[Carpet] checkpoint all timelevels ?

[Ian Hawke]

On Mon, 2006-09-11 at 16:49 +0200, Thomas Radke wrote:

> > Restriction may use all active timelevels (up to 3). There is no good
> > algorithm for generating past timelevels (i.e., no way of guaranteeing
> > identical results) other than doing the evolution. So they all need to
> > be stored.
> 
> But I thought that restriction isn't needed when recovering initial data 
> ? At least it's not there in our nice Carpet scheduling picture 
> http://www.carpetcode.org/doc/scheduling.pdf (page 9).

Sorry, I meant prolongation.

Prolongation will be required the first timestep after initialization.
Assuming a 2 refinement level run, there will first be a coarse grid
timestep (all fine), then a fine grid timestep (all fine), then
prolongation (ONLY FINE IF LESS THAN 3 COARSE LEVELS REQUIRED).

Essentially in the Carpet context the initialization routine has to
provide "correct" data for a number of timelevels equal to
prolongation_order_time. As this is one less than the number stored
overall, you might as well store the lot.

[In fact I can imagine scenarios where all timelevels are needed.
Suppose immediately after initialization through recovery some quantity
at analysis is needed (through interpolation) at a time not commensurate
with the coarse grid. Provided this is in the "past" and within the
bounds of the timelevels, CarpetInterp will provide this IN A NORMAL RUN
by interpolating in time using all available timelevels. So, in order
for the results to agree after recovery, all timelevels are required]

Ian

-- 
Ian Hawke <ih3 at soton.ac.uk>