St. Andrews, Scotland J. Huston QCD Backgrounds to New Physics I J. Huston …thanks to Weiming Yao,...

St. Andrews, ScotlandJ. Huston

QCD Backgrounds to New Physics IJ. Huston

…thanks to Weiming Yao, John Campbell, Bruce KnutesonNigel Glover for transparencies


Home again

Houston Renfrewshire

A large village located 6 miles (9.6 km) north-west of Paisley, the name derives from 'Hew's town', a 13th century village around the castle of Hugo de Kilpeter. In 1781 the castle was partly demolished and a New Town of 35 houses was erected. The Crosslee Mill (1793, demolished 1986) was used for cotton spinning but in 1916-17 a new factory was constructed for 'spinning cordite fuses'. The Mercat Cross has a sundial from the early 18th century and a shaft which may be 14th century; Houston House (17th century) and Barochan House (17th century and

1896) are other notable buildings.

don’t worry about the extra o


Run 2 Monte Carlo Workshop

Transparencies, video links to individual talks and links to programs can all be found at http://www-theory.fnal.gov/runiimc/

I’ll be referring to some of these programs in the course of my talk


Les Houches Two workshops on “Physics at

TeV Colliders” have been held so far, in 1999 and 2001 (May 21-June 1)

Working groups on QCD/SM, Higgs, Beyond Standard Model

See web page:

http://wwwlapp.in2p3.fr/conferences/LesHouches/Houches2001/

especially for links to writeups from 1999

QCD writeup (hep-ph/0005114) is an excellent pedagogical review for many of the issues discussed here and at this SS


Other useful references (for pdf uncertainties)

LHC Guide to Parton Distributions and Cross Sections, J. Huston; http://www.pa.msu.edu/~huston/lhc/lhc_pdfnote.ps

The QCD and Standard Model Working Group: Summary Report from Les Houches; hep-ph/0005114

Parton Distributions Working Group, Tevatron Run 2 Workshop; hep-ph/0006300

A QCD Analysis of HERA and fixed target structure function data, M. Botje; hep-ph/9912439

Global fit to the charged leptons DIS data…, S. Alekhin; hep-ph/0011002 Walter Giele’s presentation to the QCD group on Jan. 12; http://www-

cdf.fnal.gov/internal/physics/qcd/qcd99_internal_meetings.html Uncertainties of Predictions from Parton Distributions I: the Lagrange

Multiplier Method, D. Stump, J. Huston et al.;hep-ph/0101051 Uncertainties of Predictions from Parton Distributions II: the Hessian

Method, J. Pumplin, J. Huston et al.; hep-ph/0101032


Other references…

…to the ability to calculate QCD backgrounds …to the need to do so


One person was more offended by the book than Carlo Rubbia

“Stirling was a dapper Englishman, in his mid-thirties…”

p. 228


Monojets in UA1 UA1 monojets (1983-1984)

Possible signature of new physics (SUSY, etc)

A number of backgrounds were identified, but each was noted as being too small to account for the observed signal

pp->Z + jets

|_ pp->W + jets

|_ |_ hadrons +

pp->W + jets

|_ l + pp->W + jets

|_ |_ l +

jet

…but the sum was not “The sum of many small things is a

big thing.” G. Altarelli

Can calculate from first principles or calibrate to observed cross sections for Z->e+e- and W->e

Ellis, Kleiss, Stirling PL 167B, 1986.


Signatures of New Physics W’s, jets, ’s, b quarks, ET

…pretty much the same as signatures for SM physics How do we find new physics? By showing that it’s not old physics.

can be modifications to the rate of production …or modification to the kinematics, e.g.angular distributions

Crucial to understand the QCD dynamics of both backgrounds to any new physics and to the new physics itself

Some backgrounds can be measured in situ …but may still want to predict in advance, e.g. QCD backgrounds to

H-> For some backgrounds, need to rely on theoretical calculations, e.g. ttbb

backgrounds to ttH


Theoretical Predictions for New (Old) Physics

There are a variety of programs available for comparison of data to theory and/or predictions.

Tree level Leading log Monte Carlo

NnLO

Resummed

Important to know strengths/weaknesses of each.

In general, agree quite well…but before you appeal to new physics, check theME. (for example using Comphep)Can have ME corrections to MC or MC corrections to ME. (in CDF->HERPRT)

Perhaps biggest effort…include NLO MEcorrections in Monte Carlo programs…correct normalizations. Correct shapes. NnLO needed for precision physics.

Resummed description describes soft gluoneffects (better than MC’s)…has correct normalization (but need HO to get it); resummed predictions include non-perturbative effects correctly…may have to be put in by hand in MC’sthreshold kT

W,Z, Higgsdijet, direct

b space(ResBos)

qt space

Where possible, normalize to existing data.…in addition, worry about pdf, fragmentation uncertainties


W + Jet(s) at the Tevatron

Good testing ground for parton showers, matrix elements, NLO

Background for new physics or old physics (top production)

Reasonable agreement for the leading order comparisons using VECBOS (but large scale dependence)

•Good agreement with NLO (and smaller scale dependence) for W + >= 1 jet


W + jets For W + >=n jet production, typically

use Herwig (Herprt) for additional gluon radiation and for hadronization

Can also start off with n-1 jets and generate additional jets using Herwig


More Comparisons (VECBOS and HERWIG)

Start with W + n jets from VECBOS •Start with W + (n-1) jets from VECBOS


More Comparisons Start with W + n jets from VECBOS Start with W + (n-1) jets from

VECBOS


When good Monte Carlos go bad

Consider W + jet(s) sample Compare data (Run 0 CDF) to

VECBOS+HERPRT (Herwig radiation+hadronization interface to VECBOS) normalized to W+X jets

Starting with W + 1 jet rate in data, Herwig predicts 1 W+ >4 jet events; in data observe 10

…factor of ~2 every jet very dependent on kinematic

situation, though jet ET cuts center-of-mass energy etc

#events >1 jet >2 jets >3 jets>4 jet

pT>10 GeV/c

Data 920 213 4210

VECBOS + HERPRT (Q=<pT>)

W + 1jet 920 178 21 1

W + 2jet ----- 213 43 6

W + 3jet ----- ----- 42 10

VECBOS+HERPRT(Q=mW)

W + 1jet 920 176 24 2

W + 2jet ---- 213 46 6

W + 3 jet ---- ----- 42 7


Factors get worse at the LHC


Why? Some reasons given by the experts (Mangano, Yuan,Ilyin…)

Herwig (any Monte Carlo) only has collinear part of matrix element for gluon emission; underestimate for the wide angle emission that leads to widely separated jets

phase space: Herwig has ordering in virtualities for gluon emission while this is not present in exact matrix element calculations; more phase space for gluon emission in exact matrix element calculations

in case of exact matrix element, there are interferences among all of the different diagrams; these interferences become large when emissions take place at large angles (don’t know a priori whether interference is positive or negative)

unitarity of Herwig evolution: multijet events in Herwig will always be a fraction of the 2 jet rate, since multijet events all start from the 2-jet hard process

all “K-factors” from higher order are missed.


Tree Level Calculations

Leading order matrix element calculations describe multi-body configurations better than parton showers

Many programs exist for calculation of multi-body final states at tree-level

references at beginning of talk; see Run 2 MC workshop

CompHep includes SM Lagrangian and several other

models, including MSSM deals with matrix elements squared calculates leading order 2->4-6 in the final

state taking into account all of QCD and EW diagrams

color flow information; interface exits to Pythia

great user interface Grace

similar to CompHep Madgraph

SM + MSSM deals with helicity amplitudes “unlimited” external particles (12?) color flow information not much user interfacing yet

Alpha + O’Mega does not use Feynman diagrams gg->10 g (5,348,843,500 diagrams)


An example using CompHep


Monte Carlo Interfaces To obtain full predictability for a

theoretical calculation, would like to interface to a Monte Carlo program (Herwig, Pythia, Isajet)

parton showering (additional jets) hadronization detector simulation

Some interfaces already exist VECBOS->Herwig (HERPRT) CompHep->Pythia

A general interface accord was reached at the 2001 Les Houches workshop

All of the matrix element programs mentioned will output 4-vector and color flow information in such a way as to be universally readable by all Monte Carlo programs

CompHep, Grace, Madgraph, Alpha, etc, etc

->Herwig, Pythia, Isajet


Les Houches accord (draft)


Parton Showering Determination of the Higgs signal

requires an understanding of the Higgs pT distribution at both LHC and Tevatron

for example, for gg->HX->X, the shape of the signal pT distribution is harder than that of the background; this can be used to advantage

To reliably predict the Higgs pT distribution, especially for low to medium pT region, have to include effects of soft gluon radiation

can either use parton showering a la Herwig, Pythia, ISAJET or kT resummation a la ResBos

parton showering resums primarily the (universal) leading logs while an analytic kT resummation can resum all logs with Q2/pT

2 in their arguments; but expect predictions to be similar and Monte Carlos offer a more useful format

Where possible it’s best to compare pT predictions to a similar data set to insure correctness of formalism; if data is not available, compare MC’s to a resummed calculation or at least to another Monte Carlo

all parton showers are not equal

Note the large difference between PYTHIA versions5.7 and 6.1. Which one is correct?


Change in PYTHIA Older version of PYTHIA has more events at

moderate pT

Two changes from 5.7 to 6.1 A cut has been placed on the combination of z

and Q2 values in a branching: u=Q2-s(1-z)<0 where s refers to the subsystem of hard scattering plus shower partons

corner of emissions that do not respect this requirement occurs when Q2 value of space-like emitting parton is little changed and z value of branching is close to unity

necessary if matrix element corrections are to be made to process

net result is substantial reduction in amount of gluon radiation

In principle affects all processes; in practice only gg initial states

Parameter for minimum gluon energy emitted in space-like showers is modified by extra factor corresponding to 1/ factor for boost to hard subprocess frame

result is increase in gluon radiation

The above are choices, not bugs; which version is more correct?

->Compare to ResBos

S. Mrenna80 GeV Higgs generated at the Tevatron with Pythia


Comparison of PYTHIA and ResBos for Higgs Production at LHC

ResBos agrees much better with the more recent version of PYTHIA

Suppression of gluon radiation leading to a decrease in the average pT of the produced Higgs

Affects the ability of CMS to choose to the correct vertex to associate with the diphoton pair

Note that PYTHIA does not describe the high pT end well unless Qmax

2 is set to s (14 TeV)

Again, ResBos has the correct matrix element matching at high pT; setting Qmax

2=s allows enough additional gluon radiation to mimic the matrix element


Comparisons with Herwig at the LHC

HERWIG (v5.6) similar in shape in PYTHIA 6.1 (and perhaps even more similar in shape to ResBos)

Is there something similar to the u-hat cut that regulates the HERWIG behavior?

Herwig treatment of color coherence? (study in progress)


The need for higher order…


What would we like?

Bruce Knuteson’s wishlist from the Run 2 Monte Carlo workshop

…all at NLO


What are we likely to get?


MCFM (Monte Carlo for Femtobarn Processes) J. Campbell and K.Ellis

Goal is to provide a unified description of processes involving heavy quarks, leptons and missing energy at NLO accuracy

There have so far been three main applications of this Monte Carlo, each associated with a different paper.

Calculation of the Wbb background to a WH signal at the Tevatron. R.K.Ellis, Sinisa Veseli, Phys. Rev. D60:011501 (1999), hep-ph/9810489.

Vector boson pair production at the Tevatron, including all spin correlations of the boson decay products.

J.M.Campbell, R.K.Ellis, Phys. Rev. D60:113006 (1999), hep-ph/9905386.

Calculation of the Zbb and other backgrounds to a ZH signal at the Tevatron.

J.M.Campbell, R.K.Ellis, FERMILAB-PUB-00-145-T, June 2000, hep-ph/0006304.

The last of these references contains the most

details of our method.


Higgs backgrounds using MCFM


Wbbar and Zbbar


Recent example of data vs Monte Carlo

There is a discovery potential at the Tevatron during Run 2 for a relatively light Higgs (especially if Higgs mass is 115 GeV)

…but small signal to background ratio makes understanding of backgrounds very important

CDF and ATLAS recently went through similar exercises regarding this background

CDF using Run 1 data ATLAS using Monte Carlo

predictions


Data vs Monte Carlo


Sleuth strategy Consider recent major discoveries in

hep W,Z bosons CERN 1983 top quark Fermilab 1995 tau neutrino Fermilab 2000 Higgs Boson? CERN 2000

In all cases, predictions were definite, aside from mass

Plethora of models that appear daily on hep-ph

Is it possible to perform a “generic” search?

Transparencies from Bruce Knuteson talk at Moriond 2001


W2j

We consider exclusive final statesWe consider exclusive final states

We assume the existence of standard object definitions

These define e, μ, , , j, b, ET, W, and Z fi

All events that contain the same numbers of each of these objects belong to the same final state

Step 1: Exclusive final statesSleuth Bruce Knuteson

Steps:Steps:

1)1)

eμET

Z4j

eET jj eE

T 3j

W3jeee

ZWμμjj eμE

T j

μμμeee


Step 2: VariablesSleuth Bruce Knuteson

2) 2) Define variablesDefine variables

What is it we’re looking for?

The physics responsible for EWSB

What do we know about it?

Its natural scale is a few hundred GeV

What characteristics will such events have?

Final state objects with large transverse momentum

What variables do we want to look at?

pT’s


If the final state contains Then consider the variable

1 or more lepton

1 or more /W/Z

1 or more jet

missing ET

∑ lTp

∑ ZWTp

//γ

TE

∑ jTp

(adjust slightly for idiosyncrasies of each experiment)

Sleuth Bruce Knuteson Step 2: Variables


Input: 1 data file, estimated backgrounds

transform variables into the unit box define regions about sets of data points

Voronoi diagrams define the “interestingness” of an arbitrary region

the probability that the background within that region fluctuates up to or beyond the observed number of events

search the data to find the most interesting region, R determine P, the fraction of hypothetical similar experiments (hse’s) in

which you would see something more interesting than R Take account of the fact that we have looked in many different places

For each final state . . .

Output: R, P

3) 3) Search for regions of excess (more data events than expected from Search for regions of excess (more data events than expected from background) within that variable spacebackground) within that variable space

Step 3: Search for regions of excessSleuth Bruce Knuteson


If the data contain no new physics, Sleuth will find to be random in (0,1)

If we find small, we have something interesting

If the data contain new physics, Sleuth will hopefully find to be small

If we find large, is there no new physics in our data?

or have we just missed it?

How sensitive is Sleuth to new physics?

Impossible to answer, in general

(Sensitive to what new physics?)

But we can provide an answer for specific cases

SensitivitySleuth Bruce Knuteson


tt provides a reasonable sensitivity check [cf. DØ PRL (1997, 125 pb-1)]

in eμET 2j: find > 2 in 25% of an ensemble of mock experiments

[cf. dedicated search: 2.75 (3 events with 0.2 expected)]

in W 4j: find > 3 in 25% of an ensemble of mock experiments

[cf. dedicated search: 2.6 (19 events with 8.7 expected) w/o b-tag]

[cf. dedicated search: 3.6 (11 events with 2.5 expected) w/ b-tag]

Would we have “discovered” top with Sleuth? No. But results are nonetheless encouraging.

Lessons: b-tagging, combination of channels important for top

other sensitivity checks (WW, leptoquarks) give similarly sensible results

SensitivitySleuth Bruce Knuteson


Results

Results agree well with expectationNo evidence of new physics is observed

DØ data


Sleuth is a quasi-model-independent search strategy for new high pT physics

Defines final states and variables

Systematically searches for and quantifies regions of excess

Sleuth allows an a posteriori analysis of interesting events

Sleuth appears sensitive to new physics

SleuthSleuth finds no evidence of new physics in DØ data finds no evidence of new physics in DØ data

Sleuth has the potential for being a very useful tool Looking forward to Run IILooking forward to Run II hep-ex/0006011 PRD

hep-ex/0011067 PRDhep-ex/0011071 PRL

ConclusionsConclusionsDD


Fragmentation: Higgs-> and Backgrounds

One of the most useful search modes for the discovery of the Higgs in the 100-150 GeV mass range at the LHC is in the two photon mode

Higgs-> has very large backgrounds from QCD sources

Diphoton production o and oo production; jets

fragmenting into very high z o’s

With excellent diphoton mass resolution, can try to resolve Higgs “bump”

Still important to understand level of background


Diphoton Backgrounds in ATLAS Again, for a H-> search at the

LHC, face irreducible backgrounds

from QCD and reducible

backgrounds from o and oo

in range from 70 to 170 GeV

jet-jet cross section is estimated

to be a factor of 2E6 times the cross section and -jet a factor

of 8E2 larger Need rejection factors of 2E7 and

8E3 respectively PYTHIA results seem to indicate

that reducible backgrounds are

comfortably less than reducible

ones …but how to normalize PYTHIA

predictions for very high z fragmentation of jets; fragmentation not known well at high z and certainly not for gluon jets


Different models predict different high z fragmentation

Backgrounds to production in Higgs mass region arise from fragmentation of jets to high z o’s

Pythia and Herwig predict very different rates for high z

all fragmentation is not equal Example of a background that can

be measured in situ, but nice to be able to predict the environment beforehand

DIPHOX (see Run 2 MC workshop) program can calculate , o, and oo cross sections to NLO

comparisons underway to Tevatron data

gg-> and qqbar-> at NNLO may be available soon

B. Webber, hep-ph/9912399


PDF Uncertainties

What’s unknown about PDF’s

the gluon distribution strange and anti-strange

quarks details in the {u,d} quark

sector; up/down differences and ratios

heavy quark distributions

of quark distributions (q + qbar) is well-determined over wide range of x and Q2

Quark distributions primarily determined from DIS and DY data sets which have large statistics and systematic errors in few percent range (±3% for 10-4<x<0.75)

Individual quark flavors, though may have uncertainties larger than that on the sum; important, for example, for W asymmetry

information on dbar and ubar comes at small x from HERA and at medium x from fixed target DY production on H2 and D2 targets

Note dbar≠ubar

strange quark sea determined from dimuon production in DIS (CCFR)

d/u at large x comes from FT DY production on H2 and D2 and lepton asymmetry in W production


Example:Jets at the Tevatron Both experiments compare to NLO QCD

calculations D0: JETRAD, modified Snowmass

clustering(Rsep=1.3, F=R=ETmax/2 CDF: EKS, Snowmass clustering

(Rsep=1.3 (2.0 in some previous

comparisons),F=R=ETjet/2

•In Run 1a, CDF observed an excess in thejet cross section at high ET, outside the range of the theoretical uncertainties shown


Similar excess observed in Run 1B


Exotic explanations


Non-exotic explanationsModify the gluon distribution at high x


Tevatron Jets and the high x gluon

Best fit to CDF and D0 central jet cross sections provided by

CTEQ5HJ pdf’s


D0 jet cross section as function of rapidity

ET (GeV)

d2

dE

T d

(

fb/G

eV) 0.0 0.5

0.5 1.0 1.0 1.5 1.5 2.0 2.0 3.0

DØ PreliminaryDØ Preliminary√↵

=

−

1

pb 92 Ldt √↵

=

−

1

pb 92 LdtRun 1BRun 1B

Nominal cross sections & statistical errors only

JETRAD=ET

max/2

CTEQ4HJ provides bestdescription of data

How reliable is NLO theory inthis region?K-factors?


Chisquares for recent pdf’s

PDF χ2 χ2/ dof ProbCTEQ3M 121.56 1.35 0.01CTEQ4M 92.46 1.03 0.41CTEQ4HJ 59.38 0.66 0.99MRST 113.78 1.26 0.05MRSTgD 155.52 1.73 <0.01MRSTgU 85.09 0.95 0.63

•For 90 data points, are the chisquaresfor CTEQ4M and MRSTgU “good”?•Compared to CTEQ4HJ?


A new tool PDF uncertainties are important both for precision measurements (W/Z

cross sections) as well as for studies of potential new physics (a la jet cross sections at high ET)

Most Monte Carlo/matrix element programs have “central” pdf’s built in, or can easily interface to PDFLIB

Determining the pdf uncertainty for a particular cross section/distribution might require the use of many pdf’s

CTEQ Hessian pdf errors require using 33 pdf’s GKK on the order of 100

Too clumsy to attempt to includes grids for calculation of all of these pdf’s with the MC programs

->Les Houches accord #2 Each pdf can be specified by a few lines of information, if MC programs can

perform the evolution Fast evolution routine will be included in new releases to construct grids for

each pdf


Conclusions Great opportunity at Run 2 at the Tevatron

for discovery of new physics; even better opportunity when the LHC turns on

In order to be believeable, we must understand the QCD backgrounds to any new physics

Don’t rely totally on Monte Carlos and certainly not on one Monte Carlo alone

In the words of Ronald Reagan, “Trust but Verify”, if possible, theoretical predictions/formalisms with data

existing Run 1 data/Run 2 data background” data to be taken at the

LHC If no data, then verify with more complete

theoretical treatments Many new tools/links between old tools are

now being developed to make this job easier for experimenters

Hopefully, we’ll find many more of the type of the event on the right to try them out on

St. Andrews, Scotland J. Huston QCD Backgrounds to New Physics I J. Huston …thanks to Weiming Yao,...

Documents

Transcript of St. Andrews, Scotland J. Huston QCD Backgrounds to New Physics I J. Huston …thanks to Weiming Yao,...