Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 4 — IVI↓opt,VIVI_{\downarrow opt,V}IVI↓opt,V​ and IVI↑pes,VIVI_{\uparrow pes,V}IVI↑pes,V​ as maxima over ρV(p)\rho_V(p)ρV​(p) and σV(p)\sigma_V(p)σV​(p)

Proved
BoundedParamMDP.IntervalEval.lemma4_restricted_max

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

dynamic-programmingmarkov-decision-processesp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1robust-mdp

Let M↕M_\updownarrowM↕​ be a bounded-parameter MDP with finite state set QQQ and finite nonempty action set AAA, and let ρV,σV\rho_V,\sigma_VρV​,σV​, IVI↓opt,VIVI_{\downarrow opt,V}IVI↓opt,V​ and IVI↑pes,VIVI_{\uparrow pes,V}IVI↑pes,V​ be as in the definitions of interval value iteration.

Lemma 4. For any value functions V,V′V,V'V,V′ and state ppp,

IVI↓opt,V(V′)(p)=max⁡α∈ρV(p) min⁡M∈M↕VIM,α(V′)(p),IVI↑pes,V(V′)(p)=max⁡α∈σV(p) max⁡M∈M↕VIM,α(V′)(p).IVI_{\downarrow opt,V}(V')(p)=\max_{\alpha\in\rho_V(p)}\ \min_{M\in M_\updownarrow}VI_{M,\alpha}(V')(p),\qquad IVI_{\uparrow pes,V}(V')(p)=\max_{\alpha\in\sigma_V(p)}\ \max_{M\in M_\updownarrow}VI_{M,\alpha}(V')(p).IVI↓opt,V​(V′)(p)=α∈ρV​(p)max​ M∈M↕​min​VIM,α​(V′)(p),IVI↑pes,V​(V′)(p)=α∈σV​(p)max​ M∈M↕​max​VIM,α​(V′)(p).

It isolates how the action choice of IVI↕optIVI_{\updownarrow opt}IVI↕opt​ (respectively IVI↕pesIVI_{\updownarrow pes}IVI↕pes​) is driven by the fixed bound VVV, and is the form used to show that these operators contract.

Formalization Note. The paper prints the inner operator of the second line as min⁡M∈M↕\min_{M\in M_\updownarrow}minM∈M↕​​. Since IVI↑pes,VIVI_{\uparrow pes,V}IVI↑pes,V​ is an upper bound, eq. (29) and the proof of Theorem 13(b) (p. 47) make it max⁡M∈M↕\max_{M\in M_\updownarrow}maxM∈M↕​​, which is what is stated here. Each "max⁡α∈S\max_{\alpha\in S}maxα∈S​" is stated as: the value is the greatest element of the set {f(α):α∈S}\{f(\alpha):\alpha\in S\}{f(α):α∈S}.

Preamble
import Mathlib
import Definitions.Def_BoundedParamMDP_Optimal_MDP
import Definitions.Def_BoundedParamMDP_Optimal_BMDP
import Definitions.Def_BoundedParamMDP_IntervalEval_IntervalValueIteration
Formal statement
namespace BoundedParamMDP.IntervalEval

/-- Lemma 4 (Givan–Leach–Dean 2000, p. 25, eq. (32)), with the misprinted `min` of the second
line read as `max`: for any value functions `V, V'` and state `p`,
`IVI↓_{opt,V}(V')(p) = max_{α ∈ ρ_V(p)} min_{M ∈ M↕} VI_{M,α}(V')(p)` and
`IVI↑_{pes,V}(V')(p) = max_{α ∈ σ_V(p)} max_{M ∈ M↕} VI_{M,α}(V')(p)`; each maximum over a set
of actions is stated as "the greatest element of the set of values". -/
theorem lemma4_restricted_max {Q A : Type*} [Fintype Q] [Fintype A] [Nonempty A]
    (B : BoundedParamMDP.Optimal.BMDP Q A) (V V' : Q → ℝ) (p : Q) :
    IsGreatest ((fun α => ⨅ M : BoundedParamMDP.Optimal.Member B, BoundedParamMDP.Optimal.VIact M.1 α V' p) '' rho B V p)
      (IVIloOptV B V V' p) ∧
    IsGreatest ((fun α => ⨆ M : BoundedParamMDP.Optimal.Member B, BoundedParamMDP.Optimal.VIact M.1 α V' p) '' sigma B V p)
      (IVIhiPesV B V V' p) := by sorry

end BoundedParamMDP.IntervalEval
Source
Givan, Leach, Dean, Bounded-parameter Markov decision processes, Artificial Intelligence (2000), DOI 10.1016/S0004-3702(00)00047-3, manuscript of May 22, 2000, p. 25, Lemma 4, eq. (32) (second line misprinted min corrected to max)
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 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