NOTE-1229: Zurek — Quantum Darwinism

Contribution

A review of a programme, not a single new result. It joins three lines of Zurek’s earlier work into one account of “the classical”: einselection (decoherence selects pointer states), quantum Darwinism (the environment holds redundant records of those states, and observers learn them from fragments of it), and envariance (a derivation of the Born rule from symmetries of entangled states). Its own summary: these “mark considerable progress towards settling the quantum measurement problem” (abstract, p. 1), and in the Discussion, “The essence of the measurement problem – accounting for axioms (iv) and (v) – has been largely settled” (p. 9). The new emphasis over decoherence theory is the change of question: not which states survive information leaks to E, but “What information about the system can be found out from fragments of E?” (p. 2).

Key insight

“Tracing out” the environment is not what observers do: “Observers eavesdrop on the environment. Vast majority of our data comes from fragments of E. Environment is a witness to the state of the system” (p. 2). What makes a state classical is then not only that it survives monitoring but that it is copied many times into the environment: “Only states that produce multiple informational offspring – multiple imprints on the environment – can be found out from small fragments of E” (p. 2). When many disjoint fragments each carry nearly all the information about the same observable, many observers can learn it “independently, and without perturbing it. This is how preferred states of S become objective. Objective existence – hallmark of classicality – emerges from the quantum substrate as a consequence of redundancy” (p. 2).

Assumptions

• A given split into system and environment, and a given subdivision of the environment. “We recognize the subdivision of E into subsystems … The only requirement for a subsystem is that it should be individually accessible to measurements; observables of different subsystems commute” (Fig. 1 caption, p. 2). Fragments F are “non-overlapping collections of subsystems of E”. Which subsystems are accessible is taken as given.
• Postulates (i)–(iii) only, for sections III–IV: states are vectors in Hilbert space, evolution is unitary, and “Immediate repetition of a measurement yields the same outcome” (p. 4). The collapse postulate (iv) and the Born rule (v) are what is to be derived. Tensor-product composition is used throughout (“often cited as additional ‘axiom (0)’”, p. 9).
• Three “facts of life” for the Born-rule derivation (p. 8): “(1) When a system is not acted upon by a unitary transformation, its state remains unaffected”; “(2) Predictions regarding measurement outcomes on S (including their probabilities) can be inferred from the state of S”; “(3) When S is entangled with other systems (e.g., the environment) the state of S alone is determined by the state of the whole SE.” Also normalization (probabilities of an event and its complement sum to one), but not, it says, additivity in general (p. 8, Box).
• The redundancy results of section II use the Born rule. Mutual information is defined through von Neumann entropies of reduced density matrices (eqs. 1–3), and the paper says tracing out “relies on pk = |ψk|², Born’s rule we want to derive. We have imposed that ban while deriving and analyzing Eq. (5), but relaxed it to plot Fig. 3” (p. 6). The ordering the Discussion proposes, deriving Born first and then using reduced density matrices “with every right” (p. 9), is how the paper answers the circularity.
• Model-specific dynamics for the quantitative results: a spin coupled to many spins by an Ising Hamiltonian (Fig. 3, N = 50, couplings drawn from (0, 1]) and quantum Brownian motion of an underdamped oscillator (p. 3). The results are cited from refs. [12]–[15]; none is derived in this text.

Key results

Decoherence and einselection (section I, pp. 1–2)

• Pointer states. “Interactions that depend on a certain observable correlate it with the environment, so its eigenstates are singled out, and phase relations between such pointer states are lost” (p. 1). “Negative selection due to decoherence is the essence of environment-induced superselection, or einselection” (p. 1).
• ρ_S = Tr_E |Ψ_SE⟩⟨Ψ_SE| (eq. 1, p. 1). The predictability sieve sorts pure states by the entropy their evolved ρ_S produces; pointer states produce least, superpositions produce it “rapidly, at decoherence rates, especially when S is macroscopic” (pp. 1–2). Position is preferred because interactions “tend to depend on distance”; for a weakly damped oscillator coherent states are einselected (p. 2).

Environment as a witness (section II, pp. 2–4)

• ρ_SF = Tr_{E/F} |Ψ_SE⟩⟨Ψ_SE| (eq. 2) and I(S : F) = H_S + H_F − H_{S,F} (eq. 3), p. 2.
• The plateau (Fig. 2 caption, p. 3). For a global pure state of SE, I(S : F_f) for a typical fraction f is antisymmetric about f = ½. For Haar-random states, small fragments hold almost nothing and “once a threshold f = ½ is attained, nearly all information is in principle at hand”, so “such random states … exhibit no redundancy”. For states made by decoherence, I(S : F_f) “quickly rises to H_S (entropy of S due to decoherence) … More, up to 2H_S, can be obtained only through global measurements on S and nearly all E. H_S is therefore the classically accessible information.” The plotted curves are schematic in the extraction; the text says model calculations “exhibit behavior illustrated by the red plot” (p. 3).
• Redundancy, eq. (4), p. 3: “Redundancy is the number of independent fragments of the environment that supply almost all classical information about S, i.e., (1 − δ)H_S”: R_δ = 1/f_δ, with f_δ the fraction needed. “Rapid rise and gradual leveling of I(S : F_f), Fig. 2, implies redundancy” (p. 3). “The redundancy plateau is a classical plateau” (p. 3).
• Objectivity by redundancy. “Large redundancy implies objectivity: The state of the system can be found out indirectly and independently by many observers, who will agree about their conclusions” (Fig. 2 caption). “Observables of different F’s commute – such measurements are independent. Yet, underlying correlations mean that their outcomes imply the same state of the system, as if S were classical” (p. 3). Records are damaged by reading them (“we destroy photons!”), so error correction by redundancy matters: “with many (R_δ) copies, state of S can be found out by ∼ R_δ observers … Consensus between copies suggests objective existence of the state of S” (p. 3).
• Models (p. 3). Spin environments [12–14]: “the same number of spins suffices to reach the plateau”, so redundancy is “proportional to the number of the environment subsystems that interact with the system”. Quantum Brownian motion [15, 16]: I(S : F) ≈ H_S + ½ ln(f/(1 − f)) (reconstructed from the extraction) and R_δ ≈ s^{2δ} for an initially squeezed state with squeeze factor s; “Similar equation should hold for more general ‘Schrödinger cat’ states”.
• Branch structure (pp. 3, 5–6). Decoherence states “are far from random. Roughly speaking, they have a branch structure” (p. 3). Eq. (6), p. 6: |Ψ_SE⟩ = Σ_k ψ_k |s_k⟩|e_k^(1)⟩…|e_k^(N)⟩ = Σ_k ψ_k |s_k⟩|ε_k⟩, “to a good approximation, a multi-system Schmidt decomposition”, “a very atypical minimally entangled superposition” (p. 6).
• Shannon versus von Neumann (p. 4). With Shannon entropies of optimal local observables the I(S : F_f) graph “for small f would look very similar”, but never exceeds H_S, “antisymmetry is lost, and the plateau continues until the end at f ∼ 1”. The region above H_S needs global measurements, which is “why decoherence is so hard to undo”.
• Mixed environments (p. 4). An environment with initial entropy h per bit is a noisy channel: a bit from E yields only 1 − h bits about S. “a completely mixed E (h = 1) is useless (even though it can still induce decoherence!)”. A partly mixed one still “can still climb the same redundancy plateau at H_S [19]”, needing a larger fragment. Ref. [19] is “in preparation”.
• Which environments carry records (p. 4). Photons are ideal: they “do not interact with each other” and are far from equilibrium with our surroundings. “Air molecules scatter from one another, so that whatever record they may have gathered becomes effectively undecipherable.” So decoherence and record-keeping can be done by different environments.

From copying to quantum jumps (section III, pp. 4–6)

• Eq. (5), p. 5. If |u⟩ and |v⟩ each leave an imprint and stay unperturbed, |u⟩|e₀⟩ → |u⟩|e_u⟩, |v⟩|e₀⟩ → |v⟩|e_v⟩, unitarity preserves the scalar product: ⟨u|v⟩ = ⟨u|v⟩⟨e_u|e_v⟩. So “(a) either ⟨e_u|e_v⟩ = 1 (which means that copying was completely unsuccessful), or; (b) ⟨u|v⟩ = 0”. “As a consequence of unitarity, only sets of orthogonal states … can be so copied, explaining selection of a set of outcomes – terminal points of quantum jumps” (p. 5). It “relies on just two values of the scalar product – 0 and 1 – and, thus, does not appeal to Born’s rule” (p. 5). No-cloning is evaded because “copying of observables” is not cloning of unknown states (p. 4).
• Selectivity (Fig. 3, p. 5). With Shannon-based quantities for σ(μ) = cos μ σ_z + sin μ σ_x, the whole environment “knows” many observables (Fig. 3a), but R_{δ=0.1}(σ) “is sharply peaked around the pointer observable: Redundancy is a very selective criterion” (Fig. 3b). Local random measurements on m environment spins reveal “pointer states – and only pointer states” (Fig. 3c). A theorem cited from [12, 24]: “only pointer states can be discovered from fractions of E. Other observables … can be deduced only to the extent they are correlated with the pointer observable” (p. 5).
• Collapse as the rise of I (p. 6). “This transition from uncertainty … to certainty (once a sufficiently long branch fragment becomes known) accounts for perception of ‘collapse’. … Collapse is brief compared to the ensuing period of certainty about the outcome, as f_δ ≪ 1, but, nevertheless, not instantaneous.”
• Apparatus (p. 6). It suffices that records “are ‘repeatably accessible’. Transfer of responsibility for repeatability from a quantum S to a (still quantum) A allows one to model non-orthogonal measurement outcomes (POVM’s)”; the apparatus pointer states are then the ones copied into E. “Indeed, we usually experience ‘quantum jumps’ via an apparatus pointer.”
• Stance (p. 6). “our discussion is interpretation-free, and this is a virtue.” “Quantum states acquire objective existence when reproduced in many copies. … And there is only one copy of the Universe. Treating its state as if it really existed [26, 27, 28] seems unwarranted and ‘classical’.”

Probabilities from entanglement (section IV and Box, pp. 6–8)

• Envariance, eq. (7), p. 7: u_S ⊗ 1_E |Ψ_SE⟩ = |Φ_SE⟩; 1_S ⊗ u_E |Φ_SE⟩ = |Ψ_SE⟩. A phase change of Schmidt coefficients made on S can be undone on E, so by facts (1)–(3) phases cannot matter for S. “Such loss of phase significance for S entangled with E implies decoherence” (pp. 7–8).
• Equiprobability (p. 8; Fig. 4). For equal |ψ_k| a swap in S can be undone by a counterswap in E, so p(0) = p(1) = ½.
• Born’s rule by fine-graining (Box, p. 8). For |ψ_SE⟩ ∝ √2 |0⟩|+⟩ + |2⟩|2⟩ (reconstructed), a controlled operation from E to an ancilla C gives three equal-amplitude Schmidt terms, so p₂ = ⅓ and, by normalization, p₀ = ⅔. The extension to commensurate probabilities “is conceptually straightforward but notationally cumbersome”; the noncommensurate case “is settled with an appeal to continuity” (p. 8).

Discussion (section V, p. 9)

• The two contested postulates are claimed derived from the first three. “How Darwinian is Quantum Darwinism?”: survival of the fittest, with fitness as the ability to procreate; no-cloning implies “competition for resources – space in E – so that only pointer states can multiply”. The Universe’s memory is finite, so “old events will be eventually ‘overwritten’ by new ones … And if there is no record of an event, has it really happened?” Whether quantum Darwinism is “in some way behind the familiar natural selection” is raised and not answered.

Claims

id	claim	strength	support
C1	Only sets of orthogonal states can be copied repeatably into an environment while left unperturbed; the argument uses unitarity and repeatability, not the Born rule	strong	eq. (5), p. 5: a two-line derivation from the preservation of scalar products
C2	States of SE produced by decoherence put almost all of the system’s information H_S into small fragments of E, so redundancy R_δ = 1/f_δ is large, unlike Haar-random states	moderate	cited model calculations (spin environments [12–14], quantum Brownian motion [15, 16]); schematic Fig. 2; not derived here
C3	Redundancy is high only for the pointer observable; other observables are learnable from fragments only through their correlation with it	moderate	Fig. 3 for one spin model (N = 50); a theorem cited from [12, 24], not stated or proved here
C4	Redundancy is what makes pointer states objective: many observers can learn them independently, indirectly and without perturbing S, and agree	weak	an interpretive identification (p. 2, Fig. 2 caption, p. 3); independence rests on fragments’ observables commuting plus “underlying correlations” (p. 3); no result shows that each fragment’s record determines the same outcome for every observer
C5	The apparent collapse is the steep initial rise of I(S : F_f): brief but not instantaneous	weak	an interpretation of the plateau (p. 6)
C6	Envariance gives the Born rule without assuming it or additivity of probabilities	moderate	Fig. 4, eq. (7), Box (pp. 7–8); rests on facts (1)–(3) and the equal probability of Schmidt partners; commensurate and continuity extensions are asserted; critiques [36, 37] are cited, not answered
C7	A mixed environment still reaches the same plateau H_S, more slowly	weak	asserted with ref. [19] “in preparation” (p. 4)
C8	The measurement problem has been “largely settled”	weak	the author’s assessment (p. 9); the paper also says “Much work remains to be done”

Concepts

• Einselection — environment-induced superselection: “Under scrutiny of the environment, only pointer states remain unchanged. Other states decohere into mixtures of stable pointer states” (p. 1).
• Pointer states — the states singled out by the observable the system–environment interaction depends on; at the top of the predictability sieve (pp. 1–2).
• Fragment F_f — a non-overlapping collection of a fraction f of the environment’s subsystems, each “individually accessible to measurements” (Fig. 1).
• Redundancy R_δ — the number of disjoint fragments each holding (1 − δ)H_S of the information about S; R_δ = 1/f_δ (eq. 4).
• Redundancy (classical) plateau — the level H_S at which I(S : F_f) levels off; “additional bits only confirm what is already known” (p. 3).
• Branch — a product term of eq. (6), labelled by a pointer state; its “bud” is the pointer state, its fragments imply it (pp. 3, 6).
• Envariance — entanglement-assisted invariance: a transformation of S that can be undone by acting on E alone (eq. 7).
• Quantum Darwinism — “the proliferation, in the environment, of multiple records of selected states of a quantum system” (abstract).

Connections

• Bohr (ref. [1]): the classical apparatus the paper replaces with a quantum environment; outcome selection “parallels Bohr’s insistence” but “follows from purely quantum Eq. (5)” (p. 6).
• Everett, DeWitt (refs. [26]–[28]): relative states “come to mind”, but “We abstain” from the reality of other branches (p. 6). The “existential interpretation” of ref. [38] is named in the reference list only.
• Zurek 1981, 1982 (refs. [6], [7]): pointer basis and einselection; [7] is the paper Rovelli credits with RQM’s conclusion (NOTE-1227). Not held.
• Ollivier, Poulin and Zurek (refs. [12], [24]), Blume-Kohout and Zurek (refs. [13]–[15]): the model results and the selectivity theorem this review relies on. Not held.
• Wootters–Zurek, Dieks (refs. [20], [21]): no-cloning, evaded by copying observables.
• Gleason (ref. [35]) and the critiques Schlosshauer and Fine, Barnum (refs. [36], [37]) for the Born-rule derivation.
• Wheeler’s “It from bit” (ref. [39]), named as one way to think of the transition.
• Nothing in it on spectrum broadcast structures or strong quantum Darwinism, which postdate it; the paper’s only refinement of mutual information is the switch to Shannon entropies of specific observables (Fig. 3; p. 5).

Bearing on the record

• QUESTION-285: the “redundant record” rung, as the paper has it, and what it adds to the “stable record” rung. The exchange’s description is right on the main points: pointer observables are copied into many fragments (eq. 4, Fig. 2, p. 3), and observers learn them “independently, and without perturbing it” (p. 2), “indirectly” (p. 3). Two things the question can use. First, the paper separates rungs the question keeps apart: decoherence (correlation with E) can happen without usable records, since a completely mixed environment “is useless (even though it can still induce decoherence!)” and air molecules decohere a system while their records become “effectively undecipherable” (p. 4). So correlation and accessible record come apart, with the medium’s structure deciding which. Second, the paper ties the record rung to repeatability, its postulate (iii), and derives the outcome set from it (eq. 5): what can be recorded repeatedly without disturbance is exactly an orthogonal set. That is a physical criterion for “measurement” as separating alternatives into distinguishable outcomes, without an observer. Its p. 6 move, “Transfer of responsibility for repeatability from a quantum S to a (still quantum) A”, is an apparatus in TERM-089’s operational sense, with differentiation (copying) and retention (the pointer states) and nothing of use or revision. The Discussion touches the stable-record rung too: memory is finite, “old events will be eventually ‘overwritten’”, “if there is no record of an event, has it really happened?” (p. 9), a question about persistence the paper raises and does not pursue (Di Biagio and Rovelli, LIT-1666, unread, is the record’s other source for it).
• The exchange’s caveat is not in this paper. “Mutual information alone does not guarantee independently accessible classical records” is not said here; the paper asserts the opposite inference: “Rapid rise and gradual leveling of I(S : F_f), Fig. 2, implies redundancy” (p. 3), and “Large redundancy implies objectivity” (Fig. 2). What it does have is a partial version of the worry: it notes that von Neumann mutual information can rise above H_S toward 2H_S through global, quantum correlations (Fig. 2, p. 4), calls H_S “the classically accessible information” on that ground, and switches to Shannon entropies of specific observables to ask “What observable is this information about?” (p. 5). Independence of observers is argued from the commuting of different fragments’ observables plus “underlying correlations” (p. 3), not from any condition on the joint state of system and fragments. The A19 attribution of the caveat to later work is consistent with this reading; that work is not held and was not checked. THEORY-1055 files the finding with this limit.
• THEORY-621: a different route to inter-observer agreement, and a shared refusal. THEORY-621’s promote_when asks for a result that what one system records about another agrees across observers. Zurek gives a mechanism for agreement: observers who read disjoint fragments of a redundantly imprinted environment find “the same state of the system” (p. 3). But the setting is not RQM’s: facts are not indexed to systems, probabilities come from the Born rule (derived by envariance, section IV), and agreement is shown only for pointer observables in models, and argued, not proved, for observers in general. So it does not meet the promote_when; it suggests what such a result might look like (consensus through a shared medium). The two agree that the universe has no quantum state to take as real: “there is only one copy of the Universe. Treating its state as if it really existed … seems unwarranted and ‘classical’” (p. 6), against THEORY-621’s “no quantum state of the universe”. Zurek’s reason is different (objectivity needs copies), and he calls his account “interpretation-free” (p. 6). Rovelli’s “stable facts” via decoherence (THEORY-621’s p. 3) is the einselection half of this paper.
• CLAIM-478: weak, and only for the first half. The paper’s dynamics have a scale separation in kind: superpositions “produce entropy rapidly, at decoherence rates”, while pointer states produce least (pp. 1–2), so after decoherence the pointer basis is what persists and fixes the set of possible outcomes for whatever reads the environment. That is the claim’s “separation makes a constraint” in a quantum case. But the paper gives no time scales, and its “Collapse is brief compared to the ensuing period of certainty … as f_δ ≪ 1” (p. 6) is a ratio of fragment sizes standing in for times, not a derived separation. Support of the shape, no more.
• THEORY-017: the paper takes its parts from outside. Quantum Darwinism needs a system–environment split and a subdivision of E into “individually accessible” subsystems whose observables commute (Fig. 1). The paper supplies both by assumption. That is THEORY-017’s point (a factorisation is extra data), here in a work whose objectivity claim depends on which factorisation is chosen.
• NOTE-044 (Grangier and Auffèves) names quantum Darwinism as the reductionist programme its postulate excludes, without engaging it. This reading is what that note’s C6 would be checked against.
• CLAIM-342. Redundancy is a way evidence can reach many observers without passing through any one observer’s reading of it (p. 3). That is a physical instance of evidence whose arrival does not pass through the interpretation it tests, at the level of records. It says nothing of judgement loops; noted only.
• No instruction for machine-learning practice.

Limitations

• A review. The quantitative results (plateau shape, redundancy scaling, selectivity, mixed environments) are cited from refs. [12]–[19], two of them “in preparation”. Fig. 2 is schematic. The theorem that only pointer states can be discovered from fragments is cited, not stated.
• Objectivity is an identification, not a theorem. “Large redundancy implies objectivity” (Fig. 2) and “Consensus between copies suggests objective existence” (p. 3) move from an information measure to a property of observers’ access. The paper does not give conditions under which fragment records are each sufficient and mutually consistent for every observer.
• The S/E split and the fragment structure are assumed.
• The Born-rule derivation depends on facts (1)–(3) and on equal probabilities for Schmidt partners; its extension beyond the worked example is asserted; the cited critiques are not answered here.
• “Largely settled” is the author’s verdict; single outcomes are addressed by “perception of ‘collapse’” (p. 6) while the paper abstains on the reality of other branches.
• Extraction. Figure contents are lost; the QBM expression and the Box’s state are reconstructed. The published journal text was not compared.

Open questions

• Under what condition on the joint state of S and several fragments does a mutual-information plateau guarantee that distinct observers, each reading one fragment, obtain the same pointer outcome without disturbing S? This is the question the later strong-Darwinism and spectrum-broadcast work is standardly cited for (THEORY-1055).
• What fixes which environments preserve records (photons) and which scramble them (air), quantitatively, and over what times (p. 4)?
• The paper’s last question: is quantum Darwinism “in some way behind the familiar natural selection” (p. 9)?
• Do the assumptions of the envariance argument hold up against refs. [36], [37]?