Theorem 7.3 — weak interactions around a Gaussian reference: explicit bound (7.19) on , and
OpenQDSGibbs.Supplement.theorem_7_3Let . For a finite and the Fermi–Hubbard Hamiltonian (7.12), consider the fermionic block channels (7.18) on all nonempty labelled cubes of side , with rates , and their fermionic DS constant (7.11). For every , and integer satisfying
one has
where the last factor is at .
In particular, fix and finite bounds , set , let be an integer with
and let satisfy (7.20) with , . Then for the channels at time whenever , and .
The three losses in (7.19) are the bulk relaxation, the Gaussian boundary contribution and the interaction perturbation; at the bound holds at every finite temperature.
Formalization Note The fermionic influence and transport norm are those of (7.10) and (7.7), built on the published Bakshi–Moitra–Tan fermionic partial trace and transport plans. , , are tMaxL, uMax, omegaMax, and the convention is uLoss.
Differences from the source:
- encoding: Fock space H_Λ of dimension 4^{|Λ|} with the CAR embedding and a fixed Majorana ordering → matrices on Fin M → Fin 2 (M = 2|Λ| modes) with Jordan–Wigner c from the published GaussTesting.annihilator, for every mode ordering ord : Λ × Fin 2 ≃ Fin M; reason: the same algebra; changing the ordering conjugates by a unitary and leaves every quantity invariant (p.64), so the statement for every ordering is the statement for the fixed one
- encoding: κ(H) = min_j Σ_B ℓ^{−D}[1 − D^F_{Φ_B}(j)] ≥ x → FermiKappaGe: Σ_B ℓ^{−D} D^F_{Φ_B}(j) ≤ Σ_B ℓ^{−D} − x in ℝ≥0∞ for every site j; reason: the same inequality, since Σ_B ℓ^{−D} ≥ 1 ≥ x (each site lies in ℓ^D cubes)
- encoding: finite Λ ⊂ ℤ^D → a nonempty Finset Λ; reason: excludes only the empty lattice, where the minimum over sites is undefined
- encoding: βU_max[1 + log(4/(βU_max))]^{4D}, interpreted as 0 at U_max = 0 → uLoss D (β·U_max), defined by cases (0 at argument 0); reason: the source's convention, avoiding Lean's log 0 = 0
- reading: first set T0, then choose ℓ with (7.21), finally choose U* > 0 so that (7.20) holds at t_max = t*, U_max = U* → for every integer ℓ ≥ 1 satisfying (7.21) and every U* > 0 satisfying (7.20) at (t*, U*); reason: the source's choices are arbitrary among those satisfying the stated conditions, so the conclusion is claimed for each of them
import Mathlib import Definitions.Def_QDSGibbs_Supplement_fermiBlockChannel import Definitions.Def_QDSGibbs_Supplement_fermiInfluence open scoped Matrix ENNReal
namespace QDSGibbs.Supplement
/-- **Theorem 7.3 (Weak interactions around a Gaussian reference)** (Rouzé–Stilck França, *Rapid
mixing of Gibbs samplers via quantum Dobrushin–Shlosman conditions*, arXiv:2610.03578, held file
sha256 ede55815…, Theorem 7.3, p.67). For every `D ≥ 1`, the fermionic block channels (7.18) of the
Fermi–Hubbard Hamiltonian (7.12) on the labelled cubes of side `ℓ` with rates `ℓ^{−D}` satisfy
`κ(H) ≥ 1 − 4e^{−T/2} − (T/(5ℓ)) [32D(1 + 2Dβt_max)]^{16D²} (1 + βω_max)^{3D+1} e^{(6D²+2D)βt_max+T}
− 2^{72D+20} D^{12D+1} T (ℓ+2)^{4D²+5D−1} (1 + βt_max)^{8D+1} e^{4Dβt_max} · βU_max [1 + log(4/(βU_max))]^{4D}`
(7.19) for every `β > 0`, `T ≥ 0` and integer `ℓ ≥ 1` satisfying
`βU_max (ℓ+2)^{8D²+10D−2} (1 + βt_max)^{16D+2} e^{8Dβt_max} ≤ 1/(9 · 2^{176D+46} D^{32D+2})` (7.20), the
`U_max`-factor being `0` at `U_max = 0`. In particular, for finite bounds `t_*`, `ω_*`, with
`T_0 = 2 log 16`, every integer `ℓ ≥ 1` satisfying (7.21) and every `U_* > 0` satisfying (7.20) with
`t_max = t_*`, `U_max = U_*` give `κ(H) ≥ 1/4` whenever `t_max ≤ t_*`, `ω_max ≤ ω_*`, `U_max ≤ U_*`.
**Formalization Note**
* The lattice `Λ ⊂ ℤ^D` is a nonempty `Finset`; `ord : Λ × Fin 2 ≃ Fin M` is the Jordan–Wigner mode
ordering (any ordering); `H = fhLattice Λ ord t U μ` (the published `FermiHubbardMix.fhHam` on the
nearest-neighbour graph). The blocks are all nonempty cubes `cube Λ ℓ z`, `z ∈ cubeLabels Λ ℓ`.
* `κ(H) ≥ x` is `FermiKappaGe` with rate `ℓ^{−D}` (no subtraction under the sum; see its docstring),
using the fermionic influence (7.10) and transport norm (7.7).
* The `U_max = 0` convention is `uLoss`. `t_max`, `U_max`, `ω_max` are `tMaxL`, `uMax`, `omegaMax`. -/
theorem theorem_7_3 (D : ℕ) (hD : 1 ≤ D) :
(∀ (Λ : Finset (Fin D → ℤ)), Λ.Nonempty → ∀ (M : ℕ) (ord : Λ × Fin 2 ≃ Fin M)
(t : Sym2 Λ → ℝ) (U μ : Λ → ℝ) (β T : ℝ) (ℓ : ℕ), 0 < β → 0 ≤ T → 1 ≤ ℓ →
β * uMax U * ((ℓ : ℝ) + 2) ^ (8 * D ^ 2 + 10 * D - 2) *
(1 + β * tMaxL Λ t) ^ (16 * D + 2) * Real.exp (8 * (D : ℝ) * β * tMaxL Λ t) ≤
1 / (9 * (2 : ℝ) ^ (176 * D + 46) * (D : ℝ) ^ (32 * D + 2)) →
FermiKappaGe ord (cubeLabels Λ ℓ)
(fun z => fermiBlockChannel β ord (fhLattice Λ ord t U μ) (cube Λ ℓ z) T)
(((ℓ : ℝ) ^ D)⁻¹)
(1 - 4 * Real.exp (-T / 2)
- T / (5 * (ℓ : ℝ)) * (32 * (D : ℝ) * (1 + 2 * (D : ℝ) * β * tMaxL Λ t)) ^ (16 * D ^ 2) *
(1 + β * omegaMax U μ) ^ (3 * D + 1) *
Real.exp ((6 * (D : ℝ) ^ 2 + 2 * (D : ℝ)) * β * tMaxL Λ t + T)
- (2 : ℝ) ^ (72 * D + 20) * (D : ℝ) ^ (12 * D + 1) * T * ((ℓ : ℝ) + 2) ^ (4 * D ^ 2 + 5 * D - 1) *
(1 + β * tMaxL Λ t) ^ (8 * D + 1) * Real.exp (4 * (D : ℝ) * β * tMaxL Λ t) *
uLoss D (β * uMax U))) ∧
(∀ (β tstar ωstar : ℝ), 0 < β → 0 ≤ tstar → 0 ≤ ωstar →
∀ ℓ : ℕ, 1 ≤ ℓ →
4 * (2 * Real.log 16) / 5 * (32 * (D : ℝ) * (1 + 2 * (D : ℝ) * β * tstar)) ^ (16 * D ^ 2) *
(1 + β * ωstar) ^ (3 * D + 1) *
Real.exp ((6 * (D : ℝ) ^ 2 + 2 * (D : ℝ)) * β * tstar + 2 * Real.log 16) ≤ (ℓ : ℝ) →
∀ Ustar : ℝ, 0 < Ustar →
β * Ustar * ((ℓ : ℝ) + 2) ^ (8 * D ^ 2 + 10 * D - 2) * (1 + β * tstar) ^ (16 * D + 2) *
Real.exp (8 * (D : ℝ) * β * tstar) ≤
1 / (9 * (2 : ℝ) ^ (176 * D + 46) * (D : ℝ) ^ (32 * D + 2)) →
∀ (Λ : Finset (Fin D → ℤ)), Λ.Nonempty → ∀ (M : ℕ) (ord : Λ × Fin 2 ≃ Fin M)
(t : Sym2 Λ → ℝ) (U μ : Λ → ℝ),
tMaxL Λ t ≤ tstar → omegaMax U μ ≤ ωstar → uMax U ≤ Ustar →
FermiKappaGe ord (cubeLabels Λ ℓ)
(fun z => fermiBlockChannel β ord (fhLattice Λ ord t U μ) (cube Λ ℓ z) (2 * Real.log 16))
(((ℓ : ℝ) ^ D)⁻¹) (1 / 4)) := by sorry
end QDSGibbs.Supplement