CENTER FOR COMPUTATIONAL MATHEMATICS COLLOQUIUM
UNIVERSITY OF COLORADO AT DENVER
TITLE: Splitting Methods for Problems with Different Time Scales
SPEAKER: Jerry Browning, Colorado State University and NOAA Forecast
Systems Laboratory (joint work with Heinz Kreiss, UCLA)
DATE: Wednesday, October 18, 1995
PLACE: Math Conference Room - Suite 540
UCD Building, 1250 14th St., Denver
TIME: 2:30 pm (Refreshments served at 2:15 pm)
ABSTRACT
The time step for the leapfrog scheme for a symmetric hyperbolic system
with multiple time scales is limited by the Courant-Friedrichs-Lewy condition
based on the fastest speed present. However, in many physical cases, most of
the energy is in the slowest wave, and for this wave the use of the above time
step implies that the time truncation error is much smaller than the spatial
truncation error. A number of methods have been proposed to overcome this
imbalance, e.g. the semi-implicit method and the additive splitting technique
originally proposed by Marchuk with variations attributable to Strang, Klemp
and Wilhelmson. An analysis of the Marchuk splitting method for multiple time
scale systems shows that if a time step based on the slow speed is used, the
accuracy of the method cannot be proved and in practice the method is quite
inaccurate. If a time step is chosen that is between the two extremes, then the
Klemp and Wilhelmson method can be used, but only if an ad hoc stabilization
mechanism is added. The additional computational burden required to maintain
the accuracy and the stability of the split-explicit method leads to the
conclusion that it is no more efficient than the leapfrog method trivially
modified to handle computationally expensive smooth forcing terms.
Using the mathematical analysis developed in a previous manuscript,
it is shown that splitting schemes are not appropriate for badly skewed
hyperbolic systems. In a number of atmospheric models, the semi-implicit
method is used to treat the badly skewed vertical sound wave terms. This
leads to the excitation of the high-frequency waves in a nonphysical manner.
It is also shown that this is equivalent to solving the primitive equations,
i.e. a model using this method for the large-scale case will be ill-posed at
the lateral boundaries. The multiscale system for meteorology was introduced
by Browning and Kreiss to overcome exactly these problems.