Big idea
A matrix iteration has the form xk+1 = A xk. After many steps, eigenvectors often become visible because Akv = λkv.
growthdecayoscillationrankingMarkov chainspower iteration
1. Two-direction iteration
Change the two eigenvalues of a diagonal matrix. One direction may grow while another decays.
2. Power iteration
Repeated multiplication plus normalization can find a dominant eigenvector. The matrix is symmetric: A = [[a,b],[b,d]].
3. Markov chain weather model
Adjust transition probabilities. The columns represent today's weather; the entries describe tomorrow's distribution.
4. PageRank-style ranking
Importance flows through links. Damping adds a chance of jumping randomly to any page.
Columns are source pages; rows are destination pages. The rank vector is the stable distribution of importance.
5. Iteration as smoothing
Repeated local averaging removes sharp details. This is a simple model of diffusion.
Reflection
Use the interactive tools above to answer these questions:
- When does iteration converge to zero?
- When does it explode?
- Why does normalization help power iteration?
- What is the difference between one-step movement and long-run behavior?
- How are ranking and Markov chains both eigenvector problems?