[aspect-devel] Progress in writing the mantle convection code Aspect

[Timo Heister]

> You can see the over/undershoot goes to around 8-10% in the first few steps and decreases soon after, but it never fully disappears.

I tried a couple of things:
1. This is not because of the difference in the time discretization in
the first two steps (where we don't have two solutions to extrapolate
from).
2. This is not because of a too large time step size (CFL=0.5 gives
basically the same result)
3. Linear elements for temperature don't help.
4. Unsurprisingly, increasing beta helps. With 0.2 instead of 0.078
the overshoot is 1% instead of 10%.

-- 
Timo Heister
http://www.math.clemson.edu/~heister/