Min-plus powers record cheapest decision walks
ProvedMDPComplexity.AverageCost.min_plus_walksgraph-theorymin-plusp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1shortest-walk
Let be the min-plus adjacency matrix of a finite deterministic process. For states and an integer , the entry is exactly when there is no -arc decision walk from to . Whenever there is such a walk, the entry is the least of their total costs:
This identifies the matrix computation used for shortest closed walks. Parallel decisions are compared by cost; an absent arc has the extended value .
Formalization Note The zeroth power is the min-plus identity, so its diagonal entries are zero and other entries are infinite. A walk records each decision arc, not only its state sequence.
Preamble
import Definitions.Def_MDPComplexity_AverageCost_Model
Formal statement
namespace MDPComplexity.AverageCost
variable {S : Type*} [Fintype S] [DecidableEq S]
/-- §3, p. 446: the kth min-plus power records exactly the cheapest k-arc
decision walk, and infinity records the absence of such a walk. -/
theorem min_plus_walks (M : DetMDP S) (k : ℕ) (u v : S) :
(M.minPlusPow k u v = ⊤ ↔
¬ ∃ W : M.Walk k, W.start = u ∧ W.finish = v) ∧
(∀ W : M.Walk k, W.start = u → W.finish = v →
∃ r : ℝ, M.minPlusPow k u v = (r : WithTop ℝ) ∧
IsLeast {x : ℝ | ∃ W' : M.Walk k,
W'.start = u ∧ W'.finish = v ∧ x = W'.cost} r) := by sorry
end MDPComplexity.AverageCost
Source
Papadimitriou and Tsitsiklis, The Complexity of Markov Decision Processes, Math. Oper. Res. 12(3) (1987), p. 446, §3 The infinite horizon undiscounted case, proof of Theorem 3, https://doi.org/10.1287/moor.12.3.441
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.