Auto-regressive inference for large N

[billshoo]

In SSMKernelDPLR._setup_linear and SSMKernelDPLR._step_state_linear, it is documented that:

"Version of the step function that has time O(N) instead of O(N^2) per step, which takes advantage of the DPLR form and bilinear discretization."

Is there any reference, in the series of wonderful S4 papers, for the inference formulas used in this implementation?

Given that it is currently about 2x slower, what's the future plan for this segment of the code?