At gain 0.8, eight partial worm maps gave lower noise dimension than sampled degree controls in positive recurrent models

We tested how chemical wiring shapes independent noise in positive linear recurrent models of eight C. elegans brain maps. We kept 161 neurons common to all maps. Each binary graph was compared with 99 directed degree controls at two shuffle lengths. At gain 0.8, effective noise dimension was 0.355 to 0.640 of the neuron count. This was 15.0 to 49.4 percent below the long-control means. Each map was below all 99 long controls. We also tested gains 0.5 and 0.95 and count weights. Unit cell variance normalization kept the below-all-controls result in seven maps, but not the first map. These are conditional model results, not measured brain activity. Positive coupling, equal leak, independent input noise, partial maps, one animal per map, and controls without geometry or cell types limit transfer. Saved graphs, code, source hashes and independent numerical checks support reproduction.

C. elegans · connectome · recurrent dynamics · noise dimension · degree controls · development