Adam Jermyn

I suppose the counterargument would be that maybe the solver learns something by evaluating with a small but non-negative L, and we'd be cutting off that iteration prematurely by raising...

Hmmm actually thinking more, can't there be physical setups where L < 0(temporarily) at the surface? I'm thinking of irradiated stars...

I don't think photospheres make sense in that case...

Like in that case the radiation field doesn't become strictly outward-facing at infinity, and I think you need that to define a photosphere?

Agreed. I don't think get_phot_info is the right place for this change. -Adam On Mon, Jun 21, 2021 at 11:31 AM, Warrick Ball ***@***.***> wrote: > That's what I'm thinking...

Numerical jacobians will be computed with finite differences, which loses ~most precision right off the bat. Small differences then get magnified quickly... -Adam On Tue, Oct 05, 2021 at 9:54...

Right, but imagine if those two large numbers are different. Then the subtraction is doing the same thing, but it's amplifying a small difference still. -Adam On Tue, Oct 05,...

Note that those `pow` calls aren't to integer powers. Under the hood CRMATH implement those as calls to `log`, a multiplication, and a call to `exp` (more or less). So...

Just FYI I've removed `cash_karp` now that we no longer use it anywhere.

Added labels 'rainy day' and 'good first issue' (subject to some experience with numerical analysis).