SFH/sSFR prior undercoverage

[Nikhil0504]

Check if the prior is being under sampled for the outliers or if it actually can't be fit.

[changhoonhahn]

You should provide some text or explanation instead of using github issues as a todo list.

[Nikhil0504]

12 worst continuum only outliers using Spender 10D with a flow trained for finding the outliers.

1. 39627752139128905 logp=-67.31
2. 39627788436637535 logp=-66.88
3. 39628449266011446 logp=-65.74
4. 39633322460057381 logp=-59.74
5. 39633127722713135 logp=-53.67
6. 39627896930701693 logp=-49.84
7. 39633140817331167 logp=-48.03
8. 39627769688098973 logp=-47.34
9. 39633428001328544 logp=-45.35
10. 39627847618267567 logp=-45.15
11. 39627854840856744 logp=-44.33
12. 39627842731903789 logp=-44.27

[Nikhil0504]

Running some MAP tests for one of these galaxies above (TARGETID=39633140817331167)

1. Using just a pure power law - dust_type=0: reduced chi sq=65.6, DN_4000:1.58
2. Using KC13 dust law - dust_type=4: reduced chi sq = 11.0, DN_4000:1.58

Updating the dust law certainly fixed the slope in the redder end but I wasn't able to reproduce the DN_4000 dip well.

[Nikhil0504]

@changhoonhahn
So, I engineered more older SFHs to see if I can reproduce the spectra better and get a deeper DN_4000 break.

Seems like older SFHs could reproduce the break but the reduced chi square is around 4.9 for all the runs with the engineered fits.

[Nikhil0504]

high res and binned spectra

[changhoonhahn]

@Nikhil0504 what do you mean by

high res and binned spectra

What is the difference?

It seems the more quiescent SFHs address the discrepancy around ~4000A. The other discrepancy is the offset around 6500-7200A. Any thoughts on what could be causing that?

[Nikhil0504]

What is the difference?

high res is just higher resolution plot with dpi=300

[changhoonhahn]

@Nikhil0504 putting this here so we don't forget.

• Does increasing the lookback time resolution of the SFH (more bins for cosmic time < 11) help with the discrepancy in the continuum at 6500-7200A?

[Nikhil0504]

Having more bins for cosmic time < 11 doesn't seem to help with the discrepancy in the continuum at 6500-7200A

[changhoonhahn]

@Nikhil0504 My takeaway is the opposite. old x5 (orange) seems to boost the SED from 6600-7300 closer to the observed spectrum. Am I misunderstanding the figure?

It's hard to tell if the fit is worse at >8500A because of the legend but otherwise I don't quite understand how chi2 can be higher for oldx5.