Why Linearity Is Universal

Part 1, Chapter 1: Linear Algebra Toolkit

Learning objectives

State the two defining properties of a linear function: additivity and homogeneity
Recognize linear models in machine learning, computer graphics, quantum mechanics, and search algorithms
Identify the affine-vs-linear distinction that catches beginners
Predict that a small-scale phenomenon will be approximated by a linear model

Linear algebra is the most universally applicable branch of mathematics. Almost every quantitative discipline reduces, at some level, to manipulating vectors and matrices. The reason is structural: most phenomena, when examined closely enough, can be approximated by linear models, and linear models are the only models with closed-form solutions, predictable behaviour, and efficient computation. If you understand linearity, you understand the first thing every applied field reaches for.

The two-line definition

A function $T$ from one vector space to another is linear when it preserves two operations:

Additivity: $T(\mathbf{u} + \mathbf{v}) = T(\mathbf{u}) + T(\mathbf{v})$ , splitting an input through addition splits the output the same way.
Homogeneity: $T(c \mathbf{u}) = c\, T(\mathbf{u})$ for every scalar $c$ , scaling an input scales the output by the same factor.

Combined: $T(c_1 \mathbf{u}_1 + c_2 \mathbf{u}_2) = c_1 T(\mathbf{u}_1) + c_2 T(\mathbf{u}_2)$ . This deceptively modest requirement has enormous consequences: it forces $T$ to be completely determined by its values on a finite basis. Once you know $T(\mathbf{e}_1), T(\mathbf{e}_2), \ldots$ , you know $T$ everywhere, this is the superposition principle.

Why the universe seems to be linear (locally)

The vast majority of real systems are not linear. Populations saturate; markets fluctuate non-monotonically; friction grows quadratically with speed. Yet linear algebra dominates applied mathematics because of one fact: any smooth function looks linear when you zoom in. This is the geometric content of the derivative, the Jacobian matrix is the best linear approximation to a non-linear map at a point. From this, calculus and optimisation rest entirely on linear algebra.

The matrix-multiplier above demonstrates how a $2 \times 2$ matrix transforms the plane. Notice that the unit grid stretches and shears but stays a grid of parallelograms, that is linearity in action. Try a rotation matrix (e.g., $[\cos\theta, -\sin\theta; \sin\theta, \cos\theta]$ ) and watch shapes rotate while straightness, parallelism, and the origin all stay fixed.

Where this shows up

Machine learning: Every neural-network layer is a linear map $W\mathbf{x} + \mathbf{b}$ followed by a non-linearity. The "deep" in deep learning is really a composition of many linear layers with non-linear activations between them, pure linear-algebra at the bottom.

Computer graphics: Every 3D rotation, scaling, and projection on screen is a matrix-vector multiplication. Game engines, CAD software, and AR/VR all run on linear transformations applied at frame rate.

Quantum mechanics: Quantum states are unit vectors in a Hilbert space. Observables are self-adjoint linear operators (matrices, in the finite-dim case). The whole theory is linear algebra dressed in physics notation.

PageRank and search: Google's original algorithm computed the dominant eigenvector of a giant stochastic matrix encoding the web's link structure. Page importance is a linear-algebra problem.

Image and audio compression: JPEG (Discrete Cosine Transform), MP3 (Modified DCT), and HEVC (integer transforms) all apply linear transformations that decorrelate signal coefficients before quantisation. The reason these formats compress well is that natural signals have low-rank linear structure.

Pause and think: Is the function $f(x) = 3x + 5$ linear in the technical sense above? Why or why not? Hint: compute $f(0)$ . What does additivity force $T(\mathbf{0})$ to be for any linear $T$ ?

Try it

Predict first: is $T(x, y) = (2x + y,\ x - 3y)$ linear? Verify by computing $T(1, 0) + T(0, 1)$ and comparing to $T(1, 1)$ .

The function $g(x) = x^2$ is not linear. Find a specific pair $u, v$ where $g(u + v) \neq g(u) + g(v)$ . Justify in one sentence why $x^2$ fails additivity.

Use the matrix widget above to apply two consecutive rotations: by $30^\circ$ then by $60^\circ$ . The result should equal a single rotation by $90^\circ$ . Composition of linear maps stays linear, that is the chapter's thesis in miniature.

True or false: if $T$ is linear and $T(\mathbf{v}) = \mathbf{0}$ for some non-zero $\mathbf{v}$ , then $T$ cannot be invertible. Justify. (Hint: an invertible function maps distinct inputs to distinct outputs.)

A trap to watch for

The phrase "linear" in everyday English (a "linear" line through a chart) is more permissive than the mathematical definition. The function $f(x) = 3x + 5$ is sometimes called "linear" because its graph is a straight line, but it is not linear in the technical sense: $f(0) = 5 \neq 0$ , which violates additivity (every linear function must send the zero vector to the zero vector). Technically $f$ is affine, a linear function plus a constant. The distinction matters because non-zero-preserving maps do not inherit linear algebra's clean structure: their composition, inversion, and matrix representation all need extra care.

What you now know

You can identify linear maps by checking the two-line definition, recognise the affine-vs-linear distinction that catches beginners, and locate linearity in real engineering disciplines. The next sections build the toolkit: vectors and the standard basis (§1.2), linear maps as matrices (§1.3), basis and dimension (§1.4), and finally the determinant and eigenvalues that unlock everything downstream.

Mark section complete →

References

Garrity, T. (2002). All the Mathematics You Missed: But Need to Know for Graduate School. Cambridge University Press, ch. 1.

Strang, G. (2016). Introduction to Linear Algebra (5th ed.). Wellesley-Cambridge Press, ch. 1-2.

Axler, S. (2015). Linear Algebra Done Right (3rd ed.). Springer, ch. 1.

Hoffman, K., Kunze, R. (1971). Linear Algebra (2nd ed.). Prentice-Hall, ch. 1-2.

Trefethen, L. N., Bau, D. (1997). Numerical Linear Algebra. SIAM, ch. 1 (introduction to linear maps and matrices).

This page is prerendered for SEO and accessibility. The interactive widgets above hydrate on JavaScript load.