Hyperexponential coordinates #166
Conversation
There was a problem hiding this comment.
We should probably make sure this doesn't break anything and also add a test. Other than that, 👍 nice improvement.
I would like someday to fully compactify the domain and go all the way to physical infinity at
| @@ -191,6 +202,9 @@ class McKinneyGammieRyan { | |||
| Real x0_ = 0; // start point of smooth region | |||
| Real smooth_ = 0.5; | |||
| Real norm_; | |||
| Real hexp_br_ = 1000.; | |||
| Real hexp_nsq_ = 1.; | |||
There was a problem hiding this comment.
As I discovered when I tried to run things with this value, nsq=1 is dangerous (it is the default in HARMPI and what was used in Tchekhovskoy '11 afaict though).
The first derivative of the function r(X1) is discontinuous for this value, and it exacerbates the problems with passing material over the break radius. The effect may be much less if you use numerical first derivatives, as HARMPI does.
There was a problem hiding this comment.
Ah ok, I just pulled out the default values from kharma (as I interpreted them). Could you remind me what your preferred values are and I can update these defaults?
There was a problem hiding this comment.
Yep, it was default in KHARMA because I lifted that value straight from HARMPI. I've changed it but not committed the change yet. n=2 is a better default, but still suffers a degree of the same problem in my testing. Personally I think the "best" value is the highest one you can get away with, i.e. which does not spaghetti the outer zones into complete oblivion for your case. I haven't found a combination that eliminates problems moving material over r_br though.
TBH for AGN I'm trying out the atanh trick, and a couple more @Yurlungur gave me.
There was a problem hiding this comment.
Thanks -- updated to hexp_nsq = 4
Does copy and pasting count? Lol |
You changed the names! (Honestly I like your naming better, and hadn't considered adding HExp to FMKS since I'm using it so differently, so I'll probably steal it back even more literally) |
The original FMKS unit test was actually good for this -- the answer is unchanged in the region below the hyperexponential break radius. I also updated that test to sample a region outside the break radius. 2D torus runs fine for a short time, and grid looks correct. |
brryan
changed the title
|
Is this ready to go in @brryan ? |
|
@Yurlungur This is ready to go in, but I think it would be easiest to merge #165 first when everyone thinks thats ready, then I can deal with the merge conflicts and undoing the less-good fix to |
PR Summary
An alternative (going back to ~Tchekhovskoy+ 2011) to having an outer boundary near a disk is to use hyperexponential coordinates such that, beyond some break radius, zones grow radially in size very rapidly with radius.
This PR implements the hyperexponential coordinates based on how they are implemented in KHARMA.
PR Checklist
scripts/bash/format.sh.flatgcovtreatment in Maintain sanity (or madness) #165