What the occupation law of a finite Markov chain determines: time reversal and principal minors
The paper asks what the law of the occupation times of a finite continuous-time Markov chain determines about its generator Q, both exactly and at the large-deviation scale given by the Donsker–Varadhan functional.
Occupation laws are reduced to the rational function μ(diag x − Q)^{-1}1. Irreducibility of multiaffine determinant polynomials (Al Ahmadieh, Hartfiel–Loewy) and Loewy's theorem on principal minors, applied to a generator bordered by the start, are used to characterize the generators with equal laws. Explicit examples are checked in exact arithmetic with Python scripts. Finite and algebraic steps are formalized in Lean 4 with Mathlib, with the Perron–Frobenius positive null vector taken as a hypothesis.
From the stationary start, one-time occupation law determines Q up to (Z/2)^k, generated by time reversal and commuting reversals of memoryless blocks; generically only time reversal remains. From any start, Q is determined up to lumped reversals, and the Donsker–Varadhan functional determines exactly the principal minors of −Q.
| Observation | What it determines | Source |
|---|---|---|
| the path | Q | |
| law of O(T) at one T, start μ without renewal blocks | Q if μ≠π; Q up to time reversal if μ=π | Thm. B |
| law of O(T) at one T, stationary start | Q up to (Z/2)^k (time reversal and memoryless-block reversals) | Thm. A |
| law of O(T) at one T, any start μ | Q up to lumped reversals | Thm. B |
| functional J_V | principal minors of −Q | Prop. 1.1 |
