markov chains


Summary

Algorithm to find the equilibrium vector/eigenvector associated to 1

the normal algorithm for eigenvectors would work as well

Adjacency matrix

convert into a probability transition matrix by scaling each column such that the total is 1

Concept

Probability vector

  • all entries are nonnegative and sum to 1

different from unit vectors

(Left) stochastic matrix

  • columns are probability vectors

Markov chain, cumulative probability

  • sequence of probability vectors(state vectors), that represents the probability of diffferent outcomes
  • alongside a stochastic matrix, which transforms the current state vector to the subsequent state vector

1 is always an eigenvalue of a stochastic matrix

Steady-state vector/equilibrium vector

  • a state vector that remains the same regardless of how many experiments
  • is the eigenvector associsated to the eigenvalue 1
  • if the Markov chain converges, it will converge to this equilibrium vector

Regular stochastic matrix

  • there is a power where all the entries are more than 0
  • every state can reach every state

recall the product of triangular matrices is also triangular, ie. a triangular stochastic matrix is not regular

Markov chain with regular stochastic matrix

  • will converge to the unique equilibrium vector
Octave
octave
# Finding the equilibrium vector of a converging markov chain
rref([P-eye(3), [0; 0; 0]; 1 1 1 1])

Application

Probability

  • Sheldon only patronizes three stalls in the school canteen, the mixed rice, noodle, and mala hotpot stall for lunch everyday. He never buys from same stall two days in a row. If he buys from the mixed rice stall on a certain day, there is a 40% chance he will patronize the noodles stall the next day. If he buys from the noodle stall on a certain day, there is a 50% chance he will eat mala hotpot the next day. If he eats mala hotpot on a certain day, there is a 60% chance he will patronize the mixed rice the next day.

Outgoing connections on adjacency matrix, on a graph

s1s2s3s4