Density Matrices, Ontology, and Coherence: Why Penrose Was Right, Decoherence Was Insufficient, and Type-III QFT Forces a Non-Strict Notion of State

Density Matrices, Ontology, and Coherence: Why Penrose Was Right, Decoherence Was Insufficient, and Type-III QFT Forces a Non-Strict Notion of State Andrei T. Patrascu FAST Foundation, Destin FL, 32541, USA email: andrei.patr[email protected] Density matrices play a central role in quantum theory, yet their ontological status has long been contested. Penrose famously argued that density matrices are fundamentally inadequate, as they conflate epistemic mixtures with physically real superpositions, and therefore fail to represent what truly exists. Decoherence-based approaches, while explaining the dynamical suppression of interference and the emergence of effective classicality, do not resolve this tension: they leave improper mixtures intact and do not restore a well-defined ontology. In this work we show that these difficulties are neither interpretational accidents nor signals of incomplete dynamics, but structural consequences of the algebraic nature of quantum theory itself. In local quantum field theory, observable algebras are of Type III, admitting neither traces nor density matrices in the classical sense, rendering any ontology based on global density operators inconsistent from the outset. We reformulate Penrose’s critique in algebraic and categorical terms and argue that what fails is not quantum mechanics, but the assumption of a strict, globally factorizable state. We propose a coherence-first framework in which quantum states are treated as local algebraic states organized over a category of contexts and glued via non-strict descent. Failures of strict global extension appear as controlled higher-categorical coherence defects rather than as paradoxes or information loss. Within this framework, density matrices are reinterpreted as context-dependent coordinatizations valid only in Type I regimes, not as fundamental ontic objects. We show that this approach absorbs the valid core of Penrose’s critique without invoking objective collapse, resolves the Type III “no density matrix” problem structurally, and clarifies why decoherence alone cannot address the ontological questions at stake. I. INTRODUCTION: THE DENSITY MATRIX AS A FAULT LINE A. The long-standing tension The density matrix is one of the most successful pieces of quantum technology: it is compact, operationally complete for laboratory predictions, and indispensable for open-system dynamics, quantum information, and statistical mechanics. Yet precisely because it is so effective, it has become a conceptual fault line. The same mathematical object is routinely asked to play two roles that, from an ontological perspective, need not be compatible. This tension appears in its simplest form as the distinction between proper (epistemic) mixtures and improper mixtures produced by entanglement and restriction to a subsystem [1–3]. In the proper reading, a mixed state represents ignorance: the system is really in one of several pure states, but we do not know which. The density matrix then summarizes a classical probability distribution over preparations. In the improper reading, the mixed state is not a reflection of ignorance at all: it arises even when the global state is pure, because the subsystem description is obtained by restricting attention to a subalgebra or by tracing out degrees of freedom. In that case, the reduced state is fixed by correlations, not by an underlying classical lottery. The same density operator thus admits two incompatible narratives: one in which it encodes incomplete knowledge about a definite underlying situation, and one in which it encodes objective nonseparability without any fact of the matter about “which pure state” the subsystem is in [2]. This ambiguity is not merely interpretational. It is tightly entangled with the measurement problem as it is usually formulated. In the textbook measurement story, a unitary interaction correlates a system with an apparatus (and, realistically, an environment), producing a global entangled state. The reduced state of the apparatus then becomes mixed; one often appeals to decoherence to explain why interference between outcome branches becomes inaccessible for all practical purposes [ 4 – 6 ]. However, decoherence does not by itself transform an improper mixture into a proper one: it explains the dynamical stability of certain effective classical descriptions (pointer structures, robust records), but it does not select a unique realized outcome, nor does it restore an underlying “ontic” pure state for the subsystem [ 5 , 6 ]. The density matrix remains the same kind of object as before—and the proper/improper tension remains. This is the sense in which the density matrix functions as a fault line: it is simultaneously (i) the correct operational interface for predictions within a context, and (ii) the locus where demands for a 2 context-independent ontology become most strained. Penrose sharpened this point by arguing that density matrices are not merely incomplete descriptions but are, in a deep sense, the wrong kind of object to carry ontological weight: they can blur the distinction between epistemic uncertainty and objective indefiniteness, and they inherit a non-uniqueness of ensemble decompositions that frustrates any attempt to read them as “what is really there” [ 7 ]. While one may disagree with Penrose’s proposed remedy (objective reduction), the underlying diagnosis—that density matrices are structurally ill-suited as fundamental ontic descriptors—is difficult to dismiss and becomes even more acute in the presence of entanglement and field-theoretic locality [5, 7]. The persistence of this tension across otherwise very different research programs is a sign that something structural is being misidentified. The recurring pattern is that we expect a single, globally factorizable state-description to exist in the background, and we treat density matrices as if they were pieces of that global ontology. But the operational role of the density matrix is intrinsically contextual (relative to a chosen algebra of observables, a coarse-graining, a decoder, or a subsystem split), and its “reduced” character is not simply a matter of forgetting information. The fact that the same formal object can be read as either ignorance or entanglement is thus not an accident of language: it is an indicator that the concept of “state of a part” is more subtle than the Type I, factorized Hilbert-space intuition suggests [ 2 , 6 ]. In the remainder of this paper we will argue that this subtlety is forced on us by the algebraic structure of quantum theory in its natural habitat (local QFT), and that the appropriate replacement is not a collapse postulate but a coherence-first notion of state as non-strict descent data across contexts. B. Penrose’s intervention Penrose’s critique enters precisely at the point where the density-matrix formalism is most frequently overextended: the temptation to treat ρ as a fundamental representation of what exists. In Penrose’s view, the density operator is an excellent operational device, but it is ontologically ambiguous in a way that cannot be cured by improved bookkeeping or by “interpretational” rephrasing alone [ 7 , 8 ]. The reason is structural: a given density matrix admits infinitely many distinct ensemble decompositions ρ=X i pi|ψiihψi|, pi≥0,X i pi= 1,(1) and, except in special cases, there is no physically privileged choice among them. If one insists that ρ describes a situation in which the system is really in one definite pure state |ψii but we do not know which, then Eq. (1) would be read as revealing the underlying ontology. Penrose’s point is that this reading cannot be made consistent in general: the decomposition is not unique, so the alleged underlying “real pure state” is not well-defined by ρ itself [ 2 , 8 ]. In other words, the density matrix does not carry its own ontological semantics. It packages correct statistics while underdetermining any claim about what is the case. This observation interacts sharply with the standard measurement narrative. In unitary quantum mechanics, measurement is modeled as entangling system and apparatus (and typically an environment). The reduced state of the apparatus becomes mixed and often approximately diagonal in a pointer basis due to decoherence. But the resulting mixed state is an improper mixture: it arises from entanglement and restriction, not from ignorance about an underlying definite outcome [ 2 , 5 , 6 ]. Penrose’s critique is that appealing to the reduced density matrix as if it were an ontic description of the apparatus is therefore a sleight of hand: it replaces the question “what is the actual outcome?” by a statistical object that is compatible with many inequivalent ontological stories. Decoherence can explain why interference terms become dynamically suppressed for relevant observables, but it does not, on its own, turn an improper mixture into a proper one [ 5 , 6 ]. Thus the ontological gap persists even in the most sophisticated open-systems treatments. Penrose’s intervention is often summarized as an argument for objective collapse, but the part of his reasoning that is most relevant for the present paper is independent of that proposed remedy. The core claim is that the density matrix is not merely “incomplete knowledge”; rather, it is a contextual summary of expectations that does not define a unique underlying state of affairs. Put differently: Penrose identifies a mismatch between the operational adequacy of ρ and the ontological demands often placed upon it [ 7 ]. This mismatch is not cured by choosing a different interpretation, because it is rooted in the way density matrices are constructed: by coarse-graining, by restriction to a subsystem, and by tracing out degrees of freedom. These operations presuppose a subsystem structure and a notion of “discarding” that is already 3 nontrivial in quantum theory, and becomes sharply problematic in quantum field theory and gravity where the very existence of a global trace or a tensor factorization cannot be taken for granted [9, 11]. In this sense, Penrose’s critique targets structure rather than interpretation. It is a challenge to the assumption that the state of a part is a locally well-defined ontic object that can be encoded by a density operator, independent of the context in which that part is identified and probed. His proposed resolution is to modify dynamics so that one outcome is objectively selected, motivated in part by gravitational considerations [ 7 ]. Our aim in this paper is different: we will argue that the structural core of Penrose’s critique is correct, but that it does not force objective collapse. Instead, it points to a misidentification of what should count as a “state” in regimes where global factorization fails. We will show that once states are formulated as algebraic objects organized over contexts and glued by (generally non-strict) descent, the ontological ambiguity highlighted by Penrose becomes a controlled coherence phenomenon rather than a crisis demanding new non-unitary dynamics. C. The overlooked fact The debates around density matrices, mixtures, and measurement are often conducted as if the densityoperator formalism were universally available whenever one wishes to “describe a subsystem.” In the natural habitat of relativistic quantum physics, this presumption fails at the most basic structural level. The local observable algebras of relativistic quantum field theory are (under very general and physically standard assumptions) not Type I; rather, they are typically Type III von Neumann algebras [ 9 – 11 ]. This is not an esoteric classification detail. It is the mathematical encoding of the fact that the naive Hilbert-space picture of subsystems and reduced density matrices is not the correct starting point for local QFT. To state the point sharply: in the Type I setting, normal states can be represented by trace-class operators ρ via ωρ ( A ) = Tr ( ρA ), and reduced states are obtained by partial trace. In the Type III setting, there is no faithful normal trace on the algebra, and the standard notion of a density matrix for a local region is not available. One can still speak of states as positive normalized linear functionals ω : A → C , but one cannot, in general, encode them by density operators on a factorized Hilbert space associated to the region [ 9 , 10 ]. Put plainly: the object whose ontological status we argue about in the Type I discussion may not even exist in the fundamental local formulation of QFT. This structural fact has immediate consequences for how one should interpret “subsystems” and “reductions.” In gauge theories and in gravity-like settings, physical degrees of freedom are constrained, and the Hilbert space does not factorize cleanly across spatial regions; even in QFT without gauge constraints, the Reeh–Schlieder property and related results reflect a deep entanglement structure of the vacuum that obstructs naive separability [ 9 , 11 ]. In the algebraic language, the correct notion of “localization” is the assignment of an algebra A ( O )to each spacetime region O , together with isotony and locality properties of the net. The fundamental objects are the algebras and their states, not density matrices and partial traces. When one insists on describing a region by a density operator, one is already importing additional Type I structure that is not intrinsic to the local theory. One may object that in practice one often introduces regulators, lattices, split-property approximations, or “buffer regions” that allow an approximate factorization and hence an approximate density-matrix description. Such constructions can be extremely useful, but their very role is to approximate Type III behaviour by Type I models. They do not license the claim that density matrices exist fundamentally; rather, they underline that density matrices are effective coordinatizations valid only after auxiliary choices and idealizations [ 9 , 10 ]. If the conceptual problem is whether density matrices can serve as ontic descriptors, then the Type III fact tells us that the answer must be negative in the regimes where quantum field theory is expected to be fundamental. This is the overlooked pivot on which the present paper turns: if local QFT is Type III, then the usual demand that “the subsystem has a density matrix” is not merely hard to satisfy—it is ill-posed as a fundamental requirement. Consequently, any proposed resolution of the measurement/ontology tension that presumes the universal availability of density operators and partial traces is already inconsistent with the structure of local QFT. In particular, neither a “better interpretation” of density matrices nor a purely decoherence-based story can, by itself, address the Type III obstruction, because both presuppose the very objects (traces, reduced density operators) whose absence is the structural content of Type III locality [ 5 , 6 ]. Penrose’s critique of the ontological inadequacy of density matrices therefore gains an unexpected reinforcement: in the field-theoretic setting, the density matrix is not only ontologically ambiguous, it is not even a natural fundamental object. 4 The remainder of the paper develops a response that takes this overlooked fact seriously. We will treat states as algebraic objects organized over a category of contexts and glued by descent. In this formulation, density matrices reappear only as coordinate representations in Type I regimes or effective strictifications, while the Type III case is handled intrinsically without forcing a trace-based ontology. D. Thesis of the paper We can now state the thesis in a form that is both sharp and structurally testable. Thesis. The foundational difficulty is not that density matrices provide an incomplete description of an underlying, globally well-defined quantum reality. Rather, the difficulty is that the very use of density matrices as fundamental ontic objects presupposes a strict global ontology—a globally factorizable state space equipped with traces and partial traces—that quantum theory, and in particular local Type III quantum field theory, does not admit [ 9 – 11 ]. The proper/improper mixture ambiguity emphasized by Penrose is therefore not an interpretational blemish to be removed by selecting a narrative, but a symptom of a deeper category error: confusing context-dependent state data with globally defined ontic structure [2, 7]. Programmatic claim. Once states are formulated as algebraic objects organized over contexts, the correct notion of “globality” is not a single density operator on a globally factorized Hilbert space, but agluing class of compatible local states together with their transition data. In general this gluing is non-strict: compatibility holds up to equivalence and may carry controlled coherence defects. These defects are not paradoxes nor information loss; they are the intrinsic data of a higher-categorical (non-strict) descent structure. In this setting, density matrices are reinterpreted as coordinate representations that exist only in Type I (or effectively Type I) regimes, i.e. when the descent data strictifies and a trace-class presentation becomes available. Consequences. This viewpoint has three immediate consequences that organize the remainder of the paper. 1. The valid core of Penrose’s critique is retained: density matrices do not carry unambiguous ontological content. However, the conclusion that one must introduce objective collapse does not follow; the correct response is to revise the notion of state/globality rather than to modify unitary dynamics [5–7]. 2. Decoherence is reclassified: it explains the dynamical emergence of stable effective classical subalgebras and the practical suppression of interference, but it does not and cannot produce a fundamental density-matrix ontology, especially in Type III settings where density matrices are not intrinsic objects [5, 6, 9]. 3. The Type III “no density matrix” fact is not an obstruction to be circumvented, but the structural evidence that the correct ontology is non-factorizable and must be expressed in terms of algebraic states and (generally non-strict) descent across contexts [10, 11]. Roadmap. After reviewing Penrose’s critique in algebraic terms and contrasting it with decoherencebased accounts, we emphasize the Type III obstruction as the decisive structural input. We then formulate a coherence-first framework in which local state assignments form descent data over a category of contexts, and in which failures of strictification are captured as controlled higher-categorical coherence defects. The resulting picture absorbs the ontological insight in Penrose’s intervention without invoking collapse, and resolves the Type III “no density matrix” problem intrinsically by replacing density operators with context-relative coordinatizations of algebraic state data. II. WHAT DENSITY MATRICES ASSUME (AND WHY THIS MATTERS) A. Density matrices as Type I objects The density operator formalism is not merely a convenient notation for quantum probabilities. It encodes a specific package of structural assumptions about what a “state” is and how subsystems are organized. These assumptions are essentially Type I in the sense of von Neumann algebra theory: they rely on traces, tensor-factorization, and partial traces as canonical operations. In this subsection we make these assumptions explicit, because later sections will show that each of them becomes non-canonical or fails outright in the Type III setting [9, 10]. 5 (i) Trace as a canonical pairing. In the textbook (Type I) setting one represents observables by bounded operators on a Hilbert space H , and represents normal states by positive trace-class operators ρ with unit trace. Expectation values are given by the trace pairing ωρ(A) = TrH(ρ A),(2) and the state space is identified with a convex set of density operators. This is more than an operational rule: it presupposes that there exists a preferred trace functional TrH ( · )against which all normal states can be expressed. In Type I factors this is natural and essentially unique up to normalization; it is not a generic property of von Neumann algebras [ 10 ]. The moment the trace ceases to be available or canonical, the ontology “a state is a density matrix” loses its meaning. (ii) Tensor factorization as subsystem structure. A second Type I assumption is that physical composition corresponds to a tensor product decomposition HAB ∼ =HA⊗ HB,B(HAB)∼ =B(HA)⊗ B(HB),(3) so that “subsystem” is represented by a factor in a global Hilbert space and, correspondingly, by an inclusion of operator algebras. This factorization is often treated as a kinematical axiom. But it is already a strong ontological commitment: it encodes the claim that the degrees of freedom of the whole are composed of independent degrees of freedom of the parts, in a way that is globally well-defined and independent of context. In gauge theories and in local QFT, this factorization becomes subtle or fails (even before one reaches gravity), and the correct notion of localization is algebraic rather than Hilbert-factorial [9, 11]. (iii) Partial trace as “discarding” and as canonical restriction. Given (3), the partial trace defines a canonical map from global states to reduced states: ρAB 7→ ρA:= TrB(ρAB), ρAB 7→ ρB:= TrA(ρAB).(4) Operationally, (4) reproduces the statistics of measurements on A alone. Ontologically, it is routinely read as “forgetting” or “discarding” the B -degrees of freedom while leaving the state of A intact. That reading is precisely where the Type I assumptions become hidden: (4) is only canonical once a preferred tensor-product split and a trace are given. Without a preferred factorization, “trace out B ” is not even a well-typed operation. Without a trace, it is not available. Thus the very mechanism by which density matrices are supposed to describe subsystems presupposes a global structure that QFT will not generally provide [9, 10]. (iv) Subsystem ontology as an implicit axiom. Taken together, (2) – (4) amount to an implicit axiom of subsystem ontology: that there exist local “parts” carrying their own state objects (density matrices) which can be obtained by canonical reduction from a global state, and which can be recombined into a global state by tensoring and conditioning. In classical probability this axiom is so natural that it is rarely stated; in quantum theory it already fails in the presence of entanglement if interpreted naively (proper vs. improper mixtures), and in local QFT it fails structurally because the local algebras are not Type I [ 2 , 9 ]. Penrose’s complaint can be read as an early warning that treating ρ as an ontic “state of the system” is overreach [ 7 ]. The Type III fact goes further: it tells us that the Type I package is not universally available, hence the overreach is not merely philosophical but mathematically ill-posed in the fundamental setting. Summary. The density-matrix formalism packages three canonicalities—trace, factorization, and partial trace—into a single object ρ . Whenever any of these canonicalities is absent, ρ ceases to be the right fundamental carrier of state information. The remainder of Section II makes the other hidden assumptions explicit; later sections show how a coherence-first, context-organized reformulation retains operational predictivity while avoiding the Type I ontological commitments. B. Ontological locality hidden in the formalism The Type I package reviewed in Section II A is often presented as a neutral operational framework. However, once one moves from prediction to ontology, it quietly imports a further assumption that is rarely stated explicitly: a notion of ontological locality (or, more precisely, ontological separability) according to which “the state of a region” is a well-defined piece of reality that exists independently of how the region is embedded into a larger whole. In this subsection we isolate how that assumption enters through two closely related habits: reading restriction as forgetting, and treating extension as a default possibility. 6 (i) “Restriction” as “forgetting”. In Type I quantum mechanics one often speaks as if the reduced density matrix ρA = TrB ( ρAB )were obtained by “discarding” system B while leaving system A otherwise unaffected. This language is harmless for operational bookkeeping, but it is ontologically loaded. It suggests that (a) there is a pre-existing global state ρAB , (b) A carries an intrinsic state, and (c) the operation of taking TrB merely removes information that is irrelevant to A rather than changing the very meaning of “state of A .” In classical probability, marginalization can often be read as forgetting because the underlying sample space factorizes and the marginal is literally a projection of a joint distribution. In quantum theory, the situation is different: the reduced state of A is not, in general, a projection of an underlying ontic state of A ; it is a context-relative summary of expectations for observables in B ( HA )induced by the global preparation [ 2 , 3 ]. The proper/improper mixture distinction is precisely the warning sign that “forgetting” is not the right ontology-level interpretation of restriction. Algebraically, the point becomes even sharper. The primitive notion of subsystem in local QFT is the inclusion of observable algebras A ( O1 ) ⊂ A ( O2 )for regions O1⊂ O2 . Restricting a state ω from A ( O2 ) to A(O1)is always well-defined as a map of functionals, ω7→ ω|A(O1).(5) But this restriction is not canonically representable by a density operator in Type III settings, and it should not be interpreted as discarding degrees of freedom in a factorized Hilbert space [ 9 , 10 ]. The “forgetting” picture presupposes ontological separability that is not provided by the algebraic structure. (ii) The extension problem and its hidden premise. A second place where ontological locality hides is in the extension mindset: given local states, one assumes there should exist a global state that “contains” them as marginals. Concretely, one asks: Given density matrices ρA and ρB (and perhaps additional marginals on overlaps), does there exist a global density matrix ρAB such that TrB(ρAB ) = ρAand TrA(ρAB) = ρB? This question is often treated as a technical constraint satisfaction problem. Conceptually, it is much more: it presumes that the correct ontological picture is a single global object ρAB from which local states are derived by forgetting. In other words, the extension problem is already phrased in the language of strict global ontology. If local states fail to extend, one is tempted to conclude that something is “missing” (hidden variables, collapse, etc.). But once Type III locality is taken seriously, the default expectation of a global density matrix is ill-typed: local state data need not arise from a trace-class global operator, and the subsystem splits that make partial trace meaningful may not exist intrinsically [9, 11]. From the coherence-first perspective developed later in this paper, the correct replacement of the extension problem is a descent question: given local state assignments to contexts (regions, algebras, or decoder choices) together with compatibility data on overlaps, does there exist a coherent gluing class, possibly non-strict, that realizes these locals as restrictions? The critical shift is that one no longer assumes a unique strict global object as the goal. Instead, one allows the possibility that the best global notion is an equivalence class of gluings with controlled coherence defects. In this setting, failure of strict extension is not evidence of incompleteness; it is evidence that ontological locality (in the sense of globally factorizable subsystems with intrinsic density matrices) is an extra assumption that does not hold generically. (iii) Why this matters for the Penrose/decoherence debate. Penrose’s critique can be re-read as targeting precisely this hidden ontological locality: if reduced density matrices are not unambiguous carriers of “what is real,” then the act of restriction cannot be interpreted as mere forgetting [ 2 , 7 ]. Decoherence, in turn, explains the stabilization of certain effective classical descriptions but does not supply the missing ontological separability; it produces robust approximate diagonality in a chosen pointer context, not a canonical global object whose marginals are the reduced states [ 5 , 6 ]. Thus the formalism’s hidden ontological locality is exactly where Penrose is right to be dissatisfied, and exactly where decoherence is structurally unable to deliver what his critique demands. Summary. The phrase “state of a subsystem” is ontologically innocent only if one assumes (i) a canonical notion of restriction as forgetting and (ii) the default existence of strict global extensions. Both assumptions are Type I artefacts. In Type III QFT they fail structurally, and in entangled Type I settings they already fail conceptually if read ontically. This motivates the replacement developed later: states as context-relative algebraic objects glued by (generally non-strict) descent, in which compatibility replaces separability and coherence defects replace the demand for strict global ontology. 7 C. The extension problem as a structural diagnostic The preceding subsection argued that “restriction as forgetting” and the default expectation of global extension are Type I artefacts that smuggle ontological separability into the formalism. Here we sharpen this point by treating the extension problem not as a nuisance but as a diagnostic tool: it tells us, with mathematical precision, when the assumption of a strict globally factorizable ontology is being imposed where it does not belong. (i) The basic extension question. In its simplest Type I form, the extension problem asks whether given local states can arise as marginals of a joint state. For two subsystems Aand Bthis becomes: Given ρAand ρB,does there exist ρAB ≥0,Tr(ρAB)=1such that TrB(ρAB) = ρA,TrA(ρAB) = ρB? (6) In more refined settings one prescribes additional marginals on overlaps (e.g. ρAC , ρBC for threeparty systems) and asks for a global ρABC consistent with them (the “quantum marginal” problem). Operationally, these are compatibility constraints on predicted statistics. Ontologically, however, (6) is a test of whether the local descriptions can be regarded as pieces of a single global state in the Type I sense. (ii) Why failure to extend is not ignorance. A widespread intuition is: if local states fail to extend, then we must be “missing information” about the joint system. This is the wrong inference. Missing information would correspond to multiple possible extensions (non-uniqueness), not to non-existence. Non-existence is a structural incompatibility: the local assignments cannot simultaneously be realized as marginals of any global density operator compatible with the assumed subsystem split. That incompatibility is not repaired by learning more; it is repaired only by changing the modeling assumptions that made (6) the relevant question. There are two conceptually distinct sources of failure. First, even in Type I quantum mechanics, entanglement and monogamy constraints imply that not all collections of local marginals can arise from a single global state; local consistency constraints are nontrivial, and certain sets of reduced states are simply incompatible with any joint ρ [ 3 ]. Second, and more importantly for this paper, the very typing of (6) presupposes canonical tensor factorization and partial trace. In local QFT and in constrained systems, the subsystem split may not exist intrinsically, and the relevant notion of locality is algebraic inclusion rather than Hilbert-space tensoring [ 9 , 11 ]. In that case, the demand for an extension as a density matrix is not merely too strong; it is conceptually misplaced. (iii) Extension as a test of factorizable ontology. To see why, it is useful to restate the extension problem in algebraic terms. Let AA and AB be commuting subalgebras of a global algebra A (in Type I one may take A = B ( HA⊗ HB )). A “local state” is a state on AA or AB . The extension problem is then: Given states ωA on AA and ωB on AB , does there exist a state ω on A such that ω|AA = ωA and ω|AB=ωB? When A is Type I and the inclusion is induced by a tensor product split, this reduces to (6) . In general, it is a question about whether local functionals are jointly realizable on a larger algebra. Failure of extension then indicates not ignorance but the breakdown of a specific ontological picture: that local states are independent pieces of reality that can be freely assembled. In the coherence-first reading developed later, the correct global object is not a single ρAB but a coherent gluing class; extension becomes a special case of descent strictification. (iv) Early sign of non-factorizable ontology. The key interpretive point is therefore: When extension fails, what fails is not our knowledge of a pre-existing global density matrix, but the assumption that such a globally factorizable density-matrix ontology exists at all. This is precisely the ontological tension Penrose is pointing to, expressed structurally rather than rhetorically [ 7 ]. Penrose emphasizes that ρ does not specify “what is really the case.” The extension diagnostic shows one concrete reason why: local statistical descriptions do not, in general, determine (nor even guarantee the existence of) a single global density operator consistent with a presumed subsystem ontology. In Type III QFT, the diagnostic becomes even more decisive: the absence of traces and canonical tensor factorization means that the density-matrix extension problem is not the correct fundamental question. The early warning sign is already visible in the Type I setting (entanglement and incompatible marginals), but the structural obstruction is enforced by Type III locality [9, 10]. (v) From extension to descent. This motivates the conceptual shift that the remainder of the paper develops. Instead of treating extension failure as a deficit to be repaired (by hidden variables, collapse, or 8 “more information”), we treat it as evidence that the correct ontology is non-factorizable. The appropriate replacement is not “find ρAB” but: Given local states on contexts and compatibility on overlaps, determine the coherent gluing class (if any), (7) allowing for non-strict descent and controlled coherence defects. In this reformulation, density matrices are recognized as coordinate representations that exist only when the underlying descent data strictifies into a Type I presentation; their absence in Type III is then no longer a problem but a structural guide. III. PENROSE’S CRITIQUE REWRITTEN ALGEBRAICALLY A. Proper vs. improper mixtures (clean formulation) Penrose’s central complaint about density matrices becomes substantially sharper once it is stated in the algebraic language in which “state” is primary and “density operator” is a contingent coordinatization. In this subsection we separate (i) algebraic states versus ensemble decompositions, and (ii) the non-uniqueness of decompositions as the precise locus of ontological ambiguity. (i) States are functionals; density matrices are representations. Let A be a C∗ -algebra (or a von Neumann algebra in the normal setting). A state is a positive normalized linear functional ω:A → C, ω(A∗A)≥0, ω(1)=1.(8) When A=B(H)is Type I and ωis normal, there exists a unique trace-class operator ρsuch that ω(A) = Tr(ρA),(9) and one identifies the state with its density operator. This identification is not intrinsic in general von Neumann algebraic settings, and it is exactly what breaks down for Type III local algebras [ 9 , 10 ]. The algebraic viewpoint therefore prevents the category error “ ρ is the state” from being imposed universally: ρis a coordinate presentation of ωthat exists only under additional Type I structure. (ii) Proper vs. improper mixtures in algebraic terms. In Type I quantum mechanics, one commonly writes a mixed state as an ensemble decomposition ρ=X i pi|ψiihψi|, pi≥0,X i pi= 1,(10) and interprets this either as: (P) Proper mixture (epistemic): the system is really in one of the pure states |ψii , chosen with classical probabilities pi;ρencodes ignorance about which. (I) Improper mixture (entanglement/restriction): ρ is the reduced state of a subsystem obtained by restriction/partial trace from a global pure state; there is no fact of the matter that the subsystem “is in” any particular |ψii. Algebraically, the distinction is not psychological but structural. In case (I) , one has a larger algebra AAB (or a larger Hilbert space), a global pure state Ωon AAB, and the local state on AAis the restriction ωA= Ω|AA.(11) In this situation, the mixedness of ωA is not due to ignorance about an underlying pure state on A , but due to the fact that the restriction of a pure global functional need not be pure on the subalgebra. This is the algebraic content of “improper mixture” [ 2 , 3 ]. Importantly, (11) makes sense even when AA is Type III and no density matrix exists: the restriction map is always defined at the level of functionals, whereas “partial trace” is not fundamental [9, 10]. (iii) Non-uniqueness of decompositions as ontological ambiguity. Penrose’s key move is to notice that (10) does not define an underlying reality unless a decomposition is physically privileged. But in general it is not: the same ρ admits infinitely many distinct ensemble decompositions, including decompositions into inequivalent families of pure states. This is not a technical nuisance; it is the precise mathematical content of the underdetermination Penrose points to [7, 8]. 9 A particularly clean statement of this non-uniqueness is given by the Hughston–Jozsa–Wootters (HJW) theorem: for a given density matrix ρ on HA , every ensemble decomposition of ρ can be realized by choosing a purification | Ψ i∈HA⊗HB and performing an appropriate measurement on the ancillary system B [ 12 ]. Operationally, this means that different decompositions correspond to different measurement contexts on a purifying system. Ontologically, it means there is no context-free sense in which ρ encodes “the” underlying pure state that is merely unknown. The ambiguity is not removable without adding extra structure (a preferred decomposition rule, a collapse postulate, hidden variables, etc.). This is the structural core that Penrose correctly identifies. (iv) The algebraic punchline. In the algebraic viewpoint, Penrose’s critique can be formulated as the following statement: The object ρ (when it exists) is at best a representation of a state functional ω relative to Type I structure, and its ensemble decompositions are not intrinsic. Therefore ρ cannot carry unambiguous ontological semantics. This formulation isolates what is structural (non-uniqueness of decompositions; restriction producing mixed local states; non-existence of ρ in Type III) from what is interpretational (how one narrates the meaning of a mixed state). It also prepares the central move of this paper: rather than enforcing a unique global ontology by objective collapse, we will treat “state” as context-organized algebraic data glued by (generally non-strict) descent, in which the non-uniqueness Penrose highlights becomes controlled coherence rather than pathology. B. What Penrose gets exactly right Having isolated the structural content of the proper/improper distinction and the non-uniqueness of ensemble decompositions, we can now state succinctly what Penrose gets exactly right. The point is not to endorse any specific dynamical mechanism he proposes (objective reduction), but to preserve the valid diagnostic content of his critique in a form compatible with algebraic QFT and, later, with a coherence-first descent reformulation. (i) Density matrices do not encode “what is”. The first correct point is that the density matrix, even when it exists, is not an ontic descriptor. In Type I quantum mechanics ρ is an efficient representative of a state functional ω via ω ( A ) = Tr ( ρA ), but it carries no intrinsic interpretation as “the underlying state of affairs” because its ensemble decompositions are non-unique. By the HJW theorem, different decompositions of the same ρ are in one-to-one correspondence with different measurement contexts on a purifying ancilla [ 12 ]. Thus the decomposition is not a latent property of the system described by ρ ; it is a relational feature of a larger experimental context. Penrose’s insistence that ρ cannot, by itself, encode “what is” is therefore mathematically well-founded [7, 8]. (ii) Reduced states do not determine global reality. Second, and more sharply, reduced states cannot be treated as locally complete descriptions of the world, because restriction is many-to-one: distinct global states can induce the same local state. In Type I language, many inequivalent joint states ρAB share the same marginal ρA = TrB ( ρAB ). In algebraic language, many inequivalent global state functionals Ω on AAB restrict to the same ωA = Ω |AA . This is not an accident; it is structurally inevitable because the restriction map forgets relational information between AA and its complement (or, more generally, between a context and the larger context in which it sits). Penrose is therefore correct to reject the idea that a reduced density matrix supplies a locally sufficient ontology: local statistics do not fix the global state of affairs [2, 7]. This point becomes even more forceful in local QFT. There the correct notion of localization is not a Hilbert factor but a net of subalgebras A ( O ), and local algebras are typically Type III [ 9 , 10 ]. The local state ωO (a functional on A ( O )) can be the restriction of many inequivalent global states, while the very attempt to represent ωO by a density operator is generally ill-posed. Thus, the claim “the reduced density matrix determines reality in the region” is not merely philosophically dubious; it is structurally incompatible with the field-theoretic notion of locality. (iii) Ontology cannot be statistical. Third, Penrose is correct that “statistical” objects cannot be promoted to ontology without additional structure. A density matrix is a convex object, and convex decompositions are not unique. If one treats the convex weights as reflecting classical ignorance, one implicitly assumes a hidden-variable story that selects a privileged decomposition or a privileged underlying sample space. But quantum theory provides no such canonical structure: the convex geometry of the state space admits many inequivalent extremal decompositions that correspond to distinct operational 16 •and allow non-strict gluing with controlled coherence defects rather than insisting on strict global sections. This is precisely the direction pursued in Sections VI–VII: the Type III obstruction is not circumvented but embraced as evidence for a coherence-first ontology. Summary. Type III locality makes the discussion non-negotiable: density matrices cannot be the fundamental ontic carriers of quantum state information even in principle. Consequently, Penrose’s ontological discomfort is not an optional interpretational stance; it is forced upon us by the structure of local quantum theory. The remaining question is how to retain operational predictivity and decoherence-based emergence of classical records while replacing density-matrix ontology with an intrinsically Type IIIcompatible notion of state. The answer proposed here is a context-organized, non-strict descent formulation in which density matrices appear only as effective coordinatizations when strictification is available. C. Consequence We can now state the consequence in the sharpest possible form. Claim. Any ontology that takes global density matrices (and their reductions by partial trace) as fundamental primitives is incompatible with local quantum field theory as such. This is not an attack on the practical usefulness of density matrices. It is a statement about foundational typing. The argument is short: 1. In local QFT, localization is expressed by a net of von Neumann algebras O 7→ A ( O )and states are positive normalized linear functionals on these algebras [9]. 2. The local algebras A ( O )are typically Type III and therefore admit no trace suitable to represent normal states by density operators [10]. 3. The subsystem split required to define a canonical partial trace is not intrinsic: locality is algebraic inclusion, not Hilbert-space factorization, and the correct notion of reduction is restriction of functionals [9, 11]. Hence there is, in general, no canonical assignment O 7→ ρO with ρO a density operator representing the local state in O . Any ontology that presupposes such an assignment is therefore not describing QFT; it is describing a Type I approximation to QFT (with cutoffs, buffers, or split-property insertions) and then reifying that approximation as fundamental. Why “global density matrices” do not rescue the situation. One might attempt to evade this conclusion by insisting that although local density matrices fail, a global density matrix on some Hilbert space still exists, and that local states are obtained by tracing out the complement. This maneuver fails for the same structural reasons. First, in field theory there is no canonical Hilbert-space tensor factorization associated with spatial regions, so the notion of tracing out a complement is not intrinsic (Section V A). Second, even if one works in a particular representation in which a global Hilbert space is available, the map from a global density operator to “the state in a region” depends on additional, non-canonical choices (factorizations, split inclusions, coarse-grainings). The resulting reduced density matrices are therefore not representation-invariant fundamental objects, but context-dependent coordinatizations [ 9 , 10 ]. In short, the appeal to a global ρ does not produce an intrinsic local ontology, and local QFT is precisely about locality. What replaces density-matrix ontology. Once this consequence is accepted, the correct foundational replacement is forced: the fundamental state concept must be formulated directly in terms of algebraic states and their restrictions, and the correct global notion must be a gluing concept compatible with the net structure. This sets the stage for the remainder of the paper. In Sections VI–VII we will formulate the appropriate replacement as a context-organized, (generally non-strict) descent structure: local algebraic states form the primary data; globality is coherence of these data across contexts; and failures of strict global extension appear as controlled coherence defects rather than as missing information or a need for collapse. In that framework, density matrices reappear only where they are legitimate: as Type I coordinatizations in regimes where strictification is available. 17 Summary. Local QFT does not merely suggest caution about density-matrix ontology; it rules it out as a fundamental principle. The only coherent way forward is to elevate algebraic states and coherence/gluing conditions to the foundational level, treating density operators as contingent charts rather than as ontic primitives. VI. REFRAMING THE PROBLEM: STATES AS DESCENT DATA A. Contexts instead of subsystems The preceding sections narrow the options to a clean conclusion: if density matrices are not intrinsic objects in Type III local QFT, then the foundational language must be reorganized. The reorganization we propose begins with a simple replacement of primitives. Instead of starting from subsystems (Hilbertspace factors) and then asking for reduced density matrices, we start from contexts—operationally and algebraically defined modes of access—and treat states as data assigned to these contexts, together with compatibility conditions. This shift is minimal yet decisive: it is compatible with Type III locality, it retains the operational content of quantum theory, and it makes precise the sense in which “globality” is a gluing notion rather than a single density operator. (i) What is a context? By a context we mean any specification that renders “state” a well-typed notion of expectation assignment. The key point is that in quantum theory the notion of state is inseparable from the algebra of observables to which it assigns expectations. In practice, different choices of what one counts as an observable, what one treats as accessible, and what one treats as background correspond to different contexts. For the purposes of this paper, the most important classes of contexts are: 1. Regions: spacetime regions Oin AQFT, with associated local algebras A(O)[9]. 2. Algebras / subalgebras: a chosen von Neumann algebra M or a physically motivated subalgebra C ⊂ M (e.g. a pointer subalgebra selected by decoherence) [5, 6]. 3. Decoders / coarse-grainings: a specification of which observables are actually read out and which distinctions are ignored; mathematically, a coarse-graining map, conditional expectation, or a choice of commutative (approximately classical) subalgebra that defines the effective description. 4. Measurement frames: a choice of compatible set of observables, POVM, or measurement procedure defining what counts as an outcome record; operationally, this fixes the event algebra relevant for predictions. In all cases, a context is not an “extra interpretation” laid on top of a context-free state; it is the structure relative to which the state concept is defined. (ii) Contexts form a category (or groupoid) of access. Contexts are related by refinement, restriction, and change-of-description. The simplest relation is inclusion of regions O1⊂ O2 , inducing A ( O1 ) ⊂ A ( O2 ). More generally, a coarse-graining corresponds to a map from a finer algebra to a coarser one, or to a functorial passage from a more informative to a less informative description. These relations can be organized as a category C of contexts, whose morphisms represent admissible changes of access (restriction, coarse-graining, decoder change, etc.). The exact choice of C depends on the application, but the logical role is stable: it is the scaffolding on which “state” becomes local data. (iii) States as assignments to contexts. Given a context U∈C with associated algebra A ( U ), a state in that context is an algebraic state functional ωU:A(U)→C, ωU(A∗A)≥0, ωU(1)=1.(15) If V→U is a morphism in C expressing that V is a refinement of U (or that V embeds into U ), then there is an induced algebra map A ( V ) ,→ A ( U )or a suitable pullback/pushforward between observable structures, and hence a canonical restriction (or induced) map on states. In the simplest inclusion case, this is just restriction of functionals: ωU7−→ ωV:= ωU|A(V).(16) Crucially, (15) – (16) make no reference to traces, density matrices, partial traces, or Hilbert-space factorization. They are therefore meaningful in Type III settings by construction [9, 10]. 18 (iv) Why “decoder” belongs at the same level as “region.” It is important not to reintroduce Type I ontology by stealth. If one thinks of a region as the only context, one might still be tempted to search for a single “true” state of the region. But operationally, what is accessible in a region depends on the chosen measurement frame and on which subalgebra of observables is effectively classical (stable records). Decoherence precisely teaches that the relevant effective classical algebra is selected by the interaction and the environment; thus the “pointer” context is part of the physical specification [ 5 , 6 ]. In other words, the operational notion of a “state of the region” is already decoder-dependent. Treating decoder choice as part of the context makes this dependence explicit and prevents a category error: the local state is not a context-free ontic object; it is a consistent expectation assignment relative to an access structure. (v) From contexts to gluing: why this reframes the problem. Once contexts replace subsystems, the central question shifts. The Type I extension problem asked for a global density matrix whose partial traces reproduce local density matrices. In the context language, the corresponding question is: Given local state assignments {ωU} for contexts U and compatibility on overlaps/refinements, do these data glue to a coherent global structure? If not strictly, what is the controlled defect? This is a descent question. It makes sense in Type III QFT because it is phrased in terms of algebras and restrictions, not traces and tensor products. It also absorbs the correct part of Penrose’s critique: “ ρ does not encode what is” becomes “local coordinatizations of ωU do not define a unique global ontology.” Finally, it places decoherence in the correct role: decoherence selects stable effective subcontexts (pointer subalgebras) but does not by itself solve the gluing problem. Summary. Replacing subsystems by contexts is the minimal reorganization that makes the state concept compatible with Type III locality while keeping quantum operational content intact. Regions, local algebras, decoder/coarse-graining choices, and measurement frames are all contexts in the same logical sense: they define the domain on which a state functional lives. The remainder of Section VI develops how families of such local state assignments are glued by descent (typically non-strict), and how failures of strictification appear as controlled coherence defects rather than as paradoxes or a need for collapse. B. Algebraic states as local sections The context framework of Section VI A becomes powerful once we recognize that it provides exactly the right notion of “local section” for quantum states. In this subsection we (i) restate states as positive linear functionals (the intrinsic object in all von Neumann types), and (ii) explain why restriction is not ontologically “forgetting” but rather a change of domain that may discard relational data without implying the existence of a locally complete state of affairs. (i) States as positive linear functionals (intrinsic, Type I–III). For each context U with algebra A ( U ), a state is a positive normalized functional ωU as in (15) . This is the unique notion of “state” that is simultaneously: •intrinsic to the algebraic formulation of QFT (nets of local algebras) [9], • independent of the existence of a trace or density operator representation (hence valid in Type III) [10], •and compatible with the operational meaning of expectation values. When A ( U )happens to be Type I and ωU is normal, one may represent it by a density operator ρU via (9) . In that sense, density matrices are charts on the state functional, not the state itself. This inversion of viewpoint is what allows us to carry Penrose’s diagnosis into the field-theoretic setting without importing Type I ontology. (ii) Local sections over a context category. Given a context category C and a functor (or assignment) U7→ A(U), we can view the state assignment U7−→ ωU∈St(A(U)) (17) as a presheaf-like object: a family of local data living over contexts. Compatibility along a morphism f:V→Uis expressed by the requirement that the state on Vagrees with the restriction (or pullback) of the state on U, ωV=f∗(ωU),(18) 19 with f∗ induced by the corresponding algebra map. In the simplest inclusion case, f∗ is just restriction of the functional to a subalgebra, as in (16) . A global state in the strict sense would then be a global section: a compatible family satisfying (18) for all morphisms. In the next subsection we will explain why this strict global section need not exist uniquely (or at all), and why non-strict descent is the right replacement. (iii) Restriction is not “forgetting”. The language of “forgetting” is inherited from classical marginalization, where a joint distribution on X×Y canonically projects to a marginal on X . In the quantum context picture, the primary operation is domain change: one restricts a functional on a larger algebra to a smaller algebra. This operation always exists, but it does not carry the ontology-level meaning “remove irrelevant variables.” Instead, it discards relational information: correlations and phase relations that are not expressible in the smaller algebra. This is the algebraic way to restate what the proper/improper distinction already indicated (Section III A). If Ωis a global state on a larger context U and V→U is a restriction, then ωV = Ω |A(V) need not correspond to any ontic “state of V ” that exists independently of U ; it is merely the expectations accessible in context V given the global preparation. Multiple inequivalent global states can restrict to the same local state. Thus restriction is information-losing in a structural sense, but it does not imply that the lost information was “local reality in the discarded degrees of freedom.” It may be irreducibly relational. (iv) Type III makes the point unavoidable. In Type I models one can still tempt oneself to say: “fine, but surely behind the restriction there is a joint density matrix and we just traced something out.” Type III locality forbids this as a general foundational move: there may be no density matrix representing ωV at all, and there is no canonical tensor split that would make partial tracing meaningful [ 9 , 10 ]. Restriction is the only intrinsic notion of reduction. Therefore, ontological stories that depend on interpreting restriction as forgetting are not merely philosophically optional; they are structurally incompatible with local QFT. (v) Penrose and decoherence revisited in this language. In this context-functional viewpoint, Penrose’s critique is naturally absorbed: the non-uniqueness of ensemble decompositions is the statement that a Type I chart ρU on ωU does not carry intrinsic ontology [ 7 , 12 ]. Decoherence, by contrast, is reinterpreted as a dynamical mechanism that selects stable subcontexts (pointer subalgebras) and makes certain restrictions effectively classical for practical purposes [ 5 , 6 ]. Neither statement requires a collapse postulate. What remains is to clarify how families of local sections glue into a coherent global structure when strict global sections fail or are non-unique. This is the descent problem developed next. Summary. Algebraic states are intrinsically defined in all von Neumann types and therefore provide the correct “local section” notion for QFT. Restriction is the fundamental reduction operation and should be read as a change of accessible algebra, not as ontological forgetting. With this retyping, density matrices become optional coordinate charts, Penrose’s critique becomes a structural statement about non-intrinsic decompositions, and decoherence becomes a context-selection mechanism rather than an ontological completion. C. Descent vs. strict global states We can now articulate the precise point at which the density-matrix ontology (and, more generally, strict global ontology) fails: not at the level of local state assignment, but at the level of gluing. The language of descent provides a disciplined way to express this failure without turning it into paradox or incompleteness. In this subsection we explain (i) when gluing fails strictly, and (ii) why higher coherence appears as the correct replacement structure. (i) The strict picture: global sections as unique global states. In the simplest (and historically dominant) picture, one imagines that there exists a single global state object Ωfrom which all local states are obtained by restriction. In the context formulation of Section VI B, this corresponds to a strict global section: a family {ωU}U∈Csuch that for every morphism f:V→Uone has exact compatibility ωV=f∗(ωU).(19) If such a strict global section exists and is unique, then “the state of the world” is well-defined in the strong sense, and local states are merely its restrictions. In Type I language this strictness is often silently conflated with the existence of a global density operator whose marginals are the reduced density matrices. 20 However, even in Type I quantum mechanics, strict globality is not guaranteed if one starts from local data: the extension problem of Section II C shows that collections of local marginals need not extend to a single global ρ . In Type III QFT, strict globality in the density-matrix sense is not even the right target. The correct question is not whether a single global density operator exists, but whether the local functional data admit a coherent gluing. (ii) Descent: gluing local data into a global object. Descent is the general mathematical principle that replaces “global section” by “gluing of local data with compatibility on overlaps.” Concretely, one considers a cover of a context (or a global context) by subcontexts {Ui}and specifies: 1. local objects (here: local states) ωion Ui, 2. comparison data on overlaps Uij := Ui∩Ujidentifying ωi|Uij with ωj|Uij , 3. and higher compatibility on triple overlaps Uijk. In a strict sheaf-like situation, compatibility on overlaps would be literal equality, and higher compatibility would be automatic. In the quantum situation of interest, the correct compatibility is weaker: local descriptions often agree only up to equivalence (gauge, inner automorphism, decoder change), and the higher compatibilities encode nontrivial obstruction data. (iii) When gluing fails strictly. Strict gluing fails whenever one cannot identify local state assignments across overlaps by literal equality in a single common representation. There are several physically natural reasons: •Gauge / constraint structure: observables and states may be defined only modulo gauge, and identifications across regions are not strict but mediated by equivalences. •Representation/choice dependence: local restrictions may live naturally in different GNS representations; insisting on a single global Hilbert-space picture can be non-canonical. •Decoder dependence: effective descriptions (pointer algebras, coarse-grainings) are contextselected; local “classical” descriptions need not match strictly across different coarse-grainings. In such cases, demanding a unique strict global state is the wrong requirement. One should instead record the data of how local states are related, and ask whether these relations satisfy coherence constraints. (iv) Why higher coherence appears. Once compatibility is weakened from equality to equivalence, higher coherence becomes unavoidable. The simplest way to see this is on triple overlaps. Suppose we have equivalences (isomorphisms in an appropriate groupoid of presentations) φij :ωi|Uij ∼ = −−→ ωj|Uij .(20) On a triple overlap Uijk there are two ways to compare ωi and ωk : directly via φik , or via φjk ◦φij . In a strict setting these must coincide; in a non-strict setting one requires them to coincide up to a specified 2-morphism (a coherent transformation between transformations), αijk :φjk ◦φij ⇒φik on Uijk.(21) Now one must demand that these αijk themselves satisfy a consistency condition on quadruple overlaps (a pentagon-type condition), which introduces 3-morphisms, and so on. This is the standard mechanism by which “weak” gluing produces higher coherence data. The presence of such higher morphisms is not a pathology: it is the correct bookkeeping of consistent non-strict identification. (v) Interpretation: from missing global states to controlled defects. The key conceptual shift is therefore: Failure of strict gluing is not absence of structure; it is the presence of higher coherence structure. In the language of this paper, the data {φij, αijk, . . .} encode precisely how far the local descriptions are from strictifying into a unique global state. When the coherence data can be trivialized (all αijk equivalent to identities, etc.), one recovers strict globality and, in Type I regimes, density-matrix descriptions can be globally consistent. When they cannot be trivialized, the resulting obstruction is physical: it is the controlled coherence defect that replaces the demand for strict ontology. This is the structural mechanism by which Penrose’s “ontological ambiguity” is absorbed without collapse: the ambiguity is reinterpreted as nontrivial higher descent data rather than as a deficiency requiring non-unitary dynamics. 21 (vi) Bridge to the next section. The remainder of the paper makes this interpretation explicit. Section VII will formalize how non-strict descent and its coherence morphisms are read physically as higher-categorical coherence breakdown (HCCB), and how density matrices reappear as mere coordinatizations when the descent data strictifies. This is the sense in which the “Type III no density matrix” obstruction is resolved: the fundamental objects are local algebraic states and their coherent gluing data, not trace-class operators. Summary. Strict global states correspond to strict global sections. In quantum theory (and decisively in Type III QFT), gluing is generally non-strict: local descriptions match only up to equivalence, and higher coherence morphisms encode the consistency of these equivalences on multiple overlaps. The resulting structure is richer than strict ontology and supplies the correct replacement for density-matrix ontology: globality as coherent descent rather than as a single density operator. VII. HIGHER-CATEGORICAL COHERENCE BREAKDOWN (HCCB) A. Non-strict descent Section VI C explained why the correct replacement for strict global states is a descent structure in which identifications are not required to be literal equalities. We now sharpen this into the non-strict (higher-categorical) viewpoint that underlies higher-categorical coherence breakdown (HCCB). The key move is to treat “state data” as living not in a set but in a groupoid (or higher groupoid) of presentations, where compatibility is formulated up to equivalence, and failures of strict associativity are recorded as higher morphisms rather than interpreted as inconsistency. (i) Compatibility up to equivalence. In realistic quantum settings, local descriptions of the same physics may differ by transformations that should not be treated as physical distinctions: gauge transformations, inner automorphisms, changes of decoding/coarse-graining conventions, or changes of representation. Consequently, the right compatibility notion on overlaps is not equality but equivalence. Concretely, for local contexts Uiand overlaps Uij one replaces strict equalities ωi|Uij =ωj|Uij by specified equivalences between the restricted state data: φij :ωi|Uij ∼ −−→ ωj|Uij .(22) Here the φij live in whatever groupoid captures the “same physics” identifications appropriate to the problem (e.g. gauge-equivalence of presentations, unitary equivalence of GNS representations, decoder equivalence of effective descriptions). The point is not the particular choice of groupoid, but the logical role: physical compatibility is expressed as existence of an equivalence, not as equality of raw representatives. (ii) Failure of strict associativity as structure. Once compatibility is formulated by equivalences (22) , strict associativity is no longer automatic. On triple overlaps Uijk there are two a priori distinct composites from ωito ωk: φik and φjk ◦φij . Non-strict descent does not demand these coincide as equal maps. Instead, it supplies a specified 2-morphism witnessing their coherence: αijk :φjk ◦φij ⇒φik on Uijk.(23) On quadruple overlaps Uijkl one must then require that the two ways of composing the α ’s agree up to a 3-morphism, and so on. This “tower” of higher coherence is not optional; it is the correct bookkeeping of consistent gluing when equality is too strict a demand. In short: Non-associativity at the level of representatives is not inconsistency; it is the presence of higher coherence data. This is the precise categorical meaning of “coherence breakdown” in your usage: strict associativity fails, but the failure is controlled by higher morphisms that satisfy higher coherence conditions. The world does not become ill-defined; rather, the correct global object is weak (defined up to coherent equivalence) rather than strict. 22 (iii) Relation to density matrices and Penrose’s ambiguity. In a Type I regime, choosing a density matrix ρ is choosing a representative in a chart. Different ensemble decompositions of ρ correspond to different contextual presentations, and the HJW theorem makes this dependence explicit [ 12 ]. Penrose reads this as ontological ambiguity of ρ . In the non-strict descent view, the ambiguity is reinterpreted as the existence of multiple equivalent presentations related by coherence data: what is invariant is not the representative but the equivalence class together with its gluing constraints. In Type III regimes, where density matrices cease to exist as intrinsic objects, the non-strict descent formulation remains meaningful because it is stated directly in terms of algebraic state data and equivalences among their presentations. (iv) Why this is the right replacement for “strict ontology”. Penrose’s extra assumption (Section III C) was the demand for a unique strict global state. Non-strict descent provides the structurally natural alternative: globality is not a single object but a coherent gluing class. This move retains all operational predictions (because local expectations are still well-defined) while avoiding the demand that the global object strictify. In particular, it replaces the collapse-driven requirement “pick one branch” by the coherence-driven requirement “maintain consistent overlap equivalences and higher associators.” Summary. Non-strict descent replaces equality of local state data by equivalence (22) and replaces strict associativity by higher coherence morphisms (23) . The resulting structure is not weaker in content but richer: it records precisely the controlled ways in which strict global ontology fails. This is the first formal ingredient of HCCB. In the next subsection we identify the physically meaningful invariants of this structure as coherence defects, and explain how they resolve the Type III “no density matrix” obstruction without collapse. B. Coherence defects as physical data Non-strict descent introduces additional structure beyond local states: overlap equivalences φij and higher coherence morphisms αijk , etc. (Section VII A). The crucial interpretive step—and the point where HCCB becomes physics rather than category theory—is to recognize that the failure to strictify these data is not noise, not ignorance, and not an artifact of sloppy modeling. It is, in general, a stable, classifiable obstruction that carries physical meaning. (i) Defects are not noise. Noise is contingent: it depends on experimental imperfections and disappears in an ideal limit. Coherence defects, by contrast, are defined by the internal consistency of the gluing data. They persist even in a perfectly controlled mathematical model. In the simplest setting, the defect is the mismatch between φik and φjk ◦φij on triple overlaps, captured by αijk in (23) . If all such mismatches can be removed by a redefinition of the overlap maps (a “gauge” change of descent data), then the structure strictifies. If not, the remaining obstruction is invariant under such redefinitions and therefore physical. This is the same logical pattern by which gauge-invariant curvature replaces gauge-dependent connection coefficients: the defect is the gauge-invariant residue of an attempted strictification. (ii) Defects are not ignorance. Ignorance corresponds to multiple compatible completions: many possible strict global states consistent with the same local data. A coherence defect corresponds to the opposite: the non-existence of any strict completion consistent with the coherence constraints. In the language of Section II C, ignorance would mean non-uniqueness of extension; a defect means non-existence of strict extension but existence of a coherent weak extension. Thus, defects are not epistemic gaps; they are structural features of how the theory organizes compatibility across contexts. This is precisely why Penrose’s critique is both right and incomplete. He correctly identifies that density matrices and their decompositions underdetermine ontology. But the conclusion “there must be a hidden unique state” is not forced. The algebraic and descent reformulations show that the underdetermination can reflect the fact that the correct global object is weak (defined only up to coherent equivalence), and the residual obstruction to strictification is a genuine invariant rather than missing information. (iii) Stable, classifiable obstructions. The formal reason defects are physically meaningful is that they are cohomological in nature: they behave like obstruction classes. Concretely, changing the overlap equivalences φij by “gauge” transformations modifies αijk by a coboundary-type adjustment; what survives is an equivalence class of defects. In familiar cases this mechanism yields: •Anomalies: failure of a symmetry to be realized consistently at the quantum level, encoded by cocycles rather than by local counterterms. •Projective representations: associativity holds only up to a phase, and the phase defines a cohomology class (a central extension) rather than an inconsistency. 23 •Gerbe-like structures: local trivializations exist but cannot be glued globally without a higher cocycle. We will not need the full apparatus of these examples here; the relevant point is the pattern: controlled non-strictness produces invariants. These invariants are stable under refinements of presentation and therefore constitute physical data. (iv) How coherence defects resolve the Type III “no density matrix” issue. The Type III obstruction says there is no intrinsic local density matrix and no canonical partial trace. The coherence-first replacement is: the fundamental objects are local algebraic states together with their coherent gluing data. In this replacement, “global state” is not a trace-class operator but a descent object. The coherence defects are precisely what one must keep track of when strict global density-matrix ontology fails. This also clarifies why decoherence cannot complete ontology: decoherence may select stable pointer subalgebras and thereby simplify the local state data, but it does not remove coherence defects that are intrinsic to the gluing structure across contexts. Decoherence changes which contexts are operationally salient; it does not trivialize the higher obstruction classes that encode non-factorizable ontology. (v) Physical reading: HCCB as controlled non-closure. In your terminology, higher-categorical coherence breakdown (HCCB) is the controlled failure of the “closure” conditions that would allow strict global ontology. The correct physical invariant is not the failure itself (which depends on representatives) but the class of that failure under allowed redefinitions of presentation. This is why we emphasize “defect” rather than “breakdown”: the latter sounds like inconsistency, whereas the former indicates a stable, classifiable feature that one can track, constrain, and potentially measure indirectly via its implications for compatible observable sets. Summary. Coherence defects are the gauge-invariant residues of attempted strictification of non-strict descent data. They are not experimental noise and not ignorance about a hidden strict global state. They are stable, classifiable obstruction data—the correct replacement for strict density-matrix ontology in Type III settings and beyond. In the next section we show how this viewpoint absorbs Penrose’s diagnosis without collapse and yields a precise statement of how density matrices reappear only as context-dependent coordinatizations when strictification is available. C. Density matrices reinterpreted We can now state the central conceptual replacement of this paper in a single sentence: Density matrices are not fundamental state objects; they are local coordinatizations of algebraic state data, valid only when the underlying descent structure strictifies to Type I. This subsection makes that statement precise and explains why it resolves both Penrose’s critique and the Type III obstruction without invoking collapse. (i) Density matrices as charts on the space of states. In the algebraic viewpoint, the fundamental object is a state functional ω : A → C . When A = B ( H )is Type I and ω is normal, one can represent ω by a unique trace-class operator ρ such that ω ( A ) = Tr ( ρA )for all A∈ A (cf. (9) ). This representation is best understood as choosing coordinates on ω relative to the trace pairing. The density matrix ρ is therefore achart-dependent coordinate on the state, not the state itself. Different choices of representation (e.g. different embeddings into a Type I factor, different split inclusions, different coarse-grainings) correspond to different coordinate charts. (ii) “Reduced density matrices” as coordinatized restrictions. Similarly, the reduced density matrix of a subsystem is, in this view, a coordinatized form of a restriction. The intrinsic operation is restriction of functionals, Ω7→ Ω|A(U), and only when A ( U )is (effectively) Type I does one rewrite this restricted functional as Tr ( ρU· ). The usual partial trace formula is a special case that requires additional Type I structure: a tensor factorization and a trace. When these are absent (as in Type III QFT), the restriction still exists and still carries all operational content, while the density-matrix coordinatization does not. (iii) Validity domain: only when descent strictifies. The non-strict descent structure described in Sections VII A–VII B is the intrinsic global object. A density-matrix description becomes globally meaningful only when this structure strictifies. Concretely, strictification means that one can choose representatives and overlap identifications such that all higher coherence morphisms become trivial (equivalent to identities), yielding a strict global section in the sense of (19). In that situation: 24 • local state functionals can be consistently represented in a common Type I chart (at least effectively), •the overlap equivalences become strict equalities in that chart, •and one may legitimately speak of density matrices as if they were globally compatible objects. Outside this strictification domain, insisting on density matrices as fundamental is exactly the mistake that generates the Penrose ambiguity (non-intrinsic decompositions) and conflicts with Type III locality (non-existence of traces and partial traces). (iv) Penrose absorbed: ambiguity becomes chart dependence. Penrose’s complaint that ρ does not encode “what is” is reinterpreted here as the statement that ρ is a chart-dependent coordinate on ω and therefore cannot carry intrinsic ontology. The HJW theorem makes this concrete by relating different decompositions of the same ρ to different contexts on a purification [ 12 ]. In our language, this is exactly what one expects from a coordinate object: it changes under changes of context. What is invariant is not the decomposition but the equivalence class of the underlying descent data. Therefore, the valid core of Penrose’s critique is retained, while the move to objective collapse is avoided: the issue is not that unitarity fails, but that one should not reify chart-dependent coordinates as ontic primitives. (v) Type III resolved: “no density matrix” is not a deficiency. In Type III QFT, the non-existence of density matrices for local regions is no longer a crisis. It simply means that the Type I chart ρ is unavailable. The intrinsic state functional ωO exists, its restrictions exist, and the descent/coherence structure exists. Operational predictions remain intact. The correct conclusion is not to force a densitymatrix ontology by approximation and then declare it fundamental, but to recognize that the appropriate ontology is the non-strict descent object and its coherence defects. Density matrices then reappear exactly where they should: in effective regimes where strictification is available (e.g. finite-dimensional approximations, split-property coordinatizations, coarse-grained pointer contexts). Key conceptual replacement. This subsection supplies the paper’s central replacement rule: Replace “the state is a density matrix” by “the state is a context-organized algebraic functional glued by (generally non-strict) descent; density matrices are local coordinates when strictification permits.” This rule simultaneously (i) validates Penrose’s diagnosis about ontological inadequacy of ρ , (ii) explains why decoherence cannot by itself restore ontology (it selects effective contexts but does not enforce strictification), and (iii) resolves the Type III “no density matrix” obstruction by removing the demand that ρexist fundamentally. VIII. ABSORBING PENROSE WITHOUT COLLAPSE A. Retained from Penrose We are now in a position to state precisely what is retained from Penrose once the theory is reformulated in algebraic and descent terms. The retained content is not a rhetorical stance about “interpretations”; it is a set of structurally correct statements that survive independently of whether one endorses objective reduction. In fact, the preceding sections show that these statements become more compelling once Type III locality is taken seriously. (i) Density matrices are not ontic. The first retained point is Penrose’s core diagnosis: density matrices do not encode “what is” [ 7 , 8 ]. In our reformulation, this is strengthened and clarified. A density matrix ρ is a Type I coordinatization of an algebraic state functional ω (Section VII C). Its non-uniqueness of ensemble decompositions is then not a paradox but the expected coordinate dependence of a chart object. The HJW theorem makes this dependence operational: different decompositions correspond to different contexts on a purification [ 12 ]. Thus Penrose’s dissatisfaction with ρ as an ontic carrier is exactly right—and it is not merely an interpretational preference. It follows from the structural underdetermination of ontology by ρ in Type I settings and from the non-existence of ρ in Type III local QFT settings (Section V) [9, 10]. (ii) Improper mixtures are fundamental (and unavoidable). The second retained point is that improper mixtures are not a technical artifact to be eliminated; they are intrinsic to quantum theory. In the algebraic formulation, a global pure state Ωon a larger algebra generically restricts to a mixed state on a subalgebra, ωU= Ω|A(U), 25 and this fact persists (indeed becomes the primary notion of locality) in Type III QFT where density matrices are not available. Improper mixtures therefore cannot be dismissed as “mere ignorance.” They are the natural expression of nonseparability: local access does not canonically factorize global reality into independent local ontic states [ 2 , 9 ]. This is exactly why Penrose is right to regard the improper-mixture structure as a genuine ontological challenge. In the coherence-first formulation, this challenge is met not by forcing improper mixtures to become proper, but by recognizing that they are precisely what one should expect when the underlying global object is a non-strict descent datum. Improper mixtures are the local shadows of a coherent global structure that need not strictify into a unique global density-matrix ontology. Thus, the “fundamentality” of improper mixtures is retained, but reinterpreted: they are not symptoms of missing dynamics; they are the local manifestations of non-factorizable, coherence-governed globality. Summary. We retain from Penrose the two statements that are structurally correct and sharpened by the Type III perspective: 1. Density matrices are not ontic carriers of reality; at best they are context-dependent coordinatizations of algebraic state data [7, 12]. 2. Improper mixtures are fundamental: restriction of global states to local contexts generically produces mixed local states, and this is the correct notion of locality in QFT [2, 9]. The next subsections identify what we do not retain (the collapse-demanding extra assumption) and explain how the coherence-first descent framework supplies the missing ontology without violating unitarity. B. Rejected Penrose’s correct diagnosis does not logically force his proposed remedy. The present framework therefore rejects two closely linked requirements that drive the objective-reduction conclusion. We reject them not by fiat, but because they are (i) additional postulates beyond the algebraic content of quantum theory, and (ii) misaligned with the structural facts of Type III locality and non-strict descent. (i) Rejection of the need for objective collapse. The first rejected step is the claim that resolving the ontological ambiguity of density matrices requires a modification of unitary dynamics by an objective collapse mechanism [ 7 ]. In the coherence-first formulation, the ambiguity does not signal that the dynamics fails; it signals that a Type I coordinatization ( ρ ) has been mistakenly treated as an ontic primitive. Once density matrices are reinterpreted as local charts on algebraic state data (Section VII C), Penrose’s ontological complaint is absorbed structurally: the fundamental objects are state functionals and their coherent gluing data. No non-unitary dynamics is required to “select” a decomposition of ρ , because decompositions are not fundamental physical variables; they are presentation-dependent. Moreover, decoherence already accounts for the emergence of stable classical records and the practical suppression of interference for relevant observables (Section IV). The remaining gap—why one outcome is experienced—is not repaired by enforcing collapse at the level of ρ , because in the Type III setting ρ is not even the intrinsic state object. In our framework, definiteness is modeled as context-relative strictification within an operative subcontext (e.g. pointer subalgebra), while the global object remains unitary and coherence-governed. Collapse is therefore not required to obtain effective classicality, nor is it required to provide a consistent ontology once globality is understood as non-strict descent. (ii) Rejection of the need for a unique strict global state. The second rejected requirement is the demand for a unique realized global state (a strict global section), which is the hidden assumption isolated in Section III C. This demand is stronger than what quantum theory provides and stronger than what Type III locality permits as a foundational principle. In the descent formulation, the correct global object need not be a unique strict section; it may be a coherent gluing class with nontrivial coherence morphisms and stable obstruction data (Sections VI C–VII B). Insisting on unique strict globality is precisely what turns non-strict coherence into an apparent crisis. In particular, the Type III obstruction (Section V) implies that “global density matrix with local partial traces” is not the right ontological template for local QFT. If a unique global state is insisted upon, it must be formulated directly in algebraic terms, and even then it will generally be defined only up to coherent equivalence across contexts rather than as a single globally factorizable object. The correct invariant content lies in the descent data and their coherence defects, not in a uniquely singled-out global representative. 32 algebras, decoding maps), •algebraic state data on these contexts, • and a non-strict descent principle controlling gluing and identifying the physically meaningful coherence defects. In such a formulation, the familiar paradoxes—information loss, cloning tensions, interior/exterior duplication, and the sharpness of entanglement wedge reconstruction—are reorganized as questions about strictification domains and obstruction classes. This is a natural continuation of the present paper’s main message: many of the apparent ontological crises in quantum gravity arise from reifying Type I density-matrix intuition in regimes where the correct objects are Type III and the correct globality is coherence-first. C. Closing sentence Closing statement. The failure of density matrices in Type III quantum theory is not a deficiency of quantum mechanics, but a signal that ontology itself must be formulated in non-strict, coherence-theoretic terms: states are contextual algebraic data glued by descent, and what is physically real is the resulting coherence class (with its defects), not a globally factorizable density operator. [1] J. von Neumann, Mathematische Grundlagen der Quantenmechanik, Springer, Berlin (1932). English translation: Mathematical Foundations of Quantum Mechanics, Princeton University Press (1955). [2] B. d’Espagnat, Conceptual Foundations of Quantum Mechanics, 2nd ed., Addison–Wesley, Reading, MA (1976). [3] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, 10th Anniversary Edition, Cambridge University Press (2010). [4] E. Joos, H. D. Zeh, C. Kiefer, D. J. W. Giulini, J. Kupsch, and I.-O. Stamatescu, Decoherence and the Appearance of a Classical World in Quantum Theory, 2nd ed., Springer, Berlin (2003). [5] W. H. Zurek, “Decoherence, einselection, and the quantum origins of the classical,” Rev. Mod. Phys. 75 (2003) 715–775. [6] M. Schlosshauer, Decoherence and the Quantum-to-Classical Transition, Springer, Berlin (2007). [7] R. Penrose, “On gravity’s role in quantum state reduction,” General Relativity and Gravitation 28 (1996) 581–600. [8] R. Penrose, The Emperor’s New Mind: Concerning Computers, Minds, and the Laws of Physics, Oxford University Press, Oxford (1989). [9] R. Haag, Local Quantum Physics: Fields, Particles, Algebras, Springer, Berlin (1992). [10] M. Takesaki, Theory of Operator Algebras I, 2nd ed., Encyclopaedia of Mathematical Sciences 124 , Springer, Berlin (2002). [11] S. J. Summers, “On the independence of local algebras in quantum field theory,” Reviews in Mathematical Physics 2(1990) 201–247. [12] L. P. Hughston, R. Jozsa, and W. K. Wootters, “A complete classification of quantum ensembles having a given density matrix,” Physics Letters A 183 (1993) 14–18.