Table of Contents | ||
---|---|---|
|
Update on the HZZ 250 GeV +80/-30 250/fb Analysis for 17-19 September 2013:
...
...
outline | true |
---|
...
Analysis Procedure:
- divide into 4 jet / 6 jet topologies
- this is ZH with H->ZZ* , Z-> nn,ll,qq
- apply preselection depending on topology
- train/apply TMVA'sTMVAs
- validate with cut table
- check remaining backgrounds
Distributions before preselection with only a cut on the reconstructed Higgs mass:
...
...
Preselection:
Simplified and making only clean cuts ...
Code Block |
---|
Evis<140.0: (4 Jet category) #y34>0.0#hmass>95. && hmass<140. #PTvisJETS>25.0 && PTvisJETS<70.0 #nTrks>5. #ej1<120. Evis>140: (6 jet category) #y34>0.0 #hmass>90. && hmass<160. #ej1<120. #jetthrust<0.98 #nTrks>5 |
Distributions after preselection:
aaa
(Note: the lepton pair mass plot has the entries at less than 10 GeV randomly set to 10 and 130 GeV so that the signal peak can be seen.)
...
- Evis JETS
- Mvis JETS
- PTvis JETS
- B Likeness
- C likeness
- Number of HE Electrons
- Higgs Mass
- Z energy
- Zstr energy
- Z cos theta
- Zstr cos theta
- Z mass
- Zstr mass
- azimuthal separation of Z jets
- thrust
- Number Charged Tracks
- Number of electrons
- Number of muons
----NEW VARIABLES:
- y34
- y56
- lepton pair (PDG ID1 = -ID2) mass closest to Mz
- jet pair mass closest to mW
jet pair1 - mW
+
jet pair2 - mW
Performance of different MVA options:
BEFORE
Some signals are just not reasonable to try to select:
1) decays like l+l-H where H->ffnunu
Even in cases where the ff are leptons the initial l+l- are more energetic
and those were being selected.
eeH, H -> cc ee
nu nu H, H -> nu nu nu nu
nu nu H, H -> nu nu nu nu
|
| HZZ generated sample stats |
|
|
|
|
|
|
|
|
| events | weight |
|
I108056.001.stdhep | 10000 | 20000 | 1.29E-004 | e1e1h_zz.Gwhizard- |
I108055.001.stdhep | 10000 | 20000 | 4.36E-003 | e1e1h_zz.Gwhizard- |
I108054.001.stdhep | 10000 | 20000 | 4.11E-004 | e1e1h_zz.Gwhizard- |
I108053.001.stdhep | 10000 | 20000 | 2.70E-005 | e1e1h_zz.Gwhizard- |
I108058.001.stdhep | 10000 | 20000 | 4.29E-003 | e2e2h_zz.Gwhizard- |
I108057.001.stdhep | 10000 | 20000 | 4.00E-004 | e2e2h_zz.Gwhizard- |
I108060.001.stdhep | 10000 | 20000 | 4.28E-003 | e3e3h_zz.Gwhizard- |
I108059.001.stdhep | 10000 | 20000 | 4.00E-004 | e3e3h_zz.Gwhizard- |
|
| Sum = | 1.43E-002 | 10.40% |
I108062.001.stdhep | 10000 | 20000 | 2.54E-002 | nnh_zz.Gwhizard-1_ |
I108061.001.stdhep | 10000 | 20000 | 3.01E-003 | nnh_zz.Gwhizard-1_ |
|
| Sum = | 2.84E-002 | 20.68% |
I108064.001.stdhep | 10000 | 20000 | 8.67E-002 | qqh_zz.Gwhizard-1_ |
I108063.001.stdhep | 10000 | 20000 | 8.08E-003 | qqh_zz.Gwhizard-1_ |
|
| Sum = | 9.47E-002 | 68.93% |
|
|
|
|
|
|
|
|
|
|
|
| All = | 1.37E-001 |
|
Mass of lepton pair with mass closest to mZ:
The following also shows the jet pair mass for those that have mass closest to mW:
Note: Compared to last week plots and tables there was also a simple programming error which I've already fixed.
...
Performance of different MVA options:
BEFORE
Code Block |
---|
[neal@localhost weights]$ grep -A 8 -i optimal- ../hzzv14-presel-v4-new-common0410results.txt
--- Classifier ( #signal, #backgr.) Optimal-cut S/sqrt(S+B) NSig NBkg EffSig EffBkg
-- ------------------------------------------------- |
Code Block |
[neal@localhost weights]$ grep -A 8 -i optimal- ../hzzv14-presel-v4-new-common0410results.txt --- Classifier ( #signal, #backgr.) Optimal-cut S/sqrt(S+B) NSig NBkg EffSig EffBkg -- --------------------------------------------------------------------------------------------- --- Cuts: ( 664.0744,1312202.5) -0.0050 0 0 0 0 0 --- Likelihood: ( 664.0744,1312202.5) 1.0000 0.728991 644.0848 779980.1 0.9699 0.5944 --- Fisher: ( 664.0744,1312202.5) 0.0120 1.70335 258.2099 22721.13 0.3888 0.01732 --- BDTG: ( 664.0744,1312202.5) -0.9823 1.17754 538.5353 208621.3 0.811 0.159 --- BDT: ( 664.0744,1312202.5) -0.0462 3.42709 100.6751 762.2891 0.1516 0.0005809 -- -------------------------------------------- --- Cuts: ( 664.0744,1312202.5) ---------------------------------------------- |
NOW:
Code Block |
---|
if ((fp = fopen("HZZ-TMVA-vars-ffh_zz_all_SM_background_+80e-_-30e+-v6-mini.txt","r"))==NULL) exit(0); [neal@localhost test]$ source storeresultsHZZv4.sh hzz-withlep-withmW-n2-Mhiggspresel-plusptntrkp8g16y34nvtx --- ================================================================================================== --- Classifier ( #signal, #backgr.) Optimal-cut S/sqrt(S+B) NSig NBkg EffSig EffBkg --- 0.0050 0 0 0 0 0 --- Likelihood: ( 664.0744,1312202.5) 1.0000 0.728991 644.0848 779980.1 0.9699 0.5944 --- Fisher: ( 664.0744,1312202.5) 0.0120 1.70335 258.2099 22721.13 0.3888 0.01732 --- BDTG: ( 664.0744,1312202.5) -0.9823 1.17754 538.5353 208621.3 0.811 0.159 --- BDT: ( 664.0744,1312202.5) -0.0462 3.42709 100.6751 762.2891 0.1516 0.0005809 -- --------------------------------------------------------------------------------------------- |
NOW:
Code Block |
---|
if ((fp = fopen("HZZ-TMVA-vars-ffh_zz_all_SM_background_+80e-_-30e+-v6-mini.txt","r"))==NULL) exit(0); [neal@localhost test]$ source storeresultsHZZv4.sh hzz-withlep-withmW-n2-------------------------- --- Cuts: (1035.8644, 1196975) -0.0050 0 0 0 Mhiggspresel-plusptntrkp8g16y34nvtx --- ================================================================================================== --- Classifier ( #signal, #backgr.) Optimal-cut S/sqrt(S+B) NSig 0NBkg EffSig 0EffBkg --- Likelihood: (1035.8644, 1196975) 1.0000 1.66886 315.8824 35510.98 0.3049 0.02967 --- Fisher: (1035.8644, 1196975) 0.0047 2.38122 586.8862 60157.76 0.5666 0.05026 --- BDTG: (1035.8644, 1196975) -0.9795 2.625 293.8675 12238.81 0.2837 0.01022 --- BDT: (1035.8644, 1196975) -0.0427 4.9889 146.0711 711.2005 0.141 0.0005942 --- -------------------------------------------------------------------------------------------------- ------------------------------------------------------------- |
Spent a lot of time playing with different preselection cuts and then settled for only making cuts where
I was cutting almost no signal away and then letting the MVA's do the rest.
Some signals are just not reasonable to try to select:
1) decays like l+l-H where H->ffnunu
Even in cases where the ff are leptons the initial l+l- are more energetic
and those were being selected.
Changed to checking all opposite signed lepton pairs for consistency with coming
from a Z.
|
| HZZ generated sample stats |
|
|
|
|
|
|
|
|
| events | weight |
|
I108056.001.stdhep | 10000 | 20000 | 1.29E-004 | e1e1h_zz.Gwhizard- |
I108055.001.stdhep | 10000 | 20000 | 4.36E-003 | e1e1h_zz.Gwhizard- |
I108054.001.stdhep | 10000 | 20000 | 4.11E-004 | e1e1h_zz.Gwhizard- |
I108053.001.stdhep | 10000 | 20000 | 2.70E-005 | e1e1h_zz.Gwhizard- |
I108058.001.stdhep | 10000 | 20000 | 4.29E-003 | e2e2h_zz.Gwhizard- |
I108057.001.stdhep | 10000 | 20000 | 4.00E-004 | e2e2h_zz.Gwhizard- |
I108060.001.stdhep | 10000 | 20000 | 4.28E-003 | e3e3h_zz.Gwhizard- |
I108059.001.stdhep | 10000 | 20000 | 4.00E-004 | e3e3h_zz.Gwhizard- |
|
| Sum = | 1.43E-002 | 10.40% |
I108062.001.stdhep | 10000 | 20000 | 2.54E-002 | nnh_zz.Gwhizard-1_ |
I108061.001.stdhep | 10000 | 20000 | 3.01E-003 | nnh_zz.Gwhizard-1_ |
|
| Sum = | 2.84E-002 | 20.68% |
I108064.001.stdhep | 10000 | 20000 | 8.67E-002 | qqh_zz.Gwhizard-1_ |
I108063.001.stdhep | 10000 | 20000 | 8.08E-003 | qqh_zz.Gwhizard-1_ |
|
| Sum = | 9.47E-002 | 68.93% |
|
|
|
|
|
|
|
|
|
|
|
| All = | 1.37E-001 |
|
Cuts: (1035.8644, 1196975) -0.0050 0 0 0 0 0
--- Likelihood: (1035.8644, 1196975) 1.0000 1.66886 315.8824 35510.98 0.3049 0.02967
--- Fisher: (1035.8644, 1196975) 0.0047 2.38122 586.8862 60157.76 0.5666 0.05026
--- BDTG: (1035.8644, 1196975) -0.9795 2.625 293.8675 12238.81 0.2837 0.01022
--- BDT: (1035.8644, 1196975) -0.0427 4.9889 146.0711 711.2005 0.141 0.0005942
--- --------------------------------------------------------------------------------------------------
^^^^^^
|
BDT plots:
2) There was also a simple programming error which I've already fixed.
...
Cut table for BDT: (NEW)NEW:
...
Code Block |
---|
$ sed 's/\./,/g' zzhpassing.dat | awk '{if (NF==18) print $3,$7,$15}' | sort -n | uniq -c | awk '{if (NF==4) aa+=($1*$4);print $1*$4"\t"$0} END{print "Sum of weights = "aa}' | sort -k 1,1 -n Sum of weights = 762,5 12,5 1 106561 4f_sz 12,500000 12,5 1 106564 4f_sw 12,500000 12,5 1 106573 4f_zz 12,500000 12,5 1 106580 4f_zz 12,500000 12,5 1 106607 2f_z_ 12,500000 25 2 106551 4f_ww 12,500000 25 2 106562 4f_sz 12,500000 37,5 3 106575 4f_zz 12,500000 37,5 3 106608 2f_z_ 12,500000 50 4 106578 4f_ww 12,500000 62,5 5 106572 4f_sz 12,500000 100 8 106574 106574 4f_zz 12,500000 125 10 106577 4f_ww 12,500000 237,5 19 106576 4f_zz 12,500000 125 10 106577 4f_ww 12,500000 237,5 19 106576 4f_zz 12,500000 |
...
<----
|
Plans:
- just noticed that the preselection nTrks cut is on the PFO count and not the #charged tracks distribution shown so I will switch to
using the charged tracks count in the preselection - may still be able to do better with the leptonic Z decay selection
- high evis events with well reconstructed leptonic Z decays and a clear 6 jet topology (using y56 cut)
- write report