Physics and Generator Tuning P e t e r S k a n d s ( C E R N T h e o r e t i c a l P h y s i c s D e p t ) C M S P h y s i c s C o m p a r i s o n s a n d G e n e r a t o r Tu n e s M e e t i n g O c t o b e r 2 0 1 3 , C E R N
What is Tuning? Theory Experiment Adjust this to agree with this 2 P. S k a n d s
What is Tuning? Theory Experiment Adjust this to agree with this → Science 2 P. S k a n d s
In Practice VINCIA … PYTHIA Real Universe “Virtual Colliders” → Experiments & Data = Simulation Codes Particle Physics Models, Particle Accelerators, Detectors, Simplifications, Algorithms, … Statistical Analyses, Calibrations → Simulated Particle Collisions → Published Measurements Events Histograms 3 P. S k a n d s
Resources Data Preservation: HEPDATA Online database of experimental results Please make sure published results make it there Analysis Preservation: RIVET Large library of encoded analyses + data comparisons Main analysis & constraint package for event generators All your analysis are belong to RIVET Updated validation plots: MCPLOTS.CERN.CH Online plots made from Rivet analyses Want to help? Connect to Test4Theory (LHC@home 2.0) Reproducible tuning: PROFESSOR Automated tuning (& more) 4 P. S k a n d s
(Test4Theory) The ¡LHC@home ¡2.0 ¡project ¡Test4Theory ¡allows ¡users ¡to ¡par:cipate ¡in ¡running ¡ simula:ons ¡of ¡high-‑energy ¡par:cle ¡physics ¡using ¡their ¡home ¡computers. The ¡results ¡are ¡submiAed ¡to ¡a ¡database ¡which ¡is ¡used ¡as ¡a ¡common ¡resource ¡by ¡both ¡ experimental ¡and ¡theore:cal ¡scien:sts ¡working ¡on ¡the ¡Large ¡Hadron ¡Collider ¡at ¡CERN. New Users/ July 4 th 2012 Day May June July Aug Sep Monday Feb 18 2013 9:28 PM 5 P. S k a n d s
(mcplots.cern.ch) mcplots.cern.ch • Explicit tables of data & MC points • Run cards for each generator • Link to experimental reference paper • Steering file for plotting program • (Will also add link to RIVET analysis) 6
Current Methods Manual Tunes Tuning done by hand/eye (few parameters and observables at a time) Common sense (and experience) → subjective judgement of importance of each observable, and tails vs averages Theoretically motivated uncertainty variations can be included 7 P. S k a n d s
Current Methods Manual Tunes Tuning done by hand/eye (few parameters and observables at a time) Common sense (and experience) → subjective judgement of importance of each observable, and tails vs averages Theoretically motivated uncertainty variations can be included Automated Tunes (Professor, Profit?) Sense and experience encoded as elaborate sets of weights + “sensible” parameter ranges → faster & “easier” than manual Does not relieve you from critical judgement Are/were ranges, weights, and observables included indeed “sensible”? Are tuning interpolations looking stable and convergent? Are there strong correlations / flat directions? Do some parameters end up at the end of their physical ranges? “Data-driven” uncertainty variations do not reflect intrinsic theory uncertainties (cf PDF “errors”!) → Systematic mis-tuning? 7 P. S k a n d s
Quo Vadis? Not only central tunes *) This is intended as a cultural reference, not a religious one Your experimental (and other user-end) colleagues are relying on you for serious uncertainty estimates Modeling uncertainties are intrinsically non-universal. Including data uncertainties only → lower bound (cf PDFs) A serious uncertainty estimate includes some modeling variation (irrespectively of, and in addition to, what data allows) 8 P. S k a n d s
Quo Vadis? Not only central tunes *) This is intended as a cultural reference, not a religious one Your experimental (and other user-end) colleagues are relying on you for serious uncertainty estimates Modeling uncertainties are intrinsically non-universal. Including data uncertainties only → lower bound (cf PDFs) A serious uncertainty estimate includes some modeling variation (irrespectively of, and in addition to, what data allows) Not only global tunes Your theoretical (MC author) colleagues are relying on you for stringent tests of the underlying physics models, not just ‘best fits’ (which may obscure “tensions”) Tuning can be done to several complementary data sets. All give same parameters → universality ok → model ok Some give different parameters → universality is breaking down → can point to where → feedback to authors → improved models 8 P. S k a n d s
Example: α s Theory: default is factor 2 µ R variation → lots/less of FSR! Use this to define a theory uncertainty associated with α s (e.g., done in Perugia tunes) Data-driven (expect smaller?) : define variations by ~ 2- sigma consistent with 3-jet observables Use as cross check on theory uncertainty. How much variation does data actually allow (for the included observables)? Decide (if you dare) to reduce nominal factor 2, keeping in mind that a larger theory uncertainty is still needed to evaluate uncertainty on extrapolating to other observables/processes. Bonus! Can re-use the data-driven ones … Retune string parameters, using the data-driven large/small α s → hadronization variations for use with central α s → can add more systematic “mistunings” to explore uncertainty envelope better 9 P. S k a n d s
Global Tunes vs Model Tests Do independent tunes for several complementary “windows” on same physics Similar observables at different CM energies Schulz, Skands, arXiv:1103.3649 Similar observables, ee vs pp Same collider, different observable ranges E.g., for different pTjet, different Q 2 , different cuts, … Example: 3-parameter tuning at 630, 900, 1800, and 7000 GeV ⊥ with √ s Evolution of p 0 Evolution of PARP( 78 ) with √ s Evolution of PARP( 83 ) with √ s 1 3 . 5 3 PARP( 78 ) PARP( 83 ) ⊥ / GeV Global fit Global fit 3 Minuit result Minuit result 2 . 5 0 . 8 p 0 Combined uncertainty Combined uncertainty 2 . 5 2 0 . 6 2 1 . 5 1 . 5 0 . 4 1 Minuit result 1 Combined uncertainty √ s 0 . 2 1800 GeV ) 0 . 27 ± 0 . 02 (Global fit) p 0 ⊥ = 2.19 ± 0.06 · ( 0 . 5 √ s 0 . 5 1800 GeV ) 0 . 25 (Perugia 0 ) p 0 ⊥ = 1.99 · ( 0 0 0 10 3 10 3 10 3 √ s / GeV √ s / GeV √ s / GeV √ √ √ pT0 for MPI Impact-parameter profile CR Strength 10 P. S k a n d s
What is Tuning? FSR pQCD Parameters α s (m Z ) The value of the strong coupling at the Z pole Governs overall amount of radiation α s Running Renormalization Scheme and Scale for α s 1- vs 2-loop running, MSbar / CMW scheme, µ R ~ p T2 Matching S u b l e a d i n g L o g s 11 P. S k a n d s
What is Tuning? FSR pQCD Parameters α s (m Z ) The value of the strong coupling at the Z pole Governs overall amount of radiation α s Running Renormalization Scheme and Scale for α s 1- vs 2-loop running, MSbar / CMW scheme, µ R ~ p T2 Additional Matrix Elements included? Matching At tree level / one-loop level? Using what matching scheme? S u b l e a d i n g L o g s 11 P. S k a n d s
What is Tuning? FSR pQCD Parameters α s (m Z ) The value of the strong coupling at the Z pole Governs overall amount of radiation α s Running Renormalization Scheme and Scale for α s 1- vs 2-loop running, MSbar / CMW scheme, µ R ~ p T2 Additional Matrix Elements included? Matching At tree level / one-loop level? Using what matching scheme? Ordering variable, coherence treatment, effective 1 → 3 (or 2 → 4), recoil strategy, … S u b l e a d i n g L o g s Branching Kinematics (z definitions, local vs global momentum conservation), hard parton starting scales / phase-space cutoffs, masses, non-singular terms, … 11 P. S k a n d s
Value of Strong Coupling PYTHIA 8 (hadronization on) vs LEP: Thrust Major �� i | � p i · � n | � 1 − T → 1 T = max � i | � p i | 2 � n 1 − T → 0 Minor 1/N dN/d(Minor) 1/N dN/d(1-T) 1/N dN/d(Major) 1/N dN/d(O) 1-Thrust (udsc) Major Minor Oblateness Delphi Delphi Delphi L3 10 10 10 10 Pythia Pythia Pythia Pythia 1 1 1 1 -1 Oblateness -1 -1 -1 10 10 10 10 Minor 1-T Major = Major - Minor -2 -2 -2 -2 10 10 10 10 V I N C I A R O O T V I N C I A R O O T V I N C I A R O O T V I N C I A R O O T Data from CERN-PPE-96-120 Data from Phys.Rept. 399 (2004) 71 Data from CERN-PPE-96-120 Data from CERN-PPE-96-120 -3 Pythia 8.165 -3 Pythia 8.165 -3 Pythia 8.165 -3 Pythia 8.165 10 10 10 10 1.4 1.4 1.4 1.4 Theory/Data Theory/Data Theory/Data Theory/Data 1.2 1.2 1.2 1.2 1 1 1 1 0.8 0.8 0.8 0.8 0.6 0.6 0.6 0.6 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 0 0.2 0.4 0.6 0 0.2 0.4 0.6 1-T (udsc) Major Minor O Note: Value of Strong coupling is α s (M Z ) = 0.12 12 P. S k a n d s
Recommend
More recommend
