AFTD Auto-Formalizing Theoretical Domains

ppt_qutrit_source_normalized_gram_if_three_lemmas

theoremverified

The explicit source has a normalized PPT-invariant Gram factor on every rank-two projection, conditional on basis selection, source partial-transpose invariance, and real positive-block normalization.

Statement

theorem ppt_qutrit_source_normalized_gram_if_three_lemmas
    (hselect :   ∀ P : Matrix (Fin 3) (Fin 3) ℂ, P.IsHermitian → P*P=P → P.rank=2 →
    ∃ U : Matrix (Fin 3) (Fin 2) ℂ,
      Uᴴ*U=1 ∧ U*Uᴴ=P ∧
      (let W : Matrix (Fin 3) (Fin 2) ℂ := ppt_qutrit_steered_normalizer * U
       let lam : Fin 4 → ℝ := fun t => if t = 0 then 3257/6884 else if t = 3 then 27/6884 else 450/1721
       let G : Matrix (Fin 3) (Fin 4) ℂ := fun i t =>
        (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x 0) * ppt_qutrit_bell_vector t (i,x))
       G.rank=3))
    (hstate : ∀ i x j y : Fin 3,
      ppt_qutrit_bell_state (i,x) (j,y) = ppt_qutrit_bell_state (j,x) (i,y))
    (hnorm : ∀ R : Matrix (Fin 3) (Fin 3) ℝ, R.PosDef →
      ∃ H : Matrix (Fin 3) (Fin 3) ℝ, H.transpose = H ∧ IsUnit H ∧ H*R*H = 1) :
  ∀ P : Matrix (Fin 3) (Fin 3) ℂ, P.IsHermitian → P*P=P → P.rank=2 →
    ∃ (U : Matrix (Fin 3) (Fin 2) ℂ) (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ)
      (T : Matrix (Fin 3) (Fin 3) ℂ),
      Uᴴ*U=1 ∧ U*Uᴴ=P ∧
      (let V : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; V*Vᴴ=1) ∧
      (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧
      (let W : Matrix (Fin 3) (Fin 2) ℂ := ppt_qutrit_steered_normalizer * U
       let lam : Fin 4 → ℝ := fun t => if t = 0 then 3257/6884 else if t = 3 then 27/6884 else 450/1721
       ∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4,
        (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x p) * ppt_qutrit_bell_vector t (i,x)) =
          ∑ a : Fin 3, T i a * F (a,p) t)
Source
Source-specific helper for OP-693, based on https://arxiv.org/abs/1405.4502.
Verified
12 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 4
  • ppt_qutrit_bell_stateThe explicit two-qutrit positive-partial-transpose state in the Bell-violation example.
  • ppt_qutrit_steered_normalizerThe diagonal inverse square root of the second-qutrit reduced state.
  • ppt_qutrit_bell_vectorFour explicit real vectors in two-qutrit space used in the PPT Bell-violation example.
  • FThe prime field $\mathrm{GF}(p)$, realized as $\mathbb{Z}/p\mathbb{Z}$ via ZMod.
Used by 3

The proof, explained

The theorem proves that any rank-2 Hermitian projection P on ℂ³ admits an isometric embedding U and a normalized, partial-transpose invariant Gram factor F for the steered Bell state. First, basis selection (hypothesis hselect) yields U such that the compressed source has full rank 3. Second, partial-transpose invariance of the Bell state (hypothesis hstate) ensures that the compressed Gram matrix inherits partial-transpose symmetry. Finally, positive-definite block normalization (hypothesis hnorm) is applied to the real part of the leading block of the Gram matrix, yielding the normalized Gram factor F with V Vᴴ = 1 and the transformation matrix T.

  1. Partial-transpose invariance on the first factor is preserved under filtering on the second factor.

    For any indices a, c ∈ Fin 3 and p, q ∈ Fin m, the filtered entries are given by ∑_{x, y} star (W x p) * ρ (a,x) (c,y) * W y q. Applying Finset.sum_congr twice to move inside the double sum, we substitute the hypothesis hpt, yielding the required partial-transpose symmetry.

    lines 13–18

    uses Finset.sum_congr

      intro a c p q
      apply Finset.sum_congr rfl
      intro x hx
      apply Finset.sum_congr rfl
      intro y hy
      rw [hpt a c x y]
  2. Partial-transpose invariance on the first factor is preserved under a bilateral filter H on the first factor.

    Given partial-transpose symmetry ρ (i,p) (j,q) = ρ (j,p) (i,q), the congruence ∑_{i, j} H a i * ρ (i,p) (j,q) * H c j is transformed by swapping summation order with Finset.sum_comm, entering the sum with Finset.sum_congr, rewriting with hpt, and commuting scalar products with ring.

    lines 27–34

    uses Finset.sum_comm, Finset.sum_congr

      intro a c p q
      rw [Finset.sum_comm]
      apply Finset.sum_congr rfl
      intro i hi
      apply Finset.sum_congr rfl
      intro j hj
      rw [hpt i j p q]
      ring
  3. For any Hermitian matrix X with partial-transpose invariance, its off-diagonal block is symmetric and its diagonal blocks have entirely real entries.

    For the off-diagonal block B_{i j} = X (i,0) (j,1), transposing gives Bᵀ_{i j} = X (j,0) (i,1) = X (i,0) (j,1) by hypothesis hpt. For the diagonal block entries, Hermiticity gives star (X (i,p) (j,p)) = X (j,p) (i,p), which equals X (i,p) (j,p) by hpt, proving the entries are self-conjugate (real).

    lines 42–48

      constructor
      · ext i j
        exact hpt j i 0 1
      · intro i j p
        calc
          star (X (i,p) (j,p)) = X (j,p) (i,p) := hX.apply (j,p) (i,p)
          _ = X (i,p) (j,p) := hpt j i p p
  4. The Gram matrix of a filtered tensor product (T v_t) ⊗ z_t satisfies the expected bipartite congruence identity. automation

    Expanding matrix-vector products and finite summations over Fin 3 and Fin 4 using simp with Matrix.mulVec, dotProduct, Fin.sum_univ_three, Fin.sum_univ_four, star_mul, and star_sum, the polynomial identity in the matrix entries is verified by ring.

    lines 56–57

    uses Matrix.mulVec, dotProduct, Fin.sum_univ_three, Fin.sum_univ_four, star_mul, star_sum

      simp [Matrix.mulVec, dotProduct, Fin.sum_univ_three, Fin.sum_univ_four, star_mul, star_sum]
      ring
  5. The Gram matrix of (T ⊗ I) F equals the congruence T (F Fᴴ) Tᴴ. automation

    Expanding matrix multiplication and conjugate transposes into sums over Fin 3 and Fin 4 via simp using Matrix.mul_apply, Matrix.conjTranspose_apply, Fin.sum_univ_three, Fin.sum_univ_four, and star_mul, the resulting polynomial expression simplifies to the congruence by ring.

    lines 66–67

    uses Matrix.mul_apply, Matrix.conjTranspose_apply, Fin.sum_univ_three, Fin.sum_univ_four, star_mul

      simp [Matrix.mul_apply, Matrix.conjTranspose_apply, Fin.sum_univ_three, Fin.sum_univ_four, star_mul]
      ring
  6. Extract the real positive-definite leading block R = Re(U₀ U₀ᴴ), obtain its normalizer H via hypothesis hnorm, and construct invertible complex filter matrices E and D with D E = 1 and E C E = 1.

    Let U₀ = G (·, 0) and C = U₀ U₀ᴴ. Since G Gᴴ is Hermitian (Matrix.isHermitian_mul_conjTranspose_self) and PT-invariant, lemma in767_actualnorm_pt_hermitian_blocks shows that the entries of C are all real. Thus the real part matrix R embeds into ℂ as C, verified using Complex.ext and linarith. Applying hypothesis hnorm to R gives a symmetric invertible matrix H with H * R * H = 1. Taking the inverse unit J = H⁻¹ with Units.inv_mul and embedding into ℂ yields E and D satisfying D * E = 1 and E * C * E = 1 by Matrix.map_mul and Matrix.map_one.

    lines 81–115

    uses Matrix.isHermitian_mul_conjTranspose_self, in767_actualnorm_pt_hermitian_blocks, Complex.ext, Units.inv_mul, Matrix.map_mul, Matrix.map_one

      let U0 : Matrix (Fin 3) (Fin 4) ℂ := fun i t => G (i,0) t
      let C : Matrix (Fin 3) (Fin 3) ℂ := U0*U0ᴴ
      let R : Matrix (Fin 3) (Fin 3) ℝ := fun i j => (C i j).re
      have hX : (G*Gᴴ).IsHermitian := Matrix.isHermitian_mul_conjTranspose_self G
      obtain ⟨_, hreal⟩ := in767_actualnorm_pt_hermitian_blocks (G*Gᴴ) hX hpt
      have hCreal : ∀ i j, star (C i j) = C i j := by
        intro i j
        simpa [C, U0, Matrix.mul_apply, Matrix.conjTranspose_apply] using hreal i j 0
      have hR : R.map (algebraMap ℝ ℂ) = C := by
        ext i j
        have hh := hCreal i j
        apply Complex.ext <;> simp [R, Complex.ext_iff] at hh ⊢ <;> linarith
      obtain ⟨H, hHsym, hHunit, hH⟩ := hnorm R hpos
      obtain ⟨Hu, hHu⟩ := hHunit
      let J : Matrix (Fin 3) (Fin 3) ℝ := ↑(Hu⁻¹)
      have hJH : J * H = 1 := by
        rw [← hHu]
        exact Units.inv_mul Hu
      let E : Matrix (Fin 3) (Fin 3) ℂ := H.map (algebraMap ℝ ℂ)
      let D : Matrix (Fin 3) (Fin 3) ℂ := J.map (algebraMap ℝ ℂ)
      have hDE : D * E = 1 := by
        have hh := congrArg (fun M : Matrix (Fin 3) (Fin 3) ℝ => M.map (algebraMap ℝ ℂ)) hJH
        simp only [Matrix.map_mul, Matrix.map_one, map_zero, map_one] at hh
        exact hh
      have hEreal : ∀ i j, star (E i j) = E i j := by intro i j; simp [E]
      have hEherm : Eᴴ = E := by
        ext i j
        have hh := congrArg (fun M : Matrix (Fin 3) (Fin 3) ℝ => M i j) hHsym
        simpa [E, Matrix.conjTranspose_apply, Matrix.transpose_apply] using congrArg (algebraMap ℝ ℂ) hh
      have hEC : E*C*E = 1 := by
        have hh := congrArg (fun M : Matrix (Fin 3) (Fin 3) ℝ => M.map (algebraMap ℝ ℂ)) hH
        simp only [Matrix.map_mul, Matrix.map_one, map_zero, map_one] at hh
        change E * (R.map (algebraMap ℝ ℂ)) * E = 1 at hh
        rw [hR] at hh
        exact hh
  7. The normalized factor F = (E ⊗ I) G satisfies U Uᴴ = 1, preserves partial-transpose invariance of F Fᴴ, and reconstructs G as G = (D ⊗ I) F.

    Define F = (E ⊗ I) G and U = F (·, 0) = E U₀. Then U Uᴴ = E C Eᴴ = E C E = 1 using Matrix.conjTranspose_mul, symmetry of E, and Matrix.mul_assoc. By lemma in767_actualnorm_four_column_gram_congruence, F Fᴴ = (E ⊗ I) (G Gᴴ) (Eᵀ ⊗ I), which is PT-invariant by lemma in767_actualnorm_ppt_under_real_first_filter. Finally, G = (D ⊗ I) F follows from D * E = 1 via Matrix.mul_assoc and Matrix.one_mul.

    lines 116–135

    uses Matrix.conjTranspose_mul, Matrix.mul_assoc, in767_actualnorm_four_column_gram_congruence, in767_actualnorm_ppt_under_real_first_filter, Matrix.one_mul

      let F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ := fun ip t => ∑ i : Fin 3, E ip.1 i * G (i,ip.2) t
      let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t
      have hUF : U = E*U0 := by ext i t; simp [U, F, U0, Matrix.mul_apply]
      refine ⟨F, D, ?_, ?_, ?_⟩
      · change U * Uᴴ = 1
        rw [hUF, Matrix.conjTranspose_mul, hEherm]
        simpa [C, Matrix.mul_assoc] using hEC
      · intro i j p q
        have hcongr (i j : Fin 3) (p q : Fin 2) : (F*Fᴴ) (i,p) (j,q) =
            ∑ a : Fin 3, ∑ b : Fin 3, E i a * (G*Gᴴ) (a,p) (b,q) * E j b := by
          change (∑ t : Fin 4, (∑ a : Fin 3, E i a * G (a,p) t) *
            star (∑ b : Fin 3, E j b * G (b,q) t)) = _
          rw [in767_actualnorm_four_column_gram_congruence]
          simp only [hEreal]
        rw [hcongr, hcongr]
        exact in767_actualnorm_ppt_under_real_first_filter (G*Gᴴ) E hpt i j p q
      · intro i p t
        let Y : Matrix (Fin 3) (Fin 4) ℂ := fun j t => G (j,p) t
        change Y i t = (D * (E * Y)) i t
        rw [← Matrix.mul_assoc, hDE, Matrix.one_mul]
  8. If U has full row rank (injective vector multiplication), the entrywise real matrix R = Re(U Uᴴ) is strictly positive definite.

    The complex Gram matrix C = U Uᴴ is positive definite by Matrix.PosDef.mul_conjTranspose_self. Its real part R is Hermitian by Matrix.IsHermitian.ext. For any non-zero real vector x, viewing it as a non-zero complex vector x_c, the quadratic form star x ⬝ᵥ (R *ᵥ x) equals Re(star x_c ⬝ᵥ (C *ᵥ x_c)) > 0. Using Matrix.PosDef.of_dotProduct_mulVec_pos, Fin.sum_univ_three, and Complex.mul_re, R is positive definite.

    lines 141–163

    uses Matrix.PosDef.mul_conjTranspose_self, Matrix.IsHermitian.ext, Matrix.PosDef.of_dotProduct_mulVec_pos, Fin.sum_univ_three, Complex.mul_re

      let C : Matrix (Fin 3) (Fin 3) ℂ := U*Uᴴ
      let R : Matrix (Fin 3) (Fin 3) ℝ := fun i j => (C i j).re
      have hC : C.PosDef := Matrix.PosDef.mul_conjTranspose_self U hinj
      have hRherm : R.IsHermitian := by
        apply Matrix.IsHermitian.ext
        intro i j
        have hh := congrArg Complex.re (hC.isHermitian.apply i j)
        simpa [R, C] using hh
      apply Matrix.PosDef.of_dotProduct_mulVec_pos hRherm
      intro x hx
      let xc : Fin 3 → ℂ := fun i => x i
      have hxc : xc ≠ 0 := by
        intro h
        apply hx
        ext i
        have hh := congrFun h i
        change (x i : ℂ) = 0 at hh
        exact_mod_cast hh
      have hp := (Complex.pos_iff.mp (hC.dotProduct_mulVec_pos hxc)).1
      have heq : star x ⬝ᵥ (R *ᵥ x) = (star xc ⬝ᵥ (C *ᵥ xc)).re := by
        simp [dotProduct, Matrix.mulVec, Fin.sum_univ_three, Complex.mul_re, R, xc]
      rw [heq]
      exact hp
  9. Any 3 × 4 complex matrix U with rank 3 induces an injective linear map under left vector multiplication v ↦ v * U.

    Consider the linear map f = Uᵀ.mulVecLin from ℂ³ to ℂ⁴. By Matrix.rank_transpose, its image has dimension 3, matching the domain dimension. By the rank-nullity theorem LinearMap.finrank_range_add_finrank_ker, its kernel has dimension 0, so ker(f) = ⊥ and f is injective by LinearMap.ker_eq_bot. Transposition identifies f with U.vecMul.

    lines 167–179

    uses Matrix.rank_transpose, LinearMap.finrank_range_add_finrank_ker, LinearMap.ker_eq_bot

      let f := U.transpose.mulVecLin
      have hdim : Module.finrank ℂ (Fin 3 → ℂ) = 3 := by simp [Module.finrank_pi]
      have hrange : Module.finrank ℂ (LinearMap.range f) = 3 := by
        change U.transpose.rank = 3
        rw [Matrix.rank_transpose]
        exact hrank
      have hk := LinearMap.finrank_range_add_finrank_ker f
      rw [hrange, hdim] at hk
      have hker : LinearMap.ker f = ⊥ := Submodule.finrank_eq_zero.mp (by omega)
      have hf : Function.Injective f := LinearMap.ker_eq_bot.mp hker
      intro x y h
      apply hf
      simpa [f, Matrix.mulVec_transpose] using h
  10. A PT-invariant Gram factor G whose leading block has rank 3 can be normalized using positive-definite block normalization.

    The rank condition implies U = G (·, 0) has injective vector multiplication via in767_actualnorm_full_row_rank_vecmul_injective. This ensures R = Re(U Uᴴ) is positive definite by in767_actualnorm_real_gram_positive. Applying in767_actualnorm_normalize_pt_gram_factor then yields the desired normalized Gram factor.

    lines 188–191

    uses in767_actualnorm_full_row_rank_vecmul_injective, in767_actualnorm_real_gram_positive, in767_actualnorm_normalize_pt_gram_factor

        ∃ (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ) (T : Matrix (Fin 3) (Fin 3) ℂ),
          (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; U*Uᴴ = 1) ∧
          (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧
          (∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4, G (i,p) t = ∑ a : Fin 3, T i a * F (a,p) t) := by
  11. The rank-four compressed source matrix expands algebraically into the factorized Gram form. automation

    Since each weight lam(t) ≥ 0, Real.sq_sqrt gives (Real.sqrt (lam t))² = lam(t). Expanding the outer sums over Fin 3 and Fin 4 using simp with Fin.sum_univ_three, Fin.sum_univ_four, star_mul, and star_add, both sides simplify to identical polynomial expressions.

    lines 207–212

    uses Real.sq_sqrt, Fin.sum_univ_three, Fin.sum_univ_four, star_mul, star_add

      have hsq (t : Fin 4) : (Real.sqrt (lam t) : ℂ)^2 = (lam t : ℂ) := by
        exact_mod_cast Real.sq_sqrt (hlam t)
      simp [Fin.sum_univ_three, Fin.sum_univ_four, star_mul, star_add]
      ring_nf
      simp only [hsq]
      ring
  12. The compressed rank-four source matrix G has a PT-invariant Gram matrix G Gᴴ, and can be normalized.

    Define G as the weighted source compressed by W. By lemma in767_actualnorm_compressed_rankfour_gram, G Gᴴ = (I ⊗ W)ᴴ ρ (I ⊗ W) where ρ is the uncompressed source density matrix. Since ρ is PT-invariant by hypothesis, lemma in767_actualnorm_ppt_under_second_factor shows G Gᴴ is PT-invariant. Applying lemma in767_actualnorm_normalize_pt_gram_factor_of_rank completes the normalization.

    lines 225–237

    uses in767_actualnorm_compressed_rankfour_gram, in767_actualnorm_ppt_under_second_factor, in767_actualnorm_normalize_pt_gram_factor_of_rank

        ∃ (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ) (T : Matrix (Fin 3) (Fin 3) ℂ),
          (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; U*Uᴴ = 1) ∧
          (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧
          (∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4,
            (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x p) * source t (i,x)) =
            ∑ a : Fin 3, T i a * F (a,p) t) := by
      let G : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ := fun ip t =>
        (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x ip.2) * source t (ip.1,x))
      let ρ : Matrix (Fin 3 × Fin 3) (Fin 3 × Fin 3) ℂ := fun ax cy =>
        ∑ t : Fin 4, (lam t : ℂ) * source t ax * star (source t cy)
      have hGpt : ∀ i j : Fin 3, ∀ p q : Fin 2, (G*Gᴴ) (i,p) (j,q) = (G*Gᴴ) (j,p) (i,q) := by
        intro i j p q
        have hi := (in767_actualnorm_compressed_rankfour_gram source lam W hlam i j p q).symm
  13. Introduce the rank-2 projection P, apply basis selection to obtain the isometric factor U with full compression rank, and define the steering matrix W and weight function lam.

    Given a rank-2 Hermitian projection P, hypothesis hselect provides an isometry U : ℂ² → ℂ³ with Uᴴ * U = 1 and U * Uᴴ = P, such that the compression G under the steered normalizer has rank 3. Define the steered matrix W = ppt_qutrit_steered_normalizer * U and the 4-element weight distribution lam.

    lines 269–272

      intro P hP hPP hRank
      obtain ⟨U, hU, hUP, hfull⟩ := hselect P hP hPP hRank
      let W : Matrix (Fin 3) (Fin 2) ℂ := ppt_qutrit_steered_normalizer * U
      let lam : Fin 4 → ℝ := fun t => if t = 0 then 3257/6884 else if t = 3 then 27/6884 else 450/1721
  14. All weights lam(t) are non-negative. automation

    Case analysis on t ∈ Fin 4 shows that each value of lam(t) is a positive rational number (3257/6884, 27/6884, or 450/1721), closed by split_ifs and norm_num.

    lines 273–276

      have hlam : ∀ t, 0 ≤ lam t := by
        intro t
        dsimp [lam]
        split_ifs <;> norm_num
  15. The weighted Bell state satisfies partial-transpose invariance across the two qutrit factors. automation

    By hypothesis hstate, the state ppt_qutrit_bell_state is invariant under partial transpose. Expanding its definition via Fin.sum_univ_four, Matrix.sum_apply, Matrix.smul_apply, Matrix.vecMulVec_apply, Pi.star_apply, Complex.real_smul, and mul_assoc, the symmetry translates into the required vector identity.

    lines 277–283

    uses ppt_qutrit_bell_state, Fin.sum_univ_four, Matrix.sum_apply, Matrix.smul_apply, Matrix.vecMulVec_apply, Pi.star_apply, Complex.real_smul, mul_assoc

      have hpt : ∀ i j x y : Fin 3,
          (∑ t : Fin 4, (lam t : ℂ) * ppt_qutrit_bell_vector t (i,x) * star (ppt_qutrit_bell_vector t (j,y))) =
          ∑ t : Fin 4, (lam t : ℂ) * ppt_qutrit_bell_vector t (j,x) * star (ppt_qutrit_bell_vector t (i,y)) := by
        intro i j x y
        have hh := hstate i x j y
        simpa [ppt_qutrit_bell_state, lam, Fin.sum_univ_four, Matrix.sum_apply, Matrix.smul_apply,
          Matrix.vecMulVec_apply, Pi.star_apply, Complex.real_smul, mul_assoc] using hh
  16. Apply the compressed Gram normalization lemma to obtain F and T, completing the proof.

    Lemma in767_actualnorm_source_compressed_gram_normalization combines non-negativity of lam, partial-transpose invariance of the Bell state, full rank of the compressed block, and hypothesis hnorm to produce the normalized factor F and transition matrix T. Together with U, they satisfy all required properties.

    lines 284–286

    uses in767_actualnorm_source_compressed_gram_normalization

      obtain ⟨F, T, hupper, hptF, hrel⟩ :=
        in767_actualnorm_source_compressed_gram_normalization ppt_qutrit_bell_vector lam W hlam hpt hfull hnorm
      exact ⟨U, F, T, hU, hUP, hupper, hptF, hrel⟩
Finset.sum_congr
Congruence for finite sums: if two index sets are equal and the summands agree at each element, their sums are equal
Finset.sum_comm
Swapping the order of summation in a double finite sum
Matrix.mulVec
Matrix-vector multiplication, computing (A v)_i = ∑_j A_{i j} v_j
dotProduct
The standard scalar dot product of two vectors, ∑_i u_i v_i
Fin.sum_univ_three
Expanding a summation over Fin 3 into f(0) + f(1) + f(2)
Fin.sum_univ_four
Expanding a summation over Fin 4 into f(0) + f(1) + f(2) + f(3)
star_mul
The conjugate of a product satisfies (r * s)* = s* * r*
star_sum
The conjugate commutes with finite sums: (∑ x)* = ∑ x*
Matrix.mul_apply
Entrywise definition of matrix multiplication: (A * B)_{i k} = ∑_j A_{i j} * B_{j k}
Matrix.conjTranspose_apply
Entrywise definition of conjugate transpose: (Aᴴ)_{i j} = (A_{j i})*
Matrix.isHermitian_mul_conjTranspose_self
The matrix A * Aᴴ is always Hermitian for any matrix A
in767_actualnorm_pt_hermitian_blocks
For a PT-invariant Hermitian matrix on ℂ³ ⊗ ℂ², off-diagonal blocks are symmetric and diagonal block entries are real
Complex.ext
Two complex numbers are equal if and only if their real and imaginary parts are equal
Units.inv_mul
The product of an invertible element's inverse with itself is 1
Matrix.map_mul
An entrywise ring homomorphism distributes over matrix multiplication
Matrix.map_one
An entrywise ring homomorphism maps the identity matrix to the identity matrix
Matrix.conjTranspose_mul
The conjugate transpose of a product satisfies (M * N)ᴴ = Nᴴ * Mᴴ
Matrix.mul_assoc
Associativity of matrix multiplication: (L * M) * N = L * (M * N)
in767_actualnorm_four_column_gram_congruence
Congruence identity expressing the Gram matrix of (T ⊗ I) F in terms of T (F Fᴴ) Tᴴ
in767_actualnorm_ppt_under_real_first_filter
Partial-transpose invariance on the first factor is preserved under congruence by a real matrix filter
Matrix.one_mul
Multiplying any matrix on the left by the identity matrix leaves it unchanged: 1 * M = M
Matrix.PosDef.mul_conjTranspose_self
If v ↦ v * A is injective, the Gram matrix A * Aᴴ is positive definite
Matrix.IsHermitian.ext
A matrix is Hermitian if star (A j i) = A i j for all indices i, j
Matrix.PosDef.of_dotProduct_mulVec_pos
A Hermitian matrix is positive definite if star x ⬝ᵥ M *ᵥ x > 0 for all non-zero vectors x
Complex.mul_re
The real part of a product of complex numbers: Re(z * w) = Re(z) * Re(w) - Im(z) * Im(w)
Matrix.rank_transpose
The row rank of a matrix equals its column rank: rank(Aᵀ) = rank(A)
LinearMap.finrank_range_add_finrank_ker
The rank-nullity theorem: dim(range f) + dim(ker f) = dim V
LinearMap.ker_eq_bot
A linear map has trivial kernel if and only if it is injective
in767_actualnorm_full_row_rank_vecmul_injective
A matrix with full row rank induces an injective linear map under left vector multiplication
in767_actualnorm_real_gram_positive
The entrywise real part of U * Uᴴ is positive definite whenever U has full row rank
in767_actualnorm_normalize_pt_gram_factor
Normalization of a PT-invariant Gram factor using positive-definite block normalization
Real.sq_sqrt
The square of the square root of a non-negative real number is the number itself
star_add
Complex conjugation distributes over addition: star (r + s) = star r + star s
in767_actualnorm_compressed_rankfour_gram
Algebraic expansion expressing the compressed Gram matrix in factorized Gram form
in767_actualnorm_ppt_under_second_factor
Partial-transpose invariance on the first factor is preserved under filtering on the second factor
in767_actualnorm_normalize_pt_gram_factor_of_rank
Constructs a normalized PT-invariant Gram factor whenever the leading block has full rank
ppt_qutrit_bell_state
The bipartite density matrix defined as the weighted sum of outer products of the four Bell vectors
Matrix.sum_apply
Evaluating a finite sum of matrices entrywise equals the sum of the entries
Matrix.smul_apply
Evaluating a scalar multiple of a matrix entrywise equals the scalar times the entry
Matrix.vecMulVec_apply
Evaluating the outer product u vᵀ at (i, j) equals u_i * v_j
Pi.star_apply
Evaluating the pointwise complex conjugate of a vector equals the conjugate of the entry
Complex.real_smul
Real scalar multiplication on ℂ coincides with ordinary multiplication in ℂ
mul_assoc
Associativity of multiplication: a * b * c = a * (b * c)
in767_actualnorm_source_compressed_gram_normalization
The explicit rank-four source compressed by W admits a normalized PT-invariant Gram factorization given full rank and positive normalization

Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.

Lean source view module on GitHub

open Matrixopen scoped BigOperators Matrix ComplexOrderset_option backward.isDefEq.respectTransparency falseset_option linter.unusedSimpArgs falseset_option linter.unnecessarySeqFocus falseprivate lemma in767_actualnorm_ppt_under_second_factor    {n m : ℕ} (ρ : Matrix (Fin 3 × Fin n) (Fin 3 × Fin n) ℂ)    (W : Matrix (Fin n) (Fin m) ℂ)    (hpt : ∀ a c : Fin 3, ∀ x y : Fin n, ρ (a,x) (c,y) = ρ (c,x) (a,y)) :    ∀ a c : Fin 3, ∀ p q : Fin m,      (∑ x : Fin n, ∑ y : Fin n, star (W x p) * ρ (a,x) (c,y) * W y q) =      ∑ x : Fin n, ∑ y : Fin n, star (W x p) * ρ (c,x) (a,y) * W y q := by  intro a c p q  apply Finset.sum_congr rfl  intro x hx  apply Finset.sum_congr rfl  intro y hy  rw [hpt a c x y]private lemma in767_actualnorm_ppt_under_real_first_filter    {m : ℕ} (ρ : Matrix (Fin 3 × Fin m) (Fin 3 × Fin m) ℂ)    (H : Matrix (Fin 3) (Fin 3) ℂ)    (hpt : ∀ i j : Fin 3, ∀ p q : Fin m, ρ (i,p) (j,q) = ρ (j,p) (i,q)) :    ∀ a c : Fin 3, ∀ p q : Fin m,      (∑ i : Fin 3, ∑ j : Fin 3, H a i * ρ (i,p) (j,q) * H c j) =      ∑ i : Fin 3, ∑ j : Fin 3, H c i * ρ (i,p) (j,q) * H a j := by  intro a c p q  rw [Finset.sum_comm]  apply Finset.sum_congr rfl  intro i hi  apply Finset.sum_congr rfl  intro j hj  rw [hpt i j p q]  ringprivate lemma in767_actualnorm_pt_hermitian_blocks    (X : Matrix (Fin 3 × Fin 2) (Fin 3 × Fin 2) ℂ) (hX : X.IsHermitian)    (hpt : ∀ i j : Fin 3, ∀ p q : Fin 2, X (i,p) (j,q) = X (j,p) (i,q)) :    (let B : Matrix (Fin 3) (Fin 3) ℂ := fun i j => X (i,0) (j,1); B.transpose = B) ∧    (∀ i j : Fin 3, ∀ p : Fin 2, star (X (i,p) (j,p)) = X (i,p) (j,p)) := by  constructor  · ext i j    exact hpt j i 0 1  · intro i j p    calc      star (X (i,p) (j,p)) = X (j,p) (i,p) := hX.apply (j,p) (i,p)      _ = X (i,p) (j,p) := hpt j i p pprivate lemma in767_actualnorm_bipartite_product_congruence    (T : Matrix (Fin 3) (Fin 3) ℂ) (v : Fin 4 → Fin 3 → ℂ) (z : Fin 4 → Fin 2 → ℂ)    (i j : Fin 3) (p q : Fin 2) :    (∑ a : Fin 3, ∑ b : Fin 3, T i a *      (∑ t : Fin 4, (v t a * z t p) * star (v t b * z t q)) * star (T j b)) =      ∑ t : Fin 4, ((T *ᵥ v t) i * z t p) * star ((T *ᵥ v t) j * z t q) := by  simp [Matrix.mulVec, dotProduct, Fin.sum_univ_three, Fin.sum_univ_four, star_mul, star_sum]  ringprivate lemma in767_actualnorm_four_column_gram_congruence    (T : Matrix (Fin 3) (Fin 3) ℂ) (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ)    (i j : Fin 3) (p q : Fin 2) :    (∑ t : Fin 4, (∑ a : Fin 3, T i a * F (a,p) t) *      star (∑ b : Fin 3, T j b * F (b,q) t)) =      ∑ a : Fin 3, ∑ b : Fin 3, T i a * (F*Fᴴ) (a,p) (b,q) * star (T j b) := by  simp [Matrix.mul_apply, Matrix.conjTranspose_apply, Fin.sum_univ_three, Fin.sum_univ_four, star_mul]  ringprivate lemma in767_actualnorm_normalize_pt_gram_factor    (G : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ)    (hpt : ∀ i j : Fin 3, ∀ p q : Fin 2, (G*Gᴴ) (i,p) (j,q) = (G*Gᴴ) (j,p) (i,q))    (hpos : (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => G (i,0) t      let R : Matrix (Fin 3) (Fin 3) ℝ := fun i j => ((U*Uᴴ) i j).re      R.PosDef))    (hnorm : ∀ R : Matrix (Fin 3) (Fin 3) ℝ, R.PosDef →      ∃ H : Matrix (Fin 3) (Fin 3) ℝ, H.transpose = H ∧ IsUnit H ∧ H*R*H = 1) :    ∃ (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ) (T : Matrix (Fin 3) (Fin 3) ℂ),      (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; U*Uᴴ = 1) ∧      (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧      (∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4, G (i,p) t = ∑ a : Fin 3, T i a * F (a,p) t) := by  let U0 : Matrix (Fin 3) (Fin 4) ℂ := fun i t => G (i,0) t  let C : Matrix (Fin 3) (Fin 3) ℂ := U0*U0ᴴ  let R : Matrix (Fin 3) (Fin 3) ℝ := fun i j => (C i j).re  have hX : (G*Gᴴ).IsHermitian := Matrix.isHermitian_mul_conjTranspose_self G  obtain ⟨_, hreal⟩ := in767_actualnorm_pt_hermitian_blocks (G*Gᴴ) hX hpt  have hCreal : ∀ i j, star (C i j) = C i j := by    intro i j    simpa [C, U0, Matrix.mul_apply, Matrix.conjTranspose_apply] using hreal i j 0  have hR : R.map (algebraMap ℝ ℂ) = C := by    ext i j    have hh := hCreal i j    apply Complex.ext <;> simp [R, Complex.ext_iff] at hh ⊢ <;> linarith  obtain ⟨H, hHsym, hHunit, hH⟩ := hnorm R hpos  obtain ⟨Hu, hHu⟩ := hHunit  let J : Matrix (Fin 3) (Fin 3) ℝ := ↑(Hu⁻¹)  have hJH : J * H = 1 := by    rw [← hHu]    exact Units.inv_mul Hu  let E : Matrix (Fin 3) (Fin 3) ℂ := H.map (algebraMap ℝ ℂ)  let D : Matrix (Fin 3) (Fin 3) ℂ := J.map (algebraMap ℝ ℂ)  have hDE : D * E = 1 := by    have hh := congrArg (fun M : Matrix (Fin 3) (Fin 3) ℝ => M.map (algebraMap ℝ ℂ)) hJH    simp only [Matrix.map_mul, Matrix.map_one, map_zero, map_one] at hh    exact hh  have hEreal : ∀ i j, star (E i j) = E i j := by intro i j; simp [E]  have hEherm : Eᴴ = E := by    ext i j    have hh := congrArg (fun M : Matrix (Fin 3) (Fin 3) ℝ => M i j) hHsym    simpa [E, Matrix.conjTranspose_apply, Matrix.transpose_apply] using congrArg (algebraMap ℝ ℂ) hh  have hEC : E*C*E = 1 := by    have hh := congrArg (fun M : Matrix (Fin 3) (Fin 3) ℝ => M.map (algebraMap ℝ ℂ)) hH    simp only [Matrix.map_mul, Matrix.map_one, map_zero, map_one] at hh    change E * (R.map (algebraMap ℝ ℂ)) * E = 1 at hh    rw [hR] at hh    exact hh  let F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ := fun ip t => ∑ i : Fin 3, E ip.1 i * G (i,ip.2) t  let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t  have hUF : U = E*U0 := by ext i t; simp [U, F, U0, Matrix.mul_apply]  refine ⟨F, D, ?_, ?_, ?_⟩  · change U * Uᴴ = 1    rw [hUF, Matrix.conjTranspose_mul, hEherm]    simpa [C, Matrix.mul_assoc] using hEC  · intro i j p q    have hcongr (i j : Fin 3) (p q : Fin 2) : (F*Fᴴ) (i,p) (j,q) =        ∑ a : Fin 3, ∑ b : Fin 3, E i a * (G*Gᴴ) (a,p) (b,q) * E j b := by      change (∑ t : Fin 4, (∑ a : Fin 3, E i a * G (a,p) t) *        star (∑ b : Fin 3, E j b * G (b,q) t)) = _      rw [in767_actualnorm_four_column_gram_congruence]      simp only [hEreal]    rw [hcongr, hcongr]    exact in767_actualnorm_ppt_under_real_first_filter (G*Gᴴ) E hpt i j p q  · intro i p t    let Y : Matrix (Fin 3) (Fin 4) ℂ := fun j t => G (j,p) t    change Y i t = (D * (E * Y)) i t    rw [← Matrix.mul_assoc, hDE, Matrix.one_mul]private lemma in767_actualnorm_real_gram_positive    (U : Matrix (Fin 3) (Fin 4) ℂ) (hinj : Function.Injective U.vecMul) :    (let R : Matrix (Fin 3) (Fin 3) ℝ := fun i j => ((U*Uᴴ) i j).re; R.PosDef) := by  let C : Matrix (Fin 3) (Fin 3) ℂ := U*Uᴴ  let R : Matrix (Fin 3) (Fin 3) ℝ := fun i j => (C i j).re  have hC : C.PosDef := Matrix.PosDef.mul_conjTranspose_self U hinj  have hRherm : R.IsHermitian := by    apply Matrix.IsHermitian.ext    intro i j    have hh := congrArg Complex.re (hC.isHermitian.apply i j)    simpa [R, C] using hh  apply Matrix.PosDef.of_dotProduct_mulVec_pos hRherm  intro x hx  let xc : Fin 3 → ℂ := fun i => x i  have hxc : xc ≠ 0 := by    intro h    apply hx    ext i    have hh := congrFun h i    change (x i : ℂ) = 0 at hh    exact_mod_cast hh  have hp := (Complex.pos_iff.mp (hC.dotProduct_mulVec_pos hxc)).1  have heq : star x ⬝ᵥ (R *ᵥ x) = (star xc ⬝ᵥ (C *ᵥ xc)).re := by    simp [dotProduct, Matrix.mulVec, Fin.sum_univ_three, Complex.mul_re, R, xc]  rw [heq]  exact hpprivate lemma in767_actualnorm_full_row_rank_vecmul_injective    (U : Matrix (Fin 3) (Fin 4) ℂ) (hrank : U.rank = 3) : Function.Injective U.vecMul := by  let f := U.transpose.mulVecLin  have hdim : Module.finrank ℂ (Fin 3 → ℂ) = 3 := by simp [Module.finrank_pi]  have hrange : Module.finrank ℂ (LinearMap.range f) = 3 := by    change U.transpose.rank = 3    rw [Matrix.rank_transpose]    exact hrank  have hk := LinearMap.finrank_range_add_finrank_ker f  rw [hrange, hdim] at hk  have hker : LinearMap.ker f = ⊥ := Submodule.finrank_eq_zero.mp (by omega)  have hf : Function.Injective f := LinearMap.ker_eq_bot.mp hker  intro x y h  apply hf  simpa [f, Matrix.mulVec_transpose] using hprivate lemma in767_actualnorm_normalize_pt_gram_factor_of_rank    (G : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ)    (hpt : ∀ i j : Fin 3, ∀ p q : Fin 2, (G*Gᴴ) (i,p) (j,q) = (G*Gᴴ) (j,p) (i,q))    (hrank : (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => G (i,0) t; U.rank = 3))    (hnorm : ∀ R : Matrix (Fin 3) (Fin 3) ℝ, R.PosDef →      ∃ H : Matrix (Fin 3) (Fin 3) ℝ, H.transpose = H ∧ IsUnit H ∧ H*R*H = 1) :    ∃ (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ) (T : Matrix (Fin 3) (Fin 3) ℂ),      (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; U*Uᴴ = 1) ∧      (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧      (∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4, G (i,p) t = ∑ a : Fin 3, T i a * F (a,p) t) := by  let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => G (i,0) t  have hinj : Function.Injective U.vecMul := in767_actualnorm_full_row_rank_vecmul_injective U hrank  have hpos := in767_actualnorm_real_gram_positive U hinj  exact in767_actualnorm_normalize_pt_gram_factor G hpt hpos hnormprivate lemma in767_actualnorm_compressed_rankfour_gram    (v : Fin 4 → Fin 3 × Fin 3 → ℂ) (lam : Fin 4 → ℝ)    (W : Matrix (Fin 3) (Fin 2) ℂ) (hlam : ∀ t, 0 ≤ lam t)    (i j : Fin 3) (p q : Fin 2) :    (∑ x : Fin 3, ∑ y : Fin 3, star (W x p) *      (∑ t : Fin 4, (lam t : ℂ) * v t (i,x) * star (v t (j,y))) * W y q) =    ∑ t : Fin 4,      ((Real.sqrt (lam t) : ℂ) * ∑ x : Fin 3, star (W x p) * v t (i,x)) *      star ((Real.sqrt (lam t) : ℂ) * ∑ y : Fin 3, star (W y q) * v t (j,y)) := by  have hsq (t : Fin 4) : (Real.sqrt (lam t) : ℂ)^2 = (lam t : ℂ) := by    exact_mod_cast Real.sq_sqrt (hlam t)  simp [Fin.sum_univ_three, Fin.sum_univ_four, star_mul, star_add]  ring_nf  simp only [hsq]  ringprivate lemma in767_actualnorm_source_compressed_gram_normalization    (source : Fin 4 → Fin 3 × Fin 3 → ℂ) (lam : Fin 4 → ℝ) (W : Matrix (Fin 3) (Fin 2) ℂ)    (hlam : ∀ t, 0 ≤ lam t)    (hsourcept : ∀ i j x y : Fin 3,      (∑ t : Fin 4, (lam t : ℂ) * source t (i,x) * star (source t (j,y))) =      ∑ t : Fin 4, (lam t : ℂ) * source t (j,x) * star (source t (i,y)))    (hrank : (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t =>      (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x 0) * source t (i,x)); U.rank=3))    (hnorm : ∀ R : Matrix (Fin 3) (Fin 3) ℝ, R.PosDef →      ∃ H : Matrix (Fin 3) (Fin 3) ℝ, H.transpose = H ∧ IsUnit H ∧ H*R*H = 1) :    ∃ (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ) (T : Matrix (Fin 3) (Fin 3) ℂ),      (let U : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; U*Uᴴ = 1) ∧      (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧      (∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4,        (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x p) * source t (i,x)) =        ∑ a : Fin 3, T i a * F (a,p) t) := by  let G : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ := fun ip t =>    (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x ip.2) * source t (ip.1,x))  let ρ : Matrix (Fin 3 × Fin 3) (Fin 3 × Fin 3) ℂ := fun ax cy =>    ∑ t : Fin 4, (lam t : ℂ) * source t ax * star (source t cy)  have hGpt : ∀ i j : Fin 3, ∀ p q : Fin 2, (G*Gᴴ) (i,p) (j,q) = (G*Gᴴ) (j,p) (i,q) := by    intro i j p q    have hi := (in767_actualnorm_compressed_rankfour_gram source lam W hlam i j p q).symm    have hj := (in767_actualnorm_compressed_rankfour_gram source lam W hlam j i p q).symm    change (G*Gᴴ) (i,p) (j,q) = _ at hi    change (G*Gᴴ) (j,p) (i,q) = _ at hj    rw [hi, hj]    exact in767_actualnorm_ppt_under_second_factor ρ W hsourcept i j p q  exact in767_actualnorm_normalize_pt_gram_factor_of_rank G hGpt hrank hnormtheorem ppt_qutrit_source_normalized_gram_if_three_lemmas    (hselect :   ∀ P : Matrix (Fin 3) (Fin 3) ℂ, P.IsHermitian → P*P=P → P.rank=2 →    ∃ U : Matrix (Fin 3) (Fin 2) ℂ,      Uᴴ*U=1 ∧ U*Uᴴ=P ∧      (let W : Matrix (Fin 3) (Fin 2) ℂ := ppt_qutrit_steered_normalizer * U       let lam : Fin 4 → ℝ := fun t => if t = 0 then 3257/6884 else if t = 3 then 27/6884 else 450/1721       let G : Matrix (Fin 3) (Fin 4) ℂ := fun i t =>        (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x 0) * ppt_qutrit_bell_vector t (i,x))       G.rank=3))    (hstate : ∀ i x j y : Fin 3,      ppt_qutrit_bell_state (i,x) (j,y) = ppt_qutrit_bell_state (j,x) (i,y))    (hnorm : ∀ R : Matrix (Fin 3) (Fin 3) ℝ, R.PosDef →      ∃ H : Matrix (Fin 3) (Fin 3) ℝ, H.transpose = H ∧ IsUnit H ∧ H*R*H = 1) :  ∀ P : Matrix (Fin 3) (Fin 3) ℂ, P.IsHermitian → P*P=P → P.rank=2 →    ∃ (U : Matrix (Fin 3) (Fin 2) ℂ) (F : Matrix (Fin 3 × Fin 2) (Fin 4) ℂ)      (T : Matrix (Fin 3) (Fin 3) ℂ),      Uᴴ*U=1 ∧ U*Uᴴ=P ∧      (let V : Matrix (Fin 3) (Fin 4) ℂ := fun i t => F (i,0) t; V*Vᴴ=1) ∧      (∀ i j : Fin 3, ∀ p q : Fin 2, (F*Fᴴ) (i,p) (j,q) = (F*Fᴴ) (j,p) (i,q)) ∧      (let W : Matrix (Fin 3) (Fin 2) ℂ := ppt_qutrit_steered_normalizer * U       let lam : Fin 4 → ℝ := fun t => if t = 0 then 3257/6884 else if t = 3 then 27/6884 else 450/1721       ∀ i : Fin 3, ∀ p : Fin 2, ∀ t : Fin 4,        (Real.sqrt (lam t) : ℂ) * (∑ x : Fin 3, star (W x p) * ppt_qutrit_bell_vector t (i,x)) =          ∑ a : Fin 3, T i a * F (a,p) t) := by  intro P hP hPP hRank  obtain ⟨U, hU, hUP, hfull⟩ := hselect P hP hPP hRank  let W : Matrix (Fin 3) (Fin 2) ℂ := ppt_qutrit_steered_normalizer * U  let lam : Fin 4 → ℝ := fun t => if t = 0 then 3257/6884 else if t = 3 then 27/6884 else 450/1721  have hlam : ∀ t, 0 ≤ lam t := by    intro t    dsimp [lam]    split_ifs <;> norm_num  have hpt : ∀ i j x y : Fin 3,      (∑ t : Fin 4, (lam t : ℂ) * ppt_qutrit_bell_vector t (i,x) * star (ppt_qutrit_bell_vector t (j,y))) =      ∑ t : Fin 4, (lam t : ℂ) * ppt_qutrit_bell_vector t (j,x) * star (ppt_qutrit_bell_vector t (i,y)) := by    intro i j x y    have hh := hstate i x j y    simpa [ppt_qutrit_bell_state, lam, Fin.sum_univ_four, Matrix.sum_apply, Matrix.smul_apply,      Matrix.vecMulVec_apply, Pi.star_apply, Complex.real_smul, mul_assoc] using hh  obtain ⟨F, T, hupper, hptF, hrel⟩ :=    in767_actualnorm_source_compressed_gram_normalization ppt_qutrit_bell_vector lam W hlam hpt hfull hnorm  exact ⟨U, F, T, hU, hUP, hupper, hptF, hrel⟩
See it in the graphDiscuss this resultChallenge it