Differences

This shows you the differences between two versions of the page.

--- 2015:groups:higgs:dmhiggs:monoh [2015/06/30 21:09]
harrison.prosper [Information/HowTo]
+++ 2015:groups:higgs:dmhiggs:monoh [2016/02/03 12:11] (current)
nicola.de_filippis
@@ Line 2: / Line 2: @@
 **Members:** Veronica Sanz, Harrison Prosper, Sabine Kraml, Daniel Schmeier, Sezen Sekmen, Giacomo Polesello, Jory Sonneveld, Nishita Desai, Camillo Carillo, Sanjoy Biswas, Monika Wielers, Ilaria Brivio, Jose Miguel No
+====== Task List ======
+The goal of the study we started  in LesHouches is:
+  * derive sensitivity of monoHiggs analyses to probe DM mass hypotheses for different models, at 13 TeV
+  * compare that with what obtained by H->invisible particles analysis and other mono-X analyses
+Below there is a possible list of items to be addressed in preparation for the Les Houches proceedings. Please add more or correct my list if needed:
+  * finalize the list of models to quote and compare; prepare UFO and lhe events for all of them and store in a place accessible by everybody. Documentation about those models has to be also prepared.
+  * finalize the list of background samples to be used and those not yet produced. We need to be sure about cross section. For example in the case of 4l Sanjoy pointed that we need to produce ZH with Z->ll and H->ZZ->2l2nu.
+  * we need to run on full statistics of signal and background samples.
+  * we should have already a working recipe for delphes fast simulation; if not we need to fix that. We need to verify the values of the efficiency  of muons/electron and photons and the fake rate if any, by comparing with current CMS/ATLAS values for the 4l and gamma gamma analyses.  We need to verify the ATLAS-like approach with the smearing for photons and check the MET.
+  * we need to check if we can include also H->WW channel !
+  * we have a preliminary implementation of the 4l and gamma gamma analysis in delphes but we need to check if the efficiency is correct and there are no bugs. …eventually we need to change and optimize cuts for 13 TeV scenario.
+  * we do prepare a list of plots to be produced (like outflow, 4l, -photon invariant mass —signal + background, etc..)
+  * we need to configure and run the tool for limits and include systematics in the statistical treatment.
+  * produce plots about the upper limit on the cross section and compare different models -> turn that in limits on couplings or whatever.
+  * compare sensitivity with H->invisible particle analysis.
@@ Line 21: / Line 41: @@
 ==== Parton Events ====
 The **lhe files** for the different models are here:
+==2HDM==
+  * https://cernbox.cern.ch/index.php/s/LALgzaMjbgj8rNX
 ==hhxx==
@@ Line 39: / Line 62: @@
    * wget https://test-carrillo.web.cern.ch/test-carrillo/higgs/lhe/Higgs_Zprime_nohdecay_1000GeV_13TeV.lhe.gz.orig
-==Higgs Scalar==
-   * wget https://test-carrillo.web.cern.ch/test-carrillo/higgs/lhe/Higgs_scalar_nohdecay_1000GeV_13TeV.lhe.gz.orig
-   * wget https://test-carrillo.web.cern.ch/test-carrillo/higgs/lhe/Higgs_scalar_nohdecay_100GeV_13TeV.lhe.gz.orig
-   * wget https://test-carrillo.web.cern.ch/test-carrillo/higgs/lhe/Higgs_scalar_nohdecay_10GeV_13TeV.lhe.gz.orig
-   * wget https://test-carrillo.web.cern.ch/test-carrillo/higgs/lhe/Higgs_scalar_nohdecay_1GeV_13TeV.lhe.gz.orig
-   * wget https://test-carrillo.web.cern.ch/test-carrillo/higgs/lhe/Higgs_scalar_nohdecay_500GeV_13TeV.lhe.gz.orig
 == Background for 4l final state: ==
@@ Line 86: / Line 103: @@
 ==== Cross Sections ====
+^ Cross sections x BR(H->ZZ->4l, l=e,mu only) (in pb) (2HDM, M_zprime=variable, M_A0=300 GeV, tan(beta)=1, gZ=0.8) ^^
+^ Model File ^ sigma x Br (pb) ^
+|MZP600_MA0300 | 0.04669*1.25E-04 |
+|MZP800_MA0300 | 0.05174*1.25E-04 |
+|MZP1000_MA0300 | 0.04197*1.25E-04 |
+|MZP1200_MA0300 | 0.03176*1.25E-04 |
+|MZP1400_MA0300 | 0.02356*1.25E-04 |
+|MZP1700_MA0300 | 0.01510*1.25E-04 |
+|MZP2000_MA0300 | 0.009734*1.25E-04 |
+|MZP2500_MA0300 | 0.004860*1.25E-04 |
 ^ Cross sections x BR(H->ZZ->4l) (in pb) (M_zprime=1TeV, M_scalar=1TeV) ^^
@@ Line 100: / Line 129: @@
 |Higgs_Zprime_nohdecay_1000GeV_13TeV | 0.00000000003287160 |
 |Higgs_scalar_nohdecay_1GeV_13TeV | 0.000849144 |
-|Higgs_scalar_nohdecay_10GeV_13TeV | 0.000809598 |
-|Higgs_scalar_nohdecay_100GeV_13TeV | 0.00000002508564 |
-|Higgs_scalar_nohdecay_500GeV_13TeV | 0.0000000001063428 |
-|Higgs_scalar_nohdecay_1000GeV_13TeV | 0.000000000000027086640  |
 ^Cross sections x BR(H->gg) (in pb) (M_zprime=1TeV, M_scalar=1TeV):^^
@@ Line 117: / Line 142: @@
 |Higgs_Zprime_nohdecay_500GeV_13TeV | 0.0000001219116 |
 |Higgs_Zprime_nohdecay_1000GeV_13TeV | 0.000000000271548 |
-|Higgs_scalar_nohdecay_1GeV_13TeV | 6.94031999999999982e-03 |
-|Higgs_scalar_nohdecay_10GeV_13TeV | 6.68723999999999968e-03 |
-|Higgs_scalar_nohdecay_100GeV_13TeV | 2.07229199999999993e-07 |
-|Higgs_scalar_nohdecay_500GeV_13TeV | 8.78483999999999983e-10|
-|Higgs_scalar_nohdecay_1000GeV_13TeV | 2.23759199999999990e-13|
 ==== Settings ====
@@ Line 134: / Line 154: @@
 Let me start with the necessary installation steps:
   - Download and install HepMC from here [[http://lcgapp.cern.ch/project/simu/HepMC/download/HepMC-2.06.09.tar.gz]]. In the following I will call the directory /hepmcdir
-  - Download the latest Pythia8 from here [[http://home.thep.lu.se/~torbjorn/pythia8/pythia8209.tgz]]. Note that I had problems using any older version (even 8205!).
+  - Download the latest Pythia8 from here [[http://home.thep.lu.se/~torbjorn/pythia8/pythia8209.tgz]]. Note that I had problems using any older version (even 8205!). If you do have an older version of Pythia lying around, set your library path to start with this pythia version: 'export LD_LIBRARY_PATH=/pythia8dir/lib:$LD_LIBRARY_PATH';
   - <nowiki>./configure --enable-shared --with-hepmc2=/hepmcdir; make; make install</nowiki>. Even if you have Pythia, if you did not compile it with enable-shared options before you **cannot** run the Pythia-Delphes chain! In that case you have to install Pythia8 anew! In the following I will call the directory /pythiadir
-  - Download and install Delphes from here [[http://cp3.irmp.ucl.ac.be/downloads/Delphes-3.2.0.tar.gz]]
+  - Download and install the last verion of Delphes (currently 3.3.2) via github: do 'git clone https://github.com/delphes/delphes.git'
   - Within the Delphes folder, do 'export PYTHIA8=/pythia8dir; make HAS_PYTHIA8=true DelphesPythia8'
   - In order to run Delphes, do './DelphesPythia8 detectorcard.tcl pythiasettings.in output.root'.
-  * For the detector card, you can use the one in delphes/cards/delphes_card_CMS_pileup.tcl until we have decided on the actual settings. **If you want to use the above PileUp sample, you have to explicitly put the file path in the "set PileUpFile" line!**
+  * For the detector card, you can use the one in delphes/cards/delphes_card_CMS_pileup.tcl but that actually does not match the performance of the muon/electron reconstruction and isolation so we are now using the file: '/afs/cern.ch/user/n/ndefilip/delphes_card_CMS_pileup.tcl'. Since we are working on tunings we could still decide to change  the actual settings. **If you want to use the above PileUp sample, you have to explicitly put the file path in the "set PileUpFile" line!** . Make sure to uncomment any lines in the ROOT Tree writer at the bottom of the file
+  add Branch TrackMerger/tracks Track Track
+  add Branch Calorimeter/towers Tower Tower
+as this information is used in the analysis (see github link below).
   * For the pythiasettings, there is a working example below (which you can just store in a text file)
@@ Line 204: / Line 229: @@
 > Daniel: This does not seem to be the standard way ATLAS/CMS set limits, which is based on the frequentist-CLs. Why are we not using that one?
-> Harrison: It seems reasonable to me to favor statistical methods that are well-founded, by which I mean methods that professional statisticians (rather than amateurs like us) have shown to be sound. Just as people defer to physicists about physics, I think it reasonable to defer to statisticians about statistics. The method implemented in this package is Bayesian, which is well-founded in this sense. By contrast, CLs is not.  Moreover, 1) asymptotic CLs may be overly optimistic when observed counts are low; 2) the ATLAS/CMS method requires performing two fits per toy experiment and therefore requires a rule to deal with non-converging or bad fits; 3) the use of profile likelihoods, which underlies the ATLAS/CMS method, when there are multiple nuisance parameters in the problem can yield results that may be too optimistic, and 4) worst of all, CLs (when done correctly) is a huge sink of CPU and manpower for computing numbers which, in the end, simply say that nothing was found!
+> Harrison: It seems reasonable to me to favor statistical methods that are well-founded, by which I mean methods that professional statisticians (rather than amateurs like us) have shown to be sound. Just as people defer to physicists about physics, I think it reasonable to defer to statisticians about statistics. The method implemented in this package is Bayesian, which is well-founded in this sense. By contrast, CLs is not.  Moreover, 1) asymptotic CLs may be overly optimistic when observed counts are low; 2) the ATLAS/CMS method requires performing two fits per toy experiment and therefore requires a rule to deal with non-converging or bad fits; 3) the use of profile likelihoods, which underlies the ATLAS/CMS method, when there are multiple nuisance parameters in the problem can yield results that may be too optimistic, and 4) worst of all, CLs (when done correctly) is a huge sink of CPU and manpower for computing numbers which, in the end, simply say that nothing was found! Because of the enormous amount of effort consumed computing CLs during Run I, the CMS SUSY group is revisiting the question of whether limit setting using CLs is sensible for Run II. High energy physicists are prone to a "Masters of the Universe" complex, which is manifested in our dismissal of the wisdom of other professionals. Alas, since I am a high energy physicist //ipso facto// I am prone to some level of arrogance...but I am not so arrogant as to dismiss two and a half centuries of work on statistics in favor of something like CLs! Surely, Laplace, Fisher, Neyman, Bernardo, Berger and dozens more have useful things to teach us?\\
-Because of the enormous amount of effort wasted computing CLs during Run I, the CMS SUSY group is revisiting the question of limit setting. We high energy physicists tend to think we are Masters of the Universe who do not need to pay attention to the wisdom of other professionals. Since I am a high energy physicist //ipso facto// I am arrogant...but not so arrogant as to dismiss two and a half centuries of work on statistics in favor of a half-baked, home-grown, idea like CLs!\\
+> Daniel: Since you mention the disadvantages of CLs (which certainly are true!), I have to write a few lines to 'defend' it: Bayesian statistics might be mathematically well founded (more than CLs), but scientifically raises the huge issue that one has to put in a prior degree of belief in a theory via $\pi(\sigma)$. From a theoretical point of view, there is no reason why to assume this to be flat, lognormal, spiky or anything else. I could have a model which describes Nature (= the observation) perfectly for one particular parameter setting but completely fails for any other and if my prior is anything different than a delta-distribution, a Bayesian approach will exclude it due to the marginalisation.
+> I don't want to raise a discussion about which approach is the best, because neither one is. My only concern is - and that was the main point behind my original question - that for the sake of comparison, one should stick to the statistical methods which have been defined as standard, unless there is a strong reason why the standard approach is not applicable to the given scenario. For the models we have, I don't see any such reason. Moreover, the ATLAS study linked in the References section determines their limit using CLs and we most likely want to compare our result to theirs. If we use different means to quantify our results, I fear it will be hard to disentangle the physics from the statistics.
 This standalone tool can be used to compute upper limits on a cross section (or signal strength) given one or more observed counts, $N$. The package can be downloaded from ''github'' using
@@ Line 295: / Line 323: @@
 **Final remarks**
 Ideally, every analysis chain comprises two independent modules: event by event, the first module takes as input the nominal (experiment or framework-dependent) objects and a particular point in the space of systematic parameters and outputs the re-calibrated framework-independent objects, ideally, as vectors of ''TParticle'' objects, while the second module takes as input the re-calibrated ''TParticle'' objects and outputs histograms and, or, flat ntuples. The second module should have no knowledge of the provenance of the ''TParticle'' objects. If we wrote our analysis codes this way, they would become framework-independent reusable modules. But, I've just realized I have been dreaming!
+====== Proceedings ======
+We started to work on the proceeding. The deadline is end of February. You can find the last version at link:
+{{:2015:groups:higgs:dmhiggs:proceeding.zip|}}

Les Houches

User Tools

Site Tools

Differences

Page Tools