Resolve OP-48: can GPT framework-axioms live on a Type III₁ algebra? (split-property test)

Status: SETTLED as a sharp disjunction (medium confidence) — OP-48 / OP-GPT-TYPE3 is a NO-GO for exact composition on a single Type III₁ factor and a state-/collar-dependent YES for nested regions under the split property; full operational reconstruction remains OPEN. didItMove = sharpened-obstruction. Last updated: 2026-06-08 Iteration: 4

This note executes the single "next move" the iteration-3 merge note named (see 2026-06-08-iter3-merge-attempt-operational-substrate.md, "Next move"): test whether the GPT framework-fixing axioms — local tomography / tensor composition, purification, continuous reversibility — admit a formulation on a Type III₁ von Neumann algebra, or whether the split property rescues them. Iteration 3 left this as [OPEN — domain-of-validity mismatch + missing-theorem gap; explicitly NOT a no-go]. Iteration 4 carries the split-property rescue as far as it goes, locates exactly where it breaks, and settles OP-48 as a disjunction with a sharp dividing line, not a single verdict. The referees attacked the furthest claim and the two registry items; their corrections (tag granularity; "canonical-given-Ω" phrasing; canonicity-tax vs. existence) are folded in below.

Goal

Settle OP-48 (OP-GPT-TYPE3): can the GPT/operational framework-fixing axioms — local tomography ( $\dim V_{AB} = \dim V_A \cdot \dim V_B$ ), purification, continuous reversibility — be formulated on a Type III₁ von Neumann algebra $\mathcal{M} = \mathcal{A}(O)$ or its normal state space $\mathcal{S}_*(\mathcal{M})$ ? Three admissible verdicts:

a genuine no-go (composition needs factorization, impossible on a single III₁ factor);
a split-approximate yes (the split property restores factorization for nested regions, defeating the no-go);
a disjunction with a precisely located dividing line between the two.

Subgoals carried over from iteration 3: (b) check whether the 2026 σ-affinity + normal-steering rigidity theorem (Torres Alegre, arXiv:2602.09056) extends from the Born form to composition; (c) decide whether a state-independent, collar-free composition surrogate exists.

Setup (real math)

The obstruction object: a single Type III₁ factor

A single local algebra in relativistic QFT is the hyperfinite Type III₁ factor $\mathcal{M} = \mathcal{A}(O)$ (Connes–Haag–Araki) ESTABLISHED. It has

\text{no trace},\quad \text{no minimal projections},\quad \text{no density matrix},\quad \text{no factorization } \mathcal{H} \ne \mathcal{H}_{\mathcal{M}} \otimes \mathcal{H}_{\mathcal{M}'}.

The GPT axiom that does the framework work is local tomography,

\dim V_{AB} = \dim V_A \cdot \dim V_B,

which presupposes $\mathcal{H}_{AB} = \mathcal{H}_A \otimes \mathcal{H}_B$ with local density matrices ESTABLISHED. On a single III₁ factor this presupposition is false, so local tomography is not even statable verbatim. This is the iteration-3 BREAK 1 (see 2026-06-08-iter3-merge-attempt-operational-substrate.md).

The candidate rescue: the split property

Composition in algebraic QFT is built not from a single factor but from nested inclusions $O_1 \Subset O_2$ (read " $O_1$ relatively compact in $O_2$ ", i.e. a nonzero collar / buffer separates them). A net satisfying a nuclearity condition (Buchholz–Wichmann; the $L^2$ -nuclearity of Buchholz–D'Antoni–Longo) admits an intermediate Type I factor $N$ with

\mathcal{A}(O_1) \subset N \subset \mathcal{A}(O_2),

which is precisely the split property (Doplicher–Longo, Inventiones math. 75, 493–536, 1984) ESTABLISHED, web-confirmed. Existence of an intermediate Type I factor yields:

an exact unitary factorization $\mathcal{H} \cong \mathcal{H}_1 \otimes \mathcal{H}_2$ with $\mathcal{A}(O_1) \subset B(\mathcal{H}_1) \otimes \mathbf{1}$ ;
statistical (W*-)independence: any normal state on $\mathcal{A}(O_1)$ and any on $\mathcal{A}(O_2)'$ are jointly realizable, and product states $\psi_1 \otimes \psi_2$ exist;
local density matrices on the intermediate Type I factor $N$ .

So a local-tomography-type statement is formulable for split inclusions — far stronger than "III₁ has no factorization, full stop."

The canonical intermediate factor

Given a vector $\Omega$ cyclic and separating for the joint algebra (so Tomita–Takesaki applies), the standard split inclusion singles out a canonical intermediate Type I factor via the modular conjugation $J$ of the relative commutant:

N \;=\; \mathcal{A}(O_1) \vee J\,\mathcal{A}(O_1)\,J, \qquad\text{equivalently}\qquad F \;=\; N \vee J N J \text{ as the unique } J\text{-invariant factor},

(Doplicher–Longo 1984; Tanimoto, Conformal Covariance and the Split Property, CMP 2018) ESTABLISHED, web-confirmed. Without choosing such a vector there are infinitely many intermediate Type I factors and the factorization is "highly non-unique." Canonicity is conferred by a choice of cyclic–separating reference vector — the vacuum $\Omega$ being the distinguished physical choice, not the unique mathematical one (any faithful normal GNS vector yields its own canonical factor, invariant only up to a unitary-equivalence class). [Referee correction folded in: "made canonical only by the vacuum" overstated exclusivity — the precise statement is "canonical relative to a chosen cyclic–separating vector, the vacuum being the distinguished physical choice."]

The unknown for subgoal (b): does composition follow from Born-form rigidity?

The 2026 rigidity theorem (Torres Alegre, arXiv:2602.09056) proves, from no-signaling (NSS) + normal steering via purification (NS) + σ-affinity, that any probability rule $P(\phi\mid\psi) = \Phi(\tau(\psi,\phi))$ with $\Phi$ continuous nondecreasing, $\Phi(0)=0$ , $\Phi(1)=1$ , must satisfy

\Phi(p) = p \qquad (\Phi = \mathrm{id}),

the linear Born form as the unique no-signaling fixed point, "via von Neumann algebras through normal state spaces and the GNS representation." The question is whether this extends from the single-system probability rule to the composition rule (tensor product + complex field).

The attempt

(a) The split-property rescue, carried to its limit — VERIFIED far

I verified the following chain, each link ESTABLISHED, web-confirmed:

Nuclearity $\Rightarrow$ split. Buchholz–Wichmann nuclearity (finite phase-space volume $\leftrightarrow$ almost-finite-dimensional state space; energy-level density of states) implies the split property for the free field and restricts local degrees of freedom (Buchholz–Wichmann, CMP, 1986). $L^2$ -nuclearity via modular embeddings $\Phi_p$ is equivalent to the (quasi-)split inclusion (Buchholz–D'Antoni–Longo, Nuclear maps and modular structures I–II, CMP, 1990), giving an explicit modular link between nuclearity and intermediate Type I factors.
Split $\Rightarrow$ exact factorization. The intermediate Type I factor delivers $\mathcal{H} \cong \mathcal{H}_1 \otimes \mathcal{H}_2$ with statistical independence and local density matrices — local tomography is reinstated locally, for the nested pair.
Standard split $\Rightarrow$ canonical factor. With a cyclic–separating $\Omega$ , the factorization is canonical via $J$ (Doplicher–Longo 1984; Tanimoto 2018).
Operational reading. Local preparability of states is equivalent to the split property (Buchholz, Local preparability of states and the split property, Lett. Math. Phys., 1987): local operations preparing a local state imply interpolating Type I factors. So the split structure is exactly the operational "I can prepare here without disturbing there" requirement — the natural GPT preparation axiom in AQFT dress.

Conclusion of (a): the would-be hard no-go is defeated as a no-go. A local-tomography-type statement, with genuine factorization and local density matrices, IS formulable for split inclusions.

(b) Does Born-form rigidity reach composition? — NO (verified verbatim)

I fetched arXiv:2602.09056 (Torres Alegre, Causal Rigidity of Born-Type Probability Rules in Infinite-Dimensional Operational Theories, 7 Feb 2026) and its finite-dimensional predecessor arXiv:2512.12636 (Torres Alegre, Causal Consistency Selects the Born Rule: A Derivation from Steering in Generalized Probabilistic Theories, 14 Dec 2025). Both:

treat only the single-system probability rule $P = \Phi(\tau) \Rightarrow \Phi = \mathrm{id}$ ;
contain no derivation of composition, tensor product, or local tomography. Verified verbatim in the predecessor: it "does not address composition, tensor products, or local tomography for multi-system scenarios."

So the 2026 rigidity track fixes the probability rule, not the composition rule. [Referee note: this is a negative-existence claim over 2018–2026 — thorough but inherently OPEN, correctly hedged with "yet."]

Where it breaks / what moved

The rescue breaks at three precisely-located, web-verified steps. The net effect is to sharpen the iteration-3 obstruction from "domain-of-validity mismatch + missing-theorem gap" to a located no-go-vs-conditional-yes line with two named failure modes plus a verified missing theorem. This is a sharpened-obstruction, not a positive reconstruction.

BREAK 1 — Non-canonicity without a reference state (state-dependence)

If the split property holds there are infinitely many intermediate Type I factors; the factorization is "highly non-unique." It is made canonical only by choosing a cyclic–separating vector (physically, the vacuum $\Omega$ ) and its modular conjugation $J$ ESTABLISHED, web-confirmed. So any canonical, choice-free local density matrix and tomographic factorization are state-dependent — they presuppose a preferred reference state, the very geometry-importing input Program A was meant to generate. This is the BW/CHM circularity recurring at the factorization stage (iteration-3 BREAK 3 was at the entropy stage).

Referee-sharpened distinction (load-bearing): the bare split property buys existence of a factorization with statistical independence for free — that part needs no $\Omega$ . What $\Omega$ 's modular data buys is canonicity — a unique, well-defined factorization / density matrix / entanglement entropy (the canonical entanglement entropy is the von Neumann entropy of $\Omega$ restricted to $F$ ; Hollands–Longo, Von Neumann Entropy in QFT, CMP 2020, arXiv:1911.09390). So the split property is downgraded from "supplies factorization for free" to "supplies non-canonical factorization for free; a canonical one re-imports a preferred reference state." This is a canonicity tax, and it requires only $\Omega$ cyclic–separating — it coincides with the geometric BW/CHM circularity only under the additional hypothesis that the modular flow be geometric (the step isolated as [A-CP-KILLING], see ../ASSUMPTIONS_LEDGER.md). INFERENCE, high

A subtlety the referee insists on: statistical independence is split-property independence (a product state on the intermediate Type I factor / joint preparability), not factorization of the vacuum $\Omega$ itself — the vacuum does not tensor-factorize across the split.

BREAK 2 — Collar-dependence / failure at touching (UV singularity)

The split factorization requires a nonzero collar: $O_1 \Subset O_2$ with a finite buffer / safety-corridor. It fails when regions touch (UV singularities; universal divergence of entanglement entropy for adjacent regions) (Fewster, The split property for QFT in flat and curved spacetimes, Abh. Math. Semin. Univ. Hambg., 2016) ESTABLISHED, web-confirmed. There is no sharp factorization across a shared boundary, so the operationally natural "compose adjacent subsystems" has no exact tensor structure. [Referee addendum, if anything an under-claim: the split property can fail even at finite nonzero separation for Hagedorn / $T\bar T$ spectra (arXiv:2005.08972, PRD 102 045009) — so the collar leg is robust, not fragile.]

BREAK 3 — No composition-fixing theorem (verified)

The 2026 rigidity result fixes the probability rule, not the composition rule (subgoal (b), above). No 2018–2026 work reconstructs the full GPT framework (tensor composition + complex field) on Type III normal states or split inclusions OPEN; the existence-negative is thorough but inherently OPEN.

What moved

Iteration 3 logged this as an OPEN "domain-of-validity mismatch + missing-theorem gap, explicitly not a no-go." Iteration 4 moves it to a settled disjunction with a sharp line: no-go for exact composition on one factor; state-/collar-dependent yes for nested regions; with two named, web-verified failure modes (state-dependence; collar-dependence) and a verified missing-theorem leg (no composition-fixing theorem). The obstruction is now located and quantified, not merely flagged.

Furthest defensible claim

INFERENCE, high On a single Type III₁ factor, exact GPT local tomography / tensor composition is impossible (no factorization, no density matrix; Connes–Haag–Araki) ESTABLISHED. But the split property reinstates an exact factorization $\mathcal{H} \cong \mathcal{H}_1 \otimes \mathcal{H}_2$ with local density matrices and split-property statistical independence for nested regions $O_1 \Subset O_2$ separated by a nonzero collar (Buchholz; Doplicher–Longo 1984) ESTABLISHED; a canonical such factorization is singled out by a cyclic–separating reference state $\Omega$ and its modular conjugation $J$ (the vacuum being the distinguished physical choice) ESTABLISHED. So OP-48 is a no-go for exact composition on one factor and a state-/collar-dependent YES for nested regions — NOT a hard no-go — while no theorem (incl. arXiv:2602.09056 and its predecessor arXiv:2512.12636) yet fixes composition itself OPEN.

Tagging, per referee: the composite verdict is best tagged INFERENCE, high (its sub-facts are ESTABLISHED, but BREAK 1's circularity reading is INFERENCE and BREAK 3 is OPEN), not a flat ESTABLISHED. The source-JSON's tag: established on the furthest claim flattened a mixed-tag statement; this note adopts the referee's down-tag.

If it fails / weaker statement

If the state- and collar-dependence are shown ELIMINABLE — a state-independent, collar-free composition surrogate that single-handedly fixes tensor composition and the complex field on III₁ normal states — then OP-48 flips to a full YES, and the GPT framework becomes co-fundamental with modular geometry, upgrading [REG-SYNTH-6] (see ../HYPOTHESES.md) from a Born-form sublayer to the full framework.
If the vacuum-dependence is shown IRREDUCIBLE — every canonical split factorization must fix a geometric reference state — then the "split-approximate yes" collapses back to the iteration-3 thin methodological bridge: composition on III₁ is parasitic on pre-chosen geometry, and the A+B merge is re-description at the framework layer.
Weakest defensible fallback: even if the canonical construction's geometry-import is judged fatal, the existence (non-canonical) factorization with statistical independence still defeats the bare no-go. The honest floor is "no exact composition on one factor; non-canonical factorization for free on nested regions; canonical factorization costs a preferred state." OPEN where between these poles the verdict sits.

Proposed registry items (with referee verdicts)

Proposed statement. OP-48 resolves as a disjunction: (a) NO-GO for exact GPT local tomography / tensor composition on a single Type III₁ factor (no factorization / density matrix) ESTABLISHED; (b) state-/collar-dependent YES for nested regions $O_1 \Subset O_2$ under the split property — an exact $\mathcal{H} \cong \mathcal{H}_1 \otimes \mathcal{H}_2$ with local density matrices and statistical independence, made canonical only by the vacuum $\Omega$ (via modular $J$ of the relative commutant) and requiring a nonzero collar (fails for touching regions) ESTABLISHED; (c) OPEN whether a state-independent, collar-free composition surrogate can single out QM from split-composition axioms — no theorem, including arXiv:2602.09056 and its finite-dim predecessor arXiv:2512.12636, fixes composition itself. [tag: established]

Referee verdict — keep: true; severity: MINOR. Both citations verified real and on-point. Structural physics all confirmed (Connes–Haag–Araki for horn (a); Doplicher–Longo 1984, free-field split Buchholz 1974, $N \vee J N J$ , genuine factorization + W*-independence, nonzero-collar requirement and touching failure for horn (b)). The three horns address different objects (single factor; nested inclusion; state-independent collar-free surrogate), so the disjunction is internally consistent — a correct domain-of-validity decomposition, not a contradiction. Two minor refinements: (1) "made CANONICAL only by the vacuum" overstates exclusivity — canonicity is conferred by a choice of cyclic–separating vector, the vacuum being the distinguished physical choice (a different faithful normal GNS vector yields a different canonical factor; only the unitary-equivalence class is invariant). (2) Tag granularity: the bundle is mixed — (a) ESTABLISHED, (b) ESTABLISHED, (c) explicitly OPEN. Use per-horn tags rather than a single flat established; the meta-claim "OP-48 resolves AS this disjunction" is ESTABLISHED, but horn (c) itself is OPEN.

Refined statement (adopted). OP-48 resolves as a disjunction over different algebraic objects (no internal conflict): (a) ESTABLISHED NO-GO for exact local tomography / tensor composition on a single Type III₁ factor — no tensor factorization, no density matrix, no trace (Connes–Haag–Araki). (b) ESTABLISHED For nested regions $O_1 \Subset O_2$ under the split property (Buchholz 1974 for the free field; Doplicher–Longo 1984 in general), an exact $\mathcal{H} \cong \mathcal{H}_1 \otimes \mathcal{H}_2$ with local density matrices and statistical (W*-)independence exists via an intermediate Type I factor; a canonical such factor is $N \vee J N J$ , with $J$ the modular conjugation of the relative commutant (the collar) w.r.t. a chosen common cyclic–separating vector — the vacuum $\Omega$ the distinguished physical choice, not the unique one. It requires a nonzero collar and fails for touching / adjacent regions. (c) OPEN Whether a state-independent, collar-free composition surrogate can single out QM from split-composition axioms is unresolved: no theorem — arXiv:2602.09056 (infinite-dim) or arXiv:2512.12636 (finite-dim predecessor) — fixes composition; both lift only the Born-rule form. Bundle tag: MIXED — ESTABLISHED meta-resolution into a disjunction whose third horn is genuinely OPEN.

[HYP-SPLIT-GEOM] — new-hypothesis — keep (severity: MINOR) — refined

Proposed statement. The split-property "rescue" of GPT composition on III₁ does not escape the BW/CHM circularity but relocates it: the canonical intermediate Type I factor (hence the tomographic factorization and local density matrix) is selected by the vacuum $\Omega$ 's modular conjugation, so composition presupposes the very geometric reference state Program A aims to generate. [tag: inference]

Referee verdict — keep: true; severity: MINOR. Two scope corrections. (1) Existence vs. canonicity: the bare split property guarantees only existence of an intermediate Type I factor, of which there are infinitely many (Doplicher–Longo 1984; Hollands–Longo, CMP 2020 / arXiv:1911.09390); $J$ -selection is load-bearing for canonicity (a unique density matrix / entanglement entropy), not for the mere existence of a factorization with statistical independence. Factorization + a (non-unique) local density matrix exist for ANY split inclusion without invoking $\Omega$ 's modular data. (2) "Geometric" overshoots: a standard split inclusion needs $\Omega$ cyclic–separating (so $J$ exists) — strictly weaker than the BW/CHM demand that $\Omega$ be the symmetric vacuum whose modular flow is a geometric boost. The relocation lands on "a preferred cyclic–separating reference state with well-defined modular data"; it coincides with the geometric BW/CHM circularity only under the additional geometric-modular-flow hypothesis (the step isolated in [A-CP-KILLING], see ../ASSUMPTIONS_LEDGER.md). So the correct verb is "adds a canonicity-cost that compounds, and under the geometric-modular-flow hypothesis coincides with, the BW/CHM circularity" — not "is the BW/CHM circularity." Net: the hypothesis strengthens, not refutes, the encodes-not-generates verdict.

Refined statement (adopted). INFERENCE The split-property softening of BREAK 1 supplies factorization and statistical independence via some intermediate Type I factor, but such factors are non-unique (infinitely many). The canonical intermediate Type I factor — hence a canonical, choice-independent tomographic factorization and a canonical local density matrix / entanglement entropy — is selected by the modular conjugation $J$ of the cyclic–separating reference state $\Omega$ (Doplicher–Longo 1984; Hollands–Longo 2020). Therefore instantiating a canonical local-tomography-type GPT axiom on a split inclusion presupposes a preferred reference state with well-defined modular data. This is a distinct circularity from the entropy-matching / geometric-modular-flow input already logged ([A-CP-KILLING]; iteration-3 BREAK 3): it requires only $\Omega$ cyclic–separating, not its modular flow being geometric, and coincides with BW/CHM only under the additional symmetric-vacuum hypothesis. Net: the split property removes the bare no-go but, for a canonical construction, re-imports a preferred state — compounding rather than equalling BW/CHM, and reinforcing the encodes-not-generates verdict.

Verdict

SETTLED as a sharp disjunction (medium confidence). OP-48 / OP-GPT-TYPE3 is:

a NO-GO for exact composition / local tomography on a single Type III₁ factor ESTABLISHED — no factorization, no density matrix; and
a qualified, state- and collar-dependent YES for nested local algebras under the split property [ESTABLISHED — exact factorization $\mathcal{H} \cong \mathcal{H}_1 \otimes \mathcal{H}_2$ , canonical given a cyclic–separating $\Omega$ , with a nonzero collar].

The split property defeats the would-be hard no-go (a local-tomography-type statement IS formulable for split inclusions) but does not deliver a clean operational reconstruction, for two web-verified reasons: (1) the canonical factorization is state-dependent (selected by $\Omega$ 's modular $J$ ; infinitely many non-canonical factors otherwise), re-importing the BW/CHM geometric input as a canonicity tax; (2) it requires a nonzero collar and fails for touching / adjacent regions, so "compose adjacent subsystems" has no exact tensor structure.

Subgoal (b): the 2026 σ-affinity + normal-steering result cannot currently be extended from Born form to composition — verified that it (arXiv:2602.09056) and its finite-dim predecessor (arXiv:2512.12636), both Torres Alegre, treat only the single-system probability rule and explicitly do not derive composition.

Subgoal (c): NO-GO for exact composition on one factor; split-approximate (state-/collar-dependent) yes for nested regions; the full operational reconstruction singling out QM from split-composition axioms remains OPEN — no such theorem exists.

This sharpens the iteration-3 obstruction (no longer merely a "missing-theorem gap" but a located no-go-vs-conditional-yes line with two named, verified failure modes plus a verified missing theorem) without advancing to a positive reconstruction. didItMove = sharpened-obstruction. Confidence: medium (the structural facts are ESTABLISHED; the "irreducibility of the vacuum-dependence" that would harden the no-go is INFERENCE/OPEN).

Open subquestions

OPEN — decides HYP-SPLIT-GEOM Is the vacuum-state-dependence of the canonical split factorization irreducible — must every canonical intermediate Type I factor fix a geometric reference state — or does a state-independent composition surrogate exist on III₁ normal state spaces?
OPEN Can the 2026 causal-rigidity theorem (σ-affinity + normal steering) be strengthened to fix the composition rule (tensor product + complex field) on infinite-dim normal state spaces, not just the probability rule $\Phi = \mathrm{id}$ , without assuming factorization?
OPEN Does a collar-free / coincidence-limit composition axiom exist that survives the touching-region UV singularity, or is the finite collar an irreducible obstruction to composing adjacent operational subsystems?
OPEN Is there any complete GPT-style reconstruction that uniquely singles out QM using split-based composition + a modular-compatible continuous-reversibility surrogate + σ-additive purification on the hyperfinite III₁ normal state space? (No such theorem currently exists.)
OPEN — unverified Is the peer-review status of arXiv:2602.09056 and arXiv:2512.12636 (both Torres Alegre) confirmed, and does "normal steering via purification" covertly require a Type I/II structure that would shrink even the Born-form sublayer?