Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4 — under monotonicity and N-stage contraction, f is the unique fixed point of E and of A, and ρ(A^N v, f) ≤ cρ(v, f) for v ≤ f

Proved
DenardoDP.NStage.theorem4

by mikedeng1 · Oct 5, 2026 · Mathlib 0df444a (Lean v4.33.1)

contraction-mappingdynamic-programmingmonotone-operatorp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let Ω\OmegaΩ be a set, DxD_xDx​ a decision set at each x∈Ωx\in\Omegax∈Ω, Δ=×xDx\Delta=\times_xD_xΔ=×x​Dx​ the policies, VVV the bounded real functions on Ω\OmegaΩ with the sup metric ρ\rhoρ, h(x,dx,v)h(x,d_x,v)h(x,dx​,v) the return, HδH_\deltaHδ​ the policy operators [Hδv](x)=h(x,δx,v)[H_\delta v](x)=h(x,\delta_x,v)[Hδ​v](x)=h(x,δx​,v) and AAA the maximization operator (Av)(x)=sup⁡dx∈Dxh(x,dx,v)(Av)(x)=\sup_{d_x\in D_x}h(x,d_x,v)(Av)(x)=supdx​∈Dx​​h(x,dx​,v), both mapping VVV into VVV. Suppose:

  • the monotonicity assumption: u≥vu\ge vu≥v implies Hδu≥HδvH_\delta u\ge H_\delta vHδ​u≥Hδ​v for every δ\deltaδ;
  • the N-stage contraction assumption: for a positive integer NNN and c<1c<1c<1, independent of δ\deltaδ, each HδNH_\delta^NHδN​ has modulus ccc or less and each HδH_\deltaHδ​ has modulus 111 or less.

Let vδv_\deltavδ​ be the fixed point of HδNH_\delta^NHδN​, let EEE be the operator (Ev)(x)=sup⁡δ(HδNv)(x)(Ev)(x)=\sup_\delta(H_\delta^Nv)(x)(Ev)(x)=supδ​(HδN​v)(x), and let f(x)=sup⁡δvδ(x)f(x)=\sup_\delta v_\delta(x)f(x)=supδ​vδ​(x) be the optimal return. Then:

  1. (a) vδv_\deltavδ​ is the unique fixed point of HδH_\deltaHδ​;
  2. (b) ρ(vδ,v)≤ρ(Hδv,v) N/(1−c)\rho(v_\delta,v)\le\rho(H_\delta v,v)\,N/(1-c)ρ(vδ​,v)≤ρ(Hδ​v,v)N/(1−c) for every v∈Vv\in Vv∈V;
  3. (c) EEE is a contraction mapping of modulus ccc or less;
  4. (d) f∈Vf\in Vf∈V, and fff is the unique fixed point of EEE and the unique fixed point of AAA:
Ef=f,Af=f;Ef=f,\qquad Af=f;Ef=f,Af=f;
  1. (e) if v≤fv\le fv≤f, then
ρ(ANv,f)≤c ρ(v,f).\rho(A^Nv,f)\le c\,\rho(v,f).ρ(ANv,f)≤cρ(v,f).

This is the main result of §5: even when no single HδH_\deltaHδ​ is a contraction, the optimality equation v=Avv=Avv=Av has exactly one bounded solution, and it is the optimal return. Part (e) gives geometric convergence of successive approximations AkNv→fA^{kN}v\to fAkNv→f from below.

Formalization Note The family vvv is supplied with HδNvδ=vδH_\delta^Nv_\delta=v_\deltaHδN​vδ​=vδ​ (the paper's definition of vδv_\deltavδ​ in §5; such a family exists and is unique because HδNH_\delta^NHδN​ is a contraction on the complete space VVV); part (a) is a conclusion. EEE is taken as an operator V→VV\to VV→V with IsNStageSupOperator H N E (its value at each point is the least upper bound of {(HδNv)(x)}δ\{(H_\delta^Nv)(x)\}_\delta{(HδN​v)(x)}δ​); such an operator exists, which is the milestone Theorem 4 (a)–(c), and is then unique. fff is not defined as a fixed point: part (d) asserts that some f∈Vf\in Vf∈V is the pointwise least upper bound of {vδ(x)}δ\{v_\delta(x)\}_\delta{vδ​(x)}δ​ (which determines fff) and that this fff satisfies Ef=fEf=fEf=f, Af=fAf=fAf=f, is the only fixed point of either operator, and satisfies (e). u≤vu\le vu≤v is pointwise, and ANA^NAN is the NNN-fold composition.

Preamble
import Mathlib
import Definitions.Def_DenardoDP_NStage_Model
Formal statement
namespace DenardoDP.NStage

theorem theorem4 {Ω : Type*} {D : Ω → Type*}
    (h : (x : Ω) → D x → DenardoDP.Contraction.BFun Ω → ℝ) (H : ((x : Ω) → D x) → DenardoDP.Contraction.BFun Ω → DenardoDP.Contraction.BFun Ω)
    (A : DenardoDP.Contraction.BFun Ω → DenardoDP.Contraction.BFun Ω) (v : ((x : Ω) → D x) → DenardoDP.Contraction.BFun Ω) (N : ℕ) (c : ℝ)
    (hH : DenardoDP.Contraction.IsPolicyOperator h H) (hA : DenardoDP.Contraction.IsMaxOperator h A)
    (hmono : DenardoDP.Contraction.MonotonicityAssumption H) (hN : NStageContractionAssumption H N c)
    (hv : ∀ δ, (H δ)^[N] (v δ) = v δ)
    (E : DenardoDP.Contraction.BFun Ω → DenardoDP.Contraction.BFun Ω) (hE : IsNStageSupOperator H N E) :
    (∀ δ, H δ (v δ) = v δ ∧ ∀ w, H δ w = w → w = v δ) ∧
    (∀ δ w, dist (v δ) w ≤ dist (H δ w) w * N / (1 - c)) ∧
    DenardoDP.Contraction.ModulusLE E c ∧
    ∃ f : DenardoDP.Contraction.BFun Ω, DenardoDP.Contraction.IsOptimalReturn v f ∧
      E f = f ∧ A f = f ∧ (∀ w, E w = w → w = f) ∧ (∀ w, A w = w → w = f) ∧
      (∀ w, DenardoDP.Contraction.PLe w f → dist (A^[N] w) f ≤ c * dist w f) := by sorry

end DenardoDP.NStage
Source
Denardo, Contraction Mappings in the Theory Underlying Dynamic Programming, SIAM Review 9(2) (1967), p. 169, Theorem 4 (proof pp. 169–170)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me