}\,\rho_{01}=\sqrt{0.9048}\times0.5=0.9512\times0.5=\mathbf{0.4756}$ . So $\rho'=\begin{pmatrix}0.5476 & 0.4756\\ 0.4756 & 0.4524\end{pmatrix}$ .

Step 4. Compare the two decay rates. Population moved by a factor $1-\gamma=e^{-t/T_1}$ , but coherence moved by $\sqrt{1-\gamma}=e^{-t/2T_1}$ . The coherence envelope therefore has time constant $2T_1$ , i.e. $T_2=2T_1=200\ \mu\mathrm{s}$ for pure amplitude damping. Any additional pure dephasing $T_\phi$ enters through $1/T_2=1/(2T_1)+1/T_\phi$ and can only shorten $T_2$ — which is why $T_2\le2T_1$ is a hard inequality, saturated exactly when relaxation is the sole noise source.

2 · Depolarizing strength from an average gate fidelity (fully worked)

A gate is characterized as a depolarizing channel with $p=0.03$ . Find the Bloch contraction factor and the average gate fidelity, and interpret both.

Step 1. Contraction: the Bloch vector shrinks by $1-\tfrac{4p}{3}=1-\tfrac{4(0.03)}{3}=1-0.04=\mathbf{0.96}$ . Every pure state moves 4% of the way toward the maximally mixed center.

Step 2. Average gate fidelity. Using the mixture form $\mathcal{E}(\rho)=(1-\tfrac{4p}{3})\rho+\tfrac{4p}{3}\tfrac{I}{2}$ , for any pure input $|\psi\rangle$ the overlap is $\langle\psi|\mathcal{E}(|\psi\rangle\langle\psi|)|\psi\rangle=(1-\tfrac{4p}{3})+\tfrac{4p}{3}\cdot\tfrac12=1-\tfrac{2p}{3}$ . Because this is independent of $|\psi\rangle$ , the average is the same value: $\bar F=1-\tfrac{2p}{3}=1-0.02=\mathbf{0.98}$ .

Step 3. Interpret. A per-gate error of a few percent is far above the surface-code threshold ( $\sim1\%$ error, i.e. $p\lesssim0.01$ – $0.02$ depending on the noise model), so a device at $p=0.03$ would see logical error rates grow rather than shrink under encoding. The gate must be improved to $p<0.01$ before error correction helps rather than hurts.

3 · Dephasing sends |+⟩ to the maximally mixed state

A qubit in $\rho=|+\rangle\langle+|$ passes through a phase-flip channel with $p_Z=\tfrac12$ . Find the output and its purity.

Step 1. The phase-flip channel multiplies off-diagonal coherences by $1-2p_Z$ . At $p_Z=\tfrac12$ this factor is $1-2(\tfrac12)=0$ , so both off-diagonal entries of $|+\rangle\langle+|=\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$ vanish while the diagonal is untouched. The output is $\rho'=\tfrac12\begin{pmatrix}1&0\\0&1\end{pmatrix}=\tfrac{I}{2}$ .

Step 2. Purity: $\mathrm{Tr}(\rho'^2)=\mathrm{Tr}((\tfrac{I}{2})^2)=\mathrm{Tr}(\tfrac{I}{4})=\tfrac12$ . The state has become maximally mixed — coherence fully destroyed, populations unchanged, purity fallen from 1 to 1/2 — the extreme of decoherence with zero energy exchange.

Physics of the portrait

The theory of decoherence turned a philosophical worry — why do we never see macroscopic superpositions? — into calculable channels. Its modern shape is owed above all to the physicist who insisted that the environment is not a nuisance to be minimized but the very agent that selects a classical world out of a quantum one.

Origins · 1 minWojciech Zurek and how this was discovered

Wojciech Zurek

b. 1951 · decoherence, einselection, and pointer states

Wojciech Hubert Zurek, a Polish-American theoretical physicist at Los Alamos, gave decoherence the operational teeth that make this whole article possible. Building on H. Dieter Zeh's 1970 proposal that entanglement with an environment suppresses interference, Zurek showed in the early 1980s that the environment does not merely blur a quantum system — it continuously monitors it, and in doing so selects a preferred set of "pointer states" that survive the coupling while their superpositions are rapidly destroyed. He named this environment-induced superselection, or einselection, and it explains exactly why a qubit coupled to $Z$ -noise decoheres in the computational basis: the pointer basis is the one the environment measures. That same picture is the intuition behind the phase-damping channel — a phase leaking into inaccessible modes kills coherence with no energy lost. Zurek connected the abstract CPTP formalism of Kraus and Stinespring to the physical mechanism of information flow into the environment; his 2003 review "Decoherence, einselection, and the quantum origins of the classical" became the standard reference. He is also known, with William Wootters, for the 1982 no-cloning theorem — itself a statement about why quantum information cannot simply be copied out of noise's reach. His fingerprints are on both halves of this subject: why qubits decohere, and why that decoherence is so hard to undo.

History

1955

— Stinespring proves the dilation theorem for completely positive maps

1970

— Zeh argues that entanglement with the environment destroys interference

1971

— Kraus casts general state changes as the operator-sum representation

1976

— Lindblad, Gorini, Kossakowski and Sudarshan derive the Markovian master equation

1981

— Zurek introduces pointer states and environment-induced superselection

Going further

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, Chapter 8 (quantum noise and quantum operations) and Chapter 10 (error correction) · Preskill, Lecture Notes for Physics 219, Chapters 3 and 7 · Wilde, Quantum Information Theory, Chapter 4 (noisy quantum channels) · Watrous, The Theory of Quantum Information, Chapter 2 (channels and operations) · Kitaev, Shen and Vyalyi, Classical and Quantum Computation, Chapter 11.

Provenance. All channel definitions, Kraus operators, and Bloch-ball transformations follow Nielsen and Chuang Chapter 8 (verify section 8.3.3 depolarizing, 8.3.4 amplitude damping, 8.3.5 phase damping) and Preskill Chapter 3. A verifier should: (1) confirm $\sum_k E_k^\dagger E_k=I$ for each channel; (2) recompute the depolarizing contraction $1-\tfrac{4p}{3}$ and validity range $0\le p\le\tfrac34$ ; (3) check the amplitude-damping output (7), the average fidelity $\bar F=1-\tfrac{2p}{3}$ , and the identity $T_2=2T_1$ for pure damping via $\sqrt{1-\gamma}=e^{-t/2T_1}$ ; (4) verify the discretization argument against Nielsen and Chuang Theorem 10.1. No device-specific numbers are fabricated; the worked $T_1=100\ \mu\mathrm{s}$ example is illustrative and the threshold figures are quoted as standard ranges.

Prerequisites. Density Operators and Mixed States · Pauli Group and Clifford Group · Unitary Evolution and the Born Rule. Leads to. Discretization of Errors · Three-Qubit Bit-Flip Code · Knill–Laflamme Conditions · Stabilizer Formalism.

Ask the tutor: Ask me to run a specific state through any of the three channels and watch its Bloch vector, purity, or fidelity change step by step.

What exactly makes a map a "quantum channel" and not just any linear map?

Two conditions, both physical. First, trace preservation: $\mathrm{Tr}\,\mathcal{E}(\rho)=\mathrm{Tr}\,\rho$ , so probabilities still sum to one. Second, and more subtly, complete positivity: not only must $\mathcal{E}$ map states to states, but $\mathcal{E}\otimes\mathrm{id}$ must do so too, even when the qubit is entangled with an untouched partner. Ordinary positivity is not enough — the transpose map is positive but not completely positive, and applying it to half of an entangled pair yields a negative "probability", which is why transposition is unphysical. Every completely positive trace-preserving (CPTP) map has a Kraus form and a Stinespring dilation, and conversely; these are three names for the same set of allowed noise processes.

Is decoherence a genuinely new, irreversible law of physics?

No, and this is the conceptual centerpiece. The Stinespring dilation says any channel is a unitary $U$ acting on the system plus a fresh environment, followed by discarding the environment. Unitary evolution is perfectly reversible and information-preserving; the apparent irreversibility of decoherence comes entirely from the partial trace — we choose to ignore, or simply cannot access, the environment degrees of freedom that now hold the missing information. If you could grab the environment and run $U^{-1}$ , the coherence would come back (this is the physics behind spin echoes and dynamical decoupling). So decoherence is not fundamental irreversibility; it is bookkeeping loss, unitary quantum mechanics on a Hilbert space too large to track.

Why is amplitude damping singled out as "not a Pauli channel"?

Because it has a direction. Pauli channels — depolarizing, dephasing, bit-flip — are mixtures of unitaries, and every such mixture fixes the maximally mixed state $I/2$ (they are unital). Amplitude damping does not fix $I/2$ : it pushes every state toward the ground state $|0\rangle$ , so its fixed point is a pole of the Bloch ball, not the center. Its jump operator $E_1=\sqrt{\gamma}\,|0\rangle\langle1|$ is non-unitary and non-invertible, mapping two dimensions onto one and discarding information about the excited amplitude. You can still expand its Kraus operators in the Pauli basis for error correction, but you must not model it as a random Pauli flip; that would wrongly predict a unital, direction-free noise where the real device is strongly biased toward $|0\rangle$ .

The Kraus operators are not unique — so which ones are "real"?

None of them individually. Any two Kraus sets related by $F_j=\sum_k u_{jk}E_k$ for an isometry $u$ produce the identical channel $\mathcal{E}$ , and $\mathcal{E}$ is the only thing you can measure. This mirrors the freedom to "unravel" the same open-system dynamics into different quantum-trajectory pictures: one unraveling of amplitude damping says "either a photon was emitted or it wasn't", another mixes those alternatives, and both reproduce the same $\rho'$ . The lesson is to attach physical meaning to the map, to its fixed points and its Bloch-ball action, but not to any single Kraus operator's story about what "happened". When a particular dilation corresponds to a real measurement you actually perform on the environment, then that unraveling becomes physically meaningful — but only because of the extra measurement, not the algebra.

How are T₁ and T₂ related to γ and the dephasing parameter?

$T_1$ is the amplitude-damping timescale: the decay probability over time $t$ is $\gamma=1-e^{-t/T_1}$ , so the excited-state population falls as $e^{-t/T_1}$ . Coherence decay is governed by $T_2$ , and it has two contributions. Amplitude damping alone shrinks coherence by $\sqrt{1-\gamma}=e^{-t/2T_1}$ , contributing a rate $1/(2T_1)$ . Pure dephasing — random $Z$ noise with no energy exchange — adds an independent rate $1/T_\phi$ . The two combine as $1/T_2=1/(2T_1)+1/T_\phi$ , so $T_2\le2T_1$ always, with equality only when there is no pure dephasing. This is why hardware reports both numbers: $T_1$ fixes how fast energy leaks, and the gap between $T_2$ and $2T_1$ tells you how much extra phase noise the device carries.

Why does syndrome measurement "discretize" a continuous error?

Because measurement collapses a superposition, and after the Pauli expansion the error is literally a superposition of discrete Pauli-corrupted codewords. A small over-rotation writes the state as $\alpha\,|\psi_L\rangle+\beta\,X|\psi_L\rangle+\dots$ with tiny $\beta$ . Measuring the syndrome asks "which Pauli happened?" and the Born rule returns a definite answer — usually "no error", occasionally "an $X$ on qubit 3" — with the squared amplitudes as probabilities. The continuous parameter $\beta$ does not survive; you are left with one discrete Pauli error, undone by one discrete recovery gate. The analog nature of the noise is projected away, and a finite code defeats a continuum — the single most important reason quantum error correction is possible at all.