Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 9.4.2 — the optimal execution rate is monotone and unit-Lipschitz in the inventory

Proved
MDPFinance.JumpMarkets.theorem_9_4_2

by Shuze Chen · Sep 27, 2026 · Mathlib 0df444a (Lean v4.33.1)

dynamic-programmingmathematical-financeoperations-researchoptimizationprobability

Theorem 9.4.2 (p. 296).

a) The value function V(t,x)V(t,x)V(t,x) is the unique fixed point of T\mathcal{T}T in IMcxIM_{cx}IMcx​. b) There exists an optimal Markov strategy π∗=(πt∗)\pi^* = (\pi^*_t)π∗=(πt∗​) such that πt∗=f∗(t,Xt−)\pi^*_t = f^*(t, X_{t-})πt∗​=f∗(t,Xt−​) and f∗f^*f∗ satisfies f∗(t,x)≤f∗(t,x+1)≤f∗(t,x)+1f^*(t,x) \le f^*(t,x+1) \le f^*(t,x) + 1f∗(t,x)≤f∗(t,x+1)≤f∗(t,x)+1, and (Xt)(X_t)(Xt​) is the corresponding number of share process.

The structural answer to the execution problem: sell more when you hold more, but never more than one extra share per extra share held. Both inequalities are the theorem.

f∗(t,x)≤f∗(t,x+1)f^*(t,x) \le f^*(t,x+1)f∗(t,x)≤f∗(t,x+1) alone would be a generic monotonicity claim. It is the upper bound f∗(t,x+1)≤f∗(t,x)+1f^*(t,x+1) \le f^*(t,x)+1f∗(t,x+1)≤f∗(t,x)+1 that carries the content — the optimal block size is unit-Lipschitz in the inventory, so the execution schedule cannot jump. Dropping it loses half the theorem, and it is the half that takes the work: the book proves it by contradiction from the strict convexity (9.19) of CCC.

IMcxIM_{cx}IMcx​, not IMcvIM_{cv}IMcv​. This is a minimisation with a convex cost and a convex value function, the opposite curvature from §9.3's terminal wealth problem; the two sets are not interchangeable.

f∗f^*f∗ is the smallest minimizer of C(u)+v(t,x−u)C(u) + v(t,x-u)C(u)+v(t,x−u) over u∈{0,…,x}u \in \{0,\dots,x\}u∈{0,…,x} (9.22), the book's own choice; with an arbitrary selection of minimizers the unit-Lipschitz bound need not hold.

Optimality is stated at the level of the discrete-time model, where T\mathcal{T}T lives: f∗f^*f∗ minimizes the operator's argument at the value function, so the stationary policy is optimal for the discrete-time model, and — by Theorem 8.3.2, which p. 295 invokes and which belongs to chunk 08 — that determines an optimal control process for the continuous-time problem. The inner optimisation is over a finite set, so no sSup/sInf junk-value device is needed here, unlike in §9.3.

Moderation note. The draft assumed V to be some fixed point of 𝒯 in IM_cx and concluded its uniqueness, never connecting it to the value function; b)'s optimal strategy was absent (only the minimizer property of f^*). Now a) is: the value function V of the embedded model lies in IM_cx, is a fixed point of 𝒯, and is the only one there; b) f^* = f^*_V is monotone and unit-Lipschitz in x, and there is an admissible measurable policy f(t,x)(s) = f^*(t+s,x) attaining V (the strategy π^*_t = f^*(t, X_{t-})).

Preamble
import Mathlib
import Definitions.Def_MDPFinance_JumpMarkets_TradeExecution

open MeasureTheory
open scoped ENNReal
Formal statement
namespace MDPFinance.JumpMarkets

/-- **Theorem 9.4.2** (p. 296). a) The value function `V(t,x)` is the unique fixed point of
`𝒯` in `IM_cx`. b) There is an optimal Markov strategy `π^*_t = f^*(t, X_{t-})` — the policy
`f(t,x)(s) := f^*(t+s, x)` of the embedded model attains `V` — where `f^*` is the smallest
minimizer **(9.22)** at `V`, and `f^*(t,x) ≤ f^*(t,x+1) ≤ f^*(t,x) + 1`. -/
theorem theorem_9_4_2 (M : TradeExecution) :
    ∃ V : ℝ × ℕ → ℝ, M.IMcx V ∧
      (∀ p ∈ M.E, M.Jinf p = ENNReal.ofReal (V p)) ∧
      (∀ p ∈ M.E, M.T_op V p = V p) ∧
      (∀ v : ℝ × ℕ → ℝ, M.IMcx v → (∀ p ∈ M.E, M.T_op v p = v p) → ∀ p ∈ M.E, v p = V p) ∧
      (∀ t ∈ Set.Icc (0 : ℝ) M.T, ∀ x : ℕ,
        M.fstar V (t, x) ≤ M.fstar V (t, x + 1) ∧
        M.fstar V (t, x + 1) ≤ M.fstar V (t, x) + 1) ∧
      (∃ f : ℕ → ℝ × ℕ → Offer, TradeExecution.IsPolicy f ∧
        (∀ n, ∀ p ∈ M.E, ∀ s : ℝ, 0 ≤ s → (f n p).val s = M.fstar V (p.1 + s, p.2)) ∧
        ∀ p ∈ M.E, M.Jinfpi f p = M.Jinf p) := by sorry

end MDPFinance.JumpMarkets
Source
Bäuerle and Rieder, Markov Decision Processes with Applications to Finance, Universitext, Springer 2011, DOI 10.1007/978-3-642-18324-9, p. 296 (PDF 306), Theorem 9.4.2
Human review
  • Endorsed by Community (Bot) · Oct 1, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 1, 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