Data-based background predictions for new particle searches at the LHC

61
a-based background predictio for new particle searches at the LHC David Stuart Univ. of California, Santa Barbara Texas A&M Seminar March 24, 2010

description

Data-based background predictions for new particle searches at the LHC. David Stuart Univ. of California, Santa Barbara Texas A&M Seminar March 24, 2010. Motivation. Searching for new physics at the LHC. Potentially fast. With a large step in energy, the LHC could start up with a bang. - PowerPoint PPT Presentation

Transcript of Data-based background predictions for new particle searches at the LHC

Page 1: Data-based background predictions  for new particle searches at the LHC

Data-based background predictions for new particle searches

at the LHC

David Stuart

Univ. of California, Santa Barbara

Texas A&M SeminarMarch 24, 2010

Page 2: Data-based background predictions  for new particle searches at the LHC

2

Motivation

Searching for new physics at the LHC.

Potentially fast.With a large step in energy, the LHC could start up with a bang.

Page 3: Data-based background predictions  for new particle searches at the LHC

3

Motivation

Searching for new physics at the LHC.

Potentially fast.

But many models; on which to bet?

Do they have something in common?

Page 4: Data-based background predictions  for new particle searches at the LHC

4

Motivation

Searching for new physics at the LHC.

Potentially fast.

But many models; on which to bet?

Do they have something in common?(other than being wrong)

Page 5: Data-based background predictions  for new particle searches at the LHC

5

Motivation

Searching for new physics at the LHC.

Even within 1 model, many parameters…

Signature driven searches are more general.

But, which signature is best?

Page 6: Data-based background predictions  for new particle searches at the LHC

6

Motivation

Searching for new physics at the LHC.

Search broadly for any non-SM in all signatures?

Page 7: Data-based background predictions  for new particle searches at the LHC

7

Motivation

Searching for new physics at the LHC.

Search broadly for any non-SM in all signatures?

But signatures are not precisely predicted.pdfs, higher orders, detector effects…

e.g., Z+jetsq

Z+

-

Page 8: Data-based background predictions  for new particle searches at the LHC

8

Motivation

Monte Carlo predictions?

Sophisticated, higher order modeling,

e.g., ALPGEN.

Elaborate simulation of detector response.

Page 9: Data-based background predictions  for new particle searches at the LHC

9

Motivation

Monte Carlo predictions?

Sophisticated, higher order modeling,

e.g., ALPGEN.

Elaborate simulation of detector response.

Both are software…Only trust in so far as validated with data.

Page 10: Data-based background predictions  for new particle searches at the LHC

10

Motivation

Data validation challenges:

Slow.

Fit away signal?

Page 11: Data-based background predictions  for new particle searches at the LHC

11

Motivation

Data validation challenges:

Slow.

Fit away signal?

Would be nice to turn off new physics temporarily.

Page 12: Data-based background predictions  for new particle searches at the LHC

12

A simple discriminator

Most new physics is high mass

Most SM physics is low mass

Page 13: Data-based background predictions  for new particle searches at the LHC

13

A simple discriminator

Most new physics is high mass Produced at threshold, i.e. at rest. Decay products ≈ isotropic Decay products peaked at zero rapidity

Most SM physics is low mass

Produced ≈ uniform in rapidity

Page 14: Data-based background predictions  for new particle searches at the LHC

14

A simple discriminator

Validate SM in forward events

and

Search for new physics in central events

Page 15: Data-based background predictions  for new particle searches at the LHC

15

Start with the Z+jets signature

• Insert favorite model motivation here.

• Clean dilepton signature

• Easy to trigger and reconstruct

• Very little background

A simple signature

Page 16: Data-based background predictions  for new particle searches at the LHC

16

Start with the Z+jets signature

• Insert favorite model motivation here.

• Clean dilepton signature

• Easy to trigger and reconstruct

• Very little background

…except Z+jets.

A simple signature

Page 17: Data-based background predictions  for new particle searches at the LHC

17

Z+jets

SM falls ≈ exponentially with NJ.

Signal would appear at large NJ.

Page 18: Data-based background predictions  for new particle searches at the LHC

18

Forward control sample

SM Z rapidity is ≈ flat since the Z is light.

Forward events are a control sample for ≈ all NJ.

Signal is central.

ALPGEN+Pythia+PYCELL

Page 19: Data-based background predictions  for new particle searches at the LHC

19

Forward control sample

SM Z rapidity is ≈ flat since the Z is light.

Forward events are a control sample for ≈ all NJ.

Signal is central.After acceptance cuts the conclusion is the same.

Page 20: Data-based background predictions  for new particle searches at the LHC

20

MethodDefine the fraction of central events with:

R(NJ) = ncentral(NJ) / (ncentral(NJ) + nforwardNJ))

where we define central as |<1 and forward as |>1.3

Measure R(NJ) at low NJ. Extrapolate linear fit to high NJ.

Page 21: Data-based background predictions  for new particle searches at the LHC

21

MethodPredict number of central events with high NJ as:

ncentral(NJ) = nforward(NJ) * R(NJ) / (1-R(NJ))

From low NJ fit.

{{Measured

Dominant uncertainty is from fluctuations in nforward(NJ).

Page 22: Data-based background predictions  for new particle searches at the LHC

22

Does it work?Check self consistency in Monte Carlo…

L = 1 fb-1

Predicted

Actual

Page 23: Data-based background predictions  for new particle searches at the LHC

23

Does it work with signal?Not focused on sensitivity to any specific model,

but using LM4 as a benchmark:

L = 1 fb-1

Predicted w/o signal

Predicted w/ signal

Actual w/ signal

Page 24: Data-based background predictions  for new particle searches at the LHC

24GeneralizingThe basic premise (low-mass broad rapidity range) generalizes beyond Z’s.

Page 25: Data-based background predictions  for new particle searches at the LHC

25

Does it work, generally?Check self consistency in each mode…

Predicted

Actual

Z W

multijets

Page 26: Data-based background predictions  for new particle searches at the LHC

26

Does it work robustly?Check for robustness against mis-modeling. E.g.,

• Eta dependence of lepton efficiencies.• Eta dependence of jet efficiencies.• Changes in higher order Monte Carlo effects.

Expect robustness since data-based prediction:

• Measures lepton efficiencies in the low NJ bins

• Measures jet effects in events with forward Z’s.

• Measures NJ dependence in the fit.

As long as correlations between lepton and jet effects are a slowly varying function of NJ, the R(NJ) fit will account for it.

Page 27: Data-based background predictions  for new particle searches at the LHC

27

Does it work robustly?Tests with artificially introduced mis-modeling.

Z W j

Alpgen #partons Lepton inefficiencies Jet inefficiencies

Pulls are shown for two highest ET jet bins for each test. Alpgen test = even #partons only and odd #partons only. Lepton test = 30% efficiency changes globally and forward only. Jet test = 30% efficiency changes globally and forward only.

Page 28: Data-based background predictions  for new particle searches at the LHC

28

R(NJ)

Beyond using R(NJ) to predict the central yield and count events there,

R(NJ) is potentially of general interest as a search variable.

Page 29: Data-based background predictions  for new particle searches at the LHC

29

R(NJ)The central fraction, R(NJ), is potentially of general interest.

“Minbias” example:

Here, “NJ” uses tracksabove 3 GeV as jet proxies.

The highest pT track is therapidity tag.

R(NJ) ≈ 1/2 because tracks flat in and central ≈ forward

for tracking coverage.

Changing bounds wouldmove R(NJ) but notchange its shape.

R(N

J)

Page 30: Data-based background predictions  for new particle searches at the LHC

30

R(NJ)The central fraction, R(NJ), is potentially of general interest.

W and Z are light and so similar to Minbias.

Acceptance difference apparent.

Page 31: Data-based background predictions  for new particle searches at the LHC

31

R(NJ)The central fraction, R(NJ), is potentially of general interest.

W and Z are light and so similar to Minbias.

Acceptance difference apparent.

+jets and jet+jetsare non-flat but still linear.

Page 32: Data-based background predictions  for new particle searches at the LHC

32

R(NJ)The central fraction, R(NJ), is potentially of general interest.

W and Z are light and so similar to Minbias.

Acceptance difference apparent.

+jets and jet+jetsare non-flat but still linear.

SUSY model points are dominantly central.

Page 33: Data-based background predictions  for new particle searches at the LHC

33

R(NJ)(-1)

We have also explored another variable that tries to take advantage of the general expectation that the NJ spectrum should be falling.

L = 1 fb-1

Predicted w/o signal

Predicted w/ signal

Actual w/ signal

Without MET cut.

Clear signal when there is an increase with NJ, or even a decrease in the slope.

R(NJ)(-1) = ncentral(NJ) / (ncentral(NJ) + nforward(NJ-1))

Page 34: Data-based background predictions  for new particle searches at the LHC

34

R(NJ)(-1)

We have also explored another variable that tries to take advantage of the general expectation that the NJ spectrum should be falling.

Z+jetsZ+jets plus LM4

≈ S

Page 35: Data-based background predictions  for new particle searches at the LHC

35

R(NJ)(-2)

Can “leverage” that to use the forward events from two jet bins previous.

Z+jetsZ+jets plus LM4

≈ S2

This really just represents our generic expectation that for the SM, NJ should ≈ fall exponentially and be uniform in rapidity, while for a heavy particle production is central and increases with NJ. Similar plots can be made for , jet, W.

Page 36: Data-based background predictions  for new particle searches at the LHC

36

What about Missing ET?

Would like to predict V+jets+MET for a Supersymmetry search.

Is there a SUSY-less sample from which to measure MET?

Page 37: Data-based background predictions  for new particle searches at the LHC

37

Missing ET in Z+jets

The Z is well measured. The MET comes from the detector’s response to the jet system.

Page 38: Data-based background predictions  for new particle searches at the LHC

38

Missing ET in Z+jets

For each Z+jet event, find an event w/ a comparable jet system and use its MET as a prediction.

Huge QCD x-section makes such events SUSY free.

Page 39: Data-based background predictions  for new particle searches at the LHC

39

Missing ET in Z+jets

For each Z+jet event, use a MET template measured from events with a comparable jet system in O(1) pb-1.

Templates measured in bins of NJ and JT = j ET.

Page 40: Data-based background predictions  for new particle searches at the LHC

40

Missing ET in Z+jetsExample of template parameterization

Background predictionData distribution

For each data event...

Page 41: Data-based background predictions  for new particle searches at the LHC

41

Missing ET in Z+jetsExample of template parameterization

Background predictionData distribution

For each data event,look up the appropriate template.Sum these, each withunit normalization, to get the fullbackground prediction

N JETSpT>50 GeV

sumETBin 1

sumET Bin 2

sumETBin 3

sumET Bin 4…

2

3

4…

Page 42: Data-based background predictions  for new particle searches at the LHC

42

Missing ET in Z+jets, MC closure test

Page 43: Data-based background predictions  for new particle searches at the LHC

43

Missing ET in Z+jets, MC closure test

Page 44: Data-based background predictions  for new particle searches at the LHC

44

Missing ET in Z+jets, MC closure test

Page 45: Data-based background predictions  for new particle searches at the LHC

45

Missing ET in Z+jets, MC closure tests

“Scaled” includes a low MET normalization, which is important for low NJ.

Page 46: Data-based background predictions  for new particle searches at the LHC

46

Missing ET in +jets, MC closure test

Page 47: Data-based background predictions  for new particle searches at the LHC

47

Missing ET in +jets, MC closure test

Page 48: Data-based background predictions  for new particle searches at the LHC

48

Missing ET in +jets, MC closure test

Page 49: Data-based background predictions  for new particle searches at the LHC

49

Missing ET in +jets, MC closure tests

“Scaled” includes a low MET normalization, which is important for low NJ.

Page 50: Data-based background predictions  for new particle searches at the LHC

50

Missing ET in Z/+jets, robustness tests

Various detectoreffects could addMET tails.

Check robustnesswith MC tests,applied equally toall samples.

Page 51: Data-based background predictions  for new particle searches at the LHC

51

Missing ET in Z/+jets, robustness tests

• R=0.8 hole at (h,f)=(0,0)• Double gaussian smearing• Randomly add 50-100 GeV “noise jets”• Vary nJ slope by ±50%.• Jet energy scale sensitivity.

Page 52: Data-based background predictions  for new particle searches at the LHC

52

Missing ET in W()+jetsW

Can predict W+jets, with forward/central,

but not ttW+jets

because top is heavy.

Page 53: Data-based background predictions  for new particle searches at the LHC

53

Missing ET in W()+jetsW

Templates can predict the fake MET in W+jet events, but we also need to predict the real MET, i.e., the pT.

Can predict W+jets, with forward/central,

but not ttW+jets

because top is heavy.

Page 54: Data-based background predictions  for new particle searches at the LHC

54

Missing ET in W(+jetsW

Templates can predict the fake MET in W+jet events, but we also need to predict the real MET, i.e., the pT.

But, pT spectrum is ≈ same as pT spectrum,

if we ignore V-A or randomize W polarization.

Can predict W+jets, with forward/central,

but not ttW+jets

because top is heavy.

Page 55: Data-based background predictions  for new particle searches at the LHC

55

Missing ET in +jets

Pretend we could detect and apply templates.

Mismatch due to b-jet dominance.

But, neutrino pT dominates MET.

Page 56: Data-based background predictions  for new particle searches at the LHC

56

Missing ET in +jets

Combining template prediction with pT spectrum

gives a prediction for the full MET distribution.

Page 57: Data-based background predictions  for new particle searches at the LHC

57

Missing ET in +jets

The same approach predicts W shape,

if polarization is random.

Page 58: Data-based background predictions  for new particle searches at the LHC

58

Comparison with signal

Benchmark points (LM4 and LM1) stand out with 200/pb at 14 TeV.LM4=(m0=210,m1/2=285); LM1=(60,250). tan()=10.

Page 59: Data-based background predictions  for new particle searches at the LHC

59

SummaryExplored data-based background predictions that avoid reliance on MC.

Rapidity is a simple discriminator that relies only on kinematics.

It provides a data-based background prediction that:

• Avoids generator and detector modeling uncertainties by measuring a ratio.

• Fails to discover anything that it shouldn’t, even when reality bites.

QCD based templating gives an in situ prediction of MET distribution.

Charged lepton pT predicts neutrino pT.

We will validate these methods with low NJ data soon.

Work done by Victor Pavlunin. More details are available in:PRD78:035012 arXiv:0806.2338 & PRD81:035005 arXiv:0906.5016

Page 60: Data-based background predictions  for new particle searches at the LHC
Page 61: Data-based background predictions  for new particle searches at the LHC

Standard Model backgrounds to Z+jets

+ +

+q Z

e+

e-