WorksPhysics and Astronomy
The Structure of Shared Classical Facts in Quantum Systems
Abstract
Perfectly matching local outcomes need not constitute a nondisturbing shared classical record. For finite-dimensional multipartite states, we prove that nontrivial exact records are equivalent to a spectrum sharing structure with a unique finest decomposition into locally distinguishable sectors, where entanglement may persist. For a fixed complete pointer, exact sharing is equivalent to strong quantum Darwinism together with global nondisturbance, while spectrum broadcast structure additionally requires conditional independence. Arbitrarily weak full-rank admixtures eliminate exact sharing, but two-sided distance bounds quantify robustness through disturbance and disagreement, while information bounds connect accurate local readouts to observer consensus.
Introduction
A photograph, a detector click, and a memory register may all convey the same fact even though they occupy different physical systems. Explaining how such independently accessible records emerge from quantum mechanics is a central problem in the quantum-to-classical transition. Decoherence shows how interactions with an environment suppress interference between preferred pointer alternatives in a reduced state [1, 2, 3, 4, 5]. Yet local decoherence alone does not identify which fragments store the information, whether separate observers can access it, or whether reading it changes the larger state.
Quantum Darwinism (QD) addresses the proliferation of information into environmental fragments [6, 7, 8, 9, 10]. Its characteristic mutual-information plateau indicates that many fragments correlate strongly with the same system [11, 12, 13, 14, 15]. A plateau is not, however, a direct certificate of readable classical records: quantum mutual information also counts correlations that local measurements cannot turn into classical information about a specified pointer [16, 17, 18, 19]. Strong quantum Darwinism (SQD) remedies this by requiring the relevant information to be classically accessible and excluding discord from the system–fragment correlations [20, 21].
Spectrum broadcast structure (SBS) gives a complementary state-level description [21, 22, 23, 24]. In an SBS state, distinct system-pointer values occupy orthogonal conditional supports in every observed fragment, allowing independent nondisturbing readout. SBS additionally requires those fragments to be uncorrelated conditional on the pointer value. That product assumption is stronger than the existence of a shared classical record. Indeed, SQD supplemented by a suitable multipartite independence condition yields SBS [20, 25, 26]. These approaches illuminate information redundancy, local accessibility, and conditional independence, but do not make them interchangeable.
Nondisturbance is not new to structural accounts of objectivity: Bohr’s requirement was used to derive the form of objective pointer records in a system–environment setting [23, 27]. A complementary line of work uses quantum error correction and recoverable logical algebras to describe classical information available on multiple fragments [28, 29, 30]. These results motivate a distinction between the local recoverability of encoded information and a nondisturbing measurement of a particular full state. Measuring a locally represented logical observable can still dephase a coherent encoded state when its outcome is ignored.
Here we fix a multipartite density operator and a partition into disjoint observers’ laboratories, without designating a privileged system or a pointer in advance. We ask whether some nonconstant classical label admits locally calibrated readouts that agree with certainty and whose outcome-forgotten action preserves the entire tested state. Both the label and the readouts are allowed to depend on that state, while the observer partition remains fixed. This state-dependent operational question differs from identifying which logical observables are representable on a code subspace, and global nondisturbance cannot be inferred from reduced system–fragment states alone.
The distinction already appears for two qubits. Both the classical mixture
and the Bell state yield perfectly matching computational-basis outcomes. But only the mixture is preserved by either en route local measurement: measuring the Bell state instead removes its in-branch coherence and produces . Agreement alone therefore does not establish an already available, nondisturbing shared fact.
Our contributions follow this operational question. First, Sec. II proves that exact shared records are precisely the nontrivial spectrum sharing structures: locally distinguishable orthogonal record sectors, with a unique finest decomposition of the occupied state space. Here “spectrum” denotes record sectors, not necessarily eigenspaces of the density operator. Conditional entanglement is allowed. Second, Sec. III characterizes the extra full-state nondisturbance requirement for a prescribed complete pointer and distinguishes the resulting state family from SQD and SBS. Third, Sec. IV derives two-sided distance bounds for a fixed local readout in terms of its disturbance and disagreement, and connects reliable readout to observer consensus. The proofs and counterexamples are collected in the appendices. Finally, Sec. V discusses the physical interpretation and the open problem of record-formation dynamics.
Structure of exact shared facts
Two requirements for a shared record
Consider a finite-dimensional multipartite state on , where and labels the observers. Each observer controls only its own tensor factor. Any inaccessible environment has already been traced out, and the partition into observers’ laboratories is held fixed throughout the test.
To ask whether one classical label is already available to everyone, let observer perform a projective readout , with , from a common finite set of calibrated outcomes . Calibration specifies which detector outcomes represent the same event, without requiring identical physical detectors. Write for the projector on the complete system, where acts as the identity outside laboratory .
The readouts define an exact shared record if (i) forgetting the outcome of any one local measurement leaves the entire tested state unchanged and (ii) any two observers obtain the same calibrated answer with probability one. For projective measurements these requirements are
The first condition is equivalent to for each observer. It concerns the full state, not only the observer’s reduced state or the probabilities of the measured outcomes. The second line gives the probability of contradictory reports because measurements on disjoint laboratories commute. It also implies that the probability is independent of .
A nontrivial exact shared fact is an exact shared record with at least two outcomes of strictly positive probability. For the fixed observer partition , let be the class of states that admit some nontrivial exact shared record. In defining , neither the local measurements nor their calibrated label set is prescribed in advance.
The two requirements play different roles. The Bell state in the Introduction gives perfectly matching outcomes, but either unread local measurement removes the coherence between and . Conversely, local measurements can preserve a state yet produce occasional disagreements, as in the noisy example of Sec. IV. Exact sharing requires both nondisturbance and agreement [23, 27, 31, 32].
Locally distinguishable sectors and their characterization
The operational conditions in Eq. (2) have a state-level counterpart: the state decomposes into orthogonal sectors, and the label of each sector is locally readable in every laboratory. For each occupied label (so ), define
Here is the joint projector for unanimous outcome , is the normalized state conditioned on that outcome, and is the local subspace associated with in laboratory . Perfect agreement gives .
The corresponding state-level condition is formalized as follows.
Definition 1 (Spectrum sharing structure). A finite-dimensional multipartite state on has a spectrum sharing structure if there exist a finite label set , positive numbers with , normalized states , and subspaces such that
and, for each observer ,
The direct sum means that different labels occupy orthogonal global blocks, with no coherence between them. The state has no support outside their sum. The structure is nontrivial if .
This definition expresses the operational requirements as conditions on the state alone. The following theorem shows that the two descriptions are equivalent and, moreover, identifies the unique finest decomposition into shared support sectors.
Theorem 1 (Exact shared facts and spectrum sharing). For a finite-dimensional multipartite state with at least two disjoint observer subsystems, the following statements are equivalent:
The state admits a nontrivial exact shared record satisfying Eq. (2).
The state has a nontrivial spectrum sharing structure in the sense of Definition 1.
Moreover, for the fixed observer partition, there is a unique finest decomposition into shared support sectors,
up to relabeling. If denotes the orthogonal projections onto these nonzero sectors, then
Every other exact shared record is obtained by combining some of these finest labels. The projectors act on the support of , so their sum is the projector onto . When no nontrivial shared fact exists, this finest decomposition has only one occupied sector.
Both Eqs. (4) and (5) are essential. Eq. (4) gives globally orthogonal blocks and confines each block to the product of its local record subspaces, whereas Eq. (5) makes different labels orthogonal within every individual laboratory. Thus, global orthogonality alone does not ensure that each observer can distinguish the sectors. Here “spectrum” refers to these record sectors, not necessarily to eigenspaces of . The conditional states need not factorize and may remain entangled across laboratories.
Conversely, suppose the state has the structure in Eqs. (4) and (5). In each laboratory, measure the orthogonal subspaces , assigning any unused orthogonal complement to one of the outcomes. The support condition ensures that every state gives outcome with certainty in every laboratory. The absence of coherence between different sectors then ensures that forgetting any one outcome leaves unchanged. These readouts consequently satisfy Eq. (2).
Finally, each exact readout induces the same projector on in every laboratory. Projectors associated with two different readouts can therefore be represented in different laboratories and commute. Their common refinement gives the unique finest decomposition, and every other exact record is a coarse-graining of it. The projector argument is given in Appendix A.
Scope and implications of the characterization
The Bell-state example in the Introduction illustrates that perfectly correlated local outcomes do not necessarily constitute a nondisturbing shared fact: an unread local measurement may still disturb the joint state. Thus, the theorem characterizes shared records, not correlated measurement statistics alone [17, 19, 33].
A related distinction is that recovering an observable’s outcome statistics is not the same as reading an existing record without disturbing the state. Consider the repetition encoding and . Either physical qubit reveals the logical distribution. However, for an encoded superposition
an unread measurement of either qubit produces
destroying the coherence between the two logical branches. Recovering the logical statistics is therefore weaker than the exact-record criterion established here. In operator-algebra quantum error correction, local representatives of encoded observables, including central elements of recoverable algebras, are studied at the level of a code subspace [28, 29, 30]. Our test instead asks whether the actual state is invariant under an unread local projective measurement. The two notions should not be identified merely because the same classical outcome can be obtained locally. This is consistent with the no-broadcasting theorem: the logical label is classical, whereas an arbitrary quantum superposition is not redundantly readable without disturbing its coherence [32, 34].
The theorem identifies which nonconstant classical labels are already locally readable by all observers. It does not require the entire quantum state, or even a complete pointer basis of a designated subsystem, to be shared. To compare this general shared-record criterion with frameworks based on a designated system pointer, Sec. III fixes such a pointer and examines its relation to quantum Darwinism and spectrum broadcast structure.
Shared facts and quantum objectivity
For comparison with quantum Darwinism, fix one finite-dimensional state , a designated system , and disjoint environmental fragments . Abbreviate the collection of fragments as . For our shared-record test, and each count as separate observers, so . All exact criteria below refer to the same full state and this fixed partition. Write for von Neumann entropy and for quantum mutual information. For a transparent fixed-partition comparison, we use the idealized exact equality below as the QD criterion for every displayed fragment. This is not the same as testing an approximate or fragment-size-averaged mutual-information plateau in a dynamical model. Conventional quantum Darwinism (QD) is represented here by
To compare complete-pointer records, fix a rank-one orthonormal basis of and restrict the state families to states whose is diagonal in this basis. Thus is an eigenbasis of (including any degenerate eigenspaces), , and only labels with need be distinguished. We consider states with at least two occupied pointer labels (), consistently with the nontrivial-record convention of Sec. II. These assumptions are important when translating entropic criteria into an exact state form and comparing the state families below. Strong quantum Darwinism (SQD) requires all the correlation to be accessible as classical information about this complete pointer:
Here maximizes the classical mutual information over local measurements on and . The Holevo quantity is , with the conditional fragment state [35]. We use for the Shannon entropy of the pointer outcomes. By the diagonal-pointer assumption, . Under this assumption, ideal SQD is equivalent to the pairwise state form
Thus each pairwise state records the full pointer, but none of these marginals by itself tests whether the complete state retains coherence between distinct pointer branches [20, 21]. Classical accessibility in each pairwise marginal and nondisturbance of the global state are therefore separate requirements. The equivalence between the entropic and state-based forms is derived in Appendix C.
Spectrum broadcast structure (SBS) further requires independence of the fragments conditioned on the pointer value:
The standard SQD–SBS relation is therefore an implication from conditional product structure to locally accessible information [20, 25, 26].
For this fixed partition and complete pointer , let denote the class of states for which the prescribed rank-one projective measurement and matching, calibrated projective readouts on every fragment satisfy both conditions of Eq. (2) on the full state. Thus requires nondisturbing access to every occupied value of the specified complete system pointer. By contrast, requires only some nonconstant exact shared event, which may be a coarser observable of and need not correspond to the chosen pointer . The definition of does not single out a designated system. Equivalently, it only requires a nontrivial spectrum sharing structure by Theorem 1. Consequently, on this fixed observer partition, but a state in need not belong to . All comparisons between , , and use this fixed pointer and the diagonal-pointer state family.
Proposition 1 (Exact pointer-record overlap). For this nontrivial complete rank-one pointer and fixed fragments,
where means for each occupied . Equivalently, the global state has the form
No factorization of is required. Within this family SBS holds precisely when for each occupied label.
The proposition (proved in Appendix C) separates local readability from global nondisturbance: certifies the former in the marginals, while excludes coherence between pointer sectors in the full tested state. Earlier identifications of SQD with pointer objectivity [20] are thus not being contradicted: the present criterion explicitly tests a stronger, full-state outcome-forgotten readout on a specified multipartite partition. For instance, a coherent GHZ state has the ideal SQD pairwise marginals, although reading any one pointer record dephases the complete pure state (Example C in Appendix D). The fixed-pointer hierarchy is therefore
The general shared-event class is broader than and has no general inclusion relation with QD or SQD.
The strict distinction between and SBS is already visible in a simple example. Let , where each holds an orthogonal copy of a pointer bit, and let the two registers share . The state
has perfectly readable, nondisturbing pointer copies in and . Yet for , the conditional state is entangled, so the fragments are not conditionally independent and the state is not SBS. The explicit verification is in Example D of Appendix D. Figure 1 summarizes the full hierarchy and its other witnesses.
Figure 1. Fixed-partition hierarchy and its generalization. The outer Euler sets compare exact shared facts R with QD and SQD. Distinct boundary styles also distinguish the sets in grayscale reproduction. The nested sets distinguish SBS R SQD. Here A denotes an incomplete shared label, B two Bell pairs, C a coherent GHZ state, D an exact pointer record with conditional entanglement, and E a shared tag accompanying two Bell pairs. The state families are compared for a common observer partition and, when specified, the same complete pointer . The diagram does not represent a random-partition consensus probability.
Examples A–E in Figure 1 show how the boundaries differ. A and B separate general shared records from QD in both directions. C satisfies SQD but fails global nondisturbance. D is an exact pointer record that is not SBS. E belongs to but not SQD.
Conditional correlations explain the gap between and SBS. For the conditional states in Eq. (12), define
where is quantum relative entropy, with the usual support convention. Within R, if and only if the state is SBS. Otherwise, conditional correlations, including entanglement, may remain despite exact local access to the pointer.
These comparisons depend on the tested state and the chosen observer partition. For fixed disjoint laboratories, a record shared by all members of a group is also shared by any subgroup with at least two observers. Adding observers can only restrict the set of shared records. We assume local projective readouts and nondisturbance of the full state, rather than arbitrary decoding instruments. Randomly sampled fragment partitions are considered separately in Sec. IV C.
From exact shared facts to robust records
Exactness and rank-changing noise
Suppose the shared classical bit in Eq. (1) is contaminated by a small amount of completely random two-qubit noise:
Here is the noise strength. With probability we prepare the perfectly matched bit, while with probability we prepare a uniformly random two-bit string. The probability of disagreement is therefore . Both observers can still read in the computational basis without disturbing the state, because it is diagonal in that basis.
Yet the exact-record property changes abruptly. At , the two observers share one exact bit. For every , even if extremely small, has nonzero weight on all four computational-basis strings. In particular, mismatched outcomes are now possible, so no nontrivial exact shared fact remains. A more general statement follows.
Corollary 1 (No nontrivial exact shared fact at full rank). If a multipartite state has full rank on the tensor product of two or more disjoint observer subsystems, then it admits no nontrivial exact shared record for that observer partition.
The reason is elementary. At full rank, every pair of nonzero local projectors on distinct laboratories has strictly positive joint probability. If two different shared labels both occur, some pair of mismatched outcomes must then have positive probability, contradicting Eq. (2). A proof is given in Appendix A. When the support is smaller, several different local projectors may agree on the part of Hilbert space that is actually occupied.
It would be misleading to conclude that a tiny suddenly erases a macroscopic memory. A detector with a one-in-a-million disagreement probability is nearly as useful as a perfect detector, despite failing an exact all-or-nothing test. To describe real records, we therefore keep the proposed local measurements fixed and quantify their imperfections.
Operational errors and distance to ideal records
Let each observer attempt to read the same finite label using projectors . We do not assume that these measurements are nondisturbing or that they agree. Two directly interpretable errors are enough for our bound.
First, imagine that all observers make their measurements and then throw away the recorded outcomes. The resulting state is
The tuple lists the outcomes in all laboratories. The map removes coherence between alternatives distinguished by the complete set of local measurements. We quantify its disturbance by .
Second, let project onto the event that every observer reports the same particular value , and let collect all such events. The disagreement probability is . In summary,
For a Hermitian operator, means the sum of the absolute values of its eigenvalues. We use the unhalved trace norm throughout. The conventional trace distance between density operators is often defined with an extra factor . Thus means no disturbance, and means perfect agreement. The noisy shared bit has and .
What do small values of these errors guarantee? Define as the family of ideal states that give only matching local outcomes and have no coherence between different values of the shared label:
For this specified choice of local measurements, the quantity
is the distance to the nearest ideal shared-record state. It asks how much the physical state would need to change before the proposed records became perfect. The optimization does not alter which measurements the observers are allowed to perform.
Theorem 2 (Accuracy of a fixed shared readout). For any family of local projective measurements with , the distance to the ideal shared-record states satisfies
In particular, both errors vanish exactly if and only if .
This bound has a useful reading without the proof. If the unread measurement strongly changes the state, or if the observers frequently disagree, the state cannot be close to a perfect shared record. This is the lower bound. Conversely, when both effects are small, there is an ideal state nearby that makes the same local projectors into a perfect, nondisturbing record. This is the upper bound.
The upper estimate can be understood as a two-step repair procedure. First perform the unread local measurement, producing . This costs at most in trace norm. Then keep only the subspace in which all observers agree and normalize the remaining state. Because the first step has already removed coherence between different outcome strings, this second correction costs exactly , not a term proportional to . Appendix B gives the proof and explains why the coefficient cannot be improved in general.
This result has a different target from earlier geometric bounds on proximity to spectrum broadcast structure, which use pointer dephasing and distinguishability of conditional environmental states [20, 36]. The ideal set here fixes the observers’ proposed projectors but does not require conditional independence of the fragments. Moreover, Eq. (20) supplies both a lower and an upper estimate using directly interpretable errors of that same readout. It therefore quantifies nondisturbing agreement rather than proximity to the more restrictive SBS family.
For the noisy bit in Eq. (15), the bound is particularly transparent:
The distance from a perfect shared record goes smoothly to zero as the noise vanishes, even though no exact shared fact remains for any positive .
The errors are also stable if the prepared state changes slightly while the detector settings remain fixed. If two states are separated by an unhalved trace-norm distance , their disagreement parameters differ by at most and their disagreement probabilities by at most . Thus the easily evaluated expression changes by no more than . The estimates follow directly from the fact that quantum measurements cannot increase trace distance. Details are in Appendix B. We are not claiming that the identity of an optimal detector remains unchanged when it is reoptimized after each perturbation.
Accessible information and observer consensus
The fixed-pointer inclusion in Eq. (13) concerns exact states for one partition. The errors above also give one-sided statements for approximate records. Suppose the specified local measurements include a complete rank-one pointer on . Let be the calibrated outcome at fragment and . Write , with , for binary entropy. For pointer labels, Fano’s inequality [37] yields
If is diagonal in , then . For , may be replaced by in the lower bound. The qualification about the full pointer matters: an incomplete tag as in Eq. (50) of Appendix D is not enough to certify SQD of the entire system.
There is also a geometric implication. Let denote the global states for which every marginal has the ideal form of Eq. (9). Every ideal complete-pointer record belongs to this family, so Theorem 2 gives
The converse fails for the coherent GHZ state of Appendix D: its SQD marginals are ideal, but the unread measurement gives , , and . Conversely, in that appendix even when the fragments are conditionally entangled, so exact sharing does not require SBS.
The question then shifts from which event is shared for a fixed cover to how often different covers reveal it. Following the distinction between redundancy (existence of a suitable partition) and consensus (success under a distribution of partitions) [38, 39], let be a family of partitions of the environment into disjoint equally sized fragments, sampled from a probability distribution . For an allowed accessible-information deficit , define the success probability
For the same complete pointer and tested full state, fix a nonnegative error tolerance and require local readouts to satisfy . Define
Readouts may be optimized separately for each partition, but the pointer, full state, and sampling distribution remain fixed.
Proposition 2 (Shared-record success bounds accessible-information consensus). Let be a complete rank-one eigenbasis of with labels, and let . With
any distribution over the partitions satisfies
The same one-way implication holds for the largest successful partition size and for any prescribed probability threshold.
The proof in Appendix C applies the Fano bound to each successful partition before averaging over . It uses no conditional independence. The reverse implication is false: the GHZ state of Appendix D has full accessible pointer information on its natural partition, yet an unread measurement disturbs its global coherence. This is why a fixed-partition hierarchy must not be confused with a random-partition consensus score. A partition engineered to expose the available records may be rare under a different sampling distribution [38, 39].
Finally, the number of possible shared labels is not the same as the information usually carried by them. If the finest shared-sector decomposition contains occupied labels, a uniformly distributed sector label carries bits of Shannon information. The actual average information is the Shannon entropy . A record with many possible outcomes may carry little information if one outcome occurs almost every time. For imperfect readouts it is therefore useful to report the outcome probabilities together with and [40, 41].
Discussion and outlook
The state-level results obtained here have three complementary parts. Theorem 1 gives a unique finest locally readable record-sector structure without prescribing a pointer. Proposition 1 locates complete-pointer sharing relative to SQD and SBS by adding a full-state nondisturbance condition. Theorem 2 bounds approximate sharing through disturbance and disagreement. These conclusions neither require conditional fragment independence nor identify accessible information with nondisturbing readout. In particular, the loss of exact sharing under full-rank noise need not imply an abrupt loss of practical reliability, and Proposition 2 relates reliable records to consensus only in one direction.
These statements concern the properties of a state at a given time, not the interaction that prepared it. In the usual decoherence and quantum Darwinism picture, a system pointer becomes correlated with the environment [4, 6, 8, 42], schematically as
Tracing out inaccessible degrees of freedom may suppress interference between pointer values, but this does not by itself ensure that the fragments available to separate observers can distinguish those values. Suppression of pointer coherence and local readability are distinct requirements [43, 44, 45].
A useful bridge from this physical picture to approximate records can be stated without specifying a dynamical model. Let be the state of the system and its accessible fragments, and define the system-pointer dephasing map by
The residual pointer coherence is
For orthogonal fragment readout projectors , set and . The probability that at least one fragment fails to identify the pointer after dephasing is . If , then is an ideal shared-record state for these readouts, and
Thus small residual pointer coherence and small fragment-identification error suffice to place the state close to an ideal shared record, without requiring conditional independence. The square-root term accounts for the possible disturbance caused by projecting the fragments onto the matching-outcome subspace. A proof is given in Appendix B.
If denotes the pointer probabilities and the conditional state of fragment , the mean local identification error is
The union bound gives
regardless of correlations or entanglement between fragments. This estimate and Eq. (30) complement, rather than replace, the fixed-readout bound in Eq. (20): the former uses pointer coherence and fragment-identification errors, whereas the latter uses the disturbance of the complete local readout and its disagreement probability . These are state-level sufficient conditions, not equations governing the time evolution of records.
Neither bound determines how quickly records form. In particular, decoherence of the system pointer alone need not make information accessible to individual fragments [43, 44, 46, 47, 48]. In environments with memory, the quality of a record can vary with time [49, 50, 51, 52]. For fixed detectors, and can track such changes, but nondisturbance at one instant does not establish nondisturbance in a multitime experiment [53, 54].
A natural direction for future work is therefore a dynamical theory of shared-record formation under specified system–environment interactions [36, 55, 56, 57, 58, 59]. Such a study could track , , , and for physically motivated fragment partitions, determine when local readability develops relative to decoherence, and investigate record-formation rates, persistence, and memory-dependent effects. These are questions about the production and temporal stability of records, distinct from the state-level existence and robustness criteria established here. The present results provide a quantitative benchmark for addressing them.
Acknowledgments
AI-assisted scientific reasoning, literature organization and drafting were used in preparing this manuscript with OpenAI GPT-6. The author is responsible for all the contents.
Proof of the shared-sector structure
Proof of the spectrum sharing characterization
Let the local projective measurements satisfy Eq. (2). Since , sequential unread measurements in the different laboratories leave unchanged. Their projectors commute because the laboratories are disjoint, so the resulting state is
If the outcome string has two unequal entries, its projector is bounded by the product of the corresponding two local projectors. The probability of that string therefore vanishes by the second condition of Eq. (2). A positive density matrix with zero probability on a projector has no support on that projector. Only the strings remain, which gives the block decomposition in Eq. (4).
For each occupied label , put
The are orthogonal as varies, and . This establishes the nontrivial spectrum sharing structure whenever the original record is nontrivial.
Conversely, suppose has a nontrivial decomposition satisfying both Eqs. (4) and (5). Let be the orthogonal projector onto . Because these subspaces are pairwise orthogonal, is also a projector. Fix one label and define a complete local readout by
The arbitrary assignment of unused local subspaces cannot affect any outcome on the support of . For all ,
It follows that every local measurement commutes with the full and that all observers obtain the same label with certainty. Because at least two are positive, the resulting record is nontrivial. This completes the proof of the principal equivalence without using operator algebras.
Uniqueness of the finest shared-sector decomposition
We now prove the uniqueness claim without introducing an operator algebra. Let project onto . Any local projector of an exact shared readout commutes with and therefore with . By the first part of the proof, the occupied part of the state lies entirely in the matching-outcome subspaces . Hence
Thus every observer implements the same projector on the occupied state space, even if their physical projectors differ on subspaces that are never occupied. The are orthogonal projections summing to .
Consider two possibly different exact shared readouts, with projectors and on the support. Choose different observers and . According to Eq. (36), can be represented on the support by a local projector in laboratory , and by one in laboratory . Local projectors acting on disjoint tensor factors commute. Since both commute with , we get
This argument applies to every pair of projectors belonging to any exact shared readouts, not just to projectors from a single measurement.
A family of commuting Hermitian projectors on a finite-dimensional space has a common orthogonal eigenspace decomposition. Let be its nonzero joint-eigenspace projectors on . These subspaces are uniquely determined by the outcomes of all exact shared readouts. There are at most nonzero subspaces. Consequently a finite list of shared projectors, call it , suffices to distinguish all the joint eigenspaces. Each is the projector onto one definite binary pattern of eigenvalues of this finite list.
It remains to show that this common refinement can itself be read locally. For each , select a local projector in each laboratory whose restriction to is , as in Eq. (36). Define the Hermitian local observable
Each commutes with because every summand does. On its eigenvalue is the binary code of the joint eigenvalues of . Different occupied sectors have different codes, and the code is identical for all observers. Measuring the spectral projectors of therefore identifies the same sector in every laboratory without disturbing . Any eigenvalues occurring only outside can be assigned to an arbitrary existing outcome. This produces a complete projective readout of the finest shared sectors.
Every projector of every other exact shared readout is a sum of some , by their construction as common joint eigenspaces. Thus all such records are coarse-grainings of this unique finest one, up to relabeling. Since all commute with , the sectors give Eq. (6).
Proof of the full-rank corollary
If is full rank, every nonzero positive projector has . Suppose two different labels and both have nonzero probability in an exact shared readout. Choose any two distinct laboratories and . Then and are nonzero, so their product on the full Hilbert space is a nonzero projector. Consequently
contrary to the perfect-agreement requirement in Eq. (2). This proves Corollary 1.
Robustness and branching-state estimates
Fixed-readout bounds and stability
Let and write . The equality of the probabilities follows because is a sum of complete measurement blocks, so does not change its expectation. Also . If , define
The difference is block diagonal with respect to . On the first block it is negative and has trace . On the second it is positive and has trace . Thus . The triangle inequality gives
This explains why the upper error is linear in : the pinched state has no off-diagonal terms connecting the matching and mismatching outcomes. If we instead projected an arbitrary coherent state directly, a standard gentle-measurement estimate would yield a weaker square-root correction.
For the lower bound, take any . By definition . Contractivity of the trace norm under the quantum channel gives
Furthermore . Using the variational characterization of the trace distance for this binary test gives . Taking the infimum over proves Eq. (20). The equivalence between and follows directly from and support on .
For the noisy-bit family, is diagonal in the local measurement basis, so . The states and each occur with probability , giving . The normalized matching-state projection in Eq. (40) has trace-norm distance from . The lower bound then proves . This also shows that the factor multiplying in the general upper bound cannot be made smaller. At , the state has two finest shared labels. For every , no nontrivial shared label exists.
To see perturbative stability, suppose for two states tested with the same local projectors. Then
The second inequality uses the fact that is a projector and has trace zero. The combined upper-bound expression therefore changes by at most .
Pointer dephasing and fragment-readout estimate
Let and let be the projector onto identical pointer and fragment readouts. Because is block diagonal in the pointer basis and includes that basis, the normalized projected state has no coherence between distinct shared pointer values. It therefore has the sector structure required of an ideal shared record.
The gentle measurement inequality [60] for the projector gives . Normalizing the successful outcome changes its trace norm by exactly :
Together with the definition , the triangle inequality yields Eq. (30).
For Eq. (31), condition on pointer value . If all observed fragments return the correct , then the joint matching projector succeeds. The event that this does not happen is the union of the individual failure events. Since the local projectors act on different fragments, their outcomes have an ordinary joint probability distribution, whether or not the fragment state factorizes. Summing the usual union bound over , weighted by , gives .
Fixed-pointer comparisons and consensus bounds
From the entropic SQD criterion to the pairwise form
We spell out why the diagonal-pointer hypothesis in Sec. III is needed. Let be a complete rank-one eigenbasis of and consider one fragment . Dephasing in this basis gives . Its mutual information is precisely . As dephasing leaves both and unchanged under the eigenbasis assumption,
The entropy increase under projective dephasing vanishes only when . Moreover, if and only if the occupied conditional fragment states have mutually orthogonal supports: a classical label can then be identified perfectly from that fragment. Since here, the equalities in Eq. (8) therefore imply Eq. (9). Conversely, that pairwise form makes the pointer perfectly readable and saturates all three equalities. Without a diagonal pointer marginal, the entropic equalities alone need not specify a complete readout of .
Pointer-overlap identity and conditional independence
Fix the complete rank-one pointer basis of Sec. III, with diagonal in , and retain only labels with . Suppose first that has the shared pointer record . By nondisturbance of the local system pointer measurement, the complete state is block diagonal:
The fragment readout projectors identify with certainty, so the support of lies in the range of . The projectors have orthogonal ranges for different . Thus every has the ideal form in Eq. (9), and hence is SQD with the common pointer . This proves .
Conversely, suppose . Nondisturbance of the system pointer again gives Eq. (45). For each , SQD gives mutually orthogonal supports . The positivity of implies that it has no component outside the range of on fragment : indeed
(The implication follows because a positive operator has zero weight in a subspace only when its rows and columns vanish there.) As this holds for all , . Reading the indicated local support projectors then perfectly agrees with the system pointer and commutes with the entire state. Therefore , proving Eq. (11) and the global form Eq. (12).
Within this class, conditional fragment independence is equivalent to for all . The identity
follows by expanding the logarithm of a tensor product. Positivity of relative entropy, with equality only for identical states, then establishes Eq. (14) and the equivalence of with SBS within . In particular, the proof makes no assumption of pairwise or higher-order independence before setting .
Derivation of the geometric and accessible-information bounds
Let the fixed readout include the complete rank-one system pointer , and let . Every state satisfies and is therefore of the form in Eq. (12). In particular, , where the SQD family allows the choice of orthogonal fragment support spaces to vary. Taking the infimum over the larger family gives . The upper bound in Eq. (23) is Theorem 2.
For any local fragment measurement whose calibrated result has error rate , introduce the binary indicator of the error event. Knowing and this indicator leaves at most further bits to identify an incorrectly reported value of . Therefore the elementary Fano argument yields
Subtracting from gives the second inequality in Eq. (22). The first holds because the accessible mutual information maximizes over local measurements. Measuring on and the displayed on its fragment is one allowed choice. Moreover, the event is contained in the event that not all observers agree, so .
If is diagonal in the rank-one pointer basis, . The function is increasing on . If for any set of candidate local readouts on a partition , then , and therefore
Thus, for each partition separately, shared-record success implies accessible-information success with the indicated threshold. Taking probabilities over any gives Eq. (27). Taking maximum partition sizes gives the corresponding redundancy implication. The reverse is false because controlling the measured information does not control the global coherence removed by unread measurements, as the GHZ example shows.
Counterexamples for the fixed-pointer hierarchy
The five examples below realize the regions marked A–E in Figure 1. Unless otherwise stated, the tested partition consists of the designated system and two disjoint fragments .
A: A shared event need not encode all of . Let contain a label qubit and an independent maximally mixed qubit . With two label fragments, consider
The label is shared without disturbance, whereas the fragments contain no information about :
Consequently , but QD and SQD fail for the complete pointer of .
B: QD without a shared fact. Let , let denote a Bell projector, and take
Each fragment satisfies , but . Thus QD holds whereas SQD fails. The complete state is pure and admits no nontrivial exact shared event.
C: SQD without global nondisturbance. For , take the coherent GHZ state
Each marginal is a classical copy of the system label:
Yet an unread pointer measurement removes the coherence. Hence , and the complete pure state has no nontrivial exact shared event. Tracing out first produces a different tested state with a shared classical bit, illustrating the importance of the system boundary.
D: An exact pointer record need not be SBS. Let , where stores the pointer label and is a quantum register. Put on for , and define
Each reads without disturbance, so
For , the conditional state on is entangled, and . For , it is SBS. The conditional-correlation measure in Eq. (14) gives
which ranges from zero to two bits. Each conditional state is pure, while its one-qubit reductions have eigenvalues and . Thus exact pointer sharing can coexist with nonzero conditional entanglement.
E: A shared fact and QD need not imply SQD. Add a copied classical tag to the two Bell pairs of region B. Let and , with
The tag is an exact shared event, while and . Hence but does not satisfy SQD for the complete pointer.
References
References
- [1]W. H. Zurek, “Pointer basis of quantum apparatus: Into what mixture does the wave packet collapse?” Phys. Rev. D 24, 1516 (1981).DOI
- [2]W. H. Zurek, “Environment-induced superselection rules,” Phys. Rev. D 26, 1862 (1982).DOI
- [3]E. Joos and H. D. Zeh, “The emergence of classical properties through interaction with the environment,” Z. Phys. B Condens. Matter 59, 23–43 (1985).DOI
- [4]W. H. Zurek, “Decoherence, einselection, and the quantum origins of the classical,” Rev. Mod. Phys. 75, 715 (2003).DOI
- [5]M. Schlosshauer, “Decoherence, the measurement problem, and interpretations of quantum mechanics,” Rev. Mod. Phys. 76, 1267–1305 (2005).DOI
- [6]H. Ollivier, D. Poulin, and W. H. Zurek, “Objective properties from subjective quantum states: Environment as a witness,” Phys. Rev. Lett. 93, 220401 (2004).DOI
- [7]R. Blume-Kohout and W. H. Zurek, “Quantum Darwinism: Entanglement, branches, and the emergent classicality of redundantly stored quantum information,” Phys. Rev. A 73, 062310 (2006).
- [8]W. H. Zurek, “Quantum Darwinism,” Nat. Phys. 5, 181 (2009).
- [9]F. G. S. L. Brandão, M. Piani, and P. Horodecki, “Generic emergence of classical features in quantum Darwinism,” Nat. Commun. 6, 7908 (2015).DOI
- [10]W. H. Zurek, “Quantum theory of the classical: Einselection, envariance, quantum Darwinism and extantons,” Entropy 24, 1520 (2022).DOI
- [11]R. Blume-Kohout and W. H. Zurek, “Quantum Darwinism in quantum Brownian motion,” Phys. Rev. Lett. 101, 240405 (2008).
- [12]C. J. Riedel and W. H. Zurek, “Quantum Darwinism in an everyday environment: Huge redundancy in scattered photons,” Phys. Rev. Lett. 105, 020404 (2010).DOI
- [13]M. Zwolak, C. J. Riedel, and W. H. Zurek, “Amplification, redundancy, and quantum Chernoff information,” Phys. Rev. Lett. 112, 140406 (2014).pubmed.ncbi.nlm.nih.gov/24765928
- [14]M. A. Ciampini, G. Pinna, P. Mataloni, and M. Paternostro, “Experimental signature of quantum Darwinism in photonic cluster states,” Phys. Rev. A 98, 020101 (2018).DOI
- [15]T. K. Unden, D. Louzon, M. Zwolak, W. H. Zurek, and F. Jelezko, “Revealing the emergence of classicality using nitrogen-vacancy centers,” Phys. Rev. Lett. 123, 140402 (2019).DOI
- [16]L. Henderson and V. Vedral, “Classical, quantum and total correlations,” J. Phys. A: Math. Gen. 34, 6899–6905 (2001).DOI
- [17]H. Ollivier and W. H. Zurek, “Quantum discord: A measure of the quantumness of correlations,” Phys. Rev. Lett. 88, 017901 (2001).pubmed.ncbi.nlm.nih.gov/11800986
- [18]M. Zwolak and W. H. Zurek, “Complementarity of quantum discord and classically accessible information,” Sci. Rep. 3, 1729 (2013).arxiv.org/abs/1303.4659
- [19]T. P. Le and A. Olaya-Castro, “Objectivity (or lack thereof): Comparison between predictions of quantum Darwinism and spectrum broadcast structure,” Phys. Rev. A 98, 032103 (2018).DOI
- [20]T. P. Le and A. Olaya-Castro, “Strong quantum Darwinism and strong independence are equivalent to spectrum broadcast structure,” Phys. Rev. Lett. 122, 010403 (2019).DOI
- [21]J. K. Korbicz, “Roads to objectivity: Quantum Darwinism, spectrum broadcast structures, and strong quantum Darwinism—a review,” Quantum 5, 571 (2021).DOI
- [22]J. K. Korbicz, P. Horodecki, and R. Horodecki, “Objectivity in a noisy photonic environment through quantum state information broadcasting,” Phys. Rev. Lett. 112, 120402 (2014).pubmed.ncbi.nlm.nih.gov/24724630
- [23]R. Horodecki, J. K. Korbicz, and P. Horodecki, “Quantum origins of objectivity,” Phys. Rev. A 91, 032122 (2015).DOI
- [24]J. K. Korbicz, E. A. Aguilar, P. Cwikliński, and P. Horodecki, “Generic appearance of objective results in quantum measurements,” Phys. Rev. A 96, 032124 (2017).DOI
- [25]T. P. Le and A. Olaya-Castro, “Witnessing non-objectivity in the framework of strong quantum Darwinism,” Quantum Sci. Technol. 5, 045012 (2020).DOI
- [26]A. Feller, B. Roussel, I. Frérot, and P. Degiovanni, “Comment on Strong quantum Darwinism and strong independence are equivalent to spectrum broadcast structure,” Phys. Rev. Lett. 126, 188901 (2021).DOI
- [27]N. Bohr, “Can quantum-mechanical description of physical reality be considered complete?” Phys. Rev. 48, 696 (1935).
- [28]C. Beny, A. Kempf, and D. W. Kribs, “Generalization of quantum error correction via the Heisenberg picture,” Phys. Rev. Lett. 98, 100502 (2007).pubmed.ncbi.nlm.nih.gov/17358519
- [29]C. Beny, A. Kempf, and D. W. Kribs, “Quantum error correction of observables,” Phys. Rev. A 76, 042303 (2007).DOI
- [30]Marin Girard, Gong Cheng, and ChunJun Cao, “Demystifying objectivity with operator algebra quantum error correction,” (2026), arXiv:2606.06588.arxiv.org/abs/2606.06588
- [31]C. M. Scandolo, R. Salazar, J. K. Korbicz, and P. Horodecki, “Universal structure of objective states in all fundamental causal theories,” Phys. Rev. Research 3, 033148 (2021).DOI
- [32]M. Piani, P. Horodecki, and R. Horodecki, “No-local-broadcasting theorem for multipartite quantum correlations,” Phys. Rev. Lett. 100, 090502 (2008).pubmed.ncbi.nlm.nih.gov/18352686
- [33]S. Luo, “Using measurement-induced disturbance to characterize correlations as classical or quantum,” Phys. Rev. A 77, 022301 (2008).DOI
- [34]H. Barnum, C. M. Caves, C. A. Fuchs, R. Jozsa, and B. Schumacher, “Noncommuting mixed states cannot be broadcast,” Phys. Rev. Lett. 76, 2818 (1996).pubmed.ncbi.nlm.nih.gov/10060796
- [35]A. S. Holevo, “Bounds for the quantity of information transmitted by a quantum communication channel,” Problems of Information Transmission 9, 177–183 (1973).
- [36]P. Mironowicz, J. K. Korbicz, and P. Horodecki, “Monitoring of the process of system information broadcasting in time,” Phys. Rev. Lett. 118, 150501 (2017).DOI
- [37]R. M. Fano, Transmission of Information: A Statistical Theory of Communication (MIT Press, Cambridge, MA, 1961).
- [38]D. A. Chisholm, L. Innocenti, and G. M. Palma, “The meaning of redundancy and consensus in quantum objectivity,” Quantum 7, 1074 (2023).DOI
- [39]D. A. Chisholm, L. Innocenti, and G. M. Palma, “Importance of using the averaged mutual information when quantifying quantum objectivity,” Phys. Rev. A 110, 012218 (2024).DOI
- [40]B. Groisman, S. Popescu, and A. Winter, “Quantum, classical, and total amount of correlations in a quantum state,” Phys. Rev. A 72, 032317 (2005).DOI
- [41]C. A. Fuchs and J. van de Graaf, “Cryptographic distinguishability measures for quantum-mechanical states,” IEEE Trans. Inf. Theory 45, 1216 (1999).DOI
- [42]M. Zwolak, C. J. Riedel, and W. H. Zurek, “Amplification, decoherence and the acquisition of information by spin environments,” Sci. Rep. 6, 25277 (2016).DOI
- [43]G. Garcia-Perez, D. A. Chisholm, M. A. C. Rossi, G. M. Palma, and S. Maniscalco, “Decoherence without entanglement and quantum Darwinism,” Phys. Rev. Research 2, 012061 (2020).
- [44]T.-H. Lee and J. K. Korbicz, “Complementarity between decoherence and information retrieval from the environment,” Phys. Rev. A 109, 032221 (2024).DOI
- [45]K. Roszak and J. K. Korbicz, “Entanglement and objectivity in pure dephasing models,” Phys. Rev. A 100, 062127 (2019).DOI
- [46]S. Lorenzo, M. Paternostro, and G. M. Palma, “Anti-Zeno-based dynamical control of the unfolding of quantum Darwinism,” Phys. Rev. Research 2, 013164 (2020).DOI
- [47]D. Girolami, A. Touil, B. Yan, S. Deffner, and W. H. Zurek, “Redundantly amplified information suppresses quantum correlations in many-body systems,” Phys. Rev. Lett. 129, 014041 (2022).DOI
- [48]J. Gu, “Hamiltonian thresholds for objective records,” Phys. Rev. A 114, L020201 (2026).DOI
- [49]C. J. Riedel, W. H. Zurek, and M. Zwolak, “The rise and fall of redundancy in decoherence and quantum Darwinism,” New J. Phys. 14, 083010 (2012).DOI
- [50]G. Pleasance and B. M. Garraway, “Application of quantum Darwinism to a structured environment,” Phys. Rev. A 96, 062105 (2017).DOI
- [51]A. Lampo, J. Tuziermski, M. Lewenstein, and J. K. Korbicz, “Objectivity in non-Markovian spin-boson model,” Phys. Rev. A 96, 012120 (2017).DOI
- [52]P. Mironowicz, P. Należyty, P. Horodecki, and J. K. Korbicz, “System information propagation for spin structures,” Phys. Rev. A 98, 022124 (2018).DOI
- [53]F. Sakuldee, P. Taranto, and S. Milz, “Connecting commutativity and classicality for multimtime quantum processes,” Phys. Rev. A 106, 022416 (2022).DOI
- [54]P. Strasberg, “Classicality with(out) decoherence: Concepts, relation to Markovianity, and a random matrix theory approach,” SciPost Phys. 15, 024 (2023).DOI
- [55]J. Tuziermski and J. K. Korbicz, “Dynamical objectivity in quantum Brownian motion,” Europhys. Lett. 112, 40008 (2015).
- [56]J. Tuziermski and J. K. Korbicz, “Analytical studies of spectrum broadcast structures in quantum Brownian motion,” J. Phys. A 49, 445301 (2016).
- [57]S. Campbell, B. Çakmak, O. E. Müstecaplıoğlu, M. Paternostro, and B. Vacchini, “Collisional unfolding of quantum Darwinism,” Phys. Rev. A 99, 042103 (2019).DOI
- [58]E. Doucet and S. Deffner, “Classifying two-body Hamiltonians for quantum Darwinism,” Phys. Rev. X 14, 041064 (2024).DOI
- [59]B. Ferte and X. Cao, “Solvable model of quantum-Darwinism-encoding transitions,” Phys. Rev. Lett. 132, 110201 (2024).DOI
- [60]A. Winter, “Coding theorem and strong converse for quantum channels,” IEEE Trans. Inf. Theory 45, 2481–2485 (1999).DOI
