The problem The current state of the game Future challenges Beyond the Standard Model with global fits: then, now and tomorrow Pat Scott Imperial College London Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
76
Embed
Beyond the Standard Model with global fits: then, now and ... · Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow (Cline,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The problemThe current state of the game
Future challenges
Beyond the Standard Model with global fits:then, now and tomorrow
Pat Scott
Imperial College London
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Outline
1 The problemIntroductionGlobal fitsIncluding astroparticle observables
2 The current state of the gamePresent limitsCoverageScanning challenges
3 Future challengesRespectable LHC likelihoodsParameter space→ Theory space
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Many reasons to look for physics Beyond the Standard Model(BSM):
Higgs mass (hierarchy problem + vacuum stability)Dark matter existsBaryon asymmetryNeutrino masses and mixings
So what do we do about it?Make new particles at high-E collidersStudy rare processes at high-L collidersHunt for dark matter (direct + indirect detection)Look at cosmological observables (CMB, reionisation, etc)Look for impacts of unexpected or missing neutrinos
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Many reasons to look for physics Beyond the Standard Model(BSM):
Higgs mass (hierarchy problem + vacuum stability)Dark matter existsBaryon asymmetryNeutrino masses and mixings
So what do we do about it?Make new particles at high-E collidersStudy rare processes at high-L collidersHunt for dark matter (direct + indirect detection)Look at cosmological observables (CMB, reionisation, etc)Look for impacts of unexpected or missing neutrinos
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Many reasons to look for physics Beyond the Standard Model(BSM):
Higgs mass (hierarchy problem + vacuum stability)Dark matter existsBaryon asymmetryNeutrino masses and mixings
So what do we do about it?Make new particles at high-E collidersStudy rare processes at high-L collidersHunt for dark matter (direct + indirect detection)Look at cosmological observables (CMB, reionisation, etc)Look for impacts of unexpected or missing neutrinos
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
That’s all well and good if there are only 2 parameters and fewsearches. . .
QuestionWhat if there are many parameters?
AnswerNeed to
scan the parameter space (smart numerics)interpret the combined results (Bayesian / frequentist)project down to parameter planes of interest (marginalise /profile)
→ global fits
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
That’s all well and good if there are only 2 parameters and fewsearches. . .
QuestionWhat if there are many parameters?
AnswerNeed to
scan the parameter space (smart numerics)interpret the combined results (Bayesian / frequentist)project down to parameter planes of interest (marginalise /profile)
→ global fits
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
PS, Conrad et al JCAP, 0909.3300Ripken, Conrad & PS JCAP, 1012.3939
PS, Savage, Edsjö & The IceCubeCollab. JCAP, 1207.0810Silverwood, PS et al JCAP, 1210.0844
Strege et al JCAP, 1212.2636
Cline & PS JCAP, 1301.5908
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Outline
1 The problemIntroductionGlobal fitsIncluding astroparticle observables
2 The current state of the gamePresent limitsCoverageScanning challenges
3 Future challengesRespectable LHC likelihoodsParameter space→ Theory space
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Current constraints: CMSSM±ε
CMSSM, profile likelihoodsHiggsSignals + resimulation of LHC CMSSM limitsATLAS 0-lepton SUSY searches, 20.3 fb−1, 8 TeV
0h 0A 0H+H
1
0χ2
0χ3
0χ4
0χ1
+χ2
+χRl
~Ll
~1
τ∼
2τ∼
Rq~
Lq~
1b~
2b~
1t~
2t~ g~
Part
icle
Mass (
GeV
)
0
500
1000
1500
2000
2500
3000
3500 Environmentσ1
Environmentσ2
Best Fit Value
Fittino (PoS EPS-HEP 2013)
→ stau coannihilation + all elsedecoupled
MasterCode (EPJC 74:2922)
→ stau coannihilation+ Higgs funnel
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
What gives? Probably FeynHiggs v2.9 vs 2.10.Maybe also g − 2 calculation and DD likelihood.
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Current constraints: CMSSM±ε
CMSSM, profile likelihoodsHiggsSignals + resimulation of LHC CMSSM limitsATLAS 0-lepton SUSY searches, 20.3 fb−1, 8 TeV
0h 0A 0H+H
1
0χ2
0χ3
0χ4
0χ1
+χ2
+χRl
~Ll
~1
τ∼
2τ∼
Rq~
Lq~
1b~
2b~
1t~
2t~ g~
Part
icle
Mass (
GeV
)
0
500
1000
1500
2000
2500
3000
3500 Environmentσ1
Environmentσ2
Best Fit Value
Fittino (PoS EPS-HEP 2013)
→ stau coannihilation + all elsedecoupled
MasterCode (EPJC 74:2922)
→ stau coannihilation+ Higgs funnel
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
What gives? Probably FeynHiggs v2.9 vs 2.10.Maybe also g − 2 calculation and DD likelihood.
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Current constraints: low-scale MSSM
SuperBayeS (1405.0622)
15-parameter weak-scale MSSM
profile likelihood
latest B/D and DM constraints
‘tall poppy’ analysis:post-processed tiny subset ofbest points with collider limits
ATLAS 0 and 3-lepton SUSYsearches, 4.7 fb−1, 7 TeV
Strege et al. (2014)
mχ1
0 (GeV)
log(
σ pSI (
pb))
All data
XENON100
LUX
10 100 1000 10000
−25
−20
−15
−10
−5
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Current issues: Coverage
Test statistic: a measure on data used to construct statistical tests (e.g. χ2, lnL, etc.)Coverage: the percentage of the time that a supposed ‘x%’ confidence regionactually contains the true value
Distribution of the test statistic and design of the test it’s used in determinecoverage.
p-value calculation requires the test statistic distribution to be well known.
We don’t *really* usually know the distribution of our teststatistic in BSM global fits, as it is too expensive to Monte Carlo
coverage is rarely spot-on unless mapping from parameters todata-space is linear(Akrami, Savage, PS et al JCAP, 1011.4297, Bridges et al JHEP, 1011.4306, Strege et al PRD, 1201.3631)
p-value assessments of goodness of fit should be viewed with seriousscepticism (→MasterCode)
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Current issues: Coverage
Test statistic: a measure on data used to construct statistical tests (e.g. χ2, lnL, etc.)Coverage: the percentage of the time that a supposed ‘x%’ confidence regionactually contains the true value
Distribution of the test statistic and design of the test it’s used in determinecoverage.
p-value calculation requires the test statistic distribution to be well known.
We don’t *really* usually know the distribution of our teststatistic in BSM global fits, as it is too expensive to Monte Carlo
coverage is rarely spot-on unless mapping from parameters todata-space is linear(Akrami, Savage, PS et al JCAP, 1011.4297, Bridges et al JHEP, 1011.4306, Strege et al PRD, 1201.3631)
p-value assessments of goodness of fit should be viewed with seriousscepticism (→MasterCode)
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Fittino, arXiv:1410.6035
2χ
0 5 10 15 20 25 30 35 40 45 50
Fra
ctio
ns
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
CMSSMToy Fits
(NDF = 22)2χBest Fit point30.42
0.5) %±P = ( 2.5
The problemThe current state of the game
Future challenges
Present limitsCoverageScanning challenges
Current issues: Scanning algorithms
Convergence remains an issue, especially for profile likelihoodMessy likelihood =⇒ best-fit point can be (and often is) easilymissed (Akrami, PS et al JHEP, 0910.3950, Feroz et al JHEP, 1101.3296)
frequentist CLs are off, as isolikelihood levels are chosen incorrectlycan impact coverage (overcoverage, or masking of undercoverage dueto non-χ2 TS distribution)need to use multiple priors and scanning algorithms (one optimised forprofile likelihoods?)
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Outline
1 The problemIntroductionGlobal fitsIncluding astroparticle observables
2 The current state of the gamePresent limitsCoverageScanning challenges
3 Future challengesRespectable LHC likelihoodsParameter space→ Theory space
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
The LHC likelihood monster
Time per point:
O(minute) in best cases
Time per point for global fits to converge:
O(seconds) in worst cases
Challenge:
About 2 orders of magnitude too slow to actually include LHCdata in global fits properly
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
The LHC likelihood monster
Time per point:
O(minute) in best cases
Time per point for global fits to converge:
O(seconds) in worst cases
Challenge:
About 2 orders of magnitude too slow to actually include LHCdata in global fits properly
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
The LHC likelihood monster
Time per point:
O(minute) in best cases
Time per point for global fits to converge:
O(seconds) in worst cases
Challenge:
About 2 orders of magnitude too slow to actually include LHCdata in global fits properly
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Taming the LHC monster
Zeroth Order Response:“Just use the published limits and ignore the dependence onother parameters”
Obviously naughty – plotted limits assume CMSSM, and fix twoof the parameters
Don’t really know dependence on other parametersDon’t have a likelihood function, just a lineCan’t use this at all for non-CMSSM global fits – e.g.MSSM-25
(Early) SuperBayeS
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Taming the LHC monster
Zeroth Order Response:“Just use the published limits and ignore the dependence onother parameters”
Obviously naughty – plotted limits assume CMSSM, and fix twoof the parameters
Don’t really know dependence on other parametersDon’t have a likelihood function, just a lineCan’t use this at all for non-CMSSM global fits – e.g.MSSM-25
(Early) SuperBayeS
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Taming the LHC monster
First Order Response:“Test if things depend on the other parameters (hope not),re-simulate published exclusion curve”
Not that great, but OK in some casesAt least have some sort of likelihood this timeStill a bit screwed if things do depend a lot on otherparameters, butallows (potentially shaky) extrapolation, also tonon-CMSSM models
Fittino, Mastercode
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Taming the LHC monster
First Order Response:“Test if things depend on the other parameters (hope not),re-simulate published exclusion curve”
Not that great, but OK in some casesAt least have some sort of likelihood this timeStill a bit screwed if things do depend a lot on otherparameters, butallows (potentially shaky) extrapolation, also tonon-CMSSM models
Fittino, Mastercode
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Taming the LHC monster
Second Order Response:“That’s ridiculous. I’ve never met a calculation I can’t speed up.There must be some way to have my cake and eat it too”
Maybe – this is the challenge.Interpolated likelihoods (how to choose nodes?)Neural network functional approximation (how to trainaccurately?)Some sort of smart reduction based on event topology?Something else?
Balázs, Buckley, Farmer, White et al (1106.4613,1205.1568); GAMBIT
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Taming the LHC monster
Second Order Response:“That’s ridiculous. I’ve never met a calculation I can’t speed up.There must be some way to have my cake and eat it too”
Maybe – this is the challenge.Interpolated likelihoods (how to choose nodes?)Neural network functional approximation (how to trainaccurately?)Some sort of smart reduction based on event topology?Something else?
Balázs, Buckley, Farmer, White et al (1106.4613,1205.1568); GAMBIT
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
CMSSM, SMS 6= BSM
(SMS = Simplified Model Spectrum)
Want to do model comparison to actually work out which theoryis right. . .
Challenge:
How do I easily adapt a global fit to different BSM theories?
Somehow, we must recast things quickly to a new theorydatalikelihood functionsscanning code ‘housekeeping’even predictions
=⇒ a new, very abstract global fitting framework
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
CMSSM, SMS 6= BSM
(SMS = Simplified Model Spectrum)
Want to do model comparison to actually work out which theoryis right. . .
Challenge:
How do I easily adapt a global fit to different BSM theories?
Somehow, we must recast things quickly to a new theorydatalikelihood functionsscanning code ‘housekeeping’even predictions
=⇒ a new, very abstract global fitting frameworkPat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Hitting the wall
Issues with current global fit codes:Strongly wedded to a few theories (e.g. constrained MSSM/ mSUGRA)Strongly wedded to a few theory calculatorsAll datasets and observables basically hardcodedRough or non-existent treatment of most experiments(astroparticle + collider especially)Sub-optimal statistical methods / search algorithms=⇒ already hitting the wall on theories, data &computational methods
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
GAMBIT: a second-generation global fit code
GAMBIT: Global And Modular BSM Inference Tool
Overriding principles of GAMBIT: flexibility and modularityGeneral enough to allow fast definition of new datasets andtheoretical modelsPlug and play scanning, physics and likelihood packagesExtensive model database – not just small modifications toconstrained MSSM (NUHM, etc), and not just SUSY!Extensive observable/data libraries (likelihood modules)Many statistical options – Bayesian/frequentist, likelihooddefinitions, scanning algorithmsA smart and fast LHC likelihood calculatorMassively parallelFull open-source code release
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
The GAMBIT Collaboration
26 Members, 15 institutions, 9 countries8 Experiments, 4 major theory codes
Fermi-LAT J. Conrad, J. Edsjö, G. Martinez, P. Scott (leader)CTA C. Balázs, T. Bringmann, J. Conrad, M. White (dep. leader)HESS J. ConradATLAS A. Buckley, P. Jackson, C. Rogan, A. Saavedra, M. WhiteLHCb M. Chrzaszcz, N. SerraIceCube J. Edsjö, C. Savage, P. ScottAMS-02 A. PutzeCDMS, DM-ICE L. HsuDARWIN, XENON J. ConradTheory P. Athron, C. Balázs, T. Bringmann, J. Cornell, L.-A. Dal, J. Edsjö,
B. Farmer, A. Krislock, A. Kvellestad, N. Mahmoudi, M. Pato,A. Raklev, C. Savage, P. Scott, C. Weniger, M. White
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
So what’s so much better about GAMBIT?Aspect GAMBIT MasterCode SuperBayeS Fittino Rizzo et al.Design Modular, Adaptive Monolithic Monolithic (∼)Monolithic MonolithicStatistics Frequentist, Bayesian Frequentist Freq./Bayes. Frequentist NoneScanners Differential evolution, genetic algo-
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
LHC ATLAS+CMS multi-analysis with neu-ral net and fast detector simulation.Higgs multi-channel with correlationsand no SM assumptions. Full flavourinc. complete B → Xs ll and B →K∗ ll angular set.
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Closing remarks
Robust analysis of dark matter and BSM physics requiresmulti-messenger global fitsGAMBIT is coming:→ Lots of interesting particle, astronomical, cosmological and
astroparticle observables to include in global fits→ Serious theoretical, experimental, statistical and
computational detail to work though→ Oslo is already in the thick of it
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
The problemThe current state of the game
Future challenges
Respectable LHC likelihoodsParameter space→ Theory space
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Backup Slides
Outline
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow
Backup Slides
GAMBIT: sneak peek
Pat Scott – Oct 29 – Oslo Theory Seminar Beyond the Standard Model global fits: then, now and tomorrow