To find the maximum we differentiate Equation (including Equation for ) and set the derivatives to zero. This gives two sets of equations, those for the differentials with respect to
and those for the differentials with respect to
These simultaneous equations are nonlinear and coupled (remembering that the that appear in them are functions of the and the ). However they can be remarkably simplified. Equations can be rewritten
The left hand side depends on i only, so write it as .
The right hand side then becomes
which is a great simplification: for a given set of , the unknown quantities are given by the n unknown quantities .
The are given by Equation . If is zero then is 1: if not then
If these n equations are satisfied, with Equation used to define the , then all the
The method adopted by HMCLNL is (for a given set of
), to solve
equations for the
, thus giving the
via
equation . The log likelihood may then be calculated using
equation . The maximum of the likelihood may then be found
using numerical means - HMCMLL
uses MINUIT to perform this maximisation
(this is equivalent to solving equations ).
Although there are n equations , to be solved numerically, this does not present a problem. They are not coupled, and each equation clearly has one and only one solution in the `allowed' region for , i.e. the region where the are all positive, which lies between and t=1 ( being the largest of the ). t=0 is a suitable place to start, and Newton's method readily gives a solution. Special considerations apply when there are no events from one or more of the MC sources - more details of the solution can be found in [17].