CarpetX: fix bug in hermite prolongation#347
CarpetX: fix bug in hermite prolongation#347lwJi wants to merge 1 commit intoEinsteinToolkit:mainfrom
Conversation
rhaas80
left a comment
rhaas80
left a comment
There was a problem hiding this comment.
this will need some discussion, in particular one wants to be very sure that the coefficients are actually correct (eg were they tested to be "Hermite" like in that they give continuous derivatives?).
| 56 / T(96), | ||
| 56 / T(96), | ||
| -9 / T(96), | ||
| +1 / T(96) |
There was a problem hiding this comment.
while I agree that Hermite of same order needs more points, I suspect that we may need to (same as Carpet) read "order" really as "this many ghost points" to be able to mix operators of different orders using the same number of ghost points.
There was a problem hiding this comment.
I'm not sure. If we do that, are we forced to only mix low order of Hermite with high order other method?
|
|
||
| extern prolongate_3d_rf2<VC, VC, VC, HERMITE, HERMITE, HERMITE, 1, 1, 1, | ||
| extern prolongate_3d_rf2<VC, VC, VC, POLY, POLY, POLY, 1, 1, 1, | ||
| FB_NONE> |
There was a problem hiding this comment.
if these are actually the same as Langrangian, does it make sense to completely remove them instead and error out?
There was a problem hiding this comment.
I guess we can do that. If the use want to use 1st order Hermite, it can error out and say please use Poly for 1st order case
| assert(y1 == y); | ||
| // assert(y1 == y); | ||
| const T tol = 100 * std::numeric_limits<T>::epsilon(); | ||
| assert(std::abs(y1 - y) <= tol); |
There was a problem hiding this comment.
ah, but if we carefully chose the test problem such that roundoff is not a problem (see comment in code above) why the (hardcoded!) tol?
There was a problem hiding this comment.
Yeah. That's just a quick fix. I can take a look and see if I can make y1==y work again.
2 participants
This PR is trying to fix bug in the Hermite type of prolongation.