Minimisation

The ensemble Q was fitted to the posterior pdf by minimising $I_{\mathrm KL}(Q;P)$. The particular technique for minimisation is not important regarding the results; everybody can use their favourite algorithm. In these simulations, a variation of Newton's method was used. The parameters $\hat{s}(t)$ and $\tilde{s}(t)$ were iterated 15 times for each measurement vector x(t). Other parameters were updated after going through all the measurements, and this was iterated 200 times. The number of iterations was chosen conservatively. The details of the minimisation procedure will be given in [10].