Detailed balance, full balance, and time-reversed rates
DefinitionKellyStochasticNetworks_BalanceThe reversibility apparatus of Chapter 1 of Kelly and Yudovina, Stochastic Networks.
Let be a state space, a matrix of transition rates and a collection of real numbers. Three notions are introduced.
- Detailed balance. and are in detailed balance when
In equilibrium this says transitions from to occur as frequently as transitions from to . This is equation (1.4) of the book.
- Full balance (the equilibrium equations). satisfies the equilibrium equations for when
This is equation (1.2). The sums range over the whole, possibly countably infinite, state space.
- The time-reversed rates. For strictly positive, the transition rates of the reversed process computed in Proposition 1.1 are
These three notions are the whole method of the book's first three chapters: an equilibrium distribution is found by guessing that the process is reversible and solving the detailed balance equations, which are a two-term recursion, in place of the equilibrium equations, which are not.
Formalization Note The state space is an arbitrary type, so the same predicates serve
finite and countable chains. Full balance is written with Lean's unconditional sum tsum;
over a finite state space this is the ordinary finite sum, and over a countable one it is the
sum of the family whenever that family is summable.
import Mathlib
namespace KellyStochasticNetworks
/-- **Detailed balance.** A collection of numbers `π` and a matrix of transition rates `q`
on a state space `S` satisfy detailed balance when `π j * q j k = π k * q k j` for all
states `j, k`. This is equation (1.4) of Kelly–Yudovina, *Stochastic Networks*. -/
def DetailedBalance {S : Type*} (π : S → ℝ) (q : S → S → ℝ) : Prop :=
∀ j k : S, π j * q j k = π k * q k j
/-- **The equilibrium (full balance) equations.** `π` satisfies the equilibrium equations for
the transition rates `q` when `π j * ∑ k, q j k = ∑ k, π k * q k j` for every state `j`.
This is equation (1.2) of Kelly–Yudovina, *Stochastic Networks*; the sums range over the whole
(countable) state space, so they are written as unconditional sums. -/
def FullBalance {S : Type*} (π : S → ℝ) (q : S → S → ℝ) : Prop :=
∀ j : S, π j * (∑' k : S, q j k) = ∑' k : S, π k * q k j
/-- **The transition rates of the time-reversed process**, `q' j k = π k * q k j / π j`,
as computed in Proposition 1.1 of Kelly–Yudovina, *Stochastic Networks*. -/
noncomputable def reversedRates {S : Type*} (π : S → ℝ) (q : S → S → ℝ) : S → S → ℝ :=
fun j k => π k * q k j / π j
end KellyStochasticNetworks