[MITgcm-support] OS7MP vs SOM

[Dimitris Menemenlis]

Christopher, I only have experience with OS7MP.  SOM was not implemented for
cube sphere at time that I was fishing around for a less diffusive advection
scheme.  We use it for an eddy-permitting global configuration (18-km horizontal
grid spacing) and it provides significant reduction in spurious diffusivity
relative to 33 with little performance hit but our typical configuration has
rather large, 85x85, tiles.  I am under impression that SOM requires a bunch 
more additional 3D arrays per tracer field?

Also take a look at following presentation that Alistair Adcroft made at the 
ECCO2 meeting this past January:
http://ecco2.org/meetings/2007/ECCO2_07/Jan22/Adcroft/JPL%20ECCO2%20Jan07.ppt

D.

-- 
Dimitris Menemenlis <menemenlis at jpl.nasa.gov>
Jet Propulsion Lab, California Institute of Technology
MS 300-323, 4800 Oak Grove Dr, Pasadena CA 91109-8099
tel: 818-354-1656; cell: 818-625-6498; fax: 818-393-6720


> I'm currently using the 3rd order direct space-time advection scheme with a
> flux limiter (scheme 33) in a series of eddy-resolving simulations. However,
> I'm finding that the advection scheme is doing a lot of mixing (as much or
> more than the background Laplacian diffusivity), so I'm shopping around for a
> less diffusive advection scheme. The "News" section of the web site mentions
> that two new low-diffusivity schemes have recently been implemented: the
> one-step 7th order monotonicity preserving scheme (OS7MP) and Prather's
> second order moments scheme (SOM). Does anyone have any experience with these
> two schemes? I'm especially interested in the performance trade-offs involved
> in two schemes. I know that OS7MP requires a 4 point halo, which could have a
> large performance impact in my case, since my tiles are rather small. As near
> as I can tell, SOM requires nothing more than the standard 3 point halo, but
> perhaps other performance costs creep in elsewhere?