three_qubit_three_pairwise_zero_axes
For any three real 3 by 3 pair-correlation matrices there are unit axes for the three parties which simultaneously zero all three pair correlations.
Statement
theorem three_qubit_three_pairwise_zero_axes (A B C : Matrix (Fin 3) (Fin 3) ℝ) :
∃ a b c : Fin 3 → ℝ,
(∑ i, (a i)^2) = 1 ∧ (∑ i, (b i)^2) = 1 ∧ (∑ i, (c i)^2) = 1 ∧
(∑ i, ∑ j, a i * A i j * b j) = 0 ∧
(∑ i, ∑ j, a i * B i j * c j) = 0 ∧
(∑ i, ∑ j, b i * C i j * c j) = 0- Source
- Filed by the prover as a step towards another result
- Verified
- 15 Nov 2026
- Axioms
Classical.choiceQuot.soundpropext
The proof, explained
The theorem states that for any three real 3 by 3 correlation matrices A, B, and C, there exist unit vectors a, b, and c in ℝ³ such that aᵀ A b = 0, aᵀ B c = 0, and bᵀ C c = 0. The proof proceeds sequentially: it fixes a arbitrarily as the standard unit vector (1, 0, 0), then uses rank-nullity in ℝ³ to find a unit vector b orthogonal to the row vector a A. Finally, applying the same rank-nullity argument to the two constraints given by a B and b C yields a unit vector c orthogonal to both, zeroing all three pair correlations simultaneously.
The map v ↦ (u ⬝ᵥ v, w ⬝ᵥ v) from ℝ³ to ℝ² is a linear map.
Additivity and scalability follow componentwise: dot product distributes over addition via dotProduct_add and pulls scalars out via dotProduct_smul.
uses
dotProduct_add,dotProduct_smulmap_add' v1 v2 := by ext i fin_cases i <;> simp [dotProduct_add] map_smul' c v := by ext i fin_cases i <;> simp [dotProduct_smul]
The kernel of the linear map f(v) = (u ⬝ᵥ v, w ⬝ᵥ v) has strictly positive dimension.
The domain ℝ³ has dimension 3 and the codomain ℝ² has dimension 2 (using finrank_pi and Fintype.card_fin). Since the range has dimension at most 2 by Submodule.finrank_le, the rank-nullity theorem gives dim(ker f) = 3 - dim(range f) ≥ 1 > 0.
uses
finrank_pi,Fintype.card_fin,Submodule.finrank_lelet f := three_qubit_three_pairwise_zero_axes_two_linear_forms u w have h_dim : finrank ℝ (Fin 3 → ℝ) = 3 := by rw [finrank_pi ℝ, Fintype.card_fin] have h_codim : finrank ℝ (Fin 2 → ℝ) = 2 := by rw [finrank_pi ℝ, Fintype.card_fin] have h_range_le : finrank ℝ f.range ≤ finrank ℝ (Fin 2 → ℝ) := Submodule.finrank_le f.range have h_sum := f.finrank_range_add_finrank_ker have h_ker_pos : 0 < finrank ℝ f.ker := by omega
There exists a non-zero vector in ℝ³ orthogonal to both u and w.
Because dim(ker f) > 0, there exists a non-zero element v in ker f by finrank_pos_iff_exists_ne_zero.mp. Evaluating f(v) = 0 at indices 0 and 1 via LinearMap.mem_ker.mp and congr_fun yields u ⬝ᵥ v = 0 and w ⬝ᵥ v = 0.
uses
finrank_pos_iff_exists_ne_zero.mp,congr_fun,LinearMap.mem_ker.mpobtain ⟨⟨v, hv_mem⟩, hv_ne⟩ := finrank_pos_iff_exists_ne_zero.mp h_ker_pos refine ⟨v, ?_, ?_, ?_⟩ · intro h apply hv_ne ext exact congr_fun h _ · have h_zero : f v = 0 := LinearMap.mem_ker.mp hv_mem have h0 := congr_fun h_zero 0 change (![u ⬝ᵥ v, w ⬝ᵥ v] : Fin 2 → ℝ) 0 = 0 at h0 exact h0 · have h_zero : f v = 0 := LinearMap.mem_ker.mp hv_mem have h1 := congr_fun h_zero 1 change (![u ⬝ᵥ v, w ⬝ᵥ v] : Fin 2 → ℝ) 1 = 0 at h1 exact h1
The dot product of any non-zero vector in ℝ³ with itself is strictly positive.
The self-dot product v ⬝ᵥ v equals ∑ i, (v i)². Each squared term is non-negative by sq_nonneg, and since v ≠ 0, at least one component is non-zero, making its square strictly positive. By Finset.sum_pos', the sum is strictly positive.
uses
Finset.sum_pos',sq_nonneg,Finset.mem_univ,sq_eq_zero_iff.mp,le_antisymmhave : (v ⬝ᵥ v) = ∑ i, (v i)^2 := by simp [dotProduct, sq] rw [this] apply Finset.sum_pos' · intro i _ exact sq_nonneg (v i) · contrapose! hv ext i have := hv i (Finset.mem_univ i) exact sq_eq_zero_iff.mp (le_antisymm this (sq_nonneg (v i)))
Normalizing a non-zero vector by the reciprocal of the square root of its norm squared yields a vector of unit norm.
Because v ⬝ᵥ v > 0, its square root squared equals v ⬝ᵥ v by Real.sq_sqrt. Expanding the dot product of c = (√(v ⬝ᵥ v))⁻¹ • v with itself pulls out the scalar factors, giving ((√(v ⬝ᵥ v))⁻¹)² * (v ⬝ᵥ v) = (v ⬝ᵥ v)⁻¹ * (v ⬝ᵥ v) = 1 by inv_mul_cancel₀.
uses
three_qubit_three_pairwise_zero_axes_dotProduct_self_pos,Real.sq_sqrt,le_of_lt,smul_dotProduct,dotProduct_smul,smul_eq_mul,inv_pow,inv_mul_cancel₀,ne_of_gtintro c have hpos := three_qubit_three_pairwise_zero_axes_dotProduct_self_pos hv have hsq : (Real.sqrt (v ⬝ᵥ v)) ^ 2 = v ⬝ᵥ v := Real.sq_sqrt (le_of_lt hpos) dsimp [c] rw [smul_dotProduct, dotProduct_smul, smul_eq_mul, smul_eq_mul] have h2 : (Real.sqrt (v ⬝ᵥ v))⁻¹ * ((Real.sqrt (v ⬝ᵥ v))⁻¹ * (v ⬝ᵥ v)) = ((Real.sqrt (v ⬝ᵥ v))⁻¹)^2 * (v ⬝ᵥ v) := by ring rw [h2, inv_pow, hsq] exact inv_mul_cancel₀ (ne_of_gt hpos)
Scaling a vector preserves orthogonality.
If u ⬝ᵥ v = 0, then u ⬝ᵥ (k • v) = k * (u ⬝ᵥ v) = k * 0 = 0 by dotProduct_smul, smul_eq_mul, and mul_zero.
uses
dotProduct_smul,smul_eq_mul,mul_zerolemma three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero {u v : Fin 3 → ℝ} (k : ℝ) (h : u ⬝ᵥ v = 0) : u ⬝ᵥ (k • v) = 0 := by rw [dotProduct_smul, smul_eq_mul, h, mul_zero]
For any two vectors u, w in ℝ³, there exists a unit vector orthogonal to both.
We obtain a non-zero vector v orthogonal to u and w via three_qubit_three_pairwise_zero_axes_exists_ne_zero_orthogonal, then normalize it by k = (√(v ⬝ᵥ v))⁻¹. The normalized vector k • v has unit norm by three_qubit_three_pairwise_zero_axes_normalize_dotProduct and remains orthogonal to u and w by three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero.
uses
three_qubit_three_pairwise_zero_axes_exists_ne_zero_orthogonal,three_qubit_three_pairwise_zero_axes_normalize_dotProduct,three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zeroobtain ⟨v, hv, hu, hw⟩ := three_qubit_three_pairwise_zero_axes_exists_ne_zero_orthogonal u w let k := (Real.sqrt (v ⬝ᵥ v))⁻¹ refine ⟨k • v, ?_, ?_, ?_⟩ · have h1 := three_qubit_three_pairwise_zero_axes_normalize_dotProduct hv have h2 : (k • v) ⬝ᵥ (k • v) = ∑ i, ((k • v) i)^2 := by simp [dotProduct, sq] rw [← h2] exact h1 · exact three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero k hu · exact three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero k hw
The bilinear form ∑ i, ∑ j, a i * A i j * b j equals the dot product (a A) ⬝ᵥ b.
Expanding the vector-matrix product and the dot product, swapping the summation order using Finset.sum_comm, and distributing multiplication with Finset.sum_mul shows that the two expressions are identical.
uses
Finset.sum_comm,Finset.sum_mul,mul_assocsimp only [vecMul, dotProduct] rw [Finset.sum_comm] simp only [Finset.sum_mul, mul_assoc]
Choose a unit vector a for the first party.
We choose the standard basis vector a = (1, 0, 0), whose sum of squared components is 1 by expanding the sum over Fin 3 via Fin.sum_univ_three.
uses
Fin.sum_univ_threelet a : Fin 3 → ℝ := ![(1 : ℝ), 0, 0] have ha : (∑ i, (a i)^2) = 1 := by simp [a, Fin.sum_univ_three]
Choose unit vectors b and c orthogonal to the required vectors.
Using three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal, we find a unit vector b orthogonal to a A (and 0), and then a unit vector c orthogonal to both a B and b C.
uses
three_qubit_three_pairwise_zero_axes_exists_unit_orthogonalobtain ⟨b, hb, hab, -⟩ := three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal (vecMul a A) 0 obtain ⟨c, hc, hac, hbc⟩ := three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal (vecMul a B) (vecMul b C)
The vectors a, b, c simultaneously zero all three pair correlations.
Rewriting each bilinear form as a dot product via three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq, the expressions reduce to (a A) ⬝ᵥ b = 0, (a B) ⬝ᵥ c = 0, and (b C) ⬝ᵥ c = 0, which hold by the choice of b and c.
uses
three_qubit_three_pairwise_zero_axes_matrix_bilinear_eqrefine ⟨a, b, c, ha, hb, hc, ?_, ?_, ?_⟩ · rw [three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq]; exact hab · rw [three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq]; exact hac · rw [three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq]; exact hbc
dotProduct_add- The dot product distributes over addition in the second argument: u ⬝ᵥ (v + w) = u ⬝ᵥ v + u ⬝ᵥ w.
dotProduct_smul- Scaling the second argument in a dot product pulls out the scalar factor: v ⬝ᵥ (c • w) = c • (v ⬝ᵥ w).
finrank_pi- The dimension of a function space (ι → R) from a finite index set ι to a field R is the cardinality of ι.
Fintype.card_fin- The cardinality of the finite set Fin n is n.
Submodule.finrank_le- The dimension of a submodule is at most the dimension of the ambient module.
finrank_pos_iff_exists_ne_zero.mp- If a finite-dimensional module has strictly positive dimension, it contains a non-zero element.
congr_fun- Equal functions take equal values at any argument.
LinearMap.mem_ker.mp- An element belongs to the kernel of a linear map if and only if the map sends it to zero.
Finset.sum_pos'- A finite sum of non-negative terms with at least one strictly positive term is strictly positive.
sq_nonneg- The square of any real number is non-negative.
Finset.mem_univ- Every element of a finite type belongs to the universal finset univ.
sq_eq_zero_iff.mp- If the square of a real number is zero, the number itself is zero.
le_antisymm- If a ≤ b and b ≤ a, then a = b.
three_qubit_three_pairwise_zero_axes_dotProduct_self_pos- The dot product of any non-zero vector in ℝ³ with itself is strictly positive.
Real.sq_sqrt- For any non-negative real number x, the square of its square root is x.
le_of_lt- A strict inequality a < b implies a ≤ b.
smul_dotProduct- Scaling the first argument in a dot product pulls out the scalar factor: (c • v) ⬝ᵥ w = c • (v ⬝ᵥ w).
smul_eq_mul- Scalar multiplication on the real numbers coincides with ordinary multiplication: a • b = a * b.
inv_pow- The n-th power of an inverse is the inverse of the n-th power: (a⁻¹)^n = (a^n)⁻¹.
inv_mul_cancel₀- Multiplying a non-zero element by its inverse yields 1: a⁻¹ * a = 1.
ne_of_gt- If a > b, then a ≠ b.
mul_zero- Multiplying any number by zero yields zero: a * 0 = 0.
three_qubit_three_pairwise_zero_axes_exists_ne_zero_orthogonal- For any two vectors in ℝ³, there exists a non-zero vector orthogonal to both.
three_qubit_three_pairwise_zero_axes_normalize_dotProduct- Scaling a non-zero vector by the reciprocal of the square root of its self-dot-product yields a vector of unit norm.
three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero- Scaling a vector preserves its orthogonality to any vector u: if u ⬝ᵥ v = 0, then u ⬝ᵥ (k • v) = 0.
Finset.sum_comm- The order of summation in a double finite sum can be exchanged.
Finset.sum_mul- Multiplication distributes over a finite sum: (∑ i, f i) * a = ∑ i, f i * a.
mul_assoc- Multiplication is associative: (a * b) * c = a * (b * c).
Fin.sum_univ_three- A sum over Fin 3 equals the sum of the components at indices 0, 1, and 2.
three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal- For any two vectors in ℝ³, there exists a unit vector orthogonal to both.
three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq- The matrix bilinear form ∑ i, ∑ j, a i * A i j * b j equals the dot product of the row-vector matrix product a A with b.
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 Moduledef three_qubit_three_pairwise_zero_axes_two_linear_forms (u w : Fin 3 → ℝ) : (Fin 3 → ℝ) →ₗ[ℝ] (Fin 2 → ℝ) where toFun v := ![u ⬝ᵥ v, w ⬝ᵥ v] map_add' v1 v2 := by ext i fin_cases i <;> simp [dotProduct_add] map_smul' c v := by ext i fin_cases i <;> simp [dotProduct_smul]lemma three_qubit_three_pairwise_zero_axes_exists_ne_zero_orthogonal (u w : Fin 3 → ℝ) : ∃ v : Fin 3 → ℝ, v ≠ 0 ∧ u ⬝ᵥ v = 0 ∧ w ⬝ᵥ v = 0 := by let f := three_qubit_three_pairwise_zero_axes_two_linear_forms u w have h_dim : finrank ℝ (Fin 3 → ℝ) = 3 := by rw [finrank_pi ℝ, Fintype.card_fin] have h_codim : finrank ℝ (Fin 2 → ℝ) = 2 := by rw [finrank_pi ℝ, Fintype.card_fin] have h_range_le : finrank ℝ f.range ≤ finrank ℝ (Fin 2 → ℝ) := Submodule.finrank_le f.range have h_sum := f.finrank_range_add_finrank_ker have h_ker_pos : 0 < finrank ℝ f.ker := by omega obtain ⟨⟨v, hv_mem⟩, hv_ne⟩ := finrank_pos_iff_exists_ne_zero.mp h_ker_pos refine ⟨v, ?_, ?_, ?_⟩ · intro h apply hv_ne ext exact congr_fun h _ · have h_zero : f v = 0 := LinearMap.mem_ker.mp hv_mem have h0 := congr_fun h_zero 0 change (![u ⬝ᵥ v, w ⬝ᵥ v] : Fin 2 → ℝ) 0 = 0 at h0 exact h0 · have h_zero : f v = 0 := LinearMap.mem_ker.mp hv_mem have h1 := congr_fun h_zero 1 change (![u ⬝ᵥ v, w ⬝ᵥ v] : Fin 2 → ℝ) 1 = 0 at h1 exact h1lemma three_qubit_three_pairwise_zero_axes_dotProduct_self_pos {v : Fin 3 → ℝ} (hv : v ≠ 0) : 0 < v ⬝ᵥ v := by have : (v ⬝ᵥ v) = ∑ i, (v i)^2 := by simp [dotProduct, sq] rw [this] apply Finset.sum_pos' · intro i _ exact sq_nonneg (v i) · contrapose! hv ext i have := hv i (Finset.mem_univ i) exact sq_eq_zero_iff.mp (le_antisymm this (sq_nonneg (v i)))lemma three_qubit_three_pairwise_zero_axes_normalize_dotProduct {v : Fin 3 → ℝ} (hv : v ≠ 0) : let c := (Real.sqrt (v ⬝ᵥ v))⁻¹ • v c ⬝ᵥ c = 1 := by intro c have hpos := three_qubit_three_pairwise_zero_axes_dotProduct_self_pos hv have hsq : (Real.sqrt (v ⬝ᵥ v)) ^ 2 = v ⬝ᵥ v := Real.sq_sqrt (le_of_lt hpos) dsimp [c] rw [smul_dotProduct, dotProduct_smul, smul_eq_mul, smul_eq_mul] have h2 : (Real.sqrt (v ⬝ᵥ v))⁻¹ * ((Real.sqrt (v ⬝ᵥ v))⁻¹ * (v ⬝ᵥ v)) = ((Real.sqrt (v ⬝ᵥ v))⁻¹)^2 * (v ⬝ᵥ v) := by ring rw [h2, inv_pow, hsq] exact inv_mul_cancel₀ (ne_of_gt hpos)lemma three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero {u v : Fin 3 → ℝ} (k : ℝ) (h : u ⬝ᵥ v = 0) : u ⬝ᵥ (k • v) = 0 := by rw [dotProduct_smul, smul_eq_mul, h, mul_zero]lemma three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal (u w : Fin 3 → ℝ) : ∃ c : Fin 3 → ℝ, (∑ i, (c i)^2) = 1 ∧ u ⬝ᵥ c = 0 ∧ w ⬝ᵥ c = 0 := by obtain ⟨v, hv, hu, hw⟩ := three_qubit_three_pairwise_zero_axes_exists_ne_zero_orthogonal u w let k := (Real.sqrt (v ⬝ᵥ v))⁻¹ refine ⟨k • v, ?_, ?_, ?_⟩ · have h1 := three_qubit_three_pairwise_zero_axes_normalize_dotProduct hv have h2 : (k • v) ⬝ᵥ (k • v) = ∑ i, ((k • v) i)^2 := by simp [dotProduct, sq] rw [← h2] exact h1 · exact three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero k hu · exact three_qubit_three_pairwise_zero_axes_dotProduct_smul_eq_zero k hwlemma three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq (A : Matrix (Fin 3) (Fin 3) ℝ) (a b : Fin 3 → ℝ) : (∑ i, ∑ j, a i * A i j * b j) = (vecMul a A) ⬝ᵥ b := by simp only [vecMul, dotProduct] rw [Finset.sum_comm] simp only [Finset.sum_mul, mul_assoc]theorem three_qubit_three_pairwise_zero_axes (A B C : Matrix (Fin 3) (Fin 3) ℝ) : ∃ a b c : Fin 3 → ℝ, (∑ i, (a i)^2) = 1 ∧ (∑ i, (b i)^2) = 1 ∧ (∑ i, (c i)^2) = 1 ∧ (∑ i, ∑ j, a i * A i j * b j) = 0 ∧ (∑ i, ∑ j, a i * B i j * c j) = 0 ∧ (∑ i, ∑ j, b i * C i j * c j) = 0 := by let a : Fin 3 → ℝ := ![(1 : ℝ), 0, 0] have ha : (∑ i, (a i)^2) = 1 := by simp [a, Fin.sum_univ_three] obtain ⟨b, hb, hab, -⟩ := three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal (vecMul a A) 0 obtain ⟨c, hc, hac, hbc⟩ := three_qubit_three_pairwise_zero_axes_exists_unit_orthogonal (vecMul a B) (vecMul b C) refine ⟨a, b, c, ha, hb, hc, ?_, ?_, ?_⟩ · rw [three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq]; exact hab · rw [three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq]; exact hac · rw [three_qubit_three_pairwise_zero_axes_matrix_bilinear_eq]; exact hbc