Eigenvalues in dynamics and applications
Eigenvalues predict how systems evolve: populations, vibrations, web rankings, and differential equations. This lesson connects the theory to real behaviour.
Discrete dynamical systems
Many systems evolve in steps: \(\mathbf{x}_{k+1} = A\mathbf{x}_k\). The eigenvalues decide stability: if every eigenvalue has \(|\lambda| < 1\), all states decay to zero; if some \(|\lambda| > 1\), most states grow without bound. Eigenvalues with \(|\lambda| = 1\) give steady or periodic behaviour, as in Markov chains, whose largest eigenvalue is exactly 1.
Population models
Leslie matrices, introduced by Patrick Leslie in 1945, model populations divided into age classes, with birth rates in the first row and survival rates below the diagonal. The dominant eigenvalue gives the long-run growth rate per period, and its eigenvector gives the stable age distribution that the population approaches.
PageRank
Google's PageRank models a random surfer following links, occasionally jumping to a random page. The importance scores form the eigenvector, with eigenvalue 1, of a huge stochastic matrix built from the web's links. The power method, repeatedly multiplying a starting vector by the matrix, converges to it; with billions of pages, this was one of the largest eigenvector computations ever performed routinely.
Differential equations
For the continuous system \(\mathbf{x}'(t) = A\mathbf{x}(t)\), each eigenpair gives a solution \(e^{\lambda t}\mathbf{v}\). If A is diagonalisable, the general solution is \(\mathbf{x}(t) = c_1e^{\lambda_1 t}\mathbf{v}_1 + \dots + c_ne^{\lambda_n t}\mathbf{v}_n\). Stability now depends on real parts: all negative means solutions decay to equilibrium; any positive means growth. Complex eigenvalues produce oscillations.
Vibrations
Mechanical structures, from guitar strings to bridges and buildings, have natural frequencies of vibration given by eigenvalues of their mass and stiffness matrices, with mode shapes given by eigenvectors. Engineers ensure natural frequencies avoid excitation frequencies from wind, traffic, or earthquakes, since resonance can cause dangerous oscillations.
Quantum mechanics and chemistry
In quantum mechanics, measurable quantities are represented by operators, and the possible measurement results are their eigenvalues; energy levels of atoms and molecules are eigenvalues of the Hamiltonian operator. Computational chemistry solves enormous eigenvalue problems to predict molecular structure and reactions.
Spectral graph theory
The eigenvalues of matrices built from networks, such as the adjacency matrix and the graph Laplacian, reveal structure: how well connected a network is, natural clusters (spectral clustering), and how quickly information or disease spreads. The second-smallest Laplacian eigenvalue, the algebraic connectivity, measures how hard a network is to split.
Worked example
A two-age-class population has Leslie matrix \(L = \begin{bmatrix} 1 & 2 \\ 0.5 & 0 \end{bmatrix}\): juveniles produce 1 offspring, adults 2, and half of juveniles survive to adulthood. The characteristic equation is \(\lambda^2 - \lambda - 1 = 0\), with dominant root \(\lambda = \frac{1 + \sqrt{5}}{2} \approx 1.618\), so the population eventually grows about 61.8% per period. An eigenvector is (2, 0.618) (from \(0.5x_1 = 1.618x_2\)), so about 76% are juveniles in the long run.
Key takeaways
In x_{k+1} = A x_k, eigenvalues with |lambda| < 1 give decay, |lambda| > 1 growth, and |lambda| = 1 steady states, as in Markov chains. Leslie matrices (1945) give population growth rates and stable age structures from the dominant eigenpair. PageRank is the eigenvalue-1 eigenvector of a web link matrix, computed by the power method. For x' = A x, solutions combine e^{lambda t} v, with stability set by the real parts of eigenvalues. Natural vibration frequencies and mode shapes are eigenvalues and eigenvectors. Quantum energy levels are eigenvalues. Graph Laplacian eigenvalues reveal network connectivity and clusters.
Flashcard
All |lambda| < 1 means decay to zero
Check your understanding
For x' = Ax, the matrix A has eigenvalues -2 and -0.5. What happens to solutions over time?
