"Eigenvalue" may be the single most wall-like word in mathematics: it sounds abstract, it is usually taught as pure symbol-shuffling, and it guards the door to half of physics. Here is the secret the notation hides: an eigenvalue is a pitch. A connected thing — a chain of springs, a drumhead, a network — has a small set of natural ways it likes to wobble, and each way rings at its own rate. In this section you will pluck a chain and read off, live, exactly which of its natural wobbles your pluck was made of; then step through a real network's wobbles one at a time; then build your own network and play its chord. Only after all that will we say the word.
Strip a network down to its skeleton and you get a graph: dots (nodes) and lines between them (edges). Nothing else. The dots can be neurons, people, oscillators, atoms; the lines say who directly influences whom. On this site the lines carry the coupling strengths K you met in the phase primer — an edge is a channel through which one node nudges another. Everything in this module works for weighted couplings too, but today one line = one unit of nudge, so the shape alone does the talking.
Start with the simplest possible network: six masses in a row, joined by springs. Pull one mass aside and let go. The chain does something that looks complicated — but it is not doing anything new. Every motion this chain can ever make is a mixture of six special patterns: the whole chain swaying together, a see-saw with the left half up and the right half down, an S-curve, and progressively wigglier shapes. Each special pattern, left alone, would just oscillate in place at its own rate, forever holding its shape. Your pluck picked a particular recipe of them, and the meters below the chain show you that recipe, live.
That "shape that vibrates alone" is the whole idea of this module. Hold onto it.
Tap a drumhead and it rings in a few characteristic patterns: the whole skin billowing together, the left half up while the right half is down, four alternating quadrants. Each pattern rings at its own pitch — the gentle whole-skin motion is the low boom, the wigglier patterns are the higher overtones. (If you read the foundations essay on memory, this is exactly the physics behind Chladni figures — and the sound workshop lets you pour sand on those patterns.)
A network is a drum with a stranger shape. Imagine each node holding a value that gets pulled toward its neighbors' values — exactly the nudging dynamics this whole site runs on. Then the network has its own set of special wobble-patterns, called modes, each ringing at its own rate. Here is a network of two tight six-node clusters joined by a single edge — that lone connection is called a bridge, and it becomes the main character of section 03. Step through its modes:
Each of those patterns has a defining property, and it is worth saying carefully because it is the entire content of the scary word. When the network's pulling-toward-neighbors machinery acts on an arbitrary shape of displacements, it warps it: some parts snap back hard, others barely move, and the shape that comes out is different from the shape that went in. But the special patterns — the spring chain's see-saw, the network's cluster-vs-cluster — keep their shape. The network's response to that pattern is the same pattern, just scaled — pushed back with some overall strength. Nothing about its geometry changes; only its size.
That shape-keeping pattern is called an eigenvector (German eigen, "its own" — the network's own pattern). The scaling number — how hard the network pushes back on it, its stiffness — is the eigenvalue, written λ (lambda). Stiff pattern, big λ, fast vibration, high pitch. Gentle pattern, small λ, slow vibration, low pitch. The full set of a network's eigenvalues is called its spectrum — the same word as a spectrum of light or sound, and not by coincidence: it is literally the network's chord, the set of pitches it is capable of ringing at. That is the whole idea. Everything else about eigen-anything is bookkeeping.
The spectrum is not decoration on top of a graph — it is the graph, heard rather than seen. The fastest way to believe that is to change a graph and listen. Below, eight nodes: click two nodes in a row to add or remove the edge between them, or start from a preset. Every click recomputes the real spectrum; the coloring shows the gentlest genuine mode, and the bars are the chord.
You now own the picture: a network has its own patterns (eigenvectors), each with a stiffness that is also a pitch (eigenvalues; together, the spectrum). Section 02 opens the machine that generates all of it — a single, almost embarrassingly simple question each node asks about its neighbors — and shows the same machine running the site's coupling dynamics. Then section 03 takes the gentlest genuine mode you met here and turns it into the network's built-in early-warning system.