What Is Decoherence?

Decoherence is the suppression of observable interference in a reduced system state, relative to specified alternatives. A common mechanism is entanglement with an uncontrolled environment: the interference becomes encoded in joint correlations rather than accessible to measurements on the system alone. Averaging over uncontrolled phases can also suppress interference without requiring entanglement with the physical noise source.

The calculation below starts from an initially pure, uncorrelated environment. It shows exactly how distinguishable environmental records reduce interference, and why this reduced-state effect is not by itself a selection of one outcome.

Helpful background. Reduced Density Operators explains the partial trace used to predict measurements on a subsystem.

Environmental overlap and reduced coherence

Start with orthogonal system alternatives, $|c_0|^2+|c_1|^2=1$, and a normalized environment vector:

$$\left( c_0|0\rangle+c_1|1\rangle \right) |E_{\mathrm{ready}}\rangle.$$

Suppose the interaction correlates the two system alternatives with different environmental states:

$$\left( c_0|0\rangle+c_1|1\rangle \right) |E_{\mathrm{ready}}\rangle \longrightarrow c_0|0\rangle|E_0\rangle + c_1|1\rangle|E_1\rangle.$$

The total state is still a pure superposition on the larger Hilbert space. But the reduced state of the system is

$$\rho_S = \operatorname{Tr}_E|\Psi\rangle\langle\Psi|.$$

For

$$|\Psi\rangle = c_0|0\rangle|E_0\rangle + c_1|1\rangle|E_1\rangle,$$

one obtains

$$\rho_S ={} |c_0|^2|0\rangle\langle0| + |c_1|^2|1\rangle\langle1| + c_0c_1^*\langle E_1|E_0\rangle |0\rangle\langle1| + c_0^*c_1\langle E_0|E_1\rangle |1\rangle\langle0|.$$

The off-diagonal terms are multiplied by the environmental overlap

$$D=\langle E_1|E_0\rangle.$$

When the environmental states are nearly orthogonal,

$$|D|\ll1,$$

the reduced state is approximately diagonal in the $\{|0\rangle,|1\rangle\}$ basis:

$$\rho_S \approx |c_0|^2|0\rangle\langle0| + |c_1|^2|1\rangle\langle1|.$$

The diagonal approximation can be quantified. Its trace distance from the exact reduced state is $|c_0c_1D|$. Consequently, the probabilities of any one system-only measurement event differ by at most that amount. For an equal superposition and $|D|=0.02$, ignoring the residual coherence changes any such probability by at most $0.01$.

What Is Suppressed

Coherence is basis dependent. In the basis $\{|0\rangle,|1\rangle\}$, the coherence is carried by matrix elements such as

$$\rho_{01} = \langle0|\rho|1\rangle.$$

In the model above,

$$\rho_{01} \longmapsto D\,\rho_{01}.$$

If $D$ becomes small, observables that depend on the relative phase between $|0\rangle$ and $|1\rangle$ lose their interference signal.

For example, the operator

$$X_{01} = |0\rangle\langle1| + |1\rangle\langle0|$$

has expectation value

$$\langle X_{01}\rangle = 2\operatorname{Re}\rho_{10}.$$

After decoherence,

$$\langle X_{01}\rangle = 2\operatorname{Re} \left( D^*c_0^*c_1 \right).$$

To distinguish a phase shift from lost visibility, scan the output basis $|+_\theta\rangle=(|0\rangle+e^{i\theta}|1\rangle)/\sqrt2$. Its probability is

$$p_+(\theta)=\frac12+\operatorname{Re}[e^{i\theta}c_0c_1^*D].$$

The fringe visibility is $2|c_0c_1||D|$, equal to $|D|$ for balanced alternatives. For example, $D=i$ makes the original $X$ expectation vanish but merely shifts a full-contrast fringe. A single zero interference signal is not enough to establish decoherence.

The basis dependence of these off-diagonal terms is treated in Coherence and Preferred Bases.

Why the Environment Matters

The environment need not be a thermal bath in the narrow sense. It can include:

• scattered photons;
• air molecules;
• phonons in a solid;
• unobserved detector modes;
• uncontrolled electromagnetic fields;
• internal degrees of freedom of an apparatus;
• a deliberately ignored ancilla or measurement record.

For an initially pure environment, distinguishable conditional environment vectors directly give the overlap above. With a mixed environment, the coherence factor is instead a trace of conditional environment evolutions; a purification can restore the pure-state picture. The physical environment itself need not contain distinguishable records of the basis alternatives. Environment-Induced Decoherence gives an explicit mixed-environment example.

This is why decoherence is closely related to measurement. A measurement apparatus correlates a system with a pointer. Environmental decoherence then helps make different pointer alternatives robust and effectively noninterfering for later local observers.

For rate mechanisms, scattering-induced spatial decoherence, and many-fragment amplification, see Environment-Induced Decoherence. For the practical question of which time to report and how to estimate it, see Decoherence Timescales.

Decoherence as a Channel

The two-branch model induces a dephasing channel on the system:

$$\rho = \begin{pmatrix} \rho_{00}&\rho_{01}\\ \rho_{10}&\rho_{11} \end{pmatrix} \quad\longmapsto\quad \begin{pmatrix} \rho_{00}&D\rho_{01}\\ D^*\rho_{10}&\rho_{11} \end{pmatrix}.$$

The diagonal probabilities are unchanged in this pure-dephasing model, while the off-diagonal terms are suppressed.

More generally, if an environment monitors alternatives labeled by $a$, a useful schematic form is

$$\rho_{ab}(t) = D_{ab}(t)\rho_{ab}(0), \qquad D_{aa}(t)=1.$$

For $a\ne b$, the factors are overlaps or averaged phases determined by the interaction. Their whole matrix must be positive semidefinite; independent pairwise bounds are insufficient in higher dimensions, as shown in Dephasing Channel. When $|D_{ab}(t)|$ becomes small, interference between alternatives $a$ and $b$ is locally unavailable.

For a broader catalogue of such maps, see Common Noise Channels.

Preferred Bases and Pointer Structure

Decoherence does not suppress coherence equally in every basis. It suppresses coherence in the basis, subspaces, or overcomplete family of states that the environment distinguishes.

An idealized monitoring interaction has the form

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

When the system Hamiltonian also preserves these sectors, or can be neglected over the interaction interval, the environment evolves conditionally on the label $a$. Distinct coupling operators do not guarantee distinguishable environment outputs for every initial environment state. The states $|a\rangle$ are therefore stable labels for the environmental records, while superpositions of different $a$ values become entangled with distinguishable environmental states.

The robust alternatives are often called pointer states. In realistic systems, the pointer structure may be approximate, time-dependent, coarse-grained, or overcomplete. It is selected by the interaction Hamiltonian, the self-Hamiltonian, and the way information is amplified into the environment. The effective-superselection viewpoint is developed in Einselection.

This is why diagonalizing $\rho_S$ at one time is not enough to identify a physical classical basis. The eigenbasis of a density matrix is a mathematical fact; a pointer basis is a dynamical and operational fact.

Classical Appearance

Decoherence helps explain why macroscopic records can be treated with classical probabilities. A detector pointer, dust grain position, or current in a macroscopic circuit is never perfectly isolated. Different macroscopic alternatives scatter photons, disturb nearby molecules, excite different phonons, or leave different internal records.

If many independent environmental fragments carry partial information, the total overlap can become very small:

$$D_{\mathrm{tot}} = \prod_{k=1}^N \langle E_{1}^{(k)}|E_{0}^{(k)}\rangle.$$

Even if each factor is close to one, a large number of fragments can make the product tiny. This is one reason macroscopic superpositions are so hard to observe in ordinary conditions.

When many fragments also carry redundant information about the same pointer alternative, the same mechanism becomes the bridge to Quantum Darwinism Preview.

The resulting reduced state supports an effective classical probability calculus for the alternatives:

$$p(a) = \operatorname{Tr}(\rho_S P_a),$$

by the Born rule. The approximation is the neglect of later interference between the alternatives, not the probability formula itself. Decoherence does not automatically make their subsequent motion classical.

Decoherence as a Classical-Limit Bridge places this reduced-state result alongside closed-system wave-packet, phase-space, and semiclassical mechanisms without duplicating the open-system derivation.

Decoherence Is Approximate

Decoherence is often extremely fast, but it is still a physical process with assumptions:

• the environment must actually become correlated with the alternatives;
• the environmental records must remain inaccessible or uncontrolled;
• the relevant overlaps must become small on the timescale of interest;
• the recoherence time must be long compared with the experiment;
• the basis of suppression must be the basis relevant to the observed classical record.

Small systems can recohere when suitable dynamics refocus phases or reverse correlations. A quantum eraser can reveal fringes in conditioned subsets, but merely discarding or manipulating an environment record without conditioning cannot change the system-only reduced state. Restoring unconditional interference generally requires an appropriate reversal or a correction acting on the system.

For macroscopic environments, recoherence is usually not forbidden by the formalism, but it is practically inaccessible because phase information has spread into too many uncontrolled degrees of freedom.

Decoherence Versus Collapse

Decoherence is not the same as wavefunction collapse.

In the two-branch model, the global state after interaction is

$$c_0|0\rangle|E_0\rangle + c_1|1\rangle|E_1\rangle.$$

This is still a superposition in the full system-environment Hilbert space. The reduced density operator of the system looks diagonal because the environment has been traced out.

Therefore decoherence explains:

• why interference between alternatives becomes locally unobservable;
• why certain records are stable;
• why classical probability rules become effective for those records;
• why reversing a macroscopic measurement interaction is practically impossible.

It does not by itself explain:

• why one individual outcome is experienced rather than another;
• whether the global state literally collapses;
• which interpretation of quantum mechanics is correct;
• how to replace the Born rule;
• why a diagonal reduced state should always be treated as a proper ignorance mixture.

Those questions belong to measurement theory and foundations. Decoherence supplies a crucial dynamical mechanism, but it does not remove the need to state one’s interpretive or operational assumptions.

For the dedicated boundary page, see What Decoherence Does Not Solve.

Proper and Improper Mixtures

After decoherence, the reduced state may look like

$$\rho_S \approx \sum_a p_a |a\rangle\langle a|.$$

This has the same local statistics as a classical ensemble that prepared $|a\rangle$ with probability $p_a$. But in the decoherence model, the diagonal state came from tracing out an entangled environment.

That makes it an improper mixture: a mixed reduced state arising from correlations in a larger quantum state. It is locally indistinguishable from a proper mixture for measurements on $S$ alone, but it has a different global interpretation.

For the open-system distinction, see Proper and Improper Mixtures. For the basic density-matrix background, see Classical Mixtures vs Quantum Superpositions and Reduced Density Operators.

Exercises

1. Trace out the environment. Starting from

$$|\Psi\rangle = c_0|0\rangle|E_0\rangle + c_1|1\rangle|E_1\rangle,$$

derive the reduced density operator $\rho_S$ and identify the decoherence factor.

Solution

The projector is

$$\begin{aligned} |\Psi\rangle\langle\Psi| ={}& |c_0|^2|0\rangle\langle0|\otimes|E_0\rangle\langle E_0|\\ &+ |c_1|^2|1\rangle\langle1|\otimes|E_1\rangle\langle E_1|\\ &+ c_0c_1^*|0\rangle\langle1|\otimes|E_0\rangle\langle E_1|\\ &+ c_0^*c_1|1\rangle\langle0|\otimes|E_1\rangle\langle E_0|. \end{aligned}$$

Tracing over $E$ gives

$$\rho_S ={} |c_0|^2|0\rangle\langle0| + |c_1|^2|1\rangle\langle1| + c_0c_1^*\langle E_1|E_0\rangle |0\rangle\langle1| + c_0^*c_1\langle E_0|E_1\rangle |1\rangle\langle0|.$$

The decoherence factor multiplying $\rho_{01}$ is $D=\langle E_1|E_0\rangle$.

2. Many small records. Suppose $N$ independent environmental fragments each have overlap $d$, with $0<|d|<1$. Show that the total coherence factor decays exponentially with $N$.

Solution

The total overlap is

$$D_{\mathrm{tot}} = \prod_{k=1}^N d = d^N.$$

Its magnitude is

$$|D_{\mathrm{tot}}| = |d|^N = e^{N\ln|d|}.$$

Since $0<|d|<1$, $\ln|d|<0$, so the coherence factor decreases exponentially with the number of fragments. The endpoint $d=0$ gives exact suppression after the first fragment, without using a logarithm.

3. Decoherence is not collapse. In the two-branch model, assume $\langle E_1|E_0\rangle=0$. Is the global state a single branch? Explain using the density operator.

Solution

No. The global pure state is still

$$|\Psi\rangle = c_0|0\rangle|E_0\rangle + c_1|1\rangle|E_1\rangle.$$

The reduced density operator of $S$ is diagonal:

$$\rho_S = |c_0|^2|0\rangle\langle0| + |c_1|^2|1\rangle\langle1|.$$

The diagonal form means local measurements on $S$ cannot detect interference between the two branches. It does not mean that the global state has been replaced by one branch unless an additional collapse postulate, interpretation, or conditioning rule is added.

References

• M. Schlosshauer, “Quantum decoherence,” Physics Reports 831, 1–57 (2019). Author’s text.
• H. D. Zeh, “On the interpretation of measurement in quantum theory,” Foundations of Physics 1, 69–76 (1970). DOI.
• W. H. Zurek, “Decoherence, einselection, and the quantum origins of the classical,” Reviews of Modern Physics 75, 715–775 (2003). DOI.