Error analysis of tau-leap simulation
David F. Anderson,
University of Wisconsin - Madison
While exact simulation methods exist for discrete-stochastic models
of biochemical reaction networks, they are oftentimes too inefficient
for use because the number of computations scales linearly with the
number of reaction events; thus, approximate algorithms are used.
Stochastically modeled reaction networks often have ``natural scales''
and it is crucial that these be accounted for when developing and
analyzing approximation methods. We have recently demonstrated
fact by showing that a midpoint type algorithm thought to be no more
accurate than an Euler type method is in fact an order of magnitude
more accurate in a certain scaling--something previously observed only
through examples. I will describe the analysis performed and show
we reach fundamentally different conclusions than previous analyses.