Phase tracing issues

[jorindevandevis]

• Sometimes the phase tracing takes very long. One reason for this is that the tracer will try to follow the phase all the way to T = 0 (if it exists). In principle, there is no lower limit on the temperature that we could probe in WallGo, but in practice, one does not reach this minimum temperature. It is thus a waste of computational time to trace it all the way to T = 0, and it would be better to impose a minimum temperature. Need to come up with a good estimate for that.
• Sometimes the phase tracing proceeds without problem, but WallGo finds a run-away wall, which it wouldn't find for a different value of TemperatureScale. Need to figure out what causes this, and throw an error/warning earlier.

[niemilau]

Wasn't the first point supposed to be solved by looking at the hydro template model to find the relevant temperature range? We struggled with this already a very long time ago

[jorindevandevis]

Yeah, but I realized at some point that there is a maximum temperature (which you can estimate), but in principle no minimum.
If you want to estimate the "strongest shock" (which you need to find the smallest possible velocity for Tn), you have to set the temperature inside the bubble to zero. We already approximate this, using some minimum temperature in hydro.
In most models vmin is in fact 0, so the strongest shock is irrelevant, so we really don't need to trace down all the way to T = 0.

[jorindevandevis]

First point addressed in #283

[jorindevandevis]

Second point partially addressed in #283

Remaining issue is when temperatureScale is chosen too small. This could also result in runaway walls or the pressure computation not converging. It seems like the problem is more pressing when the momentum grid size is small.
@benoitlaurent96 or @og113 , do you have an idea what goes wrong when dT is too small? Naively, I would have thought that it just takes unnecessarily long, but that does not seem to be the case.

[benoitlaurent96]

Second point partially addressed in #283

Remaining issue is when temperatureScale is chosen too small. This could also result in runaway walls or the pressure computation not converging. It seems like the problem is more pressing when the momentum grid size is small. @benoitlaurent96 or @og113 , do you have an idea what goes wrong when dT is too small? Naively, I would have thought that it just takes unnecessarily long, but that does not seem to be the case.

temperatureScale has an impact on essentially 2 things. The first one is the step size during the phase tracing. Having it small will slow down this part of the calculation, but it shouldn't have any negative impact beyond that. The second one is the step size of the derivatives. Here, there shouldn't be any impact on the time it takes to do the calculation, but if it's too small, the temperature derivatives can be very inaccurate. So my guess would be on that second option.

[jorindevandevis]

Mostly resolved in #363