A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Linear algebra models diverse real-world systems: networks via adjacency matrices, population dynamics via Leslie matrices, image compression via SVD, least-squares fitting for data, and solutions to systems of differential equations via matrix exponentials. Eigenvalues reveal stability and long-term behavior. These applications demonstrate the unifying power of linear algebra across science and engineering.
You've built up the core machinery of linear algebra — matrix multiplication, linear transformations, eigenvalues, and eigenvectors. The payoff comes now: these tools aren't just abstract algebra; they are the language scientists and engineers use to model systems where many quantities interact simultaneously. Whenever a real-world situation involves multiple variables that influence each other linearly, linear algebra is the right framework.
Networks offer the most direct example. Model a system of nodes and connections — intersections and roads, web pages and links, proteins and interactions — as a matrix where each entry records the connection strength between two nodes. Powers of this matrix count paths of various lengths. The PageRank algorithm, which ranked web pages for early search engines, is essentially an eigenvector computation: the dominant eigenvector of the web's link matrix represents a "steady-state" distribution of random web traffic. The key insight is that the eigenvalue λ₁ and its eigenvector tell you the long-run behavior of any process that repeatedly applies the matrix — traffic, disease spread, rumor propagation.
Population dynamics shows eigenvalues predicting the future. A Leslie matrix models an age-structured population: rows represent how individuals in each age class reproduce or survive into the next. Multiply the population vector by the Leslie matrix once and you get next year's population. The dominant eigenvalue λ₁ determines the long-term growth rate: if λ₁ > 1 the population grows geometrically, if λ₁ < 1 it declines, if λ₁ = 1 it stabilizes. The corresponding eigenvector gives the stable age distribution — the proportions of each age class that the population approaches regardless of how it started. This is the same eigenvector computation you've seen abstractly, now giving concrete biological predictions.
Least-squares regression brings linear algebra into data analysis. When you want to fit a model to n data points with n > (number of parameters), there is generally no exact solution — the system Ax = b is overdetermined. The least-squares solution minimizes the total squared error and is given by the normal equations AᵀAx = Aᵀb: compute AᵀA (a square matrix), compute Aᵀb, then solve. The geometric interpretation is elegant: Ax̂ is the projection of b onto the column space of A — the closest achievable output to the observed data. Singular Value Decomposition (SVD) deepens this: factoring any matrix as A = UΣVᵀ provides the best rank-k approximation of A (keep only the k largest singular values and their corresponding columns). This approximation is the basis of image compression, noise reduction, and dimensionality reduction across machine learning — the same matrix decomposition, applied at scale.