Combined measurements of Higgs boson couplings in proton–proton collisions at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{s}=13\,\text {Te}\text {V} $$\end{document}s=13Te

Combined measurements of the production and decay rates of the Higgs boson, as well as its couplings to vector bosons and fermions, are presented. The analysis uses the LHC proton–proton collision data set recorded with the CMS detector in 2016 at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{s}=13\,\text {Te}\text {V} $$\end{document}s=13Te, corresponding to an integrated luminosity of 35.9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\,\text {fb}^{-1}} $$\end{document}fb-1. The combination is based on analyses targeting the five main Higgs boson production mechanisms (gluon fusion, vector boson fusion, and associated production with a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {W}$$\end{document}W or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {Z}$$\end{document}Z boson, or a top quark-antiquark pair) and the following decay modes: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {H} \rightarrow \gamma \gamma $$\end{document}H→γγ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {Z}\mathrm {Z}$$\end{document}ZZ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {W}\mathrm {W}$$\end{document}WW, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {\tau }\mathrm {\tau }$$\end{document}ττ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {b} \mathrm {b} $$\end{document}bb, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm {\mu }\mathrm {\mu }$$\end{document}μμ. Searches for invisible Higgs boson decays are also considered. The best-fit ratio of the signal yield to the standard model expectation is measured to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu =1.17\pm 0.10$$\end{document}μ=1.17±0.10, assuming a Higgs boson mass of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$125.09\,\text {Ge}\text {V} $$\end{document}125.09Ge. Additional results are given for various assumptions on the scaling behavior of the production and decay modes, including generic parametrizations based on ratios of cross sections and branching fractions or couplings. The results are compatible with the standard model predictions in all parametrizations considered. In addition, constraints are placed on various two Higgs doublet models.

The analyses included in this combination target production via gluon fusion (ggH), vector boson fusion (VBF), associated production with a vector boson (VH, V= W or Z), and associated production with a pair of top quarks (ttH). The prediction for ggH production has advanced to next-to-next-to-next-to-leading order (N 3 LO) in perturbative quantum chromo-dynamics (QCD) [29,30] and next-to-leading order (NLO) for electroweak (EW) corrections, reducing its uncertainty from +7.6% −8.1% (next-to-NLO) to +4.6% −6.7% . The calculations of the VBF and VH cross sections are performed at next-to-NLO QCD and NLO EW accuracy, while the calculation of the ttH cross section is performed at NLO QCD and NLO EW accuracy. The updated theoretical predictions used for the various production and decay modes in this paper can be found in Refs.  and are summarized in Ref. [52]. Examples of leading-order (LO) Feynman diagrams for these production processes can be seen in Figs. 2 and 3. In addition to the five main production processes, the contributions due to Higgs boson production in association with a single top quark (tH) and either a W boson (tHW) or a quark (tHq), as shown in Fig. 4, are included in the analyses that have some sensitivity to them. For certain measurements in this paper, such as ggH production and H → γγ decay, the interference between the diagrams that contribute to the process is considered. In addition, the tH cross section is small in the SM, being approximately 14% of the ttH cross section, due to the destructive interference between the diagrams shown in Fig. 4, which involve the coupling of the Higgs boson to W bosons (tHW process) and top quarks (tHq process). This interference becomes constructive, however, when the relative sign between these couplings is negative, and so the tH process is sensitive to the relative sign of the HWW and ttH couplings.
The ATLAS and CMS Collaborations have published combined measurements of Higgs bo- son production rates, decay rates, and coupling modifiers with the √ s = 7 and 8 TeV LHC Run 1 data [53,54]. A combination of the Run 1 ATLAS and CMS analyses has also been performed [55]. All results were found to be in agreement, within their uncertainties, with the predictions of the SM. In this paper, due to the larger integrated luminosity and increased signal cross section at √ s = 13 TeV, the measured precision for several parameters of interest has significantly increased with respect to Ref. [55]. In particular, the predicted cross sections for the dominant ggH production mode and the ttH production mode increase by factors of approximately 2.3 and 3.8, respectively, between √ s = 8 and 13 TeV. In addition, some of the theoretical predictions have improved, as mentioned earlier.
This paper is organized as follows: A brief description of the CMS detector is given in Section 2, Section 3 provides a summary of the various analyses included in the combination, and Section 4 describes the modifications made to these analyses to ensure a common signal and uncertainty model. Section 5 outlines the statistical procedure used to derive the results, and Section 6 outlines the treatment of the systematic uncertainties. Section 7 reports the results of the signal parametrizations in terms of signal strength modifiers and fiducial cross sections, while Section 8 describes the results obtained from an alternative set of signal parametrizations in terms of Higgs boson couplings. Section 9 details interpretations in terms of various two Higgs doublet models. The paper is summarized in Section 10.

The CMS detector
The central feature of the CMS apparatus is a superconducting solenoid of 6 m internal diameter, providing a magnetic field of 3.8 T. Within the solenoid volume are a silicon pixel and strip tracker, a lead tungstate crystal electromagnetic calorimeter, and a brass and scintillator hadron calorimeter, each composed of a barrel and two endcap sections. Forward calorimeters extend the pseudorapidity coverage provided by the barrel and endcap detectors. Muons are detected in gas-ionization chambers embedded in the steel flux-return yoke outside the solenoid. A more detailed description of the CMS detector, together with a definition of the coordinate system used and the relevant kinematic variables, can be found in Ref. [56].

Analyses included in the combination
In this section, the individual analyses included in the combination are briefly described. More detailed information on each analysis can be found in the corresponding references. Many of the analyses use multiple event categories in order to discriminate between the different Higgs boson production processes to increase the sensitivity of their measurements. This is achieved through selections that require the presence of additional leptons or jets, as expected in the decay of a W or Z boson in the WH and ZH modes, or in top quark decays in the ttH mode, and that exploit the distinctive kinematic properties of the final state objects, such as the presence of two jets with a large separation in pseudorapidity ∆η jj , and a large invariant mass m jj , in the VBF topology. In some categories, the kinematic features of an event as a whole are used to select particular production processes. For example, requiring a large missing transverse momentum p miss T , defined as the magnitude of the negative vector sum over the transverse momenta p T of all particles reconstructed in an event, targets ZH production in which the Z boson decays to neutrinos. In total, up to 265 event categories are combined, and there are over 5500 nuisance parameters corresponding to various sources of experimental and theoretical systematic uncertainty. A summary of these channels, which are described in more detail in the following sections, is shown in Table 1.

H → γγ
The H → γγ analysis [57] provides good sensitivity to nearly all Higgs boson production processes. Since the H → γγ decay proceeds mainly through W-and top-loop processes, interference effects make its branching fraction sensitive to the relative sign of the fermion and vector boson couplings. The analysis measures a narrow signal peak in the diphoton invariant mass (m γγ ) spectrum over a smoothly falling continuum background, originating mainly from prompt, nonresonant diphoton production, or from events where at least one jet is misidentified as an isolated photon. A dedicated boosted decision tree (BDT [58]) discriminator is used to select and categorize events; it is trained using the diphoton kinematic variables, photon isolation and identification variables, and the per-event m γγ resolution.
Exclusive event categories are defined using dedicated selections based on additional reconstructed objects to separate the different Higgs boson production mechanisms. The presence of additional leptons, p miss T , or jets is used to classify events into one of the following categories: ttH leptonic, ttH hadronic, ZH leptonic, WH leptonic, loose VH leptonic with low p miss T requirement, VBF, VH p miss T , and VH hadronic. The selection of ttH multijet events is enhanced using a dedicated BDT discriminator in the ttH hadronic category. The VBF category is divided into three subcategories of increasing purity using another dedicated BDT discriminator, which is trained to distinguish the distinctive kinematic features of the VBF jets from those arising in ggH production. Finally, the remaining events are divided into four untagged categories with increasing signal purity, using the diphoton BDT discriminator.
In each event class, the background in the signal region (SR) is estimated from a fit to the observed m γγ distribution in data. The uncertainty due to the choice of the function used to describe the shape of the background is incorporated into the statistical procedure. This is achieved by maximizing the likelihood with respect to a discrete variable that selects which of the functional forms is used to describe the background [59]. The parameters of the background functions contribute to the statistical uncertainty of the measurements.
The dominant experimental uncertainties in the measurement of the rate of Higgs boson decay in the H → γγ channel are related to the modeling of the electromagnetic shower shape observables used in the photon identification and the background shape parametrization. Table 1: Summary of the channels in the analyses included in this combination. The first and second columns indicate the decay mode and/or production mechanism targeted by an analysis. Notes on the expected composition of the signal are given in the third column. In some cases, the composition for subdivisions of different final states or categories that target the same production process is not given. In these cases an approximate range for the composition is provided. Where the numbers do not sum to 100%, the remaining contributions are from other production processes. The third column provides the total number of categories per production tag, excluding control regions. Finally, where relevant, the fourth column specifies the approximate expected relative mass resolution for the SM Higgs boson.

H → ZZ
Despite the H → ZZ ( * ) → 4 ( = e or µ) decay having the lowest branching fraction of the decay channels considered, it also has the lowest background contamination, resulting in very good sensitivity to production processes with large cross sections, such as ggH. It is also the most important channel in constraining the HZZ coupling. The H → ZZ ( * ) → 4 [60] analysis measures a narrow four-lepton invariant mass peak over a small continuum background. The dominant irreducible background in this channel is due to nonresonant ZZ production with both Z bosons decaying to a pair of charged leptons, and is estimated from simulation. To further separate signal and background, a discriminant D kin bkg is constructed using the leadingorder matrix elements for signal and background. The signal is extracted by performing a two-dimensional (2D) maximum likelihood fit to the four-lepton invariant mass, and D kin bkg distributions. These distributions are fit separately for the 4e, 4µ, and 2e2µ/2µ2e events to better model the different mass resolutions and background rates arising from jets misidentified as leptons.
To separate the different Higgs boson production mechanisms, the following categories are defined on the basis of the presence of jets, b-tagged jets, leptons, p miss T , and various matrix element discriminants that make use of the information about the additional objects: VBF (1and 2-jet), VH hadronic, VH leptonic, ttH, VH p miss T , and untagged categories. In the H → ZZ ( * ) → 4 channel, the dominant experimental uncertainties are related to the lepton efficiencies and the determination of the Z+jets background from data.

H → WW
The H → WW ( * ) → ν ν analysis [61] profits from the fact that the H → WW decay mode has one of the largest branching fractions and has a relatively low-background final state. As a result, this decay channel has very good sensitivity to most production processes, in particular ggH and VBF. Imposing tight lepton identification criteria and requiring the absence of btagged jets helps to reduce the misidentified lepton and top quark backgrounds, respectively. Several event categories with varying signal-to-background ratios are defined to improve the sensitivity to the signal. Events are selected that contain two leptons, denoted 2 , which may be of different or same flavor. The different-flavor eµ channel dominates the sensitivity since it has the largest branching fraction and is the least contaminated by backgrounds. The sameflavor ee and µµ final states are also considered, although their sensitivity is limited by the contamination from Drell-Yan (DY) background events with misreconstructed p miss T . Given the large background contribution from tt production in both the different-flavor and same-flavor final states, events are further categorized into categories with 0, 1, and 2 associated jets, with the 0-jet category dominating the overall sensitivity. In addition, events are further categorized on the basis of the p T of the subleading lepton, since the background from misidentified leptons is larger in the low-p T region. For each category, dedicated control regions (CRs) for the tt and DY backgrounds in the same-flavor final states, and for the tt and WW backgrounds in the different-flavor final states, are fitted jointly with the SRs. In the different-flavor final state, dedicated 2-jet categories are included to enhance the sensitivity to VBF and VH production mechanisms.
The analysis also includes categories that are sensitive to the associated production of the Higgs boson with a vector boson that decays leptonically. Two 3 categories that are sensitive to WH production are defined by requiring the presence of a total of three leptons (electrons or muons). The two are distinguished by whether or not they contain a pair of leptons with the same flavor and opposite sign. Events with four charged leptons, in which one pair is consistent with a Z boson decay, are separated into two categories depending on whether the remaining pair consists of same-flavor leptons or not. These 4 categories are sensitive to the ZH production mode. Dedicated CRs, enriched in Zγ and WZ or ZZ backgrounds, are included when fitting for the signal, to constrain the backgrounds in the 3 and 4 categories, respectively.
The signal extraction method depends on the event category. In the different-flavor 2 categories that target the ggH production mode, the signal extraction is based on 2D binned templates of the dilepton invariant mass m versus the transverse mass m T , defined as m T = √ 2p T p miss T (1 − cos ∆φ), where p T is the dilepton transverse momentum, and ∆φ is the azimuthal angle between the p T and p miss T vectors. Instead, in the different-flavor, 2-jet, VBF and VH 2 categories, the m distribution of the events is used. In the same-flavor 2 category, a counting analysis is performed. In the 3 categories, the signal extraction is based on binned templates of the minimum ∆R between pairs of leptons with opposite charges. The quantity ∆R is defined as ∆R = √ (∆η) 2 + (∆φ) 2 , where ∆η denotes the difference in pseudorapidities between the leptons. Finally, independent counting measurements are performed in the 4 categories to extract the signal.
When measuring the rate of Higgs boson production in the H → WW decay channel, the dominant experimental uncertainties arise from the determination of the top quark pair, WW and DY backgrounds from data, and the uncertainties related to the p T and η dependent lepton reconstruction and identification efficiencies.

H → ττ
The H → ττ analysis [62] benefits from a relatively large branching fraction and a reasonable mass resolution of ≈10-20%, providing competitive sensitivity to both the ggH and VBF production processes. It also provides the best sensitivity for the direct measurement of a fermionic Higgs boson coupling. The analysis utilizes the four most sensitive ττ final states: eµ, eτ h , µτ h , and τ h τ h , where τ h denotes a hadronically decaying τlepton. In the analysis of each ττ channel, events are divided into three categories labeled 0-jet, boosted, and VBF. The analysis is performed using 2D binned templates to fit for signal and background contributions in the data in each category. One of the two observables used in the fit is always the ττ invariant mass, but the other varies with the category.
The VBF category requires the presence of two additional jets with large m jj and ∆η jj , designed to increase the purity of VBF events. The m jj variable is used as the second observable in the VBF category. The 0-jet category does not have much sensitivity to the signal, but is useful to constrain systematic uncertainties in the background model. The reconstructed τ h candidate decay mode (in the µτ h and eτ h decay channels) or the p T of the muon (in the eµ channel) is used as the second observable in the 0-jet category. The boosted category contains all remaining events, and uses the p T of the ττ system as the second observable. The high p T bins increase the sensitivity to ggH production in this category. Additional CRs are included in order to provide an estimate of the major backgrounds, including QCD dijet, top quark pair, and Z(ττ)+jets backgrounds. There is a nonnegligible contribution from the H → WW process in some categories, and this is treated consistently as an H → WW signal in this combined analysis.
The p miss T and τ h energy scale uncertainties are the dominant experimental uncertainties in the measurement of the Higgs boson production rate in the H → ττ decay channel, followed by the uncertainties in the determination of the Z(ττ)+jets background from data.

VH production with H → bb
The H → bb decay has the largest expected branching fraction in the SM (58.1% for m H = 125.09 GeV) and a reasonable mass resolution of 15%. By requiring VH production it is possible to increase the signal purity with respect to the inclusive case for which the background from QCD multijet production is dominant. The analysis of the H → bb decay targeting VH production (VH(bb)) [63] provides the best sensitivity to the WH and ZH processes as well as to the bbH coupling. Selected events are categorized based on the presence of two b-tagged jets, and two (Z( )bb), one (W( ν)bb) or no (Z(νν)bb) electrons or muons in the final state. The Z( )bb categories are subdivided into low-boost (50 < p T (Z) < 150 GeV) and high boost (p T (Z) > 150 GeV) regions. Events selected in the Z(νν)bb category are further required to have p miss T > 170 GeV.
A multivariate regression technique, similar to those in Refs. [64][65][66], is used to calibrate the energy measurement of the b-tagged jets to improve the dijet mass resolution. The signal is extracted via fits to the output of a BDT discriminator in several event categories.
The main backgrounds come from Z or W boson production in association with light-and heavy-flavor (LF and HF) jets, as well as from top quark pair and diboson production. Dedicated CRs for each signal category, enhanced in these backgrounds, are included in the fit to constrain their normalizations. A total of 23 CRs for the Z+LF, Z+HF, W+LF, W+HF and tt backgrounds are included. In each CR, the distribution of the combined multivariate b tagging discriminator [67, 68] is used to separate the LF and HF contributions. The dominant experimental uncertainties in measuring the rate of H → bb decays in this channel are related to the determination of these backgrounds, and uncertainties in the b tagging discriminator shapes and efficiencies.

Boosted H production with H → bb
The H → bb decay is also measured in an analysis that targets inclusive production of the Higgs boson [69], exploiting the higher signal to background ratio at high p T (H) (the transverse momentum of the Higgs boson). The decay products of a high-p T H → bb system are reconstructed as one large radius jet using the anti-k T algorithm [70,71] with a distance parameter of 0.8 (AK8 jet). To isolate the Higgs boson signal and reduce the rate of triggered events, an AK8 jet with p T > 450 GeV is required. The highest p T jet in the event is assumed to be the Higgs boson candidate, the H jet. The soft-drop algorithm [72,73] is used to reconstruct the jet mass m SD , which peaks at the Higgs boson mass for signal events, and is lower for jets from quark-and gluon-initiated background processes. Events containing substantial p miss T , or identified and isolated electrons, muons or τleptons are vetoed to reduce the background contributions from vector boson production and top quark processes.
In order to select events in which the H jet is most likely to contain two b quarks, the doubleb-tagger algorithm [74] optimized for this specific topology is used. The main background component, QCD multijet production, is estimated from a signal-depleted CR that fails the double-b tagger requirement. The selected events are divided according to the jet p T into six bins of increasing width from 450 GeV to 1 TeV. The signal is extracted by performing a combined binned fit to the distribution of m SD in data, in the SR and CRs of each p T category, and in the tt enriched CR.
The dominant experimental uncertainties when measuring the rate of H → bb decays in this channel are the uncertainties related to the extrapolation of the QCD multijet and top quark pair backgrounds from the CRs.

ttH production
Measurements of the rate of the ttH production process provide a direct test of the Higgs boson's coupling to top quarks. A recent measurement by CMS combining the √ s = 7, 8 and 13 TeV datasets was able to establish the first 5σ observation of the ttH production process [75]. Dedicated analyses targeting the H → leptons [76] and H → bb [77,78] decay channels using √ s = 13 TeV data are described in this section.

ttH production with H → leptons
The analysis of ttH production with H → leptons [76] is mainly sensitive to the Higgs boson decaying to ττ, WW or ZZ with electrons, muons and/or τ h in the final state. This analysis provides the best sensitivity to the ttH production process. The main irreducible backgrounds come from ttV and diboson production. The contributions from rare processes, such as triboson, tttt and ttWW, are also evaluated. Reducible backgrounds contain misidentified leptons or leptons with misidentified charge. A method based on CRs in data is used to estimate these contributions. Events are categorized according to their lepton content. The light-lepton (e/µ) categories are defined as: • 2 ss: Events with two leptons having the same sign and at least four additional jets. A veto on the presence of hadronic tau decays is applied. Further categories based on lepton charge, flavor and the number of b-tagged jets are defined within this class. A BDT discriminator is used for signal extraction.
• 3 : Events containing three leptons, with the sum of lepton charges equal to ±1, and at least two additional jets of which one or two are b tagged. The output of a BDT discriminator is used for signal extraction.
• 4 : Events with four leptons, with an explicit veto on H → ZZ ( * ) → 4 events as these are selected by the analysis described in Section 3.2. The signal is extracted from a counting analysis.
The τ h categories, which require the presence of hadronically decaying τleptons, are defined as: • 1 +2τ h : Events with two oppositely charged τ h candidates and an additional e/µ, at least three additional jets, and at least one b-tagged jet.
• 2 ss+1τ h : Events containing three leptons, with sum of lepton charges equal to ±1, and at least two additional jets of which one or two are b tagged. These events are further sorted into two subcategories based on whether or not all of the jets expected in the ttH process are reconstructed.
• 3 +1τ h : Events with three light leptons, one τ h and at least two additional jets and one b-tagged jet.
In the e/µ and τ h categories, the dominant experimental uncertainties on the measurement of the rate of Higgs boson production in the ttH mode are related to the lepton reconstruction efficiencies, and the estimation of the reducible background contributions from data.

ttH production with H → bb
There are two analyses that target the associated production of the Higgs boson with a pair of top quarks in the H → bb decay mode [77,78]. The leptonic analysis requires at least one lepton to be present in the final state, from the tt decay system, while the hadronic analysis selects events in the all-hadronic final state. These analyses provide good sensitivity to the ttH production process and improve the precision in the measurement of the bbH coupling.
In the leptonic analysis, events are sorted into the 1 or 2 classes, depending on the presence of one or two well-identified leptons. Events are further categorized based on the number of reconstructed jets (Nj) and the number of jets that are tagged as b jets (Nb) in each event.
The largest background is due to top quark pair production with additional jets that contain heavy flavor hadrons. In the 1 class, three categories are used: 4j ≥3b, 5j ≥3b, and 6j ≥3b.
In each category a multi-classification deep neural network (DNN) [79] is trained to assign a probability for each event to belong to the signal or one of the five main background processes. The inputs are the kinematic features of the event and variables used in the b tagging algorithm. A matrix element method (MEM) discriminator [80][81][82][83][84], designed to separate the ttH signal from the ttbb background, is also included as an input. Six classes are then defined on the basis of the most probable hypothesis for each event, yielding a total of 18 categories. The output of the DNN classifier for a given process is used to extract the signal in the 1 categories. In the 2 class, there are two jet categories: ≥4j 3b and ≥4j ≥4b. A BDT discriminator is used to separate signal from the major backgrounds, where the input variables are similar to those used in the DNN for the 1 categories. In the ≥4j 3b category, the BDT output distribution is used as the final discriminant. The ≥4j ≥4b category is further divided into two subcategories, and the distribution of the MEM discriminator output is used for the signal extraction.
The all-hadronic final-state analysis selects events that contain at least seven jets, at least three of which are tagged as b jets. These events are divided into seven categories: 7j 3b, 7j ≥4b, 8j 3b, 8j ≥4b, ≥9j 3b, and ≥9j ≥4b. Events containing electrons or muons are vetoed to maintain an orthogonal selection to the leptonic final state analysis. A likelihood-based discriminator that distinguishes between quark and gluon jets [85, 86] is used to separate signal containing quark jets (from the W boson decays) from the dominant QCD multijet background, which consists largely of gluon jets. Other important backgrounds come from tt+jets processes. The signal is extracted from a fit to a MEM based likelihood ratio discriminator, which compares the probability of each event to come from either the signal or the ttbb background process.
The dominant experimental uncertainties in the measurement of the rate of ttH production with H → bb decay in the leptonic and all-hadronic final states are due to uncertainties in the determination of the ttbb backgrounds and b tagging efficiencies. In the all-hadronic final state, the uncertainty in the determination of the QCD multijet background also has a significant contribution to the overall systematic uncertainty.

Search for H → µµ
The H → µµ search [87] is the only analysis included here that is sensitive to the coupling of the Higgs boson to second-generation fermions. The analysis searches for a narrow peak in the dimuon invariant mass (m µµ ) spectrum above a large continuum background from DY production of muon pairs. Events are categorized using variables that are uncorrelated with m µµ , in order to avoid introducing an irregular shape in the background spectrum. Variables that distinguish between the ggH and VBF signals, and the DY and tt backgrounds, are used as inputs to a BDT discriminator, the output of which is used to define event categories with varying signal-to-background ratios. The categories are further divided based on the momentum of the muon with the largest |η| in the dimuon pair, to exploit the differences in the m µµ resolution.
Since there are more variables associated with VBF production that can be used to separate signal and background, the events with the highest BDT output value are most compatible with that process.
In each event category, the background is estimated from a fit to the observed m µµ distribution.
As in the H → γγ analysis, the parameters of the functions used to describe the background contribute to the statistical uncertainty in the measurements. This is the dominant source of uncertainty in constraining the rate of Higgs boson decay in the H → µµ channel. The observed upper limit on the cross section times branching fraction of H → µµ obtained in Ref.
[87] is 2.93 times the SM value.

Search for H → invisible
The direct search for the Higgs boson decaying into particles that cannot be detected provides a constraint on the invisible Higgs boson branching fraction (B inv ), which is predicted to be enhanced in BSM scenarios [25][26][27][28]88]. The search is performed using events with large p miss T , and containing additional particles consistent with Higgs  Events selected in the VBF category are required to contain two jets, with a large m jj and a large ∆η jj . The VH hadronic and ggH categories comprise events containing either a high-p T AK8 jet, consistent with a boosted, hadronically decaying vector boson, or a jet from initialstate radiation, reconstructed in the fiducial volume of the tracker. The dominant backgrounds in these categories are due to the Z(νν)+ jets and W( ν)+ jets processes. These are estimated from dedicated lepton and photon CRs in data. The signal is extracted by performing a fit to the m jj distribution in the VBF category, and by fitting the p miss T distribution in the VH and ggH hadronic categories. In all three categories, the dominant uncertainties are related to the extrapolation of the lepton and photon CRs to determine the Z(νν)+ jets and W( ν)+ jets backgrounds in the SR.
The ZH leptonic category is defined by selecting events that contain a pair of oppositely charged electrons or muons consistent with the decay of a Z boson. In this category, a dedicated BDT discriminator is used to separate the signal from the background, and the lowest scoring events are used to constrain the overall normalization of the DY background process. The dominant backgrounds arise from Z( )Z(νν) and W( ν)Z(νν) diboson production and are estimated using a combination of CRs in data containing additional leptons, and simulated events. The dominant uncertainty in this category is related to theoretical uncertainties in the higher-order corrections used in the simulation of these backgrounds.
The observed upper limit on the branching fraction of H → invisible assuming SM Higgs production rates is 26% [89]. As described later in Section 8, the H → invisible analyses are included only in models for which a nonzero invisible branching fraction of the Higgs boson is considered.

Modifications to the input analyses
This section describes the changes in each analysis, as implemented for this combination, compared to their respective publications.

Gluon fusion modeling
In order to consistently combine the various analyses, it is necessary to use the same theoretical predictions for the signal. The most significant difference between the input analyses is the modeling of the dominant ggH production mode in the H → ZZ, H → ττ, H → γγ, and H → WW channels. The published results in these channels used different generators with next-to-leading order matrix elements merged with parton showering (NLO+PS). In order to synchronize these channels and take advantage of the most accurate simulation of ggH available, a reweighting is applied. Gluon fusion events are generated using the POWHEG 2.0 [92][93][94][95], MADGRAPH5 aMC@NLO version 2.2.2 [96,97], and NNLOPS [98,99] generators. The NNLOPS simulation, which is the highest order parton shower matched ggH simulation available, includes the effects of finite quark masses. Events are separated into 0, 1, 2, and ≥3 jet bins, where the jets used for counting are clustered from all stable particles, excluding the decay products of the Higgs boson or associated vector bosons, and have p T > 30 GeV. The sums of weights in each sample are first normalized to the inclusive N 3 LO cross section. The ratio of the p T (H) distribution from the NNLOPS generator to that from the POWHEG or MADGRAPH5 aMC@NLO generators in each jet bin is applied to the ggH signal samples. The reweighting procedure has been checked against fully simulated NNLOPS samples in the H → γγ and H → ττ channels and was found to give results compatible within the statistical uncertainty of the simulated samples. The H → µµ and boosted H → bb analyses, which are much less sensitive to ggH production than other channels, use the NLO+PS simulation.

Theoretical uncertainties in gluon fusion
The ggH cross section uncertainty scheme for the H → ZZ and H → ττ channels has been updated to the one proposed in Ref. [52], as already used in the H → γγ and H → WW analyses. This uncertainty scheme includes 9 nuisance parameters accounting for uncertainties in the cross section prediction for exclusive jet bins (including the migration between the 0and 1-jet, as well as between the 1-and ≥2-jet bins), the 2-jet and ≥3-jet VBF phase spaces, different p T (H) regions, and the uncertainty in the p T (H) distribution due to missing higherorder finite top quark mass corrections. The boosted H → bb search, which is only sensitive to ggH in the high p T (H) tail, uses a dedicated prediction in this region, and hence the theoretical uncertainties assigned are assumed to be uncorrelated with the other channels.

Statistical uncertainties in simulation
In the combination, many of the nuisance parameters originate from the use of a limited number of Monte Carlo events to determine SM signal and background expectations. Some of the input channels have been modified to use the "Barlow-Beeston lite" approach, which assigns a single nuisance parameter per bin that scales the total bin yield [100,101]. This differs from the previous implementation, which utilized separate nuisance parameters for each process per bin. With the Barlow-Beeston approach, the maximum likelihood estimator for each of these nuisance parameters is independent from the others, and can be solved for analytically. This has been found to provide a significant reduction in the minimization time, while reproducing the results obtained with the full treatment to within 1%.

Combination procedure
The overall statistical methodology used in this combination is the same as the one developed by the ATLAS and CMS Collaborations, and described in Ref. [55]. The procedures used in this paper are described in more detail in Refs. [14,102,103] and are based on the standard LHC data modeling and handling toolkits ROOFIT [104] and ROOSTATS [105].
The parameters of interest (POI) α for a particular model are estimated with their corresponding confidence intervals using a profile likelihood ratio test statistic q( α) [106], in which exper-imental or theoretical uncertainties are incorporated via nuisance parameters (NP) θ: The likelihood functions in the numerator and denominator of Eq. (1) are constructed using products of signal and background probability density functions for the various discriminating variables used in the input analyses, as well as constraint terms for certain NPs. The probability density functions are derived from simulation for the signal and from both data and simulation for the background. The quantitiesˆ α andˆ θ denote the unconditional maximum likelihood estimates of the parameter values, whileˆ θ α denotes the conditional maximum likelihood estimate for fixed values of the parameters of interest α. The choice of the POIs, e.g., signal strengths (µ), couplings modifiers, production cross sections, branching fractions or related ratios of the above quantities, depends on the specific model under consideration, while the remaining parameters are treated as NPs. An individual NP represents a single source of systematic uncertainty, and its effect is therefore considered fully correlated between all of the input analyses included in the fit.
For each model considered, the maximum likelihood estimatesˆ α are identified as the best fit parameter values. The 1σ and 2σ confidence level (CL) intervals for one-dimensional measurements of each POI are determined as the union of intervals for which q( α) < 1 and q( α) < 4, respectively, unless otherwise stated. In models with more than one POI, these intervals are determined treating the other POIs as NPs. The differences between the boundaries of the 1σ and 2σ CL intervals and the best fit value yield the ±1σ and ±2σ uncertainties on the measurement. In cases where a physical boundary restricts the interval, we report a truncated interval and determine the uncertainty from that interval. (See Fig. 6 and Table 3, for example). In these cases, the intervals are not expected to maintain coverage. In the case where the intervals are not contiguous, the interval that contains the best fit point is used to determine these uncertainties. The 2D 1σ and 2σ CL regions are determined from the set of parameter values for which q( α) < 2.30 and q( α) < 6.18, respectively, unless otherwise stated.
The likelihood functions are constructed with respect to either the observed data or an Asimov data set [106] constructed using the expected values of the POIs for the SM, in order to obtain the observed or expected results, respectively. Because of fluctuations in the observed data the observed intervals may differ from the expected ones.
Finally, the SM predictions for the production and decay rates of the Higgs boson depend on the mass of the Higgs boson, m H . For all measurements in this paper, the mass is taken to be m H = 125.09 ± 0.21(stat) ± 0.11(syst) GeV, determined from the ATLAS and CMS combined measurement, from the LHC Run 1 data, using the high-resolution H → γγ and H → ZZ ( * ) → 4 channels [107].

Systematic uncertainties
For many of the POIs, the systematic uncertainties in their determination are expected to be as large as, or larger than, the statistical uncertainties. The theoretical uncertainties affecting the signal are among the most important contributions to the systematic uncertainties. The uncertainties in the total cross section prediction for the signal processes arising from the parton distribution functions, the renormalization and factorization scales used in the calculations and the branching fraction predictions are correlated between all channels. Instead, theoretical uncertainties that affect kinematic distributions and cause migrations between analysis categories are largely uncorrelated between the channels. An exception is the set of theoretical uncertainties for the ggH production mode, where the correlation scheme described in Section 4.2 is used to correlate both the normalization and shape uncertainties between channels. The theoretical uncertainties affecting the background predictions, including the parton distribution function uncertainties, are assumed to be uncorrelated with those affecting the signal predictions [102], with the exception of the uncertainties from the underlying event and parton shower model.
The majority of the systematic uncertainties arising from experimental sources are uncorrelated between the input channels, with a few exceptions. The uncertainties in the integrated luminosity measurement [108], and in the modeling of additional collisions in the event (pileup), are correlated between all of the input channels. Certain channels, namely the H → ττ, VH(bb), and ttH(bb) channels are able to further constrain the jet energy scale uncertainties determined in auxiliary measurements. The jet energy scale uncertainty in these channels is decomposed into several nuisance parameters corresponding to different sources of uncertainty (for example, different flavor composition and kinematic regions) that are correlated among these channels but uncorrelated with the other channels. An independent jet energy scale uncertainty is assumed to be correlated between the channels that are not sensitive to the different sources of uncertainty. The uncertainties in the b tagging efficiency are correlated between the ttH channels, but are uncorrelated from the VH(bb) channel, which is sensitive to different kinematic regions. A separate set of NPs is used to describe the uncertainty in the b tagging efficiency in the H → WW, H → γγ, and H → ZZ channels. The uncertainty in the efficiency of the double-b-tagger algorithm described in Section 3.6 is taken to be uncorrelated from the single b tagging uncertainties. Finally, the uncertainties in the lepton efficiency and misidentification rate in the ttH-τ h and ttH-e/µ event classes are correlated, since the same reconstruction and identification algorithms were used. In other channels, different algorithms were used and therefore the uncertainties are assumed to be uncorrelated.
The free parameters describing the shapes and normalizations of the background models, and parameters that allow for the choice of the background parametrization in each of the H → γγ analysis categories are fully determined by the data without any additional constraints, and are therefore assigned to the statistical uncertainty of a measurement. The remaining uncertainties are assigned to groups of systematic uncertainties.

Signal strength and cross section fits
The signal strength modifier µ, defined as the ratio between the measured Higgs boson yield and its SM expectation, has been used extensively to characterize the Higgs boson yields. However, the specific meaning of µ varies depending on the analysis. For a specific production and decay channel i → H → f , the signal strengths for the production, µ i , and for the decay, µ f , are defined as: respectively. Here σ i (i = ggH, VBF, WH, ZH, ttH) and B f ( f = ZZ, WW, γγ, ττ, bb, µµ) are, respectively, the production cross section for i → H and the branching fraction for H → f . The subscript "SM" refers to their respective SM predictions, so by definition, the SM corresponds to µ i = µ f = 1. Since σ i and B f cannot be separately measured without additional assumptions, only the product of µ i and µ f can be extracted experimentally, leading to a signal strength µ f i for the combined production and decay: This parametrization makes use of the narrow width approximation, and the reliability of this approximation was studied in Ref. [109] and found to be adequate for global fits.
In this section, results are presented for several signal strength parametrizations starting with a single global signal strength µ, which is the most restrictive in terms of the number of assumptions. Further parametrizations are defined by relaxing the constraint that all production and decay rates scale with a common signal strength modifier.
The combined measurement of the common signal strength modifier at m H = 125.09 GeV is, where the total uncertainty has been decomposed into statistical, signal theoretical systematic, and other systematic components. The largest single source of uncertainty apart from the signal theoretical systematic uncertainties is the integrated luminosity (∆µ/µ = 2.5%), which is correlated between all of the input analyses. In this measurement and others, however, the other systematic uncertainty component is mostly dominated by uncertainties that only affect a single input analysis.
Relaxing the assumption of a common production mode scaling, but still assuming the relative SM branching fractions, leads to a parametrization with five production signal strength modifiers: µ ggH , µ VBF , µ WH , µ ZH , and µ ttH . In this parametrization, as well as all subsequent parametrizations involving signal strengths or cross sections, the tH production is assumed to scale like ttH. Conversely, relaxing the common decay mode scaling, but assuming the relative SM production cross sections, leads to one with the modifiers: µ γγ , µ ZZ , µ WW , µ ττ , µ µµ , and µ bb . Results of the fits in these two parametrizations are summarized in Fig. 5. The numerical values, including the decomposition of the uncertainties into statistical and systematic components, and the corresponding expected uncertainties, are given in Table 2.
[55], can be attributed to the combined effects of an increased ggH production cross section, and a reduction in the associated theoretical uncertainties. The improvements in the precision are up to ∼20% for the VBF and VH production rates compared to Ref. [54] . The uncertainty in the measurement of the ttH production rate is reduced by around 50% compared to Ref.
[55]. This is in part due to the increase in the ttH cross section between 8 and 13 TeV, but also due to the inclusion of additional exclusive event categories for this production process.
The most generic signal strength parametrization has one signal strength parameter for each production and decay mode combination, µ f i . Given the five production and six decay modes listed above, this implies a model with 30 parameters of interest. However not all can be experimentally constrained in this combination. There is no dedicated analysis from CMS at √ s = 13 TeV targeting WH and ZH production with H → ττ decay, or VBF production with H → bb decay, therefore these signal strength modifiers are fixed to the SM expectation and are not included in the maximum likelihood fit. Likewise, the WH, ZH, and ttH production rates with H → µµ decay are fixed to the SM expectation. In the case of WH, ZH, and ttH production  Figure 5: Summary plot of the fit to the per-production mode (left) and per-decay mode (right) signal strength modifiers. The thick and thin horizontal bars indicate the ±1σ and ±2σ uncertainties, respectively. Also shown are the ±1σ systematic components of the uncertainties. The last point in the per-production mode summary plot is taken from a separate fit and indicates the result of the combined overall signal strength µ.
with H → ZZ decay, as well as ZH production with H → γγ decay, the background contamination is sufficiently low so that a negative signal strength can result in an overall negative event yield. Therefore, these signal strengths are restricted to nonnegative values. Figure 6 summarizes the results in this model along with the 1σ CL intervals. The numerical values, including the uncertainty decomposition into statistical and systematic parts, and the corresponding expected uncertainties, are given in Table 3.

Ratios of cross sections and branching fractions, relative to ggH → ZZ
Results are presented for a model based on the ratios of cross sections and branching fractions. These are given relative to a well-measured reference process, chosen to be ggH → ZZ (µ ZZ ggH ). Using ratios has the advantage that some systematic or theoretical uncertainties common to both the numerator and denominator cancel. The following ratios are used: µ γγ /µ ZZ , µ WW /µ ZZ , µ ττ /µ ZZ ,µ µµ /µ ZZ , µ bb /µ ZZ , µ VBF /µ ggH , µ WH /µ ggH , µ ZH /µ ggH , and µ ttH /µ ggH . These results are summarized in Fig. 7, and the numerical values are given in Table 4. The uncertainties in the SM predictions are included in the measurements.

Stage 0 simplified template cross sections
Measurements of production cross sections, which are complementary to the signal strength parametrization, are made for seven processes defined according to the simplified template cross sections proposed in Ref. [52]. The results given here are for the stage 0 fiducial regions defined by the rapidity of the Higgs boson |y H | < 2.5. All input analyses have a negligible acceptance for |y H | > 2.5. Defining the fiducial region in this way reduces the theoretical uncertainty that would otherwise apply while extrapolating to the fully inclusive phase space. Subsequent stages propose splitting the fiducial regions into a number of smaller ones, for example based on ranges of the Higgs boson p T . The measured cross sections are defined as:   Figure 7: Summary of the cross section and branching fraction ratio model. The thick and thin horizontal bars indicate the ±1σ and ±2σ uncertainties, respectively. Also shown are the ±1σ systematic components of the uncertainties. • σ ggH+bbH : gluon fusion and b-associated production. While Ref. [52] proposes separate bins for these modes, they are merged here because of the current lack of sensitivity to the associated production with b quarks.
• σ H+V(qq) : Associated production with a Z or W boson, either quark or gluon initiated, in which the vector boson decays hadronically.
• σ H+Z( /νν) : Associated production with a Z boson, in which the Z boson decays leptonically. While Ref. [52] proposes separate bins for the quark-and gluon-initiated modes, they are merged here because they cannot easily be distinguished experimentally, and therefore, their measurements would be highly anticorrelated.
• σ H+W( ν) : Associated production with a W boson, in which the W decays leptonically.
• σ ttH+tH : Associated production with a pair of top quarks or a single top quark. While Ref. [52] proposes separate bins for these modes, they are merged here because of the lack of a dedicated analysis targeting tH production in this combination.
In addition to the cross sections, the Higgs boson branching fractions are also included as POIs via ratios with respect to B ZZ . A summary of the results in this model, normalized to the expected SM cross sections, is given in Fig. 8 and Table 5. Since cross sections are measured and not signal strength modifiers, the theoretical uncertainties in these cross sections do not enter as sources of uncertainty. In Fig. 8, the uncertainties in the SM predictions are indicated by gray bands.

Measurements of Higgs boson couplings
In the κ-framework [110], coupling modifiers are introduced in order to test for deviations in the couplings of the Higgs boson to other particles. In order to measure the individual Higgs couplings in this framework, some assumption must be made to constrain the total Higgs boson width since it cannot be directly measured at the LHC. Unless stated otherwise, it is assumed that there are no BSM contributions to the total Higgs boson width. With this assumption, the cross section times branching fraction for a production process i and decay f can be expressed as, where Γ H ( κ) is the total width of the Higgs boson and Γ f ( κ) is the partial width of the Higgs boson decay to the final state f . A set of coupling modifiers, κ, is introduced to parameterize potential deviations in the bosonic and fermionic couplings of the Higgs boson from the SM predictions. For a given production process or decay mode j, a coupling modifier κ j is defined such that, In the SM, all κ j values are positive and equal to unity. In this parametrization it is assumed that the higher-order accuracy of the QCD and electroweak corrections to the SM cross sections and branching fractions is preserved when the values of κ j deviate from unity. While this does not hold in general, for the parameter ranges considered in this paper the dominant higher-order QCD corrections largely factorize from the rescaling of the couplings, therefore the approach is considered valid. Individual coupling modifiers, corresponding to tree-level Higgs boson couplings to the different particles, are introduced, as well as effective coupling modifiers κ g and κ γ that describe ggH production and H → γγ decay. This is possible because the presence of any BSM particles in these loops is not expected to significantly change the corresponding kinematic properties of the processes. This approach is not possible for gg → ZH production, which occurs at leading order through box and triangular loop diagrams, because a tree-level contact interaction from BSM physics would likely exhibit a kinematic structure very different from the SM, and is expected to be highly suppressed [111]. Other possible BSM effects on the gg → ZH process are related to modifications of the HZZ and ttH vertices, which are best taken into account, within the limitation of the framework, by resolving the loop in terms of the corresponding coupling modifiers, κ Z and κ t . More details on the development of this framework as well as its theoretical and phenomenological foundations and extensions can be found, for example, in Refs. [112][113][114][115][116][117][118][119][120][121][122][123][124][125][126].
The normalization scaling effects of each of the κ parameters are given in Table 6. Loop processes such as ggH and H → γγ can be studied through either the effective coupling modifiers, thereby providing sensitivity to potential BSM physics in the loops, or the modifiers of the SM particles themselves. Interference between different diagrams, such as those that contribute to gg → ZH, provides some sensitivity to relative signs between Higgs boson couplings to different particles. Modifications to the kinematic distributions of the tH production are also expected when the relative sign of κ t and κ W is negative. These effects were studied and the distributions of the final observables were found to be insensitive with the present dataset to the relative sign of κ t and κ W . Table 6: Normalization scaling factors for all relevant production cross sections and decay partial widths. For the κ parameters representing loop processes, the resolved scaling in terms of the fundamental SM couplings is also given.
Effective Loops Interference scaling factor Resolved scaling factor Production

Generic model within κ-framework assuming resolved loops
Under the assumption that there are no BSM particles contributing to the ggH production or H → γγ decay loops, these processes can be expressed in terms of the coupling modifiers to the SM particles as described previously. There are six free coupling parameters: κ W , κ Z , κ t , κ τ , κ b , and κ µ . Without loss of generality, the value of κ t is restricted to be positive, while both negative and positive values of κ W , κ Z and κ b are allowed. In this model, the rates of the ggH and H → γγ processes, which occur through loop diagrams at leading order, are resolved, meaning that they are described by the functions of κ W , κ Z , κ τ , and κ b given in Table 6. The results of the fits with this parametrization are given in Fig. 9 and Table 7. The rate of the H → ZZ decay and ZH production depend only on the absolute value of κ Z . The interference between the two diagrams shown in Fig. 3, however, allows contributions from the gg → ZH production mode to break the degeneracy between the signs, leading to a positive value of κ Z being preferred. As these contributions are typically small compared to other production modes, the 1σ and 2σ intervals also include negative values of κ Z . Although a negative value of κ b is preferred in this model, the difference in q between the best fit point and the minimum in the region κ b > 0 is smaller than 0.1.
An additional fit is performed using a phenomenological parametrization relating the masses of the fermions and vector bosons to the corresponding κ modifiers using two parameters, denoted M and [127,128]. In such a model one can relate the coupling modifiers to The 1σ and 2σ CL regions in the (M, ) fit are shown in Fig. 10 (left). The results of the fit using the six parameter κ model are plotted versus the particle masses in Fig. 10 (right), and the result of the (M, ) fit is also shown for comparison. For the b quark, since the best fit point for κ b is negative, the absolute value of this coupling modifier is shown. In order to show both the Yukawa and vector boson couplings in the same plot, a "reduced" vector boson coupling

Generic model within κ-framework with effective loops
The results of the fits to the generic κ model where the ggH and H → γγ loops are scaled using the effective coupling modifiers κ g and κ γ are given in Fig. 11 and Table 8. In this parametrization, additional contributions from BSM decays are allowed for by rewriting the total width of the Higgs boson, relative to its SM value, as, where κ H is defined in Table 6.
Two different model assumptions are made concerning the BSM branching fraction. In the first parametrization, it is assumed that B BSM = B inv + B undet = 0, whereas in the second, B inv and B undet are allowed to vary as POIs, and instead the constraint |κ W |, |κ Z | ≤ 1 is imposed. This avoids a complete degeneracy in the total width where all of the coupling modifiers can be scaled equally to account for a non-zero B undet . The parameter B undet represents the total branching fraction to any final state that is not detected by the channels included in this combined analysis. The likelihood scan for the B inv parameter in this model, and the 2D likelihood scan of B inv vs. B undet are given in Fig. 12. The 68 and 95% CL regions for Fig. 12 (right) are determined as the regions for which q(B undet , B inv ) < 2.28 and 5.99, respectively. The 95% CL upper limits of B inv < 0.22 and B undet < 0.38 are determined, corresponding to the value for which q < 3.84 [106]. The uncertainty in the measurement of κ t is reduced by nearly 40%    . This improvement is because of the improved sensitivity to the ttH production mode as described in Section 7. In both of the generic κ models, the best fit point for κ W is negative. The value of q(κ W ) as a function of κ W in the two cases is shown in Fig. 13. While different combinations of signs for κ W and κ Z are shown, the minimum value of q across all combinations is used to determine the best fit point and the 1σ and 2σ CL regions.

CMS
The preferred negative value of κ W is due to the interference between some of the diagrams describing tH production, which contributes in several analyses entering the combination. In particular, the excess in the ttH tagged categories of the H → γγ analysis can be accounted for by a negative value of κ W as this increases the contribution of tH production. In these models, the H → γγ decay is treated as an effective coupling so that it has no dependence on κ W . This means that a negative value of κ W will not result in excesses in the other categories of the H → γγ analysis.
Using Eq. (7), this model is also reinterpreted as a constraint on the total Higgs boson width, and the corresponding likelihood scan is shown in Fig. 14. Using this parametrization, the total Higgs boson width relative to the SM expectation is determined to be Γ H /Γ SM H = 0.98 +0.31 −0.25 . The different behavior between the observed and expected likelihood scans for large Γ H /Γ SM H is due to the preference in data for the κ t κ W < 0 relative sign combination.
An additional fit is performed assuming that the only BSM contributions to the Higgs couplings appear in the loop-induced ggH and H → γγ processes. In this fit, κ g and κ γ are the POIs, B inv and B undet are floated, and the other couplings are fixed to their SM predictions. The best fit point and the 1σ and 2σ CL regions in the κ g -κ γ plane for this model are shown in Fig. 15.

Generic model with effective loops and coupling modifier ratios
An analogous parametrization to the ratios of cross sections and branching fractions described in the previous section can be derived in terms of ratios of the coupling modifiers (λ ij = κ i /κ j ). In this parametrization a reference combined coupling modifier is defined that accounts for   Figure 13: Scan of the test statistic q as a function of κ W in the generic κ model assuming B BSM = 0 (left) and allowing B inv and B undet to float (right). The different colored lines indicate the value of q for different combinations of signs for κ W and κ Z . The solid black line shows the minimum value of q(κ W ) in each case and is used to determine the best fit point and the 1σ and 2σ CL regions. The scan in the right figure is truncated because of the constraints of |κ W | ≤ 1 and |κ Z | ≤ 1, which are imposed in this model.  modifications to the total event yield of a specific production times decay process, thereby avoiding the need for assumptions on the total Higgs boson width. The reference coupling modifier is taken to be κ gZ = κ g κ Z /κ H . The remaining parameters of interest are ratios of the form: λ Zg , λ tg , λ WZ , λ γZ , λ τZ , λ bZ . A summary of the results in this model is given in Fig. 16, and the numerical values along with the ±1σ uncertainties are shown in Table 9.

Fits of vector boson and fermion coupling modifiers
A more constrained version of the loop-resolved κ model is defined by assuming a common scaling of all vector boson and fermion couplings, respectively. Two models are defined: one in which all signal processes are scaled according to these two κ V and κ F parameters, and one in which separate κ f V and κ f F parameters are defined for each of the five decay processes. The best fit points and the 1σ and 2σ CL regions in the κ V -κ F plane for both models are shown in Fig. 17, and the results are summarized in  Figure 17: The 1σ and 2σ CL regions in the κ F vs. κ V parameter space for the model assuming a common scaling of all the vector boson and fermion couplings.

Benchmark models with resolved loops to test the symmetry of fermion couplings
Several BSM models predict the existence of an extended Higgs sector. In such scenarios, the couplings to up-and down-type fermions, or to leptons and quarks, can be separately modified.
In order to probe such models, parametrizations are introduced in which the couplings of the Higgs boson to fermions are scaled either by separate common modifiers for up-type (κ u ) and down-type (κ d ) fermions or by separate common modifiers for quarks κ q and leptons κ l (l = e, µ, τ). Figure 18 shows the results of the fits where the ratio of the couplings to up-and down-type fermions λ du = κ d /κ u is determined along with the ratio λ Vu = κ V /κ u and κ uu = κ 2 u /Γ H . Also shown are the results of the fit where the ratio of the coupling to leptons and to quarks λ lq = κ l /κ q is determined along with the ratio λ Vq = κ V /κ q and κ qq = κ 2 q /Γ H . The results of these two parametrizations are summarized in Table 11.  Table 12 shows a summary of the compatibility of the different models considered, as described in Sections 7 and 8, with the SM predictions. For each model, the value of q at the values of the POIs for the SM expectation (q SM ) is converted to a p-value with respect to the SM. This is done assuming q is distributed according to a χ 2 function with the number of degrees of freedom equal to the number of POIs. This p-value is found to be greater than 5% for all parametrizations.

Constraints on benchmark two Higgs doublet models
The generic models described in Section 8.5 can also be interpreted in the context of explicit benchmark BSM models that contain a second Higgs doublet (2HDM) [130][131][132]. Only models with CP conservation and a discrete Z 2 symmetry to prevent tree-level flavor changing neutral currents are considered. Under these assumptions, four 2HDM types are possible, referred to as Types I, II, III, and IV. Each of these 2HDMs contain seven free parameters. Under the additional assumption that the Higgs boson with a mass of 125.09 GeV is the lightest CP-even, neutral Higgs boson in the extended Higgs sector, the predicted rates for its production and decay are sensitive at leading order to only two 2HDM parameters: the angles α and β that diagonalize the mass-squared matrices of the scalars and pseudoscalars. These two parameters are conventionally substituted by cos(β − α) and tan β, without loss of generality. In all of the 2HDMs, the coupling of the Higgs boson to vector bosons is modified by a factor sin(β − α). The 2HDM types differ in how the fermions couple to the Higgs doublets. In the Type I model, all fermions couple to just one of the Higgs doublets. In Type II, the up-type fermions couple to one of the Higgs doublets, while the down-type fermions and the right-handed leptons couple to the second. In Type III 2HDM, also referred to as "lepton-specific", the quarks couple to one of the Higgs doublets and the right-handed leptons couple to the other. In the Type IV 2HDM, also referred to as "flipped", the up-type fermions and right-handed leptons couple to one of the Higgs doublets, while the down-type quarks couple to the other. Table 13 shows the relation between the coupling modifiers to vector bosons, quarks and leptons and the 2HDM model parameters.
The minimal supersymmetric standard model (MSSM) [133][134][135][136] is a specific example of a 2HDM of Type II that includes additional particle content compared to the SM. The additional strong constraints given by the nontrivial fermion-boson symmetry fix all mass relations between the Higgs bosons and the angle α, at tree-level, leaving only two free parameters to fully constrain the MSSM Higgs sector, usually chosen to be m A and tan β. The hMSSM scenario [137,138], in particular, is an effective MSSM model, trading the precise knowledge of m H against unknown higher-order corrections such that m H = 125.09 GeV across the m A , tan β parameter space. Another requirement of the scenario is that H be identified as the lightest of   [139], which are shown in Table 13 and completed by Eqs. (8) and (9). Although many other MSSM benchmark models have also been defined [140], the lack of analytic expressions for the Higgs boson couplings renders these models technically more challenging to consider and they are therefore beyond the scope of this paper.
To set constraints on the 2HDM model parameters, 3-dimensional likelihood scans of the parametrizations described in Section 8.5 (with necessary modifications to the lepton coupling modifiers to describe the Type IV 2HDM) are performed. A test-statistic is then defined, for example in the Types I, II and hMSSM scenarios, whereλ du ,λ Vu ,κ uu are the values of the POIs that maximize the likelihood. An interpolation scheme is used to determine the value of q as a function of cos(β − α) and tan β, or m A and tan β, for the Types I and II, or hMSSM scenarios, respectively, using the relations in Table 13.
A second quantity q is defined as, where L max is the maximum likelihood value attained in the planes of cos(β − α)-tan β, or m A -tan β. Allowed regions are obtained as the regions in each plane for which the value of ∆q = q − q is less than 5.99, corresponding to a 95% confidence region for a χ 2 function with two degrees of freedom. A similar procedure is performed using the model with the parameters λ lq , λ Vq and κ qq , to determine the allowed region for the Type III scenario. Figure 19 shows the results of the fits for the different 2HDM benchmark scenarios. The lobe features that can be seen in the Types II, III, and IV constraints for cos(β − α) > 0 are due to negative values of κ d , κ τ , and κ b , which are not excluded with the current sensitivity. In all of these 2HDM models, the Higgs boson couplings are the same as those predicted in the SM for cos(β − α) = 0.
The results for the hMSSM scenario are also shown in Fig. 19. The constraints observed are more stringent than those expected under the SM. This is due to the best fit value of λ du being smaller than 1, while in the hMSSM for tan β > 1, λ du is strictly greater than 1 and asymptotically approaches unity only at large m A . Therefore the observed data disfavors small values of m A , leading to the stronger constraint.

Summary
A set of combined measurements of Higgs boson production and decay rates has been presented, along with the consequential constraints placed on its couplings to standard model (SM) particles, and on the parameter spaces of several beyond the standard model (BSM) scenarios. The combination is based on analyses targeting the gluon fusion and vector boson fusion production modes, and associated production with a vector boson or a pair of top quarks. The analyses included in the combination target Higgs boson decay in the H → ZZ, WW, γγ, ττ, bb, and µµ channels, using 13 TeV proton-proton collision data collected in 2016 and corresponding to an integrated luminosity of 35.9 fb −1 . Additionally, searches for invisible Higgs boson decays are included to increase the sensitivity to potential interactions with BSM particles.
Measurements of the Higgs boson production cross section times branching fraction in each of the channels are presented, along with a generic parametrization in terms of ratios of production cross sections and branching fractions, which makes no assumptions about the Higgs boson total width. The combined signal yield relative to the SM prediction has been measured as 1.17 ± 0.10 at m H = 125.09 GeV. An improvement in the measured precision of the gluon fusion production rate of around ∼50% is achieved compared to previous ATLAS and CMS measurements. Additionally, a set of fiducial Higgs boson cross sections, in the context of the simplified template cross section framework, is presented for the first time from a combination of six decay channels. Furthermore, interpretations are provided in the context of a leadingorder coupling modifier framework, including variants for which effective couplings to the photon and gluon are introduced. All of the results presented are compatible with the SM prediction. The invisible (undetected) branching fraction of the Higgs boson is constrained to be less than 22 (38%) at 95% Confidence Level. The results are additionally interpreted in two BSM models, the minimal supersymmetric model and the generic two Higgs doublet model. The constraints placed on the parameter spaces of these models are complementary to those that can be obtained from direct searches for additional Higgs bosons.  [55] ATLAS and CMS Collaborations, "Measurements of the Higgs boson production and decay rates and constraints on its couplings from a combined ATLAS and CMS analysis of the LHC pp collision data at √ s = 7 and 8 TeV", JHEP 08 (2016) 45, doi:10.1007/JHEP08(2016)045, arXiv:1606.02266.