Physicist’s Summary Asher Kaboth 21 Sept 2016
Thank you! ๏ Thank you to the organizers! ๏ Thank you to the panel members for the interesting discussion! ๏ Thank you to the attendees for all your contributions! 2
Reminder PhyStat- ν Kashiwa has an in-progress summary document of the discussions there: www.hep.ph.ic.ac.uk/~yoshiu/PhyStat-nu- IPMU-2016-Summary-Draft Let’s think about a summary document for this meeting! 3
Pictures of Cute Animals are Obligatory ν μ ν e ν τ 4
6 A ToDo List Possible Future Neutrino Prizes: • Nature of the Neutrino David Moore (Majorana (2) v Dirac (4) ) • Observing CPV in Neutrino Sector ( sin δ 6 = 0 ) Pilar Coloma, Christopher Backhouse, Shao-Feng Ge • Demonstrating the Existence of the Sterile Neutrinos Aixin Tan, Zarko Pavlovic • Observation of New Physics in Neutrino Sector? Neutrino Decay, Non- Standard Interactions, ..... • • A convincing Model of Neutrino Masses and Mixing with confirmed predictions. Everyone, basically! 5
Starting Point Almost here! •One thing I learned: • collaboration might converge on high-level statistical procedure. Put in likelihood / probability model and turn the crank. • Practical improvements to analysis mainly lie in techniques used for modeling the data ! (eg. systematics, ND->FD extrapolation, etc.) • Useful to factorize discussion & software in terms of modeling and high-level statistical procedure 2 This is still a good idea! 6
Oscillation Analyses Statistical Issues for the Long-Baseline Neutrino Experiment Analysis Techniques Solar Neutrino Researcher Statistical Approaches for IceCube, DeepCore, and PINGU Neutrino Oscillation Analyses PhysStat- ν Joshua Hignight Christopher Backhouse for the IceCube-PINGU Collaboration n E M A C California Institute of Technology C S T I S T I T A S G O A L T C A February 5, 2015 September 21 st , 2016 Scott Oser University of British Columbia PhyStat- n 2016 September 21 st , 2016 Joshua Hignight PhyStat- ν Fermilab 2016 1 / 20 September 20, 2016 C. Backhouse (Caltech) LBL analysis February 5, 2015 1 / 30 S e n s i t i v i t y t o C P v i o l a t i o n i n n e u t r i n o o s c i l l a t i o n e x p e r i m e n t s P i l a r C o l o m a Short-baseline F e r m i l a b analysis techniques Statistical Methods used in B a s e d o n : B l e n n o w , P C , F e r n a n d e z - Ma r t i n e z , a r X i v : 1 4 0 7 . 3 2 7 4 [ h e p - p h ] Reactor Neutrino Experiments J H E P 1 5 0 3 ( 2 0 1 5 ) 0 0 5 Zarko Pavlovic P h y s t a t - n u Wo r k s h o p Xin Qian F e r m i l a b t h S e p 1 9 , 2 0 1 6 BNL PhyStat-nu Fermilab 2016 1 7
Good Points ๏ It looks like most experiments consider their approximations! ๏ There’s a wide variety of methods, frequently on the same experiment 8
Things to Work On ๏ My biggest request: show the diagnostics! ๏ There’s lots of algorithms:MCMC, F-C, MultiNest, etc ๏ Diagnostics for each are different, but all important ๏ What do we communicate to the future? 9
Consensus? ๏ We’re pretty much on the right track! ๏ Treatment of systematics is important here, especially in model tests 10
Unfolding ๏ Lots of discussion here! ๏ What to do in different situations? https://arxiv.org/pdf/ 1607.07038v1.pdf 11
Cross Section Unfolding 3 3 × 10 × 10 Statistical Errors Only Statistical Errors Only 1.4 2 MINER A Tracker CCQE MINER A Tracker CCQE ν • ν → 1.4 ν • ν → Events / 0.05 GeV 1.2 1.2 2 Events / 0.05 GeV 1 1 0.8 0.8 Data Data 0.6 Monte Carlo Monte Carlo 0.6 0.4 0.4 POT Normalized POT Normalized 0.2 0.2 1.01e+20 POT 1.01e+20 POT 0 0 0 0.5 1 1.5 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2 2 2 2 Reconstructed Q (GeV ) Q (GeV ) QE QE high x = more elastic 12
Daya Bay Unfolding simpler usage Stat+Sys 13
Consensus? ๏ My sense is that there’s a preference for not unfolding—and if doing so, show more diagnostics ๏ There should be more investment by experimentalists in providing information to outside the experiment to go from physics to detector quantities 14
Comparing Models Example of model-dependent NME uncertainty: This shows up in a number s of places! Several different PRD 92 , 012002 (2015) techniques, but problems with inputs, too. t s” Workshop Idea econcile MiniBooNE, MINERvA ees of : [QE: PRD93 no.7, 072010 , and P is 15 arXiv:1507.08204 ce,
Generative Modeling http://indico.ipmu.jp/indico/getFile.py/access? contribId=22&sessionId=5&resId=0&materialId=slides&confId=82 Fundamental Theory Auxiliary Theory Detector Effects Data Summary Treat all of these probabilistically 16
• Conceptually: Prob(detector response | particles ) • Implementation: Monte Carlo integration over micro-physics •Consequence: cannot evaluate likelihood for a given event Detector Effects Data Summary 17
New Ideas from Statisticians Post-Selection Inference Classical Inference start selected end start end selection data data data model inference model inference Post-Selection Inference Todd Ku ff ner Washington University in St. Louis Bayesian, Fiducial, and Frequentist (BFF): Best Friends Forever? PhyStat ν 2016 BFF 1/21 Fermilab Xiao-Li Meng Xiao-Li Meng Choose Your Replication! Department of Statistics, Harvard University Basu Ex Summary Liu & Meng (2106) There Is Individualized Treatment. Why Not Individualized Inference? Annual Review of Statistics and Its Application , 3: 79-111 Liu & Meng (2014). A Fruitful Resolution To Simpson’s Paradox via Multi-Resolution Inference. The American Statistician , 68: 17-29. Meng (2014). A Trio of Inference Problems That Could Win You a Nobel Prize in Statistics (if you help fund it) . In the Past, Present, and Future of Statistical Science (Eds: X. Lin, et. al.) , 535-560. 18
Final Thoughts ๏ It’s so great to see the neutrino community discussing and integrating these issues! ๏ Clearly combinations, unfolding, and systematic uncertainties are on your minds—good! ๏ Let’s keep this momentum going: ๏ Future PhyStat- ν ! ๏ Think about: does your experiment need a statistics committee? What would that look like? What are you taking back to your experiment and analysis? 19
Recommend
More recommend
