Regularity of the spectral projector and reduced resolvent along a gapped linear interpolation
ProvedQAlgorithms.Nannicini.linInterp_spectral_projector_regularThere is a universal constant with the following property. Let be Hermitian matrices on , let , and let
Suppose is continuous and, for every , is a simple eigenvalue of whose distance to the rest of the spectrum is at least . Let be the orthogonal projector onto the eigenline of and the reduced resolvent. Then and are differentiable on , is continuous, and
(operator norms), together with the algebraic identities , , , , , and every -eigenvector being fixed by .
This is the first-order perturbation theory of an isolated eigenvalue (Kato) specialised to a linear path; it is the spectral input for the adiabatic theorem along linear interpolations (Corollary 9.24).
Formalization note. are continuous -linear maps on EuclideanSpace ℂ ι, derivatives are HasDerivWithinAt on Set.Icc 0 1, the norm of is specNorm, and the spectral data are stated as the listed pointwise identities rather than by a definition of .
import Definitions.Def_QAlgorithms_Defs_Adiabatic set_option autoImplicit false
universe u
namespace QAlgorithms.Nannicini
/-- Kato's perturbation theory for the gapped linear interpolation `H(s) = (1 − s)H_I + sH_F`,
`D = H_F − H_I`. If `λ(s)` is a continuous simple eigenvalue of `H(s)` with gap `γ > 0` on `[0, 1]`,
then the spectral projector `P(s)` and the reduced resolvent `R(s) = (H(s) − λ(s))⁻¹(1 − P(s))`
are differentiable on `[0, 1]`, with `P′` continuous, Kato's formula `P′ = −R D P − P D R`, and
`‖P‖ ≤ 1`, `‖R‖ ≤ 1/γ`, `‖R′‖ ≤ K₀‖D‖/γ²` for a universal constant `K₀`. -/
theorem linInterp_spectral_projector_regular : ∃ K0 : ℝ, ∀ {ι : Type u} [Fintype ι]
[DecidableEq ι] (HI HF : Matrix ι ι ℂ), HI.IsHermitian → HF.IsHermitian →
∀ (lam : ℝ → ℝ), ContinuousOn lam (Set.Icc 0 1) →
∀ (γ : ℝ), 0 < γ →
(∀ s ∈ Set.Icc (0 : ℝ) 1, HasGapAround (linInterp HI HF s) (lam s) γ) →
∃ P P' R R' : ℝ → (EuclideanSpace ℂ ι →L[ℂ] EuclideanSpace ℂ ι),
ContinuousOn P' (Set.Icc 0 1) ∧
∀ s ∈ Set.Icc (0 : ℝ) 1,
HasDerivWithinAt P (P' s) (Set.Icc 0 1) s ∧
HasDerivWithinAt R (R' s) (Set.Icc 0 1) s ∧
(∀ x, P s (P s x) = P s x) ∧
(∀ x y, inner ℂ (P s x) y = inner ℂ x (P s y)) ∧
(∀ x, act (linInterp HI HF s) x = (lam s : ℂ) • x → P s x = x) ∧
(∀ x, act (linInterp HI HF s) (P s x) = (lam s : ℂ) • P s x) ∧
(∀ x, act (linInterp HI HF s) (R s x) - (lam s : ℂ) • R s x = x - P s x) ∧
(∀ x, R s (P s x) = 0) ∧ (∀ x, P s (R s x) = 0) ∧
(∀ x, P' s x = -R s (act (HF - HI) (P s x)) - P s (act (HF - HI) (R s x))) ∧
‖P s‖ ≤ 1 ∧ ‖R s‖ ≤ 1 / γ ∧ ‖R' s‖ ≤ K0 * specNorm (HF - HI) / γ ^ 2 := by
sorry
end QAlgorithms.Nannicini