Package neurord.numeric.stochastic Description
These are the functions for the innner loops of the calculation. Once the algorithms
are settled, they should probably be hand-converted to c for better performance.
The main class is the StepGenerator that generates the number of particles that
make a particular transition given the total number, the probability of one particle
going, and a random number.
There are two flavours - the DiscretePStepGenerator delegates to a set of
FixedPStepGenerators according to the value of p. This is applicable if the geometry,
reaction scheme and timestepping are controlled in such a way that there are
only a relativeley small number of different probability values in play (up to
a few hundred say). The interpolating step generator works for continuous p,
but is rather slower (though there is plenty of scope for optimization).
Each case ends up in a NGoTable which stores the cumulative probability and does the
lookup.
The mixed stochastic/continuous calculation takes a ReactionTable for the reactions and a
VolumeGrid for the morphology and does the calculation. For high number densities the
update algorithm should use Dufort-Frankel. For lower densities, various exact and
approximate stochastic methods a la Blackwell.