The impact of velocity update frequency on time accuracy for mantle
convection particle methods
Abstract
Computing the velocity field is an expensive process for mantle
convection codes. This has implications for particle methods used to
model the advection of quantities such as temperature or composition.
A common choice for the numerical treatment of particle trajectories
is classical fourth-order Runge–Kutta (ERK4)
integration, which involves a velocity computation at each of its
four stages. To reduce the cost per time step, it is possible to
evaluate the velocity for a subset of the four time integration
stages. We explore two such alternative schemes, in which velocities
are only computed for: a) stage 1 on odd-numbered time steps and
stages 2–4 for even-numbered time steps, and b) stage 1 for all time
steps. A theoretical analysis of stability and accuracy is presented
for all schemes. It was found that the alternative schemes are
first-order accurate with stability regions different from that of
ERK4. The efficiency and accuracy of the alternate schemes were
compared against ERK4 in four test problems covering isothermal,
thermal, and thermochemical flows. Exact solutions were used as
reference solutions when available. In agreement with theory, the
alternate schemes were observed to be first-order accurate for all
test problems. Accordingly, they may be used to efficiently compute
solutions to within modest error tolerances. For small error
tolerances, however, ERK4 was the most efficient.