finding Cepheids through GDR2

Goal:

can one use GDR2 only to select a type I Cepheid sample for targetting; the basic idea is: 

a) they must be luminous absK<0, they must vary >0.3mag peak-to-peak, and they are blueish.

Dust extinction is the bain, of course.

How can one define 'photometric variability' in GDR2?

Via the photometric noise: sqrt(g.phot_g_n_obs)/g.phot_g_mean_flux_over_error

across the many epochs:

This depends on G-band magnitude, but for G<18 the "actual photon noise" is small:

There are many sources with excess noise, which at the bright end turns out to be most
commonly intrinsic variability:

What does Gaia DR2 "variability" quantify?

that looks good! For a sinusoid: 4*rms = peak-to-peak.

Query:

SELECT top 70000 * , sqrt(phot_g_n_obs)/phot_g_mean_flux_over_error as variability

FROM gaiadr2.gaia_source

WHERE

bp_rp < 2.

and

sqrt(phot_g_n_obs)/phot_g_mean_flux_over_error > 0.08

and

phot_g_mean_mag < 17.

and

b between -10. and 10.

and

phot_g_mean_mag - 1.75*bp_rp < 13.

and

parallax+parallax_error < power(10.,(10.-(phot_g_mean_mag - 1.75*bp_rp))/5.)

Rationale: 

-- vary by >0.32mag peak-to-peak

-- have a 'predicted' W1mag (== phot_g_mean_mag - 1.75*bp_rp) < 13.

-- have an W1 abs mag < 0: parallax+parallax_error < power(10.,(10.-(phot_g_mean_mag - 1.75*bp_rp))/5.)

That yields 27000 candidates, of which 6000 have good astrometry...

Of the 700 GaiaDR2 Cepheids, 95% get picked up that way; the rest is all typeII Cepheids.

The query output is named Cepheid_searches_GDR2only_v3-result.fits
and can be found here:
https://www.dropbox.com/s/7xwsnvruk8an58w/Cepheid_searches_GDR2only_v3-result.fits?dl=0

This is what the distribution looks like  [NB: we can do better with the X-axis by taking a spectroscopic-survey-trained estimate of Teff, derived from Gaia,2Mass & WISE photometry]

In this plot, I have used (J-K) - 0.25*(G-K) as a self-dereddened color; Ideally, I'd like to have Teff as the X-axis. YST to the rescue?

Now we need to look at the contamination by other types of variable stars (RV Tau, W Vir, RRL):

We do this by looking at "dereddened color" vs "abs. mag." (however lousy):

and compare this to the Cepheids from Gaia DR2 (in blue):

If we then plot the lump centered on (0.2,-5), 825 sources, they look like this

And the possible contaminant's sky distribution looks like this:

are these stars (at (0.1,0) in the color-absmag plane) where the instability strip crosses the main sequence? (Delta Scuti?)

Update January 25, 2019

I have done a broader candidate selection, Yuan-Sen Ting has then estimated their T_eff, as follows: we train a neural net on all APOGEE stars to predict T_eff(APOGEE) from BP,G,RP,J,H,K,W1,

and apply it to the candidates. Initial cross-validation indicates a precision of ~250-300K; in the range 4000K t o 8000K.

With this, one gets a candidate set that looks like this:

This shows the different classes of variables even more nicely: luminous red variables at 3500K, (presumed) RV Tau (at 4600K), and (presumed) RRL (& beta Ceph, whatever) at (6200K,0).

Comparison with the Gaia DR2 paper Cepheids (black) shows where the classical Cepheids should lie (and shows Gaia DR2's misclassification rate).

That suggest to select (in a more stringent fashion) like this

which leads to an on-sky distribution like that:

This sample selection includes (>90%) if the Gaia DR2 Cepheid I, and basically all Cepheid I selected by variability from WISE in a recent 2018 paper (incl de Grijs,  check reference).
It will be interesting to see what the stars near the GC (|l|<45) are.

Addendum (Feb 18, 2019)

HWR is discovering that there are analogous variability measures in BP and RP. Taking the 

"Cepheid candidate take 7" sample, it's fun to look at the distribution of the ratio of variability

in BP and RP. This should be followed-up.

Note that this is a funky X-axis -- the sqrt(phot_rp_n_obs) is missing. The unmarked stripe  at (0.015,1.5) are RV Tau (what the f...).