Environment-Induced Decoherence

Environment-induced decoherence is the suppression of interference caused by ordinary interactions with degrees of freedom outside the chosen system. No conscious observer is required. Photons, gas molecules, phonons, detector electronics, uncontrolled modes, and deliberately ignored ancillas can all act as environments when they carry away information about system alternatives.

The essential mechanism is environmental monitoring:

$$\text{different system alternatives} \quad\longrightarrow\quad \text{different environmental records}.$$

When those records are ignored, the reduced state of the system loses coherence between the alternatives that the environment has distinguished. This page explains how that statement becomes a rate estimate, a scattering model, or a master-equation term.

Helpful background. What Is Decoherence? derives the reduced-state overlap. Kraus operators and Lindblad evolution connect single interactions to rates.

Conditional environment states and coherence

First assume orthonormal alternatives $\{\lvert a\rangle\}$ and an initially independent, normalized pure environment $\lvert E_0\rangle$. A monitoring interaction has the schematic action

$$\lvert a\rangle\lvert E_0\rangle \longmapsto \lvert a\rangle\lvert E_a(t)\rangle.$$

For an initial superposition,

$$\left( \sum_a c_a\lvert a\rangle \right) \lvert E_0\rangle \longmapsto \sum_a c_a\lvert a\rangle\lvert E_a(t)\rangle.$$

The total state may remain pure. The reduced density matrix of the system has entries

$$\rho_{ab}(t) = c_ac_b^* \langle E_b(t)\vert E_a(t)\rangle.$$

Thus the environmental overlap

$$D_{ab}(t) = \langle E_b(t)\vert E_a(t)\rangle$$

is the decoherence factor. If $\lvert E_a(t)\rangle$ and $\lvert E_b(t)\rangle$ become nearly orthogonal, then the system alone no longer shows interference between $a$ and $b$.

In this pure-environment model, the environment acts like an unread measuring apparatus. Its distinguishability limits local interference. For a mixed initial environment $\rho_E$, a controlled unitary $U=\sum_a|a\rangle\langle a|\otimes U_a$ instead gives

$$D_{ab}=\operatorname{Tr}(U_a\rho_EU_b^\dagger).$$

The distinction matters. Take a qubit environment in $I_E/2$, $U_0=I_E$, and $U_1=Z_E$. Then $D_{01}=\operatorname{Tr}(Z_E)/2=0$, although the conditional environment states are both $I_E/2$. With system input $|+\rangle$, the joint output is the separable mixture

$$\frac12|+\rangle\langle+|\otimes|0\rangle\langle0| +\frac12|-\rangle\langle-|\otimes|1\rangle\langle1|.$$

The physical environment records a random phase flip, not which $Z$-basis alternative occurred. Decoherence need not imply system–environment entanglement or distinguishable basis records in a mixed environment. Purifying the environment restores an overlap description on the larger space.

The Interaction Selects the Alternatives

A common idealized interaction is

$$H_{\mathrm{int}} = \sum_a \lvert a\rangle\langle a\rvert \otimes B_a.$$

The system label $a$ changes the environmental Hamiltonian. If the system Hamiltonian preserves these sectors, or is negligible during the interaction, these are candidate stable alternatives. Actual record formation also depends on the environment state: different conditional Hamiltonians may act only by phases on that state.

More generally, if

$$H_I = \sum_\alpha S_\alpha\otimes B_\alpha,$$

then the operators $S_\alpha$ identify what the environment can learn about the system. If the relevant $S_\alpha$ commute and share an eigenbasis, decoherence tends to suppress coherence between their eigenspaces. If they do not commute, the preferred structure can be approximate, time dependent, or determined by a compromise between the system Hamiltonian and the coupling.

For the basis-dependence issue, see Coherence and Preferred Bases. For robustness criteria, see Pointer States.

When this monitoring produces effectively stable sectors or record states, the result is often described as Einselection.

Many Weak Records

Macroscopic decoherence is often fast because many weak environmental records multiply.

Suppose the environment is composed of fragments $E_1,\ldots,E_N$, and each fragment becomes slightly correlated with the system alternative. The total environment state is approximately

$$\lvert E_a\rangle = \bigotimes_{k=1}^N \lvert E_a^{(k)}\rangle.$$

The decoherence factor is then

$$D_{ab} = \prod_{k=1}^N \langle E_b^{(k)}\vert E_a^{(k)}\rangle.$$

If each fragment has the same magnitude of overlap,

$$\left| \langle E_b^{(k)}\vert E_a^{(k)}\rangle \right| = r, \qquad 0\le r\le1,$$

then

$$|D_{ab}| = r^N.$$

Even if $r$ is close to one, $r^N$ can be tiny for large $N$. For $r=1-\epsilon$ with $\epsilon\ll1$,

$$r^N \approx e^{-N\epsilon}.$$

The approximation uses $\ln(1-\epsilon)=-\epsilon+O(\epsilon^2)$; small relative error in the overlap requires $N\epsilon^2\ll1$. The exact $r^N$ remains valid even when this approximation fails. Conditional product states are essential: initially independent fragments can become correlated, invalidating a naive product rule.

When many fragments carry redundant information about the same pointer label, this many-record structure is also the starting point for Quantum Darwinism Preview.

Independent events and a decoherence rate

Suppose identical independent events arrive as a Poisson process of rate $R$, and each event multiplies the chosen coherence by a factor $\eta_{ab}$, averaged over any independent event parameters. Averaging over the event count gives

$$\begin{aligned} D_{ab}(t) &=\sum_{n=0}^{\infty}e^{-Rt}\frac{(Rt)^n}{n!}\eta_{ab}^n\\ &=e^{Rt(\eta_{ab}-1)}. \end{aligned}$$

Within this model the magnitude decay rate is exactly

$$\Gamma_{ab}=R(1-\operatorname{Re}\eta_{ab}),$$

while $R\operatorname{Im}\eta_{ab}$ produces a phase drift. Decay can occur even for $|\eta_{ab}|=1$: random numbers of individually coherent kicks scramble the phase. For $\eta_{ab}=-1$, fixed even and odd event counts preserve full coherence with opposite signs, but the unobserved Poisson count gives $D=e^{-2Rt}$.

This result requires independent events, a state-independent rate, and the same coherence basis between events. Correlated scatterings or an evolving basis need a more detailed model. Decoherence Timescales explains how this rate differs from relaxation and echo times.

Scattering-induced spatial decoherence

In a one-dimensional heavy-particle model, neglect displacement during a collision and let each event impart a momentum kick $Q$ drawn independently from a normalized distribution $P(Q)$. The corresponding unitary $e^{iQx/\hbar}$ multiplies $\rho(x,x')$ by $e^{iQ(x-x')/\hbar}$. Poisson-distributed collisions give

$$\partial_t\rho(x,x')=-F(x-x')\rho(x,x'), \qquad F(\Delta x)=R\left[1-\int dQ\,P(Q)e^{iQ\Delta x/\hbar}\right].$$

Here we have omitted free motion and other dissipative terms. The diagonal is unchanged because $F(0)=0$. In general $F$ is complex, with nonnegative real part; for a symmetric kick distribution it is real and nonnegative. Positivity of the whole map follows from averaging physical kicks, not merely from assuming $F\ge0$ pointwise.

For Gaussian kicks of mean zero and variance $\sigma_Q^2$, define the resolution length $\ell=\hbar/\sigma_Q$. Then

$$F(\Delta x)=R\left[1-e^{-\Delta x^2/(2\ell^2)}\right].$$

For $|\Delta x|\ll\ell$, this becomes $F\simeq\Lambda\Delta x^2$ with $\Lambda=R/(2\ell^2)$. For $|\Delta x|\gg\ell$, the rate saturates at $R$: one collision already resolves the two positions, so increasing separation adds little distinguishability.

As a numerical example, for $R=1000\,\mathrm{s}^{-1}$, separation $0.1\ell$ gives $F\approx4.99\,\mathrm{s}^{-1}$ and an $e^{-1}$ time of about $0.201$ seconds. At separation $5\ell$, the time is approximately $0.001$ seconds. Extending the small-separation quadratic approximation to $5\ell$ would overestimate the rate by a factor of about $12.5$.

Real scattering requires the momentum-transfer distribution, differential cross section, recoil, and geometry of the experiment. Saturation is typical for broad continuous kick distributions; monotonic increase with separation is not a universal theorem for every environment.

Markovian Dephasing Form

In a Markovian approximation, environment-induced decoherence often appears as exponential decay of off-diagonal elements:

$$\rho_{ab}(t) = e^{-\Gamma_{ab}t} e^{-i\Omega_{ab}t} \rho_{ab}(0), \qquad a\ne b.$$

Here $\Gamma_{ab}$ is a decoherence rate and $\Omega_{ab}$ includes coherent phase shifts. Populations may be constant in a pure-dephasing model or may also relax if the environment exchanges energy.

A useful Lindblad diagnostic comes from diagonal Lindblad operators. Suppose

$$L_\mu\lvert a\rangle = \ell_{\mu a}\lvert a\rangle.$$

For a generator $\sum_\mu\gamma_\mu\mathcal D[L_\mu]$ with $\gamma_\mu\ge0$, the full contribution is

$$\left.\dot\rho_{ab}\right|_{\mathrm{diss}} =\sum_\mu\gamma_\mu \left[\ell_{\mu a}\ell_{\mu b}^* -\frac{|\ell_{\mu a}|^2+|\ell_{\mu b}|^2}{2}\right]\rho_{ab}.$$

Its real part gives the magnitude-decay rate

$$\Gamma_{ab}=\frac12\sum_\mu\gamma_\mu |\ell_{\mu a}-\ell_{\mu b}|^2.$$

There is also a phase term $i\sum_\mu\gamma_\mu\operatorname{Im}(\ell_{\mu a}\ell_{\mu b}^*)\rho_{ab}$. It vanishes for Hermitian diagonal Lindblad operators, but cannot simply be omitted for complex eigenvalues. For example, $L=\operatorname{diag}(1,i)$ with rate $\gamma$ gives $\dot\rho_{01}=-\gamma(1+i)\rho_{01}$, rather than pure real decay.

If every relevant $\ell_{\mu a}=\ell_{\mu b}$, both terms vanish for that pair: this dissipator does not distinguish or rotate their coherence. Additional Hamiltonian motion or other noise may still change it.

For the channel version, see Dephasing Channel. For the generator version, see Pure Dephasing Master Equation.

Relation to Dissipation

Decoherence need not involve appreciable energy relaxation. Frequency noise commuting with a qubit Hamiltonian preserves its energy populations. Scattering can strongly resolve position while transferring negligible energy to a heavy object, but finite recoil still produces momentum diffusion; unchanged position-diagonal entries do not prove exact energy conservation.

Dissipation can also cause decoherence. Spontaneous emission, thermal relaxation, cavity loss, and phonon emission all leak information and energy into the environment. In those cases, the same interaction may both suppress coherences and change populations.

The practical distinction is:

Process	Environment learns	Energy exchanged?
pure dephasing	a basis label, or an uncontrolled random phase	no, when the coupling commutes with the Hamiltonian
elastic scattering	position or path information	often negligible
amplitude damping	whether an excitation was emitted	yes
thermalization	energy and transition history	yes
detector readout	measurement outcome information	depends on apparatus

See Dephasing vs Dissipation for the operational distinction.

Records, Accessibility, and Reversibility

Decoherence is not literal destruction of phase information in the closed system plus environment. In a unitary model, the phase information is stored in nonlocal correlations. Recoherence is possible in principle if one controls the relevant environmental degrees of freedom and reverses the entangling dynamics.

In practice, macroscopic recoherence is usually fantastically hard because:

• many environmental fragments are involved;
• the fragments disperse into uncontrolled modes;
• records are amplified into thermodynamic degrees of freedom;
• phases become sensitive to microscopic details;
• later interactions create further correlations.

This is a practical irreversibility, not a new fundamental projection postulate. The distinction matters when discussing Proper and Improper Mixtures and What Decoherence Does Not Solve.

Conditional Versus Unconditional Descriptions

If the environmental record is ignored, the system is described by an unconditional reduced state. That state can decohere.

If the record is actually measured, the appropriate state may be conditioned on the record. In quantum trajectory language, the conditioned state can remain pure or become purer even though the unconditional average decoheres.

There is no contradiction. The two descriptions answer different questions:

$$\text{unconditional state} = \text{average over records},$$

while

$$\text{conditional state} = \text{state assigned after one record is known}.$$

For continuous monitored dynamics, see Stochastic Master Equations.

Exercises

1. Product of weak records. Suppose $N$ independent environmental fragments each leave overlap magnitude $r$ between two alternatives. Show that the total overlap magnitude is $r^N$ and approximate it for $r=1-\epsilon$ with $\epsilon\ll1$.

Solution

For independent fragments,

$$D = \prod_{k=1}^N \langle E_b^{(k)}\vert E_a^{(k)}\rangle.$$

If each factor has magnitude $r$, then

$$|D|=r^N.$$

For $r=1-\epsilon$ with $\epsilon\ll1$,

$$r^N = \exp\left(N\ln(1-\epsilon)\right) \approx \exp(-N\epsilon).$$

The logarithmic error is $O(N\epsilon^2)$, so $N\epsilon^2\ll1$ controls the relative error. Many weak records can suppress coherence strongly while satisfying that condition.

2. No distinguishability, no decoherence. In the two-branch model, show that if $\lvert E_a(t)\rangle=\lvert E_b(t)\rangle$ then coherence between $a$ and $b$ is not suppressed by the environment.

Solution

The coherence factor is

$$D_{ab}(t) = \langle E_b(t)\vert E_a(t)\rangle.$$

If the two environment states are identical and normalized, then

$$D_{ab}(t)=1.$$

The off-diagonal element remains

$$\rho_{ab}(t)=\rho_{ab}(0)$$

up to any Hamiltonian phases. The environment has not learned which alternative occurred.

3. Spatial decoherence time. If $F(\Delta x)=\Lambda|\Delta x|^2$, estimate the time at which the coherence between two positions separated by $\Delta x$ has decayed by a factor $e^{-1}$.

Solution

The off-diagonal term decays as

$$\exp[-\Lambda t|\Delta x|^2].$$

Set the exponent equal to $-1$:

$$\Lambda t_{\mathrm{dec}}|\Delta x|^2=1.$$

Thus

$$t_{\mathrm{dec}} = \frac{1} {\Lambda|\Delta x|^2}.$$

This assumes $\Lambda>0$, nonzero fixed separation, and the small-separation regime. At $\Delta x=0$ there is no such decay; at large separation the quadratic estimate must be replaced by the full decoherence function.

4. Diagonal Lindblad operators. Suppose one Lindblad operator satisfies $L\lvert a\rangle=\ell_a\lvert a\rangle$. Show from the formula in the text that coherence between $a$ and $b$ is protected from this noise when $\ell_a=\ell_b$.

Solution

The dephasing contribution is proportional to

$$|\ell_a-\ell_b|^2.$$

If $\ell_a=\ell_b$, this factor vanishes, so this Lindblad operator does not distinguish the two alternatives. The corresponding coherence is not damped by that noise channel.

5. Conditional versus unconditional. Explain why an unconditional state can decohere even when a perfectly monitored environment would allow a conditioned pure-state description.

Solution

The unconditional state averages over all possible environmental records. Different records generally correspond to different conditional states, phases, jumps, or diffusive updates. Averaging them removes record-dependent information and can suppress off-diagonal terms.

If the record is known, the state assignment is conditioned on that record and need not be the same mixed average. The difference is not a contradiction; it is the distinction between ignoring and retaining the record.

References

• H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press (2002). Publisher.
• K. Hornberger, “Introduction to decoherence theory,” in A. Buchleitner, C. Viviescas, and M. Tiersch (eds.), Entanglement and Decoherence: Foundations and Modern Trends, Lecture Notes in Physics 768, Springer (2009). Author’s chapter.
• M. Schlosshauer, “Quantum decoherence,” Physics Reports 831, 1–57 (2019). Author’s text.
• W. H. Zurek, “Decoherence, einselection, and the quantum origins of the classical,” Reviews of Modern Physics 75, 715–775 (2003). DOI.