davidwhogg/chemicaltangents

chemicaltangents's Issues

need a name!

Since this is a trial project, the general idea needs a name. And I mean a name like "chemical tagging" not a name like "The Tractor".

Audit the marginal likelihood math

Let's see how far Hogg's intuition got us!

add a halo component to the dynamical model

We are waiting on @HWRix for feedback on parameters...

Could we be using tracers that are too distant?

Try doing a more conservative cut on distance so that we're sure Gaia uncertainties aren't fucking us up!

make or find a Jeans methodology

required for #18

changing repo name

I am changing the repo name because the long name is way too annoying.

The new name is still a temporary name! See #1

Look for radial gradients in disk density

easy method: split data into R > 8.1 kpc and R < 8.1 kpc and compare.

Make better "reference to O" plots; refactor all this.

Plot should go from

Fe/H, X/Fe

to

O/H, X/O

and keeping the plots in periodic-table order

compare to Jeans

For any data set we can do this, we can also do Jeans.

Let's use fake data (#17) to show that we ALWAYS BEAT JEANS. or when we do, why?

Show that abundance distributions are not symmetric around the midplane

Can we SEE in the APOGEE data (like we can see in GALAH) that the abundance ratio distribution is not symmetric around the midplane? Make some simple visualizations of this, and share with APOGEE/GALAH teams to ask for input!

Visualize residuals

Show that there is little dependence at the optimum but dependencies when we move the parameters away from optimal.

Try running model with un-marginalized likelihood

Just sample over the 4 current params + the linear params + var!

plot Zn/H and other elements vs housekeeping data

In particular: scatter plots against the four SNR measures.

run on fake data

I want two fake data things:

• one-d data made to order, perfectly in line with our assumptions.
• one of Loebman's fake Gaia Catalogs.

These will demonstrate accuracy and information (Fisher, eg).

Look at the spatial dependence of the Solar position

Re-do z_sun inference for different x and y slices or boxels.

make trivial, assumption-satisfying fake data

• choose x,y positions.
• choose vertical energies and abundances from sensible choices.
• integrate for a random amount of time.
• noisify everything sensibly.
• apply dumb Galah-like selection in x,y,z.
• make plots

APOGEE RC vs. GALAH

I ran on YST's red clump stars from APOGEE and on all RGB stars with SNR > 20 in GALAH. See a few sample plots below; bug?!

[O/H]

APOGEE RC

GALAH

[Fe/H]

APOGEE RC

GALAH

Spergel comment on Jeans

DNS say: Isn't this no better than splitting the sample by metallicity and doing Jeans in each metallicity slice? I am sure the answer to that is no, this is much better! But I'd like an answer to that.

get two codes to give same answers!

somehow @adrn and @davidwhogg get different likelihood values.

does it make sense to use Ba and Li?

they are so age-sensitive...?

Al/Fe and O/Fe are inconsistent in GALAH

Demonstrate this by plotting ratios. Not sure what to do with this point.