1 00:00:04,710 --> 00:00:10,350 Thank you. Yes, I will talk about Higgs measurements in a decay, um, to second 2 00:00:10,350 --> 00:00:18,180 generation fermions at the LHC. Well, decays to the second generations are, of course, 3 00:00:18,180 --> 00:00:24,570 more difficult to see at the LHC, because of very low branching ratio and 4 00:00:24,570 --> 00:00:29,250 the couplings of the Higgs to the second generation. This is illustrated on the 5 00:00:29,280 --> 00:00:34,440 figure on the right, where, um, particle mass and therefore coupling, which is 6 00:00:34,440 --> 00:00:40,410 proportional to the mass, is shown. In this talk, I will present direct searches for 7 00:00:40,410 --> 00:00:48,900 Higgs to cc bar and Higgs to mu mu decays. So, let's start with Higgs to cc bar. 8 00:00:48,900 --> 00:00:57,840 And let's first look at the analysis by CMS. This analysis is done at 13 TeV with partial run two 9 00:00:57,840 --> 00:01:06,000 data set. The search for Higgs to cc bar is done in the production of Higgs 10 00:01:06,060 --> 00:01:11,550 associated with a vector boson, as shown on the diagram here. The reason for that of course is... 11 00:01:11,550 --> 00:01:20,880 when Higgs decays hadronically, the dominant QCD background is too large, so we 12 00:01:20,880 --> 00:01:26,910 have to have leptons from vector boson decays to tag an event, to have a cleaner 13 00:01:26,910 --> 00:01:33,300 selection. So, three channels by CMS analysis are pursued. First is when 14 00:01:33,300 --> 00:01:37,350 Z boson decays to two leptons; Z bozon decays to two neutrinos and a W boson 15 00:01:37,350 --> 00:01:44,100 dacaying to lepton and neutrino. And here by lepton I mean a muon or an electron, 16 00:01:44,100 --> 00:01:51,450 so, two channels in each category. Higgs dacaying to cc bar can be ... 17 00:01:51,630 --> 00:01:59,190 and by CMS two analysis are pursued - it can be reconstructed as two small cone jets using 18 00:01:59,460 --> 00:02:07,770 anti-kt jet algorithm with parameter zero ... with the cone parameter zero point four and then 19 00:02:07,770 --> 00:02:13,260 a c-tagger is developed to tag both of these jets as c-jets or light-flavor jets, 20 00:02:13,290 --> 00:02:19,470 or b-flavor jets. And the second approach is to look for a large cone... cone jet. 21 00:02:19,470 --> 00:02:27,960 In this case it's anti-kt with cone parameter one point five, and in this case a different tagger has to 22 00:02:27,960 --> 00:02:39,000 be used to tag a jet as two c-like object inside. It is more used to select boosted jets 23 00:02:39,720 --> 00:02:45,630 boosted Higgs. The dominant background in all those channels is 24 00:02:45,630 --> 00:02:52,650 Z plus jets and W plus jets, with some small contribution from tt bar and QCD in the 25 00:02:52,650 --> 00:02:59,970 zero lepton channel. So first, let me talk about the first analysis, when two jets are 26 00:03:00,000 --> 00:03:05,850 reconstructed, in this case a deep neural network tagger is used, which is depicted 27 00:03:05,850 --> 00:03:11,610 here. It's rather straightforward deep neural network with four layers, each has 28 00:03:11,640 --> 00:03:16,170 one hundred nodes using as input jet kinematics, tracks and secondary vertex inputs, 29 00:03:16,170 --> 00:03:21,510 total of 66 features. And it outputs probabilities for a given jet 30 00:03:21,510 --> 00:03:29,250 being B flavor, C flavor and light flavor. Of course, B-jets are background, 31 00:03:29,250 --> 00:03:34,200 when we look for C-jets. So we have to look for two discriminators. One discriminator is 32 00:03:34,200 --> 00:03:42,150 to separate C-jets versus light jets. And for that the variable C versus L is 33 00:03:42,150 --> 00:03:47,760 constructed as a probability of jet being a C divided by probability of C plus light, 34 00:03:48,180 --> 00:03:53,940 and C versus B, which is similar - probability of C divided by probability of C plus B. 35 00:03:53,940 --> 00:04:01,800 Figure on the right shows the efficiency of this tagger in three dimensions. 36 00:04:01,800 --> 00:04:07,140 On Z-axis is a C-jet eficiency and you can see equidistant 37 00:04:07,140 --> 00:04:14,100 line of efficiencies. So for example, CMS working point is around thirty percent. And for each 38 00:04:14,100 --> 00:04:19,800 point you can see what is B tagging jet efficiency or in other words mistag rate for B-jets, 39 00:04:19,830 --> 00:04:26,610 and light jets on the Y-axis. And in CMS there is a working point used for 40 00:04:26,610 --> 00:04:33,270 pre-selection of jets is about twenty eight percent C-jet efficiency with fifteen percent B-jet mistag rate 41 00:04:33,270 --> 00:04:40,530 and four percent light jet mistag rate. Further selection involves of course selection of objects, 42 00:04:40,530 --> 00:04:45,660 such as "two lepton", "no lepton" and "one lepton" in different channels. 43 00:04:45,660 --> 00:04:50,440 In two lepton channel, you want to have a Z peak, so we select on dilepton mass 44 00:04:50,880 --> 00:04:59,190 and we cut on pT of two leptons of 50 GeV, 150 GeV, so we 45 00:04:59,190 --> 00:05:05,670 have two categories in this channel. And then by missing ... cutting on missing energy 46 00:05:05,670 --> 00:05:12,480 in zero lepton channel and pT of the W boson candidate in one lepton channel. 47 00:05:12,480 --> 00:05:20,550 And then we have a pre-selection of the jet - one of the jet from the Higgs, um, 48 00:05:20,550 --> 00:05:27,180 on C versus L and C versus B taggers, as I described previously, and then later on these C versus L 49 00:05:27,180 --> 00:05:33,600 and C versus B scores are used in the BDT. And finally, there is also a selection on 50 00:05:34,290 --> 00:05:41,250 invariant mass of the two jets in this range. The backgrounds in this analysis are 51 00:05:41,700 --> 00:05:47,580 constrained from control region, taking the shapes from monte carlo. okay, so, this 52 00:05:47,580 --> 00:05:53,190 figure here shows how the control regions are defined, which is done in a plane of 53 00:05:53,190 --> 00:05:57,840 C versus L and C versus B discriminator on one of the jets. 54 00:05:57,840 --> 00:06:04,680 Signal region is shown here, as ... as I mentioned already, and then control regions are defined 55 00:06:04,710 --> 00:06:11,070 inverting one of the cuts. So, V plus light ... light jet flavor is constructed 56 00:06:11,070 --> 00:06:17,670 by inverting C versus L cut, V plus heavy flavor jets are constructed inverting 57 00:06:17,670 --> 00:06:23,250 C versus B cut and then also tt bar control region, in addition, invert 58 00:06:25,200 --> 00:06:27,750 vector boson invariant mass in two lepton channel. 59 00:06:29,190 --> 00:06:34,230 We have another control region, which is V plus two C-tagged jets, which is the most 60 00:06:34,230 --> 00:06:39,540 difficult one, because it basically is signal region but inverting Higgs boson 61 00:06:39,540 --> 00:06:47,250 invariant ... invariant mass, Higgs candidate invariant mass. The first ... sorry ... the light flavor 62 00:06:47,250 --> 00:06:53,040 control region and tt bar control region are very pure in light flavor and tt bar respectively. 63 00:06:53,070 --> 00:06:58,890 Heavy flavor and C-tag flavor are not so pure. Nevertheless, they are very useful 64 00:06:58,890 --> 00:07:04,860 to construct these corresponding backgrounds. Finally, BDT is trained to separate 65 00:07:04,860 --> 00:07:10,500 signal and backgrounds, which use sixteen to twenty input features, including C versus L 66 00:07:10,530 --> 00:07:20,010 and C versus B scores of all jets ... of both jets, and invariant mass of the two jets, um, and, 67 00:07:20,010 --> 00:07:24,270 on the figures here you see the distributions of the BDT score after the final fit, when 68 00:07:24,300 --> 00:07:29,160 background is also included in the fit. So they are normalized. And on the left you 69 00:07:29,160 --> 00:07:33,900 can see in the two lepton channel dominant background is green, which is Z plus 70 00:07:34,770 --> 00:07:41,430 jets, in different shades are different kind of jets - either B-jets, C-jets etcetera. 71 00:07:41,430 --> 00:07:46,260 on ... in the middle is one lepen channel. So the dominant background is orange, which is 72 00:07:46,260 --> 00:07:51,630 W plus jets, with some contribution from tt bar in blue - it's hard to see but it's in 73 00:07:51,630 --> 00:07:56,370 a log scale - it does contribute here. And on the right is the zero lepton channel, 74 00:07:56,550 --> 00:08:01,110 where dominant background is Z plus jets again, um, noticible here, there is 75 00:08:01,110 --> 00:08:06,330 some contribution from QCD, although it's in a low BDT score, so, actually 76 00:08:06,360 --> 00:08:14,850 khm, does not affect much the ... the sensitivity of the result. And the results for this 77 00:08:14,880 --> 00:08:22,590 analysis is, of course, a limit of thirt ... of seventy five observed - seventy five times the Standard Model 78 00:08:22,590 --> 00:08:28,320 with expected of thirty eight. So, there is a slight excess, although with ... with uncertainties of course, 79 00:08:28,320 --> 00:08:33,840 so, we have a limit from this channel. Now, I mentioned in the first slide, 80 00:08:33,840 --> 00:08:39,000 we have a second kind of analysis, when a jet ... Higgs is reconstructed using a 81 00:08:39,000 --> 00:08:47,970 single large cone jet. For this a dedicated ... well, first of all, jet sub ... jet algorithms 82 00:08:48,030 --> 00:08:53,040 is developed, but in addition jet B-tagging and C-tagging algorithm, using 83 00:08:53,370 --> 00:08:58,380 rather complicated neural network is developed, so. It's described here - it uses 84 00:08:58,650 --> 00:09:03,000 a lot more variables, hundred Particle Flow candidates, meaning all their four vectors, 85 00:09:03,000 --> 00:09:11,880 using some, um, high level features, secondary vertices etcetera. It has more 86 00:09:11,880 --> 00:09:17,550 complicated structure, using convolution neural networks. And more details you can 87 00:09:17,550 --> 00:09:26,400 see in the talk by Lucas in a dedicated session. But the results for this analysis 88 00:09:26,430 --> 00:09:34,830 is slightly less sensitive - is a limit of thirty nine expected limit of forty nine times the Standard Model 89 00:09:35,220 --> 00:09:38,340 Nevertheless, we can use this analysis to combine with, um, um 90 00:09:39,809 --> 00:09:42,419 resolved jet analysis by, khm 91 00:09:43,890 --> 00:09:48,780 splitting events for more ... to more boosted and less boosted, using the pT of 92 00:09:48,780 --> 00:09:54,660 the vector boson - so there is a cut of 300 GeV, less and above - and combining that we 93 00:09:54,660 --> 00:10:00,540 can improve slightly the final limit ... and now it is thirty seven times the Standard Model 94 00:10:00,540 --> 00:10:09,240 expected and seventy times the Standard Model observed. Now, this was analysis by CMS, here is analysis by ATLAS. 95 00:10:09,240 --> 00:10:15,810 This is also done on the same data set - thirty six inverse femtobarns, however, only a Z to LL 96 00:10:15,810 --> 00:10:25,290 channel is used by C.. by ATLAS. There are two BDTs developed for separation C-jets 97 00:10:25,290 --> 00:10:30,210 against light and C-jets against B-jets and on the right there is a similar picture 98 00:10:30,240 --> 00:10:34,770 depicting ... depicting efficiencies and the cross shows the working point, which 99 00:10:34,770 --> 00:10:42,540 corresponds to forty one percent C-tag efficiency with twenty five percent mistag rate for B and five percent 100 00:10:42,540 --> 00:10:51,210 mistag rate for light flavor jets, And, the results are then extracted by fitting invariant mass 101 00:10:51,210 --> 00:11:01,470 of the two jets, which is shown on the plot figure ... on the figure here. And ... yeah, the 102 00:11:01,470 --> 00:11:05,670 normalization of the dominant background, which is Z plus jet is also floating in 103 00:11:05,670 --> 00:11:11,340 this fit. So it's kind of constraining the background here as well. And the result by 104 00:11:11,340 --> 00:11:16,170 ATLAS here is a limit of one hundred ten times the Standard Model 105 00:11:17,460 --> 00:11:19,110 with expected of one hundred fifty 106 00:11:20,820 --> 00:11:23,040 Allright, this was Higgs to 107 00:11:24,419 --> 00:11:28,919 cc bar. Now let's move on to ... to Higgs to mumu. Higgs to mumu is a ... 108 00:11:28,919 --> 00:11:36,809 in some sense, more straightforward - is a fit of dimuon invariant mass, looking 109 00:11:36,809 --> 00:11:44,699 for a bump at one hundred twenty five GeV. The difficulty here is to find the right model to fit the 110 00:11:44,699 --> 00:11:51,119 background. In CMS, in this analysis, this modified Breit-Wigner used with [...] 111 00:11:51,119 --> 00:11:58,559 multiplied by Bernstein polynomials. BDT then used to categorize the events, enhancing 112 00:11:58,589 --> 00:12:04,649 the VBF, when there are two jets in the events, eta of the muons are also used to further 113 00:12:05,159 --> 00:12:15,479 separate high ... better resolution events with less ... better ... and the ... the results of this 114 00:12:15,509 --> 00:12:21,899 is an upper limit of about three times the Standard Model. And by ATLAS, there is a 115 00:12:21,899 --> 00:12:30,239 result using the full statistics of one hundred forty inverse femtobarn. And here I just show some 116 00:12:30,239 --> 00:12:36,779 improvements that has done with respect to thirty six inverse femtobarn analysis, which is using 117 00:12:36,989 --> 00:12:42,089 BDT for categorization - there are four ... four categories, depending on number of jets 118 00:12:42,089 --> 00:12:49,679 and, in each BDT - high, medium and low categories used. Here are some variables 119 00:12:49,679 --> 00:12:56,429 that are used in this BDT. Another improvement is including QED FSR recovery 120 00:12:56,429 --> 00:13:01,289 from muons, which is shown on the picture here. It does not affect many events, 121 00:13:01,289 --> 00:13:08,039 but it gives some improvement to the final results. And then, there is a background 122 00:13:08,879 --> 00:13:14,039 function is Drell-Yan lineshape convoluted with Gaussian and multiplied by 123 00:13:14,699 --> 00:13:23,069 power of ... power functions. And this is the result by CMS [edit: by ATLAS]. On the ... on these two figures 124 00:13:23,069 --> 00:13:28,259 are shown are events in all categories - on the left is without weights and on the right 125 00:13:28,349 --> 00:13:33,599 weighted in each category by signal over background. And the result is the limit on 126 00:13:33,959 --> 00:13:42,389 on mu of one point seven times ... times the Standard Model. It's dominated by ... by statistical 127 00:13:42,389 --> 00:13:51,329 uncertainty as ... as in CMS as well. In this slide, I show the summary of previous 128 00:13:51,329 --> 00:13:57,239 results for Higgs to mumu research as well. I showed the ones shown ... depicted with the 129 00:13:57,239 --> 00:14:03,449 star. Unfortunately, there is no yet results for full run two data from CMS. 130 00:14:03,449 --> 00:14:08,009 But I think this table gives you some idea of what would you expect from CMS 131 00:14:08,009 --> 00:14:14,549 comparing ATLAS and CMS results ... from previous years. And this is a summary. 132 00:14:14,549 --> 00:14:20,339 I've presented you recent results on CMS, and AT... from CMS and ATLAS on Higgs boson decay 133 00:14:20,339 --> 00:14:25,499 to second generations. There is no yet significant excess observed against the 134 00:14:25,499 --> 00:14:33,299 backgrounds. The best limit on Higgs to cc bar is from AT... from CMS and it's seventy times 135 00:14:33,299 --> 00:14:38,159 the Standard Model. There are other ways to constrain coupling of the Higgs to 136 00:14:38,159 --> 00:14:44,399 charms, which I didn't describe here. For example, most notable is from Higgs pT 137 00:14:44,399 --> 00:14:49,919 distributions, which is mentioned in ... in the plenary and also in the dedicated 138 00:14:49,919 --> 00:14:54,509 talk in the parallel session. There is also Higgs to Jay Psi Gamma dacay, constraint 139 00:14:54,509 --> 00:15:00,869 from width of the Higgs and from global fits. As for Higgs to mumu the best results so far 140 00:15:00,899 --> 00:15:06,779 is from ATLAS and the limit is one point seven times the Standard Model and we are looking 141 00:15:06,779 --> 00:15:11,489 forward to CMS result with full Run two data. Thank you. 142 00:15:15,240 --> 00:15:21,390 Thanks a lot Andrey, thank you very much for this very nice talk. So, any 143 00:15:21,390 --> 00:15:26,250 questions? Yes, please, feel free to raise your hands and we can take questions. 144 00:15:26,670 --> 00:15:27,390 Okay, then. 145 00:15:29,039 --> 00:15:30,719 Fabio, please go ahead. 146 00:15:33,299 --> 00:15:41,789 Hi, Andrey, thanks for the nice talk. This is Fabio Cerruti. So, I ... um ... well, maybe 147 00:15:41,789 --> 00:15:46,769 it's more a general comment. I mean, if I look to the performance of ... this is true 148 00:15:46,769 --> 00:15:52,979 for ATLAS and CMS ... of the c-tagger, essentially. You have a factor 149 00:15:52,979 --> 00:15:59,849 two inefficiency for B versus ... for beauty versus charm. So considering that in 150 00:15:59,849 --> 00:16:06,299 the Standard Model the two branching ratios are ... I mean, Higgs to bb is twenty times the Higgs to charm, it's 151 00:16:06,299 --> 00:16:15,389 expected. Does it really make sense to try to measure Higs to charm? Wouldn't be more 152 00:16:16,229 --> 00:16:22,199 meaningful to try to make a simultaneous measurement of the two, maybe binning in a 153 00:16:22,289 --> 00:16:31,769 kind of more charm-like, more bottom-like ... selection, because if I guess yeah, 154 00:16:31,769 --> 00:16:36,299 you fix Higgs to BB to Standard Model, but after your selection, you still have ten 155 00:16:36,299 --> 00:16:41,969 more times Higgs to BB then Higgs to CC and the mass is very similar, right? So ... 156 00:16:42,960 --> 00:16:47,340 Yes, you're correct. You're correct that we do fix Higgs to BB to Standard Model. 157 00:16:47,370 --> 00:16:52,710 That's how we deal with it now. And you're correct that it would be better to have a 158 00:16:52,710 --> 00:16:57,300 simultaneous fit for Higgs to BB and Higgs to CC bar, and there is such a plan for 159 00:16:57,300 --> 00:17:02,340 legacy analysis at CMS, at least, because I'm from CMS, I can tell you. It's of 160 00:17:02,340 --> 00:17:07,590 course, more difficult to do for many reasons. But yes, you're correct. It makes 161 00:17:07,590 --> 00:17:14,550 sense to have a simultaneous fit for Higgs to BB and Higgs to CC and have an analysis 162 00:17:14,550 --> 00:17:25,440 that aims to do such a thing. It's not trivial, but - yes. 163 00:17:25,440 --> 00:17:30,180 In addition to that, I can say that there are some improvements from CMS and probably from ATLAS in 164 00:17:30,180 --> 00:17:36,780 the taggers. So in this CMS analysis, we use so called Deep CSV tagger. There is now a 165 00:17:36,780 --> 00:17:42,810 new tagger called Deep flavor, or sometimes called Deep jet, which is a bit more 166 00:17:42,840 --> 00:17:46,500 elaborate Deep Neural Network with more features and it gives some more 167 00:17:46,500 --> 00:17:52,410 performance to separate C versus B and also C versus light, so, there will be 168 00:17:52,410 --> 00:17:53,370 some improvements there. 169 00:17:55,980 --> 00:17:59,970 Thanks a lot. And we have another question from Benedict. Please go ahead. 170 00:18:01,380 --> 00:18:06,240 Yeah, for the Higgs to CC, can you comment on what the relative contributions of 171 00:18:06,240 --> 00:18:12,780 systematic and statistical uncertainties are to ... yeah, limiting the precision of 172 00:18:12,780 --> 00:18:13,620 the analysis? 173 00:18:14,160 --> 00:18:19,830 Yeah, this is an interesting question. I have this table from CMS and the limiting 174 00:18:19,920 --> 00:18:26,250 factor here is statistical. So, this is here shown on mu - best fit mu value 175 00:18:26,250 --> 00:18:33,240 which is in this case thirty seven, u-um ... expected, I think, yes. And then statistical 176 00:18:33,270 --> 00:18:39,120 uncertainty are dominated. With ... with experimental uncertainly coming from 177 00:18:39,930 --> 00:18:45,960 mainly charm-tagging efficiencies, but also from theory. So for example, theo... signal 178 00:18:45,960 --> 00:18:54,090 theory make some large uncertainties. However, I also have ... yes, this table from 179 00:18:54,120 --> 00:18:59,700 ATLAS, which is again, breakdown of the uncertainties and here statistical 180 00:18:59,700 --> 00:19:07,350 uncertainty is ... is forty nine percent, but systematic uncertainty is dominant. And the dominant 181 00:19:07,440 --> 00:19:15,990 here is the flavor tagging uncertainty. So ... this is something that I expect ATLAS to 182 00:19:15,990 --> 00:19:21,960 improve - bring down this uncertainty on the flavor tagging, and so that it is 183 00:19:21,990 --> 00:19:29,400 become statistically dominant as well. I don't know if that answers your question. 184 00:19:29,400 --> 00:19:34,140 So there are different splitting of uncertainties between CMS and ATLAS and 185 00:19:34,140 --> 00:19:40,290 CMS, we managed to bring down a lot of systematic uncertainties, so statistical 186 00:19:40,350 --> 00:19:41,670 so far dominant one. 187 00:19:43,710 --> 00:19:48,030 Okay, thanks a lot for that. Okay, so we should co...