Jw. Sandilands et Jc. Mcconnell, EVALUATION OF A REDUCED JACOBIAN CHEMICAL SOLVER, JOURNAL OF GEOPHYSICAL RESEARCH-ATMOSPHERES, 102(D15), 1997, pp. 19073-19087
To allow the use of detailed chemical modules in three-dimensional glo
bal forecast and climate models, it is necessary to have efficient and
accurate solvers for the chemical species that allow for diurnal forc
ing. The aim of this work was to develop a fast chemistry solver based
on the implicit Euler finite difference approximation to the chemical
continuity equation for a set of chemical species. This finite differ
ence form of the continuity equation is appealing because of its mass-
conserving properties. We employ two algorithms for solution: a basic
Newton solver and a modification of this method, which we have dubbed
the reduced Jacobian (RJ) solver. Both methods yield equivalent soluti
ons, but the RJ solver can be 10 times more efficient on a scalar mach
ine for fixed time stepping. Both methods are evaluated for accuracy a
nd timing against the Gear solver. The RJ algorithm breaks the Jacobia
n matrix into smaller units that may be solved more efficiently. To do
this, we have made use of the interactiveness of ''family'' species.
The reduction in computational time makes this method a good candidate
for three-dimensional modeling of atmospheric chemistry. In this pape
r we discuss the method and present a chemical box model comparison of
these three techniques for a chemical set comprising O-x, HOx, N-x, C
lOx, BrOx, and CH4 oxidation gas-phase chemistry, at various heights i
n the atmosphere ranging from the upper troposphere to the mesosphere.
We have also tested a modification that reduces errors at sunrise and
sunset without seriously compromising the efficiency of the method. F
rom our tests it would appear that the RJ solver is robust at all heig
hts tested and should be readily adapted to other chemical mechanisms.