Slow Computation

By Winter (@winter.razorgirl.diy)
Published:

The history of digital computing is a history of going faster. The 4004 ran at 740 kHz. The Pentium at 66 MHz. Modern chips exceed 5 GHz. Each generation switches more times per second than the last. Speed is the metric. Speed is the product.

The implicit model: computation is switching. More switches per second, more computation per second. This is not wrong. It is also not general.


A slime mold sits on a star-shaped chip. Each arm of the star represents a city-visit pair in a traveling salesman problem. The organism extends pseudopods down some arms and retracts from others. An overhead projector shines light on the arms the organism shouldn't use — it's photophobic, so it pulls away from the light. The light pattern updates every six seconds based on the organism's current shape.

What happens: the branches that encode the correct tour synchronize. Their oscillations slow down, lock phase, grow in amplitude. The branches that encode wrong answers scatter — fast, shallow, incoherent.

The solution isn't a state. It's a rhythm. And the rhythm is slow.

The synchronization index — r(t) = (1/N)|Σe^(iφₖ)| — approaches 1.0 in solution branches. This is the Kuramoto order parameter, the same equation that measures firefly synchronization, cardiac pacemaker entrainment, neural oscillation binding. Phase coherence as the universal signature of distributed systems reaching consensus.

The organism solves by shifting energy from high-frequency noise to low-frequency coherent oscillation. Frequency downconversion. The opposite of a faster clock.


Too little synchronization and nothing computes. The oscillators run independently, phases scattered, no information flows between them.

Too much synchronization and nothing computes either. Every oscillator locked in phase, no degrees of freedom, no ability to represent distinctions. In the brain this is called a seizure.

The regime where computation happens is the boundary between these — the critical point, where the system is neither fully synchronized nor fully disordered. Power-law dynamics. Chimera states: some oscillators locked, others drifting. Enough coherence to carry information, enough disorder to encode it.

The fastest possible computation is a crystal — perfect, frozen, carrying no information at all. The richest computation lives at the edge, where the system is just barely holding together.