AFTD Auto-Formalizing Theoretical Domains

real_positive_qutrit_gram_normalizer

theoremverified

A positive-definite real qutrit Gram block has an invertible symmetric normalizing filter.

Statement

theorem real_positive_qutrit_gram_normalizer :
  ∀ A : Matrix (Fin 3) (Fin 3) ℝ, A.PosDef →
    ∃ H : Matrix (Fin 3) (Fin 3) ℝ,
      H.transpose = H ∧ IsUnit H ∧ H*A*H = 1
Source
Local compatibility helper for OP-693.
Verified
12 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Used by 3

The proof, explained

The proof constructs the normalizer H as the inverse of the positive square root S = √A obtained via the continuous functional calculus. Because A is positive-definite, S is symmetric and invertible, which implies that its inverse H = S⁻¹ is also symmetric and invertible. Expanding A = S * S then yields H * A * H = (S⁻¹ * S) * (S * S⁻¹) = 1.

  1. The continuous functional calculus square root S = √A is positive semidefinite and symmetric (Sᵀ = S).

    We define S as the functional calculus square root √A. Since A is positive-definite, it is non-negative, and therefore its square root S is non-negative and positive semidefinite (CFC.sqrt_nonneg). A positive semidefinite matrix is Hermitian, and over the real numbers conjugate transpose equals the transpose (Matrix.conjTranspose_eq_transpose_of_trivial), so S is symmetric.

    lines 7–10

    uses CFC.sqrt_nonneg, Matrix.conjTranspose_eq_transpose_of_trivial

      let S : Matrix (Fin 3) (Fin 3) ℝ := CFC.sqrt A
      have hSpos : S.PosSemidef := (CFC.sqrt_nonneg A).posSemidef
      have hSsym : S.transpose=S := by
        simpa only [Matrix.conjTranspose_eq_transpose_of_trivial] using hSpos.isHermitian.eq
  2. S squares to A, and S is invertible with an invertible determinant.

    By properties of the continuous functional calculus, S * S = A (CFC.sq_sqrt, pow_two). Since A is positive-definite, it is invertible, which implies its factor S is also invertible (isUnit_of_mul_isUnit_left). Over a commutative ring, invertibility of a matrix is equivalent to invertibility of its determinant (Matrix.isUnit_iff_isUnit_det), so det(S) is invertible.

    lines 11–15

    uses CFC.sq_sqrt, pow_two, isUnit_of_mul_isUnit_left, Matrix.isUnit_iff_isUnit_det

      have hSq : S*S=A := by
        have h := CFC.sq_sqrt A hA.posSemidef.nonneg
        simpa only [pow_two] using h
      have hSunit : IsUnit S := isUnit_of_mul_isUnit_left (hSq ▸ hA.isUnit)
      have hSdet : IsUnit S.det := (Matrix.isUnit_iff_isUnit_det S).mp hSunit
  3. The inverse H = S⁻¹ is invertible, symmetric (Hᵀ = H), and satisfies H * S = 1 and S * H = 1.

    We set H = S⁻¹. Since S is invertible, H is also invertible (Matrix.isUnit_nonsing_inv_iff.mpr). The inverse of a Hermitian matrix is Hermitian, which over ℝ means H is symmetric: Hᵀ = H (Matrix.conjTranspose_eq_transpose_of_trivial). Because det(S) is invertible, H is a two-sided inverse of S, so H * S = 1 and S * H = 1 (Matrix.nonsing_inv_mul, Matrix.mul_nonsing_inv).

    lines 16–22

    uses Matrix.isUnit_nonsing_inv_iff.mpr, Matrix.conjTranspose_eq_transpose_of_trivial, Matrix.nonsing_inv_mul, Matrix.mul_nonsing_inv

      let H : Matrix (Fin 3) (Fin 3) ℝ := S⁻¹
      have hHunit : IsUnit H := Matrix.isUnit_nonsing_inv_iff.mpr hSunit
      have hHsym : H.transpose=H := by
        have h := hSpos.isHermitian.inv
        simpa only [Matrix.conjTranspose_eq_transpose_of_trivial] using h.eq
      have hHS : H*S=1 := Matrix.nonsing_inv_mul S hSdet
      have hSH : S*H=1 := Matrix.mul_nonsing_inv S hSdet
  4. H satisfies the normalization condition H * A * H = 1, completing the proof of the helper lemma.

    We provide H, its symmetry, and its invertibility as the witness. To establish H * A * H = 1, we replace A by S * S, reassociate matrix multiplication to (H * S) * (S * H) using Matrix.mul_assoc, and substitute H * S = 1 and S * H = 1 to conclude the product is 1 (Matrix.one_mul).

    lines 23–27

    uses Matrix.mul_assoc, Matrix.one_mul

      refine ⟨H, hHsym, hHunit, ?_⟩
      calc
        H*A*H = H*(S*S)*H := by rw [hSq]
        _ = (H*S)*(S*H) := by simp only [Matrix.mul_assoc]
        _ = 1 := by rw [hHS,hSH,Matrix.one_mul]
  5. Every positive-definite real 3 × 3 matrix has an invertible symmetric normalizer.

    We introduce the matrix A and the assumption that A is positive-definite, and apply the helper lemma in767_posnormalproof_real_positive_normalizer.

    lines 33–34

    uses in767_posnormalproof_real_positive_normalizer

      intro A hA
      exact in767_posnormalproof_real_positive_normalizer A hA
CFC.sqrt_nonneg
The functional calculus square root of any element is non-negative (positive semidefinite).
Matrix.conjTranspose_eq_transpose_of_trivial
For matrices with trivial conjugation (such as real matrices), the conjugate transpose equals the ordinary transpose.
CFC.sq_sqrt
The square of the functional calculus square root of a non-negative element equals the original element.
pow_two
The square of an element a² equals a * a.
isUnit_of_mul_isUnit_left
If a product x * y is invertible in a Dedekind-finite monoid, then the left factor x is invertible.
Matrix.isUnit_iff_isUnit_det
A square matrix is invertible if and only if its determinant is invertible.
Matrix.isUnit_nonsing_inv_iff.mpr
If a matrix is invertible, its inverse is also invertible.
Matrix.nonsing_inv_mul
If a square matrix has an invertible determinant, then its inverse multiplied by the matrix on the right equals the identity.
Matrix.mul_nonsing_inv
If a square matrix has an invertible determinant, then the matrix multiplied by its inverse on the right equals the identity.
Matrix.mul_assoc
Multiplication of matrices is associative: (L * M) * N = L * (M * N).
Matrix.one_mul
Multiplying the identity matrix on the left leaves any matrix unchanged: 1 * M = M.
in767_posnormalproof_real_positive_normalizer
Any positive-definite real 3 × 3 matrix has an invertible symmetric normalizer H such that H * A * H = 1.

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 BigOperators Matrix MatrixOrder ComplexOrderset_option backward.isDefEq.respectTransparency falseprivate lemma in767_posnormalproof_real_positive_normalizer    (A : Matrix (Fin 3) (Fin 3) ℝ) (hA : A.PosDef) :    ∃ H : Matrix (Fin 3) (Fin 3) ℝ, H.transpose = H ∧ IsUnit H ∧ H*A*H=1 := by  let S : Matrix (Fin 3) (Fin 3) ℝ := CFC.sqrt A  have hSpos : S.PosSemidef := (CFC.sqrt_nonneg A).posSemidef  have hSsym : S.transpose=S := by    simpa only [Matrix.conjTranspose_eq_transpose_of_trivial] using hSpos.isHermitian.eq  have hSq : S*S=A := by    have h := CFC.sq_sqrt A hA.posSemidef.nonneg    simpa only [pow_two] using h  have hSunit : IsUnit S := isUnit_of_mul_isUnit_left (hSq ▸ hA.isUnit)  have hSdet : IsUnit S.det := (Matrix.isUnit_iff_isUnit_det S).mp hSunit  let H : Matrix (Fin 3) (Fin 3) ℝ := S⁻¹  have hHunit : IsUnit H := Matrix.isUnit_nonsing_inv_iff.mpr hSunit  have hHsym : H.transpose=H := by    have h := hSpos.isHermitian.inv    simpa only [Matrix.conjTranspose_eq_transpose_of_trivial] using h.eq  have hHS : H*S=1 := Matrix.nonsing_inv_mul S hSdet  have hSH : S*H=1 := Matrix.mul_nonsing_inv S hSdet  refine ⟨H, hHsym, hHunit, ?_⟩  calc    H*A*H = H*(S*S)*H := by rw [hSq]    _ = (H*S)*(S*H) := by simp only [Matrix.mul_assoc]    _ = 1 := by rw [hHS,hSH,Matrix.one_mul]theorem real_positive_qutrit_gram_normalizer :  ∀ A : Matrix (Fin 3) (Fin 3) ℝ, A.PosDef →    ∃ H : Matrix (Fin 3) (Fin 3) ℝ,      H.transpose = H ∧ IsUnit H ∧ H*A*H = 1 := by  intro A hA  exact in767_posnormalproof_real_positive_normalizer A hA
See it in the graphDiscuss this resultChallenge it