Documentation

CsdLean4.CV.EntanglementDistance

ST-2: the entanglement distance on the composite arena #

Category: 3-Local (CV; entanglement geometry at finite dimension). BACKLOG #38, brick ST-2 of records-to-spacetime-scoping.md.

ST-1 (RecordInfluence.lean) gave the composite arena a causal shape from the coupling graph. This file gives it a correlational one: the mutual information of two sectors, and the separation functional d = −log I that reads it as a distance. The scoping note prices this as research rather than as something the record layer needs, and says so; it is here because the author asked for entanglement geometry as a direction.

Mutual information #

The distance, and the arena #

Honest scope #

⚠️ entDist is a separation functional, not a proved metric. No triangle inequality is proved or claimed, here or anywhere in the corpus. Calling −log I a "distance" is the physics convention this file adopts; the properties proved are the three the scoping note asked for — the sign, the product-point value, and monotonicity — and nothing beyond them.

⚠️ Monotonicity carries the corpus's SSA hypothesis. mutualInfoTri_mono is QuantumInfo.strong_subadditivity_of_relEntropy_monotone rearranged, so it inherits that theorem's explicit data-processing bound hDPI rather than assuming it silently: the corpus does not prove SSA unconditionally at this pin, and this file does not pretend otherwise.

⚠️ This is not geometry from records, and not a step toward emergent spacetime. It is a correlation functional on a finite-dimensional composite arena. The scoping note's finding is unchanged: what is missing for emergence proper is which coarse-graining projection defines the macroscopic coordinates, and that is the author's decision rather than a lemma. Nothing here feeds the record layer; ST-1 is the brick that does.

⚠️ Finite cutoff throughout (FieldConfig K N), and the arena statements are about the join (product) points; the entangled case is covered only through mutualInfo_pureDensity, as twice a marginal entropy, with no numerical value computed for any particular entangled ray.

References: records-to-spacetime-scoping.md ST-2; Mathlib/QuantumInfo/{Entropy, Subadditivity, StrongSubadditivity}.lean; CV/{CompositeArena, EntangledWeights}.lean; specs/BACKLOG.md #38.

Mutual information of a bipartite density #

Entropy depends only on the matrix, not on the Hermitian witness or on how the matrix is written.

noncomputable def CSD.CV.mutualInfo {n : Type u_1} {m : Type u_2} [Fintype n] [DecidableEq n] [Fintype m] [DecidableEq m] {ρ : Matrix (n × m) (n × m) ℂ} (hpsd : ρ.PosSemidef) :

The mutual information I(A:B) = S(ρ_A) + S(ρ_B) − S(ρ_AB) of a bipartite density, read off the two partial traces.

Equations
Instances For
    theorem CSD.CV.mutualInfo_congr_matrix {n : Type u_1} {m : Type u_2} [Fintype n] [DecidableEq n] [Fintype m] [DecidableEq m] {ρ σ : Matrix (n × m) (n × m) ℂ} (h : ρ = σ) (hρ : ρ.PosSemidef) (hσ : σ.PosSemidef) :

    Mutual information depends only on the matrix.

    theorem CSD.CV.mutualInfo_nonneg {n : Type u_1} {m : Type u_2} [Fintype n] [DecidableEq n] [Fintype m] [DecidableEq m] {ρ : Matrix (n × m) (n × m) ℂ} (hpsd : ρ.PosSemidef) (htr : ρ.trace = 1) (hpdA : (QuantumInfo.partialTraceRight ρ).PosDef) (hpdB : (QuantumInfo.partialTraceLeft ρ).PosDef) :

    ★★ Mutual information is non-negative — subadditivity of the von Neumann entropy, with the standard positive-definite support condition on the two marginals.

    theorem CSD.CV.mutualInfo_kronecker {n : Type u_1} {m : Type u_2} [Fintype n] [DecidableEq n] [Fintype m] [DecidableEq m] {ρ : Matrix n n ℂ} {σ : Matrix m m ℂ} (hpsdρ : ρ.PosSemidef) (hpsdσ : σ.PosSemidef) (htrρ : ρ.trace = 1) (htrσ : σ.trace = 1) :

    ★★★ A product state carries no mutual information. The marginals of ρ ⊗ σ are ρ and σ and the entropy is additive over the tensor product, so I = 0 exactly.

    theorem CSD.CV.pureDensity_mul_self {n : Type u_1} {m : Type u_2} [Fintype n] [Fintype m] {ψ : n × m → ℂ} (hψ : ∑ p : n × m, ‖ψ p‖ ^ 2 = 1) :

    A unit vector's pure density is a projection.

    theorem CSD.CV.mutualInfo_pureDensity {n : Type u_1} {m : Type u_2} [Fintype n] [DecidableEq n] [Fintype m] [DecidableEq m] {ψ : n × m → ℂ} (hψ : ∑ p : n × m, ‖ψ p‖ ^ 2 = 1) (hpsd : (QuantumInfo.pureDensity ψ).PosSemidef) :

    ★★★ For a pure bipartite state the mutual information is twice the marginal entropy. The joint entropy vanishes and the two marginals have equal entropy, so I(A:B) = 2 S(ρ_A): on a pure state the mutual information is the entanglement entropy, doubled.

    The entanglement distance #

    noncomputable def CSD.CV.entDist (I : ℝ) :

    The entanglement distance of a mutual information: d = −log I, with the vanishing case sent to ⊤. Unentangled sectors are infinitely far apart; a mutual information of at least 1 nat puts them at distance zero. ⚠️ This is a separation functional, not a proved metric: no triangle inequality is claimed anywhere in this file.

    Equations
    Instances For
      @[simp]
      theorem CSD.CV.entDist_of_nonpos {I : ℝ} (h : I ≤ 0) :
      theorem CSD.CV.entDist_antitone {I J : ℝ} (hI : 0 < I) (hIJ : I ≤ J) :

      ★ More correlation is less distance: the distance is antitone in the mutual information.

      theorem CSD.CV.entDist_eq_zero_of_one_le {I : ℝ} (h : 1 ≤ I) :

      ★ A mutual information of a nat or more sits at distance zero.

      On the composite arena #

      noncomputable def CSD.CV.compositeDM {K₁ K₂ N : ℕ} (x : FieldArena (K₁ + K₂) N) :
      Matrix (FieldConfig K₁ N × FieldConfig K₂ N) (FieldConfig K₁ N × FieldConfig K₂ N) ℂ

      The composite arena density in product coordinates: the matrix whose partial traces are the two sectors' reduced states.

      Equations
      Instances For
        theorem CSD.CV.mutualInfo_compositeDM_join {K₁ K₂ N : ℕ} (p : FieldArena K₁ N) (q : FieldArena K₂ N) (hpsd : (compositeDM (arenaJoin p q)).PosSemidef) :
        mutualInfo hpsd = 0

        ★★★ A product point of the composite arena carries no mutual information. The join of two single-sector points has I = 0: on the arena, zero mutual information is exactly the product-point case this theorem names, and the entanglement distance there is infinite.

        theorem CSD.CV.entDist_mutualInfo_join {K₁ K₂ N : ℕ} (p : FieldArena K₁ N) (q : FieldArena K₂ N) (hpsd : (compositeDM (arenaJoin p q)).PosSemidef) :

        ★★ The two sectors of a product point are infinitely far apart.

        Monotonicity under sector enlargement #

        ★★ Enlarging a sector cannot decrease the mutual information, hence cannot increase the entanglement distance: I(A:B) ≤ I(A:BC). This is strong subadditivity rearranged, and it carries the same hypothesis the corpus's SSA carries — the data-processing bound hDPI, left explicit rather than assumed (QuantumInfo.strong_subadditivity_of_relEntropy_monotone).