Galactic Archaeology: Summary and future outlook for #Ken75

The last talk at GASP 2016 Galactic Archaeology and Stellar Physics

Transcript

1. summary and future outlook David W Hogg
 (NYU) (MPIA) (Flatiron) (thank you, Ken Freeman & Gayandhi De Silva)
2. Part 1 (positive things)
3. the importance of external galaxies (Zasowski, Sakari talks) A few measurements in another galaxy of, say, the alpha/Fe bimodality (Sakari?), would completely change the set of available explanations. The best information about the Milky Way might come from another galaxy.
4. Part 2 (I have now said all the positive things I had to say.) (just kidding) (sort of)
5. accuracy vs precision (or how I met Ken Freeman)
6. story time vs measurements (I’m an outsider; and a methodologist.) It’s about to get scoldy time in here.
7. why take yet more data? What questions can be answered with 800,000 stars that can’t be answered with 300,000 stars? eg: Take 100,000 more stars in the MW halo or take 200 in the M81 halo?
8. so many stars, so many surveys* Do we really need 4 enormous, independently running surveys? *I work on this too, so I am also guilty
9. exploration vs exploitation Exploit only when questions are mature.
10. Ken wants a story! (He said so.) But that story should be a hypothesis generator,
 and we should use it to make measurements. We can’t take 1,000,000 8000-pixel spectra
 at SNR of 100 in search of a story.
11. key: alpha/Fe bimodality (Freeman, Chiappini, Toyouchi, Kobayashi talks) Is it weather or nucleosynthetic fundamentals? (see my external-galaxy comment)
12. key: radial migration (Freeman, Bovy talks) But why are we talking about what’s consistent with the picture, and not measuring the rates? (This doesn’t need to be story time.)
13. chemical vs spatial decompositions (Freeman, Bovy, Chiappini talks) If we had well-posed questions, we wouldn’t have the confusions that these differences generate.
14. terminology matters “thick” == “alpha rich”? “chemical tagging” Martell vs Freeman
15. photometry vs spectroscopy* (Ruiz-Dern, Feuillet, Starkenburg,
 DaCosta, Schlaufman talks) A lot of degenerate one-sigma measurements can turn into something very valuable. How far can we go, Mr. Ting?
16. aside: open data and code It is now undeniable that early, public release of code and data creates more citations and successes for the builders than any other policy. (No, putting the labels online is not a data release.)
17. nucleosynthesis vs the data How should nucleosynthetic pathways affect the ways we measure and analyze chemical abundances? Can abundance data be used to set nucleosynthetic parameters directly? (Hampel, Côté talks)
18. machine learning PCA, t-SNE, k-means are great for visualizing your data, finding outliers (Žerjal, Travern talks). They will not bring long-term scientific insights.
19. data-driven models Capitalize on what’s good about machine learning, but constrain it to make use of our knowledge.
20. combining and synchronizing data* (Ness, Hekker talks) Data-driven approaches; ideas from self-calibration.
21. dealing with nuisances The only really justifiable use of data-driven models: code approximations, line lists, log(gf) values, continuum, spectrograph calibration, etc (Lind has a good list of these!)
22. aside: utility (Binney talk) What we do depends on some product of its importance and its easiness. #LTFDFCF
23. analytic models vs simulations* Right now we only know how to do parameter estimation with rich data in simplified models. The theory really is computational. New methods for inference might be necessary. (Binney, Nataf, Gerhard, Athanassoula, Wegg talks)
24. distribution functions can ruin your whole day* Maybe even complex galaxies have simple distribution functions? (Binney) A goal for the future (Gaia): Marginalize out the distribution functions.
25. balance of resources Simulations are outrageously expensive and complex. Data sets are too. Comparisons are qualitative?
26. time domain* (This is the future for US astronomy.) Do asteroseismologists really need to be in space?
 (Stello, Huber talks) Are 3D NLTE time variations really uninteresting?
 (Lind talk)
27. gravitational radiation (I hope this is the future for US astronomy!) (Côté talk)
28. the two questions of chemical tagging* (Blanco-Cuaresmo) Two stars that were born together:
 How different will they be? Two stars that were born apart:
 How similar can they be?
29. birthday paradox It is overwhelmingly likely that two people in this room share the same birthday. (Duh. Why the hell am I saying that?)
30. chemical tagging: the pessimistic regime* (Ting talk) How we use chemical information depends critically on the detailed answers to the two questions.
31. probabilistic reasoning The prize will go to those willing to take and propagate extremely noisy information. (It always does.)
32. my themes make measurements, test hypotheses embrace probabilism and live with uncertainty
 (re-cast chemical tagging) re-evaluate the spectroscopic data taking
 (other galaxies, other modes)
33. thank you, Ken Freeman