Eq. (2) and footnote 1, p.8 — the marginalized transition exists, is unique and is stochastic
ProvedMarkovEntanglementExact.local_transition_well_definedmarkov-chainp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1reinforcement-learningvalue-decomposition
Let be a transition matrix on and a distribution with , and positive agent marginals. Then there is exactly one matrix satisfying Eq. (2), it is a transition matrix, and
The same holds for agent .
This shows that the local transition matrices of the mission are well defined and that a statement quantified over them is about exactly one matrix for each agent.
Formalization Note IsLocalTransitionA is Eq. (2) multiplied through by the marginal. The display above ties it to the printed conditional form.
Differences from the source:
- generalization: P^π_AB of a two-agent MDP and policy → an arbitrary transition matrix P on a product of two finite sets; reason: the argument uses only the stochasticity of P; P^π_AB is an instance by M12
- generalization: μ is the stationary occupancy measure of a unichain, unique and strictly positive on states → μ is any distribution on joint pairs with μ ≥ 0, Σμ = 1 and positive agent marginals; reason: the derivation uses only the positivity of the conditioning marginals (footnote 1); the paper's occupancy measure is an instance, proved in the verification file OccupancyInstance
- encoding: Eq. (2) with the conditional μ(s_B, a_B | s_A, a_A) → IsLocalTransitionA/B, Eq. (2) multiplied through by the marginal; reason: equivalent under positive marginals; the quotient form is proved in M2
Preamble
import Mathlib import Definitions.Def_markov_entanglement import Definitions.Def_MarkovEntanglementExact_IsOccupancy open scoped BigOperators open MarkovEntanglement
Formal statement
namespace MarkovEntanglementExact
/-- Eq. (2) is well defined: under the hypotheses of footnote 1 there is exactly one local
transition matrix for each agent, it is a transition matrix, and it is given by the printed
formula `P_A(a' | a) = Σ_{b'} Σ_b P_AB((a', b') | (a, b)) μ(b | a)`, with the conditional
`μ(b | a) = μ(a, b) / μ(a)`. The same holds for agent `B`.
Source: Chen & Peng, Multi-agent Markov Entanglement, arXiv:2506.02385v3, Eq. (2) and
footnote 1, p.8.
**Formalization Note.** The published `IsLocalTransitionA` is Eq. (2) multiplied through by the
marginal. The last clause of each agent ties it to the printed quotient form. -/
theorem local_transition_well_defined
{SA SB : Type*} [Fintype SA] [Fintype SB] [DecidableEq SA] [DecidableEq SB]
(P : Matrix (SA × SB) (SA × SB) ℝ) (hP : IsTransitionMatrix P)
(μ : SA × SB → ℝ) (hμ : IsOccupancy μ) :
(∃! PA : Matrix SA SA ℝ, IsLocalTransitionA P μ PA) ∧
(∃! PB : Matrix SB SB ℝ, IsLocalTransitionB P μ PB) ∧
(∀ PA : Matrix SA SA ℝ, IsLocalTransitionA P μ PA →
IsTransitionMatrix PA ∧
∀ a a', PA a a' = ∑ b', ∑ b, P (a, b) (a', b') * (μ (a, b) / marginalA μ a)) ∧
(∀ PB : Matrix SB SB ℝ, IsLocalTransitionB P μ PB →
IsTransitionMatrix PB ∧
∀ b b', PB b b' = ∑ a', ∑ a, P (a, b) (a', b') * (μ (a, b) / marginalB μ b)) := by sorry
end MarkovEntanglementExact
Source
Chen & Peng, Multi-agent Markov Entanglement, arXiv:2506.02385v3, Eq. (2) and footnote 1, p.8
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.