Weyl monotonicity: for Hermitian the eigenvalues can be paired,
ProvedBosonicLearn.loewner_eigenvalue_pairingbosonic-gaussian-statesquantum-informationquantum-learning
Weyl's monotonicity theorem for eigenvalues, in a form with explicit enumerations.
Let be Hermitian complex matrices with . Then the eigenvalues of and of can be enumerated, with multiplicity, as and so that
Role. In the proof of Lemma 12 this pairs the mean photon numbers of the thermal modes of the two reduced states ("Since , Weyl's monotonicity theorem yields for every "), which is what makes the convolution coupling of the two photon-count distributions possible.
Formalization Note "Enumeration of the eigenvalues with multiplicity" is expressed through the consequences that are used: for all real (this determines the multiset), and . The hypotheses are Matrix.PosSemidef of , and .
Preamble
import Mathlib open scoped Matrix ComplexOrder
Formal statement
namespace BosonicLearn
/-- **Weyl monotonicity: eigenvalues of Loewner-ordered Hermitian matrices can be paired.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §5.7, proof of Lemma 12, after Eq. (136): "Since B₊ ≥ B₋, Weyl's monotonicity theorem yields λ⁺_i ≥ λ⁻_i for every i."
Let `0 ⪯ B₋ ⪯ B₊ ⪯ Λ·1` be Hermitian `n × n` matrices. Then there are enumerations `λ⁻_1, …, λ⁻_n` and
`λ⁺_1, …, λ⁺_n` of their eigenvalues (with multiplicity) such that `0 ≤ λ⁻_i ≤ λ⁺_i ≤ Λ` for every `i`.
**Formalization Note.** "Enumeration of the eigenvalues" is expressed by the consequences that are used:
`det(1 + tB_±) = ∏ᵢ (1 + tλ^±_i)` for all real `t`, `Tr B_± = ∑ᵢ λ^±_i` and `Tr B₋² = ∑ᵢ (λ⁻_i)²`. -/
theorem loewner_eigenvalue_pairing {n : ℕ} (Bm Bp : Matrix (Fin n) (Fin n) ℂ) (hm : Bm.PosSemidef)
(hd : (Bp - Bm).PosSemidef) (Λ : ℝ)
(hΛ : (((Λ : ℝ) : ℂ) • (1 : Matrix (Fin n) (Fin n) ℂ) - Bp).PosSemidef) :
∃ lp lm : Fin n → ℝ, (∀ i, 0 ≤ lm i) ∧ (∀ i, lm i ≤ lp i) ∧ (∀ i, lp i ≤ Λ) ∧
(∀ t : ℝ, (1 + ((t : ℝ) : ℂ) • Bp).det = ∏ i, ((1 + t * lp i : ℝ) : ℂ)) ∧
(∀ t : ℝ, (1 + ((t : ℝ) : ℂ) • Bm).det = ∏ i, ((1 + t * lm i : ℝ) : ℂ)) ∧
Bp.trace = ((∑ i, lp i : ℝ) : ℂ) ∧ Bm.trace = ((∑ i, lm i : ℝ) : ℂ) ∧
(Bm * Bm).trace = ((∑ i, lm i ^ 2 : ℝ) : ℂ) := by sorry
end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, §5.7 (Delayed proof for Lemma 12), proof of Lemma 12, after Eq. (136) (Weyl's monotonicity theorem)