Contribution
The paper makes Schrödinger’s ensemble steering a precise property of a bipartite state in any finite-dimensional ordered-linear-space GPT (Def. 5.1). It then characterises steering order-theoretically: ω steers its B-marginal iff ω̂ restricted to the unit interval is a strong quotient map of ordered sets (Thm 5.4). Combined with a purification characterisation of homogeneity (Thm 4.1), this yields the headline:
- a theory in which every system can steer every one of its own states from a composite of two copies (universal self-steering) has only homogeneous, weakly self-dual irreducible state spaces (Prop. 5.8).
Before this paper, homogeneity and self-duality entered reconstructions as geometric postulates. After it, homogeneity and weak self-duality have an operational reading.
Key insight
A bipartite state is a positive map ω̂ : A* → B, and steering is surjectivity of that map in an order sense: every ensemble of the marginal ω_B = ω̂(u_A) must be the image of an observable, so ω̂ has to carry the effect interval [0, u_A] onto [0, ω_B] chain by chain. When the dimensions match, surjectivity is bijectivity, so a steering state is an order-isomorphism between A* and B. An order-isomorphism that can be made to hit every interior state as u_A’s image is precisely a transitive automorphism group. That is the homogeneity of the cone. Steering thus converts to geometry with almost no calculation. The remaining distance to quantum theory is the difference between “A* ≅ A somehow” and “A* ≅ A via an inner product”.
Assumptions
- Finite dimension throughout (§2). All spaces are identified with their double duals.
- Abstract state spaces (Def. 2.1). An ordered linear space A with a regular (pointed, closed, generating) cone A+ and a strictly positive order-unit functional u_A. Normalised states are Ω_A = {α ∈ A+ : u_A(α) = 1}.
- No-restriction on effects (p. 6). Every effect a ∈ [0, u_A] ⊆ A* “represents a measurement outcome”. This is “the further assumption”. Footnote 8 notes that the tensor-product notion depends on it.
- Composites lie between min and max (p. 7). A composite AB is (A*⊗B*)* ordered by any cone between A ⊗min B (generated by product states) and A ⊗max B (forms positive on product effects).
- Local tomography is adopted “as a working assumption” (p. 7). A bipartite state is fixed by its values on product effects.
- A “theory” (p. 7) is a class of abstract state spaces closed under some such composite.
- Scope of the headline results. Props 5.7–5.8 are stated for irreducible state spaces. The extension to all state spaces (p. 15) needs one more assumption: every direct summand of a state space in the theory is itself in the theory.
- Strong self-duality is not assumed. It appears only as a hypothesis in §6.
Key results
-
Thm 3.3. If A is irreducible, every order-automorphism of A lies on an extremal ray of the cone L+(A, A) of positive maps. Proof on pp. 8–9, by a basis-of-extremal-rays argument. Example 3.4 shows irreducibility is needed.
-
Cor. 3.5. For irreducible A, isomorphism states are pure in A ⊗max B.
-
Cor. 3.6. If ω̂ factors isomorphically through an irreducible face of A*, then ω̂ is extremal in the cone of positive maps.
-
Thm 4.1. Two conditions are equivalent:
- (a) A is homogeneous;
- (b) every normalised interior state is the A-marginal of an isomorphism state in B ⊗max A, for one fixed B ≅ A*.
Both directions are proved (p. 11).
-
Cor. 4.2. For irreducible A, two conditions are equivalent:
- (a) A is weakly self-dual and homogeneous;
- (b) every normalised interior state is the marginal of an isomorphism state in A ⊗max A.
-
Lemma 5.2. If ω steers ω_B, then ω̂(A+) = Face(ω_B).
-
Example 5.3. Lemma 5.2 has no converse: a classical R³+ ⊗ R²+ state satisfies the face condition but is not steering.
-
Thm 5.4. ω steers its B-marginal iff ω̂ : [0, u_A] → [0, ω_B] is a strong quotient map of ordered sets (every chain is the image of a chain). Both directions are proved (p. 14).
-
Cor. 5.5. If ω steers an interior ω_B and ω̂ is injective, then ω̂ is an order-isomorphism, and it is pure if B+ is irreducible. When dim A = dim B, the states steering an interior marginal are exactly the isomorphism states, which steer both marginals.
-
Prop. 5.7. Uniform universal steering ⇒ every irreducible state space is homogeneous. The support is thin; see corrections.
-
Prop. 5.8. Universal self-steering ⇒ every irreducible state space is homogeneous and weakly self-dual. With the direct-summand assumption, the conclusion extends to all state spaces (p. 15 and fn. 7).
-
Fact B.3. For two qubits, the steering product ⊗str (Def. B.1: the cone generated by all steering states) equals ⊗max, which is not weakly self-dual. So “include all steering states” does not yield a weakly self-dual composite, even for quantum factors.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Order-automorphisms of an irreducible cone are extremal positive maps; isomorphism states are pure | strong | proof, Thm 3.3 and Cor. 3.5 (Example 3.4 shows irreducibility is needed) |
| C2 | A cone is homogeneous iff every interior state purifies to an isomorphism state with one fixed ancilla ≅ A* | strong | proof, Thm 4.1 |
| C3 | Steering its marginal ⇔ ω̂ is a strong quotient map [0,u_A] → [0,ω_B] | strong | proof, Thm 5.4 |
| C4 | Whether every quotient map of order intervals is automatically strong | open | Remark (i), p. 14: “we suspect, but so far have been unable to prove” |
| C5 | Uniform universal steering ⇒ homogeneity of irreducible state spaces | weak | one-line citation of Cor. 5.5 + Thm 4.1 (p. 15). It covers only the injective case, and a non-injective steering map is not addressed |
| C6 | Universal self-steering ⇒ homogeneous and weakly self-dual | strong | Cor. 5.5 + Cor. 4.2 (matching dimensions make ω̂ bijective) |
| C7 | Self-steering + strong self-duality + local tomography + a qubit ⇒ finite-dimensional C*-algebraic QM (with superselection) | weak | §6 argument. It cites Koecher–Vinberg, Jordan–von Neumann–Wigner and Hanche-Olsen [19], but the operational-to-JB-tensor-product match is “established elsewhere”, and strong self-duality is a hypothesis |
| C8 | The steering product of two qubits is ⊗max, so it is not weakly self-dual | moderate | proof sketch, Fact B.3 (via decomposability of qubit positive maps) |
| C9 | ⊗str is associative | open | Conjecture B.2 |
| C10 | In a nonclassical theory where all states can be steered, the Barnum–Dahlsten–Leifer–Toner bit-commitment protocol cannot exist | weak | assertion, p. 17. The paper itself calls the connection only “suggestive” |
| C11 | Conclusive teleportation through a copy of the system forces weak self-duality | moderate (cited) | cited to [6], with extra condition fn. 9; not proved here |
Method
Pure order theory on finite-dimensional cones:
- represent bipartite states as positive maps ω̂ : A* → B (§3);
- use extremal-ray arguments for purity (Thm 3.3);
- construct a variable dual A♦ = (A*, α_o) for purification (§4);
- translate ensembles into chains in [0, ω_B] (Thm 5.4).
There are no numerical or experimental components. The examples (App. A) are square-base (gbit) and cube/hexagon cones, computed by hand.
Concepts
- Weakly self-dual (Def. 3.1): there exists some order-isomorphism η : A* ≃ A. Self-dual (strong) means the isomorphism comes from an inner product. The paper uses the word only informally, in §1 and §6.
- Homogeneous: the order-automorphism group acts transitively on the interior of A+.
- Isomorphism state (Def. 3.2): ω with ω̂ : A* → B an order-isomorphism. An automorphism state is one on A ⊗ A.
- Ensemble (p. 12): a finite family βᵢ ∈ B+ with Σβᵢ = β. The states are unnormalised, and u_B(βᵢ) are the weights.
- Steering (Def. 5.1): every ensemble of ω_B equals {ω̂(xᵢ)} for some observable {xᵢ} on A. Bisteering means steering for both marginals.
- Strong quotient map (p. 14): an order-preserving surjection under which every finite chain lifts to a chain.
- Universal / uniform universal / universal self-steering (Def. 5.6), in increasing strength:
- universal: each state has its own steering partner;
- uniform universal: one partner system per system;
- universal self-steering: that partner is a copy of the system itself.
- A♦: A* made into a state space by choosing an interior α_o as order unit. It is non-canonical.
- Steering product ⊗str (Def. B.1): the cone generated by all steering states in ⊗max.
Connections
It builds on the Barnum–Barrett–Leifer–Wilce ordered-linear-space programme ([4]–[7]: cloning/broadcasting, teleportation, entropy). It is a companion to Wilce’s “Four and a half axioms” ([32], arXiv:0912.5530, posted the same day), which the conclusion names as one strategy for bridging the weak/strong gap.
Against the GPT introduction, LIT-073 (reads/c21.md). The review is the right vocabulary source for this paper: state space, effects, ⊗min/⊗max, local tomography. Two things do not carry over:
- “Steering” is not the same notion in the two works. LIT-073’s steering is channel steering (Defs 7.25–7.27, after Plávala), used as an entanglement-assisted incompatibility test. This paper’s is Schrödinger’s ensemble steering of a marginal. The review cites this paper ([2]) but does not state its results.
- Self-duality is absent from LIT-073. The review never defines it (c21’s correction). So a reader coming from LIT-073 needs this paper’s Def. 3.1 and the Koecher–Vinberg paragraph (p. 3) directly.
The two agree on the structural fact used in Appendix B: for non-simplicial factors ⊗min ≠ ⊗max. That is why neither is weakly self-dual here (p. 23), and it matches LIT-073’s Thm 5.21 (one factor simplicial ⇔ min = max).
Against PBR, LIT-062 (reads/c20.md). There is no technical overlap. PBR works inside quantum theory with ontological models and a preparation-independence assumption. This paper works outside quantum theory with no ontic layer. Their common ancestor is Schrödinger’s 1935–36 discussion of entanglement ([28] here). Reading one does not help with the other.
Later literature (not read here, pointers only). Barnum–Duncan–Wilce, “Symmetry and self-duality in categories of probabilistic models” (arXiv:1210.0622), and Müller–Ududec (PRL 2012) give other routes to self-duality. I have not checked whether either closes this paper’s gap in its stated form.
Bearing on the record
- It supports any theory-level summary of GPT reconstructions that names steering as one of the operational routes to Jordan-algebraic state spaces. Such a document should cite Prop. 5.8, not Prop. 5.7, and should state the C*-conclusion as conditional (C7).
- Nothing in the record cites it for a claim it does not make, as far as the seeded LIT-081 goes. LIT-081’s summary sentence “Koecher–Vinberg plus Hanche-Olsen force … C*-algebraic quantum theory” should be softened to “would force, given strong self-duality, which the paper does not derive”.
- No bearing on ML practice. Nothing here carries an instruction for the Anthology of the SOTA.
Limitations
- The weak-to-strong self-duality gap is the paper’s own stated limitation (§6, p. 17). Without closing it, the chain ends at “homogeneous and weakly self-dual”, which is well short of Jordan algebras.
- The C*-step relies on a Hanche-Olsen correspondence proved “elsewhere”.
- Prop. 5.7’s proof covers only the injective case (see corrections).
- Several statements are left open:
- Remark (i): is every quotient a strong quotient?
- Conjecture B.2: is ⊗str associative?
- Questions 1–3 of Appendix B.
- The universal vs uniform universal steering gap (p. 15).
- Finite dimension, no-restriction and local tomography are standing assumptions. Real-Hilbert-space QM violates local tomography, and the paper notes that it thereby falls outside the framework (fn. 5).
- The v2 preprint shows signs of being unfinished: an author’s marginal note, dangling cross-references and typos.
Open questions
- Does homogeneity plus weak self-duality plus a reasonable composite force strong self-duality? The paper states this as the decisive question (p. 17). Showing that every homogeneous weakly self-dual cone admitting a locally tomographic composite with a qubit is self-dual would close it. So would a counterexample category.
- Is there a class of weakly-but-not-strongly self-dual homogeneous state spaces closed under a non-signalling tensor product? The steering product fails (Fact B.3).
- Does universal (non-uniform) steering already force homogeneity (p. 15)?
- Is every quotient map [0,u] → [0,v] a strong quotient (Remark (i), p. 14)? A proof would simplify Thm 5.4 to a plain quotient condition.
Corrections to the seeded skim
- The dossier’s “and Koecher–Vinberg plus Hanche-Olsen force … C-algebraic quantum theory” is not a result of this paper.* It is the §6 programme. The paper proves homogeneity and weak self-duality (Props 5.7–5.8). Strong self-duality is assumed hypothetically (“If one could motivate the stronger assumption …”, p. 16). The claim that Hanche-Olsen’s JB-tensor-product conditions are “exactly the operational requirements we’ve imposed” is deferred (“As we will establish elsewhere”, p. 16). The C*-conclusion is therefore conditional on an assumption the paper does not make and on an argument it does not give.
- Prop. 5.7 (uniform universal steering ⇒ homogeneity) is under-argued. Its proof is one line: “Corollary 5.5, combined with Theorem 4.1, establish”. Cor. 5.5 turns a steering state into an order-isomorphism only when ω̂ is injective. Thm 4.1(b) needs isomorphism states. For a steering system A_B of larger dimension than B, a steering ω̂ is a surjection that need not be injective, and the paper does not say how homogeneity follows then. The gap closes for self-steering (Prop. 5.8), where the dimensions match: ω̂ onto B+ for an interior marginal forces bijectivity. Whether Prop. 5.7 holds in general is, on this text, unverified. The paper itself flags a neighbouring gap, between universal and uniform universal steering, as open (p. 15).
- The dossier links the paper to the ψ-ontology debates (PBR, LIT-062). The text has no ontological-model content. It never mentions ψ-epistemic or ψ-ontic models, preparation independence or hidden variables. The link to LIT-062 is thematic (Schrödinger 1935–36 is ancestral to both) and nothing more.
- v2 still carries an unresolved author’s marginal note after Def. 3.1 (p. 8): “[HB: Not clear to me this follows. More argument needed…]”. It concerns the claim that η⁻¹ of a normalised isomorphism state is a bipartite effect. None of the main results depends on that sentence, but it shows the preprint was posted mid-revision.
- Cross-reference and typo errors a reader will trip on:
- “Example 2.4 above” (p. 12) and “Theorem 4.2” (p. 12) do not exist; the latter means Cor. 4.2.
- Cor. 5.5 begins “et ω”.
- In the proof of Thm 5.4, “β_j = Σ_{i≤j} β_i” should be η_i.
- In Example 5.3 the text’s description of the row and column indices is swapped against the 3×2 table.
- Def. 5.6 writes “every state β ∈ A” where B is meant.
- The dossier says “§4 (purification: every interior state has a purification in A♦⊗max A)”. That is accurate, but only with the paper’s own caveat: the ancilla A♦ = (A*, α_o) depends on α_o. The state-independent version is exactly homogeneity (Thm 4.1). So “every state can be purified” is not the general-GPT result; “every state can be purified with an ancilla chosen for it” is.