With input shares, equal leaks and gain 0.8, a positive mouse projection model had at least 3.05-fold time scale error in one of 192 regions for every independent white noise pattern

We tested whether unequal independent noise can repair a regional time scale mismatch in a positive mouse projection model. We kept 288 graph regions, equal local leaks, target input shares, and gain 0.8. The functional target had 192 regional medians from 73,560 units. Standard linear filter identities give an exact attainable time scale range for each output over all nonnegative independent white noise strengths. These ranges give a lower bound on joint model error after any common physical time scaling. Every noise pattern left at least one region with an error factor of 3.0518 or more, and at least eight regions with more than twofold error. If noise strengths differed by at most tenfold, the bounds rose to 6.5316 and 55 regions. A stricter 92-region panel kept a 2.541-fold bound. Yet noise at one selected source reversed rank correlation from -0.632 to +0.738. Thus, rank agreement can coexist with unavoidable calibrated error. The bounds apply to the fixed models and source point estimates. They do not test shared or colored noise, variable leaks, or other maps. A regional model state and a median neuron time scale are only coarsely comparable. We preserve the proof, new code, source provenance, full model coefficients, results, figures, numerical checks, and repeat instructions.

connectome · intrinsic time scale · independent noise · linear stochastic model · mouse · structure and function · model error bound