Simple Algorithm
Quantum Computing · Excited State · 22 min

Decoherence: T₁, T₂ and Noise

How the environment steals quantum information.

A qubit is never truly alone. Stray photons, fluctuating spins and wobbling control lines couple it to a vast environment, and whatever the environment learns, the qubit loses. This lesson makes that precise: noise as quantum channels, the two canonical processes behind T1T_1 and T2T_2, and the Bloch equations that every experimentalist fits.

What you’ll learn

  • Derive the Kraus (operator-sum) form of noise by tracing out an environment.
  • Write the Kraus operators of the dephasing, amplitude-damping and depolarizing channels and compute their action on the Bloch vector.
  • Prove that T2≤2T1T_2\le2T_1 and solve the Bloch equations.
  • Distinguish T2∗T_2^* from the echo T2T_2, and coherent errors from stochastic ones.

Decoherence is entanglement with the environment

Let a qubit in α∣0⟩+β∣1⟩\alpha\ket0+\beta\ket1 interact with an environment EE that starts in ∣e⟩\ket{e}. Suppose the interaction leaves ∣0⟩\ket0 and ∣1⟩\ket1 themselves alone but imprints which one is present on the environment:

(α∣0⟩+β∣1⟩)∣e⟩  ⟼  α∣0⟩∣e0⟩+β∣1⟩∣e1⟩\big(\alpha\ket{0}+\beta\ket{1}\big)\ket{e}\;\longmapsto\;\alpha\ket{0}\ket{e_0}+\beta\ket{1}\ket{e_1}
A pure-dephasing interaction. The environment states ∣e0⟩\ket{e_0}, ∣e1⟩\ket{e_1} need not be orthogonal.

Tracing out the environment with the partial trace gives the qubit's density matrix:

ρ=(∣α∣2αβ∗⟨e1∣e0⟩α∗β⟨e0∣e1⟩∣β∣2)\rho = \begin{pmatrix} \abs{\alpha}^2 & \alpha\beta^*\braket{e_1|e_0} \\ \alpha^*\beta\braket{e_0|e_1} & \abs{\beta}^2 \end{pmatrix}
The populations are untouched; the coherences are multiplied by the overlap ⟨e1∣e0⟩\braket{e_1|e_0}.
DerivationEvery noise process is a channel
  1. ρ  ⟼  Tr⁡E ⁣[ U(ρ⊗∣e0⟩ ⁣⟨e0∣)U†]\rho\;\longmapsto\;\Tr_E\!\big[\,U(\rho\otimes\dyad{e_0})U^\dagger\big]

    Couple the system to an environment in a fixed state ∣e0⟩\ket{e_0}, evolve jointly with a unitary UU, and discard the environment:

  2. =∑k⟨ek∣U∣e0⟩ ρ ⟨e0∣U†∣ek⟩= \sum_k \bra{e_k}U\ket{e_0}\,\rho\,\bra{e_0}U^\dagger\ket{e_k}

    Evaluate the trace in an orthonormal environment basis {∣ek⟩}\{\ket{e_k}\}:

  3. E(ρ)=∑kKk ρ Kk†\mathcal{E}(\rho) = \sum_k K_k\,\rho\,K_k^\dagger

    Each Kk:=⟨ek∣U∣e0⟩K_k := \bra{e_k}U\ket{e_0} is an operator on the system alone. These are the Kraus operators:

  4. ∑kKk†Kk=⟨e0∣U†(∑k∣ek⟩ ⁣⟨ek∣)U∣e0⟩=⟨e0∣U†U∣e0⟩=I\sum_k K_k^\dagger K_k = \bra{e_0}U^\dagger\Big(\sum_k\dyad{e_k}\Big)U\ket{e_0} = \bra{e_0}U^\dagger U\ket{e_0} = \id

    Unitarity of UU gives the completeness relation, so E\mathcal{E} preserves the trace:

  5. Maps of this form are exactly the completely positive, trace-preserving maps, the quantum channels. Conversely, every channel arises from some system–environment unitary (Stinespring's theorem).

Check your understanding

Which pair of operators is a valid set of Kraus operators for a qubit channel?

  1. K0=12I, K1=12XK_0 = \tfrac12\id,\ K_1 = \tfrac12X
  2. K0=0.9 I, K1=0.1 ZK_0 = \sqrt{0.9}\,\id,\ K_1 = \sqrt{0.1}\,Z
  3. K0=0.9 I, K1=0.9 XK_0 = \sqrt{0.9}\,\id,\ K_1 = \sqrt{0.9}\,X
  4. K0=I, K1=0.1 ZK_0 = \id,\ K_1 = 0.1\,Z
Show the answer

B. K0=0.9 I, K1=0.1 ZK_0 = \sqrt{0.9}\,\id,\ K_1 = \sqrt{0.1}\,Z

∑kKk†Kk=0.9 I+0.1 Z†Z=I\sum_kK_k^\dagger K_k = 0.9\,\id+0.1\,Z^\dagger Z = \id. This is the phase-flip channel: with probability 0.10.1 a ZZ error occurs.

Dephasing: the phase-flip channel

The simplest model of dephasing is the phase-flip channel: with probability pp a Pauli ZZ error occurs.

EPF(ρ)=(1−p) ρ+p ZρZ,K0=1−p I,K1=p Z\mathcal{E}_{\mathrm{PF}}(\rho) = (1-p)\,\rho+p\,Z\rho Z,\qquad K_0 = \sqrt{1-p}\,\id,\quad K_1 = \sqrt{p}\,Z

Write the qubit state as ρ=12(I+xX+yY+zZ)\rho = \tfrac12(\id+xX+yY+zZ). Since ZXZ=−XZXZ = -X, ZYZ=−YZYZ = -Y and ZZZ=ZZZZ = Z, the channel acts on the Bloch vector as (x,y,z)↦((1−2p)x, (1−2p)y, z)(x,y,z)\mapsto\big((1-2p)x,\,(1-2p)y,\,z\big): the Bloch ball is squashed toward the zz-axis. In the matrix, the coherences ρ01\rho_{01} are multiplied by 1−2p1-2p and the populations are untouched.

Where does continuous-time dephasing come from? If the qubit frequency jitters by a random amount, the qubit picks up a random phase φ\varphi, and averaging eiφe^{i\varphi} over the noise shrinks the coherence. For a Gaussian phase of variance σ2\sigma^2, ⟨eiφ⟩=e−σ2/2\langle e^{i\varphi}\rangle = e^{-\sigma^2/2}. When the frequency noise is fast (white), σ2\sigma^2 grows linearly in time and the coherence decays exponentially, ρ01(t)=ρ01(0) e−t/Tφ\rho_{01}(t) = \rho_{01}(0)\,e^{-t/T_\varphi}, with TφT_\varphi the pure-dephasing time.

Amplitude damping: energy relaxation

T1T_1 processes are irreversible energy loss: an excited qubit emits a photon (or phonon, or quasiparticle) and falls to ∣0⟩\ket0. Model the environment as a mode that starts empty and can receive one excitation:

∣0⟩∣0⟩E↦∣0⟩∣0⟩E,∣1⟩∣0⟩E↦1−γ ∣1⟩∣0⟩E+γ ∣0⟩∣1⟩E\ket{0}\ket{0}_E\mapsto\ket{0}\ket{0}_E,\qquad \ket{1}\ket{0}_E\mapsto\sqrt{1-\gamma}\,\ket{1}\ket{0}_E+\sqrt{\gamma}\,\ket{0}\ket{1}_E
With probability γ\gamma the excitation has leaked into the environment.

Reading off Kk=⟨k∣EU∣0⟩EK_k = \bra{k}_EU\ket{0}_E gives the amplitude-damping channel:

K0=(1001−γ),K1=(0γ00)=γ ∣0⟩ ⁣⟨1∣K_0 = \begin{pmatrix}1&0\\0&\sqrt{1-\gamma}\end{pmatrix},\qquad K_1 = \begin{pmatrix}0&\sqrt{\gamma}\\0&0\end{pmatrix} = \sqrt{\gamma}\,\ketbra{0}{1}
DerivationAction on the density matrix
  1. K0†K0+K1†K1=(1001−γ)+(000γ)=IK_0^\dagger K_0+K_1^\dagger K_1 = \begin{pmatrix}1&0\\0&1-\gamma\end{pmatrix}+\begin{pmatrix}0&0\\0&\gamma\end{pmatrix} = \id

    Completeness:

  2. EAD(ρ00ρ01ρ10ρ11)=(ρ00+γρ111−γ ρ011−γ ρ10(1−γ) ρ11)\mathcal{E}_{\mathrm{AD}}\begin{pmatrix}\rho_{00}&\rho_{01}\\ \rho_{10}&\rho_{11}\end{pmatrix} = \begin{pmatrix}\rho_{00}+\gamma\rho_{11} & \sqrt{1-\gamma}\,\rho_{01}\\ \sqrt{1-\gamma}\,\rho_{10} & (1-\gamma)\,\rho_{11}\end{pmatrix}

    Apply the channel to a general state:

  3. ρ11(t)=e−t/T1ρ11(0),ρ01(t)=e−t/(2T1)ρ01(0)\rho_{11}(t) = e^{-t/T_1}\rho_{11}(0),\qquad \rho_{01}(t) = e^{-t/(2T_1)}\rho_{01}(0)

    For relaxation at a constant rate, γ(t)=1−e−t/T1\gamma(t) = 1-e^{-t/T_1}. Population and coherence then decay at different rates:

  4. (x, y, z)  ⟼  (1−γ x, 1−γ y, γ+(1−γ)z)(x,\,y,\,z)\;\longmapsto\;\big(\sqrt{1-\gamma}\,x,\ \sqrt{1-\gamma}\,y,\ \gamma+(1-\gamma)z\big)

    On the Bloch vector, xx and yy shrink by 1−γ\sqrt{1-\gamma} while zz moves toward +1+1. The ball shrinks and slides toward the north pole ∣0⟩\ket0, the only fixed point. Relaxation alone gives T2=2T1T_2 = 2T_1.

Predict before you look

A qubit in ∣+⟩\ket+ undergoes amplitude damping with γ=0.36\gamma = 0.36. What is the probability (in %) of measuring 11 afterwards?

Show the answer

32 %

ρ11\rho_{11} starts at 12\tfrac12 and is multiplied by 1−γ=0.641-\gamma = 0.64, so P(1)=0.32P(1) = 0.32. The coherence shrinks only by 0.64=0.8\sqrt{0.64} = 0.8, more slowly than the population: that is the factor of 2 in T2=2T1T_2 = 2T_1.

T₂ ≤ 2T₁ and the Bloch equations

Real qubits suffer both processes at once. Coherence decays at the sum of half the relaxation rate (from amplitude damping) and the pure-dephasing rate:

1T2=12T1+1Tφ⟹T2≤2T1\frac{1}{T_2} = \frac{1}{2T_1}+\frac{1}{T_\varphi}\qquad\Longrightarrow\qquad T_2\le2T_1
Equality holds exactly when there is no pure dephasing (Tφ→∞T_\varphi\to\infty).
DerivationWhy T₂ > 2T₁ would be unphysical
  1. ρ11(t)=12e−t/T1,ρ00(t)=1−12e−t/T1,∣ρ01(t)∣=12e−t/T2\rho_{11}(t) = \tfrac12e^{-t/T_1},\qquad \rho_{00}(t) = 1-\tfrac12e^{-t/T_1},\qquad \abs{\rho_{01}(t)} = \tfrac12e^{-t/T_2}

    Suppose relaxation toward ∣0⟩\ket0 at rate 1/T11/T_1 and coherence decay at rate 1/T21/T_2. Start in ∣+⟩\ket+:

  2. ∣ρ01∣2≤ρ00 ρ11\abs{\rho_{01}}^2\le\rho_{00}\,\rho_{11}

    A density matrix must be positive semidefinite. For a 2×22\times2 matrix with nonnegative diagonal this is the condition det⁡ρ≥0\det\rho\ge0:

  3. ∣ρ01∣2ρ00 ρ11  ∼  12exp⁡ ⁣[t(1T1−2T2)]\frac{\abs{\rho_{01}}^2}{\rho_{00}\,\rho_{11}}\;\sim\;\frac12\exp\!\Big[t\Big(\frac{1}{T_1}-\frac{2}{T_2}\Big)\Big]

    At long times the left side behaves as 14e−2t/T2\tfrac14e^{-2t/T_2} and the right side as 12e−t/T1\tfrac12e^{-t/T_1}:

  4. T2>2T1  ⟹  det⁡ρ(t)<0  for large tT_2>2T_1\;\Longrightarrow\;\det\rho(t)<0\ \text{ for large } t

    If T2>2T1T_2>2T_1 the exponent is positive and the ratio eventually exceeds 1: the "state" would have a negative eigenvalue. Positivity alone forces T2≤2T1T_2\le2T_1.

Collecting everything gives the Bloch equations. Work in a frame rotating close to the qubit frequency, with a residual detuning Δ\Delta (Hamiltonian H=Δ2ZH = \tfrac\Delta2Z), and relaxation toward ∣0⟩\ket0 at zero temperature:

x˙=−Δ y−xT2,y˙=Δ x−yT2,z˙=1−zT1\dot x = -\Delta\,y-\frac{x}{T_2},\qquad \dot y = \Delta\,x-\frac{y}{T_2},\qquad \dot z = \frac{1-z}{T_1}
Solution: x+iy=(x0+iy0) eiΔte−t/T2x+iy = (x_0+iy_0)\,e^{i\Delta t}e^{-t/T_2} and z=1−(1−z0) e−t/T1z = 1-(1-z_0)\,e^{-t/T_1}. At temperature T>0T>0 the equilibrium value 11 becomes tanh⁡(ℏωq/2kBT)\tanh(\hbar\omega_q/2k_BT).
Predict before you look

A qubit has T1=80 μsT_1 = 80\,\mu\text{s} and pure-dephasing time Tφ=120 μsT_\varphi = 120\,\mu\text{s}. What is T2T_2, in μs?

Show the answer

69 μs

1/T2=1/160+1/120=7/4801/T_2 = 1/160+1/120 = 7/480, so T2≈68.6 μsT_2\approx68.6\,\mu\text{s}, shorter than both 2T1=160 μs2T_1 = 160\,\mu\text{s} and TφT_\varphi. Rates add, so times combine like resistors in parallel.

Measuring noise, and other error models

Experiments distinguish two dephasing times. A Ramsey sequence (π/2\pi/2 pulse, free evolution, π/2\pi/2 pulse) measures T2∗T_2^*, which includes slow drifts of the qubit frequency that differ from shot to shot. A Hahn echo inserts a π\pi pulse halfway: phase gained before the pulse is undone after it, as long as the frequency offset did not change in between. Echoes remove quasi-static noise, so T2∗≤T2echo≤2T1T_2^*\le T_2^{\text{echo}}\le2T_1.

ExperimentSequenceMeasures
Inversion recoveryπ\pi pulse, wait tt, measureT1T_1
Ramseyπ2\tfrac\pi2, wait tt, π2\tfrac\pi2, measureT2∗T_2^* (fringes at the detuning)
Hahn echoπ2\tfrac\pi2, wait t2\tfrac t2, π\pi, wait t2\tfrac t2, π2\tfrac\pi2, measureT2echoT_2^{\text{echo}}

Beyond single-qubit relaxation and dephasing, the standard theorist's model is the depolarizing channel, which shrinks the Bloch vector uniformly by 1−p1-p. Real devices add leakage out of the qubit subspace, crosstalk and correlated errors, and readout errors.

ρ  ↦  (1−p) ρ+p I2=(1−3p4)ρ+p4(XρX+YρY+ZρZ)\rho\;\mapsto\;(1-p)\,\rho+p\,\frac{\id}{2} = \Big(1-\frac{3p}{4}\Big)\rho+\frac p4\big(X\rho X+Y\rho Y+Z\rho Z\big)
The depolarizing channel, using I2=14(ρ+XρX+YρY+ZρZ)\tfrac{\id}{2} = \tfrac14(\rho+X\rho X+Y\rho Y+Z\rho Z) for any qubit state ρ\rho.

Key takeaways

  • Noise is described by channels ρ↦∑kKkρKk†\rho\mapsto\sum_kK_k\rho K_k^\dagger with ∑kKk†Kk=I\sum_kK_k^\dagger K_k = \id; they arise from unitary coupling to an environment that is then traced out.
  • Dephasing (ZZ errors) kills coherences and leaves populations alone; amplitude damping drives the qubit to ∣0⟩\ket0 with ρ11∝e−t/T1\rho_{11}\propto e^{-t/T_1} and ρ01∝e−t/(2T1)\rho_{01}\propto e^{-t/(2T_1)}.
  • 1/T2=1/(2T1)+1/Tφ1/T_2 = 1/(2T_1)+1/T_\varphi, so T2≤2T1T_2\le2T_1; positivity of ρ\rho forbids anything else.
  • The Bloch equations combine precession, relaxation and dephasing; Ramsey measures T2∗T_2^*, while echoes remove slow noise.
  • Coherent errors accumulate in amplitude and can be far worse than their average error rate suggests.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

What does the phase-flip channel with p=12p = \tfrac12 do to the Bloch vector (x,y,z)(x,y,z)?

  1. (x,y,z)↦(0,0,z)(x,y,z)\mapsto(0,0,z)
  2. (x,y,z)↦(0,0,0)(x,y,z)\mapsto(0,0,0)
  3. (x,y,z)↦(x,y,0)(x,y,z)\mapsto(x,y,0)
  4. (x,y,z)↦(−x,−y,z)(x,y,z)\mapsto(-x,-y,z)
Show the answer

A. (x,y,z)↦(0,0,z)(x,y,z)\mapsto(0,0,z)

xx and yy are multiplied by 1−2p=01-2p = 0 while zz is untouched: the state is projected onto the zz-axis, just as a ZZ measurement with a forgotten result would do.

Question 2

Under amplitude damping with parameter γ\gamma, by what factor is the coherence ρ01\rho_{01} multiplied?

  1. 1−γ1-\gamma
  2. 1−γ\sqrt{1-\gamma}
  3. 1−2γ1-2\gamma
  4. γ\gamma
Show the answer

B. 1−γ\sqrt{1-\gamma}

K0K_0 has 1−γ\sqrt{1-\gamma} in its ∣1⟩\ket1 entry, so ρ01↦1−γ ρ01\rho_{01}\mapsto\sqrt{1-\gamma}\,\rho_{01}, while ρ11↦(1−γ)ρ11\rho_{11}\mapsto(1-\gamma)\rho_{11}. With γ=1−e−t/T1\gamma = 1-e^{-t/T_1} the coherence decays as e−t/(2T1)e^{-t/(2T_1)}.

Question 3

A Ramsey experiment gives T2∗=5 μsT_2^* = 5\,\mu\text{s}, but a Hahn echo gives T2=60 μsT_2 = 60\,\mu\text{s}. What does this tell you?

  1. The qubit's T1T_1 must be about 5 μs5\,\mu\text{s}
  2. The echo pulse injects extra coherence into the qubit
  3. Most of the dephasing comes from slow, quasi-static frequency fluctuations
  4. The dephasing noise is white (frequency-independent)
Show the answer

C. Most of the dephasing comes from slow, quasi-static frequency fluctuations

An echo cancels phase from frequency offsets that stay constant during the sequence. A large gain from the echo means the dephasing is dominated by slow fluctuations; white noise would not be refocused at all.

Question 4

A gate over-rotates by ϵ=0.02\epsilon = 0.02 rad every time it is applied. What is the failure probability after 100 applications, compared with a stochastic error of the same per-gate probability?

  1. About the same, roughly 1% in both cases
  2. About 70% for the coherent error, versus about 1% for the stochastic one
  3. About 0.01% for the coherent error, because coherent errors cancel
  4. Zero for the coherent error, because a unitary error keeps the state pure
Show the answer

B. About 70% for the coherent error, versus about 1% for the stochastic one

Coherent: the rotations add to 100ϵ=2100\epsilon = 2 rad, so the failure probability is sin⁡2(1)≈0.71\sin^2(1)\approx0.71. Stochastic: p=sin⁡2(ϵ/2)≈10−4p = \sin^2(\epsilon/2)\approx10^{-4} per gate, so about 100p≈1%100p\approx1\%. Amplitudes add for coherent errors; probabilities add for stochastic ones.

Go deeper

Further reading

  • M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, Ch. 8, esp. §8.3 (Cambridge, 10th anniv. ed., 2010).
  • J. Preskill, Lecture Notes for Ph219/CS219: Quantum Information and Computation, Ch. 3 (Caltech).
  • P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gustavsson & W. D. Oliver, "A quantum engineer's guide to superconducting qubits", Applied Physics Reviews 6, 021318 (2019).
  • M. Schlosshauer, Decoherence and the Quantum-to-Classical Transition (Springer, 2007).