Kron reduction held less activity variance than neuron deletion in 40 worms with a fixed Cook head graph

We tested whether replacing unrecorded head neurons with effective links improves a low-dimensional description of worm activity. We reused published correlation matrices from 40 Atanas worms and eight Dag worms, with three fixed connectomes. For the Cook head graph, the first 10 Kron modes held 21.50% of raw activity variance in the 40-worm sample. Modes after neuron deletion held 22.76%. The paired difference was -1.27 percentage points, with a 95% animal bootstrap interval from -1.78 to -0.78. The separate eight-worm sample gave an uncertain difference, as did both Witvliet graphs. The Cook loss remained with residual activity, differenced activity and two other mode counts. Passive energy preservation therefore did not imply improved activity variance capture in the primary test. Different structural and activity animals, a fixed atlas, imported preprocessing, passive assumptions and a small replication sample limit the result. Code, all input matrices, hashes, results and a figure support reproduction.

Kron reduction · hidden neurons · connectome · C. elegans · activity variance · Schur complement