1. 1 QCD fits via TFractionFitter ● I took another look at the QCD fit problem: ● A fraction of the fits hangs in an infinite loop ● A (smaller) fraction of the fits doesn't converge ● Solutions: ● Exclude bins where either data, QCD template, or EWK template has <5 entries – Tested on ~100k fits → never hangs anymore! ● If the fit fails, keep trying with a different starting point: EWK fraction = [0..1.0] with step 0.05

2. 2 Excluded bins EXAMPLE: This fit fails to converge out-of-the-box But excluding some low- statistics bins fixes the problem.

3. 3 Wmunu control plots: eta ● Using latest reco and trigger SF

4. 4 μ-μ-μ+μ+ ● Signal Monte-Carlo is Powheg+Pythia ● QCD is data-driven and estimated separately in each eta bin ● QCD scale is average of 3 EWK templates (Powheg+Pythia,Powheg+Herwig, MC@NLO) ● Still seeing “choppiness” on the order of 1-2% ● A couple of bins are a lot worse. In particular, eta=-2.1 still shows a ~5% effect

5. 5 Double ratio: mu+/mu-

6. 6 μ-μ-μ+μ+ ● Signal Monte-Carlo is Powheg+Herwig ● Everything else stays the same

7. 7 μ-μ-μ+μ+ ● Signal Monte-Carlo is MC@NLOMC@NLO ● Everything else stays the same

8. 8 μ-μ-μ+μ+ ● Signal Monte-Carlo is Powheg+Pythia ● Looking at |η| here ● Note that bin at |η| = 2.1 is noticeably off due to discrepancy on C-side

9. 9 Debugging “bad” eta bin ● Second to last bin on C-side (eta ~ -2.1)

10. 10 Recall that the bin at eta=-2.1 looks fine in Z events, but is 5% off in W events. What's different between W's and Z's? One suspect is MET. Perhaps somehow the MET calculation gets screwed up whenever a muon lands in the eta=-2.1 region. I tried to look at the MET_MuonBoy term for events where muon falls in eta=-2.1 bin.

11. 11 C-Side: eta = [-2.18..- 1.95] A-Side: eta = [1.95..2.18] Nothing is horribly wrong on the C-side at the low edge (around 20 GeV, muon pT cut) But big difference in the 40..60 GeV region MET_MuonBoy in symmetric bins on C-side and A-side

12. 12 C-Side: eta = [-2.18..- 1.95] A-Side: eta = [1.95..2.18] MET_RefFinal in symmetric bins on C-side and A-side C-side data/ratio is uniformly lower

13. 13 C-Side: eta = [-2.18..- 1.95] A-Side: eta = [1.95..2.18] Muon phi in symmetric bins on C- side and A-side Notice un-natural jumpiness in C-side. Looks like a detector effect. Instrumentation difference on C-side and A- side?

14. 14 C-Side: eta = [-2.18..- 1.95] A-Side: eta = [1.95..2.18] Muon pT in symmetric bins on C- side and A-side C-side data/ratio is uniformly lower

15. 15 C-Side: eta = [-2.18..- 1.95] A-Side: eta = [1.95..2.18] Muon pT (ID and MS parts) PT (ID) PT (MS)

16. 16 Nothing obvious

17. 17 Wmunu control plots: kinematic ● In all cases, Powheg+Pythia is used for signal ● CAVEAT: in the following plots, QCD is normalized in the pT and eta inclusive region.

18. 18 μ-μ-μ+μ+ ● Right tail disagreement in MET is always seen in Powheg+Pythia

19. 19 μ-μ-μ+μ+ ● Electron plots show a similar disagreement in left tail of WMT

20. 20 μ-μ-μ+μ+ ● W pT looks good after reweighting (reweighted to Pythia8)

21. 21 μ-μ-μ+μ+ ● Lepton pT out-of-the-box looks horrible ● Largely due to the first bin at 20..25 GeV (which is dropped in the electron channel) ● Remember that QCD is normalized in a pt-inclusive region for this plot. Perhaps re-doing this plot with QCD normalization separately done in each pT slice will make it better. ● The structure around the peak (pT = 35..40 GeV) is strange, and is also seen with heavy-flavor MC (next slide)

22. 22 μ-μ-μ+μ+ ● Save as the previous plot, but QCD comes from heavy-flavor Monte-Carlo normalized to its cross-section (i.e., no template fit is done)

23. 23 Cross-check with alt. selection ● From Wc/W cross-section ratio measurement: ● Add etcone40/et<0.2 to ptcont40/pt<0.1 ● Add |z0|<1.0 and d0sig<2.5 ● QCD template: d0sig>2.5 ● QCD normalized in MET = 0..100 GeV ● Jet veto: njets <= 2 ● They use Alpgen as primary W MC ● Caveat: I don't have heavy-flavor Alpgen in my plots (didn't have the samples handy) ● Only showing mu- plots below

24. 24 MET fit – output of TFractionFitter SIG+EWK = Powheg+PythiaSIG+EWK = Alpgen+Herwig Note that TfractionFitter is allowed to modify the shape of the templates bin-by-bin (within statistics), so the agreement here looks better than in stack plots. Also, EWK template has a floating scale, too - making the agreement even better met spot-on!

25. 25 MET fit – using original templ. shape SIG+EWK = Powheg+PythiaSIG+EWK = Alpgen+Herwig Here, we use the fitted fractions (scale factors) from TfractionFitter, but original templates EWK still has a renormalized floating scale. Notice that you now see many more bin by bin variations in the ratio plot met spot-on!

26. 26 MET stack SIG+EWK = Powheg+PythiaSIG+EWK = Alpgen+Herwig Here: Template shapes are taken out-of-the-box (without bin-by-bin smoothing from TfractionFitter) EWK normalization is NOT fitted but taken as absolute (from cross-section)

27. 27 Muon pt stack SIG+EWK = Powheg+PythiaSIG+EWK = Alpgen+Herwig This is the same plot as above, BUT: Template shapes are taken out-of-the-box (without bin-by-bin variation from TfractionFitter) EWK normalization