Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-adapted
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Pure-state separability is equivalent to being a product: a Schmidt overlap witness

Statement

Assume Countable Choice (The Axiom of Countable Choice (ACω)). Let H,K be arbitrary nonzero complex Hilbert spaces, with inner products linear in the first argument, and put L=H⊗^K with the isometric inclusion κ of Hilbert tensor construction: inner product, completion and product bases. Write Tx,yz=⟨z,y⟩x as in Rank-one operator calculus. Set P={Tκ(a⊗b),κ(a⊗b):∥a∥=∥b∥=1},S=conv⁡P‾∥⋅∥1, with trace class as in Trace class operator, where the convex hull consists of finite convex combinations, and the closure is taken in the trace-class operators on L. For each unit Ψ∈L, let PΨ=TΨ,Ψ. Then PΨ∈S⟺Ψ=κ(a⊗b) for some unit a∈H, b∈K.

More precisely, let Ψ=∑r∈Jsrκ(vr⊗ur) be any finite or countably infinite positive Schmidt expansion from Infinite Schmidt decomposition and the pure product criterion, and put λr=sr2. The number m=max⁡r∈Jλr exists. Every pair of unit vectors a,b satisfies ∣⟨Ψ,κ(a⊗b)⟩∣2≤m, with equality at a Schmidt pair whose weight is m. Every σ∈S satisfies tr⁡(σPΨ)≤m, whereas tr⁡(PΨ2)=1. If Ψ is non-product then m<1 and, for every σ∈S, ∥PΨ−σ∥1≥1−m>0. The factors need not be separable; only the support of the single tensor vector has a countable Schmidt expansion. These are mathematical trace pairings, with no measurement or Born-rule assumption.

Facts & Assumptions

Given: Countable Choice; the Hilbert spaces H,K; L, κ, P, S and unit Ψ as above.

[F1]

The completed tensor product is Hilbert, ∥κ(a⊗b)∥=∥a∥∥b∥, and ⟨κ(x⊗y),κ(a⊗b)⟩=⟨x,a⟩⟨y,b⟩, with linearity in the first argument (Hilbert tensor construction: inner product, completion and product bases clauses (a), (c)).

[F2]

Under Countable Choice, an arbitrary unit tensor vector has a nonempty at most countable positive Schmidt expansion, with orthonormal families (vr), (ur) and ∑rsr2=1; it is a product vector exactly when there is one positive coefficient (Infinite Schmidt decomposition and the pure product criterion, Statement). No ambient Hilbert basis is assumed.

[F3]

Rank-one operators are positive when their two vectors agree, have trace tr⁡Tx,y=⟨x,y⟩, and for trace-class T satisfy tr⁡(TTx,y)=⟨Tx,y⟩. Trace is linear by Trace is absolutely convergent and basis independent clause 4, and ∥T∥≤∥T∥1 by Trace class is a two sided Banach operator ideal clause 1; the rank-one identities are Rank-one operator calculus clauses (b), (c), (e).

Proof

Given: The data of the statement and [F1]–[F3].

1.1givenF2algebra

[Maximum and the strict gap] By [F2], λr>0 and ∑rλr=1. Fix one r0∈J and put c=λr0>0. The set E={r:λr≥c} is nonempty and finite: any finite subset of E with N elements has weight sum at least Nc≤1, so E has at most ⌊1/c⌋ elements. The maximum m on this finite set is a maximum on J, since every weight outside E is below c≤m. If Ψ is non-product, [F2] gives at least two positive weights; at an index r∗ attaining m, another positive weight implies m<∑rλr=1.

1.2givenF1algebra

[Finite orthogonal estimates] For an orthonormal family (er), a vector x and a finite set F, expanding ∥x−∑r∈F⟨x,er⟩er∥2 gives ∥x∥2−∑r∈F∣⟨x,er⟩∣2≥0. Thus for unit a,b the sums of ∣⟨vr,a⟩∣2 and ∣⟨ur,b⟩∣2 are each at most one, using conjugate symmetry and the supremum of finite sums. For finite complex sequences (αr),(βr), Cauchy–Schwarz gives ∣∑rαrβr∣≤(∑r∣αr∣2)1/2(∑r∣βr∣2)1/2: this follows by expanding ∑r∣αr−tβr‾∣2≥0 with t=(∑rαrβr)/(∑r∣βr∣2) when the denominator is positive, and is immediate otherwise.

1.3givenF3algebra

[Exact continuity of the fixed trace witness] For every trace-class T on L, [F3] gives tr⁡(TPΨ)=⟨TΨ,Ψ⟩ and ∣tr⁡(TPΨ)∣≤∥TΨ∥∥Ψ∥≤∥T∥≤∥T∥1, since ∥Ψ∥=1. Applying this estimate to T−S proves ∣tr⁡(TPΨ)−tr⁡(SPΨ)∣≤∥T−S∥1. Thus this particular bounded trace functional is 1-Lipschitz; no unspecified continuity theorem or unbounded cyclicity is used. Also PΨΨ=Ψ, so tr⁡(PΨ2)=⟨PΨΨ,Ψ⟩=1.

2.1step 1.1step 1.2F1F2algebra

[The overlap bound, with the conjugations fixed] For a finite Schmidt truncation ΨF, [F1] gives ⟨ΨF,κ(a⊗b)⟩=∑r∈Fsr⟨vr,a⟩⟨ur,b⟩; equivalently it is the conjugate of ∑r∈Fsr⟨a,vr⟩⟨b,ur⟩. Applying step 1.2 to αr=sr⟨vr,a⟩ and βr=⟨ur,b⟩ gives a modulus bound (∑r∈Fλr∣⟨vr,a⟩∣2)1/2(∑r∈F∣⟨ur,b⟩∣2)1/2≤m. The Schmidt truncations converge in norm [F2], so continuity of the Hilbert pairing passes this bound to Ψ. At a=vr∗ and b=ur∗, orthonormality gives overlap sr∗ and squared overlap m.

3.1step 2.1step 1.3F3algebra

[Convex hull and its closure] For unit product q=κ(a⊗b), [F3] and first-linear conjugate symmetry give tr⁡(Tq,qPΨ)=⟨Tq,qΨ,Ψ⟩=⟨Ψ,q⟩⟨q,Ψ⟩=∣⟨Ψ,q⟩∣2≤m by step 2.1. Trace linearity gives the same real bound for finite convex combinations. For σ∈S and every ε>0, the definition of closure supplies a single τ∈conv⁡P with ∥σ−τ∥1<ε. Step 1.3 gives ∣tr⁡(σPΨ)−tr⁡(τPΨ)∣<ε; letting ε↓0 shows that tr⁡(σPΨ) is real and at most m. This neighborhood argument does not select a sequence of approximants.

4.1step 1.1step 1.3step 3.1F1algebra∎

[Equivalence and distance] If Ψ is non-product, steps 1.1, 1.3 and 3.1 give tr⁡(PΨ2)=1>m≥tr⁡(σPΨ) for all σ∈S, so PΨ∉S. Moreover step 1.3 gives ∥PΨ−σ∥1≥∣tr⁡((PΨ−σ)PΨ)∣=1−tr⁡(σPΨ)≥1−m. Conversely a unit product vector has PΨ∈P⊆S. If initially Ψ=κ(x⊗y) with arbitrary nonzero factors, [F1] gives ∥x∥∥y∥=1; therefore a=x/∥x∥ and b=∥x∥y are unit and Ψ=κ(a⊗b). This proves the full claimed equivalence for the declared trace-norm closure and the strict witness bound.

Depends on

Dependency tree · 2 levels

Sources