Skip to content

Dephasing Channel

The dephasing channel is the basic channel for loss of phase coherence in a preferred basis. It suppresses off-diagonal density-matrix elements while leaving the corresponding populations unchanged.

For a qubit in the ZZ basis, the defining action is

(ρ00ρ01ρ10ρ11)↦(ρ00λρ01λ∗ρ10ρ11),∣λ∣≤1.\begin{pmatrix} \rho_{00} & \rho_{01}\\ \rho_{10} & \rho_{11} \end{pmatrix} \mapsto \begin{pmatrix} \rho_{00} & \lambda\rho_{01}\\ \lambda^*\rho_{10} & \rho_{11} \end{pmatrix}, \qquad |\lambda|\le1.

The parameter λ\lambda is the coherence factor. Its magnitude is the remaining coherence; its phase is a coherent ZZ rotation that can often be separated from the irreversible dephasing.

Dephasing is not energy relaxation. In the dephasing basis, populations are unchanged:

ρ00′=ρ00,ρ11′=ρ11.\rho_{00}'=\rho_{00}, \qquad \rho_{11}'=\rho_{11}.

Only coherences are damped:

ρ01′=λρ01.\rho_{01}'=\lambda\rho_{01}.

This is why pure dephasing is the canonical model for:

  • fluctuating energy splittings,
  • unread which-path information,
  • elastic scattering that distinguishes alternatives,
  • nonselective projective measurement in a basis,
  • slow classical phase noise,
  • environment records that become correlated with basis states.

The basis matters. A channel that is pure dephasing in one basis may change populations when viewed in another basis.

For the broader physical distinction, see Dephasing vs Dissipation.

Dephasing and Amplitude Damping owns the QI-facing conversion from protocol-qualified relaxation, coherence, and equilibrium-population records to a finite idle channel, including physicality, composition, and held-out audits; this page retains the dephasing-channel definition, Kraus and Choi forms, environment-record and random-phase pictures, higher-dimensional extension, and formal semigroup limit.

For real −1≤λ≤1-1\le\lambda\le1, the qubit dephasing channel can be written as a phase-flip mixture:

Φλ(ρ)=1+λ2ρ+1−λ2ZρZ.\Phi_\lambda(\rho) = \frac{1+\lambda}{2}\rho + \frac{1-\lambda}{2}Z\rho Z.

A Kraus representation is

K0=1+λ2 I,K1=1−λ2 Z.K_0 = \sqrt{\frac{1+\lambda}{2}}\,I, \qquad K_1 = \sqrt{\frac{1-\lambda}{2}}\,Z.

The trace-preserving condition is

K0†K0+K1†K1=I.K_0^\dagger K_0+K_1^\dagger K_1=I.

Applying the channel to the density matrix gives

Φλ(ρ)=(ρ00λρ01λρ10ρ11)\Phi_\lambda(\rho) = \begin{pmatrix} \rho_{00} & \lambda\rho_{01}\\ \lambda\rho_{10} & \rho_{11} \end{pmatrix}

when λ\lambda is real. If λ\lambda is complex, write λ=∣λ∣e−iθ\lambda=|\lambda|e^{-i\theta} and treat the phase as a ZZ rotation together with damping by ∣λ∣|\lambda|.

Equivalently, use an error probability

p=1−λ2.p=\frac{1-\lambda}{2}.

Then

Φp(ρ)=(1−p)ρ+pZρZ,0≤p≤1.\Phi_p(\rho) = (1-p)\rho+pZ\rho Z, \qquad 0\le p\le1.

This is the ZZ-axis special case of a Pauli channel.

The Markovian pure-dephasing semigroup has 0≤p(t)≤1/20\le p(t)\le1/2 and λ(t)=e−Γϕt\lambda(t)=e^{-\Gamma_\phi t}; allowing p>1/2p>1/2 is still a valid phase-flip channel but not the usual monotone Markovian decay convention.

Write a qubit state as

ρ=12(I+rxX+ryY+rzZ).\rho = \frac{1}{2} \left( I+r_xX+r_yY+r_zZ \right).

For real λ\lambda, the dephasing channel acts as

(rx,ry,rz)↦(λrx,λry,rz).(r_x,r_y,r_z) \mapsto (\lambda r_x,\lambda r_y,r_z).

The Bloch sphere is contracted toward the ZZ axis. The north and south pole states are fixed, while transverse superpositions lose contrast.

Repeated dephasing multiplies coherence factors, and the fixed states are exactly diagonal in the dephasing basis. See Channel Composition and Fixed Points for the general channel language.

Complete dephasing is the case λ=0\lambda=0:

(rx,ry,rz)↦(0,0,rz).(r_x,r_y,r_z) \mapsto (0,0,r_z).

It maps every state to its diagonal part in the ZZ basis.

Using the output-input convention of Choi Matrix, the Choi matrix of the qubit dephasing channel is

JΦλ=(100λ00000000λ∗001).J_{\Phi_\lambda} = \begin{pmatrix} 1 & 0 & 0 & \lambda\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ \lambda^* & 0 & 0 & 1 \end{pmatrix}.

The nonzero block has eigenvalues

1+∣λ∣,1−∣λ∣.1+|\lambda|, \qquad 1-|\lambda|.

Therefore the channel is completely positive exactly when

∣λ∣≤1.|\lambda|\le1.

The Choi rank is 22 for 0<∣λ∣<10\lt|\lambda|\lt1, rank 11 for a unitary phase rotation with ∣λ∣=1|\lambda|=1, and rank 22 at complete dephasing for the qubit channel with λ=0\lambda=0.

An unread projective measurement in the ZZ basis gives complete dephasing:

ΔZ(ρ)=P0ρP0+P1ρP1,\Delta_Z(\rho) = P_0\rho P_0+P_1\rho P_1,

where

P0=∣0⟩⟨0∣,P1=∣1⟩⟨1∣.P_0=\lvert0\rangle\langle0\rvert, \qquad P_1=\lvert1\rangle\langle1\rvert.

In matrix form,

ΔZ((ρ00ρ01ρ10ρ11))=(ρ0000ρ11).\Delta_Z \left( \begin{pmatrix} \rho_{00} & \rho_{01}\\ \rho_{10} & \rho_{11} \end{pmatrix} \right) = \begin{pmatrix} \rho_{00} & 0\\ 0 & \rho_{11} \end{pmatrix}.

Thus an unread ideal measurement is not “no measurement.” It is a channel that removes coherence in the measured decomposition. See Measurement Backaction for the operational distinction.

Suppose the environment becomes correlated with the dephasing basis:

∣0⟩∣e⟩↦∣0⟩∣e0⟩,∣1⟩∣e⟩↦∣1⟩∣e1⟩.\lvert0\rangle\lvert e\rangle \mapsto \lvert0\rangle\lvert e_0\rangle, \qquad \lvert1\rangle\lvert e\rangle \mapsto \lvert1\rangle\lvert e_1\rangle.

For an input coherence ρ01\rho_{01}, tracing out the environment gives

ρ01↦⟨e1∣e0⟩ ρ01.\rho_{01} \mapsto \langle e_1|e_0\rangle\,\rho_{01}.

Thus

λ=⟨e1∣e0⟩.\lambda=\langle e_1|e_0\rangle.

Orthogonal environment records give complete dephasing. Identical environment states give no dephasing. Partial distinguishability gives partial dephasing.

This is the simplest channel version of Environment-Induced Decoherence.

Classical random phase noise gives another representation. Let

Uϕ=e−iϕZ/2U_\phi = e^{-i\phi Z/2}

and average over phases with probability density p(ϕ)p(\phi):

Φ(ρ)=∫dϕ p(ϕ) UϕρUϕ†.\Phi(\rho) = \int d\phi\,p(\phi)\, U_\phi\rho U_\phi^\dagger.

The off-diagonal element transforms as

ρ01↦[∫dϕ p(ϕ)e−iϕ]ρ01.\rho_{01} \mapsto \left[ \int d\phi\,p(\phi)e^{-i\phi} \right] \rho_{01}.

Therefore

λ=⟨e−iϕ⟩.\lambda = \langle e^{-i\phi}\rangle.

For a Gaussian random phase with mean 00 and variance σϕ2\sigma_\phi^2,

λ=e−σϕ2/2.\lambda=e^{-\sigma_\phi^2/2}.

Random-unitary phase averaging is a useful model for classical frequency noise. A quantum bath can produce the same channel form while carrying quantum correlations rather than a classical random label.

A Markovian pure-dephasing master equation can be written

dρdt=Γϕ2(ZρZ−ρ).\frac{d\rho}{dt} = \frac{\Gamma_\phi}{2} \left( Z\rho Z-\rho \right).

The populations are constant:

dρ00dt=0,dρ11dt=0.\frac{d\rho_{00}}{dt}=0, \qquad \frac{d\rho_{11}}{dt}=0.

The coherence obeys

dρ01dt=−Γϕρ01.\frac{d\rho_{01}}{dt} = -\Gamma_\phi\rho_{01}.

Hence

λ(t)=e−Γϕt.\lambda(t)=e^{-\Gamma_\phi t}.

The corresponding phase-flip probability is

p(t)=1−e−Γϕt2.p(t) = \frac{1-e^{-\Gamma_\phi t}}{2}.

For the exact coherence-factor model behind nonexponential dephasing, see Pure Dephasing Model. For the generator-level assumptions, rate conventions, and T2T_2 relation, see Pure Dephasing Master Equation. In T2T_2 language, this pure-dephasing contribution is often written as Tϕ−1=ΓϕT_\phi^{-1}=\Gamma_\phi, with relaxation contributions added separately.

For projectors {Pa}\{P_a\} onto a preferred orthogonal decomposition, complete dephasing is

Δ(ρ)=∑aPaρPa.\Delta(\rho) = \sum_a P_a\rho P_a.

More general partial dephasing can multiply blocks by coherence factors:

Φ(ρ)=∑a,bΓabPaρPb.\Phi(\rho) = \sum_{a,b} \Gamma_{ab} P_a\rho P_b.

For rank-one projectors, complete positivity is equivalent to the matrix Γ\Gamma being positive semidefinite with

Γaa=1.\Gamma_{aa}=1.

This Γ\Gamma can be interpreted as a Gram matrix of environment states:

Γab=⟨eb∣ea⟩.\Gamma_{ab} = \langle e_b|e_a\rangle.

The qubit channel is the two-dimensional special case.

Dephasing always refers to a basis or decomposition. Changing the basis changes what counts as population and coherence.

Confusing dephasing with depolarizing noise

Section titled “Confusing dephasing with depolarizing noise”

Dephasing contracts only transverse directions relative to the dephasing axis. Depolarizing noise contracts all Bloch directions equally.

Confusing dephasing with amplitude damping

Section titled “Confusing dephasing with amplitude damping”

Pure dephasing does not transfer population between energy eigenstates in the dephasing basis. Amplitude damping does.

If which-outcome information has leaked into an apparatus or environment and is then ignored, the reduced channel is dephasing. The coherence is not restored by merely forgetting to read the record.

Overinterpreting a negative coherence factor

Section titled “Overinterpreting a negative coherence factor”

A real λ<0\lambda\lt0 is allowed by complete positivity when ∣λ∣≤1|\lambda|\le1; it includes an additional phase flip. It is not the usual one-parameter Markovian pure-dephasing semigroup, where λ(t)≥0\lambda(t)\ge0.

Using

K0=1+λ2 I,K1=1−λ2 Z,K_0=\sqrt{\frac{1+\lambda}{2}}\,I, \qquad K_1=\sqrt{\frac{1-\lambda}{2}}\,Z,

show that ρ01↦λρ01\rho_{01}\mapsto\lambda\rho_{01} for real λ\lambda.

Solution

The channel is

Φλ(ρ)=1+λ2ρ+1−λ2ZρZ.\Phi_\lambda(\rho) = \frac{1+\lambda}{2}\rho + \frac{1-\lambda}{2}Z\rho Z.

Since Z=diag⁡(1,−1)Z=\operatorname{diag}(1,-1),

(ZρZ)01=−ρ01.(Z\rho Z)_{01}=-\rho_{01}.

Therefore

(Φλ(ρ))01=1+λ2ρ01−1−λ2ρ01=λρ01.(\Phi_\lambda(\rho))_{01} = \frac{1+\lambda}{2}\rho_{01} - \frac{1-\lambda}{2}\rho_{01} = \lambda\rho_{01}.

Show that real-λ\lambda dephasing maps (rx,ry,rz)(r_x,r_y,r_z) to (λrx,λry,rz)(\lambda r_x,\lambda r_y,r_z).

Solution

For

ρ=12(1+rzrx−iryrx+iry1−rz),\rho = \frac12 \begin{pmatrix} 1+r_z & r_x-ir_y\\ r_x+ir_y & 1-r_z \end{pmatrix},

dephasing leaves the diagonal entries fixed and multiplies off-diagonal entries by λ\lambda. Hence

rz′=rz,rx′−iry′=λ(rx−iry).r_z' = r_z, \qquad r_x'-ir_y'=\lambda(r_x-ir_y).

For real λ\lambda this gives

rx′=λrx,ry′=λry.r_x'=\lambda r_x, \qquad r_y'=\lambda r_y.

Let ϕ\phi be a zero-mean Gaussian random variable with variance σϕ2\sigma_\phi^2. Show that ⟨e−iϕ⟩=e−σϕ2/2\langle e^{-i\phi}\rangle=e^{-\sigma_\phi^2/2}.

Solution

The characteristic function of a zero-mean Gaussian with variance σϕ2\sigma_\phi^2 is

⟨eikϕ⟩=e−k2σϕ2/2.\langle e^{ik\phi}\rangle = e^{-k^2\sigma_\phi^2/2}.

Set k=−1k=-1 to obtain

⟨e−iϕ⟩=e−σϕ2/2.\langle e^{-i\phi}\rangle = e^{-\sigma_\phi^2/2}.
  • K. Kraus, States, Effects, and Operations: Fundamental Notions of Quantum Theory, Springer (1983).
  • M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, Cambridge University Press (2010).
  • H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press (2002).
  • M. Schlosshauer, Decoherence and the Quantum-to-Classical Transition, Springer (2007).
  • H. M. Wiseman and G. J. Milburn, Quantum Measurement and Control, Cambridge University Press (2010).