WEBVTT

1
00:00:02.610 --> 00:00:05.009
Julie Hogan: Advanced physics objects.

2
00:00:13.530 --> 00:00:20.280
Julie Hogan: And in this lesson, those of you who were able to read the first section ahead of time.

3
00:00:22.890 --> 00:00:34.860
Julie Hogan: zone here in the schedule, so the the first section here was an introduction to how we make corrections we've talked about now all of these selections.

4
00:00:35.910 --> 00:00:46.500
Julie Hogan: And all of those selections Unfortunately, we have no guarantee that they work exactly the same for data and for simulation and we must.

5
00:00:47.850 --> 00:01:08.370
Julie Hogan: take that into account, we must try to measure that that difference and correct for it, so this entire lesson is going to talk about corrections that we apply on simulations so I wanted to make sure we began with some time for questions so did anyone in your reading of this.

6
00:01:09.750 --> 00:01:19.710
Julie Hogan: about how to measure efficiencies and and create scale factors did anyone come up with any questions that you wanted to ask.

7
00:01:30.060 --> 00:01:32.700
Dillon Fitzgerald: yeah maybe this is too specific, but.

8
00:01:33.900 --> 00:01:37.110
Dillon Fitzgerald: If you were thinking about measuring like a cross section.

9
00:01:40.170 --> 00:01:47.520
Dillon Fitzgerald: And you know the scale factors, do you use that to correct the luminosity the yields are both.

10
00:01:50.520 --> 00:01:51.150
Julie Hogan: So.

11
00:01:51.540 --> 00:01:55.860
Julie Hogan: If you're measuring if you're wanting to measure, a cross section.

12
00:01:55.950 --> 00:02:13.500
Julie Hogan: of a certain process in data, where the scale factors would come in, is when you are when you are performing some comparison of that data to any simulation.

13
00:02:14.760 --> 00:02:34.440
Julie Hogan: So we do tend to treat the data as correct and these scale factors would apply to any simulation you're using to support your analysis or to make some comparison, perhaps to how that process has been generated in in Monte Carlo generators.

14
00:02:37.230 --> 00:02:37.560
Dillon Fitzgerald: Thank you.

15
00:02:51.360 --> 00:03:06.630
Julie Hogan: So the takeaways from this and if you weren't able, if you weren't able to read it, you can come back later the takeaways from this is that all of these choices we've made so everything we just talked about like if i'm going to select new ons of tight ID or tight isolation.

16
00:03:07.920 --> 00:03:18.540
Julie Hogan: All of those will carry along some correction that I need to apply to the simulation to ensure that it's giving me the same the same.

17
00:03:19.860 --> 00:03:22.860
Julie Hogan: Performance let's say as in as in data.

18
00:03:24.120 --> 00:03:27.000
Julie Hogan: And we do this by measuring efficiency.

19
00:03:28.080 --> 00:03:30.780
Julie Hogan: Efficiency there's a definition of it up here.

20
00:03:32.220 --> 00:03:41.340
Julie Hogan: In yellow efficiency means the number of objects of a certain type that past a certain algorithm out of the total number of objects of that type.

21
00:03:43.440 --> 00:03:59.790
Julie Hogan: We do that by measuring efficiencies in data in data we have a harder time of it, we have to use physics knowledge to isolate to isolate groups of common objects and I we shared in this page, the most common method of doing that, which is the tag and probe method.

22
00:04:00.930 --> 00:04:14.610
Julie Hogan: In simulation will often repeat that method, but it's also possible here to use our Monte Carlo truth information if if it's needed for a certain identifying a certain type of object like be corks.

23
00:04:16.020 --> 00:04:33.120
Julie Hogan: And then we form scale factors which are the ratios of these efficiencies so we always want to make the efficiency and simulation match the efficiency and data we do that with a multiplicative correction which always carries and uncertainty because that's life.

24
00:04:34.560 --> 00:04:34.920
Julie Hogan: Okay.

25
00:04:36.270 --> 00:04:38.370
Julie Hogan: So i'm going to hand it off.

26
00:04:38.670 --> 00:04:40.410
Franz Glessgen: I just a quick last question so for into it.

27
00:04:40.440 --> 00:04:41.160
Julie Hogan: yeah go ahead.

28
00:04:41.220 --> 00:04:44.640
Franz Glessgen: So at what point do you consider that skin factors to big.

29
00:04:46.260 --> 00:04:48.720
Franz Glessgen: scale factor of two What do you do, or three.

30
00:04:49.650 --> 00:04:51.120
Julie Hogan: Sure yeah um.

31
00:04:53.640 --> 00:05:08.160
Julie Hogan: that's a great question when you consider the scale factor too big, I personally would consider a scale factor too big, if it if it gets even into the 10% ish range or 15 percentage range much, much less 100%.

32
00:05:08.640 --> 00:05:13.680
Julie Hogan: And so that is part of part of the process of designing these algorithms for sure.

33
00:05:14.640 --> 00:05:29.520
Julie Hogan: When when the new on group and cms is coming up with the criteria that are tight and loose and medium if there's some criterion that is causing the performance to diverged really badly, we need to study that and we need to understand what's going wrong.

34
00:05:30.420 --> 00:05:36.450
Julie Hogan: Probably in the simulation what's going wrong to cause that and we would try to mitigate that type of effect.

35
00:05:39.210 --> 00:05:40.500
Julie Hogan: Sometimes there can be.

36
00:05:41.040 --> 00:05:44.580
Julie Hogan: yeah well we'll see some cases where you know in some tiny.

37
00:05:44.580 --> 00:05:56.490
Julie Hogan: corner of the phase space you start to get large so, for instance, like at very, very low momentum, you might have large scale factors, whereas at a medium or higher momentum range they're acceptable.

38
00:05:57.540 --> 00:06:05.730
Julie Hogan: So sometimes we'll deal with large scale factors by saying we don't recommend you use this object below a certain momentum or things like that.

39
00:06:06.750 --> 00:06:20.880
Franz Glessgen: Okay, and have a follow up question so because it's mentioned in your in your prep that those gating factor I attributed to annoys me detector you need me other other factors for that which have a large influence.

40
00:06:22.500 --> 00:06:23.010
Julie Hogan: um.

41
00:06:26.400 --> 00:06:32.880
Franz Glessgen: So, the fact is that would that would account for this 10 to 15% differences to simulation and data.

42
00:06:34.410 --> 00:06:35.250
Franz Glessgen: sure.

43
00:06:35.730 --> 00:06:36.510
So.

44
00:06:38.430 --> 00:06:48.510
Julie Hogan: it's really difficult to pin down exactly where they come from, probably most the you know all different objects will have different sources, specifically for.

45
00:06:51.120 --> 00:06:53.940
Julie Hogan: for things like jets.

46
00:06:55.590 --> 00:07:03.120
Julie Hogan: There can be differences, because of the flavor which we have trouble identifying there can be differences, because of the.

47
00:07:04.890 --> 00:07:13.590
Julie Hogan: Because of the background pile up that we're we can never you know it's changing every moment and we can never guarantee what it is.

48
00:07:14.850 --> 00:07:18.240
Julie Hogan: There there'll be a wide variety of sources.

49
00:07:20.550 --> 00:07:33.240
Julie Hogan: The best information on sources like that might be the papers on the performance of each of these objects in the detector so most the detector systems have made publications.

50
00:07:34.140 --> 00:07:43.350
Julie Hogan: You looked at one, I think, for the electrons and protons in the pre exercises, there are similar from you ons for jets and they will go into some of these sources of.

51
00:07:44.940 --> 00:07:45.870
Julie Hogan: of corrections.

52
00:07:47.010 --> 00:07:47.460
Franz Glessgen: Right okay.

53
00:07:50.970 --> 00:08:06.180
Julie Hogan: Okay i'm going to hand it to a Nina So the first The first lesson here we're going to do is on corrections for you on momentum scales so i'm going to stop sharing my screen so that she can do her own.

54
00:08:14.400 --> 00:08:15.630
Julie Hogan: Okay, can you hear me.

55
00:08:17.430 --> 00:08:23.610
Anniina Kinnunen: Yes, good okay so hi everyone i'm on Nina and i'm going to talk to you a little about me on corrections.

56
00:08:24.510 --> 00:08:28.260
Anniina Kinnunen: let's start with why we have these corrections in the first place.

57
00:08:28.890 --> 00:08:44.130
Anniina Kinnunen: So there are misalignments in the cms detector that make the reconstruction of your momentum biased and the cms reconstruction software doesn't fully correct these misalignments and that's why we need to add some additional corrections and the corrections.

58
00:08:45.750 --> 00:09:00.210
Anniina Kinnunen: are important, when you are using the meal momentum in precision measurements, because the results will be affected by the bias, but if you're using let's say only the number of meals, then the corrections are not necessary.

59
00:09:02.460 --> 00:09:13.410
The corrections we use are the mirror and momentum scale corrections, also known as the Rochester corrections and they are available to you, to the moon corrections to github page.

60
00:09:15.150 --> 00:09:22.140
Anniina Kinnunen: In the setup setup page of this lesson, you have the corning instructions for this.

61
00:09:23.760 --> 00:09:33.900
Anniina Kinnunen: repository if you have not yet cloned it and call it now, and also, if you call it yesterday, you can just try and get full to make sure that you have the latest version of it.

62
00:09:36.360 --> 00:09:50.820
Anniina Kinnunen: So the mian corrections are extracted in a two step method and the first step, the corrections are obtaining bins of the charge of the meal on the Ad the Ad fee coordinates of the me on track and we will be working with those variables.

63
00:09:52.560 --> 00:09:54.210
The beacons.

64
00:09:56.280 --> 00:10:03.120
Anniina Kinnunen: And in the second step, the corrections are just find using the mass of the Z bows on.

65
00:10:05.070 --> 00:10:14.820
The corrections are different for data and Monte Carlo because the Monte Carlo and start with no biases, but they can be induced during the reconstruction and corrections have been extracted for both.

66
00:10:16.830 --> 00:10:30.690
In this example we're going to use different one Rochester corrections and, like the assess we're only going to use a Monte Carlo dataset because the 2012 data set would take too long to run, but you can run it later if you want.

67
00:10:32.520 --> 00:10:43.440
And if you want to check the official code for the corrections it's in the register corrections directory in the github repo and the code we're kind of us now is in the test directory.

68
00:10:47.370 --> 00:10:49.620
So we can actually apply the corrections now.

69
00:10:51.540 --> 00:10:56.280
So you should have your terminal open in your vm or.

70
00:10:58.680 --> 00:11:04.200
In your docker container I have the docker container ready here, so you can just.

71
00:11:05.670 --> 00:11:10.410
enter the mirror and corrections tool and open route.

72
00:11:12.300 --> 00:11:17.790
So we're going to start by compiling the official code by running these two lines here.

73
00:11:19.320 --> 00:11:20.400
So you can just.

74
00:11:21.480 --> 00:11:24.600
copy paste those into the terminal it shouldn't take.

75
00:11:25.650 --> 00:11:26.490
too long.

76
00:11:31.890 --> 00:11:34.440
it's going to give a little warning here, but you can ignore that.

77
00:11:36.300 --> 00:11:41.370
Anniina Kinnunen: Then, before we run the analysis, which is where we actually apply the corrections i'm going to.

78
00:11:42.420 --> 00:11:45.690
Explain a little how how the corrections are applied there.

79
00:11:48.240 --> 00:11:57.930
So we have the first the main function of analysis, which is where we just call this apply corrections function and the main is just simply so that you can call.

80
00:11:59.190 --> 00:12:00.810
Anniina Kinnunen: Multiple files in one run.

81
00:12:01.920 --> 00:12:06.870
Anniina Kinnunen: And as you can see, the data file is now commented here, because it would take too long, as I said.

82
00:12:08.460 --> 00:12:28.620
Anniina Kinnunen: And here is the Monte Carlo file which we're going to use now and as a parameter we give simply the name of the Monte Carlo file the path to the file and this boolean variable which tells where the where we're using data or MC and also means that we're using MC.

83
00:12:31.320 --> 00:12:31.830
Then.

84
00:12:32.880 --> 00:12:43.380
Anniina Kinnunen: we're going to use teachers throughout this function, and the first thing apply corrections does is just open the file, we gave it gave to it and make a tape teacher out of it.

85
00:12:44.700 --> 00:13:02.400
Anniina Kinnunen: And after that, if you look at the now the whole file, then you would see we're creating some variables for holding the values we read from the tree some new branches for the corrected values and output file and a few variables for running the corrections were are also added there.

86
00:13:05.580 --> 00:13:11.370
But the next important thing in that function is where we actually loop over the events.

87
00:13:12.450 --> 00:13:16.680
So you can see here the beginning of the loop and we're going to.

88
00:13:17.730 --> 00:13:26.790
Anniina Kinnunen: need the invariant mass of a neon pair later so that's why we're selecting events that have only exactly to me on that tab opposite George.

89
00:13:27.780 --> 00:13:39.240
Anniina Kinnunen: If you want to use this code to correct all of your Amiens, then you should simply change this number of meals, to be greater than zero and then just ignore the rest, if you want to use this code later as well.

90
00:13:43.290 --> 00:13:51.600
Anniina Kinnunen: Then next we're going to make another loop inside the previous loop where we loop over all of neurons in that suck that event.

91
00:13:52.980 --> 00:13:53.490
and

92
00:13:55.230 --> 00:14:09.210
Anniina Kinnunen: Here we create at Warren spector so the official code for adding the corrections needs at Warren spector as a parameter, so we take the me on PT is a fly and mass and create at lunch vector.

93
00:14:10.290 --> 00:14:25.920
Anniina Kinnunen: And then, this few Rosa Rosa code is where we actually apply the corrections, as I said, the corrections are different for data and MC and that's why we have the two separate functions and we use the boolean variable.

94
00:14:27.090 --> 00:14:37.500
Anniina Kinnunen: We gave earlier to know whether we have data or MC and after we run this part the corrected variables are actually.

95
00:14:38.580 --> 00:14:44.940
stored in the same to inspector, and here we extract them and save them to the.

96
00:14:45.990 --> 00:14:55.200
New branches, we created earlier so, for example, if you want to use the corrected PG later, you can find it in the neon PG core.

97
00:14:57.570 --> 00:15:03.090
And then we want to correct compute the correct that environment mass as well.

98
00:15:05.100 --> 00:15:09.630
And then we just fill the tree with the new values and save it.

99
00:15:11.820 --> 00:15:20.610
Okay, now that you may have gotten an idea of how the corrections are applied, we can actually run the code, so you can just.

100
00:15:22.110 --> 00:15:23.670
take these storylines.

101
00:15:25.200 --> 00:15:25.650
and

102
00:15:26.850 --> 00:15:30.150
run them in the terminal as well.

103
00:15:31.650 --> 00:15:33.030
Anniina Kinnunen: You are going to get some.

104
00:15:34.860 --> 00:15:49.770
Warnings after running this main function but don't worry about them, as long as you get this looping over events in the end, you know that the code is working just fine and it will take a few minutes, it took me about four minutes when I run it last time.

105
00:15:50.940 --> 00:16:06.510
Anniina Kinnunen: And after that you can go to the test directory and check that you have this corrected file with the corrected values, and you can you give us a T browser or something else do if you want to look around what's actually inside.

106
00:16:07.680 --> 00:16:12.090
Anniina Kinnunen: And let me know if you have any problems with this.

107
00:16:19.650 --> 00:16:23.490
Julie Hogan: awesome thanks to Nina any questions on this on this section.

108
00:16:32.190 --> 00:16:34.050
Julie Hogan: So this is actually one of the rare.

109
00:16:36.300 --> 00:16:48.420
Julie Hogan: My maybe I shouldn't say rare it's it's actually one of the unique types of corrections that that might actually apply to data because it's it's genuinely correcting the the momentum scale to a known physics.

110
00:16:49.320 --> 00:17:00.720
Julie Hogan: correct answer, and so this is a common common tool for, especially for precision analyses that need that need very accurate you on momentum skills.

111
00:17:03.330 --> 00:17:07.560
Okay, I got please explain last three lines.

112
00:17:10.410 --> 00:17:11.010
um.

113
00:17:18.420 --> 00:17:25.380
Anniina Kinnunen: i'm not sure what do you mean the last ones, I am here or last one, so I have actually oh gosh okay.

114
00:17:26.670 --> 00:17:32.220
Anniina Kinnunen: This is just for compiling this analysis that see that we have and.

115
00:17:34.200 --> 00:17:48.720
Anniina Kinnunen: This is just for like these two are just for running the main function, so when we run these it runs the main function, we have here in beginning, so it actually just gives these to the apply corrections.

116
00:17:50.160 --> 00:17:51.690
Anniina Kinnunen: And then, all of this.

117
00:17:52.800 --> 00:17:53.280
happens.

118
00:17:54.990 --> 00:18:07.200
Julie Hogan: So what you will be able to see if you open these some of these code files, is that we actually have a class a c++ class called analysis and so Nina is creating a an object of that class called PDF.

119
00:18:08.430 --> 00:18:15.750
Julie Hogan: And then is running the main function, which would be one of the class functions in the class named analysis.

120
00:18:19.650 --> 00:18:36.060
Julie Hogan: Okay, so everyone's probably waiting for those files to show up, but if it's okay we're going to we're going to transition back i'll share my screen and we'll talk about we'll talk about similar corrections for bottom corks and for jets generically.

121
00:18:41.520 --> 00:18:45.510
Julie Hogan: So, and you know if you'll stop your share Oh yes, thanks.

122
00:18:49.860 --> 00:18:50.160
Julie Hogan: Okay.

123
00:18:50.220 --> 00:19:03.360
Julie Hogan: um there was a question, if there are similar corrections for other objects, there are electrons have have an quite similar process we're going to talk about the most commonly used.

124
00:19:04.410 --> 00:19:20.760
Julie Hogan: corrections which are for for flavor tagging of jets and for jets generically in their energy scale the others things like this new on correction electron corrections come into play, most often when people are doing.

125
00:19:21.990 --> 00:19:27.870
Julie Hogan: What we would call a precision measurement so, for instance I mostly do new physics searches.

126
00:19:28.590 --> 00:19:41.700
Julie Hogan: And for me the momentum scale of nuance out of the box is fine it's it's it's I don't need these tiny, tiny shifts to do new physics searches, but if you're into percent precision measurements, they would be useful for you.

127
00:19:43.140 --> 00:19:46.680
Julie Hogan: Okay, so the next the next page.

128
00:19:48.090 --> 00:19:50.010
Julie Hogan: is about heavy flavor tagging.

129
00:19:51.450 --> 00:19:59.700
Julie Hogan: Which is how we take jets and we identify where they came from so jets are a big conglomerate.

130
00:20:00.990 --> 00:20:09.900
Julie Hogan: of particles and they can come from any Cork, they can come from any glue on they could also, of course, come from heavier particles and so.

131
00:20:11.040 --> 00:20:22.500
Julie Hogan: we're not going to get into really heavy things like jets from top works or deaths from higgs bosons, although we can certainly point you to references for that, but we'll focus on on the beef flavor for now.

132
00:20:24.690 --> 00:20:29.310
Julie Hogan: And so, in cms we have a suite of be tagging algorithms.

133
00:20:30.750 --> 00:20:32.280
Julie Hogan: which are related to.

134
00:20:33.510 --> 00:20:40.920
Julie Hogan: To there are several different techniques, but what we're they're all trying to do is they're all trying to take the particles that exists in this Cone.

135
00:20:41.370 --> 00:20:54.840
Julie Hogan: That we have called one jet and they're trying to dig into it and analyze the features of the particles in that jet Cone to decide whether it's likely than any of that that jack came from a bottom court.

136
00:20:55.920 --> 00:21:13.140
Julie Hogan: And so features like obvious displace to decay vertices where we have you know some some let's say some we call it a secondary vertex appearing at some displaced location might point to.

137
00:21:14.370 --> 00:21:15.960
Julie Hogan: A decay of of the hadron.

138
00:21:17.100 --> 00:21:25.080
Julie Hogan: You can look even at things like the number of tracks, the number of particles that are in the JET to do some.

139
00:21:25.680 --> 00:21:40.920
Julie Hogan: Discrimination, you can look at whether any low momentum you ons or electrons appear in this jet to do some discrimination, there are all all sorts of different techniques that might point towards bottom corks.

140
00:21:42.150 --> 00:21:44.700
Julie Hogan: And the commonly used.

141
00:21:46.800 --> 00:21:59.130
Julie Hogan: algorithm this one persistent was was developed here during run one of cms and persisted well into run two is called the combined secondary vertex algorithm.

142
00:21:59.880 --> 00:22:12.540
Julie Hogan: And what this does is it tries to combine most of these decision making tools into a multivariate analysis that produces a single real value real float.

143
00:22:13.920 --> 00:22:21.000
Julie Hogan: value between zero and one to tell you whether this jet was or was not a lot of work.

144
00:22:23.550 --> 00:22:30.360
Julie Hogan: And so we have access to this information from our jets so let me come back over here.

145
00:22:32.010 --> 00:22:35.190
Julie Hogan: We have access to this information from our jets.

146
00:22:36.480 --> 00:22:40.770
Julie Hogan: In one of two different places, so we can look at.

147
00:22:42.000 --> 00:22:55.560
Julie Hogan: Either the JET analyzer or the Pacha analyzer i'm going to point you all to last year's lesson if you're interested in working with working with the chat analyzer and i'm going to focus whoops.

148
00:22:56.580 --> 00:22:58.170
Julie Hogan: i'm going to focus on.

149
00:22:59.310 --> 00:23:00.810
Julie Hogan: The Patrick analyzer.

150
00:23:08.760 --> 00:23:10.170
Julie Hogan: Am I in the right folder.

151
00:23:11.910 --> 00:23:14.940
Julie Hogan: You know, focus on the Patrick analyzer.

152
00:23:16.830 --> 00:23:31.710
Julie Hogan: Because one of the things that that creating these physics analysis toolkit jets does for you, is it links in all of the information we have stored in the route files about be tagging.

153
00:23:33.000 --> 00:23:51.210
Julie Hogan: And so I can break down all of the all of what I need to do to access the tagging information to this one line where I can ask my jet object via the be discriminate or Member function for this particular string.

154
00:23:52.290 --> 00:24:04.710
Julie Hogan: Which which I could look up with an edm dump event content, and so I can ask for the results of this string and i'm going to get a numerical value that I can store in the tree.

155
00:24:06.180 --> 00:24:07.260
Julie Hogan: Which is super handy.

156
00:24:09.150 --> 00:24:16.410
Julie Hogan: And so, when you run when you run poets We saw this yesterday i'll show it to you again.

157
00:24:17.370 --> 00:24:26.580
Julie Hogan: When we run poets, with the physics analysis toolkit turned on which is the second argument you see a printout here at the beginning.

158
00:24:27.510 --> 00:24:43.530
Julie Hogan: which tells you which be discriminate or strings you'll have access to in your in your trees so that's the best way to get the list very quickly is just to run it and you can get all the available strings from different algorithms.

159
00:24:45.660 --> 00:24:58.110
Julie Hogan: And so, then you can look if you if you want to do this, you can look I think most of you will probably have a root file by now, you could look at the distribution of these of these discriminate or values.

160
00:25:00.840 --> 00:25:07.680
Julie Hogan: And so, just like with identification with isolation, we have to then choose what we're going to do we have to choose where.

161
00:25:08.520 --> 00:25:25.290
Julie Hogan: Know numerically where are we going to place cuts on this distribution it goes from zero to one zero means not a very, very big court like and one means very the Court like but there's a huge range in between, and we have to make decisions somewhere.

162
00:25:26.550 --> 00:25:29.250
Julie Hogan: So we have set up working points.

163
00:25:30.360 --> 00:25:32.040
Julie Hogan: One of them is moose.

164
00:25:33.240 --> 00:25:42.030
Julie Hogan: And it brings along about a 10% let's say about 10% of the light court jets would get thrown into the B corp category.

165
00:25:44.850 --> 00:25:49.050
Julie Hogan: Many cms analysts work with the medium working point.

166
00:25:50.190 --> 00:26:03.390
Julie Hogan: Because that one has been scooted a little a little higher on the purity side a little lower on the efficiency side but higher on the purity side, so that only about 1% of light court jets would be miss identified as a bottom corner.

167
00:26:05.340 --> 00:26:19.530
Julie Hogan: And there is a tight working point that gets you really pure sample so they're only a 10th of a percent of the light court jets would get thrown in with bees and so, so that it would be very pure be sample if you select the tight working point.

168
00:26:24.090 --> 00:26:28.320
Julie Hogan: And, just as, as you can read in the in the intro lesson.

169
00:26:30.360 --> 00:26:43.890
Julie Hogan: All of those working points carry some difference in in in the efficiency between Monte Carlo and data so here's actually a plot, this is this is looking at the JET momentum.

170
00:26:44.910 --> 00:27:01.470
Julie Hogan: Get from lower to hire and we're looking at the efficiency for correctly identifying bottom court jets which we which we've identified either let's say with the Monte Carlo truth or with some physics knowledge inspired process for data.

171
00:27:03.660 --> 00:27:13.530
Julie Hogan: So we're looking at the efficiency and we can see that they're they're disparate between the two, and so we would we would calculate a scale factor.

172
00:27:13.920 --> 00:27:26.970
Julie Hogan: which you can see, based on the red lines here, it tends to be less than one the Monte Carlo over predicts the number of bottom Cork, the number of bottom court jets, and so we would want to reduce that rate in our analysis.

173
00:27:28.800 --> 00:27:35.790
Julie Hogan: But this efficiency is quite high 70% between 60 and 70% correct correct efficiency.

174
00:27:37.290 --> 00:27:45.360
Julie Hogan: The miss it misidentification rate for the same one we want these numbers to be much, much, much, much, much, much smaller which they are.

175
00:27:46.950 --> 00:27:51.570
Julie Hogan: So this is about a 1% this identification rate, up to a 5% rate.

176
00:27:52.950 --> 00:28:06.360
Julie Hogan: And you can see that as the momentum of the JET increases we actually we actually start to consider that light ports are be or be like, as they get to be very high momentum, more often than when they're low moment.

177
00:28:09.300 --> 00:28:20.940
Julie Hogan: And this scale factor goes the other way, the scale factor is larger than one, so we have it's more likely that excuse me that money crawl under predicts the number of light corks because it's over predicting the number equals.

178
00:28:23.790 --> 00:28:26.130
Julie Hogan: And so we need to excuse me.

179
00:28:27.600 --> 00:28:33.660
Julie Hogan: We need to apply these scale factors in order to correct the efficiency in the Monte Carlo.

180
00:28:35.880 --> 00:28:38.550
Julie Hogan: And this has all been set up for you in poet.

181
00:28:39.720 --> 00:28:41.970
Julie Hogan: So we have some references here.

182
00:28:43.080 --> 00:28:59.820
Julie Hogan: To the actual cms web pages that describe the correct procedure for accessing the scale factors which have been computed it's not on you, the analysts to compute them for yourself, they have been computed as long as you're using this algorithm.

183
00:29:01.200 --> 00:29:05.130
Julie Hogan: And they actually exist in this particular data file, which is just.

184
00:29:06.150 --> 00:29:16.590
Julie Hogan: In a happy accident, the scale factors for the csv algorithm are stored in a comma separated values file, which is called if you see at the bottom of my screen csv dot csv.

185
00:29:17.790 --> 00:29:22.440
Julie Hogan: And so you can actually download the data file for all of these scale factors.

186
00:29:23.760 --> 00:29:36.660
Julie Hogan: And there's a reference that will show you the many approved application methods, there are about six methods that are that are approved and cms to apply the scale factors in your analysis.

187
00:29:37.590 --> 00:29:46.590
Julie Hogan: And how it we have implemented one of those methods, not all of them, but one and we have already built functions for you.

188
00:29:47.340 --> 00:30:02.010
Julie Hogan: To provide all of the scale factors, based on a jet momentum and a jet security, so I can show you what that looks like you saw yesterday, all of these helper functions at the top of my file.

189
00:30:03.300 --> 00:30:08.910
Julie Hogan: With helper functions in here, where you can store your efficiencies which you should calculate for yourself.

190
00:30:09.870 --> 00:30:19.260
Julie Hogan: And you can access the scale factors, so these scale factor functions are fixed you do not need to edit them unless you would like to use a different working points.

191
00:30:20.010 --> 00:30:30.060
Julie Hogan: For the CSP algorithm we've done the media working points, but this gives you an example of how they should look, if you want to adjust to different working point.

192
00:30:32.790 --> 00:30:39.570
Julie Hogan: And then we've also implemented, one of the application methods so there's a whole big chunk of code here.

193
00:30:41.250 --> 00:30:54.180
Julie Hogan: Below some of the variables storage of a whole chunk of code that's only run on simulation so again, this is a type of correction, where the data is correct and we're we're fixing the simulation.

194
00:30:56.610 --> 00:31:00.330
Julie Hogan: And we're using the flavor of the JET.

195
00:31:02.250 --> 00:31:10.560
Julie Hogan: The efficiency for tagging either bottom corks or charm courts or light corks for this jet.

196
00:31:11.670 --> 00:31:28.230
Julie Hogan: we're using the scale factor value which we look up for each for each jet and we use that to calculate an event rate, and so you read about event weights in the in the intro lesson use that to calculate an event weight.

197
00:31:29.550 --> 00:31:34.710
Julie Hogan: And so, at the very end in the route file, let me show you a number of files.

198
00:31:36.870 --> 00:31:39.480
Julie Hogan: We have we have.

199
00:31:40.500 --> 00:31:41.850
Julie Hogan: Some branches.

200
00:31:42.900 --> 00:31:44.820
Julie Hogan: In our jet folder.

201
00:31:50.340 --> 00:31:51.630
Julie Hogan: called be tagged wait.

202
00:31:52.710 --> 00:32:00.990
Julie Hogan: which gives you an event wait for the Monte Carlo and you can see that most of them are in that most of them are in that let's say 5% range.

203
00:32:02.730 --> 00:32:06.300
Julie Hogan: And we can shift the up or down, based on the uncertainty.

204
00:32:08.730 --> 00:32:16.380
Julie Hogan: So this is a very common correction that basically all analysts to are interested in using jet flavor would apply and cms.

205
00:32:20.040 --> 00:32:22.950
Julie Hogan: Any questions on this before we hit the last lesson.

206
00:32:39.120 --> 00:32:39.660
Julie Hogan: Okay, great.

207
00:32:44.640 --> 00:32:45.270
Julie Hogan: So.

208
00:32:46.980 --> 00:32:54.240
Julie Hogan: The last the last lesson here is about the JET energy scale and the JET energy resolution.

209
00:32:55.350 --> 00:33:00.150
Julie Hogan: which has to be corrected in order to use jets and cms.

210
00:33:01.980 --> 00:33:15.870
Julie Hogan: And so we're going to go through the concepts of this and then the challenge the offline challenge today will will have you do some running of poets and dig into the effects of these corrections by looking at the data.

211
00:33:17.520 --> 00:33:18.960
Julie Hogan: So we're going to go over the concept.

212
00:33:21.360 --> 00:33:37.080
Julie Hogan: That 30% unsurprisingly, the cms detector does not measure jet energies perfectly so first of all, in this case we actually know for sure our detector does not measure jet energies perfectly.

213
00:33:38.160 --> 00:33:51.570
Julie Hogan: And plus also data and simulation do not agree, so we have two problems we have the problem that the detector does not accurately measure the JET energy and we have the problem that Monte Carlo and data do not agree.

214
00:33:52.620 --> 00:33:58.230
Julie Hogan: need a way to correct both of those problems, and that is called the JET energy corrections.

215
00:33:59.280 --> 00:34:10.500
Julie Hogan: I want to point you out to references i'm both that for 70 db data in this case the year is the year of this reference paper as opposed to the year of the data.

216
00:34:11.580 --> 00:34:22.980
Julie Hogan: So it's so a 17 EB reference Paper was published in 2011 and an ATV reference Paper was published in 2017 so you can you can actually get an idea of how much work was needed.

217
00:34:23.580 --> 00:34:32.940
Julie Hogan: To go into all of these corrections by the fact that the 2012 data was was being published in 2017 about its corrections.

218
00:34:33.720 --> 00:34:34.440
Dillon Fitzgerald: Have a quick question.

219
00:34:35.100 --> 00:34:50.310
Dillon Fitzgerald: Yes, so for the JET energy corrections is this essentially just like a normalization correction and it's been migration handled separately via unfolding, or is it sort of all wrapped in together.

220
00:34:51.000 --> 00:34:52.740
Julie Hogan: Yes, yes.

221
00:34:53.850 --> 00:34:56.190
Julie Hogan: yeah so we're gonna we'll go through some of these steps.

222
00:34:56.580 --> 00:34:57.390
Dillon Fitzgerald: Okay, thanks.

223
00:35:00.090 --> 00:35:00.420
Julie Hogan: Okay.

224
00:35:01.830 --> 00:35:08.160
Julie Hogan: So this graphic, this is a, this is a helpful graphic to show you what all is going on.

225
00:35:10.110 --> 00:35:16.620
Julie Hogan: We start over here on the left with our reconstructed jets, which is the out of the box what we get from the detector.

226
00:35:17.910 --> 00:35:23.040
Julie Hogan: And we have a suite of corrections that we actually apply on data.

227
00:35:24.360 --> 00:35:35.160
Julie Hogan: And we have a suite of corrections, some of which is a little bit smaller, you can see that's what the blanks here me the suite of corrections that are applied on simulation.

228
00:35:37.320 --> 00:35:52.050
Julie Hogan: And what you're what you're seeing in the in the in bold is the title of the correction and what you're seeing not in bold is is how this correction is derived and there's there's much more information on all about in those reference papers.

229
00:35:53.760 --> 00:35:57.540
Julie Hogan: The biggest first correction here is the pilot correction.

230
00:35:59.520 --> 00:36:06.360
Julie Hogan: pileup again is all of the other collisions that are simultaneous to the collision we which we wish to study.

231
00:36:07.380 --> 00:36:24.030
Julie Hogan: All of those other collisions put energy into the calorie limiters and much of that energy, we cannot get rid of base with some deterministic algorithm it's just backgrounds let's say background, energy and or detector.

232
00:36:25.770 --> 00:36:32.910
Julie Hogan: And so i'm going to use these diagrams to show to show how these corrections go forward, so the first plot here.

233
00:36:34.200 --> 00:36:44.400
Julie Hogan: This is a plot from simulation and on the X axis is showing you the generated momentum of the jets, so this is what we would consider the correct answer let's say this is the truth.

234
00:36:45.210 --> 00:36:57.810
Julie Hogan: This is really the momentum and the y axis is showing you the response, the response is basically the ratio of how it was reconstructed over the truth.

235
00:36:58.650 --> 00:37:12.090
Julie Hogan: So what you can see, is if there is no pileup if we simulate a single collision we actually see that the jets that we reconstruct tend to have not enough momentum.

236
00:37:13.200 --> 00:37:18.600
Julie Hogan: With respect to what was generated, especially in the low momentum regions.

237
00:37:20.130 --> 00:37:28.380
Julie Hogan: As we increase the amount of pile up in our simulation we quickly go to the other side of one, so we very quickly end up with way too much.

238
00:37:29.460 --> 00:37:34.440
Julie Hogan: Energy in our jets, especially in the low momentum ranges because this is fractional.

239
00:37:36.120 --> 00:37:44.190
Julie Hogan: You know so pilot pilot has a much smaller effect on a one TV Jeff than it does on a 30 G jet big difference.

240
00:37:45.240 --> 00:38:01.740
Julie Hogan: So this is the first correction that we have to make is for pileup, and so we have to measure what's going on in terms of how much pilot there actually is, and then we have to figure out this response curve and correct for it.

241
00:38:03.930 --> 00:38:22.200
Julie Hogan: And so that gives us that process of energy correction gives us the second graph here, as you can see what's happened all of the colors black red green and blue have collapsed themselves down to what we basically what we were seeing in the open circles in the Left graph.

242
00:38:23.250 --> 00:38:30.000
Julie Hogan: So we can collapse down to how we how we look for for pilot correct, excuse me for zero pilot.

243
00:38:32.160 --> 00:38:43.680
Julie Hogan: And then we go through this the rest of these corrections which is called the the red box here, the response correction and these residual corrections which are small.

244
00:38:44.340 --> 00:38:51.930
Julie Hogan: Often Ada dependence so that's a little bit of what you're asking about dylan been been migration shifting of aware of the JET lies.

245
00:38:54.090 --> 00:39:02.910
Julie Hogan: We are actually turning knobs to change the momentum and the direction of the JET of it for when we apply these corrections.

246
00:39:03.510 --> 00:39:19.200
Julie Hogan: And so, at the end when we end up all the way over here in our green box of calibrated jets we have got a flattish response curve, that is a flat and b flat at one, which is what we really want.

247
00:39:21.180 --> 00:39:34.920
Julie Hogan: So this entire process of of stepwise sequential corrections gives us jets whose reconstructed momentum matches their generated momentum to to a good degree.

248
00:39:37.290 --> 00:39:43.530
Julie Hogan: We also have to think about the the resolution, so if you take this.

249
00:39:44.760 --> 00:39:57.300
Julie Hogan: If you take these response curves and you actually you know grab one of these bins and you and you plot what the response looks like for all the jets it's got some goshen is shaped.

250
00:39:58.140 --> 00:40:14.940
Julie Hogan: And the width of that shape can vary between data and Monte Carlo so we have a full extra correction that takes the width of that curve and swishes or spreads it so that Monte Carlo matches what we see in data.

251
00:40:17.130 --> 00:40:27.840
Julie Hogan: And I think we've arrived already almost set the second break, so what i'm going to do is i'm just going to leave it i'm going to leave it there and and note point out to you that, actually, we have a.

252
00:40:28.380 --> 00:40:42.480
Julie Hogan: We have a significant amount of work put into making sure that you guys have a reliable way of applying these corrections, so we have tried to do our homework both to to apply.

253
00:40:42.840 --> 00:40:53.100
Julie Hogan: The corrections for you in poet, as an example of how they can be done and to provide you with the documentation on how we did that, so you have.

254
00:40:53.610 --> 00:41:08.520
Julie Hogan: In the JET analyzers in the JET analyzers of poet, a lot of code which i'll let you read through on your own and then please ask us questions about on matter most if you have any concern or question about the mathematical procedures.

255
00:41:10.560 --> 00:41:22.170
Julie Hogan: All of the JET energy corrections are applied either when we make the jets or by looking them up from text files and the JET energy resolution is applied.

256
00:41:22.800 --> 00:41:38.130
Julie Hogan: as well to all of the jets So if you work with poet as an analysis tool we have tried to set that up correctly, for you and the concept of it is covered here in this lesson page and also in the reference papers.

257
00:41:39.360 --> 00:41:54.180
Julie Hogan: We will come back to this in the challenge, particularly to look at the uncertainties, so when we get to the offline challenge, I invite you to come back to this page and read, especially about the uncertainties before you launch into those exercises.

258
00:41:55.530 --> 00:42:01.050
Julie Hogan: So, at that point and we will, I think we're in take our screenshot images and then have a break.

259
00:42:08.880 --> 00:42:13.950
Matthew Bellis: So I need to stop by sure i'll go ahead and I will go ahead and stop the recording.