Matrix Models

A matrix model is a mathematical framework that uses matrices to describe how a system evolves over discrete time steps. Instead of tracking one number at a time, a matrix model bundles all the relevant quantities of a system into a state vector and uses matrix multiplication to predict the next state. This makes it possible to analyze complex systems — from population dynamics to web-page rankings — with elegant, compact algebra. The central idea is the recurrence relation x_{n+1} = A x_n, where x_n is a column vector describing the system at step n and A is a square matrix called the transition matrix (or Leslie matrix in ecology, or stochastic matrix in probability). Each entry of A encodes how strongly one part of the system influences another. Repeated multiplication A^n x_0 gives the state after n steps, and the long-run behavior is determined entirely by the eigenvalues of A. Matrix models appear throughout science and engineering: Leslie matrices model age-structured animal populations; Markov chains model weather, genetics, and Google's PageRank; compartmental matrices model disease spread. Learning to read and construct a transition matrix is therefore one of the most practical skills linear algebra can offer a student of any quantitative discipline.

Key takeaways
  • Identify the State Vector
  • Determine the Transition Rules
  • Assemble the Matrix
  • Iterate with Matrix Multiplication
  • Interpret the Eigenvalues

Identify the State Vector

Decide which quantities fully describe your system at one moment in time and list them as entries of a column vector x. For example, in a two-city population model the state vector might be x = [P_A, P_B]^T, where P_A and P_B are the populations of city A and city B.

Determine the Transition Rules

For each state variable, write a formula for its next value in terms of the current values of all variables. These formulas must be linear (no products or powers of variables). For instance: P_A(next) = 0.85 P_A + 0.10 P_B means 85% of city A's population stays plus 10% of city B migrates to A.

Assemble the Matrix

Place the coefficients from each linear formula into the rows of a matrix A. The entry A_{ij} gives the contribution of state j to state i in the next step. In column form: if P_A(next) = 0.85 P_A + 0.10 P_B and P_B(next) = 0.15 P_A + 0.90 P_B, then A = [[0.85, 0.10],[0.15, 0.90]]. Notice the columns sum to 1 because people must go somewhere — this is a stochastic (Markov) matrix.

Iterate with Matrix Multiplication

The state after one step is x_1 = A x_0. After two steps x_2 = A x_1 = A(A x_0) = A^2 x_0. In general, x_n = A^n x_0. Computing A^n directly is tedious; instead, diagonalize A = P D P^{-1} so that A^n = P D^n P^{-1}, where D^n is found by raising each diagonal eigenvalue to the nth power.

Interpret the Eigenvalues

The dominant eigenvalue lambda_1 (the one with the largest absolute value) controls the long-run behavior. If |lambda_1| > 1 the system grows without bound; if |lambda_1| < 1 it decays to zero; if |lambda_1| = 1 it reaches a steady state. The corresponding eigenvector v_1 gives the stable distribution — the proportional makeup of the state in the long run, regardless of starting conditions.

The Key Equation

xn=An x0=P Dn P−1 x0\mathbf{x}_{n} = A^{n}\,\mathbf{x}_{0} = P\,D^{n}\,P^{-1}\,\mathbf{x}_{0}
  • \mathbf{x}_n — State vector at step n (varies (e.g., people, probability)): Column vector whose entries describe every relevant quantity of the system at discrete time n.
  • A — Transition matrix (dimensionless): Square matrix whose (i,j) entry gives the contribution of state j to state i after one time step.
  • P — Eigenvector matrix (dimensionless): Matrix whose columns are the eigenvectors of A; changes basis from eigen-coordinates to standard coordinates.
  • D^n — Diagonal eigenvalue matrix to power n (dimensionless): Diagonal matrix with entries lambda_1^n, lambda_2^n, …, making repeated multiplication trivial.
  • \mathbf{x}_0 — Initial state vector (varies): The known starting condition of the system at time step zero.

Worked Example

Problem. A nature reserve tracks rabbits in two zones, Forest (F) and Meadow (M). Each year, 80% of forest rabbits stay in the forest and 20% move to the meadow; 30% of meadow rabbits move to the forest and 70% stay in the meadow. Initially there are 400 rabbits in the forest and 200 in the meadow. (a) Write the transition matrix. (b) Find the population after 1 year. (c) Find the long-run stable distribution.

x1=[380220],x∞∝[32]⇒[360240]\mathbf{x}_1 = \begin{bmatrix}380\\220\end{bmatrix}, \quad \mathbf{x}_\infty \propto \begin{bmatrix}3\\2\end{bmatrix} \Rightarrow \begin{bmatrix}360\\240\end{bmatrix}

Answer. After 1 year: 380 forest rabbits and 220 meadow rabbits. Long-run stable distribution: 360 in forest and 240 in meadow (ratio 3:2).

Figure
Diagram
Two-Zone Rabbit Transition ModelForestx₁ (rabbits)Meadowx₂ (rabbits)0.800.700.200.30Transition Matrix AA = ⎡ 0.80 0.30 ⎤ ⎣ 0.20 0.70 ⎦Columns sum to 1.0(Stochastic matrix)Eigenvaluesλ₁ = 1.0→ Stable steady stateλ₂ = 0.5→ Decays to zero
Origins · 1 minAndrey Andreyevich Markov and how this was discovered
AA
Andrey Andreyevich Markov
1856–1922 · Russian

Markov introduced the concept of stochastic processes with the 'Markov property' — that the future state depends only on the present state, not the history. His 1906 paper defined what we now call Markov chains, exactly the framework behind transition matrices. He applied his theory first to analyze the alternation of vowels and consonants in Pushkin's poem Eugene Onegin, then to probability theory more broadly. Stochastic matrices with columns (or rows) summing to 1 are named Markov matrices in his honor. Markov and his contemporary Pavel Nekrasov had a heated debate over whether the law of large numbers required events to be independent. Markov invented his chains partly to prove Nekrasov wrong — showing that dependent events could still obey the law of large numbers.

  • Matrix Multiplication — Matrix models are driven entirely by repeated matrix multiplication x_{n+1} = A x_n, so fluency with matrix multiplication is the essential prerequisite.
  • Eigenvalues and Eigenvectors — The long-run behavior of every matrix model — whether it grows, decays, or reaches equilibrium — is determined by the dominant eigenvalue and its eigenvector.
  • Diagonalization — Diagonalizing A = PDP⁻¹ reduces computing A^n from repeated matrix multiplication to simply raising each eigenvalue to the nth power, making long-run predictions practical.
  • Markov Chains — A stochastic transition matrix is the algebraic heart of a Markov chain; understanding both topics together connects linear algebra to probability theory and real applications such as PageRank.