AFTD Auto-Formalizing Theoretical Domains

ghz_closed_network_lambda_le_two

theoremverifiedformalization

For every parameter alpha in R, all orthonormal frames e, f in O(3), and all setting indices mu, nu, mu', nu' in {0, 1, 2, 3}, the two-copy closed-network Bell measurement quantity Lambda(tau_{mu nu}, tau_{mu' nu'}) for the generalized GHZ state is at most 2.

Statement

theorem ghz_closed_network_lambda_le_two (α : ℝ) (e f : Matrix (Fin 3) (Fin 3) ℝ) (he : e ∈ Matrix.orthogonalGroup (Fin 3) ℝ) (hf : f ∈ Matrix.orthogonalGroup (Fin 3) ℝ) (μ ν μ' ν' : Fin 4) :
    bell_network_lambda (ghz_correlation_vector α e f μ ν) (ghz_correlation_vector α e f μ' ν') ≤ 2
Formalizes
arXiv:2610.35838 Appendix B
Verified
02 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 4
  • bell_network_lambdaFor vectors u, v in R^3, Lambda(u, v) is defined as (1/2) * (s + sqrt(s^2 - 4 * (u . v)^2)) where s = |u|^2 + |v|^2.
  • ghz_correlation_vectorFor parameter alpha in R, orthonormal frames e, f of R^3, and indices mu, nu in {0, 1, 2, 3}, the correlation vector tau_{mu nu} has…
  • ghz_pauli_coeffFor an orthonormal frame e of R^3 and index mu in {0, 1, 2, 3}, the Pauli coefficients of A_mu are given by A_0 = I (coefficient 1 at…
  • ghz_standard_frame_lambda_leFor the generalized GHZ state with parameter alpha and standard identity frames e = 1 and f = 1, the two-copy closed-network quantity…

Read back from the Lean

For all real numbers α, all 3 × 3 real orthogonal matrices e and f, and all indices μ, ν, μ', ν' ∈ Fin 4, the Bell network lambda quantity for the two GHZ correlation vectors u = ghz_correlation_vector α e f μ ν and v = ghz_correlation_vector α e f μ' ν' is at most 2.

Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.

The proof, explained

The theorem establishes that the two-copy closed-network Bell measurement quantity Λ(u, v) for the generalized GHZ state is bounded by 2 for all state parameters α, orthonormal frames e, f in O(3), and measurement setting quadruples. The proof uses the general upper bound Λ(u, v) ≤ u ⬝ᵥ u + v ⬝ᵥ v, reducing the problem to showing that every single-setting correlation vector has squared Euclidean norm at most 1. This norm bound is established by case analysis on the Pauli coefficient vectors: when either setting is index 0, the correlation vector collapses to a single component bounded by 1; when both settings are non-zero, the squared norm is recognized as a convex combination of two terms bounded by 1, completing the proof.

  1. The sum of squares in the norm inequality simplifies algebraically to s^2 * (x_1^2 + x_2^2) * (y_1^2 + y_2^2) + c^2 * x_3^2 * y_3^2.

    Expanding the squares in [s(x_1 y_1 - x_2 y_2)]^2 + [-s(x_1 y_2 + x_2 y_1)]^2 + (c x_3 y_3)^2 causes cross-terms to cancel, leaving s^2 (x_1^2 + x_2^2)(y_1^2 + y_2^2) + c^2 x_3^2 y_3^2, which the ring tactic verifies. Rewriting with this equality simplifies the goal.

    lines 8–12

      have h_alg : (s * (x1 * y1 - x2 * y2)) ^ 2 +
          (-s * (x1 * y2 + x2 * y1)) ^ 2 +
          (c * x3 * y3) ^ 2 =
          s ^ 2 * (x1 ^ 2 + x2 ^ 2) * (y1 ^ 2 + y2 ^ 2) + c ^ 2 * x3 ^ 2 * y3 ^ 2 := by ring
      rw [h_alg]
  2. Defining a = x_1^2 + x_2^2 and b = y_1^2 + y_2^2, we have 0 ≤ a ≤ 1, 0 ≤ b ≤ 1, x_3^2 = 1 - a, and y_3^2 = 1 - b.

    We introduce variables a = x_1^2 + x_2^2 and b = y_1^2 + y_2^2. Because squares in ℝ are nonnegative (`sq_nonneg`), a and b are nonnegative sums of squares. From the unit-sphere equations a + x_3^2 = 1 and b + y_3^2 = 1 with x_3^2 ≥ 0 and y_3^2 ≥ 0, linear arithmetic establishes that a ≤ 1, b ≤ 1, x_3^2 = 1 - a, and y_3^2 = 1 - b. Substituting these expressions into the goal eliminates x_3 and y_3.

    lines 13–29

    uses sq_nonneg

      set a := x1 ^ 2 + x2 ^ 2
      set b := y1 ^ 2 + y2 ^ 2
      have hS0 : 0 ≤ s ^ 2 := sq_nonneg s
      have hC0 : 0 ≤ c ^ 2 := sq_nonneg c
      have hx1 : 0 ≤ x1 ^ 2 := sq_nonneg x1
      have hx2 : 0 ≤ x2 ^ 2 := sq_nonneg x2
      have hx3 : 0 ≤ x3 ^ 2 := sq_nonneg x3
      have hy1 : 0 ≤ y1 ^ 2 := sq_nonneg y1
      have hy2 : 0 ≤ y2 ^ 2 := sq_nonneg y2
      have hy3 : 0 ≤ y3 ^ 2 := sq_nonneg y3
      have ha0 : 0 ≤ a := by positivity
      have hb0 : 0 ≤ b := by positivity
      have ha1 : a ≤ 1 := by linarith
      have hb1 : b ≤ 1 := by linarith
      have hx3_eq : x3 ^ 2 = 1 - a := by linarith
      have hy3_eq : y3 ^ 2 = 1 - b := by linarith
      rw [hx3_eq, hy3_eq]
  3. The expression s^2 * a * b + c^2 * (1 - a) * (1 - b) is at most 1.

    Since 0 ≤ a ≤ 1 and 0 ≤ b ≤ 1, their products satisfy a * b ≤ 1 and (1 - a) * (1 - b) ≤ 1. The target expression is a convex combination of these two terms with coefficients s^2 and c^2 satisfying s^2 + c^2 = 1 and s^2, c^2 ≥ 0, which nonlinear arithmetic (`nlinarith`) closes to bound the sum by 1.

    lines 30–32

      have hab1 : a * b ≤ 1 := by nlinarith
      have hab2 : (1 - a) * (1 - b) ≤ 1 := by nlinarith
      nlinarith
  4. Every row of an orthogonal matrix M in O(3) is a unit vector: ∑_{j=0}^2 (M_{i,j})^2 = 1 for each row index i.

    An orthogonal matrix satisfies M * M^T = 1 (`Matrix.mem_orthogonalGroup_iff`). Applying functional equality to the (i, i) diagonal entry (`congr_fun`), expanding matrix multiplication and transposition (`Matrix.mul_apply`, `Matrix.transpose_apply`), and evaluating the identity matrix diagonal (`Matrix.one_apply_eq`) over the 3 components (`Fin.sum_univ_three`) shows that the sum of squared entries in row i is 1.

    lines 39–44

    uses Matrix.mem_orthogonalGroup_iff, congr_fun, Matrix.mul_apply, Matrix.transpose_apply, Matrix.one_apply_eq, Fin.sum_univ_three

      have row_norm : ∀ i : Fin 3, (M i 0)^2 + (M i 1)^2 + (M i 2)^2 = 1 := by
        intro i
        rw [Matrix.mem_orthogonalGroup_iff] at hM
        have h := congr_fun (congr_fun hM i) i
        simp only [Matrix.mul_apply, Matrix.transpose_apply, Matrix.one_apply_eq, Fin.sum_univ_three] at h
        linarith [h]
  5. For any setting index k in {0, 1, 2, 3}, the Pauli coefficient vector of M is either (1, 0, 0, 0) (for k = 0) or has 0-th component 0 with components 1, 2, 3 having squared norm 1 (for k in {1, 2, 3}).

    Case exhaustion on k in Fin 4 (`fin_cases`) separates index 0 from indices 1, 2, 3. For k = 0, the definition of `ghz_pauli_coeff` produces (1, 0, 0, 0) trivially. For k in {1, 2, 3}, the 0-th component is 0 by definition, while the components at indices 1, 2, 3 correspond to row k - 1 of M, which is a unit vector by the row-norm property proved in the previous step.

    lines 45–49

      fin_cases k
      · left; dsimp [ghz_pauli_coeff]; exact ⟨rfl, rfl, rfl, rfl⟩
      · right; dsimp [ghz_pauli_coeff]; exact ⟨rfl, row_norm 0⟩
      · right; dsimp [ghz_pauli_coeff]; exact ⟨rfl, row_norm 1⟩
      · right; dsimp [ghz_pauli_coeff]; exact ⟨rfl, row_norm 2⟩
  6. The squared norm u ⬝ᵥ u of the GHZ correlation vector is expressed explicitly in terms of the Pauli coefficients of frames e and f and the parameter alpha.

    Unfolding the definition of the correlation vector (`ghz_correlation_vector`) and expanding the dot product on ℝ^3 (`dotProduct`, `Fin.sum_univ_three`, `sq`) expresses u ⬝ᵥ u as the sum of three squared coordinates: two containing sine factors and one containing cosine and mixed Pauli coefficient products.

    lines 55–64

    uses Fin.sum_univ_three, sq

      have h_dot : ghz_correlation_vector α e f μ ν ⬝ᵥ ghz_correlation_vector α e f μ ν =
          (Real.sin (2 * α) * (ghz_pauli_coeff e μ 1 * ghz_pauli_coeff f ν 1 - ghz_pauli_coeff e μ 2 * ghz_pauli_coeff f ν 2)) ^ 2 +
          (- Real.sin (2 * α) * (ghz_pauli_coeff e μ 1 * ghz_pauli_coeff f ν 2 + ghz_pauli_coeff e μ 2 * ghz_pauli_coeff f ν 1)) ^ 2 +
          (Real.cos (2 * α) * ghz_pauli_coeff e μ 0 * ghz_pauli_coeff f ν 0 +
           ghz_pauli_coeff e μ 0 * ghz_pauli_coeff f ν 3 +
           ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 0 +
           Real.cos (2 * α) * ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 3) ^ 2 := by
        simp only [dotProduct, Fin.sum_univ_three, sq]
        dsimp [ghz_correlation_vector]
      rw [h_dot]
  7. When the first setting index mu is 0, the correlation vector satisfies u ⬝ᵥ u ≤ 1.

    We apply `ghz_closed_network_lambda_le_two_pauli_coeff_prop` to frame e and setting mu. When mu = 0, the Pauli vector for e is (1, 0, 0, 0). Branching on setting nu for frame f: if nu = 0, f also yields (1, 0, 0, 0), so u = (0, 0, cos(2α)) and u ⬝ᵥ u = cos(2α)^2 ≤ 1 (`Real.cos_sq_le_one`). If nu ≠ 0, f has 0-th coefficient 0 and unit Euclidean norm on the rest; u simplifies to (0, 0, f_{nu, 3}), whose squared norm is at most 1 because the other components squared are nonnegative (`sq_nonneg`).

    lines 65–73

    uses ghz_closed_network_lambda_le_two_pauli_coeff_prop, Real.cos_sq_le_one, sq_nonneg

      rcases ghz_closed_network_lambda_le_two_pauli_coeff_prop e he μ with ⟨ae0, ae1, ae2, ae3⟩ | ⟨ae0, he_sum⟩
      · rcases ghz_closed_network_lambda_le_two_pauli_coeff_prop f hf ν with ⟨cf0, cf1, cf2, cf3⟩ | ⟨cf0, hf_sum⟩
        · rw [ae0, ae1, ae2, ae3, cf0, cf1, cf2, cf3]
          have hcos : (Real.cos (2 * α)) ^ 2 ≤ 1 := Real.cos_sq_le_one (2 * α)
          nlinarith
        · rw [ae0, ae1, ae2, ae3, cf0]
          have h1 : 0 ≤ (ghz_pauli_coeff f ν 1) ^ 2 := sq_nonneg _
          have h2 : 0 ≤ (ghz_pauli_coeff f ν 2) ^ 2 := sq_nonneg _
          nlinarith
  8. When mu ≠ 0 and nu = 0, the correlation vector satisfies u ⬝ᵥ u ≤ 1.

    When mu ≠ 0, frame e has 0-th component 0 and unit norm on the remaining three components. When nu = 0, frame f yields (1, 0, 0, 0). Substituting these into the formula simplifies the correlation vector to (0, 0, e_{mu, 3}), whose squared norm is at most 1 by nonnegativity of the other components squared (`sq_nonneg`) and nonlinear arithmetic (`nlinarith`).

    lines 74–78

    uses ghz_closed_network_lambda_le_two_pauli_coeff_prop, sq_nonneg

      · rcases ghz_closed_network_lambda_le_two_pauli_coeff_prop f hf ν with ⟨cf0, cf1, cf2, cf3⟩ | ⟨cf0, hf_sum⟩
        · rw [ae0, cf0, cf1, cf2, cf3]
          have h1 : 0 ≤ (ghz_pauli_coeff e μ 1) ^ 2 := sq_nonneg _
          have h2 : 0 ≤ (ghz_pauli_coeff e μ 2) ^ 2 := sq_nonneg _
          nlinarith
  9. When both mu ≠ 0 and nu ≠ 0, the correlation vector satisfies u ⬝ᵥ u ≤ 1.

    When neither setting is 0, both Pauli vectors have 0-th component equal to 0, and their remaining three components form unit vectors in ℝ^3. The third component of u simplifies via polynomial arithmetic (`ring`) to cos(2α) * e_{mu, 3} * f_{nu, 3}. Since sin(2α)^2 + cos(2α)^2 = 1 (`Real.sin_sq_add_cos_sq`), this exactly matches the algebraic bound `ghz_closed_network_lambda_le_two_norm_sq_bound`, establishing u ⬝ᵥ u ≤ 1.

    lines 79–87

    uses ghz_closed_network_lambda_le_two_norm_sq_bound, Real.sin_sq_add_cos_sq

        · rw [ae0, cf0]
          have h_simp : (Real.cos (2 * α) * 0 * 0 +
              0 * ghz_pauli_coeff f ν 3 +
              ghz_pauli_coeff e μ 3 * 0 +
              Real.cos (2 * α) * ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 3) =
              Real.cos (2 * α) * ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 3 := by ring
          rw [h_simp]
          exact ghz_closed_network_lambda_le_two_norm_sq_bound (Real.sin (2 * α)) (Real.cos (2 * α)) _ _ _ _ _ _
            (Real.sin_sq_add_cos_sq (2 * α)) he_sum hf_sum
  10. Both correlation vectors u and v have squared norm at most 1.

    Applying the previously established lemma `ghz_closed_network_lambda_le_two_norm_sq_le_one` to the setting pairs (mu, nu) and (mu', nu') shows that both correlation vectors u and v satisfy u ⬝ᵥ u ≤ 1 and v ⬝ᵥ v ≤ 1.

    lines 92–95

    uses ghz_closed_network_lambda_le_two_norm_sq_le_one

      have hu : ghz_correlation_vector α e f μ ν ⬝ᵥ ghz_correlation_vector α e f μ ν ≤ 1 :=
        ghz_closed_network_lambda_le_two_norm_sq_le_one α e f he hf μ ν
      have hv : ghz_correlation_vector α e f μ' ν' ⬝ᵥ ghz_correlation_vector α e f μ' ν' ≤ 1 :=
        ghz_closed_network_lambda_le_two_norm_sq_le_one α e f he hf μ' ν'
  11. The Bell network measurement quantity Lambda(u, v) is at most 2.

    The lemma `ghz_standard_frame_lambda_le_lambda_le_s` gives Lambda(u, v) ≤ u ⬝ᵥ u + v ⬝ᵥ v. Combining this with the bounds u ⬝ᵥ u ≤ 1 and v ⬝ᵥ v ≤ 1 via linear arithmetic (`linarith`) completes the proof that Lambda(u, v) ≤ 1 + 1 = 2.

    lines 96–98

    uses ghz_standard_frame_lambda_le_lambda_le_s

      have h_le := ghz_standard_frame_lambda_le_lambda_le_s
        (ghz_correlation_vector α e f μ ν) (ghz_correlation_vector α e f μ' ν')
      linarith
sq_nonneg
For any real number x, its square is nonnegative: 0 ≤ x^2.
Matrix.mem_orthogonalGroup_iff
A square matrix A is in the orthogonal group if and only if A * A^T = 1.
congr_fun
If two functions or operators f and g are equal, then f(x) = g(x) for every input x.
Matrix.mul_apply
The (i, k)-th entry of a matrix product M * N is the sum over j of M_{i,j} * N_{j,k}.
Matrix.transpose_apply
The (i, j)-th entry of the transpose M^T is M_{j, i}.
Matrix.one_apply_eq
The diagonal entries of the identity matrix are equal to 1, i.e., 1_{i,i} = 1.
Fin.sum_univ_three
A sum of a function f over Fin 3 equals f(0) + f(1) + f(2).
sq
The square of an element a is equal to a * a.
Real.cos_sq_le_one
For any real number x, cos(x)^2 ≤ 1.
ghz_closed_network_lambda_le_two_pauli_coeff_prop
For any orthogonal matrix M in O(3) and index k in Fin 4, the Pauli coefficient vector is either (1, 0, 0, 0) when k = 0, or has 0-th component 0 and sum of squares of components 1, 2, 3 equal to 1 when k is 1, 2, or 3.
ghz_closed_network_lambda_le_two_norm_sq_bound
For real numbers s, c, x, y with s^2 + c^2 = 1, x_1^2 + x_2^2 + x_3^2 = 1, and y_1^2 + y_2^2 + y_3^2 = 1, the sum (s * (x_1 * y_1 - x_2 * y_2))^2 + (-s * (x_1 * y_2 + x_2 * y_1))^2 + (c * x_3 * y_3)^2 is at most 1.
Real.sin_sq_add_cos_sq
For any real number x, sin(x)^2 + cos(x)^2 = 1.
ghz_closed_network_lambda_le_two_norm_sq_le_one
For any parameter alpha, orthogonal frames e, f in O(3), and setting indices mu, nu, the correlation vector has dot product with itself at most 1.
ghz_standard_frame_lambda_le_lambda_le_s
For any 3-dimensional vectors u and v, the Bell network quantity Lambda(u, v) is at most u ⬝ᵥ u + v ⬝ᵥ v.

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

lemma ghz_closed_network_lambda_le_two_norm_sq_bound (s c x1 x2 x3 y1 y2 y3 : ℝ)    (hsc : s ^ 2 + c ^ 2 = 1)    (hx : x1 ^ 2 + x2 ^ 2 + x3 ^ 2 = 1)    (hy : y1 ^ 2 + y2 ^ 2 + y3 ^ 2 = 1) :    (s * (x1 * y1 - x2 * y2)) ^ 2 +    (-s * (x1 * y2 + x2 * y1)) ^ 2 +    (c * x3 * y3) ^ 2 ≤ 1 := by  have h_alg : (s * (x1 * y1 - x2 * y2)) ^ 2 +      (-s * (x1 * y2 + x2 * y1)) ^ 2 +      (c * x3 * y3) ^ 2 =      s ^ 2 * (x1 ^ 2 + x2 ^ 2) * (y1 ^ 2 + y2 ^ 2) + c ^ 2 * x3 ^ 2 * y3 ^ 2 := by ring  rw [h_alg]  set a := x1 ^ 2 + x2 ^ 2  set b := y1 ^ 2 + y2 ^ 2  have hS0 : 0 ≤ s ^ 2 := sq_nonneg s  have hC0 : 0 ≤ c ^ 2 := sq_nonneg c  have hx1 : 0 ≤ x1 ^ 2 := sq_nonneg x1  have hx2 : 0 ≤ x2 ^ 2 := sq_nonneg x2  have hx3 : 0 ≤ x3 ^ 2 := sq_nonneg x3  have hy1 : 0 ≤ y1 ^ 2 := sq_nonneg y1  have hy2 : 0 ≤ y2 ^ 2 := sq_nonneg y2  have hy3 : 0 ≤ y3 ^ 2 := sq_nonneg y3  have ha0 : 0 ≤ a := by positivity  have hb0 : 0 ≤ b := by positivity  have ha1 : a ≤ 1 := by linarith  have hb1 : b ≤ 1 := by linarith  have hx3_eq : x3 ^ 2 = 1 - a := by linarith  have hy3_eq : y3 ^ 2 = 1 - b := by linarith  rw [hx3_eq, hy3_eq]  have hab1 : a * b ≤ 1 := by nlinarith  have hab2 : (1 - a) * (1 - b) ≤ 1 := by nlinarith  nlinarithlemma ghz_closed_network_lambda_le_two_pauli_coeff_prop (M : Matrix (Fin 3) (Fin 3) ℝ)    (hM : M ∈ Matrix.orthogonalGroup (Fin 3) ℝ) (k : Fin 4) :    (ghz_pauli_coeff M k 0 = 1 ∧ ghz_pauli_coeff M k 1 = 0 ∧ ghz_pauli_coeff M k 2 = 0 ∧ ghz_pauli_coeff M k 3 = 0) ∨    (ghz_pauli_coeff M k 0 = 0 ∧     (ghz_pauli_coeff M k 1)^2 + (ghz_pauli_coeff M k 2)^2 + (ghz_pauli_coeff M k 3)^2 = 1) := by  have row_norm : ∀ i : Fin 3, (M i 0)^2 + (M i 1)^2 + (M i 2)^2 = 1 := by    intro i    rw [Matrix.mem_orthogonalGroup_iff] at hM    have h := congr_fun (congr_fun hM i) i    simp only [Matrix.mul_apply, Matrix.transpose_apply, Matrix.one_apply_eq, Fin.sum_univ_three] at h    linarith [h]  fin_cases k  · left; dsimp [ghz_pauli_coeff]; exact ⟨rfl, rfl, rfl, rfl⟩  · right; dsimp [ghz_pauli_coeff]; exact ⟨rfl, row_norm 0⟩  · right; dsimp [ghz_pauli_coeff]; exact ⟨rfl, row_norm 1⟩  · right; dsimp [ghz_pauli_coeff]; exact ⟨rfl, row_norm 2⟩lemma ghz_closed_network_lambda_le_two_norm_sq_le_one (α : ℝ) (e f : Matrix (Fin 3) (Fin 3) ℝ)    (he : e ∈ Matrix.orthogonalGroup (Fin 3) ℝ) (hf : f ∈ Matrix.orthogonalGroup (Fin 3) ℝ)    (μ ν : Fin 4) :    ghz_correlation_vector α e f μ ν ⬝ᵥ ghz_correlation_vector α e f μ ν ≤ 1 := by  have h_dot : ghz_correlation_vector α e f μ ν ⬝ᵥ ghz_correlation_vector α e f μ ν =      (Real.sin (2 * α) * (ghz_pauli_coeff e μ 1 * ghz_pauli_coeff f ν 1 - ghz_pauli_coeff e μ 2 * ghz_pauli_coeff f ν 2)) ^ 2 +      (- Real.sin (2 * α) * (ghz_pauli_coeff e μ 1 * ghz_pauli_coeff f ν 2 + ghz_pauli_coeff e μ 2 * ghz_pauli_coeff f ν 1)) ^ 2 +      (Real.cos (2 * α) * ghz_pauli_coeff e μ 0 * ghz_pauli_coeff f ν 0 +       ghz_pauli_coeff e μ 0 * ghz_pauli_coeff f ν 3 +       ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 0 +       Real.cos (2 * α) * ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 3) ^ 2 := by    simp only [dotProduct, Fin.sum_univ_three, sq]    dsimp [ghz_correlation_vector]  rw [h_dot]  rcases ghz_closed_network_lambda_le_two_pauli_coeff_prop e he μ with ⟨ae0, ae1, ae2, ae3⟩ | ⟨ae0, he_sum⟩  · rcases ghz_closed_network_lambda_le_two_pauli_coeff_prop f hf ν with ⟨cf0, cf1, cf2, cf3⟩ | ⟨cf0, hf_sum⟩    · rw [ae0, ae1, ae2, ae3, cf0, cf1, cf2, cf3]      have hcos : (Real.cos (2 * α)) ^ 2 ≤ 1 := Real.cos_sq_le_one (2 * α)      nlinarith    · rw [ae0, ae1, ae2, ae3, cf0]      have h1 : 0 ≤ (ghz_pauli_coeff f ν 1) ^ 2 := sq_nonneg _      have h2 : 0 ≤ (ghz_pauli_coeff f ν 2) ^ 2 := sq_nonneg _      nlinarith  · rcases ghz_closed_network_lambda_le_two_pauli_coeff_prop f hf ν with ⟨cf0, cf1, cf2, cf3⟩ | ⟨cf0, hf_sum⟩    · rw [ae0, cf0, cf1, cf2, cf3]      have h1 : 0 ≤ (ghz_pauli_coeff e μ 1) ^ 2 := sq_nonneg _      have h2 : 0 ≤ (ghz_pauli_coeff e μ 2) ^ 2 := sq_nonneg _      nlinarith    · rw [ae0, cf0]      have h_simp : (Real.cos (2 * α) * 0 * 0 +          0 * ghz_pauli_coeff f ν 3 +          ghz_pauli_coeff e μ 3 * 0 +          Real.cos (2 * α) * ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 3) =          Real.cos (2 * α) * ghz_pauli_coeff e μ 3 * ghz_pauli_coeff f ν 3 := by ring      rw [h_simp]      exact ghz_closed_network_lambda_le_two_norm_sq_bound (Real.sin (2 * α)) (Real.cos (2 * α)) _ _ _ _ _ _        (Real.sin_sq_add_cos_sq (2 * α)) he_sum hf_sum/-- Universal upper bound Lambda(tau_{mu nu}, tau_{mu' nu'}) <= 2 for all alpha, orthonormal frames, and setting quadruples. -/theorem ghz_closed_network_lambda_le_two (α : ℝ) (e f : Matrix (Fin 3) (Fin 3) ℝ) (he : e ∈ Matrix.orthogonalGroup (Fin 3) ℝ) (hf : f ∈ Matrix.orthogonalGroup (Fin 3) ℝ) (μ ν μ' ν' : Fin 4) :    bell_network_lambda (ghz_correlation_vector α e f μ ν) (ghz_correlation_vector α e f μ' ν') ≤ 2 := by  have hu : ghz_correlation_vector α e f μ ν ⬝ᵥ ghz_correlation_vector α e f μ ν ≤ 1 :=    ghz_closed_network_lambda_le_two_norm_sq_le_one α e f he hf μ ν  have hv : ghz_correlation_vector α e f μ' ν' ⬝ᵥ ghz_correlation_vector α e f μ' ν' ≤ 1 :=    ghz_closed_network_lambda_le_two_norm_sq_le_one α e f he hf μ' ν'  have h_le := ghz_standard_frame_lambda_le_lambda_le_s    (ghz_correlation_vector α e f μ ν) (ghz_correlation_vector α e f μ' ν')  linarith
See it in the graphDiscuss this resultChallenge it