smoh / gaia-comoving-stars Goto Github PK

@davidwhogg - can you add a bulleted list to the discussion of anything you might want to discuss? @smoh or I will then fill in text.

I think we should be using this when we make our dumb plots (ie, our plots that just use 1/parallax as distance).

Tremaine asked me again today if there are any stars in the catalog consistent with being co-moving with the Sun. This is just a reminder for either of us to visualize / look at stars that have proper motions consistent with zero.

Brainstorm in comments below.

It just takes me back to the web viz!

could be used to test method.

We need to correlate our positions with the positions of known clusters.

We also need to detect new clusters.

Do we have a stream?

@timothydmorton - can you move any modules you add to the package to a subdirectory gwb/starmodel/

de Zeeuw catalog from Hipparcos

Here is a preview summary of the population of fitted binaries (only 1290, not 1842, so some of them failed; will need to check why).

And here is an attempt at a control sample (1842 stars randomly chosen from TGAS):

A lot of the singles seem to want to be ~10^5 years old; not sure why. For this reason, I think it would be good @smoh if you sent me a matched list of stars following similar cuts (e.g., parallax SNR) as the binary sample, but just ones that don't have co-moving companions. I think that would make a better control sample.

But apart from this, there does seem to be an excess of relatively young stars (~10^7 years) in the binary sample. But there is also weird structure in the age distribution, so I'm not sure what all is going on. Also remember that these are just scatter plots of median posterior values---I'm happy to take any suggestions for better visualizations as well.

Drop (d*pomega - 1) from y, etc.

ask hogg for details

• cite http://adsabs.harvard.edu/abs/2011ApJS..192....2S
• emphasize no cut-off for wide binaries separation
• add sentence to caption of false-positive rate plot to say that we are over-estimating because of projections

How are we dealing with the systematic uncertainty in parallax (e.g. as documented here and more vaguely in the DR1 documentation)? It looks like for now (from TGASStar.get_cov) we're taking the covariance matrix as is from the table. Perhaps this is OK since we are just interested in relative parallaxes for the purposes of identifying close pairs, but I just realized that I have also been fitting my starmodels without inflating the parallax uncertainty. Any suggestions on the best way to correct this? The Stassun & Torres paper suggest a systematic offset of -0.25 +/- 0.05 mas.

Do we include stellar parameter info inferred from photometry in Paper 1 about the candidate sample? My gut says "yes" since it seems like it is fairly easy for @timothydmorton to run on a sample with the numbers we expect, but it could also move to the population inference paper. Any strong opinions?

@adrn and I confirmed that the positions in the TGAS catalog have epoch of 2015. By my understanding, this means that in order to compare coordinates to, e.g., a catalog with J2000 coordinates, I will have to adjust both ra and dec by 15*pm mas.

The TGAS also has a tycho2_id column. So, bypassing any step of catalog querying, I can look up positions in the Tycho-2 catalog by ID, and compare those with the proper-motion-rewinded TGAS coordinates, and they should match up.

When I do this, and calculate the separation angle between the TGAS ra/dec and the Tycho-2 ra/dec for the matched sources, this is what they look like:

So this is not bad, but I would feel better if it were better. I will proceed from here, but I just wanted to note this in case anyone had thoughts.

Interestingly enough, the Tycho-2 catalog (as obtained from Vizier) has both (RAJ2000,DEJ2000) columns and (RA_ICRS, DE_ICRS) columns. The following is the separation distribution when I compare evolved TGAS coords to these ICRS-labeled columns:

When we have candidate binary pairs, the most convenient way for me to receive them would be just a set of (i,j) indices, (with i < j to avoid duplicates), where the indices refer to row numbers from the entire TGAS table. I'm preparing to receive such a list so that I can quickly turn around the StarModel fits.

I will also need a similarly sized list of apparently single stars, for me to fit as a control sample, to see if there are systematic differences in age between these two samples.

• submit to AAS journals, preferred journal = ApJ (you can't submit directly to ApJ anymore), put Spergel under "who is handling page charges" 💰💰💰
• make this repository public
• submit to arxiv, link to web visualization in comments (if it's ready)
• lunch is on me Hogg! (by which he probably means Jim Simons)

Add Simons Foundation to Hogg's affiliation list.

1. Gaia sprint
2. Gaia team

There’s a bug in line 58-60 in likelihood.py — I think it is better to set idx by hand as it is third row for every star for RV (If I’m reading things right).
When d is very large, because there is /=d**2 in line 49 of get_Cinv, Cinv_dirty can be very small, which will get deleted, and Cinv is now empty matrix.
This happens for star 2 and star 299524 in stacked_tgas.

/Users/semyeong/projects/gaia-wide-binaries/gwb/likelihood.pyc in get_Ainv_nu_Delta(d, M_dirty, Cinv_dirty, y_dirty, Vinv)
    59     Cinv = np.delete(Cinv_dirty, idx, axis=0)
    60     Cinv = np.delete(Cinv, idx, axis=1)
---> 61     _,log_detCinv = np.linalg.slogdet(Cinv/(2*np.pi))
    62 
    63     M = np.delete(M_dirty, idx, axis=0)



/Users/semyeong/anaconda2/lib/python2.7/site-packages/numpy/linalg/linalg.pyc in slogdet(a)
  1704
    """

  1705     a = asarray(a)
-> 1706     _assertNoEmpty2d(a)
  1707     _assertRankAtLeast2(a)
  1708     _assertNdSquareness(a)



/Users/semyeong/anaconda2/lib/python2.7/site-packages/numpy/linalg/linalg.pyc in _assertNoEmpty2d(*arrays)
   220     for a in arrays:
   221         if a.size == 0 and product(a.shape[-2:]) == 0:
--> 222             raise LinAlgError("Arrays cannot be empty")
   223 
   224 



LinAlgError: Arrays cannot be empty

tracebacks:

ipdb> �d

/Users/semyeong/projects/gaia-wide-binaries/gwb/likelihood.py(87)_marg_likelihood_helper()
    85     Cinv = get_Cinv(ds, data)
    86 
---> 87     Ainv,nu,Delta = get_Ainv_nu_Delta(ds, M, Cinv, y, Vinv)
    88     sgn,log_detAinv = np.linalg.slogdet(Ainv/(2*np.pi))
    89     log_detA = -log_detAinv


��ipdb> ��ds
6685.9553296605964
��ipdb> ��d

/Users/semyeong/projects/gaia-wide-binaries/gwb/likelihood.py(61)get_Ainv_nu_Delta()
    59     Cinv = np.delete(Cinv_dirty, idx, axis=0)
    60     Cinv = np.delete(Cinv, idx, axis=1)
---> 61     _,log_detCinv = np.linalg.slogdet(Cinv/(2*np.pi))
    62 
    63     M = np.delete(M_dirty, idx, axis=0)


��ipdb> ��Cinv_dirty
array([[  7.55760737e-09,  -2.53065913e-09,   0.00000000e+00],
      [ -2.53065913e-09,   9.37824516e-09,   0.00000000e+00],
      [  0.00000000e+00,   0.00000000e+00,   0.00000000e+00]])

Like we do, say, in The Joker?

Double check that the mu_alpha's in the appendix are consistent with definitions above (e.g., https://github.com/smoh/gaia-comoving-stars/blob/master/paper/ms.tex#L922)

We want to see them on the sky (ra, dec) and in the color-magnitude diagram (absolute magnitude vs G-J or G-W1 color).

@adrn Right now, you cutoff stacked tgas catalog with a S/N cut in line 39 of generate-par-sample.py, and indicies generated from this file is for that particular S/N cut applied tgas, and we need to give compute-likelihood-ratio.py that same S/N. I think this is quite prone to mistakes... and it's better to only refer to indicies in the entire tgas catalog. This may be an issue when we pass info to @timothydmorton also. Let me know what you think.