I can't find the post, I think it might have been two days ago.
Anyway the upshot was this. The function I'm using to model it looks like this:
Aexp(B(t-t0)^2 + C(t-t0)) + F
A: Essentially an integration constant, could be useful to add additional complexities like changes in how testing is conducted.
B: The parameter that 'bends the exponential'. B should be small and negitive, that way the function begins as an exponential but with a slowly decreasing time dependent R0. Initially the function looks like an exponential, later it becomes a Gausian when B dominates over C
C: Related to the initial exponential growth. Should be fit to the beginning of the sustained localized 'exponential' transmission phase. A C value of 0.9 roughly corresponds to an R0 of 2.5
t0: A lag parameter. Just to allow the function to approppiately start at the right time
F: Unphysical constant. Recomment keeping this at 0 ideally.
If you want to model the three day periodicity then you could incoperate a sinusodial variation into A (for example A=0.1cos(2pi/3t) +1 )
Originally Posted by: Quantum