slime mold network optimization · Tero et al. 2010
Each node is a food source. Each edge is a possible tube connection, starting with equal conductance. Fluid cycles through source-sink pairs; well-used tubes grow thicker while poorly-used tubes decay. No part of the system has information about the whole. The efficient network is what remains after everything less efficient has been pruned away.
The model is from Tero et al. (2010). The network is a graph: nodes are food sources, edges are potential tube connections. At each step, a source-sink pair is chosen and the pressure at every node is computed by solving Kirchhoff's circuit equations — the same physics as current through a resistor network, where resistance is length divided by conductance.
Flow magnitude drives the conductance update: tubes with high flow are reinforced; tubes with low flow decay toward zero. The rule is local — each tube segment only knows the flow through itself.
Over many iterations across all source-sink pairs, a sparse, efficient network emerges. It was not designed. It was not selected. It is the residue of a process that had no information about its own outcome.
This differs from the agent-based slime mold model (see slime), which simulates individual particles following chemical gradients. The Tero model operates directly on tube conductances — it is a model of the vascular adaptation process, not of exploration and growth.
→ entry-554: Residue