[MITgcm-support] linear EOS and Tref

[Martin Losch]

Jean-Michel, Alistair,
I am soo stupid. I am sorry I bothered you with this. Because I am only 
interested in my statistical higher order moments, I never really 
looked at the actual fields T and W. Now it turns out that they are 
almost exactly the same between different experiments (due to finite 
round off), as they should. I should have checked that much earlier.
What was the problem:
I want second, third, and fourth order moments so I save things such as 
timeaverages of theta, theta^2, theta^3, etc. To compute the moments I 
have to do this:
2nd moment = <theta^2> - <theta>^2
3rd moment = <theta^3> - <theta>^3 - 3<theta><theta^2>
etc. Well, and what happens when you don't save your variables in 
double precision and then substract large numbers from each other? Yes, 
you are left with numerical noise. I have had this before with the 
variance of bottom pressure for the Boussinesq work, do you remember? 
Anyway, thanks for having a look, and ALWAY SAVE YOUR VARIABLES IN 
DOUBLE PRECISION!!!

Puzzle solved,
Martin

On Aug 26, 2004, at 5:27 AM, Jean-Michel Campin wrote:

> Hi Martin,
>
> As you mentioned, you should get the same results (not identical
> results, because I don't expect to have the same truncation error
> in both cases) in the ~.1oC  and ~3.oC initial temp. runs.
> So, there is something strange.
>
> Now, to rule out some hypothesis (e.g., advection scheme ???),
> what happens if you change Tref so that you have the same T-Tref
> as in the 1rst run (meaning, setting Tref= 3+1-.1 =3.9 ?) ?
> I guess you will reproduce the initial solution, but I think
> it's worth to check.
>
> Cheers,
>
> Jean-Michel