mathematics//graph theory

Graph theory is the branch of mathematics that studies networks of nodes and connections, and here it is used to explain how a network organizes collective behavior. The nodes and edges stand for drones and the radio links between them, tanks and the pipes joining them, substations and transmission lines, or machines joined by how alike their vibration signatures are. A graph records which nodes are connected and how strongly; everything else (what quantity lives on the nodes, what law moves it, what the eigenvalues mean) is a separate decision layered on top. It is easily confused with computer networks (ports, gateways, RPC), which are only one of the things it can model.

Graph theory is the branch of mathematics that studies networks of nodes and connections, and here it is used to explain how a network organizes collective behavior. The nodes and edges stand for drones and the radio links between them, tanks and the pipes joining them, substations and transmission lines, or machines joined by how alike their vibration signatures are. A graph records which nodes are connected and how strongly; everything else (what quantity lives on the nodes, what law moves it, what the eigenvalues mean) is a separate decision layered on top. It is easily confused with computer networks (ports, gateways, RPC), which are only one of the things it can model.

What makes a graph something to compute with is that it is a matrix. Written as an adjacency matrix, it opens all of linear algebra to the questions an engineer asks of a network (is it connected, where is the bottleneck, how fast does something spread, which node matters most), and each of them becomes a question about eigenvalues.

adjacency and degree describes the connections. The graph Laplacian sums differences across them, and its normalized versions turn absolute sums into local averages.

nodal capacity adds storage, which is what turns a balance into a rate. diffusion and consensus describes the relaxation that follows, and the spectral gap says which patterns persist.

spectral coordinates separates two uses of the same eigenvectors: decomposing a state into modes, and giving each node new coordinates.

Two worked cases hold the pieces together: three-tank network (balance, capacity, modes and a random walk on one topology) and two-community graph (a slow mode that separates two groups).

The diameter is the longest of the shortest paths between two nodes, counted in hops, and it is the number of rounds a message needs to reach everyone. Agents that spread auction prices by keeping the highest bid they have heard agree in exactly that many rounds (consensus protocol).

How the links are distributed decides how a network fails. A network dominated by a few hubs shrugs off random failures and collapses when the hubs go (scale-free network).

The same topology can carry a physical network of levels, a random walk of probabilities, a similarity graph over data (spectral clustering), the who-hears-whom communication graph of a fleet, or the input of a GNN. State, units, capacities and clock must be declared each time; an edge can be a conductance, a dependence or a similarity, and its meaning does not travel automatically between models.