Transition Matrix · Stationary Dist

Markov Chains Explained

Represent a stochastic process with a transition matrix and study state evolution via matrix powers. Explore conditions for a unique stationary distribution and mixing rates.

📚 Fundamentals

Stationarity

A distribution π is stationary if πP = π. Under irreducible and aperiodic conditions, the chain converges to a unique π.

🧪 Examples

Weather model with states {Sunny, Cloudy, Rainy} and plausible transition probabilities; compute long-run frequencies.

❓ Frequently Asked Questions

1) What is ergodicity?
Conditions ensuring convergence to a unique stationary distribution.
2) How to find π?
Solve (Pᵀ - I)π = 0 with normalization or power iteration.
3) What is mixing time?
How fast distributions converge to stationarity.
4) Periodicity?
If returns occur at multiples >1, convergence may oscillate.
5) Absorbing states?
States that, once entered, cannot be exited (row with 1 on diagonal).
6) Reversible chains?
Satisfy detailed balance: π(i)P(i,j)=π(j)P(j,i).
7) Continuous-time chains?
Use generators instead of transition matrices.
8) Relation to MCMC?
MCMC constructs chains whose stationary distribution is the target posterior.
9) Graph interpretation?
P is a weighted adjacency matrix of a directed graph.
10) PageRank?
A Markov chain on the web graph with teleportation to ensure ergodicity.