ppt_qutrit_bell_source_full_rank_on_every_plane
The contraction of the four source vectors has full qutrit row rank for some vector in each complex two-plane.
Statement
theorem ppt_qutrit_bell_source_full_rank_on_every_plane : ∀ U : Matrix (Fin 3) (Fin 2) ℂ, Uᴴ * U = 1 → ∃ x : Fin 2 → ℂ, let z : Fin 3 → ℂ := U.mulVec x let K : Matrix (Fin 3) (Fin 4) ℂ := fun a l => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b K.rank = 3
- Source
- Source-specific rank helper for OP-693, using https://arxiv.org/abs/1405.4502.
- Verified
- 12 Nov 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 3
- ppt_qutrit_bell_vectorFour explicit real vectors in two-qutrit space used in the PPT Bell-violation example.
- qutrit_source_two_cubics_no_common_planeTwo explicit cubic maximal minors of the source vectors cannot both vanish on a complex two-dimensional plane.
- Cslib.SKI.Polynomial.evalSubstitute terms for the free variables of a polynomial
- Used by 1
- ppt_qutrit_source_projection_basis_full_rankEvery rank-two qutrit projection admits an orthonormal basis whose first normalized-source contraction has full row rank.
The proof, explained
To show that the 3 × 4 contraction matrix K(z) of the four PPT Bell vectors has full row rank (rank 3) for some vector in any complex two-plane, the proof analyzes two 3 × 3 minors of K(z). Their determinants are proportional to homogeneous cubics q(z) = x(x² - 3y²) and p(z) = x²y + (10α/3)x²z + y³ - (10α/3)y²z - yz²/18. The argument then proves that these two cubics cannot vanish simultaneously on any two-plane: if q vanishes identically, the plane must satisfy x = k y with k ∈ {0, ±√3}, and then selecting a point on the plane with y = 1 and z = 0 forces p to evaluate to k² + 1 ≠ 0, a contradiction. Thus every plane contains a vector where at least one minor is non-zero, ensuring K(z) has full rank 3.
The columns of a 3 × 2 complex isometry U are linearly independent.
Left-multiplying the equation U *ᵥ x = U *ᵥ y by Uᴴ and using Uᴴ * U = 1 shows that multiplication by U is injective. By `Matrix.mulVec_injective_iff.mp`, this injectivity is equivalent to linear independence of the column vectors of U.
uses
Matrix.mulVec_mulVec,Matrix.one_mulVec,Matrix.mulVec_injective_iff.mphave hinj : Function.Injective U.mulVec := by intro x y h have hh := congrArg (fun v => Uᴴ *ᵥ v) h simpa only [Matrix.mulVec_mulVec, hU, Matrix.one_mulVec] using hh have hcol := Matrix.mulVec_injective_iff.mp hinj have heq : ![(fun i => U i 0),(fun i => U i 1)] = U.col := by funext j i fin_cases j <;> rfl rw [heq] exact hcol
For any vector z ∈ ℂ³, the contraction matrix K(z) matches the explicit 3 × 4 matrix formula. automation
By extensionality over the row and column indices, the equality is verified component by component. Unfolding the definition of the PPT Bell vectors and expanding the sum over the three components with `Fin.sum_univ_three`, each entry simplifies to the claimed coordinate expression via ring arithmetic.
uses
Fin.sum_univ_threeext i l fin_cases i <;> fin_cases l <;> simp [ppt_qutrit_bell_vector, Fin.sum_univ_three] <;> ring
The 3 × 3 minors of K(z) on columns {0, 1, 2} and {0, 1, 3} are proportional to the cubics q(z) = z₀(z₀² - 3z₁²) and p(z) = z₀²z₁ + (10α/3)z₀²z₂ + z₁³ - (10α/3)z₁²z₂ - z₁z₂²/18, respectively. automation
Using the explicit matrix representation from `in767_rankproof_contraction_explicit`, the determinants of the two 3 × 3 submatrices are expanded via `Matrix.det_fin_three` and reduce by algebraic computation to the stated factored forms.
uses
in767_rankproof_contraction_explicit,Matrix.det_fin_threedsimp only rw [in767_rankproof_contraction_explicit] constructor <;> simp [Matrix.det_fin_three, Matrix.submatrix_apply] <;> ring
If two linearly independent vectors u, v ∈ ℂ³ satisfy u₀ = k u₁ and v₀ = k v₁ for some scalar k, then the 2 × 2 minor u₁ v₂ - u₂ v₁ is non-zero.
Suppose for contradiction that u₁ v₂ - u₂ v₁ = 0. The linear combination v₁ • u - u₁ • v vanishes in all three coordinates: coordinate 0 vanishes because u₀ = k u₁ and v₀ = k v₁, coordinate 1 vanishes identically, and coordinate 2 vanishes by assumption. Linear independence of u and v via `LinearIndependent.pair_iff.mp` forces v₁ = 0 and u₁ = 0. Consequently, the combination v₂ • u - u₂ • v also vanishes in all coordinates, forcing v₂ = 0 and u₂ = 0. This implies u = 0, which contradicts the linear independence of u and v.
uses
LinearIndependent.pair_iff.mpintro hd have hpair := LinearIndependent.pair_iff.mp hli have hz1 : (v 1) • u + (-u 1) • v = 0 := by ext i fin_cases i <;> simp only [Pi.add_apply, Pi.smul_apply, smul_eq_mul, Pi.zero_apply] · change v 1 * u 0 + (-u 1) * v 0 = 0 rw [hu, hv]; ring · change v 1 * u 1 + (-u 1) * v 1 = 0 ring · change v 1 * u 2 + (-u 1) * v 2 = 0 linear_combination -hd have hfirst := hpair (v 1) (-u 1) hz1 have hv1 : v 1=0 := hfirst.1 have hu1 : u 1=0 := by simpa using hfirst.2 have hz2 : (v 2) • u + (-u 2) • v = 0 := by ext i fin_cases i <;> simp only [Pi.add_apply, Pi.smul_apply, smul_eq_mul, Pi.zero_apply] · change v 2 * u 0 + (-u 2) * v 0 = 0 rw [hu, hv, hu1, hv1]; ring · change v 2 * u 1 + (-u 2) * v 1 = 0 rw [hu1, hv1]; ring · change v 2 * u 2 + (-u 2) * v 2 = 0 ring have hsecond := hpair (v 2) (-u 2) hz2 have hv2 : v 2=0 := hsecond.1 have hu2 : u 2=0 := by simpa using hsecond.2 have hz3 : (1 : ℂ) • u + (0 : ℂ) • v = 0 := by ext i fin_cases i <;> simp [hu, hu1, hu2] have := hpair 1 0 hz3 norm_num at this
If both cubics q and p vanish everywhere on the two-plane spanned by independent vectors u and v, then u₀ = k u₁ and v₀ = k v₁ for some k with k = 0 or k² = 3.
Assume for contradiction that both cubics vanish on the plane. Setting s = √3 (with s² = 3 by `Real.sq_sqrt`), the cubic factors as x(x - sy)(x + sy). Restricting to the affine line u + t v, the coordinate functions X_p(t) and Y_p(t) are polynomials in t. The polynomial X_p(X_p - s Y_p)(X_p + s Y_p) vanishes at all evaluation points t ∈ ℂ, hence is the zero polynomial by `Polynomial.funext`. Since the polynomial ring over ℂ has no zero divisors, one of the three linear factors must vanish identically. Evaluating at t = 0 and t = 1 yields u₀ = k u₁ and v₀ = k v₁ with k ∈ {0, s, -s}, so k = 0 or k² = 3.
uses
Real.sq_sqrt,Polynomial.funextby_contra hcontra dsimp only at hcontra push_neg at hcontra let s : ℂ := Real.sqrt 3 have hs : s*s=3 := by dsimp [s] have hh : (Real.sqrt 3)^2 = 3 := Real.sq_sqrt (by norm_num) have hh' : (Real.sqrt 3 : ℂ)^2 = (3 : ℂ) := by exact_mod_cast hh simpa [pow_two] using hh' have hfact (x y : ℂ) : x*(x-s*y)*(x+s*y)=x*(x*x-3*y*y) := by linear_combination (-x*y*y) * hs let Xp : Polynomial ℂ := C (u 0) + X * C (v 0) let Yp : Polynomial ℂ := C (u 1) + X * C (v 1) have hpoly : Xp * (Xp - C s * Yp) * (Xp + C s * Yp) = 0 := by apply Polynomial.funext intro t have hq := (hcontra 1 t).1 simp only [one_mul] at hq simp only [Polynomial.eval_mul, Polynomial.eval_sub, Polynomial.eval_add, Polynomial.eval_C, Polynomial.eval_X, Polynomial.eval_zero, Xp, Yp] rw [hfact] exact hq have hlin : Xp = 0 ∨ Xp - C s * Yp = 0 ∨ Xp + C s * Yp = 0 := by rcases mul_eq_zero.mp hpoly with h | h · exact mul_eq_zero.mp h |>.imp_right Or.inl · exact Or.inr (Or.inr h) have hplane : ∃ k : ℂ, u 0 = k*u 1 ∧ v 0 = k*v 1 ∧ (k=0 ∨ k*k=3) := by rcases hlin with h | h | h · have h0 := congrArg (fun P : Polynomial ℂ => P.eval 0) h have h1 := congrArg (fun P : Polynomial ℂ => P.eval 1) h simp [Xp] at h0 h1 refine ⟨0, by simpa using h0, ?_, Or.inl rfl⟩ simpa [h0] using h1 · have h0 := congrArg (fun P : Polynomial ℂ => P.eval 0) h have h1 := congrArg (fun P : Polynomial ℂ => P.eval 1) h simp [Xp, Yp] at h0 h1 refine ⟨s, by linear_combination h0, ?_, Or.inr hs⟩ linear_combination h1 - h0 · have h0 := congrArg (fun P : Polynomial ℂ => P.eval 0) h have h1 := congrArg (fun P : Polynomial ℂ => P.eval 1) h simp [Xp, Yp] at h0 h1 refine ⟨-s, by linear_combination h0, ?_, Or.inr ?_⟩ · linear_combination h1 - h0 · linear_combination hs
The cubics q and p cannot vanish simultaneously on any complex two-plane.
Since u₀ = k u₁ and v₀ = k v₁, Lemma `in767_rankproof_plane_yz_det_ne_zero` guarantees D = u₁ v₂ - u₂ v₁ ≠ 0. We can therefore choose linear combination coefficients c = v₂/D and d = -u₂/D so that the resulting vector has y = c u₁ + d v₁ = 1 and z = c u₂ + d v₂ = 0, which forces x = c u₀ + d v₀ = k y = k. Evaluating the cubic p at (k, 1, 0) yields k² + 1 = 0. But k = 0 gives 1 = 0 and k² = 3 gives 4 = 0, both of which are contradictions in ℂ.
uses
in767_rankproof_plane_yz_det_ne_zeroobtain ⟨k, hu, hv, hk⟩ := hplane let D : ℂ := u 1*v 2-u 2*v 1 have hD : D ≠ 0 := in767_rankproof_plane_yz_det_ne_zero u v k hu hv hli let c : ℂ := v 2 / D let d : ℂ := -u 2 / D have hy : c*u 1 + d*v 1 = 1 := by dsimp [c,d] field_simp [hD] dsimp [D] ring have hz : c*u 2 + d*v 2 = 0 := by dsimp [c,d] field_simp [hD] dsimp [D] ring have hx : c*u 0 + d*v 0 = k := by rw [hu,hv] linear_combination k*hy have hp := (hcontra c d).2 rw [hx,hy,hz] at hp rcases hk with rfl | hk · norm_num at hp · have hh : (4 : ℂ) = 0 := by linear_combination hp - hk norm_num at hh
For any 3 × 2 isometry U, there is a vector x ∈ ℂ² such that the contraction matrix K(U *ᵥ x) has rank 3.
The columns u, v of U are linearly independent by `in767_rankproof_isometry_cols_independent`. By `in767_rankproof_cubic_pair_no_common_plane`, there exists x = [c, d]ᵀ such that z = U *ᵥ x satisfies q(z) ≠ 0 or p(z) ≠ 0. By `in767_rankproof_contraction_minors`, this implies that either the {0, 1, 2} minor or the {0, 1, 3} minor of K(z) has non-zero determinant. By `Matrix.rank_of_det_ne_zero`, that 3 × 3 submatrix has rank 3. Since submatrix rank lower-bounds matrix rank (`Matrix.rank_submatrix_le`) and K has 3 rows (`Matrix.rank_le_card_height`), the rank of K(z) must be exactly 3.
uses
in767_rankproof_isometry_cols_independent,in767_rankproof_cubic_pair_no_common_plane,in767_rankproof_contraction_minors,Matrix.rank_le_card_height,Matrix.rank_of_det_ne_zero,Matrix.rank_submatrix_lelet u : Fin 3 → ℂ := fun i => U i 0 let v : Fin 3 → ℂ := fun i => U i 1 let α : ℂ := (Real.sqrt (131/2 : ℝ) : ℂ)/12 have hli : LinearIndependent ℂ ![u,v] := in767_rankproof_isometry_cols_independent U hU obtain ⟨c,d,h⟩ := in767_rankproof_cubic_pair_no_common_plane ((10/3 : ℂ)*α) u v hli let x : Fin 2 → ℂ := ![c,d] let z : Fin 3 → ℂ := U.mulVec x have hz (i : Fin 3) : z i = c*u i+d*v i := by simp [z, x, u, v, Matrix.mulVec, dotProduct, Fin.sum_univ_two, mul_comm] dsimp only at h rw [← hz 0, ← hz 1, ← hz 2] at h let K : Matrix (Fin 3) (Fin 4) ℂ := fun a l => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b have h2r : Real.sqrt (2 : ℝ) ≠ 0 := ne_of_gt (Real.sqrt_pos.2 (by norm_num)) have h3r : Real.sqrt (3 : ℝ) ≠ 0 := ne_of_gt (Real.sqrt_pos.2 (by norm_num)) have har : Real.sqrt (131/2 : ℝ) ≠ 0 := ne_of_gt (Real.sqrt_pos.2 (by norm_num)) have h2 : (Real.sqrt 2 : ℂ) ≠ 0 := by exact_mod_cast h2r have h3 : (Real.sqrt 3 : ℂ) ≠ 0 := by exact_mod_cast h3r have ha : α ≠ 0 := by dsimp [α]; exact div_ne_zero (by exact_mod_cast har) (by norm_num) have hminor := in767_rankproof_contraction_minors z change (K.submatrix id ![0,1,2]).det = _ ∧ (K.submatrix id ![0,1,3]).det = _ at hminor refine ⟨x, ?_⟩ change K.rank = 3 have hheight : K.rank ≤ 3 := by simpa using Matrix.rank_le_card_height K rcases h with hq | hp · have hdet : (K.submatrix id ![0,1,2]).det ≠ 0 := by rw [hminor.1] exact mul_ne_zero (mul_ne_zero (mul_ne_zero (inv_ne_zero h2) ha) (by norm_num)) hq have hfull : (K.submatrix id ![0,1,2]).rank = 3 := by simpa using Matrix.rank_of_det_ne_zero hdet have hle := Matrix.rank_submatrix_le K id ![0,1,2] omega · have hdet : (K.submatrix id ![0,1,3]).det ≠ 0 := by rw [hminor.2] exact mul_ne_zero (mul_ne_zero (mul_ne_zero (inv_ne_zero h2) (inv_ne_zero h3)) (by norm_num)) hp have hfull : (K.submatrix id ![0,1,3]).rank = 3 := by simpa using Matrix.rank_of_det_ne_zero hdet have hle := Matrix.rank_submatrix_le K id ![0,1,3] omega
The contraction of the four source vectors has full row rank 3 for some vector in every complex two-plane.
Given an isometry U representing the complex two-plane, applying `in767_rankproof_source_full_rank_on_plane` directly produces the required vector x and witnesses the rank condition.
uses
in767_rankproof_source_full_rank_on_planeintro U hU exact in767_rankproof_source_full_rank_on_plane U hU
Matrix.mulVec_mulVec- Matrix-vector multiplication is associative: M *ᵥ (N *ᵥ v) = (M * N) *ᵥ v.
Matrix.one_mulVec- The identity matrix acts as the identity on vectors: 1 *ᵥ v = v.
Matrix.mulVec_injective_iff.mp- If matrix-vector multiplication by M is injective, then the columns of M are linearly independent.
Fin.sum_univ_three- A sum over Fin 3 expands to f 0 + f 1 + f 2.
in767_rankproof_contraction_explicit- For any z ∈ ℂ³, the contraction of the four PPT Bell vectors against z equals the explicit 3 × 4 matrix given in coordinates.
Matrix.det_fin_three- The determinant of a 3 × 3 matrix equals the standard alternating sum of six products of its entries.
LinearIndependent.pair_iff.mp- If a pair of vectors ![u, v] is linearly independent, then s • u + t • v = 0 implies s = 0 and t = 0.
Real.sq_sqrt- For non-negative real x, (√x)² = x.
Polynomial.funext- Two polynomials over an infinite domain that evaluate to the same value everywhere are equal.
in767_rankproof_plane_yz_det_ne_zero- If two linearly independent vectors u, v ∈ ℂ³ satisfy u 0 = k * u 1 and v 0 = k * v 1, then the minor u 1 * v 2 - u 2 * v 1 is non-zero.
in767_rankproof_isometry_cols_independent- The two column vectors of a 3 × 2 complex isometry U (satisfying Uᴴ * U = 1) are linearly independent.
in767_rankproof_cubic_pair_no_common_plane- On the two-plane spanned by two linearly independent vectors u, v ∈ ℂ³, the cubics x*(x² - 3y²) and x²y + a x²z + y³ - a y²z - y z² / 18 do not vanish simultaneously everywhere.
in767_rankproof_contraction_minors- The 3 × 3 minors on columns {0, 1, 2} and {0, 1, 3} of the contraction matrix K(z) are proportional to the cubics q(z) and p(z), respectively, with non-zero scalar factors.
Matrix.rank_le_card_height- The rank of an m × n matrix is bounded above by the cardinality of its row index type m.
Matrix.rank_of_det_ne_zero- A square matrix over an integral domain with non-zero determinant has full rank equal to its dimension.
Matrix.rank_submatrix_le- The rank of any submatrix of A is at most the rank of A.
in767_rankproof_source_full_rank_on_plane- For any 3 × 2 complex isometry U, there exists x ∈ ℂ² such that the contraction matrix K(U *ᵥ x) has rank 3.
- In steps 2 and 3, verification of the matrix contraction entries and determinant formulas is carried out by automated polynomial and ring arithmetic.
- In step 7, the inequality between the submatrix rank and the ambient matrix rank is closed by the linear integer arithmetic tactic `omega`.
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
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
open Matrix Polynomialopen scoped BigOperators Matrix ComplexOrderset_option backward.isDefEq.respectTransparency falseset_option linter.unusedSimpArgs falseprivate lemma in767_rankproof_isometry_cols_independent (U : Matrix (Fin 3) (Fin 2) ℂ) (hU : Uᴴ*U=1) : LinearIndependent ℂ ![(fun i => U i 0),(fun i => U i 1)] := by have hinj : Function.Injective U.mulVec := by intro x y h have hh := congrArg (fun v => Uᴴ *ᵥ v) h simpa only [Matrix.mulVec_mulVec, hU, Matrix.one_mulVec] using hh have hcol := Matrix.mulVec_injective_iff.mp hinj have heq : ![(fun i => U i 0),(fun i => U i 1)] = U.col := by funext j i fin_cases j <;> rfl rw [heq] exact hcolprivate lemma in767_rankproof_contraction_explicit (z : Fin 3 → ℂ) : (fun (a : Fin 3) (l : Fin 4) => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b) = !![(Real.sqrt 2 : ℂ)⁻¹*z 0, ((Real.sqrt (131/2 : ℝ) : ℂ)/12)*z 1+z 2/60, ((Real.sqrt (131/2 : ℝ) : ℂ)/12)*z 0, -(Real.sqrt 3 : ℂ)⁻¹*z 1; (Real.sqrt 2 : ℂ)⁻¹*z 1, ((Real.sqrt (131/2 : ℝ) : ℂ)/12)*z 0, -((Real.sqrt (131/2 : ℝ) : ℂ)/12)*z 1+z 2/60, (Real.sqrt 3 : ℂ)⁻¹*z 0; 0, -(3/10 : ℂ)*z 1, (3/10 : ℂ)*z 0, (Real.sqrt 3 : ℂ)⁻¹*z 2] := by ext i l fin_cases i <;> fin_cases l <;> simp [ppt_qutrit_bell_vector, Fin.sum_univ_three] <;> ringprivate lemma in767_rankproof_contraction_minors (z : Fin 3 → ℂ) : let K : Matrix (Fin 3) (Fin 4) ℂ := fun a l => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b let α : ℂ := ((Real.sqrt (131/2 : ℝ) : ℂ)/12) let q : ℂ := z 0 * (z 0*z 0 - 3*z 1*z 1) let p : ℂ := z 0*z 0*z 1 + ((10/3 : ℂ)*α)*z 0*z 0*z 2 + z 1*z 1*z 1 - ((10/3 : ℂ)*α)*z 1*z 1*z 2 - z 1*z 2*z 2/18 (K.submatrix id ![0,1,2]).det = ((Real.sqrt 2 : ℂ)⁻¹)*α*(3/10)*q ∧ (K.submatrix id ![0,1,3]).det = ((Real.sqrt 2 : ℂ)⁻¹)*((Real.sqrt 3 : ℂ)⁻¹)*(3/10)*p := by dsimp only rw [in767_rankproof_contraction_explicit] constructor <;> simp [Matrix.det_fin_three, Matrix.submatrix_apply] <;> ringprivate lemma in767_rankproof_plane_yz_det_ne_zero (u v : Fin 3 → ℂ) (k : ℂ) (hu : u 0 = k*u 1) (hv : v 0 = k*v 1) (hli : LinearIndependent ℂ ![u,v]) : u 1 * v 2 - u 2 * v 1 ≠ 0 := by intro hd have hpair := LinearIndependent.pair_iff.mp hli have hz1 : (v 1) • u + (-u 1) • v = 0 := by ext i fin_cases i <;> simp only [Pi.add_apply, Pi.smul_apply, smul_eq_mul, Pi.zero_apply] · change v 1 * u 0 + (-u 1) * v 0 = 0 rw [hu, hv]; ring · change v 1 * u 1 + (-u 1) * v 1 = 0 ring · change v 1 * u 2 + (-u 1) * v 2 = 0 linear_combination -hd have hfirst := hpair (v 1) (-u 1) hz1 have hv1 : v 1=0 := hfirst.1 have hu1 : u 1=0 := by simpa using hfirst.2 have hz2 : (v 2) • u + (-u 2) • v = 0 := by ext i fin_cases i <;> simp only [Pi.add_apply, Pi.smul_apply, smul_eq_mul, Pi.zero_apply] · change v 2 * u 0 + (-u 2) * v 0 = 0 rw [hu, hv, hu1, hv1]; ring · change v 2 * u 1 + (-u 2) * v 1 = 0 rw [hu1, hv1]; ring · change v 2 * u 2 + (-u 2) * v 2 = 0 ring have hsecond := hpair (v 2) (-u 2) hz2 have hv2 : v 2=0 := hsecond.1 have hu2 : u 2=0 := by simpa using hsecond.2 have hz3 : (1 : ℂ) • u + (0 : ℂ) • v = 0 := by ext i fin_cases i <;> simp [hu, hu1, hu2] have := hpair 1 0 hz3 norm_num at thisprivate lemma in767_rankproof_cubic_pair_no_common_plane (a : ℂ) (u v : Fin 3 → ℂ) (hli : LinearIndependent ℂ ![u, v]) : ∃ c d : ℂ, let x := c*u 0+d*v 0 let y := c*u 1+d*v 1 let z := c*u 2+d*v 2 x*(x*x-3*y*y) ≠ 0 ∨ x*x*y + a*x*x*z + y*y*y - a*y*y*z - y*z*z/18 ≠ 0 := by by_contra hcontra dsimp only at hcontra push_neg at hcontra let s : ℂ := Real.sqrt 3 have hs : s*s=3 := by dsimp [s] have hh : (Real.sqrt 3)^2 = 3 := Real.sq_sqrt (by norm_num) have hh' : (Real.sqrt 3 : ℂ)^2 = (3 : ℂ) := by exact_mod_cast hh simpa [pow_two] using hh' have hfact (x y : ℂ) : x*(x-s*y)*(x+s*y)=x*(x*x-3*y*y) := by linear_combination (-x*y*y) * hs let Xp : Polynomial ℂ := C (u 0) + X * C (v 0) let Yp : Polynomial ℂ := C (u 1) + X * C (v 1) have hpoly : Xp * (Xp - C s * Yp) * (Xp + C s * Yp) = 0 := by apply Polynomial.funext intro t have hq := (hcontra 1 t).1 simp only [one_mul] at hq simp only [Polynomial.eval_mul, Polynomial.eval_sub, Polynomial.eval_add, Polynomial.eval_C, Polynomial.eval_X, Polynomial.eval_zero, Xp, Yp] rw [hfact] exact hq have hlin : Xp = 0 ∨ Xp - C s * Yp = 0 ∨ Xp + C s * Yp = 0 := by rcases mul_eq_zero.mp hpoly with h | h · exact mul_eq_zero.mp h |>.imp_right Or.inl · exact Or.inr (Or.inr h) have hplane : ∃ k : ℂ, u 0 = k*u 1 ∧ v 0 = k*v 1 ∧ (k=0 ∨ k*k=3) := by rcases hlin with h | h | h · have h0 := congrArg (fun P : Polynomial ℂ => P.eval 0) h have h1 := congrArg (fun P : Polynomial ℂ => P.eval 1) h simp [Xp] at h0 h1 refine ⟨0, by simpa using h0, ?_, Or.inl rfl⟩ simpa [h0] using h1 · have h0 := congrArg (fun P : Polynomial ℂ => P.eval 0) h have h1 := congrArg (fun P : Polynomial ℂ => P.eval 1) h simp [Xp, Yp] at h0 h1 refine ⟨s, by linear_combination h0, ?_, Or.inr hs⟩ linear_combination h1 - h0 · have h0 := congrArg (fun P : Polynomial ℂ => P.eval 0) h have h1 := congrArg (fun P : Polynomial ℂ => P.eval 1) h simp [Xp, Yp] at h0 h1 refine ⟨-s, by linear_combination h0, ?_, Or.inr ?_⟩ · linear_combination h1 - h0 · linear_combination hs obtain ⟨k, hu, hv, hk⟩ := hplane let D : ℂ := u 1*v 2-u 2*v 1 have hD : D ≠ 0 := in767_rankproof_plane_yz_det_ne_zero u v k hu hv hli let c : ℂ := v 2 / D let d : ℂ := -u 2 / D have hy : c*u 1 + d*v 1 = 1 := by dsimp [c,d] field_simp [hD] dsimp [D] ring have hz : c*u 2 + d*v 2 = 0 := by dsimp [c,d] field_simp [hD] dsimp [D] ring have hx : c*u 0 + d*v 0 = k := by rw [hu,hv] linear_combination k*hy have hp := (hcontra c d).2 rw [hx,hy,hz] at hp rcases hk with rfl | hk · norm_num at hp · have hh : (4 : ℂ) = 0 := by linear_combination hp - hk norm_num at hhprivate lemma in767_rankproof_source_full_rank_on_plane (U : Matrix (Fin 3) (Fin 2) ℂ) (hU : Uᴴ*U=1) : ∃ x : Fin 2 → ℂ, let z : Fin 3 → ℂ := U.mulVec x let K : Matrix (Fin 3) (Fin 4) ℂ := fun a l => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b K.rank = 3 := by let u : Fin 3 → ℂ := fun i => U i 0 let v : Fin 3 → ℂ := fun i => U i 1 let α : ℂ := (Real.sqrt (131/2 : ℝ) : ℂ)/12 have hli : LinearIndependent ℂ ![u,v] := in767_rankproof_isometry_cols_independent U hU obtain ⟨c,d,h⟩ := in767_rankproof_cubic_pair_no_common_plane ((10/3 : ℂ)*α) u v hli let x : Fin 2 → ℂ := ![c,d] let z : Fin 3 → ℂ := U.mulVec x have hz (i : Fin 3) : z i = c*u i+d*v i := by simp [z, x, u, v, Matrix.mulVec, dotProduct, Fin.sum_univ_two, mul_comm] dsimp only at h rw [← hz 0, ← hz 1, ← hz 2] at h let K : Matrix (Fin 3) (Fin 4) ℂ := fun a l => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b have h2r : Real.sqrt (2 : ℝ) ≠ 0 := ne_of_gt (Real.sqrt_pos.2 (by norm_num)) have h3r : Real.sqrt (3 : ℝ) ≠ 0 := ne_of_gt (Real.sqrt_pos.2 (by norm_num)) have har : Real.sqrt (131/2 : ℝ) ≠ 0 := ne_of_gt (Real.sqrt_pos.2 (by norm_num)) have h2 : (Real.sqrt 2 : ℂ) ≠ 0 := by exact_mod_cast h2r have h3 : (Real.sqrt 3 : ℂ) ≠ 0 := by exact_mod_cast h3r have ha : α ≠ 0 := by dsimp [α]; exact div_ne_zero (by exact_mod_cast har) (by norm_num) have hminor := in767_rankproof_contraction_minors z change (K.submatrix id ![0,1,2]).det = _ ∧ (K.submatrix id ![0,1,3]).det = _ at hminor refine ⟨x, ?_⟩ change K.rank = 3 have hheight : K.rank ≤ 3 := by simpa using Matrix.rank_le_card_height K rcases h with hq | hp · have hdet : (K.submatrix id ![0,1,2]).det ≠ 0 := by rw [hminor.1] exact mul_ne_zero (mul_ne_zero (mul_ne_zero (inv_ne_zero h2) ha) (by norm_num)) hq have hfull : (K.submatrix id ![0,1,2]).rank = 3 := by simpa using Matrix.rank_of_det_ne_zero hdet have hle := Matrix.rank_submatrix_le K id ![0,1,2] omega · have hdet : (K.submatrix id ![0,1,3]).det ≠ 0 := by rw [hminor.2] exact mul_ne_zero (mul_ne_zero (mul_ne_zero (inv_ne_zero h2) (inv_ne_zero h3)) (by norm_num)) hp have hfull : (K.submatrix id ![0,1,3]).rank = 3 := by simpa using Matrix.rank_of_det_ne_zero hdet have hle := Matrix.rank_submatrix_le K id ![0,1,3] omegatheorem ppt_qutrit_bell_source_full_rank_on_every_plane : ∀ U : Matrix (Fin 3) (Fin 2) ℂ, Uᴴ * U = 1 → ∃ x : Fin 2 → ℂ, let z : Fin 3 → ℂ := U.mulVec x let K : Matrix (Fin 3) (Fin 4) ℂ := fun a l => ∑ b : Fin 3, ppt_qutrit_bell_vector l (a,b) * z b K.rank = 3 := by intro U hU exact in767_rankproof_source_full_rank_on_plane U hU