Issues and Run-II Musings About the +MET Analysis

Osamu Jinnouchi (Tokyo Tech), Ryan Reece (SCIPP), Sheena

Schier (SCIPP), Bruce Schumm (SCIPP)

Prepared for the Inclusive Strong Production Sept 2014 Oxford SUSY Workshop Live Page

September 2014 2

NOTE!!!

This set of slides covers all the areas of the review within this single set.

However, the sections are clearly separate out, and the background material provided by the other sections will likely help anyone reviewing them to understand what they are.

Apologies…

September 2014 3

Conference Note public in early January 2014:

Search for Supersymmetry in Diphoton Events with Large MissingTransverse Momentum in 8 TeV pp Collision Data with the ATLASDetector

ATLAS-CONF-2014-001

Final result in preparation

September 2014 4

Diphoton+MET Personnel (beginning of R2)

Osamu Jinnouchi, Tokyo Tech Faculty

Ryan Reece, UCSC/SCIPP Post-doc

Sheena Schier, UCSC/SCIPP Ph.D. Graduate Student

Bruce Schumm Faculty

XXX, Tokyo Tech Unnamed Masters Student

Possible additional contributors (encouragement welcome!)

Khliesh Mistry, U. Pennsylvania Ph.D. Graduate Student

Brig Williams, U. Pennsylvania Faculty

September 2014 5

Summary of SRs

SP (WP): Strong (Weak) production

1 (2): High- (low-) mass bino production

MIS: Selection driven purely by expected background studies

September 2014 6

Summary of Backgrounds

Data Driven

Data Driven

Data-scaled

MC

Pure MC

September 2014 7

Opening Note: Diphoton+MET Triggers

• All Diphoton+MET SRs require two tight photons with ET > 75 GeV

• However, QCD control region makes use of a subset of [loose-tight] (as well as tight) photons with ET > 50 GeV

• The QCD control samples are statistically limited

• In RUNI, we used 2g40_loose trigger

• RUN2 proposal is medium photons of 35 and 25 GeV

• We have not had a chance to study the effect of the potential reduction in control sample sizes due to going from loose to medium.

September 2014 8

Section I: Data Driven BackgroundsBackground A: QCD Define various control samples

“t” require 1 tight, iso photon“g” = pseudophoton (reverse two PID bits)“g” can be isolated or not

•QCDg•QCDg+iso•QCDtg•QCDtg+iso

QCDtg+iso reproduces MET dist well but limited statisticsQCDtg reproduces QCDtg+iso well at high MET, fair statsQCDg provides high statistics but has higher tails

September 2014 9

Section I: Data Driven BackgroundsBackground A: QCD

Background estimated by applying all cuts save MET and then scaling control sample (QCDtg) to distribution

Could work on quality of control sample agreement with distribution (study effect of relaxing isolation requirement, etc.)

Could revisit contribution from true background which does not have jet fake and might have different MET distribution (but was seen to be small in prior versions of analysis)

Could think about new approach… ?

September 2014 10

Section I: Data Driven BackgroundsBackground B: Electroweak

• Estimated via dedicated e control sample scaled by measured e fake rate (fairly standard)

• Studies show 25% of W/Z/t- backgrounds do not have e fake, but some of this expected to be incorporate in “QCD” background estimate

Adopt +-25% systematic uncertainty. Want to reduce with further study.

Me (converted

photon)

Me (unconverted

photon)

September 2014 11

Section I: Data Driven BackgroundsBackground C: W Irreducible Background

Concern: very large K factor arises from elimination of a cancellation when gluon radiation included

Associated with higher values of W system recoil (pT,l)

Our selection places all our W background at high values of pT,l

Our MC model (Alpgen) not a full NLO model

Constrain W contribution from data

LO

NLO

September 2014 12

Section I: Data Driven BackgroundsBackground C: W Irreducible Background

Constrain W K-factor for pT,l > 100

Constrain K-factor with dedicated l control sample

W contribution (white) includes x3 “enhancement factor” relative to Alpgen

Actual analysis makes use of simultaneous fit to control and signal region, but in limit of 0 signal, W floats up to about x3 in fit.

September 2014 13

Section I: Data Driven BackgroundsBackground C: W Irreducible Background

However:*) Subsequent MC-based studies by Ben Kaplan showed good agreement between Alpgen and a full NLO calculationhttps://indico.cern.ch/event/319991/contribution/3/material/slides/0.pdf

*) A direct measurement (data!) of W production in the SM Working Group, although in a slightly less restricted kinematic space (looser selection cuts) support the x3 enhancement factor at some level

Needs to be resolved

Ben Kaplan

Can we use VBFNLO as a generator for our W background samples?

September 2014 14

Section II: Direct Use of MCUse A: Direct Estimate of Z Background

Makes use of Sherpa for kinematic distributions

Sherpa cross section corrected by ratio of Sherpa cross section to MadGraph NLO cross section calculation over same kinematic region

Result used directly to estimate Z backgrounds to both signal and l control regions

Scale uncertainty of 50% assumed – can this be reduced with further study?

Consider data-driven Z study

Does SM group have corresponding analysis?

September 2014 15

Section II: Direct Use of MCUse B: Estimating Backgrounds to l Con-trol Sample

• W, Z Z and ttbar- backgrounds to the l control sample are estimated directly from the MC

• All appropriate K-factors are used

• In the control region pT,l they are small

September 2014 16

Section II: Direct Use of MCPotential Use C: Possible tt Background

At 13/14 TeV, might the tt start to contribute at an appreciable level?

Need to explore this, perhaps even with new (?) dedicated MC sample.

September 2014 17

Section III: Signal Studies – Pointing vs. Non-Pointing Photons• For prompt photon analyses (ours!), there is a tension between the bino

decay length for low bino mass and direct decays of strongly produced states (gluinos, squarks) to gravitinos for high bino masses.

• Finessed by fine tuning of the coupling between SM and gravitational sectors – we have always felt this to be a bit unnatural

• Have talked for some time about making the bino lifetime a third dimension in the parameters space (e.g. a parameter space in the three dimensions of gluino mass, bino mass and bino lifetime)

• This would involve generating 3D grids and combining the prompt diphoton+MET and non-prompt-photon analyses

• Note that CMS already makes use of 3D parameter spaces – gluino mass, squark mass, bino mass – but the idea of having bino lifetime be a parameters in this manner would be both natural and unique to ATLAS.

September 2014 18

Section IV: Early Data StrategyTrigger Perspectives

For Run 1 (8 TeV) we used 2g40_loose (two >40 GeV loose EM objects)

Our offline SR selections made use of tight photons down to 75 GeV

Our offline CR selections made use of tight and loose photons down to 50 GeV (lower ET needed to procure statistics

For our future analysis, we believe that the appropriate trigger will be

2gXX_loose

where XX is as low as possible. It would be good to know ASAP what XX might be expected to be.

September 2014 19

Section IV: Early Data StrategyFollow-up on Run I Excess (?)

Our WP2 selection identified 5 events over a background of 2.4 0.6

1.6 sigma excess

May not warrant special attention but in the spirit of due diligence we point this out.