Kernel rms bound
ProvedDerivativeChainPhaseLock.kernel_rms_boundFor each natural number , write , with . Let , , , , , , , . Assume
Then
The named functions DerivativeChainFiniteHistory.FiniteBalancedHistoryLaw, DerivativeChainFiniteHistory.FiniteBalancedHistoryLaw.endpointVariance, DerivativeChainFiniteHistory.FiniteBalancedHistoryLaw.meanSquareCost, DerivativeChainFiniteHistory.FiniteBalancedHistoryLaw.paths, DerivativeChainPhaseLock.parentSquareCost, DerivativeChainPhaseLock.parentVariance, DerivativeChainPrivate.fromRoots retain their exact imported definitions. Use the conditional barycenter identity to compute the exact edge-energy gain. Weighted Minkowski then bounds the root-mean-square cost after extension.
Formalization Note. Natural-number subtraction is truncated. Named functions, index embeddings and field projections use their exact imported definitions.
import Mathlib import Definitions.Def_DerivativeChainAnalytic_BarycentricExtension import Definitions.Def_DerivativeChainAnalytic_FiniteBalancedHistoryLaw import Definitions.Def_DerivativeChainPrivate_foundations set_option autoImplicit false namespace DerivativeChain end DerivativeChain namespace DerivativeChainContinuation end DerivativeChainContinuation namespace DerivativeChainFiniteHistory end DerivativeChainFiniteHistory namespace DerivativeChainPhaseLock end DerivativeChainPhaseLock namespace DerivativeChainPrefixLaws end DerivativeChainPrefixLaws namespace DerivativeChainPrivate end DerivativeChainPrivate set_option autoImplicit false set_option backward.isDefEq.respectTransparency false set_option maxHeartbeats 2000000 open scoped BigOperators open Polynomial DerivativeChainPrivate DerivativeChainFiniteHistory DerivativeChainContinuation universe v open DerivativeChainPhaseLock
theorem DerivativeChainPhaseLock.kernel_rms_bound {n : ℕ} (a : Fin (n+2) → ℂ)
{b : Fin (n+1) → ℂ} {u : ℂ} {κ : Type v} [Fintype κ]
(L : FiniteBalancedHistoryLaw b u κ) (α : κ → (fromRoots a).roots.toFinset → ℝ)
(hα : ∀ i j, 0 ≤ α i j) (hαsum : ∀ i, ∑ j, α i j=1)
(hαmean : ∀ i, ∑ j, (α i j:ℂ)*(j:ℂ)=L.paths i (Fin.last n)) :
Real.sqrt (parentSquareCost a L α) ≤ Real.sqrt L.meanSquareCost+
Real.sqrt (parentVariance a L α-L.endpointVariance) := by sorry