Spectral projector and reduced resolvent at a gapped simple eigenvalue
ProvedQAlgorithms.Nannicini.gapped_spectral_dataLet be a Hermitian matrix on and let be an eigenvalue of of multiplicity one, with every other eigenvalue satisfying . Then there are linear operators and on (the spectral projector and the reduced resolvent ) such that
and for some unit vector .
Preamble
import Definitions.Def_QAlgorithms_Defs_Adiabatic set_option autoImplicit false
Formal statement
namespace QAlgorithms.Nannicini
/-- Spectral data at a simple isolated eigenvalue. If `A` is Hermitian and `λ` is a simple
eigenvalue of `A` separated from the rest of the spectrum by `γ > 0` (`HasGapAround A λ γ`), there are
the orthogonal projector `P` onto the `λ`-eigenspace and the reduced resolvent
`R = (A − λ)⁻¹(1 − P)`: `P` is a self-adjoint idempotent commuting with `A` with range in the
eigenspace, `(A − λ)R = R(A − λ) = 1 − P`, `RP = PR = 0`, `‖P‖ ≤ 1`, `‖R‖ ≤ 1/γ`, and `P` fixes a unit
vector. -/
theorem gapped_spectral_data {ι : Type*} [Fintype ι] [DecidableEq ι] (A : Matrix ι ι ℂ) (lam γ : ℝ)
(hγ : 0 < γ) (hA : HasGapAround A lam γ) :
∃ P R : EuclideanSpace ℂ ι →L[ℂ] EuclideanSpace ℂ ι,
(∀ x, P (P x) = P x) ∧
(∀ x y, inner ℂ (P x) y = inner ℂ x (P y)) ∧
(∀ x, act A (P x) = (lam : ℂ) • P x) ∧
(∀ x, P (act A x) = (lam : ℂ) • P x) ∧
(∀ x, act A (R x) - (lam : ℂ) • R x = x - P x) ∧
(∀ x, R (act A x) - (lam : ℂ) • R x = x - P x) ∧
(∀ x, R (P x) = 0) ∧ (∀ x, P (R x) = 0) ∧
‖P‖ ≤ 1 ∧ ‖R‖ ≤ 1 / γ ∧ (∃ x, ‖x‖ = 1 ∧ P x = x) := by
sorry
end QAlgorithms.Nannicini
Source
Giacomo Nannicini, Quantum algorithms for optimizers, 2025, v5, §9.2 (proof of Corollary 9.24); cf. T. Kato, Perturbation Theory for Linear Operators (1966), Ch. II §5, and Jansen–Ruskai–Seiler, J. Math. Phys. 48 (2007) 102111. https://arxiv.org/pdf/2408.07086v5