The current documentation for the aeif models does not point out two peculiarities of these models:
The default refractory period to these models is zero. This is consistent with the model definition in Brette & Gerstner, but most likely unexpected, since NEST models generally have a non-zero refractory period.
Because NEST uses an adaptive stepsize solver for the aeif ODEs with spike detection inside the update loop, aeif neurons can fire multiple spikes in a single time step if the refractory period is zero (the default). This can lead to exploding spike numbers and extreme slow-down of simulations.
Documentation should point this out clearly in note boxes and recommend users to set t_ref > 0.
The current documentation for the aeif models does not point out two peculiarities of these models:
Documentation should point this out clearly in note boxes and recommend users to set
t_ref > 0
.