AFTD Auto-Formalizing Theoretical Domains

not_universal_spectral_bound_positive_tp_dim_three

theoremverifiedoriginal

The universal spectral bound Tr(Φ) ≤ 3 m(Φ) + 6 fails for some positive trace-preserving map Φ on 3 × 3 complex matrices: Φ(X) = K X K + (3/4) P Xᵀ P with K = diag(1/2, 1/2, 1) and P = diag(1, 1, 0) is positive (a completely positive plus a completely copositive map) and trace-preserving, has operator trace 11/2 and the eigenvalue -1/2 (eigenvector E₁₂ - E₂₁), so 3 m(Φ) + 6 ≤ 9/2 < 11/2.

Statement

theorem not_universal_spectral_bound_positive_tp_dim_three :
    ¬ ∀ (Φ : Matrix (Fin 3) (Fin 3) ℂ →ₗ[ℂ] Matrix (Fin 3) (Fin 3) ℂ),
      is_positive_linear_map 3 Φ → is_trace_preserving_linear_map 3 Φ →
      ((LinearMap.trace ℂ (Matrix (Fin 3) (Fin 3) ℂ)) Φ).re ≤ 3 * spectral_min_re 3 Φ + 6
Related work
OP-733: arXiv:2610.37187 (Removing Trace Preservation from a Universal Spectral Bound for 2-Positive Maps), Sec. 9, Concluding remarks: whether positive trace-preserving maps in dimension d >= 3 can violate (1.2), a question raised in arXiv:2506.02145.
Verified
01 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 3
  • spectral_min_reThe minimal real part m(Φ) of an eigenvalue of a linear map Φ on d × d complex matrices is the infimum of the real parts of the elements…
  • is_positive_linear_mapA linear map Φ on d × d complex matrices is positive if it maps every positive semidefinite matrix to a positive semidefinite matrix.
  • is_trace_preserving_linear_mapA linear map Φ on d × d complex matrices is trace-preserving if the trace of Φ(X) equals the trace of X for all d × d complex matrices X.
Used by 1

Read back from the Lean

It is not the case that for every ℂ-linear map Φ : Matrix (Fin 3) (Fin 3) ℂ →ₗ[ℂ] Matrix (Fin 3) (Fin 3) ℂ, if Φ is positive (maps positive semidefinite matrices to positive semidefinite matrices) and trace-preserving (preserves the matrix trace of every matrix), then the real part of the trace of Φ (as a linear map on the ℂ-module of 3×3 complex matrices) is less than or equal to 3 * (the infimum of the real parts of the eigenvalues in the spectrum of Φ) + 6.

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 universal spectral bound Re(Tr(Φ)) ≤ 3 m(Φ) + 6 fails in dimension 3 for the map Φ(X) = K X K + (3/4) P Xᵀ P, with K = diag(1/2, 1/2, 1) and P = diag(1, 1, 0). The map Φ is positive because K and P are self-adjoint diagonal matrices, making Φ the sum of a completely positive map and a completely copositive map, and an explicit computation of diagonal entries confirms that Φ is trace-preserving. Its operator trace on M₃(ℂ) is 11/2, while the skew-symmetric matrix E₁₂ - E₂₁ is an eigenvector with eigenvalue -1/2, ensuring m(Φ) ≤ -1/2. Substituting these values into the conjectured inequality yields 11/2 ≤ 9/2, refuting the bound.

  1. The operation Φ(X) = K X K + (3/4) P Xᵀ P defines a ℂ-linear map on 3 × 3 complex matrices.

    Additivity (map_add') is verified by expanding matrix multiplication, transposition, and scalar multiplication over matrix addition using Matrix.mul_add, Matrix.add_mul, Matrix.transpose_add, and smul_add, followed by abelian group simplification (abel). Complex scalar multiplication (map_smul') follows similarly using Matrix.mul_smul, Matrix.smul_mul, Matrix.transpose_smul, smul_add, and RingHom.id_apply, and commuting scalars with smul_comm.

    lines 5–9

    uses Matrix.mul_add, Matrix.add_mul, Matrix.transpose_add, smul_add, Matrix.mul_smul, Matrix.smul_mul, Matrix.transpose_smul, RingHom.id_apply, smul_comm

      map_add' X Y := by
        simp only [Matrix.mul_add, Matrix.add_mul, Matrix.transpose_add, smul_add]; abel
      map_smul' c X := by
        simp only [Matrix.mul_smul, Matrix.smul_mul, Matrix.transpose_smul, smul_add, RingHom.id_apply]
        rw [smul_comm c (3/4 : ℂ)]
  2. The linear map Φ is positive: if X is positive semidefinite, then Φ(X) is positive semidefinite.

    Unfolding the definition with op733_map_apply, Φ(X) is the sum of K X K and (3/4) P Xᵀ P. Because K = diag(1/2, 1/2, 1) and P = diag(1, 1, 0) have real diagonal entries, they are self-adjoint (Matrix.diagonal_conjTranspose). Consequently, K X K = K X K† is positive semidefinite for any positive semidefinite X, and transposition preserves positive semidefiniteness, so P Xᵀ P = P Xᵀ P† is also positive semidefinite. Since 3/4 ≥ 0 (Complex.zero_le_real.mpr), their positive linear combination is positive semidefinite by Matrix.PosSemidef.add and Matrix.PosSemidef.smul.

    lines 16–30

    uses op733_map_apply, Matrix.diagonal_conjTranspose, Matrix.PosSemidef.add, Matrix.PosSemidef.smul, Complex.zero_le_real.mpr

    theorem op733_pos : is_positive_linear_map 3 op733_map := by
      classical
      intro X hX
      rw [op733_map_apply]
      have hK : Matrix.conjTranspose (Matrix.diagonal ![(1/2 : ℂ), 1/2, 1]) = Matrix.diagonal ![1/2, 1/2, 1] := by
        rw [Matrix.diagonal_conjTranspose]; congr 1; ext i; fin_cases i <;> simp
      have hP : Matrix.conjTranspose (Matrix.diagonal ![(1 : ℂ), 1, 0]) = Matrix.diagonal ![1, 1, 0] := by
        rw [Matrix.diagonal_conjTranspose]; congr 1; ext i; fin_cases i <;> simp
      refine Matrix.PosSemidef.add ?_ (Matrix.PosSemidef.smul ?_ ?_)
      · have := hX.mul_mul_conjTranspose_same (Matrix.diagonal ![(1/2 : ℂ), 1/2, 1])
        rwa [hK] at this
      · have := hX.transpose.mul_mul_conjTranspose_same (Matrix.diagonal ![(1 : ℂ), 1, 0])
        rwa [hP] at this
      · rw [show (3/4 : ℂ) = ((3/4 : ℝ) : ℂ) by push_cast; ring]
        exact Complex.zero_le_real.mpr (by norm_num)
  3. The linear map Φ is trace-preserving: Tr(Φ(X)) = Tr(X) for all 3 × 3 matrices X.

    Rewriting with op733_map_apply, the matrix trace is expressed as the sum of the three diagonal entries via Matrix.trace and Fin.sum_univ_three. Evaluating the diagonal products using Matrix.diagonal_mul and Matrix.mul_diagonal yields (1/4 X₀₀ + 3/4 X₀₀) + (1/4 X₁₁ + 3/4 X₁₁) + (1 · X₂₂ · 1 + 0) = X₀₀ + X₁₁ + X₂₂, which equals Tr(X) by polynomial simplification (ring).

    lines 32–36

    uses op733_map_apply, Matrix.trace, Fin.sum_univ_three, Matrix.diagonal_mul, Matrix.mul_diagonal

    theorem op733_tp : is_trace_preserving_linear_map 3 op733_map := by
      intro X
      rw [op733_map_apply]
      simp [Matrix.trace, Fin.sum_univ_three, Matrix.diagonal_mul, Matrix.mul_diagonal]
      ring
  4. The skew-symmetric matrix V = E₁₂ - E₂₁ satisfies Φ(V) = (-1/2) V.

    Unfolding Φ(V) via op733_map_apply, the identity is checked entrywise for all (i, j) ∈ {0, 1, 2}² (ext and fin_cases). Simplifying the diagonal matrix products and transposition (Matrix.diagonal_mul, Matrix.mul_diagonal, Matrix.transpose_apply) reduces the claim to arithmetic: entry (0, 1) gives (1/2)(1)(1/2) + (3/4)(-1) = -1/2 = (-1/2)(1), entry (1, 0) gives (1/2)(-1)(1/2) + (3/4)(1) = 1/2 = (-1/2)(-1), and all other entries vanish on both sides.

    lines 38–43

    uses op733_map_apply, Matrix.diagonal_mul, Matrix.mul_diagonal, Matrix.transpose_apply

    theorem op733_eigen : op733_map (Matrix.of ![![0, 1, 0], ![-1, 0, 0], ![0, 0, 0]]) =
        (-1/2 : ℂ) • Matrix.of ![![0, 1, 0], ![-1, 0, 0], ![0, 0, 0]] := by
      rw [op733_map_apply]
      ext i j
      fin_cases i <;> fin_cases j <;>
        simp [Matrix.diagonal_mul, Matrix.mul_diagonal, Matrix.transpose_apply] <;> norm_num
  5. The minimal real part of the spectrum of Φ, m(Φ), satisfies m(Φ) ≤ -1/2.

    Because V = E₁₂ - E₂₁ satisfies Φ(V) = (-1/2) V (op733_eigen, Module.End.mem_eigenspace_iff) and V ≠ 0 (since its (0, 1) entry is 1 ≠ 0, verified via congrFun), -1/2 is an eigenvalue of Φ and belongs to its spectrum (Module.End.HasEigenvalue.mem_spectrum, Module.End.hasEigenvalue_of_hasEigenvector). Because the spectrum of an endomorphism on a finite-dimensional space is finite (Module.End.finite_spectrum), its image under Re is finite and hence bounded below. By csInf_le, the infimum m(Φ) = inf Re(spectrum(Φ)) is bounded above by Re(-1/2) = -1/2.

    lines 45–58

    uses Module.End.HasEigenvalue.mem_spectrum, Module.End.hasEigenvalue_of_hasEigenvector, Module.End.mem_eigenspace_iff, op733_eigen, congrFun, Module.End.finite_spectrum, csInf_le

    theorem op733_min_re_le : spectral_min_re 3 op733_map ≤ -1/2 := by
      have hmem : (-1/2 : ℂ) ∈ spectrum ℂ op733_map := by
        apply Module.End.HasEigenvalue.mem_spectrum
        apply Module.End.hasEigenvalue_of_hasEigenvector (x := Matrix.of ![![0, 1, 0], ![-1, 0, 0], ![0, 0, 0]])
        refine ⟨?_, ?_⟩
        · rw [Module.End.mem_eigenspace_iff]; exact op733_eigen
        · intro h
          have := congrFun (congrFun h 0) 1
          simp at this
      unfold spectral_min_re
      have hfin : (Complex.re '' spectrum ℂ op733_map).Finite :=
        (Module.End.finite_spectrum op733_map).image _
      have := csInf_le hfin.bddBelow ⟨_, hmem, rfl⟩
      simpa using this
  6. The trace of Φ as an operator on M₃(ℂ) equals 11/2.

    By LinearMap.trace_eq_matrix_trace, the trace of Φ equals the trace of its 9 × 9 representation matrix with respect to the standard basis of M₃(ℂ). Expanding this trace using Matrix.trace, Fintype.sum_prod_type, Fin.sum_univ_three, and LinearMap.toMatrix_apply, each basis matrix Eᵢⱼ (Matrix.stdBasis_eq_single) is mapped under Φ (op733_map_apply) and its (i, j) coordinate is extracted (op733_repr_a, Matrix.diagonal_mul, Matrix.mul_diagonal, Matrix.single_apply). Evaluating the resulting sum of 9 diagonal coefficients via norm_num yields 11/2.

    lines 63–69

    uses LinearMap.trace_eq_matrix_trace, Matrix.trace, Fintype.sum_prod_type, Fin.sum_univ_three, LinearMap.toMatrix_apply, op733_map_apply, Matrix.stdBasis_eq_single, Matrix.diagonal_mul, Matrix.mul_diagonal, op733_repr_a, Matrix.single_apply

    theorem op733_trace : (LinearMap.trace ℂ (Matrix (Fin 3) (Fin 3) ℂ)) op733_map = 11/2 := by
      classical
      rw [LinearMap.trace_eq_matrix_trace ℂ (Matrix.stdBasis ℂ (Fin 3) (Fin 3))]
      simp [Matrix.trace, Fintype.sum_prod_type, Fin.sum_univ_three, LinearMap.toMatrix_apply,
        op733_map_apply, Matrix.stdBasis_eq_single, Matrix.diagonal_mul, Matrix.mul_diagonal, op733_repr_a,
        Matrix.single_apply]
      norm_num
  7. The universal spectral bound Re(Tr(Φ)) ≤ 3 m(Φ) + 6 fails for positive trace-preserving maps on 3 × 3 matrices.

    Assuming for contradiction (intro h) that the bound holds universally, we instantiate it with op733_map using its positivity (op733_pos) and trace preservation (op733_tp). Substituting the operator trace Tr(Φ) = 11/2 (op733_trace) yields 11/2 ≤ 3 m(Φ) + 6. Together with m(Φ) ≤ -1/2 (op733_min_re_le), this would require 11/2 ≤ 3(-1/2) + 6 = 9/2, a contradiction closed by linarith.

    lines 75–80

    uses op733_pos, op733_tp, op733_trace, op733_min_re_le

      intro h
      have h1 := h op733_map op733_pos op733_tp
      rw [op733_trace] at h1
      have h2 := op733_min_re_le
      norm_num at h1
      linarith
Matrix.mul_add
Matrix multiplication distributes over addition on the right: L(M + N) = LM + LN.
Matrix.add_mul
Matrix multiplication distributes over addition on the left: (L + M)N = LN + MN.
Matrix.transpose_add
The transpose of a sum of matrices is the sum of their transposes: (M + N)ᵀ = Mᵀ + Nᵀ.
smul_add
Scalar multiplication distributes over addition: a • (b₁ + b₂) = a • b₁ + a • b₂.
Matrix.mul_smul
Scalar multiplication commutes with matrix multiplication on the right: M(a • N) = a • (MN).
Matrix.smul_mul
Scalar multiplication commutes with matrix multiplication on the left: (a • M)N = a • (MN).
Matrix.transpose_smul
Transposition commutes with scalar multiplication: (c • M)ᵀ = c • Mᵀ.
RingHom.id_apply
The identity ring homomorphism maps any element to itself: id(x) = x.
smul_comm
Scalar multiplications commute: m • (n • a) = n • (m • a).
op733_map_apply
The definition of op733_map: Φ(X) = K X K + (3/4) P Xᵀ P with K = diag(1/2, 1/2, 1) and P = diag(1, 1, 0).
Matrix.diagonal_conjTranspose
The conjugate transpose of a diagonal matrix is the diagonal matrix of the complex conjugates of its diagonal entries.
Matrix.PosSemidef.add
The sum of two positive semidefinite matrices is positive semidefinite.
Matrix.PosSemidef.smul
Multiplying a positive semidefinite matrix by a non-negative scalar yields a positive semidefinite matrix.
Complex.zero_le_real.mpr
A real number is non-negative if and only if its coercion into ℂ is non-negative in the complex order.
Matrix.trace
The trace of a square matrix is the sum of its diagonal entries: Tr(M) = ∑ᵢ Mᵢᵢ.
Fin.sum_univ_three
A sum over Fin 3 expands to the sum of the values at 0, 1, and 2: ∑ᵢ f(i) = f(0) + f(1) + f(2).
Matrix.diagonal_mul
Multiplying a matrix on the left by a diagonal matrix scales row i by dᵢ: (diag(d) * M)ᵢⱼ = dᵢ * Mᵢⱼ.
Matrix.mul_diagonal
Multiplying a matrix on the right by a diagonal matrix scales column j by dⱼ: (M * diag(d))ᵢⱼ = Mᵢⱼ * dⱼ.
Matrix.transpose_apply
The (i, j) entry of the transpose of M is Mⱼᵢ.
Module.End.HasEigenvalue.mem_spectrum
Any eigenvalue of an endomorphism belongs to its spectrum.
Module.End.hasEigenvalue_of_hasEigenvector
An endomorphism has eigenvalue μ if there exists a non-zero eigenvector for μ.
Module.End.mem_eigenspace_iff
A vector x is an eigenvector for eigenvalue μ if and only if f(x) = μ • x.
op733_eigen
The skew-symmetric matrix E₁₂ - E₂₁ is an eigenvector of op733_map with eigenvalue -1/2.
congrFun
If two functions are equal, their values are equal at every argument.
Module.End.finite_spectrum
The spectrum of a linear endomorphism on a finite-dimensional vector space is finite.
csInf_le
If a set s is bounded below and a ∈ s, then the infimum of s satisfies inf s ≤ a.
LinearMap.trace_eq_matrix_trace
The trace of a linear endomorphism equals the matrix trace of its representation matrix with respect to any basis.
Fintype.sum_prod_type
A finite sum over pairs (x, y) equals the iterated sum over x and y.
LinearMap.toMatrix_apply
The (i, j) entry of the matrix representing a linear map in bases v₁, v₂ is the i-th coordinate of f(v₁(j)) in basis v₂.
Matrix.stdBasis_eq_single
The (i, j)-th element of the standard matrix basis is the single-entry elementary matrix Eᵢⱼ.
op733_repr_a
The coordinate of a matrix M at index (i, j) with respect to the standard basis is Mᵢⱼ.
Matrix.single_apply
The elementary matrix with value c at (i, j) evaluates to c at (i, j) and 0 elsewhere.
op733_pos
op733_map is a positive linear map: it preserves positive semidefiniteness of matrices.
op733_tp
op733_map is a trace-preserving linear map: Tr(Φ(X)) = Tr(X) for all matrices X.
op733_trace
The trace of op733_map as an operator on 3 × 3 complex matrices is 11/2.
op733_min_re_le
The minimal real part of the eigenvalues of op733_map is at most -1/2.

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

/-- The map X ↦ K X K + (3/4) P Xᵀ P on 3 × 3 complex matrices, K = diag(1/2, 1/2, 1), P = diag(1, 1, 0). -/noncomputable def op733_map : Matrix (Fin 3) (Fin 3) ℂ →ₗ[ℂ] Matrix (Fin 3) (Fin 3) ℂ where  toFun X := Matrix.diagonal ![1/2, 1/2, 1] * X * Matrix.diagonal ![1/2, 1/2, 1] +    (3/4 : ℂ) • (Matrix.diagonal ![1, 1, 0] * X.transpose * Matrix.diagonal ![1, 1, 0])  map_add' X Y := by    simp only [Matrix.mul_add, Matrix.add_mul, Matrix.transpose_add, smul_add]; abel  map_smul' c X := by    simp only [Matrix.mul_smul, Matrix.smul_mul, Matrix.transpose_smul, smul_add, RingHom.id_apply]    rw [smul_comm c (3/4 : ℂ)]theorem op733_map_apply (X : Matrix (Fin 3) (Fin 3) ℂ) : op733_map X =    Matrix.diagonal ![1/2, 1/2, 1] * X * Matrix.diagonal ![1/2, 1/2, 1] +    (3/4 : ℂ) • (Matrix.diagonal ![1, 1, 0] * X.transpose * Matrix.diagonal ![1, 1, 0]) := rflopen scoped ComplexOrder intheorem op733_pos : is_positive_linear_map 3 op733_map := by  classical  intro X hX  rw [op733_map_apply]  have hK : Matrix.conjTranspose (Matrix.diagonal ![(1/2 : ℂ), 1/2, 1]) = Matrix.diagonal ![1/2, 1/2, 1] := by    rw [Matrix.diagonal_conjTranspose]; congr 1; ext i; fin_cases i <;> simp  have hP : Matrix.conjTranspose (Matrix.diagonal ![(1 : ℂ), 1, 0]) = Matrix.diagonal ![1, 1, 0] := by    rw [Matrix.diagonal_conjTranspose]; congr 1; ext i; fin_cases i <;> simp  refine Matrix.PosSemidef.add ?_ (Matrix.PosSemidef.smul ?_ ?_)  · have := hX.mul_mul_conjTranspose_same (Matrix.diagonal ![(1/2 : ℂ), 1/2, 1])    rwa [hK] at this  · have := hX.transpose.mul_mul_conjTranspose_same (Matrix.diagonal ![(1 : ℂ), 1, 0])    rwa [hP] at this  · rw [show (3/4 : ℂ) = ((3/4 : ℝ) : ℂ) by push_cast; ring]    exact Complex.zero_le_real.mpr (by norm_num)theorem op733_tp : is_trace_preserving_linear_map 3 op733_map := by  intro X  rw [op733_map_apply]  simp [Matrix.trace, Fin.sum_univ_three, Matrix.diagonal_mul, Matrix.mul_diagonal]  ringtheorem op733_eigen : op733_map (Matrix.of ![![0, 1, 0], ![-1, 0, 0], ![0, 0, 0]]) =    (-1/2 : ℂ) • Matrix.of ![![0, 1, 0], ![-1, 0, 0], ![0, 0, 0]] := by  rw [op733_map_apply]  ext i j  fin_cases i <;> fin_cases j <;>    simp [Matrix.diagonal_mul, Matrix.mul_diagonal, Matrix.transpose_apply] <;> norm_numtheorem op733_min_re_le : spectral_min_re 3 op733_map ≤ -1/2 := by  have hmem : (-1/2 : ℂ) ∈ spectrum ℂ op733_map := by    apply Module.End.HasEigenvalue.mem_spectrum    apply Module.End.hasEigenvalue_of_hasEigenvector (x := Matrix.of ![![0, 1, 0], ![-1, 0, 0], ![0, 0, 0]])    refine ⟨?_, ?_⟩    · rw [Module.End.mem_eigenspace_iff]; exact op733_eigen    · intro h      have := congrFun (congrFun h 0) 1      simp at this  unfold spectral_min_re  have hfin : (Complex.re '' spectrum ℂ op733_map).Finite :=    (Module.End.finite_spectrum op733_map).image _  have := csInf_le hfin.bddBelow ⟨_, hmem, rfl⟩  simpa using thistheorem op733_repr_a (M : Matrix (Fin 3) (Fin 3) ℂ) (i j : Fin 3) :    (Matrix.stdBasis ℂ (Fin 3) (Fin 3)).repr M (i, j) = M i j := rfltheorem op733_trace : (LinearMap.trace ℂ (Matrix (Fin 3) (Fin 3) ℂ)) op733_map = 11/2 := by  classical  rw [LinearMap.trace_eq_matrix_trace ℂ (Matrix.stdBasis ℂ (Fin 3) (Fin 3))]  simp [Matrix.trace, Fintype.sum_prod_type, Fin.sum_univ_three, LinearMap.toMatrix_apply,    op733_map_apply, Matrix.stdBasis_eq_single, Matrix.diagonal_mul, Matrix.mul_diagonal, op733_repr_a,    Matrix.single_apply]  norm_numtheorem not_universal_spectral_bound_positive_tp_dim_three :    ¬ ∀ (Φ : Matrix (Fin 3) (Fin 3) ℂ →ₗ[ℂ] Matrix (Fin 3) (Fin 3) ℂ),      is_positive_linear_map 3 Φ → is_trace_preserving_linear_map 3 Φ →      ((LinearMap.trace ℂ (Matrix (Fin 3) (Fin 3) ℂ)) Φ).re ≤ 3 * spectral_min_re 3 Φ + 6 := by  intro h  have h1 := h op733_map op733_pos op733_tp  rw [op733_trace] at h1  have h2 := op733_min_re_le  norm_num at h1  linarith
See it in the graphDiscuss this resultChallenge it