Slide 1Slide 2Slide 3Slide 4Slide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Slide 20Slide 21Slide 22Slide 23Slide 24Slide 25Slide 26Slide 27Slide 28Slide 29Slide 30Slide 31Slide 32Slide 338.882 LHC PhysicsExperimental Methods and MeasurementsSearch Strategies and Observations[Lecture 14, March 30, 2009]C.Paus, 8.882 LHC Physics: Search Strategies and Observations2Organizational Issues Project 2•due April 9 Project 3 is coming up quickly afterwards•due May 2 Guest lecture planned on:•Baysian versus frequentistic approach to statistics•Michael BetancourtC.Paus, 8.882 LHC Physics: Search Strategies and Observations3Lecture Outline Search Strategies and Observations•introduction•general methods•analysis biases and how to avoid them•two examples•resonance searches: D0→μ+μ–•search for an oscillation frequency: Bs oscillationsC.Paus, 8.882 LHC Physics: Search Strategies and Observations4Introduction Different types of analysis in particle physics•measurement•yields a number with an uncertainty for a given observable•that number should say something about the Standard Model (SM)•at best: could be inconsistent with the SM: something is wrong•at least: knowledge of SM improved, better test: bread and butter •search•assume some physics model (usually new physics or even SM)•determine some characteristic observable to verify the model•analyze data concerning observable: is model supported by data?•yes: a signal was found, turn the search into a measurement•no: search yields a limit with a confidence level•at best: find physics beyond the Standard Model•at least: set a new limit for a given physics modelC.Paus, 8.882 LHC Physics: Search Strategies and Observations5Typical Design of a Search Physics assumption•search for SM Higgs at mass 115 GeV•search for Z' (mass larger then Z mass)•new physics (NP) appears in tail of high transverse momenta Choice of observable•SM Higgs: best channel H→γγ, observable mγγ•Z': prime decay channel Z' →μ+μ–, observable mμμ•NP: all final states with identifiable objects (j – jet)•μμ, μμμ, μe, μμe, ee, eee, eeμ, jj, jjj, ....•use events falling into the high pT portionC.Paus, 8.882 LHC Physics: Search Strategies and Observations6Typical Design of a Search Optimization with Monte Carlo / data•find quantity QS which defines sensitivity of analysis•classical optimization quantities: • used for search without knowledge of cross section•careful, unnatural behavior at low S• used when signal cross section is known•careful, it really optimizes the measurement if search successful•improved quantity: •see www.cmsaf.mit.edu/twiki/bin/view/Class8882/WebHome •find optimal set of cuts maximizing QS Include systematic uncertainties•in many cases the effect is small•for large uncertainties statistical methods complicated: see www.cmsaf.mit.edu/twiki/bin/view/Class8882/WebHomeC.Paus, 8.882 LHC Physics: Search Strategies and Observations7Typical Design of a Search Look at fully optimized data analysis•find a signal: determine significance•method reasonably straight forward•see examples in what follows•conventions•observation or discovery at 5 standard deviations•evidence is to mark the turf, at about 3 standard deviations•evidence became more popular since experiments take so long to accumulate statistics to have an observation•set a limit at a given confidence level (usually 95%)•once sensitivity is known confidence level straight forward•usually implies Poisson statistics: few events are found•see examples in what followsC.Paus, 8.882 LHC Physics: Search Strategies and Observations8Clever Hans Effect Astonishing horse (real story)•Hans von Osten in early 1900 could do math!•Hans added up numbers, result communicated in number of pawns on the ground•trainer cues? no, trainer was send outside, hmmm?•1907 Oskar Pfungst proposed: result should be unknown to people in the room: Hans lost all ability to add numbers•horse sensed cues from people who knew the answer•really a clever horse•imagine how 'clever' we could be ....C.Paus, 8.882 LHC Physics: Search Strategies and Observations9The Hawthorne Experiment* Study of illumination** (optimal value for productivity)•Study 1a: In the first experiment, there was no control group. The researchers experimented on three different departments; all showed an increase of productivity, whether illumination increased or decreased.•Study 1b: A control group had no change in lighting, while the experimental group got a sequence of increasing light levels. Both groups substantially increased production, and there was no difference between the groups. This naturally piqued the researchers' curiosity.** source is wikipedia* Hawthorne experiments have been intensely discussed. Read up on the web if you like.C.Paus, 8.882 LHC Physics: Search Strategies and Observations10The Hawthorne Experiment Study of illumination (optimal value for productivity)•Study 1c: The researchers decided to see what would happen if they decreased lighting. The control group got stable illumination; the other got a sequence of decreasing levels. Surprisingly, both groups steadily increased production until finally the light in experimental group got so low that they protested and production fell off.•Study 1d: This was conducted on two women only. Their production stayed constant under widely varying light levels. It was found that if the experimenter said bright was good, they said they preferred the light; the brighter they believed it to be, the more they liked it. The same was true when he said dimmer was good. If they were deceived about a change, they said they preferred it. Researchers concluded that their preference on lighting level was completely subjective - if they were told it was good, they believed it was good and preferred it, and vice versa.C.Paus, 8.882 LHC Physics: Search Strategies and Observations11Effects in Physics Measurements of the speed of light*•measurements around 1930-40 are clearly off•investigating these measurements it was concluded:•the investigator searches for the source or sources of errors, and continues to search until he gets a result close to the accepted value. Then he stops! Particle physics, even recently has many more examples* from