browndwarf / protostars Goto Github PK

fun modeling project.

What is the equilibrium distance for a given dust grain for the re-radiation disk temperature we derive?

My hunch is that the power law is so steep and correlated with A_K (and other params) that the chains will mix really slowly.

Possible Ideas:

1. Teff and logg compared to theory and observations of Class I, II, and/or III young stars
2. Spectrum with example composite model (optionally extinction curves, etc)
3. Distribution of inferred solid angle ratio
4. Distribution of r_K
5. Corner plot for a subset of other interesting parameters?
6. Visualization of priors over the posteriors?
7. Constraints from independent information-- photometry, etc.

We want to show where Class 0 protostars should appear on a diagram of logg versus effective temperature, in comparison to Class III, Class II, and Class I sources.

Inspired by a tweet from Adrian Price-Whelan:

2048 @APOGEEsurvey spectra (rows) stacked into an image, sorted by [Fe/H] #visualizingtheexpected #dte17 #wtfdata https://t.co/vtKLyzz0Xg pic.twitter.com/fbK7cMRDV7

— Adrian Price-Whelan (@adrianprw) March 1, 2017

Except we would be using model spectra, not real data.

The Starfish star.py code would be renamed to star_{*project*}.py

So for example we would have:

$protostars/code/star_protostars.py --samples=5000 

instead of having a whole bunch of star_{misc_descriptor}.py (e.g. star_BB.py ) all living in the $Starfish/scripts/ directory.

There are tradeoffs to both approaches, but I think this revised strategy is closer to the customized work I am doing for multiple different projects.

The Right Thing To Do is to rewrite Starfish with a base model class from which I can inherit and extend functionality, but in practice many of these experiments are one-offs right now, and so good design is probably not urgent. Eventually the experience and insight I gain from these experiments will give me a clearer vision on how to do this right.

uniform over 1.7 - 2.0

Should we implement an Av?
I'm of two minds about this.

Some tradeoff considerations:

1. The Av value will not be significant because the emergent spectrum comes from scattered light, which introduces a non-standard spectral shape, hampering accurate retrieval of Av anyways.
2. More evolved stellar systems will not suffer from this problem as much, so we might as well implement it anyways.
3. We can use the Av value it gives us to demonstrate that it's unphysical and therefore verify the extent to which scattered light is playing a role.
4. Of course we don't really know the overall spectral shape or if the models are accurate in the absence of scattering and extinction anyways, so there's really no point right now to use Av and not just let the Chebyshev polynomials pick up the slack.

It'd be really neat to show a comparison of what we can do now with Keck, and what JWST will be able to do. This should be considered a bonus since it's basically fake (ie synthetic) data, but could be neat.

This exercise will result in highly degenerate posteriors of solid angle, etc.

Interestingly, we might be able to put a prior on the solid angle ratio-- the disk should be larger than the Star! Of course geometrical considerations could limit the strength of the prior we can assign-- a thin ring of disk inner wall could be consistent with an apparent area less than the (proto)stellar surface area.

The model is now a stellar photosphere plus a black body:

net model = stellar photosphere + BlackBody(T, Omega)

By "most righteous" we mean including all the salient physics and instrumental nuisance parameters that we can think of, with the most information-rich portion of the spectrum.

This is a re-do of Issue #4 but with the solid angle flux-correction we figured out in Issue #5.
I will also adjust the starting parameters to match the posteriors from the Experiment 2 Run02.

Maybe logg versus veiling, or Omega ratio, or effective temperature-logg, or something that shows some differences (or similarities) among the different experiments we've done.

The Simbad-able name is: ** SH 2-68 N**

Summarize all data analysis so far in a document with figures and some text.

Right now star_BB.py computes the relative flux ratio of the star to disk like so:

F_bol1 = self.F_bol_interp.interp(p.grid)
F_bol2 = stef_boltz * p.T_BB**4.0 ## Radiance from black body
self.qq = F_bol2/F_bol1[0]

and then computes the net model like so:

model1 = self.Omega * (self.chebyshevSpectrum.k * self.flux_mean + X.dot(self.mus))
# Black body
model2 = self.Omega2 * self.qq * self.BB_lam.value * self.chebyshevSpectrum.k

# Return a "metadata blob" of both spectrum models
raw_models = [model1, model2]

net_model = model1 + model2

The absolute values of the solid angles have no meaning since we do not have absolutely flux-calibrated spectra. However the relative value of the solid angles is physically meaningful. How much bigger is the disk than the star (since they share the same distance).

However, the posterior values I'm getting for logOmega2 are much less than those for logOmega, suggesting that something is wrong in the co-factors I'm assuming for the relative flux ratio of the Phoenix models and Black Body.

TODO's:

• Assess the units of Phoenix and astropy's Black Body.
• Revisit the F_bol assumptions

This model adds a DC component to Starfish:

model = stellar_photosphere + constant

Another protostar

2.20 - 2.40 um (or 2.395 or whatever)

could be useful

• What is the wavelength range of the NIRSPEC-7 filter? Website says 1.84 - 2.63, but the short wavelength portion is inaccessible. Telluric atmosphere is present here.

• Explore metadata from Keck Archive
• Examine the spectrum of S68N
• Calculate and verify the spectral resolution and sampling
• Save ascii spectrum to `HDF5 for Starfish

Will be useful for talks and maybe the paper.