(7.2), p. 3611 — the drift error splits as 4(R^{n,1}_s + R^{n,2}_s + R^{n,3}_s + R^{n,4}_s)
ProvedGraphonMF.SparseRate.eq_7_2graphonpropagation-of-chaosrate-of-convergencesparse-graphs
For every , every solution of (4.1) and every ,
where
This is the add-and-subtract decomposition (7.2) of the paper, with replaced by as is appropriate under Condition 4.3 (§7.3); the four differences sum to the drift error, and . and are the terms Rn2, Rn4 of the setting.
Throughout, is the continuum noise, , , satisfy Condition 4.1, is a graphon, the edges are independent for and independent of the noise (Condition 4.3), solves the graphon limit system (4.2) with , and is any solution of the not-so-dense system (4.1), , . Expectations are lower Lebesgue integrals in ; the solution concepts are those of the GraphonMF_SparseRate_Setting definitions.
Preamble
import Mathlib import Definitions.Def_GraphonMF_SparseRate_Setting open MeasureTheory ProbabilityTheory Filter Topology open scoped ENNReal NNReal
Formal statement
namespace GraphonMF.SparseRate
/-- (7.2) (p. 3611), with `G` in place of `Gₙ` as under Condition 4.3 (§7.3, p. 3616): for every `n`, every
solution `Xⁿ` of (4.1) and every `s ≤ T`, the drift error of (7.1) is at most `4(R^{n,1}_s + R^{n,2}_s + R^{n,3}_s + R^{n,4}_s)`. -/
theorem eq_7_2 {Ω : Type*} [MeasurableSpace Ω] {P : Measure Ω} {d : ℕ} {T : ℝ≥0} (hT : 0 < T)
{μ0 : I → Measure (Fin d → ℝ)} {X0 : I → Ω → Fin d → ℝ} {B : I → ℝ≥0 → Ω → Fin d → ℝ}
(hN : NoiseSetting P μ0 X0 B)
{b : (Fin d → ℝ) → (Fin d → ℝ) → (Fin d → ℝ)} {σ : (Fin d → ℝ) → Matrix (Fin d) (Fin d) ℝ}
{β : ℕ → ℝ} (h41 : Cond41 μ0 b σ β) {G : I → I → ℝ} (hG : GraphonMF.DenseRate.IsGraphon G)
{ξ : (n : ℕ) → Ω → Fin n → Fin n → ℝ} (h43 : Cond43 P X0 B ξ β G)
{X : I → ℝ≥0 → Ω → Fin d → ℝ} (hX : IsGraphonSolution42 P hN T b σ G X) :
∀ (n : ℕ) (Xn : Fin n → ℝ≥0 → Ω → Fin d → ℝ),
IsParticleSolution41 P hN h43.meas T b σ β n Xn → ∀ s : ℝ≥0, s ≤ T →
(1 / (n : ℝ≥0∞)) * ∑ i : Fin n, ∫⁻ ω,
‖(1 / ((n : ℝ) * β n)) • ∑ j : Fin n, ξ n ω i j • b (Xn i s ω) (Xn j s ω) -
driftAvg P b G X (GraphonMF.DenseRate.lab n i) s (X (GraphonMF.DenseRate.lab n i) s ω)‖ₑ ^ 2 ∂P ≤
4 * (((1 / (n : ℝ≥0∞)) * ∑ i : Fin n, ∫⁻ ω,
‖(1 / (n : ℝ)) • ∑ j : Fin n, (ξ n ω i j / β n) •
(b (Xn i s ω) (Xn j s ω) - b (X (GraphonMF.DenseRate.lab n i) s ω) (Xn j s ω))‖ₑ ^ 2 ∂P) + Rn2 P b β ξ X n Xn s + ((1 / (n : ℝ≥0∞)) * ∑ i : Fin n, ∫⁻ ω,
‖(1 / (n : ℝ)) • ∑ j : Fin n, ((ξ n ω i j / β n) • b (X (GraphonMF.DenseRate.lab n i) s ω) (X (GraphonMF.DenseRate.lab n j) s ω) -
G (GraphonMF.DenseRate.lab n i) (GraphonMF.DenseRate.lab n j) • ∫ x, b (X (GraphonMF.DenseRate.lab n i) s ω) x ∂(P.map (X (GraphonMF.DenseRate.lab n j) s)))‖ₑ ^ 2 ∂P) + Rn4 P b G X n s) := by sorry
end GraphonMF.SparseRate
Source
Bayraktar, Chakraborty, Wu, Graphon mean field systems, Ann. Appl. Probab. 33(5) (2023), p. 3611, (7.2) (arXiv:2003.13180 v3, p. 26); with G in place of G_n under Condition 4.3: p. 3616, §7.3