AFTD Auto-Formalizing Theoretical Domains

ppt_qutrit_rational_binary_effect_incompatible

theoremverified

The three explicit rational qutrit effects have no joint positive parent.

Statement

theorem ppt_qutrit_rational_binary_effect_incompatible :
  ¬ ∃ G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ,
    (∀ s, (G s).PosSemidef) ∧ (∑ s, G s = 1) ∧
    ∀ j : Fin 3, (∑ s, if s j then G s else 0) = ppt_qutrit_rational_binary_effect j
Source
Helper for OP-693, derived from the Bell witness of arXiv:1405.4502; rational matrices and exact dual certificates in experiments 22874, 22875, 22880, and 22905.
Verified
12 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 1

The proof, explained

The theorem proves that three explicit rational qutrit effects are incompatible (they have no joint positive parent POVM) using semidefinite programming duality. Assuming for contradiction that a joint positive parent G exists, the proof constructs an explicit dual certificate consisting of positive semidefinite matrices Z(s) for each binary assignment s. Because both Z(s) and G(s) are positive semidefinite, the total trace inner product ∑_s Re(Tr(Z(s) * G(s))) must be non-negative; however, expanding Z(s) and applying the normalization and marginal conditions shows that this sum equals an explicit rational scalar that is verified by exact arithmetic to be strictly negative.

  1. The real part of the trace inner product of any two positive semidefinite complex matrices is non-negative: 0 ≤ Re(Tr(P * ρ)).

    By continuous functional calculus for self-adjoint operators with non-negative spectrum, P has a self-adjoint square root X such that X² = P. Using the associative and cyclic properties of the matrix trace, Tr(P * ρ) = Tr(X² * ρ) = Tr(X * ρ * X). Since ρ is positive semidefinite and X is self-adjoint, the congruence X * ρ * X = X * ρ * Xᴴ is positive semidefinite, and therefore its trace has a non-negative real part.

    lines 11–18

    uses CFC.exists_sqrt_of_isSelfAdjoint_of_quasispectrumRestricts, QuasispectrumRestricts.nnreal_of_nonneg, Matrix.mul_assoc, Matrix.trace_mul_comm, Matrix.star_eq_conjTranspose

    private lemma in767_witness_trace_psd {n : Type} [Fintype n] [DecidableEq n] {P ρ : Matrix n n ℂ}
        (hP : P.PosSemidef) (hρ : ρ.PosSemidef) : 0 ≤ (P * ρ).trace.re := by
      obtain ⟨X, hX, -, rfl⟩ := CFC.exists_sqrt_of_isSelfAdjoint_of_quasispectrumRestricts hP.isHermitian
        (QuasispectrumRestricts.nnreal_of_nonneg hP.nonneg)
      rw [Matrix.mul_assoc, Matrix.trace_mul_comm, Matrix.mul_assoc, ← Matrix.mul_assoc]
      have h := hρ.mul_mul_conjTranspose_same X
      rw [← Matrix.star_eq_conjTranspose, hX.star_eq] at h
      exact (Complex.nonneg_iff.1 h.trace_nonneg).1
  2. The four dual witness vectors evaluate to their specified rational coordinate vectors. automation

    The function defining the dual vectors is given by an if-then-else chain over index i ∈ Fin 4. For each index 0, 1, 2, 3, the conditional branches are resolved and the rational components are evaluated.

    lines 20–30

    uses if_false, if_true

    private noncomputable def in767_witness_dual_vec (i : Fin 4) : Fin 3 → ℂ :=
      if i = 0 then ![10599/1557500, 1791231/3115000, 1991/89000]
      else if i = 1 then ![-773727/1557500, -448691/1557500, 3993/178000]
      else if i = 2 then ![1565119/3115000, -455757/1557500, 1947/89000]
      else ![0, 0, 77/2000]
    
    private lemma in767_witness_dual_vec0 : in767_witness_dual_vec 0 = ![10599/1557500, 1791231/3115000, 1991/89000] := by norm_num [in767_witness_dual_vec]
    private lemma in767_witness_dual_vec1 : in767_witness_dual_vec 1 = ![-773727/1557500, -448691/1557500, 3993/178000] := by norm_num [in767_witness_dual_vec]
    private lemma in767_witness_dual_vec2 : in767_witness_dual_vec 2 = ![1565119/3115000, -455757/1557500, 1947/89000] := by simp only [in767_witness_dual_vec, show (2 : Fin 4) ≠ 0 from by decide, show (2 : Fin 4) ≠ 1 from by decide, if_false, if_true]
    private lemma in767_witness_dual_vec3 : in767_witness_dual_vec 3 = ![0, 0, 77/2000] := by simp only [in767_witness_dual_vec, show (3 : Fin 4) ≠ 0 from by decide, show (3 : Fin 4) ≠ 1 from by decide, show (3 : Fin 4) ≠ 2 from by decide, if_false]
    
  3. For every binary assignment s : Fin 3 → Bool, the dual matrix Z(s) is positive semidefinite.

    Each projector P_i is the outer product of a vector with its conjugate transpose, hence positive semidefinite. For seven of the eight boolean assignments s, Z(s) simplifies via abelian group operations to the sum of two such projectors P_a + P_b, which is positive semidefinite. For the remaining assignment s = (true, true, true), Z(s) decomposes as v * vᴴ + w * wᴴ for two explicit vectors v = (3533/5000, 0, 0) and w = (0, 3533/5000, 0), which is verified component-wise across all nine matrix entries by exact rational arithmetic.

    lines 31–69

    uses two_smul, in767_witness_dual_p_psd, Matrix.posSemidef_vecMulVec_self_star

    private noncomputable def in767_witness_dual_p (i : Fin 4) : Matrix (Fin 3) (Fin 3) ℂ :=
      Matrix.vecMulVec (in767_witness_dual_vec i) (star (in767_witness_dual_vec i))
    
    private noncomputable def in767_witness_dual_z (s : Fin 3 → Bool) : Matrix (Fin 3) (Fin 3) ℂ :=
      (2 : ℝ) • in767_witness_dual_p 3 + (if s 0 then in767_witness_dual_p 1 - in767_witness_dual_p 3 else 0) +
      (if s 1 then in767_witness_dual_p 0 - in767_witness_dual_p 3 else 0) +
      (if s 2 then in767_witness_dual_p 2 - in767_witness_dual_p 3 else 0)
    
    private lemma in767_witness_dual_p_psd (i : Fin 4) : (in767_witness_dual_p i).PosSemidef :=
      Matrix.posSemidef_vecMulVec_self_star _
    
    private lemma in767_witness_dual_all_psd (s : Fin 3 → Bool) : (in767_witness_dual_z s).PosSemidef := by
      cases h0 : s 0 <;> cases h1 : s 1 <;> cases h2 : s 2
      · have he : in767_witness_dual_z s = in767_witness_dual_p 3 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 3).add (in767_witness_dual_p_psd 3)
      · have he : in767_witness_dual_z s = in767_witness_dual_p 2 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 2).add (in767_witness_dual_p_psd 3)
      · have he : in767_witness_dual_z s = in767_witness_dual_p 0 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 0).add (in767_witness_dual_p_psd 3)
      · have he : in767_witness_dual_z s = in767_witness_dual_p 0 + in767_witness_dual_p 2 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 0).add (in767_witness_dual_p_psd 2)
      · have he : in767_witness_dual_z s = in767_witness_dual_p 1 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 1).add (in767_witness_dual_p_psd 3)
      · have he : in767_witness_dual_z s = in767_witness_dual_p 1 + in767_witness_dual_p 2 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 1).add (in767_witness_dual_p_psd 2)
      · have he : in767_witness_dual_z s = in767_witness_dual_p 0 + in767_witness_dual_p 1 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel
        rw [he]; exact (in767_witness_dual_p_psd 0).add (in767_witness_dual_p_psd 1)
      · let v : Fin 3 → ℂ := ![3533/5000, 0, 0]
        let w : Fin 3 → ℂ := ![0, 3533/5000, 0]
        have he : in767_witness_dual_z s = Matrix.vecMulVec v (star v) + Matrix.vecMulVec w (star w) := by
          simp only [in767_witness_dual_z, h0, h1, h2, if_true, in767_witness_dual_p,
            in767_witness_dual_vec0, in767_witness_dual_vec1, in767_witness_dual_vec2, in767_witness_dual_vec3, two_smul]
          ext i j
          fin_cases i <;> fin_cases j <;>
            simp only [Matrix.add_apply, Matrix.sub_apply, Matrix.vecMulVec_apply, Pi.star_apply] <;>
            norm_num [v, w, map_div₀, map_ofNat]
        rw [he]
        exact (Matrix.posSemidef_vecMulVec_self_star v).add (Matrix.posSemidef_vecMulVec_self_star w)
    
  4. The dual objective value 2 * Re(Tr(P₃)) + Re(Tr((P₁ - P₃) * E₀)) + Re(Tr((P₀ - P₃) * E₁)) + Re(Tr((P₂ - P₃) * E₂)) is strictly negative. automation

    The traces of the matrix products involving the rank-one projectors and the rational effect matrices are expanded into explicit sums of rational numbers and evaluated. Lean's exact arithmetic automation verifies that the final rational scalar is strictly less than zero.

    lines 70–78

    uses Fin.sum_univ_succ

    private lemma in767_witness_rational_dual_number :
        (2 * (in767_witness_dual_p 3).trace.re +
          ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * in767_witness_effect0).trace.re +
          ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * in767_witness_effect1).trace.re +
          ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * in767_witness_effect2).trace.re) < 0 := by
      simp only [in767_witness_dual_p, in767_witness_dual_vec0, in767_witness_dual_vec1, in767_witness_dual_vec2, in767_witness_dual_vec3]
      simp only [Matrix.trace, Matrix.diag_apply, Matrix.mul_apply, Matrix.sub_apply, Matrix.vecMulVec_apply, Pi.star_apply, Fin.sum_univ_succ]
      norm_num [in767_witness_effect0, in767_witness_effect1, in767_witness_effect2, Matrix.cons_val_two, Matrix.vecHead, Matrix.vecTail, Matrix.mul_apply, Matrix.sub_apply, Matrix.vecMulVec_apply, Pi.star_apply, Fin.sum_univ_succ, map_div₀, map_ofNat]
    
  5. The trace inner product Re(Tr(A * ·)) distributes over finite matrix sums: Re(Tr(A * ∑_s G(s))) = ∑_s Re(Tr(A * G(s))).

    Matrix multiplication distributes over finite sums, the trace commutes with summation, and taking the real part commutes with finite addition of complex numbers.

    lines 79–83

    uses Matrix.mul_sum, Matrix.trace_sum, Complex.re_sum

    private lemma in767_witness_trace_matrix_sum (A : Matrix (Fin 3) (Fin 3) ℂ)
        (G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ) :
        (A * ∑ s, G s).trace.re = ∑ s, (A * G s).trace.re := by
      simp only [Matrix.mul_sum, Matrix.trace_sum, Complex.re_sum]
    
  6. Under the assumption that a joint parent G exists, summing Re(Tr(A * G(s))) over all assignments where s(j) is true yields Re(Tr(A * E_j)).

    Suppose for contradiction that a joint positive parent G exists. Multiplying A by the marginal condition ∑_{s: s(j)=true} G(s) = E_j and applying the trace linearity lemma shows that the marginal sum of trace inner products equals Re(Tr(A * E_j)).

    lines 84–95

    uses in767_witness_trace_matrix_sum, Finset.sum_congr

    private theorem in767_witness_rational_not_joint :
        ¬ ∃ G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ,
          (∀ s, (G s).PosSemidef) ∧ (∑ s, G s = 1) ∧
          ∀ j : Fin 3, (∑ s, if s j then G s else 0) = in767_witness_rational_effect j := by
      classical
      rintro ⟨G, hG, hsum, hmarg⟩
      have hmargin (A : Matrix (Fin 3) (Fin 3) ℂ) (j : Fin 3) :
          (∑ s, if s j then (A * G s).trace.re else 0) = (A * in767_witness_rational_effect j).trace.re := by
        rw [← hmarg j, in767_witness_trace_matrix_sum]
        apply Finset.sum_congr rfl
        intro s _
        split_ifs <;> simp_all
  7. For each assignment s, the trace inner product Re(Tr(Z(s) * G(s))) splits into a base term 2 * Re(Tr(P₃ * G(s))) and conditional terms for each measurement outcome s(j).

    By case analysis on the boolean values s(0), s(1), and s(2), the definition of Z(s) is substituted. Distributing matrix multiplication and linearity of trace over matrix addition and subtraction decomposes the trace into four separate terms.

    lines 96–103

    uses two_smul, Matrix.add_mul, Matrix.sub_mul, Matrix.trace_add, Matrix.trace_sub

      have hsplit (s : Fin 3 → Bool) :
          (in767_witness_dual_z s * G s).trace.re =
            2 * (in767_witness_dual_p 3 * G s).trace.re +
            (if s 0 then ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * G s).trace.re else 0) +
            (if s 1 then ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * G s).trace.re else 0) +
            (if s 2 then ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * G s).trace.re else 0) := by
        cases h0 : s 0 <;> cases h1 : s 1 <;> cases h2 : s 2 <;>
          simp [in767_witness_dual_z, h0, h1, h2, two_smul, Matrix.add_mul, Matrix.sub_mul, Matrix.trace_add, Matrix.trace_sub] <;> ring
  8. The total sum ∑_s Re(Tr(Z(s) * G(s))) algebraically equals the dual objective value.

    Summing the pointwise expansion over all assignments s, we split the sum into four parts. The first part, 2 * ∑_s Re(Tr(P₃ * G(s))), simplifies to 2 * Re(Tr(P₃)) using the normalization condition ∑_s G(s) = 1. The remaining three parts correspond to the marginals at j = 0, 1, 2, which equal Re(Tr((P₁ - P₃) * E₀)), Re(Tr((P₀ - P₃) * E₁)), and Re(Tr((P₂ - P₃) * E₂)) respectively.

    lines 104–119

    uses Finset.sum_congr, Finset.sum_add_distrib, in767_witness_trace_matrix_sum, Matrix.mul_one

      have heq : (∑ s, (in767_witness_dual_z s * G s).trace.re) =
          2 * (in767_witness_dual_p 3).trace.re +
            ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * in767_witness_effect0).trace.re +
            ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * in767_witness_effect1).trace.re +
            ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * in767_witness_effect2).trace.re := by
        calc
          _ = ∑ s, (2 * (in767_witness_dual_p 3 * G s).trace.re +
              (if s 0 then ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * G s).trace.re else 0) +
              (if s 1 then ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * G s).trace.re else 0) +
              (if s 2 then ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * G s).trace.re else 0)) := by
            exact Finset.sum_congr rfl fun s _ => hsplit s
          _ = _ := by
            simp only [Finset.sum_add_distrib, ← Finset.mul_sum]
            rw [← in767_witness_trace_matrix_sum, hsum, Matrix.mul_one]
            rw [hmargin _ 0, hmargin _ 1, hmargin _ 2]
            simp [in767_witness_rational_effect]
  9. The existence of a joint parent G leads to a contradiction between non-negativity and strict negativity of the dual sum. automation

    Since each Z(s) is positive semidefinite and each G(s) is assumed positive semidefinite, every term Re(Tr(Z(s) * G(s))) is non-negative, so the total sum is non-negative. However, this sum was proved equal to the dual objective value, which is strictly negative. Linear arithmetic derives the contradiction.

    lines 120–125

    uses Finset.sum_nonneg, in767_witness_trace_psd, in767_witness_dual_all_psd, in767_witness_rational_dual_number

      have hnonneg : 0 ≤ ∑ s, (in767_witness_dual_z s * G s).trace.re :=
        Finset.sum_nonneg fun s _ => in767_witness_trace_psd (in767_witness_dual_all_psd s) (hG s)
      rw [heq] at hnonneg
      have := in767_witness_rational_dual_number
      linarith
    
  10. The three explicit rational qutrit effects have no joint positive parent.

    The theorem follows directly from the incompatibility result in767_witness_rational_not_joint, because the effect definitions coincide.

    lines 126–130

    uses in767_witness_rational_not_joint

    theorem ppt_qutrit_rational_binary_effect_incompatible :
      ¬ ∃ G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ,
        (∀ s, (G s).PosSemidef) ∧ (∑ s, G s = 1) ∧
        ∀ j : Fin 3, (∑ s, if s j then G s else 0) = ppt_qutrit_rational_binary_effect j := by
      exact in767_witness_rational_not_joint
CFC.exists_sqrt_of_isSelfAdjoint_of_quasispectrumRestricts
A self-adjoint element whose spectrum is non-negative has a self-adjoint square root.
Complex.re_sum
The real part of a finite sum of complex numbers equals the sum of the real parts.
Fin.sum_univ_succ
A sum over Fin (n + 1) splits into the term at 0 and the sum of the remaining n terms.
Finset.sum_add_distrib
A sum of sums can be split into two separate sums.
Finset.sum_congr
Sums over equal index sets are equal if their summands are pointwise equal.
Finset.sum_nonneg
A finite sum of non-negative elements is non-negative.
if_false
An if-then-else expression with a false condition evaluates to the else branch.
if_true
An if-then-else expression with a true condition evaluates to the then branch.
in767_witness_dual_all_psd
The dual witness matrix Z(s) is positive semidefinite for all eight assignments s.
in767_witness_dual_p_psd
Each of the rank-one dual witness matrices P_i is positive semidefinite.
in767_witness_rational_dual_number
The dual witness objective value evaluated on the three rational effects is strictly negative.
in767_witness_rational_not_joint
There exists no joint positive parent POVM for the three rational qutrit effects.
in767_witness_trace_matrix_sum
The real part of the trace of a matrix multiplied by a sum of matrices equals the sum of the trace inner products.
in767_witness_trace_psd
The trace inner product Re(Tr(P * ρ)) of two positive semidefinite matrices is non-negative.
Matrix.add_mul
Right multiplication distributes over matrix addition.
Matrix.mul_assoc
Matrix multiplication is associative.
Matrix.mul_sum
Matrix multiplication distributes on the left over a finite sum of matrices.
Matrix.mul_one
Multiplying a matrix on the right by the identity matrix leaves it unchanged.
Matrix.posSemidef_vecMulVec_self_star
The outer product of any vector with its conjugate transpose is positive semidefinite.
Matrix.star_eq_conjTranspose
The adjoint of a complex matrix is its conjugate transpose.
Matrix.sub_mul
Right multiplication distributes over matrix subtraction.
Matrix.trace_add
The trace is additive over matrix addition.
Matrix.trace_mul_comm
The trace of a product of matrices is invariant under cyclic permutation: Tr(AB) = Tr(BA).
Matrix.trace_sub
The trace is additive over matrix subtraction.
Matrix.trace_sum
The trace of a finite sum of matrices equals the sum of their traces.
QuasispectrumRestricts.nnreal_of_nonneg
An element greater than or equal to zero has its spectrum restricted to the non-negative reals.
two_smul
Multiplying by the scalar 2 equals adding an element to itself.

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 ComplexOrder Matrix BigOperatorsset_option maxRecDepth 8192set_option maxHeartbeats 4000000private noncomputable def in767_witness_effect0 : Matrix (Fin 3) (Fin 3) ℂ := !![(1215916159000 / 21192203043001), (-2086398000000 / 21192203043001), (1071090000 / 65981951167); (-2086398000000 / 21192203043001), (3622358659000 / 21192203043001), (532050000 / 65981951167); (1071090000 / 65981951167), (532050000 / 65981951167), (7650594020 / 10066309561)]private noncomputable def in767_witness_effect1 : Matrix (Fin 3) (Fin 3) ℂ := !![(316737473028500 / 1360728454130809), 0, 0; 0, (605951028500 / 1360728454130809), (-76179480000 / 4236629775103); 0, (-76179480000 / 4236629775103), (488260819680 / 646346857849)]private noncomputable def in767_witness_effect2 : Matrix (Fin 3) (Fin 3) ℂ := !![(1215916159000 / 21192203043001), (2086398000000 / 21192203043001), (-1071090000 / 65981951167); (2086398000000 / 21192203043001), (3622358659000 / 21192203043001), (532050000 / 65981951167); (-1071090000 / 65981951167), (532050000 / 65981951167), (7650594020 / 10066309561)]private noncomputable def in767_witness_rational_effect (j : Fin 3) : Matrix (Fin 3) (Fin 3) ℂ :=  if j = 0 then in767_witness_effect0 else if j = 1 then in767_witness_effect1 else in767_witness_effect2open scoped MatrixOrderprivate lemma in767_witness_trace_psd {n : Type} [Fintype n] [DecidableEq n] {P ρ : Matrix n n ℂ}    (hP : P.PosSemidef) (hρ : ρ.PosSemidef) : 0 ≤ (P * ρ).trace.re := by  obtain ⟨X, hX, -, rfl⟩ := CFC.exists_sqrt_of_isSelfAdjoint_of_quasispectrumRestricts hP.isHermitian    (QuasispectrumRestricts.nnreal_of_nonneg hP.nonneg)  rw [Matrix.mul_assoc, Matrix.trace_mul_comm, Matrix.mul_assoc, ← Matrix.mul_assoc]  have h := hρ.mul_mul_conjTranspose_same X  rw [← Matrix.star_eq_conjTranspose, hX.star_eq] at h  exact (Complex.nonneg_iff.1 h.trace_nonneg).1private noncomputable def in767_witness_dual_vec (i : Fin 4) : Fin 3 → ℂ :=  if i = 0 then ![10599/1557500, 1791231/3115000, 1991/89000]  else if i = 1 then ![-773727/1557500, -448691/1557500, 3993/178000]  else if i = 2 then ![1565119/3115000, -455757/1557500, 1947/89000]  else ![0, 0, 77/2000]private lemma in767_witness_dual_vec0 : in767_witness_dual_vec 0 = ![10599/1557500, 1791231/3115000, 1991/89000] := by norm_num [in767_witness_dual_vec]private lemma in767_witness_dual_vec1 : in767_witness_dual_vec 1 = ![-773727/1557500, -448691/1557500, 3993/178000] := by norm_num [in767_witness_dual_vec]private lemma in767_witness_dual_vec2 : in767_witness_dual_vec 2 = ![1565119/3115000, -455757/1557500, 1947/89000] := by simp only [in767_witness_dual_vec, show (2 : Fin 4) ≠ 0 from by decide, show (2 : Fin 4) ≠ 1 from by decide, if_false, if_true]private lemma in767_witness_dual_vec3 : in767_witness_dual_vec 3 = ![0, 0, 77/2000] := by simp only [in767_witness_dual_vec, show (3 : Fin 4) ≠ 0 from by decide, show (3 : Fin 4) ≠ 1 from by decide, show (3 : Fin 4) ≠ 2 from by decide, if_false]private noncomputable def in767_witness_dual_p (i : Fin 4) : Matrix (Fin 3) (Fin 3) ℂ :=  Matrix.vecMulVec (in767_witness_dual_vec i) (star (in767_witness_dual_vec i))private noncomputable def in767_witness_dual_z (s : Fin 3 → Bool) : Matrix (Fin 3) (Fin 3) ℂ :=  (2 : ℝ) • in767_witness_dual_p 3 + (if s 0 then in767_witness_dual_p 1 - in767_witness_dual_p 3 else 0) +  (if s 1 then in767_witness_dual_p 0 - in767_witness_dual_p 3 else 0) +  (if s 2 then in767_witness_dual_p 2 - in767_witness_dual_p 3 else 0)private lemma in767_witness_dual_p_psd (i : Fin 4) : (in767_witness_dual_p i).PosSemidef :=  Matrix.posSemidef_vecMulVec_self_star _private lemma in767_witness_dual_all_psd (s : Fin 3 → Bool) : (in767_witness_dual_z s).PosSemidef := by  cases h0 : s 0 <;> cases h1 : s 1 <;> cases h2 : s 2  · have he : in767_witness_dual_z s = in767_witness_dual_p 3 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 3).add (in767_witness_dual_p_psd 3)  · have he : in767_witness_dual_z s = in767_witness_dual_p 2 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 2).add (in767_witness_dual_p_psd 3)  · have he : in767_witness_dual_z s = in767_witness_dual_p 0 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 0).add (in767_witness_dual_p_psd 3)  · have he : in767_witness_dual_z s = in767_witness_dual_p 0 + in767_witness_dual_p 2 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 0).add (in767_witness_dual_p_psd 2)  · have he : in767_witness_dual_z s = in767_witness_dual_p 1 + in767_witness_dual_p 3 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 1).add (in767_witness_dual_p_psd 3)  · have he : in767_witness_dual_z s = in767_witness_dual_p 1 + in767_witness_dual_p 2 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 1).add (in767_witness_dual_p_psd 2)  · have he : in767_witness_dual_z s = in767_witness_dual_p 0 + in767_witness_dual_p 1 := by simp [in767_witness_dual_z, h0, h1, h2, two_smul] <;> abel    rw [he]; exact (in767_witness_dual_p_psd 0).add (in767_witness_dual_p_psd 1)  · let v : Fin 3 → ℂ := ![3533/5000, 0, 0]    let w : Fin 3 → ℂ := ![0, 3533/5000, 0]    have he : in767_witness_dual_z s = Matrix.vecMulVec v (star v) + Matrix.vecMulVec w (star w) := by      simp only [in767_witness_dual_z, h0, h1, h2, if_true, in767_witness_dual_p,        in767_witness_dual_vec0, in767_witness_dual_vec1, in767_witness_dual_vec2, in767_witness_dual_vec3, two_smul]      ext i j      fin_cases i <;> fin_cases j <;>        simp only [Matrix.add_apply, Matrix.sub_apply, Matrix.vecMulVec_apply, Pi.star_apply] <;>        norm_num [v, w, map_div₀, map_ofNat]    rw [he]    exact (Matrix.posSemidef_vecMulVec_self_star v).add (Matrix.posSemidef_vecMulVec_self_star w)private lemma in767_witness_rational_dual_number :    (2 * (in767_witness_dual_p 3).trace.re +      ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * in767_witness_effect0).trace.re +      ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * in767_witness_effect1).trace.re +      ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * in767_witness_effect2).trace.re) < 0 := by  simp only [in767_witness_dual_p, in767_witness_dual_vec0, in767_witness_dual_vec1, in767_witness_dual_vec2, in767_witness_dual_vec3]  simp only [Matrix.trace, Matrix.diag_apply, Matrix.mul_apply, Matrix.sub_apply, Matrix.vecMulVec_apply, Pi.star_apply, Fin.sum_univ_succ]  norm_num [in767_witness_effect0, in767_witness_effect1, in767_witness_effect2, Matrix.cons_val_two, Matrix.vecHead, Matrix.vecTail, Matrix.mul_apply, Matrix.sub_apply, Matrix.vecMulVec_apply, Pi.star_apply, Fin.sum_univ_succ, map_div₀, map_ofNat]private lemma in767_witness_trace_matrix_sum (A : Matrix (Fin 3) (Fin 3) ℂ)    (G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ) :    (A * ∑ s, G s).trace.re = ∑ s, (A * G s).trace.re := by  simp only [Matrix.mul_sum, Matrix.trace_sum, Complex.re_sum]private theorem in767_witness_rational_not_joint :    ¬ ∃ G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ,      (∀ s, (G s).PosSemidef) ∧ (∑ s, G s = 1) ∧      ∀ j : Fin 3, (∑ s, if s j then G s else 0) = in767_witness_rational_effect j := by  classical  rintro ⟨G, hG, hsum, hmarg⟩  have hmargin (A : Matrix (Fin 3) (Fin 3) ℂ) (j : Fin 3) :      (∑ s, if s j then (A * G s).trace.re else 0) = (A * in767_witness_rational_effect j).trace.re := by    rw [← hmarg j, in767_witness_trace_matrix_sum]    apply Finset.sum_congr rfl    intro s _    split_ifs <;> simp_all  have hsplit (s : Fin 3 → Bool) :      (in767_witness_dual_z s * G s).trace.re =        2 * (in767_witness_dual_p 3 * G s).trace.re +        (if s 0 then ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * G s).trace.re else 0) +        (if s 1 then ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * G s).trace.re else 0) +        (if s 2 then ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * G s).trace.re else 0) := by    cases h0 : s 0 <;> cases h1 : s 1 <;> cases h2 : s 2 <;>      simp [in767_witness_dual_z, h0, h1, h2, two_smul, Matrix.add_mul, Matrix.sub_mul, Matrix.trace_add, Matrix.trace_sub] <;> ring  have heq : (∑ s, (in767_witness_dual_z s * G s).trace.re) =      2 * (in767_witness_dual_p 3).trace.re +        ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * in767_witness_effect0).trace.re +        ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * in767_witness_effect1).trace.re +        ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * in767_witness_effect2).trace.re := by    calc      _ = ∑ s, (2 * (in767_witness_dual_p 3 * G s).trace.re +          (if s 0 then ((in767_witness_dual_p 1 - in767_witness_dual_p 3) * G s).trace.re else 0) +          (if s 1 then ((in767_witness_dual_p 0 - in767_witness_dual_p 3) * G s).trace.re else 0) +          (if s 2 then ((in767_witness_dual_p 2 - in767_witness_dual_p 3) * G s).trace.re else 0)) := by        exact Finset.sum_congr rfl fun s _ => hsplit s      _ = _ := by        simp only [Finset.sum_add_distrib, ← Finset.mul_sum]        rw [← in767_witness_trace_matrix_sum, hsum, Matrix.mul_one]        rw [hmargin _ 0, hmargin _ 1, hmargin _ 2]        simp [in767_witness_rational_effect]  have hnonneg : 0 ≤ ∑ s, (in767_witness_dual_z s * G s).trace.re :=    Finset.sum_nonneg fun s _ => in767_witness_trace_psd (in767_witness_dual_all_psd s) (hG s)  rw [heq] at hnonneg  have := in767_witness_rational_dual_number  linariththeorem ppt_qutrit_rational_binary_effect_incompatible :  ¬ ∃ G : (Fin 3 → Bool) → Matrix (Fin 3) (Fin 3) ℂ,    (∀ s, (G s).PosSemidef) ∧ (∑ s, G s = 1) ∧    ∀ j : Fin 3, (∑ s, if s j then G s else 0) = ppt_qutrit_rational_binary_effect j := by  exact in767_witness_rational_not_joint
See it in the graphDiscuss this resultChallenge it