Quantum Darwinism

Decoherence tells you why interference disappears, and einselection tells you which states survive — but neither explains the most basic feature of classical reality: that facts are objective. When you and I both look at this page, we agree on where it is, and neither of us disturbs it by looking. Objectivity is not a given in quantum mechanics, where a measurement generically perturbs its object and different observers can hold different descriptions. Quantum Darwinism, introduced by Wojciech Zurek and collaborators in the early 2000s, supplies the missing ingredient: a pointer state's value is not merely stable, it is copied redundantly into many fragments of the environment, so that many observers can each read a fragment and independently learn the same value without disturbing the system. The conceptual shift is to stop treating the environment as a monolithic sink to be traced away and instead to partition it into fragments $F_1,F_2,\dots$ — the photons scattering off this page reach many eyes, each catching a different slice. One then asks how much a fragment reveals about the system, measured by the quantum mutual information $I(S{:}F_f)$ as a function of the fragment fraction $f$ . The answer is the signature of Darwinism: a tiny fragment already carries essentially all the classically accessible information ( $H_S$ , the system's entropy), and the curve plateaus — the same classical fact is recorded over and over. Only by capturing almost the entire environment ( $f\to1$ ) does one recover the additional, non-redundant quantum (phase) information, where the mutual information rises sharply to $2H_S$ . This article owns the redundancy/objectivity theorem, the partial-information plateau, and the redundancy $R_\delta$ that counts how many independent copies exist. It marks two honest boundaries: Darwinism uses classical mutual information (the quantum correlations beyond it — discord — are developed in Quantum Discord and Non-Classical Correlations by name), and it presupposes both the Born rule and a chosen system/environment cut. What it delivers is a quantitative, falsifiable account of why the world is objective, tested in real experiments.

Key takeaways

Objectivity — many observers agreeing without disturbing the system — is explained by redundant environmental records of the pointer states.
Redundancy/objectivity theorem (owned): a pointer value is copied into many environment fragments; each fragment $F$ carries $I(S{:}F)\approx H_S$ of classical information, so $R_\delta\gg1$ independent copies exist.
Partial-information plateau (owned): $I(S{:}F_f)$ rises fast to $H_S$ for a tiny fragment fraction $f$ , then plateaus (redundant classical info), rising to $2H_S$ only near $f=1$ (the non-redundant quantum info).
Only pointer states proliferate: superpositions decohere and are not redundantly recorded, so einselection + Darwinism together pick which observable becomes objective (recalled from The Pointer Basis and Environment-Induced Superselection).
Honest scope: Darwinism uses classical mutual information and presupposes the Born rule; the quantum correlations beyond it are developed in Quantum Discord and Non-Classical Correlations by name.
Key idea: classical objectivity arises because information about the pointer states is copied redundantly into many environmental fragments — the partial-information plot's long plateau at $H_S$ measures how many independent observers can learn the same fact without disturbing it.

The core insight

Objectivity is redundancy: the environment does not merely decohere a system, it makes many copies of the pointer-state value in its many fragments. Measured by the quantum mutual information $I(S{:}F_f)$ , a small fragment already reveals almost all the classical information $H_S$ , and the curve plateaus — the same fact recorded $R_\delta\gg1$ times — so independent observers agree without disturbing the system. Recovering the extra quantum (phase) information requires nearly the whole environment.

Where this lives today

Quantum Darwinism has become an experimentally testable framework for studying the emergence of objective classical information. Photonic cluster-state experiments have engineered systems coupled to controllable environment degrees of freedom and measured information-sharing diagnostics, observing the emergence of redundant classical information in suitable regimes. Such experiments provide evidence for the information-theoretic signatures predicted by quantum Darwinism, while the precise form of the large-fragment rise depends on the state and experimental setting. Nitrogen-vacancy centers in diamond and engineered spin environments have reproduced the plateau-and-rise structure, extracting the redundancy $R_\delta$ from real data. Beyond confirmation, the framework guides quantum-to-classical boundary studies by asking when a given system--environment interaction produces a broad redundancy plateau, indicating that classical information is redundantly accessible, rather than a featureless growth of mutual information with fragment size. This provides an information-theoretic diagnostic of when a mesoscopic system can exhibit robust quasiclassical behavior. And it connects to the black-hole information problem and holography, where the redundant-encoding idea reappears in how information about a system is distributed across many degrees of freedom — the same mutual-information plateau structure the article owns, in a gravitational setting. In every case the diagnostic is the plot: the plateau at $H_S$ is objectivity made quantitative.

The physical picture

Why can a hundred people look at the moon and all agree it is there, none of them nudging it? Because the moon dumps information about its position into an astronomically redundant environment — sunlight scattering off it reaches every eye on the night side, each photon carrying a copy of "the moon is there". No single observer needs the whole environment; a tiny fragment of the scattered light already tells you the moon's position, and everyone else's tiny fragment tells them the same thing. That is what objectivity means operationally in quantum Darwinism: information about a system observable is encoded redundantly in many distinct environmental fragments, so multiple observers can independently access the same information, agree on the outcome, and leave the system essentially undisturbed. Quantum Darwinism asks whether decoherence can naturally produce this kind of redundant environmental record, and quantifies that redundancy through the information stored in environment fragments. The tool is the partial-information plot. Partition the environment into fragments and ask, for a fragment comprising a fraction $f$ of the environment, how much it tells you about the system: the quantum mutual information $I(S{:}F_f)$ . Darwinism predicts a characteristic shape. For very small $f$ , $I$ shoots up to $H_S$ — the system's entropy, which is all the classical information there is — because even a scrap of the environment already carries the full pointer-state record. Then it plateaus: adding more environment tells you nothing new, because the extra fragments carry redundant copies of the same classical fact. Only when $f$ approaches $1$ does the mutual information recover the full quantum value; for a globally pure system–environment state and a sufficiently fine-grained fragment description, this reaches $2H_S$ . The additional information is not redundantly accessible classical information but quantum correlations that require access to essentially the whole environment. The length of the plateau is the physics. A plateau spanning the fragment range $[f_\delta,\,1-f_\delta]$ means the classical record is stored $R_\delta\approx1/f_\delta$ times over — the redundancy. A large $R_\delta$ is the operational definition of an objective, classical property: it can be found out many times, independently, without disturbance. And crucially, this redundancy is associated with robust pointer states selected by the system–environment interaction: superpositions of distinct pointer states generally decohere rather than acquire equally stable redundant records. Thus the environment preferentially records the pointer observable rather than arbitrary superpositions. Einselection picks the basis; Darwinism makes it objective. The von Neumann entropy is owned by Entanglement Entropy and Measures of Entanglement, recalled here by name; the quantum mutual information $I(S{:}F)=S(\rho_S)+S(\rho_F)-S(\rho_{SF})$ is built here from those entropies.

Figure

The partial-information plot: fast rise to Hₛ, long redundant plateau, sharp climb to 2Hₛ

Redundancy and the partial-information plot

The framework is built entirely from the density-operator toolkit — reduced density matrices from partial traces (owned by The Reduced Density Matrix and Partial Trace) and the von Neumann entropy (owned by Entanglement Entropy and Measures of Entanglement), both recalled here by name; the quantum mutual information $I(S{:}F)=S(\rho_S)+S(\rho_F)-S(\rho_{SF})$ is assembled here from those entropies. We use them to prove that pointer information is redundant and to compute the plateau.

Figure

Redundant records: information about the pointer value is encoded in many environment fragments, so many observers can independently access the same information

Derivation ARedundancy of pointer records — a small fragment already carries

H_S

We show that when the environment has perfectly recorded the pointer value in each of $N$ fragments, a single fragment already carries the full classical information $H_S$ about the system, so the record exists $N$ times over. This redundancy, quantified by the mutual-information plateau, is the operational content of objectivity.

Step 1. The perfect-record state. Decoherence in the pointer basis, with each of $N$ environment fragments copying the pointer index, produces the branching state

|\Psi\rangle=\sum_i c_i\,|s_i\rangle\,|F_1^i\rangle|F_2^i\rangle\cdots|F_N^i\rangle \qquad (1)

with orthogonal records $\langle F_k^i|F_k^j\rangle=\delta_{ij}$ . For a qubit with $c_i=1/\sqrt2$ this is a GHZ-type state $\tfrac{1}{\sqrt2}(|0\rangle|0\cdots0\rangle+|1\rangle|1\cdots1\rangle)$ . Why this is the generic decohered state: einselection (owned by The Pointer Basis and Environment-Induced Superselection) makes the pointer states the ones the environment records, and independent fragments each carry a copy.

Step 2. Compute the mutual information of one fragment. The system's reduced state is $\rho_S=\sum_i|c_i|^2|s_i\rangle\langle s_i|$ with entropy $H_S=-\sum_i|c_i|^2\log|c_i|^2$ ; for the qubit $H_S=1$ bit. A single fragment $F_k$ is perfectly correlated with $S$ (its record is orthogonal per branch), so

I(S{:}F_k)=H_S+H_{F_k}-H_{SF_k}=H_S \qquad (2)

where: $I(S{:}F_k)$ is the quantum mutual information, and it equals the entire classical information $H_S$ — one fragment already tells you everything classically knowable about the system. Why: a perfect record makes $S$ and $F_k$ classically equivalent, so learning $F_k$ determines the pointer value completely.

Step 3. Read off the redundancy. Since each of the $N$ fragments independently provides $I(S:F_i)=H_S$ in this idealized perfect-record model, the same classical information is redundantly accessible from $N$ different fragments. For fragments of the chosen size, the redundancy is therefore $R_\delta=N$ : adding more environmental fragments does not increase the accessible classical information beyond $H_S$ , producing the plateau. Why the plateau is objectivity: $N$ observers can each grab a different fragment, all learn the same pointer value, and none disturbs the system — the hallmark of a classical, objective fact. *Verified numerically: for the GHZ-type record with $N=3$ and $N=5$ fragments, $H_S=1$ bit, $I(S{:}\text{one fragment})=1$ bit, and $I(S{:}\text{two fragments})=1$ bit — the plateau. The redundancy is $R=N$ .*

Go deeper · Derivation B · 1 minThe plateau-and-rise structure and the quantum information

Derivation BThe plateau-and-rise structure and the quantum information

We extend the perfect-record picture to a realistic noisy record and exhibit the full partial-information plot: a rise to $H_S$ , a plateau, and a final climb to $2H_S$ . The last climb is the non-redundant quantum information, and its presence is the sharp fingerprint that distinguishes a Darwinist (classical-plateau) environment from a scrambling one.

Step 1. A noisy record. Let each fragment record the pointer only imperfectly, $\langle F_k^0|F_k^1\rangle=c\ne0$ (a single scattering carries partial which-path information). Now one fragment gives $I(S{:}F_1)<H_S$ ; more fragments are needed to accumulate the full classical record, and the rise to the plateau is more gradual.

Step 2. The whole-environment value. For a pure global $S+E$ state, the mutual information between the system and the entire environment is $I(S{:}E)=S(\rho_S)+S(\rho_E)-S(\rho_{SE})=2S(\rho_S)=2H_S$ , because $S(\rho_{SE})=0$ and $S(\rho_E)=S(\rho_S)$ . Thus, in this idealized pure-state setting, the plot rises from the classical-information plateau $H_S$ toward $2H_S$ as $f\to1$ :

I(S{:}F_{f=1})=2H_S,\qquad I(S{:}F_{\text{plateau}})=H_S \qquad (3)

where: the plateau value $H_S$ is the redundant classical information (accessible from any small fragment), and the extra $H_S$ at $f=1$ is the quantum information — the global phase relations that only the entire environment together encodes and that no small fragment reveals. Why this shape is diagnostic: an information-scrambling environment (a quantum computer's, say) would give a straight ramp with no plateau; the long flat plateau is the signature of classical objectivity.

Step 3. Define the redundancy. Let $f_\delta$ be the smallest fragment fraction for which $I(S:F)\ge(1-\delta)H_S$ . The redundancy is then conventionally estimated as $R_\delta\sim1/f_\delta$ , representing the approximate number of disjoint fragments that can independently provide at least $(1-\delta)$ of the classical information. A macroscopic pointer has $R_\delta$ astronomically large. Verified numerically: for a noisy record with per-fragment overlap $c=0.6$ , $H_S=0.966$ bits, $I(S{:}1\text{ frag})=0.784$ , $I(S{:}2)=1.149$ — a rising staircase — and $I(S{:}\text{whole environment})=1.932\approx2H_S$ ; the fragment count reaching $0.9H_S$ is $f^\star=2$ .

Limiting cases

case	plot shape	reading
perfect records, $N$ fragments	rise to $H_S$ , flat plateau	$R_\delta\approx N$ ; maximally redundant in the ideal model
noisy records, overlap $c$	gradual rise, shorter plateau	fewer copies; still objective if $R_\delta\gg1$
scrambled environment	straight ramp to $2H_S$ , no plateau	no redundancy; not objective
whole environment, $f\to1$	climbs to $2H_S$	recovers the quantum (phase) information
superposition input	decoheres, not recorded	only pointer states proliferate
macroscopic pointer	plateau spanning almost all $f$	$R_\delta$ astronomically large

Common trapRedundancy is about classical information — it is not entanglement, and not discord

A common confusion is to read the mutual-information plateau as a statement about entanglement between the system and each fragment. It is not: the plateau measures classical correlation — how much a fragment reveals about the pointer value — and the fragments each carry a nearly classical record, not entanglement with the system. In fact the whole point is that pointer information is classically copyable (unlike quantum information, which the no-cloning theorem forbids from being duplicated). The genuinely quantum correlations — the part of $I(S{:}F)$ not captured by any classical measurement, the quantum discord — are precisely what does not proliferate and what the plateau is not about. Darwinism explains objectivity through the redundantly accessible classical information in the plateau; the additional quantum correlations that are not redundantly accessible are characterized by complementary measures such as quantum discord and other non-classical correlation measures. Reading the plateau as entanglement misidentifies the mechanism of classicality.

Worked examples

Example 1 — a perfect record makes one fragment enough (confidence builder). A qubit in $|+\rangle$ decoheres by copying its pointer value into $N=5$ environment qubits, giving the GHZ-type state $\tfrac{1}{\sqrt2}(|0\rangle|00000\rangle+|1\rangle|11111\rangle)$ . How much does one fragment reveal, and what is the redundancy?

Step 1. The system entropy. Tracing out the environment gives $\rho_S=\mathrm{diag}(\tfrac12,\tfrac12)$ , so $H_S=-2\cdot\tfrac12\log_2\tfrac12=\boxed{1\text{ bit}}$ (machine-verified) — one bit of classical information, the pointer value "0 or 1".

Step 2. One fragment's information. A single environment qubit is perfectly correlated with the system (record $|0\rangle$ for pointer $0$ , $|1\rangle$ for pointer $1$ ), so $I(S{:}F_1)=H_S=\boxed{1\text{ bit}}$ (machine-verified). One fragment already carries the entire classical record — you do not need the rest of the environment to learn the pointer value.

Step 3. The redundancy. Since each of the $5$ fragments independently gives $1$ bit, the record exists $R=\boxed{5}$ times over (machine-verified $I(S{:}\text{two fragments})=1$ bit, the plateau). Five independent observers could each grab a different qubit, all learn the same pointer value, and none disturb the system — objectivity with redundancy $5$ . A macroscopic pointer replaces $5$ with $\sim10^{9}$ or more scattered photons.

Example 2 — a noisy record and the rise to 2H_S (exam level). Now let each of $3$ fragments record the pointer imperfectly, with per-fragment overlap $\langle F^0|F^1\rangle=c=0.6$ . Trace out to find the staircase and confirm the whole-environment value.

Step 1. System entropy. The imperfect records reduce the effective mixing slightly: $\rho_S$ gives $H_S=\boxed{0.966\text{ bits}}$ (machine-verified) — close to one bit but not saturated, because the records are not perfectly orthogonal.

Step 2. The staircase. One fragment now gives only $I(S{:}F_1)=0.784$ bits — a partial record; two fragments give $I(S{:}F_1F_2)=1.149$ bits (machine-verified). The information rises with fragment number, reaching $0.9H_S$ at $f^\star=2$ fragments — the plot's climb toward the plateau, gentler than the perfect-record case because each noisy copy is weaker.

Step 3. The quantum tail. The whole environment (all 3 fragments) gives $I(S:E)=1.932$ bits $\approx 2H_S$ , consistent with the pure global-state value. The additional information beyond the classical plateau is non-redundant quantum correlation that is not accessible from small environment fragments. The characteristic shape is a classical-information plateau followed, for sufficiently large fragments, by a rise toward the full quantum mutual information; this separation between redundant classical information and non-redundant quantum correlations is the key diagnostic of quantum Darwinism.

Zurek and the fittest information

Origins · 1 minQuantum Darwinism and how this was discovered

Quantum Darwinism

2003–2009 · Los Alamos

Quantum Darwinism is Wojciech Zurek's proposal that classical objectivity is selected the way biological traits are — the fittest information is the most-copied. The name and framework emerged from Los Alamos in the early 2000s: with Harold Ollivier and David Poulin, Zurek showed (Physical Review Letters 93, 220401, 2004) that pointer-state information is redundantly proliferated into the environment while phase information is not, and he developed the partial-information plot as the quantitative signature. His 2009 Nature Physics review (5, 181) laid out the program — that the objective classical reality we experience is the small, redundantly-recorded classical residue of a vast quantum state — and connected it to experiment. The redundancy $R_\delta$ is his measure of how "objective" a property is: the more copies, the more classical. Zurek's person, and the einselection that decides which states get copied, are developed in The Pointer Basis and Environment-Induced Superselection by name; here the proliferation of records and the plateau are the concept.

History

1981

— Zurek: einselection identifies the states worth copying (pointer states)

2003

— Zurek: quantum Darwinism named; redundancy as the root of objectivity

2004

— Ollivier, Poulin and Zurek: pointer info proliferates, phase info does not (Phys. Rev. Lett. 93, 220401)

2009

— Zurek's Nature Physics review: the partial-information plot and objectivity (Nat. Phys. 5, 181)

2019

— photonic and NV-center experiments confirm the plateau-and-rise structure

Going further

References. Zurek, Nat. Phys. 5, 181 (2009) · Ollivier, Poulin & Zurek, Phys. Rev. Lett. 93, 220401 (2004) · Zurek, Rev. Mod. Phys. 75, 715 (2003) · Blume-Kohout & Zurek, Phys. Rev. A 73, 062310 (2006) · Riedel & Zurek, Phys. Rev. Lett. 105, 020404 (2010) · Ciampini et al., Phys. Rev. A 98, 020101 (2018) (photonic test) · Unden et al., Phys. Rev. Lett. 123, 140402 (2019) (NV-center test).

Provenance. The base article's physics is standard and was preserved. Reduced density matrices and the partial trace are owned by The Reduced Density Matrix and Partial Trace; the von Neumann entropy by Entanglement Entropy and Measures of Entanglement (the quantum mutual information $I(S{:}F)=S(\rho_S)+S(\rho_F)-S(\rho_{SF})$ is assembled here from those entropies); the shared Schmidt spectrum $S(\rho_S)=S(\rho_E)$ by The Schmidt Decomposition; the einselection that selects which states are copied by The Pointer Basis and Environment-Induced Superselection — all recalled by name. Owned and derived here: (1) the redundancy/objectivity theorem — one fragment of a perfect record carries $I(S{:}F)=H_S$ , so $R\approx N$ (machine-verified $H_S=1$ bit, $I=1$ bit for $N=3,5$ ); (2) the partial-information plateau-and-rise with $I(S{:}\text{plateau})=H_S$ and $I(S{:}\text{whole env})=2H_S$ (machine-verified $I=1.932\approx2H_S$ for a noisy record, staircase $0.784\to1.149$ bits); (3) the only-pointer-states-proliferate statement. The quantum correlations beyond classical redundancy (discord) are owned by Quantum Discord and Non-Classical Correlations; the einselection of the pointer basis by The Pointer Basis and Environment-Induced Superselection; the emergence of the classical world by Emergence of Classicality from Quantum Mechanics; the residual definite-outcome question by The Quantum Measurement Problem — all by name.

Builds on: The Reduced Density Matrix and Partial Trace · The Pointer Basis and Environment-Induced Superselection · Entanglement Entropy and Measures of Entanglement · Quantum Entanglement and Non-Separability · The Schmidt Decomposition

Leads to: Emergence of Classicality from Quantum Mechanics · The Quantum Measurement Problem · Quantum Discord and Non-Classical Correlations

How is this "Darwinism" — what is being selected?

The analogy is to natural selection: the fittest information survives and proliferates, where "fittest" means "most redundantly copyable". Pointer states are the fit ones — they are stable under monitoring (einselection) and their values get copied into many environment fragments, so they proliferate and become objective. Superpositions are unfit — they decohere and are not copied. So the environment "selects" the pointer basis to be the classical, objective reality by preferentially proliferating its information. The selection is of which information becomes objective, and the criterion is redundancy.

Doesn't the no-cloning theorem forbid copying quantum information?

It forbids copying arbitrary unknown quantum states, but not classical information encoded in a known orthogonal basis. Pointer states are (approximately) orthogonal and the environment copies only the which-pointer-state classical label — which is allowed, exactly as you can broadcast a classical bit but not an unknown qubit. This is why the plateau carries classical information: it is the copyable part. The non-copyable quantum (phase) information is precisely what does not proliferate and what only the whole environment holds — consistent with no-cloning, and the reason $I$ rises to $2H_S$ only at $f\to1$ .

What distinguishes a "Darwinist" environment from an ordinary decohering one?

Ordinary decoherence only requires the environment to carry away which-path information — a scrambling environment (like a chaotic quantum system) does that but hides the information so no small fragment can read it: its partial-information plot is a straight ramp with no plateau. A Darwinist environment additionally redundantly copies the information so many small fragments can each read it: its plot has the long plateau. Most natural environments (photons, air) are Darwinist because scattering broadcasts position redundantly. So Darwinism is a stronger, more specific claim than decoherence: not just information loss, but redundant, accessible, objective records.

Does Quantum Darwinism solve the measurement problem?

No. It explains objectivity — why observers agree on a pointer value without disturbing it — but it presupposes the Born rule (the $|c_i|^2$ probabilities) and does not select which single pointer value is realized in a given run. The global state remains a superposition of all branches, each redundantly recorded. So Darwinism sharpens the account of the classical world (adding objectivity to decoherence and einselection) but leaves the definite-outcome question — the measurement problem — untouched, developed in The Quantum Measurement Problem by name.

Why does the mutual information reach exactly 2H_S for the whole environment?

Because the global system+environment state is pure. For a pure bipartite state, the two reduced entropies are equal ( $S(\rho_S)=S(\rho_E)$ , the shared Schmidt spectrum) and the joint entropy is zero, so $I(S{:}E)=S(\rho_S)+S(\rho_E)-S(\rho_{SE})=2S(\rho_S)=2H_S$ . The factor of two is the fingerprint of a pure global state: the whole environment holds twice the classical information — the classical record ( $H_S$ ) plus the same amount again as quantum phase information. Any small fragment reveals only the first $H_S$ ; the second $H_S$ requires nearly everything.

Does Darwinism depend on how I cut the environment into fragments?

The plateau structure is robust to reasonable partitions — dividing the scattered photons into disjoint fragments of any moderate size reproduces the plateau, because each fragment independently carries the redundant record. It does depend on the fragments being genuine sub-environments (spatially or dynamically separable pieces an observer could actually intercept), not artificial recombinations. And, like all of decoherence theory, it presupposes a chosen system/environment cut. Within those natural choices, the redundancy $R_\delta$ and the plateau are well-defined and, importantly, measured in experiments that make specific fragment choices.

Quantum Darwinism

Where this lives today

The physical picture

Redundancy and the partial-information plot

Limiting cases

Worked examples

Zurek and the fittest information

Going further

Before this