[MITgcm-support] Negative concentrations in pTracer

[Roland Young]

 > I think it's probably just normal advection noise.  See, for example,
 > http://mitgcm.org/public/r2_manual/latest/online_documents/node72.html
 > http://mitgcm.org/public/r2_manual/latest/online_documents/node81.html

I think you're right, but I'm still having trouble with this (using 
scheme 33) as I'm getting negative concentrations that are too large to 
be removed by redistribution of the negative cells to nearby postive 
cells. I was wondering whether the temperature advection scheme should 
be the same as the tracer advection scheme? Currently temperature 
advection is the default (scheme 2).

 > Typically advection schemes with flux limiters, as is 77, should
 > remove or minimize this type of overshoot, but maybe not perfectly
 > so.  You could try using 33 or 7, see if it helps get rid of
 > undershoot.

I've heard good things about scheme 7 but as I recall it requires 
OLx=OLy=4, so will probably slow down my runs considerably as I'm using 
tiles quite thin in latitude to make the zonal filter efficient (another 
story...). Is 7 that much better than the other schemes that it's worth 
trying in any case?

Are any of the other tracer options useful here? This is what I have 
currently set

PTRACERS_SOM_Advection = /* tracer uses SOM advection scheme */
                   F
PTRACERS_ImplVertAdv = /* implicit vert. advection flag */
                   F
PTRACERS_MultiDimAdv = /* tracer uses Multi-Dim advection */
                   T
PTRACERS_AdamsBashGtr = /* apply AB on tracer tendency */
                   F
PTRACERS_AdamsBash_Tr = /* apply AB on passive tracer */
                   F
PTRACERS_diffKh = /* Laplacian Diffusivity */
                 0.000000000000000E+00
PTRACERS_diffK4 = /* Biharmonic Diffusivity */
                 0.000000000000000E+00
PTRACERS_diffKrNr = /* Vertical Diffusivity */
    33 @  0.000000000000000E+00              /* K =  1: 33 */
PTRACERS_useGMRedi = /* apply GM-Redi */
                   F
PTRACERS_useDWNSLP = /* apply DOWN-SLOPE Flow */
                   F
PTRACERS_useKPP = /* apply KPP scheme */
                   F

I tried a bit of horizontal diffusion of tracers (coefficient 
corresponding to an advection courant number around 0.01) and this 
improved it for a while by reducing the number of cells with negative 
tracer concentrations, so may be an option if it can be made stronger 
without smoothing the tracer fields too much.

Roland