AFTD Auto-Formalizing Theoretical Domains

ghz_standard_frame_lambda_le

theoremverifiedformalization

For the generalized GHZ state with parameter alpha and standard identity frames e = 1 and f = 1, the two-copy closed-network quantity Lambda(tau_{mu nu}, tau_{mu' nu'}) is at most 1 + sin^2(2*alpha) for all setting indices mu, nu, mu', nu' in {0, 1, 2, 3}.

Statement

theorem ghz_standard_frame_lambda_le (α : ℝ) (μ ν μ' ν' : Fin 4) :
    bell_network_lambda (ghz_correlation_vector α 1 1 μ ν) (ghz_correlation_vector α 1 1 μ' ν') ≤ 1 + Real.sin (2 * α) ^ 2
Formalizes
arXiv:2610.35838 Eq. 15
Verified
02 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 3
  • 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…
Used by 1
  • ghz_closed_network_lambda_le_twoFor 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…

Read back from the Lean

For all real numbers α and all indices μ, ν, μ', ν' ∈ Fin 4 (i.e., {0, 1, 2, 3}), the Bell network lambda value for the two GHZ correlation vectors ghz_correlation_vector(α, 1, 1, μ, ν) and ghz_correlation_vector(α, 1, 1, μ', ν') (evaluated with the 3×3 identity matrix as both frames) is less than or equal to 1 + sin²(2α).

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 proof bounds the two-copy network quantity Λ(u, v) by 1 + sin²(2α) for generalized GHZ correlation vectors u and v under identity measurement frames. It first establishes two geometric bounds on Λ(u, v): a general bound Λ(u, v) ≤ u ⬝ᵥ u + v ⬝ᵥ v, and a sharper bound Λ(u, v) ≤ max(u ⬝ᵥ u, v ⬝ᵥ v) when both vectors lie along the z-axis. Exhaustive case analysis over all 16 measurement settings shows that every correlation vector has norm squared at most 1, and either lies purely on the z-axis or has norm squared at most sin²(2α). The result follows by case distinction: if both vectors are on the z-axis, Λ(u, v) ≤ 1 ≤ 1 + sin²(2α); otherwise, at least one vector has norm squared bounded by sin²(2α), so the sum bound yields Λ(u, v) ≤ 1 + sin²(2α).

  1. The dot product of any vector u ∈ ℝ³ with itself is non-negative: 0 ≤ u ⬝ᵥ u.

    Rewriting the dot product as the sum ∑_{i=0}^2 u_i², each term u_i * u_i is non-negative by non-linear arithmetic, so the sum is non-negative by Finset.sum_nonneg.

    lines 2–5

    uses Finset.sum_nonneg

      change 0 ≤ ∑ i : Fin 3, u i * u i
      apply Finset.sum_nonneg
      intro i _
      nlinarith
  2. For any vectors u, v ∈ ℝ³, the network quantity satisfies Λ(u, v) ≤ u ⬝ᵥ u + v ⬝ᵥ v.

    Let s = u ⬝ᵥ u + v ⬝ᵥ v, which is non-negative since each self dot product is non-negative. The discriminant s² - 4(u ⬝ᵥ v)² is at most s², so its square root is at most √(s²) = s. Substituting this into the definition Λ(u, v) = ½(s + √(s² - 4(u ⬝ᵥ v)²)) yields Λ(u, v) ≤ ½(s + s) = s.

    lines 8–16

    uses ghz_standard_frame_lambda_le_dot_self_nonneg, add_nonneg, Real.sqrt_le_sqrt, Real.sqrt_sq

      dsimp [bell_network_lambda]
      set s := (u ⬝ᵥ u) + (v ⬝ᵥ v)
      have hs : 0 ≤ s := add_nonneg (ghz_standard_frame_lambda_le_dot_self_nonneg u) (ghz_standard_frame_lambda_le_dot_self_nonneg v)
      have hdisc_le : s ^ 2 - 4 * (u ⬝ᵥ v) ^ 2 ≤ s ^ 2 := by nlinarith
      have hsqrt_le : Real.sqrt (s ^ 2 - 4 * (u ⬝ᵥ v) ^ 2) ≤ s := by
        calc
          Real.sqrt (s ^ 2 - 4 * (u ⬝ᵥ v) ^ 2) ≤ Real.sqrt (s ^ 2) := Real.sqrt_le_sqrt hdisc_le
          _ = s := Real.sqrt_sq hs
      linarith
  3. If u, v ∈ ℝ³ have vanishing first two coordinates, their dot product is u₂ * v₂.

    Expanding the dot product over the three coordinates and substituting the hypotheses u₀ = u₁ = v₀ = v₁ = 0, the first two products vanish and only u₂ * v₂ remains.

    lines 20–22

    uses Fin.sum_univ_three

      simp only [dotProduct, Fin.sum_univ_three]
      rw [hu0, hu1, hv0, hv1]
      ring
  4. For z-axis vectors u and v, Λ(u, v) simplifies to ½(a + b + |a - b|), where a = u₂² and b = v₂².

    Applying the z-axis dot product formula gives u ⬝ᵥ u = a, v ⬝ᵥ v = b, and u ⬝ᵥ v = u₂ * v₂. The expression under the square root simplifies algebraically to (a + b)² - 4(u₂ * v₂)² = (a - b)², and its square root is |a - b|.

    lines 26–40

    uses ghz_standard_frame_lambda_le_z_axis_dot, Real.sqrt_sq_eq_abs

      have hu : u ⬝ᵥ u = (u 2) ^ 2 := by
        have := ghz_standard_frame_lambda_le_z_axis_dot u u hu0 hu1 hu0 hu1
        linarith [this]
      have hv : v ⬝ᵥ v = (v 2) ^ 2 := by
        have := ghz_standard_frame_lambda_le_z_axis_dot v v hv0 hv1 hv0 hv1
        linarith [this]
      have huv : u ⬝ᵥ v = u 2 * v 2 := ghz_standard_frame_lambda_le_z_axis_dot u v hu0 hu1 hv0 hv1
      dsimp [bell_network_lambda]
      rw [hu, hv, huv]
      set a := (u 2) ^ 2
      set b := (v 2) ^ 2
      have hdisc : (a + b) ^ 2 - 4 * (u 2 * v 2) ^ 2 = (a - b) ^ 2 := by
        dsimp [a, b]
        ring
      rw [hdisc, Real.sqrt_sq_eq_abs]
  5. For z-axis vectors u and v, Λ(u, v) ≤ max(u ⬝ᵥ u, v ⬝ᵥ v).

    By totality of order on ℝ, either a ≤ b or b ≤ a. If a ≤ b, then |a - b| = b - a, so ½(a + b + (b - a)) = b ≤ max(a, b). If b ≤ a, then |a - b| = a - b, so ½(a + b + (a - b)) = a ≤ max(a, b).

    lines 41–53

    uses le_total, abs_sub_comm, abs_of_nonneg, sub_nonneg.mpr, le_max_right, le_max_left

      cases le_total a b with
      | inl h =>
        have habs : |a - b| = b - a := abs_sub_comm a b ▸ abs_of_nonneg (sub_nonneg.mpr h)
        rw [habs]
        have : (1 / 2 : ℝ) * (a + b + (b - a)) = b := by ring
        rw [this]
        exact le_max_right a b
      | inr h =>
        have habs : |a - b| = a - b := abs_of_nonneg (sub_nonneg.mpr h)
        rw [habs]
        have : (1 / 2 : ℝ) * (a + b + (a - b)) = a := by ring
        rw [this]
        exact le_max_left a b
  6. For all setting pairs (μ, ν), the correlation vector satisfies u ⬝ᵥ u ≤ 1.

    Expanding the correlation vector with identity frames and performing an exhaustive case split over all 16 setting pairs (μ, ν) ∈ {0, 1, 2, 3}², each case evaluates to an expression bounded by 1 via the trigonometric identity sin²(2α) + cos²(2α) = 1.

    lines 57–62

    uses Fin.sum_univ_three, Matrix.one_apply, Real.sin_sq_add_cos_sq

      simp only [dotProduct, Fin.sum_univ_three]
      dsimp [ghz_correlation_vector, ghz_pauli_coeff]
      fin_cases μ <;> fin_cases ν <;> (
        simp [Matrix.one_apply]
        try nlinarith [Real.sin_sq_add_cos_sq (2 * α)]
      )
  7. For every setting pair (μ, ν), the correlation vector either satisfies u₀ = u₁ = 0 or u ⬝ᵥ u ≤ sin²(2α).

    Exhaustive case analysis across all 16 setting pairs shows that whenever the settings do not both measure in the equatorial plane, the x and y components vanish; in the remaining equatorial cases, the length squared is bounded by sin²(2α).

    lines 67–72

    uses Fin.sum_univ_three, Matrix.one_apply, sq

      dsimp [ghz_correlation_vector, ghz_pauli_coeff]
      simp only [dotProduct, Fin.sum_univ_three]
      fin_cases μ <;> fin_cases ν <;> (
        simp [Matrix.one_apply, sq]
        try nlinarith
      )
  8. Both correlation vectors u = τ_{μ ν} and v = τ_{μ' ν'} satisfy u ⬝ᵥ u ≤ 1 and v ⬝ᵥ v ≤ 1.

    We abbreviate the two correlation vectors as u and v, and invoke ghz_standard_frame_lambda_le_u_dot_u_le_one for each.

    lines 77–80

    uses ghz_standard_frame_lambda_le_u_dot_u_le_one

      set u := ghz_correlation_vector α 1 1 μ ν
      set v := ghz_correlation_vector α 1 1 μ' ν'
      have hu_le_1 : u ⬝ᵥ u ≤ 1 := ghz_standard_frame_lambda_le_u_dot_u_le_one α μ ν
      have hv_le_1 : v ⬝ᵥ v ≤ 1 := ghz_standard_frame_lambda_le_u_dot_u_le_one α μ' ν'
  9. When both u and v lie purely along the z-axis, Λ(u, v) ≤ 1 + sin²(2α).

    Using ghz_standard_frame_lambda_le_cases, when both u₀ = u₁ = 0 and v₀ = v₁ = 0, the z-axis lemma gives Λ(u, v) ≤ max(u ⬝ᵥ u, v ⬝ᵥ v) ≤ 1. Because sin²(2α) ≥ 0, this is at most 1 + sin²(2α).

    lines 81–86

    uses ghz_standard_frame_lambda_le_cases, ghz_standard_frame_lambda_le_lambda_z_axis, max_le, sq_nonneg

      rcases ghz_standard_frame_lambda_le_cases α μ ν with ⟨hu0, hu1⟩ | hu_sin
      · rcases ghz_standard_frame_lambda_le_cases α μ' ν' with ⟨hv0, hv1⟩ | hv_sin
        · have h_z := ghz_standard_frame_lambda_le_lambda_z_axis u v hu0 hu1 hv0 hv1
          have hmax : max (u ⬝ᵥ u) (v ⬝ᵥ v) ≤ 1 := max_le hu_le_1 hv_le_1
          have hsin_ge : 0 ≤ Real.sin (2 * α) ^ 2 := sq_nonneg _
          linarith
  10. When u is along the z-axis but v satisfies v ⬝ᵥ v ≤ sin²(2α), Λ(u, v) ≤ 1 + sin²(2α).

    The general bound Λ(u, v) ≤ u ⬝ᵥ u + v ⬝ᵥ v combined with u ⬝ᵥ u ≤ 1 and v ⬝ᵥ v ≤ sin²(2α) yields the required inequality by linear arithmetic.

    lines 87–88

    uses ghz_standard_frame_lambda_le_lambda_le_s

        · have h_s := ghz_standard_frame_lambda_le_lambda_le_s u v
          linarith
  11. When u satisfies u ⬝ᵥ u ≤ sin²(2α), Λ(u, v) ≤ 1 + sin²(2α) holds regardless of v.

    In the remaining branch where u ⬝ᵥ u ≤ sin²(2α), the general bound Λ(u, v) ≤ u ⬝ᵥ u + v ⬝ᵥ v combined with v ⬝ᵥ v ≤ 1 completes the proof.

    lines 89–90

    uses ghz_standard_frame_lambda_le_lambda_le_s

      · have h_s := ghz_standard_frame_lambda_le_lambda_le_s u v
        linarith
Finset.sum_nonneg
A finite sum of non-negative real numbers is non-negative.
ghz_standard_frame_lambda_le_dot_self_nonneg
The dot product of any vector in ℝ³ with itself is non-negative: 0 ≤ u ⬝ᵥ u.
add_nonneg
The sum of two non-negative real numbers is non-negative.
Real.sqrt_le_sqrt
The square root function on real numbers preserves non-strict inequality: if x ≤ y, then √x ≤ √y.
Real.sqrt_sq
For any non-negative real number x, √(x²) = x.
Fin.sum_univ_three
The sum of a function over the three-element set Fin 3 equals the sum of its values at 0, 1, and 2.
ghz_standard_frame_lambda_le_z_axis_dot
If two vectors in ℝ³ have zeros in their first two coordinates, their dot product equals the product of their third coordinates: u ⬝ᵥ v = u₂ * v₂.
Real.sqrt_sq_eq_abs
For any real number x, √(x²) = |x|.
le_total
Any two real numbers are comparable: a ≤ b or b ≤ a.
abs_sub_comm
Absolute difference is symmetric: |a - b| = |b - a|.
abs_of_nonneg
The absolute value of a non-negative real number equals itself: |a| = a.
sub_nonneg.mpr
If b ≤ a, then 0 ≤ a - b.
le_max_right
For any real numbers a and b, b ≤ max a b.
le_max_left
For any real numbers a and b, a ≤ max a b.
Matrix.one_apply
The identity matrix has entry 1 on the diagonal and 0 elsewhere.
Real.sin_sq_add_cos_sq
For any real angle x, sin²(x) + cos²(x) = 1.
sq
The square of a number equals the number multiplied by itself: a² = a * a.
ghz_standard_frame_lambda_le_u_dot_u_le_one
For any setting pair (μ, ν), the GHZ correlation vector with identity frames satisfies u ⬝ᵥ u ≤ 1.
ghz_standard_frame_lambda_le_cases
For any setting pair (μ, ν), the correlation vector either has vanishing x and y components or has u ⬝ᵥ u ≤ sin²(2α).
ghz_standard_frame_lambda_le_lambda_z_axis
If both vectors lie purely along the z-axis, then Λ(u, v) ≤ max(u ⬝ᵥ u, v ⬝ᵥ v).
max_le
If a ≤ c and b ≤ c, then max a b ≤ c.
sq_nonneg
The square of any real number is non-negative: 0 ≤ a².
ghz_standard_frame_lambda_le_lambda_le_s
For any vectors u, v ∈ ℝ³, the network quantity satisfies Λ(u, v) ≤ 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_standard_frame_lambda_le_dot_self_nonneg (u : Fin 3 → ℝ) : 0 ≤ u ⬝ᵥ u := by  change 0 ≤ ∑ i : Fin 3, u i * u i  apply Finset.sum_nonneg  intro i _  nlinarithlemma ghz_standard_frame_lambda_le_lambda_le_s (u v : Fin 3 → ℝ) : bell_network_lambda u v ≤ (u ⬝ᵥ u) + (v ⬝ᵥ v) := by  dsimp [bell_network_lambda]  set s := (u ⬝ᵥ u) + (v ⬝ᵥ v)  have hs : 0 ≤ s := add_nonneg (ghz_standard_frame_lambda_le_dot_self_nonneg u) (ghz_standard_frame_lambda_le_dot_self_nonneg v)  have hdisc_le : s ^ 2 - 4 * (u ⬝ᵥ v) ^ 2 ≤ s ^ 2 := by nlinarith  have hsqrt_le : Real.sqrt (s ^ 2 - 4 * (u ⬝ᵥ v) ^ 2) ≤ s := by    calc      Real.sqrt (s ^ 2 - 4 * (u ⬝ᵥ v) ^ 2) ≤ Real.sqrt (s ^ 2) := Real.sqrt_le_sqrt hdisc_le      _ = s := Real.sqrt_sq hs  linarithlemma ghz_standard_frame_lambda_le_z_axis_dot (u v : Fin 3 → ℝ) (hu0 : u 0 = 0) (hu1 : u 1 = 0) (hv0 : v 0 = 0) (hv1 : v 1 = 0) :    u ⬝ᵥ v = u 2 * v 2 := by  simp only [dotProduct, Fin.sum_univ_three]  rw [hu0, hu1, hv0, hv1]  ringlemma ghz_standard_frame_lambda_le_lambda_z_axis (u v : Fin 3 → ℝ) (hu0 : u 0 = 0) (hu1 : u 1 = 0) (hv0 : v 0 = 0) (hv1 : v 1 = 0) :    bell_network_lambda u v ≤ max (u ⬝ᵥ u) (v ⬝ᵥ v) := by  have hu : u ⬝ᵥ u = (u 2) ^ 2 := by    have := ghz_standard_frame_lambda_le_z_axis_dot u u hu0 hu1 hu0 hu1    linarith [this]  have hv : v ⬝ᵥ v = (v 2) ^ 2 := by    have := ghz_standard_frame_lambda_le_z_axis_dot v v hv0 hv1 hv0 hv1    linarith [this]  have huv : u ⬝ᵥ v = u 2 * v 2 := ghz_standard_frame_lambda_le_z_axis_dot u v hu0 hu1 hv0 hv1  dsimp [bell_network_lambda]  rw [hu, hv, huv]  set a := (u 2) ^ 2  set b := (v 2) ^ 2  have hdisc : (a + b) ^ 2 - 4 * (u 2 * v 2) ^ 2 = (a - b) ^ 2 := by    dsimp [a, b]    ring  rw [hdisc, Real.sqrt_sq_eq_abs]  cases le_total a b with  | inl h =>    have habs : |a - b| = b - a := abs_sub_comm a b ▸ abs_of_nonneg (sub_nonneg.mpr h)    rw [habs]    have : (1 / 2 : ℝ) * (a + b + (b - a)) = b := by ring    rw [this]    exact le_max_right a b  | inr h =>    have habs : |a - b| = a - b := abs_of_nonneg (sub_nonneg.mpr h)    rw [habs]    have : (1 / 2 : ℝ) * (a + b + (a - b)) = a := by ring    rw [this]    exact le_max_left a blemma ghz_standard_frame_lambda_le_u_dot_u_le_one (α : ℝ) (μ ν : Fin 4) :    ghz_correlation_vector α 1 1 μ ν ⬝ᵥ ghz_correlation_vector α 1 1 μ ν ≤ 1 := by  simp only [dotProduct, Fin.sum_univ_three]  dsimp [ghz_correlation_vector, ghz_pauli_coeff]  fin_cases μ <;> fin_cases ν <;> (    simp [Matrix.one_apply]    try nlinarith [Real.sin_sq_add_cos_sq (2 * α)]  )lemma ghz_standard_frame_lambda_le_cases (α : ℝ) (μ ν : Fin 4) :    ((ghz_correlation_vector α 1 1 μ ν) 0 = 0 ∧ (ghz_correlation_vector α 1 1 μ ν) 1 = 0) ∨    (ghz_correlation_vector α 1 1 μ ν ⬝ᵥ ghz_correlation_vector α 1 1 μ ν ≤ Real.sin (2 * α) ^ 2) := by  dsimp [ghz_correlation_vector, ghz_pauli_coeff]  simp only [dotProduct, Fin.sum_univ_three]  fin_cases μ <;> fin_cases ν <;> (    simp [Matrix.one_apply, sq]    try nlinarith  )/-- For standard identity frames e = I and f = I, the two-copy closed-network quantity Lambda is at most 1 + sin^2(2*alpha) for all setting quadruples. -/theorem ghz_standard_frame_lambda_le (α : ℝ) (μ ν μ' ν' : Fin 4) :    bell_network_lambda (ghz_correlation_vector α 1 1 μ ν) (ghz_correlation_vector α 1 1 μ' ν') ≤ 1 + Real.sin (2 * α) ^ 2 := by  set u := ghz_correlation_vector α 1 1 μ ν  set v := ghz_correlation_vector α 1 1 μ' ν'  have hu_le_1 : u ⬝ᵥ u ≤ 1 := ghz_standard_frame_lambda_le_u_dot_u_le_one α μ ν  have hv_le_1 : v ⬝ᵥ v ≤ 1 := ghz_standard_frame_lambda_le_u_dot_u_le_one α μ' ν'  rcases ghz_standard_frame_lambda_le_cases α μ ν with ⟨hu0, hu1⟩ | hu_sin  · rcases ghz_standard_frame_lambda_le_cases α μ' ν' with ⟨hv0, hv1⟩ | hv_sin    · have h_z := ghz_standard_frame_lambda_le_lambda_z_axis u v hu0 hu1 hv0 hv1      have hmax : max (u ⬝ᵥ u) (v ⬝ᵥ v) ≤ 1 := max_le hu_le_1 hv_le_1      have hsin_ge : 0 ≤ Real.sin (2 * α) ^ 2 := sq_nonneg _      linarith    · have h_s := ghz_standard_frame_lambda_le_lambda_le_s u v      linarith  · have h_s := ghz_standard_frame_lambda_le_lambda_le_s u v    linarith
See it in the graphDiscuss this resultChallenge it