Decoherence

Every quantum superposition is a promise of interference, and decoherence is the process that quietly breaks that promise. A perfectly isolated system evolves unitarily and keeps every off-diagonal element of its density matrix — every coherence — forever. But no laboratory object is isolated: stray photons, phonons, fluctuating fields and the measuring apparatus itself all couple to it, and the joint system-plus-environment state, though it evolves by an impeccably unitary law, entangles the two. Look only at the system — trace the environment away — and the coherences fade. This is not a new axiom bolted onto quantum mechanics; it is what unitarity does the moment the environment starts keeping records. The mechanism is the copying of information. If a pointer state $\mid s_i\rangle$ stamps a distinguishable imprint $\mid E_i\rangle$ on the environment, then a superposition $\sum_i c_i\mid s_i\rangle$ grows into the entangled state $\sum_i c_i\mid s_i\rangle\mid E_i\rangle$ . Tracing out the environment leaves each coherence between branches $i$ and $j$ multiplied by a single number — the overlap $\langle E_i\mid E_j\rangle$ of the records. A large, restless environment carries which-path information away almost instantly, so those overlaps collapse toward zero, the off-diagonals vanish, and the system behaves as a plain classical mixture. Two languages formalise this: the exact CPTP map obtained by the partial trace, and, for a memoryless bath, the Lindblad master equation whose dissipator drives the decay. Crucially, decoherence does more than randomise — it selects. It singles out a preferred pointer basis, the states least disturbed by the coupling, and this einselection is why the macroscopic world looks classical while the microscopic law stays unitary and reversible at the global level.

Key takeaways

Decoherence is the dynamical loss of quantum coherence — the suppression of the off-diagonal interference terms of a density matrix — driven by entanglement between a system and an uncontrolled environment, not by any collapse postulate.
Tracing out the environment leaves each coherence multiplied by the decoherence factor $\langle E_i\mid E_j\rangle$ , the overlap of the environment records; as records become distinguishable this factor decays to zero.
It dynamically selects a preferred pointer basis — the states robust under the system-environment coupling — which is how effective classicality emerges without modifying quantum mechanics (einselection).
The exact description is a CPTP map with Kraus operators; for a memoryless bath the Lindblad master equation applies, and the canonical single-qubit model is the phase damping channel with survival factor $e^{-t/T_2}$ .
Decoherence is fast and basis-selective, and is distinct from energy relaxation $T_1$ : pure dephasing destroys coherence with no energy exchange, obeying $1/T_2=1/(2T_1)+1/T_\phi$ .
Key idea: coherence is lost not by mysterious collapse but by entanglement with the environment — a superposition $\sum_i c_i\mid s_i\rangle$ becomes $\sum_i c_i\mid s_i\rangle\mid E_i\rangle$ , and tracing out the environment leaves the coherences proportional to the overlaps $\langle E_i\mid E_j\rangle$ , which decay toward zero in the pointer basis.

The core insight

Coherence is lost not by mysterious collapse but by entanglement with the environment: a system superposition $\sum_i c_i\mid s_i\rangle$ becomes $\sum_i c_i\mid s_i\rangle\mid E_i\rangle$ , and tracing out the environment leaves the coherences proportional to the overlaps $\langle E_i\mid E_j\rangle$ , which decay toward zero in the pointer basis.

Where this lives today

Decoherence is the central adversary of quantum technology, and taming it is the working business of every qubit platform. In superconducting and trapped-ion processors, $T_2$ is the headline number a fabrication run is judged by: coherences carry the quantum information, so their decay is the dominant error, and Ramsey and spin-echo sequences measure $T_2$ directly while separating pure dephasing from relaxation. The response is layered. Quantum error correction encodes logical qubits redundantly so decoherence errors are caught and reversed before they accumulate, with the threshold theorem promising scalable computation once physical error rates fall below a bound set by the decoherence rates. Below that, hardware fights the bath directly: decoherence-free subspaces exploit collective-coupling symmetry so that certain encoded states share identical environment records ( $\langle E_i\mid E_j\rangle=1$ ) and do not decohere, while dynamical decoupling pulse sequences average the coupling toward zero. On the foundational frontier, matter-wave interferometry with ever-larger molecules and optomechanical resonators pushes the boundary of observable superposition to test whether decoherence alone accounts for classicality or whether some intrinsic collapse mechanism lurks — a program in which Zurek's quantum Darwinism is now being probed experimentally.

The physical picture

Picture a double-slit experiment in which each path also jostles a passing gas molecule. The instant a molecule could in principle report which slit was taken, the fringe pattern washes out — even if no one ever reads that molecule. What kills the interference is not observation but the mere availability of which-path information in the environment, and that availability is measured by how distinguishable the environment states $\mid E_i\rangle$ have become. The reduced density matrix tells the same story in bookkeeping: its diagonal entries are populations (probabilities), its off-diagonals are coherences (the capacity to interfere), and decoherence is the erosion of the off-diagonals while the diagonals stay pinned — the matrix bleeds toward diagonal form in the pointer basis. The first figure follows one superposition through the entangling unitary and shows exactly where the coherence goes.

Figure

A system superposition becomes entangled with the environment under the joint unitary U, so each pointer state imprints its own record; tracing out the environment leaves the off-diagonal coherence multiplied by the overlap of those records, the decoherence factor, which falls to zero as the records grow distinguishable and the reduced state looks classical.

The pointer basis in that figure is the environment's choice, not ours. It is the set of system states that get copied cleanly into the environment without being scrambled further — usually the eigenstates of the operator $\hat A$ through which the system couples, $H_{\rm int}=\hat A\otimes\hat B$ . Superpositions of pointer states are fragile; the pointer states themselves are robust, which is why we see needles at definite dial positions and never their superpositions. Decoherence is also brutally fast for anything macroscopic, and the rate climbs steeply with the "distance" between the superposed alternatives. The second figure makes the timescale concrete for a single qubit undergoing pure dephasing — the exponential collapse of one coherence, with real numbers a hardware team would recognise.

Figure

Coherence decay for pure dephasing with T₂ = 20 μs: the off-diagonal element follows ρ₀₁(t) = ρ₀₁(0)·e^(−t/T₂), so after one T₂ it has fallen to 1/e = 0.368 of its start and after t = 10 μs to e^(−0.5) = 0.607, while the populations never move at all.

The core equations

Decoherence starts from a single, almost trivial move: let the interaction copy the system's pointer label into the environment. A superposition is dragged into an entangled state in which each branch wears its own environment record.

\Big(\textstyle\sum_i c_i\mid s_i\rangle\Big)\otimes\mid E_0\rangle \ \xrightarrow{\;U\;}\ \sum_i c_i\,\mid s_i\rangle\otimes\mid E_i\rangle \qquad (1)

where: $\mid s_i\rangle$ — system pointer states (the basis robust to the interaction); $c_i$ — amplitudes of the initial superposition; $\mid E_0\rangle$ — initial environment state; $\mid E_i\rangle$ — environment record correlated with $\mid s_i\rangle$ ; $U$ — joint unitary generated by the interaction Hamiltonian. Reading: the global state stays pure and unitary; nothing has been lost yet — the information has merely been shared. Now look at the system alone by tracing out the environment. Every off-diagonal element survives only in proportion to how indistinguishable the two records are — the overlap $\langle E_j\mid E_i\rangle$ , the decoherence factor.

\rho_S = \mathrm{Tr}_E\big(\mid\Psi\rangle\langle\Psi\mid\big)=\sum_{i,j} c_i c_j^{*}\,\langle E_j\mid E_i\rangle\;\mid s_i\rangle\langle s_j\mid \qquad (2)

where: $\rho_S$ — reduced state of the system; $\mathrm{Tr}_E$ — partial trace over the environment; $\langle E_j\mid E_i\rangle$ — record overlap. Reading: diagonal terms keep $\langle E_i\mid E_i\rangle=1$ (populations untouched), while off-diagonal terms are multiplied by an overlap that heads to zero — this one equation is the whole phenomenon. The exact map $\rho\mapsto\rho_S$ is a CPTP channel, and every such channel has an operator-sum (Kraus) form. The Kraus operators are literally slices of the system-environment unitary in an environment basis.

\rho_S=\sum_k K_k\,\rho\,K_k^{\dagger},\qquad \sum_k K_k^{\dagger}K_k=\mathbb{1},\qquad K_k=\langle E_k\mid U\mid E_0\rangle \qquad (3)

where: $K_k$ — Kraus operators acting on the system; $\mathbb{1}$ — identity on the system Hilbert space. Reading: the completeness relation $\sum_k K_k^\dagger K_k=\mathbb 1$ is trace preservation — it is just $\langle E_0\mid U^\dagger U\mid E_0\rangle=\mathbb 1$ , unitarity read in the environment basis. The canonical single-qubit model of pure decoherence is the phase damping channel: it shrinks the coherences by a survival factor while leaving both populations exactly where they were.

\rho=\begin{pmatrix}\rho_{00}&\rho_{01}\\ \rho_{10}&\rho_{11}\end{pmatrix}\ \longmapsto\ \begin{pmatrix}\rho_{00}& (1-\lambda)\,\rho_{01}\\ (1-\lambda)\,\rho_{10}&\rho_{11}\end{pmatrix} \qquad (4)

where: $\rho_{00},\rho_{11}$ — populations, unchanged (no energy exchange); $\rho_{01},\rho_{10}$ — coherences, scaled by $(1-\lambda)$ ; $\lambda$ — decoherence probability, $0\le\lambda\le1$ . Reading: this is decoherence in its purest form — information leaks out about which pointer state, but no energy moves, so it is distinct from relaxation. The channel comes from three Kraus operators: a "nothing recorded" branch and two branches that project onto the pointer states, copying which-state information into the bath.

K_0=\sqrt{1-\lambda}\,\mathbb{1},\qquad K_1=\sqrt{\lambda}\begin{pmatrix}1&0\\0&0\end{pmatrix},\qquad K_2=\sqrt{\lambda}\begin{pmatrix}0&0\\0&1\end{pmatrix} \qquad (5)

where: $K_1,K_2$ — projectors onto $\mid 0\rangle,\mid 1\rangle$ ; $\lambda$ — per-step decoherence probability, equivalently $1-\lambda=e^{-t/T_2}$ in continuous time. Reading: check completeness — $K_0^\dagger K_0+K_1^\dagger K_1+K_2^\dagger K_2=(1-\lambda)\mathbb 1+\lambda\,\mathrm{diag}(1,0)+\lambda\,\mathrm{diag}(0,1)=\mathbb 1$ . For a memoryless (Markovian) bath the differential law is the Lindblad/GKSL master equation; pure dephasing is the single choice $L=\sigma_z$ .

\frac{\mathrm{d}\rho}{\mathrm{d}t}=-\frac{i}{\hbar}[H,\rho]+\gamma\Big(L\rho L^{\dagger}-\tfrac12\{L^{\dagger}L,\rho\}\Big) \qquad (6)

where: $H$ — system Hamiltonian (coherent part); $L$ — Lindblad (jump) operator, $L=\sigma_z$ for pure dephasing; $\gamma$ — decoherence rate, $\gamma\ge0$ ; $\{\,,\,\}$ — anticommutator. Reading: the commutator rotates the state reversibly; the dissipator is the irreversible part that erases coherence, and its positivity $\gamma\ge0$ is what guarantees the map stays physical. Solving that generator gives the observable law: an exponential decay of coherence set by the dephasing time $T_2$ .

\rho_{01}(t)=\rho_{01}(0)\,\exp\!\big[-\Gamma(t)\big],\qquad \Gamma(t)\xrightarrow{\text{Markov}}\frac{t}{T_2} \qquad (7)

where: $\Gamma(t)$ — decoherence exponent fixed by the bath correlation function; $T_2$ — dephasing time, the $1/e$ decay time of coherence. Reading: in the Markovian limit $\Gamma(t)=t/T_2$ is linear (exponential decay); a structured bath makes $\Gamma(t)\sim(t/\tau_d)^2$ at short times, a qualitatively different, Gaussian collapse. Finally, the reason we never see macroscopic superpositions: the record overlap falls off Gaussianly with the separation of the alternatives, so the decoherence time is crushed below the relaxation time by the square of a tiny wavelength ratio.

\langle E_i\mid E_j\rangle \sim e^{-(x_i-x_j)^2/\ell^2}\quad\Rightarrow\quad \tau_{\rm dec}\sim \tau_{\rm rel}\Big(\frac{\lambda_{dB}}{\Delta x}\Big)^2 \qquad (8)

where: $x_i-x_j=\Delta x$ — separation of the superposed positions; $\ell,\lambda_{dB}$ — environment coherence length and thermal de Broglie wavelength; $\tau_{\rm dec},\tau_{\rm rel}$ — decoherence and relaxation timescales. Reading: because $\lambda_{dB}\ll\Delta x$ for anything macroscopic, $\tau_{\rm dec}\ll\tau_{\rm rel}$ by an astronomical factor — decoherence outruns dissipation by many orders of magnitude.

Derivation AFrom entanglement to a diagonal density matrix

The goal is to show, with no collapse postulate anywhere, that a global unitary interaction which merely copies the pointer label into the environment drives the system's reduced state to diagonal form in that pointer basis. We start from a controlled-copy interaction, evolve a generic superposition by linearity into an entangled state that is still perfectly pure globally, take the partial trace, and watch the off-diagonal terms get multiplied by record overlaps that vanish once the environment is large. The result is einselection: the diagonal Born-rule weights survive while the coherences die.

Step 1. Let the interaction be diagonal in the pointer basis, $H_{\rm int}=\hat A\otimes\hat B$ with $\hat A\mid s_i\rangle=a_i\mid s_i\rangle$ , so the pointer states are eigenstates of the coupled operator and are never flipped. Why: a state that commutes with the interaction is not scrambled by it — that stability is precisely what makes it an einselected pointer state rather than a fragile superposition.

Step 2. Because $\hat A$ is diagonal, each pointer state steers the environment into its own branch, $U\,\mid s_i\rangle\mid E_0\rangle=\mid s_i\rangle\otimes\mid E_i(t)\rangle$ with $\mid E_i(t)\rangle=e^{-ia_i\hat B t/\hbar}\mid E_0\rangle$ . Why: the system state acts as a classical control that conditionally evolves the environment, so the two become correlated without the system itself moving.

Step 3. Apply this to a superposition by linearity: $\mid\Psi(t)\rangle=\sum_i c_i\mid s_i\rangle\mid E_i(t)\rangle$ , a pure, fully entangled state in which no information has been lost globally. Why: unitarity is exact at the level of the whole world; the "loss" we are about to see is entirely a consequence of choosing to look at only one part.

Step 4. Trace out the environment. Expanding $\mid\Psi\rangle\langle\Psi\mid$ and using $\mathrm{Tr}_E(\mid E_i\rangle\langle E_j\mid)=\langle E_j\mid E_i\rangle$ gives $\rho_S(t)=\sum_{i,j}c_ic_j^{*}\,\langle E_j(t)\mid E_i(t)\rangle\,\mid s_i\rangle\langle s_j\mid$ . Why: the partial trace sends each environment operator to its overlap, so the decoherence factor lands directly on top of every coherence.

Step 5. For a large bath the branches orthogonalise rapidly, $\langle E_j(t)\mid E_i(t)\rangle\to\delta_{ij}$ , leaving $\rho_S\to\sum_i\mid c_i\mid^2\mid s_i\rangle\langle s_i\mid$ — a classical mixture in the pointer basis with the Born weights intact. Why: countless environment particles each carry off a little which-path information, and their combined record makes the branches distinguishable, killing the off-diagonals while the diagonals never budge. *Note: the diagonal limit is a proper mixture with $\mathrm{Tr}\,\rho_S=\sum_i\mid c_i\mid^2=1$ and $\mathrm{Tr}\,\rho_S^2<1$ for any genuine superposition — local purity is lost while global purity is preserved.*

Go deeper · Derivation B · 1 minCoherence decay from the dephasing Lindblad equation

Derivation BCoherence decay from the dephasing Lindblad equation

The goal is to solve the GKSL master equation for pure dephasing and extract both the exponential decay law and the identification of the rate, showing explicitly that populations are frozen while coherences fall. We take the simplest non-trivial dephasing generator — no coherent dynamics, a single Hermitian Lindblad operator $L=\sigma_z$ — write the dissipator in matrix elements, and integrate the resulting first-order equations. The payoff is the clean relation $1/T_2=2\gamma$ that ties the abstract rate to the measured coherence time and matches the phase damping channel.

Step 1. With $H=0$ and $L=\sigma_z$ (so $L^\dagger L=\mathbb 1$ ), the anticommutator term collapses to $-\tfrac12\{\mathbb 1,\rho\}=-\rho$ , leaving $\dot\rho=\gamma(\sigma_z\rho\,\sigma_z-\rho)$ . Why: because $\sigma_z$ is unitary as well as Hermitian, the Lindblad form simplifies to a single sandwich minus the state — the hallmark of a random-phase-kick process.

Step 2. Act on matrix elements using $\sigma_z\mid 0\rangle=+\mid 0\rangle$ and $\sigma_z\mid 1\rangle=-\mid 1\rangle$ , so $\langle 0\mid\sigma_z\rho\,\sigma_z\mid 1\rangle=-\rho_{01}$ while $\langle 0\mid\sigma_z\rho\,\sigma_z\mid 0\rangle=\rho_{00}$ . Why: $\sigma_z$ flips the sign of the excited amplitude, and a coherence carries one factor of each level, so it picks up the single minus sign that will drive its decay.

Step 3. The populations obey $\dot\rho_{00}=\gamma(\rho_{00}-\rho_{00})=0$ , so they are constant in time. Why: dephasing exchanges no energy — the diagonal is untouched, which is exactly what distinguishes it from $T_1$ relaxation.

Step 4. The coherence obeys $\dot\rho_{01}=\gamma(-\rho_{01}-\rho_{01})=-2\gamma\,\rho_{01}$ , a first-order linear decay. Why: the sign flip from Step 2 combines with the $-\rho$ term to give a doubled negative rate, and only the off-diagonal feels it.

Step 5. Integrate to get $\rho_{01}(t)=\rho_{01}(0)\,e^{-2\gamma t}$ , so the $1/e$ time is $T_2=1/(2\gamma)$ , matching the phase damping channel through $1-\lambda=e^{-2\gamma t}=e^{-t/T_2}$ . Why: reading off the exponent fixes the dictionary between the continuous rate $\gamma$ , the discrete channel strength $\lambda$ , and the experimentally quoted $T_2$ . *Note: trace and Hermiticity are preserved at all times, as required for a physical density matrix.*

Go deeper · Derivation C · 1 minPhase damping Kraus operators from a scattering environment

Derivation CPhase damping Kraus operators from a scattering environment

The goal is to connect the two languages of the article by building the phase damping channel of equations (4) and (5) directly from a concrete system-environment scattering model, rather than postulating the Kraus operators. We let a single environment mode scatter off the qubit so that it partly stays in place and partly gets kicked into a state that depends on which pointer state the qubit occupies, then slice the resulting unitary with the rule $K_k=\langle E_k\mid U\mid E_0\rangle$ . The which-state records that emerge are exactly the projectors that suppress coherence, closing the loop between "records in the bath" and "a CPTP channel on the qubit."

Step 1. Model the scattering with a three-state environment: $\mid 0\rangle\mid E_0\rangle\mapsto\mid 0\rangle(\sqrt{1-\lambda}\mid E_0\rangle+\sqrt{\lambda}\mid E_1\rangle)$ and $\mid 1\rangle\mid E_0\rangle\mapsto\mid 1\rangle(\sqrt{1-\lambda}\mid E_0\rangle+\sqrt{\lambda}\mid E_2\rangle)$ . Why: with probability $\lambda$ the environment is kicked into a record ( $\mid E_1\rangle$ or $\mid E_2\rangle$ ) that reveals which pointer state the qubit is in, and this unitary is norm-preserving on each branch since $(1-\lambda)+\lambda=1$ .

Step 2. Extract the Kraus operators by projecting onto the environment basis: $K_0=\langle E_0\mid U\mid E_0\rangle=\sqrt{1-\lambda}\,\mathbb 1$ , $K_1=\langle E_1\mid U\mid E_0\rangle=\sqrt{\lambda}\,\mid 0\rangle\langle 0\mid$ , and $K_2=\langle E_2\mid U\mid E_0\rangle=\sqrt{\lambda}\,\mid 1\rangle\langle 1\mid$ . Why: each environment outcome slices out one operator on the qubit, and the two "kicked" outcomes are pure projectors because the record uniquely tags the pointer state.

Step 3. Apply the channel to $\rho$ : the $K_0$ term returns $(1-\lambda)\rho$ , while $K_1\rho K_1^\dagger+K_2\rho K_2^\dagger=\lambda\,\mathrm{diag}(\rho_{00},\rho_{11})$ , so populations survive intact and the coherence is scaled by $(1-\lambda)$ . Why: the projector terms rebuild the diagonal but erase the off-diagonal, reproducing equation (4) exactly — the record overlap $\langle E_1\mid E_2\rangle=0$ is what makes the suppression complete. *Note: completeness holds explicitly — $\sum_k K_k^\dagger K_k=(1-\lambda)\mathbb 1+\lambda\,\mathrm{diag}(1,0)+\lambda\,\mathrm{diag}(0,1)=\mathbb 1$ — confirming the channel is trace preserving, and the coherence factor $(1-\lambda)$ agrees with the Lindblad result $e^{-2\gamma t}$ .*

Limiting cases

Case	Result	Meaning
Fully distinguishable environment	$\langle E_i\mid E_j\rangle\to0\ (i\neq j)$	Complete decoherence: the reduced state is a perfectly classical mixture, diagonal in the pointer basis, with all interference lost.
Indistinguishable environment	$\langle E_i\mid E_j\rangle\to1$	No decoherence: the environment records nothing, coherences are fully preserved, and the system evolves effectively unitarily.
Markovian exponential decay	$\rho_{ij}(t)=\rho_{ij}(0)\,e^{-t/T_2}$	Memoryless bath: coherences decay exponentially with the dephasing time $T_2$ — the canonical regime assumed in most quantum-computing error models.
Non-Markovian / short-time	$\rho_{ij}(t)\approx\rho_{ij}(0)\,e^{-(t/\tau_d)^2}$	Structured or slow baths give Gaussian (Zeno-like) decay at short times and can show coherence revivals — relevant to engineered environments and decoherence-free subspaces.

Common trapDecoherence is not collapse, and T₂ is not T₁

The most common misconception is that decoherence solves the measurement problem by selecting a single outcome. It does not. The global system-plus-environment state stays pure and unitary at every instant; the apparent mixture is only the local description after the partial trace, and the reduced $\rho_S$ still contains all outcomes, now merely without their mutual coherence. Decoherence explains why we see a classical mixture rather than a superposition, but it does not pick which member of the mixture is realised — that is left to whatever interpretation of quantum mechanics you favour. A second trap is conflating decoherence ( $T_2$ , loss of off-diagonal coherence) with relaxation ( $T_1$ , energy loss that changes populations). Pure dephasing destroys coherence with no energy exchange, and the pointer basis is set by the interaction Hamiltonian $H_{\rm int}=\hat A\otimes\hat B$ — not by the system's energy eigenbasis. In general $1/T_2=1/(2T_1)+1/T_\phi$ , so coherence always decays at least as fast as relaxation permits, and pure dephasing $T_\phi$ often dominates.

Worked examples

1 · Dephasing of a superconducting qubit, fully in numbers (confidence builder)

A transmon is prepared in $\mid +\rangle=\tfrac{1}{\sqrt2}(\mid 0\rangle+\mid 1\rangle)$ , so $\rho_{01}(0)=0.5$ . Its pure-dephasing time is $T_2=20\ \mu\mathrm{s}$ . After a free-evolution time $t=10\ \mu\mathrm{s}$ , find the surviving coherence, the equivalent phase damping strength $\lambda$ , the Lindblad rate $\gamma$ , and the purity of the reduced state.

Step 1. The survival factor is $e^{-t/T_2}=e^{-10/20}=e^{-0.5}=\mathbf{0.607}$ . The coherence falls to $\rho_{01}(t)=0.5\times0.607=\mathbf{0.303}$ , while the populations $\rho_{00}=\rho_{11}=0.5$ are frozen — the Bloch vector, initially pointing along $+x$ with length 1, shrinks to length 0.607 and stays in the equatorial plane.

Step 2. The phase damping channel matches this through $1-\lambda=e^{-t/T_2}=0.607$ , so $\lambda=1-0.607=\mathbf{0.393}$ — a single application of the channel with $\lambda=0.393$ reproduces $10\ \mu\mathrm{s}$ of dephasing. Consistency check: the map (4) gives $\rho_{01}\mapsto(1-\lambda)\rho_{01}=0.607\times0.5=0.303$ . ✓

Step 3. The Lindblad rate follows from $1/T_2=2\gamma$ , so $\gamma=1/(2T_2)=1/(40\ \mu\mathrm{s})=\mathbf{2.5\times10^{4}\ s^{-1}}$ . The coherence law $e^{-2\gamma t}=e^{-2\times2.5\times10^4\times10^{-5}}=e^{-0.5}=0.607$ agrees, tying the three descriptions together.

Step 4. Purity: $\mathrm{Tr}\,\rho^2=\rho_{00}^2+\rho_{11}^2+2\mid\rho_{01}\mid^2=0.25+0.25+2(0.303)^2=\mathbf{0.684}$ . The state began pure ( $\mathrm{Tr}\,\rho^2=1$ ) and has drifted toward the maximally mixed value $0.5$ — local purity is lost even though the global system-plus-bath state remains pure, the signature of decoherence.

2 · Why a dust grain never superposes (order of magnitude)

Estimate the decoherence-to-relaxation time ratio for a $1\ \mu\mathrm{m}$ dust grain ( $m\approx10^{-15}\ \mathrm{kg}$ ) placed in a superposition of two positions separated by $\Delta x=1\ \mu\mathrm{m}$ at room temperature $T=300\ \mathrm{K}$ , using equation (8).

Step 1. The thermal de Broglie wavelength is $\lambda_{dB}=h/\sqrt{2\pi m k_B T}$ . With $m=10^{-15}\ \mathrm{kg}$ : $2\pi m k_B T=2\pi(10^{-15})(1.38\times10^{-23})(300)=2.60\times10^{-35}$ , so $\sqrt{\cdot}=5.10\times10^{-18}\ \mathrm{kg\,m/s}$ and $\lambda_{dB}=6.63\times10^{-34}/5.10\times10^{-18}=\mathbf{1.3\times10^{-16}\ m}$ .

Step 2. The ratio in equation (8) is $(\lambda_{dB}/\Delta x)^2=(1.3\times10^{-16}/10^{-6})^2=(1.3\times10^{-10})^2=\mathbf{1.7\times10^{-20}}$ .

Step 3. Even with a generously long relaxation time $\tau_{\rm rel}\approx1\ \mathrm{s}$ , the decoherence time is $\tau_{\rm dec}\approx\tau_{\rm rel}\times1.7\times10^{-20}\approx\mathbf{1.7\times10^{-20}\ s}$ — twenty orders of magnitude faster than any dissipation. The grain's position superposition is destroyed essentially the instant it forms, which is exactly why macroscopic Schrödinger cats are never observed even though the underlying dynamics permit them.

3 · Combining T₁ and pure dephasing into T₂

A qubit has energy-relaxation time $T_1=30\ \mu\mathrm{s}$ and pure-dephasing time $T_\phi=60\ \mu\mathrm{s}$ . Find its coherence time $T_2$ and compare with the ceiling $2T_1$ .

Step 1. Use the standard combination $1/T_2=1/(2T_1)+1/T_\phi$ . The relaxation contribution is $1/(2T_1)=1/(60\ \mu\mathrm{s})$ and the pure-dephasing contribution is $1/T_\phi=1/(60\ \mu\mathrm{s})$ .

Step 2. Add them: $1/T_2=1/60+1/60=2/60=1/30\ \mu\mathrm{s}^{-1}$ , so $T_2=\mathbf{30\ \mu s}$ .

Step 3. The ceiling set by relaxation alone is $2T_1=60\ \mu\mathrm{s}$ , and indeed $T_2=30\ \mu\mathrm{s}<60\ \mu\mathrm{s}$ . Pure dephasing has halved the coherence time relative to the $T_1$ limit — if $T_\phi$ could be pushed to infinity (no pure dephasing), $T_2$ would rise only to its hard ceiling of $2T_1=60\ \mu\mathrm{s}$ , never beyond.

Physics of the portrait

Decoherence is where the measurement problem stopped being purely philosophical and became a calculable, unitary process. The concept was born from H. Dieter Zeh's 1970 recognition that macroscopic systems can never be isolated, and it was sharpened into a predictive theory by the physicist who named its central mechanism.

Origins · 1 minWojciech Zurek and how this was discovered

Wojciech Zurek

b. 1951 · einselection and the quantum origin of the classical

Wojciech Hubert Zurek is a Polish-American theoretical physicist at Los Alamos National Laboratory whose work turned decoherence from a foundational puzzle into a quantitative theory of the quantum-to-classical transition. Building on H. Dieter Zeh's 1970 insight that a macroscopic object is inescapably coupled to its surroundings, Zurek introduced in 1981 the idea of the pointer basis — asking, in the title of his Physical Review D paper, "into what mixture does the wave packet collapse?" — and in 1982 formalised einselection, environment-induced superselection, the dynamical rule by which the interaction Hamiltonian singles out the robust, classical states an apparatus can display. His 2003 Reviews of Modern Physics article became the standard synthesis of the field, and his later programme of quantum Darwinism explains objectivity as the redundant proliferation of pointer-state records across many fragments of the environment. Zurek's reach extends well beyond decoherence: with William Wootters he proved the no-cloning theorem in 1982, and the Kibble-Zurek mechanism bearing his name predicts topological defect formation in phase transitions across cosmology and condensed matter. The throughline is a single question — why does a unitary, superposition-friendly world look classical? — answered not by amending quantum mechanics but by taking its entanglement with the environment seriously.

History

1935

— Schrödinger names entanglement and poses the cat paradox, framing the puzzle decoherence later dissolves

1970

— Zeh publishes the first paper on environment-induced decoherence of macroscopic systems

1981

— Zurek introduces the pointer basis and asks into what mixture the wave packet collapses

1985

— Joos and Zeh compute decoherence rates for scattering environments, quantifying the effect

2003

— Zurek's Reviews of Modern Physics article synthesises decoherence and einselection

Going further

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, section 8.3 (quantum operations) and 8.3.5 (phase damping) · Preskill, Lecture Notes for Physics 219, Ch. 3 (measurement and evolution) · Watrous, The Theory of Quantum Information, Ch. 2 (channels) · Zurek, Reviews of Modern Physics 75, 715 (2003) · Breuer and Petruccione, The Theory of Open Quantum Systems, Chs. 3-4 · Schlosshauer, Decoherence and the Quantum-to-Classical Transition (2007). Key results to keep in mind: the partial-trace identity $\mathrm{Tr}_E(\mid E_i\rangle\langle E_j\mid)=\langle E_j\mid E_i\rangle$ places the decoherence factor directly on each coherence; the phase damping Kraus set $K_0=\sqrt{1-\lambda}\,\mathbb 1$ , $K_1=\sqrt{\lambda}\,\mathrm{diag}(1,0)$ , $K_2=\sqrt{\lambda}\,\mathrm{diag}(0,1)$ satisfies $\sum_k K_k^\dagger K_k=\mathbb 1$ and scales coherences by $(1-\lambda)$ ; and the Lindblad solution with $L=\sigma_z$ gives $\dot\rho_{01}=-2\gamma\rho_{01}$ and $T_2=1/(2\gamma)$ . The anchor numbers are: dephasing at $T_2=20\ \mu\mathrm{s}$ , $t=10\ \mu\mathrm{s}$ gives survival $e^{-0.5}=0.607$ , $\lambda=0.393$ , $\gamma=2.5\times10^4\ \mathrm{s}^{-1}$ , purity $0.684$ ; the dust-grain ratio $(\lambda_{dB}/\Delta x)^2=1.7\times10^{-20}$ with $\lambda_{dB}=1.3\times10^{-16}\ \mathrm{m}$ ; and the combination $1/T_2=1/(2T_1)+1/T_\phi$ gives $T_2=30\ \mu\mathrm{s}$ for $T_1=30\ \mu\mathrm{s}$ , $T_\phi=60\ \mu\mathrm{s}$ .

Prerequisites. Open Quantum Systems and Reduced Dynamics · Quantum Operations and CPTP Maps · Kraus Representation of Quantum Channels. Leads to. Phase Damping Channel · Relaxation and Dephasing: T₁ and T₂ Times · Lindblad / GKSL Master Equation · Markovian versus Non-Markovian Dynamics.

Ask the tutor: Ask me to trace out an environment of your choosing and watch the coherences decay, or to solve the Lindblad equation for any dephasing model and read off its $T_2$ step by step.

If the global state stays pure, in what sense is coherence actually lost?

It is lost locally and irretrievably for anyone who can only access the system. The full system-plus-environment state remains pure and its coherences are all still there, but they now live in correlations spread across countless environment degrees of freedom you cannot touch. Any measurement confined to the system is governed by the reduced density matrix $\rho_S$ , and in that object the off-diagonals have vanished — no local experiment can reveal the interference. In principle the coherence could be recovered by a joint operation on the system and the entire environment (a spin-echo does exactly this for a small, controlled bath), but for a realistic macroscopic environment that is hopelessly impractical, so the loss is effectively permanent.

What is the difference between decoherence (T₂) and relaxation (T₁)?

Relaxation, characterised by $T_1$ , is energy exchange with the environment that changes the populations — the diagonal elements of $\rho$ — as when an excited qubit spontaneously emits and decays to its ground state (amplitude damping). Decoherence in the pure sense, characterised by the pure-dephasing time $T_\phi$ and appearing in $T_2$ , destroys the coherences — the off-diagonal elements — with no energy exchange at all. Because relaxation also destroys coherence as a side effect, the two combine as $1/T_2=1/(2T_1)+1/T_\phi$ , which means $T_2$ can never exceed $2T_1$ . In many devices pure dephasing dominates, so $T_2\ll2T_1$ , and the two timescales are measured by different experiments — inversion recovery for $T_1$ , Ramsey or spin-echo for $T_2$ .

What exactly is the pointer basis, and who chooses it?

The pointer basis is the special set of system states that the environment leaves essentially undisturbed while it copies information about them. The environment chooses it, not the experimenter: it is fixed by the interaction Hamiltonian $H_{\rm int}=\hat A\otimes\hat B$ , and the pointer states are the eigenstates of the coupled operator $\hat A$ , because states that commute with the interaction are not scrambled by it. Superpositions of pointer states are fragile and decohere quickly; the pointer states themselves are robust and get faithfully recorded in the environment. This dynamical selection is what Zurek called einselection, and it is why a measuring apparatus shows definite, classical dial readings rather than superpositions of them — the readings are its pointer states.

Does decoherence solve the measurement problem?

It solves an important part and leaves another part open. Decoherence explains, from unitary dynamics alone, why we observe a classical mixture of definite outcomes in a well-defined pointer basis rather than a coherent superposition — it accounts for the appearance of collapse and for why interference between macroscopically distinct alternatives is unobservable. What it does not do is single out which one outcome is actually realised on a given run; the reduced density matrix still lists all outcomes with their Born-rule weights. That final selection is exactly where interpretations of quantum mechanics differ, whether many-worlds (all branches real), Bohmian (a hidden variable picks one), or objective-collapse (an extra dynamical law). Decoherence is interpretation-neutral machinery that every serious interpretation must incorporate.

Why does decoherence happen so much faster than relaxation for large objects?

Because the two processes scale with the separation of the alternatives in completely different ways. Energy relaxation is set by system-bath coupling strengths that change only slowly with size, but the decoherence rate is governed by the distinguishability of the environment records, which grows as the square of the separation between the superposed states: $\langle E_i\mid E_j\rangle\sim e^{-(\Delta x)^2/\ell^2}$ . For a macroscopic $\Delta x$ , an enormous number of environment particles each acquire a little which-path information, and their combined record makes the branches orthogonal almost instantly. The worked dust-grain estimate makes this vivid — the decoherence time comes out roughly $10^{-20}$ times the relaxation time, so the object is classical long before it can lose any appreciable energy.

Can decoherence be avoided or slowed in a quantum computer?

Not eliminated, but fought on several fronts. First, materials and design: cleaner interfaces, better shielding and lower temperatures push $T_1$ and $T_2$ up directly. Second, symmetry: if information is encoded in a decoherence-free subspace whose states share identical environment records, the coupling cannot distinguish them and they do not decohere — effective for collective noise. Third, control: dynamical decoupling applies rapid pulse sequences that average the system-environment coupling toward zero, much as a spin echo refocuses dephasing, extending usable coherence by orders of magnitude against slow noise. Fourth, redundancy: quantum error correction accepts that residual decoherence will cause errors and encodes logical qubits so those errors are detected and reversed before they accumulate — the route to arbitrarily long computations once physical error rates drop below the fault-tolerance threshold.