At 10 ms per state, opposite gain changes of 10%, 20% and 40% kept decay within 0.1% of baseline in one zebrafish linear model

We tested whether cell gain changes with a time mean of one preserve a model eye-position signal. We used a signed 289-cell core from one partial zebrafish hindbrain connectome. We made 64 balanced gain patterns. Each pattern alternated between opposite 10%, 20% or 40% gain changes. Each state lasted 0.01, 0.1, 1, 10 or 100 s. Three model calibrations gave 2,880 tests. At 0.01 s per state, every tested decay rate stayed within 0.1% of its initial value. Longer states changed decay in both directions. The 0.1 s calibration had recurrent eigenvalue 0.99. At 40% amplitude, 49 of 64 pattern pairs had a growing frozen state. Only one pattern grew over complete cycles at 1 s per state. The source article's times for each cell class gave decay in every tested cycle. For one pattern near the median rate effect, motor output reached 2.91 times its initial value. A local second-order calculation links the duration effect to activity outside the initial slow mode. Uniform gain changes gave the exact commuting control. This is a conditional result for assumed periodic gains in a linear model. It does not estimate gain changes, failure rates or eye movements in animals. The package has code, source checksums, all results, figures, a derivation and repeat instructions. Original source matrices remain external.

connectome · neural integrator · gain modulation · periodic dynamics · Floquet exponent · zebrafish · persistent activity · model stability