Determination of the strong coupling constant α s ( m Z ) in next-to-next-to-leading order QCD using H 1 jet cross section measurements H 1 Collaboration

The strong coupling constant αs is determined from inclusive jet and dijet cross sections in neutral-current deep-inelastic ep scattering (DIS) measured at HERA by the H1 collaboration using next-to-next-to-leading order (NNLO) QCD predictions. The dependence of the NNLO predictions and of the resulting value of αs(mZ) at the Z-boson mass mZ are studied as a function of the choice of the renormalisation and factorisation scales. Using inclusive jet and dijet data together, the strong coupling constant is determined to be αs(mZ) = 0.1157 (20)exp (29)th. Complementary, αs(mZ) is determined together with parton distribution functions of the proton (PDFs) from jet and inclusive DIS data measured by the H1 experiment. The value αs(mZ) = 0.1142 (28)tot obtained is consistent with the determination from jet data alone. The impact of the jet data on the PDFs is studied. The running of the strong coupling is tested at different values of the renormalisation scale and the results are found to be in agreement with expectations. Dedicated to the memory of our dear friends and colleagues Vitaliy Dodonov and Yakov Vazdik. To be submitted to EPJ C ar X iv :1 70 9. 07 25 1v 1 [ he pex ] 2 1 Se p 20 17 V. Andreev, A. Baghdasaryan, K. Begzsuren, A. Belousov, V. Bertone, A. Bolz, V. Boudry, G. Brandt, V. Brisson, D. Britzger, A. Buniatyan, A. Bylinkin, L. Bystritskaya, A.J. Campbell, K.B. Cantun Avila, K. Cerny, V. Chekelian, J.G. Contreras, J. Cvach, J. Currie, J.B. Dainton, K. Daum, C. Diaconu, M. Dobre, V. Dodonov10,†, G. Eckerlin, S. Egli, E. Elsen, L. Favart, A. Fedotov, J. Feltesse, M. Fleischer, A. Fomenko, E. Gabathuler14,†, J. Gayler, T. Gehrmann, S. Ghazaryan10,†, L. Goerlich, N. Gogitidze, M. Gouzevitch, C. Grab, A. Grebenyuk, T. Greenshaw, G. Grindhammer, C. Gwenlan, D. Haidt, R.C.W. Henderson, J. Hladkỳ, D. Hoffmann, R. Horisberger, T. Hreus, F. Huber, A. Huss, M. Jacquet, X. Janssen, A.W. Jung, H. Jung, M. Kapichine, J. Katzy, C. Kiesling, M. Klein, C. Kleinwort, R. Kogler, P. Kostka, J. Kretzschmar, D. Krücker, K. Krüger, M.P.J. Landon, W. Lange, P. Laycock, A. Lebedev, S. Levonian, K. Lipka, B. List, J. List, B. Lobodzinski, E. Malinovski, H.-U. Martyn, S.J. Maxfield, A. Mehta, A.B. Meyer, H. Meyer, J. Meyer, S. Mikocki, A. Morozov, K. Müller, Th. Naumann, P.R. Newman, C. Niebuhr, J. Niehues, G. Nowak, J.E. Olsson, D. Ozerov, C. Pascaud, G.D. Patel, E. Perez, A. Petrukhin, I. Picuric, H. Pirumov, D. Pitzl, R. Plačakytė, R. Polifka, K. Rabbertz, V. Radescu, N. Raicevic, T. Ravdandorj, P. Reimer, E. Rizvi, P. Robmann, R. Roosen, A. Rostovtsev, M. Rotaru, D. Šálek, D.P.C. Sankey, M. Sauter, E. Sauvan, S. Schmitt, L. Schoeffel, A. Schöning, F. Sefkow, S. Shushkevich, Y. Soloviev, P. Sopicki, D. South, V. Spaskov, A. Specka, M. Steder, B. Stella, U. Straumann, M.R. Sutton, T. Sykora, P.D. Thompson, D. Traynor, P. Truöl, I. Tsakov, B. Tseepeldorj, A. Valkárová, C. Vallée, P. Van Mechelen, Y. Vazdik19,†, D. Wegener, E. Wünsch, J. Žáček, Z. Zhang, R. Žlebč́ık, H. Zohrabyan, and F. Zomer 1 I. Physikalisches Institut der RWTH, Aachen, Germany 2 School of Physics and Astronomy, University of Birmingham, Birmingham, UK 3 Inter-University Institute for High Energies ULB-VUB, Brussels and Universiteit Antwerpen, Antwerp, Belgium 4 Horia Hulubei National Institute for R&D in Physics and Nuclear Engineering (IFIN-HH) , Bucharest, Romania 5 STFC, Rutherford Appleton Laboratory, Didcot, Oxfordshire, UK 6 Institute of Nuclear Physics Polish Academy of Sciences, PL-31342 Krakow, Poland 7 Institut für Physik, TU Dortmund, Dortmund, Germany 8 Joint Institute for Nuclear Research, Dubna, Russia 9 Irfu/SPP, CE Saclay, Gif-sur-Yvette, France 10 DESY, Hamburg, Germany 11 Institut für Experimentalphysik, Universität Hamburg, Hamburg, Germany 12 Physikalisches Institut, Universität Heidelberg, Heidelberg, Germany 13 Department of Physics, University of Lancaster, Lancaster, UK 14 Department of Physics, University of Liverpool, Liverpool, UK 15 School of Physics and Astronomy, Queen Mary, University of London, London, UK 16 Aix Marseille Université, CNRS/IN2P3, CPPM UMR 7346, 13288 Marseille, France 17 Departamento de Fisica Aplicada, CINVESTAV, Mérida, Yucatán, México 18 Institute for Theoretical and Experimental Physics, Moscow, Russia 19 Lebedev Physical Institute, Moscow, Russia 20 Max-Planck-Institut für Physik, München, Germany 21 LAL, Université Paris-Sud, CNRS/IN2P3, Orsay, France 22 LLR, Ecole Polytechnique, CNRS/IN2P3, Palaiseau, France 23 Faculty of Science, University of Montenegro, Podgorica, Montenegro 24 Institute of Physics, Academy of Sciences of the Czech Republic, Praha, Czech Republic 25 Faculty of Mathematics and Physics, Charles University, Praha, Czech Republic 26 Dipartimento di Fisica Università di Roma Tre and INFN Roma 3, Roma, Italy 27 Institute for Nuclear Research and Nuclear Energy, Sofia, Bulgaria 28 Institute of Physics and Technology of the Mongolian Academy of Sciences, Ulaanbaatar, Mongolia 29 Paul Scherrer Institut, Villigen, Switzerland 30 Fachbereich C, Universität Wuppertal, Wuppertal, Germany 31 Yerevan Physics Institute, Yerevan, Armenia 32 DESY, Zeuthen, Germany 33 Institut für Teilchenphysik, ETH, Zürich, Switzerland 34 Physik-Institut der Universität Zürich, Zürich, Switzerland 35 Now at IPNL, Université Claude Bernard Lyon 1, CNRS/IN2P3, Villeurbanne, France 36 Now at Lomonosov Moscow State University, Skobeltsyn Institute of Nuclear Physics, Moscow, Russia 37 Now at CERN, Geneva, Switzerland 38 Also at Ulaanbaatar University, Ulaanbaatar, Mongolia 39 Also at Department of Physics, University of Toronto, Toronto, Ontario, Canada M5S 1A7 40 Also at LAPP, Université de Savoie, CNRS/IN2P3, Annecy-le-Vieux, France 41 Now at II. Physikalisches Institut, Universität Göttingen, Göttingen, Germany 42 Now at Institute for Information Transmission Problems RAS, Moscow, Russia 43 Now at Moscow Institute of Physics and Technology, Dolgoprudny, Moscow Region, Russian Federation 44 Department of Physics, Oxford University, Oxford, UK 45 Now at Department of Physics and Astronomy, Purdue University 525 Northwestern Ave, West Lafayette, IN, 47907, USA 46 Vrije University, Department of Physics and Astronomy De Boelelaan 1081, Amsterdam, Netherlands 47 National Institute for Subatomic Physics (NIKHEF) Science Park 105, Amsterdam, Netherlands 48 Durham University, Institute for Particle Physics Phenomenology Ogden Centre for Fundamental Physics, South Road, Durham, United Kingdom 49 Karlsruher Institut für Technologie (KIT), Institut für Experimentelle Teilchenphysik (ETP) Wolfgang-Gaede-Str. 1, Karlsruhe, Germany 50 Department of Physics and Astronomy, University of Sussex, Pevensey II, Brighton, United Kingdom † Deceased ? Contributing author providing theory predictions or infrastructure a Supported by the Bundesministerium für Bildung und Forschung, FRG, under contract numbers 05H09GUF, 05H09VHC, 05H09VHF, 05H16PEA b Supported by the UK Science and Technology Facilities Council, and formerly by the UK Particle Physics and Astronomy Research Council c Supported by FNRS-FWO-Vlaanderen, IISN-IIKW and IWT and by Interuniversity Attraction Poles Programme, Belgian Science Policy d Partially Supported by Polish Ministry of Science and Higher Education, grant DPN/N168/DESY/2009 e Supported by the Ministry of Education of the Czech Republic under the project INGO-LG14033 f Supported by the Swiss National Science Foundation g Supported by CONACYT, México, grant 48778-F h Russian Foundation for Basic Research (RFBR), grant no 1329.2008.2 and Rosatom i Supported by the Romanian National Authority for Scientific Research under the contract PN 09370101 j Partially Supported by Ministry of Science of Montenegro, no. 05-1/3-3352 k Russian Foundation for Sciences, project no 14-50-00150 l Ministery of Education and Science of Russian Federation contract no 02.A03.21.0003


Introduction
The strong coupling constant is one of the least well known parameters of the Standard Model of particle physics (SM) and a precise knowledge of this coupling is crucial for precision measurements, consistency tests of the SM and searches for physics beyond the SM. It has been determined in a large variety of processes and using different techniques [1,2]. Jet production in the Breit frame [3] in neutral-current deep-inelastic ep scattering (NC DIS) is directly sensitive to the strong coupling and has a clean experimental signature with sizable cross sections. It is thus ideally suited for the precision determination of the strong coupling constant α s (m Z ) at the Z-boson mass m Z .
Cross section predictions for inclusive jet and dijet production in NC DIS are obtained within the framework of perturbative QCD (pQCD) [4], where for the past 25 years only next-to-leading order (NLO) calculations have been available [5,6]. Continuous developments enabled the advancement of these calculations [7][8][9][10], and next-to-next-toleading order (NNLO) predictions for jet production in DIS [11,12] and hadron-hadron collisions [13,14] have become available recently. The theoretical uncertainties of the NNLO predictions are substantially reduced compared to those of the NLO predictions. It is observed [11,12,15] that the NNLO predictions and the current experimental data are of comparable precision for large parts of the measured phase space.
The strong coupling constant has been determined from jet cross sections in DIS at NLO accuracy [15, 17, 21-24, 27, 30, 33-35] and the precision of α s (m Z ) of these determinations is typically limited by the scale uncertainty of the NLO calculations. Only recently an α s determination was performed using inclusive jet cross sections, where NLO calculations have been supplemented with contributions beyond NLO in the threshold resummation formalism, and a moderate reduction of the scale uncertainty was achieved [36].
Measurements of jet production cross sections in processes other than NC DIS, such as photoproduction [37,38] or in e + e − [39][40][41][42][43][44], pp [45][46][47] and pp collisions [48][49][50][51][52], have also been employed for the determination of the strong coupling constant. The corresponding predictions were at NLO accuracy in most cases, possibly supplemented with 2-loop threshold corrections or matched with next-to-leading logarithmic approximations (NLLA). An exception are 3-jet observables in e + e − collisions using predictions in NNLO accuracy [42], which are also matched to NLLA contributions [43,44]. In contrast to variables such as hadronic event shape observables [53,54] where only limited regions of the corresponding distributions are described by fixed order pQCD calculations, jet observables such as their transverse momenta typically are well described by such calculations over the full experimentally accessible range.
The presence of a proton in the initial state in lepton-hadron or hadron-hadron collisions complicates the determination of α s (m Z ) and therefore α s (m Z ) is often determined together with parton distribution functions of the proton (PDFs). Such simultaneous determinations of α s (m Z ) and PDFs were performed using jet cross sections in DIS [17,55,56] or jet cross sections at either the LHC or Tevatron [50,52,[57][58][59]. However, the absence of full NNLO corrections for jet production cross sections limited the theoretical precision of these approaches.
This article presents the first determination of α s (m Z ) making use of the recent calculations of jet production at NNLO [11][12][13][14]. These calculations are also used in this paper for the first time for the determination of PDFs. The jet cross section calculations are performed using the program NNLOJET [11,12,60].
Two strategies for the extraction of α s (m Z ) are investigated. First, described in section 3, the value of α s (m Z ) is determined in NNLO from inclusive jet and dijet cross sections [15,17,21,23,24] using pre-determined PDFs as input. In a second approach described in section 4, the value of α s (m Z ) is determined together with the PDFs. This approach is denoted as 'PDF+α s -fit' in the following and uses inclusive DIS data [61][62][63][64][65][66] in addition to normalised jet cross section data [15,21,24], both measured by the H1 experiment [67][68][69][70].

Cross section measurements
For the present analysis, measurements of jet cross sections and inclusive DIS cross sections in lepton-proton collisions performed by the H1 experiment at HERA are exploited.
Jet cross sections Cross sections for jet production in lepton-proton collisions have been measured by H1 at two different centre-of-mass energies using data from different periods of data taking. In the present analysis, inclusive jet and dijet cross sections measured in the range of negative four-momentum transfer squared 5 < Q 2 < 15 000 GeV 2 and inelasticities 0.2 < y < 0.7 are considered. An overview of the individual measurements [15,17,21,23,24] is given in table 1. Common to all data, jets are defined in the Breit frame [3] using the k t clustering algorithm [71] with a resolution parameter R = 1. The jet four-vectors are restricted to the pseudorapidity range −1 < η jet lab < 2.5 in the laboratory frame. The data sets '300 GeV', 'HERA-I' and 'HERA-II' correspond to different data taking periods and are subdivided into two kinematic ranges, the low-Q 2 (Q 2 100 GeV 2 ) and high-Q 2 (Q 2 150 GeV 2 ) domains, where different components of the H1 detector were used for the measurement of the scattered lepton. The inclusive jet cross sections are measured double-differentially as functions of Q 2 and the jet transverse momentum in the Breit frame, P jet T , where the phase space is constrained by Q 2 , y, η jet lab and P jet T , as specified in table 1. For dijets at least two jets must be identified in the η jet lab range above the relevant P jet T threshold. The double-differential dijet cross sections are measured as functions of Q 2 and the average transverse momentum of the two leading jets, P T = (P jet1 T + P jet2 T )/2. In 150 < Q 2 < 5000 GeV 2 7 < P jet T < 50 GeV P jet T > 7 GeV [17] 0.2 < y < 0.6 8.5 < P T < 35 GeV HERA-I 319 43.5 5 < Q 2 < 100 GeV 2 5 < P jet T < 80 GeV 5 < P jet T < 50 GeV [23] 0.2 < y < 0.7 5 < P T < 80 GeV HERA-II 319 290 5.5 < Q 2 < 80 GeV 2 4.5 < P jet T < 50 GeV P jet T > 4 GeV [15] 0.2 < y < 0.6 5 < P T < 50 GeV HERA-II  319  351 150 < Q 2 < 15000 GeV 2 5 < P jet T < 50 GeV 5 < P jet T < 50 GeV [15,24] 0.2 < y < 0.7 7 < P T < 50 GeV m 12 > 16 GeV Table 1: Summary of the kinematic ranges of the studied inclusive jet and dijet data sets. The ep centre-of-mass energy √ s and the integrated luminosity L are shown. Kinematic restrictions are made on the negative four-momentum transfer squared Q 2 , the inelasticity y and the jet transverse momenta P jet T as indicated. Common to all data sets is a requirement on the pseudorapidity of the jets, −1 < η jet lab < 2.5, not shown in the table. Dijet events are defined by extra cuts or on the average jet transverse momentum P T or the invariant mass of the two leading jets m 12 . The asterisk denotes a cut not present in the original work [23] but imposed for the present analysis.
order to avoid regions of phase space where the predictions exhibit an enhanced infrared sensitivity [72,73], the phase space definitions impose asymmetric cuts on the transverse momenta of the two leading jets [12]. Such an asymmetric cut may also be obtained by choosing P T larger than the minimum P jet T . For this reason, data points with P T < 7 GeV are excluded from the HERA-I low-Q 2 data set (table 1).
Data from different periods and Q 2 ranges are statistically independent, whereas dijet and inclusive jet data of the same data set are statistically correlated. These correlations have been determined for the HERA-II data sets [15,24]. Different data sets, as well as inclusive jet and dijet data of the same data set, may furthermore share individual sources of experimental uncertainties [15,56] and thus correlations are present for all data points considered.

Determination of α s (m Z ) from H1 jet cross sections
The strong coupling constant α s (m Z ) is determined from inclusive jet and dijet cross sections in NC DIS measured by the H1 collaboration and using NNLO QCD predictions.

Predictions
The cross sections for inclusive jet and dijet production for a given phase space interval i (for instance a 'bin' in the relevant physical observables) are calculated [4,74] as a convolution in the variable x of the PDFs f k and perturbatively calculated partonic cross sectionsσ i,k , where the sum runs over all parton flavours k. The calculations depend on the renormalisation scale µ R and the factorisation scale µ F . The factors c had,i account for non-perturbative effects (hadronisation corrections).
Both the f k and theσ i,k are sensitive to the strong coupling. The partonic cross sections are given in terms of the perturbative expansion in orders of α s (µ R ) For high P T jet production in the Breit frame the lowest order is n = 1. The hard coefficientsσ (n) i,k are calculated for the expansion up to O(α 3 s ) taking into account properties of the jet algorithm in the integration over the phase space. The renormalisation scale dependence ('running') of the coupling satisfies the renormalisation group equation The QCD beta-function β is known at 4-loop accuracy [75,76]. The strength of the coupling thus may be determined at an arbitrary scale, which is conventionally chosen to be the mass of the Z-boson, m Z = 91.1876 GeV [2]. Here, the calculations are performed in the modified minimal subtraction (MS) scheme in 3-loop accuracy and using 5 flavors, . The PDFs f k exhibit a dependence on α s (m Z ), which originates from the factorisation theorem [74]. This dependence can be schematically expressed as [77][78][79] with P being the QCD splitting kernels and the symbol '⊗' denoting a convolution. After fixing the x-dependence of the PDFs f k at a scale µ 0 and setting µ R = µ F , the PDF at any factorisation scale µ F is calculated as with Γ being the evolution kernel which obeys equation 4. It is here calculated in NNLO, i.e. in 3-loop accuracy [80,81], with five active flavours.
The evolution starting scale is chosen to be µ 0 = 20 GeV. This is a typical scale of the jet data studied. As a consequence, the influence of the evolution of equation 5 on the α s determination is moderate, because µ F ≈ µ 0 . The PDFs at that scale are well known, in particular the quark densities. Moreover, the latter are to a large extent insensitive to the assumption made on the strong coupling α PDF s (m Z ) during their determination, because in leading order QCD inclusive DIS is independent of α sṪ he gluon density is constrained due to QCD sum-rules and the precisely known quark densities. In the vicinity of a scale of 20 GeV threshold effects from heavy quarks are not relevant. The PDFs at µ 0 = 20 GeV are provided by the NNPDF3.1 PDF set [82] which was obtained with a nominal value of α PDF s (m Z ) = 0.118. The influence of those choices is quantified in section 3.3.
The scales µ R and µ F are chosen to be where P T denotes P jet T in the case of inclusive jet cross sections and P T for dijets. Previously, a variety of different scale definitions have been employed by H1 [15,17,21,23,24,[33][34][35]83], ZEUS [26,27,[29][30][31][32]55] and elsewhere [18,72,[84][85][86][87][88][89]. The choice adopted here was already suggested and discussed earlier [25,28,90,91]. Advantages of the scale defined in equation 6 are in its simple functional form and in the fact that it remains nonzero in either of the kinematical limits Q 2 → 0 GeV 2 and P 2 T Q 2 . This is particularly important here, since low-and high-Q 2 domains and a large range in P T are considered.
The inclusive jet and dijet NNLO predictions as a function of µ R and µ F are studied for selected phase space regions in figure 1. The dependence on the scale factor is strongest for cross sections at lower values µ R , i.e. lower values of Q 2 and P T . The NNLO predictions depend less on the scale factor than the NLO predictions. Other choices of µ R and µ F are studied with the α s fit in section 3.3.
The dependence of the inclusive jet and dijet NNLO predictions on α s (m Z ) is displayed in figure 2, where the two contributions to the α s (m Z ) dependence,σ ik and f k , are separated. The predominant sensitivity to α s (m Z ) arises fromσ i,k .
The hard coefficientsσ (n) i,k are calculated using the program NNLOJET [11,12,60], which is interfaced to fastNLO [92] to allow for computationally efficient, repeated calculations with different values of α s (m Z ), different scale choices and different PDF sets. The PDFs are included in the LHAPDF package [93]. The evolution kernels are calculated using the program APFEL++ [94] and all results are validated with the programs APFEL [95] and QCDNUM [96,97]. The α s evolution is calculated using the APFEL++ code and validated with the CRunDec code [98], and the running of the electromagnetic coupling with Q 2 is calculated using the package EPRC [99,100]. The fits are performed using the Alpos fitting framework [101].

Methodology
The value of the strong coupling constant is determined in a fit of theory predictions to H1 jet cross sections with a single free fit parameter. The goodness-of-fit quantity, which is subject to the minimisation algorithm, is defined as where ς i are the measurements and σ i the predictions (equation 1). The covariance matrices express the relative uncertainties of the data (V exp ), hadronisation correction factors (V had ) and the PDFs (V PDF ). The underlying statistical model is that the logarithm of each measurement is normal-distributed within its relative uncertainty, or equivalently the measurements follow log-normal distributions. The fit value is found using the TMinuit algorithm [102,103]. Correlations of the uncertainties among the different data sets and running periods are considered [15,56]. The hadronisation corrections and their uncertainties have been provided together with the jet cross section measurements [15,17,21,23,24]. The PDF uncertainties were provided by the authors of the respective PDF set.
To each data point a representative scale valueμ is assigned, which is calculated from the geometric mean of the bin boundaries (denoted as 'dn' and 'up') in Q 2 and P T , together with the definition of the scales in equation (6) as Effects from heavy quark masses become important at lower scales, while the NNLO calculations are performed with five massless quark flavours. Unless otherwise stated the data are selected with the conditionμ > 2m b , with m b = 4.5 GeV [56] being the mass of the b-quark.
The uncertainty calculated by TMinuit contains the experimental (exp), hadronisation (had) and PDF uncertainties (PDF). The breakdown of the uncertainties into these three components is obtained from repeated fits with V had and/or V PDF set to zero. Further uncertainties are defined in section 3.3 and will be denoted as PDFset, PDFα s , and scale uncertainties. The theory uncertainty ('th') is defined as the quadratic sum of the PDF, PDFset, PDFα s , hadronisation and scale uncertainties, and the 'total' uncertainty considers additionally the experimental uncertainty.
The value of α s (m Z ) is determined separately for each individual data set, for all inclusive jet measurements, for all dijet measurements, and for all H1 jet data taken together. The latter is denoted as 'H1 jets' in the following. In the case of fits to 'H1 jets', dijet data from the HERA-I running period however are excluded, since their statistical correlations to the respective inclusive jet data are not known (table 2).

Sensitivity of the fit to input parameters
Sensitivity to α s (m Z ) The sensitivity of the data to α s (m Z ) and the consistency of the calculations are investigated by performing fits with two free parameters representing the two distinct appearances of α s (m Z ) in equation (1), i.e. in the PDF evolution, α Γ s (m Z ), and in the partonic cross sections, ασ s (m Z ). The cross sections with the α s contributions identified separately are schematically expressed by where α s (m Z ) as of equation 5 is denoted as α Γ s (m Z ), and α s (m Z ) as of equation 2 is denoted as ασ s (m Z ). The result of such a fit performed for H1 jets is displayed in figure 3. Consistency is found for the two fitted values of α s (m Z ), where the resulting α Γ s (m Z ) tends to be larger than ασ s (m Z ). It is observed that the predominant sensitivity to α s (m Z ) arises from theσ i,k , as was already suggested by the jet cross section study (figure 2). The ellipses obtained using PDFs determined with values α PDF s (m Z ) of 0.116, 0.118 and 0.120 are consistent with each other. In the following, all fits are performed using a single fit parameter α s (m Z ). Three PDF related uncertainties are assigned to the fitted α s (m Z ) results. The 'PDF' uncertainty originates from the data used for the PDF extraction [82]. A 'PDFset' uncertainty is defined as half of the maximum difference of the results from fits using the ABMP [104], CT14 [105], HERAPDF2.0 [56], MMHT [58] or NNPDF3.1 PDF set [82]. The 'PDFα s ' uncertainty is defined as the difference of results from repeated fits using PDFs of the NNPDF3.1 series determined with α PDF s (m Z ) values differing by 0.002 [106]. This uncertainty can be considered to be uncorrelated to the PDF uncertainty [106,107]. The size of the variation includes the NNPDF3.1 PDF set determined with α PDF s (m Z ) = 0.116, where α PDF s (m Z ) is close to the fitted α s (m Z ), in particular when restricting H1 jets toμ > 28 GeV (figure 5). The variation of µ 0 in the range 10 to 90 GeV is also studied but has negligible effect on the results.

Scale variants and comparison of NLO and NNLO predictions
Studies of different choices for µ R and µ F are commonly used to estimate contributions of higher orders beyond NNLO.
The dependence of the results on µ R and µ F is studied by applying scale factors to the definition of µ R and µ F . The values of α s (m Z ) and χ 2 /n dof resulting from the fits to inclusive jet and to dijet cross sections are displayed in figure 6 indicating that the standard choice for the scales (unity scale factor) yields reasonable values of χ 2 /n dof . Figure 7 displays the resulting α s (m Z ) for fits to H1 jets. In general, variations of µ R have a larger impact on the result than those of µ F . When restricting the data toμ > 28 GeV, the scale dependence is greatly reduced.
Scale uncertainties are estimated through repeated fits with scale factors applied simultaneously to µ R and µ F . Instead of varying the scales up and down by conventional factors, in this analysis a linear error propagation to the scale factors of 0.5 and 2 is performed using the derivative determined at the nominal scale. This is justified by the almost linear dependence on the logarithm of the scale factor (figure 6 and 7) and thus symmetric scale uncertainties are presented.
Alternative choices for µ R and µ F are investigated and the results for α s (m Z ) with values of χ 2 /n dof are displayed in figure 8 for fits to inclusive jet and dijet data. The nominal scale definition µ 2 R = µ 2 F = Q 2 + P 2 T results in good agreement of theory and data in terms of χ 2 /n dof . The results obtained with alternative scale choices typically vary within the assigned scale uncertainty. This is also observed for fits to H1 jets, presented in figure 9. A representative scale of the jet data analysed here is 20 GeV. Using µ R = µ F = µ 0 = 20 GeV, simplifies the theory calculations such that equations 3-5 are not used and no running of the coupling or evolution of the PDFs is needed. For this scale choice the resulting value of α s (m Z ) (which after the fit is evolved from 20 GeV to m Z for comparisons) is also found to be consistent with the values obtained using the nominal scale.
The fits are repeated with the partonic cross sectionsσ i,k calculated only up to NLO where for better comparisons identical scale definitions and identical PDFs determined in NNLO fits are used. For inclusive jets, the values of χ 2 /n dof of the NLO fits are of comparable size for some of the studied scale choices, but are significantly worse for certain choices such as µ 2 R = µ 2 F = Q 2 . For dijets, the values of χ 2 /n dof are always higher for NLO than for NNLO calculations. The NLO calculations exhibit an enhanced sensitivity to the choice of the scale and to scale variations, as compared to NNLO, resulting in scale uncertainties of α s (m Z ) of 0.0077, 0.0081 and 0.0083 for inclusive jets, dijet and H1 jets, respectively, as compared to uncertainties of 0.0040, 0.0040 and 0.0042 in NNLO, respectively. The previously observed reduction of scale uncertainties of the cross section predictions at NNLO [11,12,15] is reflected in a corresponding reduction of the α s (m Z ) scale uncertainties.
Restricting the scaleμ In order to study the size of the uncertainties as a function ofμ, the fits to inclusive jet and to dijet cross sections are repeated using data points exceeding a given valueμ cut . The resulting uncertainties are displayed in figure 10. The experimental uncertainties are smaller for lowerμ. This is because more data are considered in the fit, but also since the data at lower values ofμ have an enhanced sensitivity to α s (m Z ) due to the running of the strong coupling. In contrast, the scale uncertainties of the NNLO cross section predictions are largest for low values ofμ, and thus decrease with increasingμ. Considering only data with values ofμ above approximately 30 GeV the experimental and scale uncertainty become similar in size.
The result obtained withμ > 28 GeV is considered as the main result of this article.
At values ofμ cut around 20 GeV the PDFα s uncertainty effectively vanishes. In other words, the fit result is insensitive to the α PDF s (m Z ) assumptions made for the PDF determination. A possible explanation is the gradual change of the fraction of gluon and quark induced processes withμ: data at lower values ofμ have contributions from low-x where the gluon PDF is dominating, whereas data at higher values ofμ have a successively higher fraction of quark induced processes. The quark PDFs are less dependent on α PDF s (m Z ) than the gluon PDF, and are well determined by inclusive DIS data.

Results
The value of the strong coupling constant α s (m Z ) The values of α s (m Z ) obtained from the fits to the data are collected in table 4 and displayed in figure 11. Good agreement between theory and data is found.
For the fits to the individual data sets the χ 2 /n dof is around unity in most cases. The α s (m Z ) values are all found to be consistent, in particular between inclusive jet and dijet measurements.
The fits to the inclusive jet data exhibit χ 2 /n dof values around unity, thus indicating the consistency of the individual data sets. The value of α s (m Z ) from 'H1 inclusive jets' has a significantly reduced experimental uncertainty compared to the results for the individual data sets. The cutμ > 28 GeV results for inclusive jets in α s (m Z ) = 0.1152 (20) exp (27) th , which is consistent with the world average [2,108].
Value of χ 2 /n dof around unity are obtained for fits to all dijet cross sections confirming their consistency. The results agree with those from inclusive jet cross sections and the world average. At high scalesμ > 28 GeV, a value α s (m Z ) = 0.1147 (24) exp (25) th is found.
The fit to H1 jets yields χ 2 /n dof = 0.98 for 200 data points and α s (m Z ) = 0.1143 (9) exp (43) th . The scale uncertainty is the largest among the theoretical uncertainties and all other uncertainties are negligible in comparison.
The α s (m Z ) value obtained from H1 jet data restricted toμ > 28 GeV is with χ 2 = 63.2 for 91 data points. Although the reduced number of data points leads to an increased experimental uncertainty, as compared to the optionμ > 2m b , it is still smaller than the scale uncertainty, which is found to be reduced significantly. All PDF related uncertainties essentially vanish 1 . Therefore, this α s (m Z ) determination is taken as the main result. This result as well as those results obtained from the inclusive jet and dijet data separately are consistent with the world average.
The main result is also found to be consistent with α s (m Z ) = 0.1165(8) exp (38) pdf,theo determined previously in NLO accuracy from normalised H1 HERA-II high-Q 2 jet cross section data [24]. That result is experimentally more precise, mainly because data at somewhat lower scales and three-jet data are included 2 . The scale uncertainty of the previous NLO fit is larger than for the present analysis in NNLO, despite of the fact that it was considered to be partially uncorrelated bin-to-bin in the previous NLO fit, whereas the present approach is more conservative.
In the present analysis, the value with the smallest total uncertainty is obtained in a fit to H1 jets restricted toμ > 42 GeV with the result α s (m Z ) = 0.1168 (22) exp (20) theo and a value of χ 2 /n dof = 37.6/40. This result, however, is obtained from a very limited number of measurements, the precision of which is limited by statistical uncertainties.
The ratio of all H1 jet cross section measurements to the NNLO predictions is displayed in figure 12. Overall good agreement between data and predictions is observed. 1 The difference of the main fit result to α PDF s (m Z ) = 0.118 is covered by the systematic variation α PDF s (m Z ) = 0.118 ± 0.002. 2 No NNLO calculation is available for three-jet production in DIS to date Running of the strong coupling constant The strong coupling is determined in fits to data points grouped into intervals [μ lo ;μ up ] ofμ. The data point grouping and the interval boundaries can be read off figure 12. The assumptions on the running of α s (µ R ) thus are for each fit restricted to a limited µ R range 3 . For a given data point its µ value is representative for the µ R range probed by the corresponding prediction, see equations 8 and 9. The fit results are for each interval shown at the representative scale µ R = μ loμup .
The results for fits to inclusive jet and to dijet cross sections, as well as to H1 jets, are presented for the ten selected intervals inμ in table 5 and are displayed in figure 13. Consistency is found for the fits to inclusive jets, dijets, and H1 jets, and the running of the strong coupling is confirmed in the accessible range of approximately 7 to 90 GeV. The lowest interval considered contains the data points withμ < 2m b , which are excluded from the main analysis. Nevertheless, these results are found to be consistent with the other α s (m Z ) determinations presented here.
The values obtained from fits to H1 jets are compared to other determinations of at least NNLO accuracy [41,44,54,109] and to results at NLO at very high scale [52] in figure 14, and consistency with the other experiments is found.
The results are consistent with results obtained from an alternative method used as a cross check, where in a single fit with ten free parameters the α s values in the ten bins are determined simultaneously.

Simultaneous α s and PDF determination
In addition to the fits described above also a fit in NNLO accuracy of α s (m Z ) together with the non-perturbative PDFs is performed which takes jet data and inclusive DIS data as input. This fit is denoted as 'PDF+α s -fit' in the following.

Methodology
The methodology of the PDF+α s -fit is closely related to PDF determinations as performed by other groups [56,58,82,104,105]. The PDFs are parametrised at a low starting scale µ 0 which is below the charm-quark mass. Heavy-quark PDFs are generated dynamically and only light-quark PDFs and the gluon distribution have to be determined in the fit.
The calculations of the splitting kernels are performed in NNLO using the program QCDNUM [96,97]. The predictions for the inclusive DIS cross sections are calculated using structure function calculations in NNLO using the zero-mass variable flavour number scheme (ZM-VFNS) [65] as implemented in QCDNUM [96,97]. Normalised jet cross sections are calculated as a ratio of jet cross sections to inclusive NC DIS, where the former are calculated as outlined in section 3.1 and the latter are calculated using ZM-VFNS structure functions using QCDNUM. For inclusive DIS predictions the scales µ 2 R and µ 2 F are both set to Q 2 and for jet predictions to Q 2 + P 2 T , as specified in equation 6. For the PDF+α s -fit all data are restricted to the range Q 2 > 10 GeV 2 in order to exclude kinematic regions where fixed-order pQCD cannot be applied reliably. For jet cross sectionsμ > 2m b is required in addition. After applying these cuts, the jet predictions receive contributions from the x-range down to 0.003, whereas without these cuts it would be 0.002. Major contributions to the data points at highest values ofμ are within the x-range 0.1 to 0.5.
The choice of the PDF parametrisations and the values of input parameters follows closely previous approaches [56,63,65,110] and are only discussed briefly here. At a starting scale µ 2 0 = 1.9 GeV 2 parton densities are attributed to the constituents of the proton. These take the functional form where f is one of g,ũ,d,Ū ,D, denoting the density of the gluon, up-valence, downvalence, up-sea, down-sea in the proton, respectively. The strange sea is set tos(x) = f sD , where f s = 0.4. Parameters f D and f E are set to zero by default, but are added for specific flavours in order to improve the fit. The parameters g A ,ũ A andd A are constrained by sum rules. The parameterŪ A is set equal toD A (1 − f s ). The parameterŪ B is set equal toD B . A total of 12 fit parameters are used to describe the PDFs.
The uncertainty obtained from the fit comprises experimental uncertainties of the data and hadronisation uncertainties of the jet cross section predictions. The resulting uncertainty of α s (m Z ) from the PDF+α s -fit is denoted as 'exp,had,PDF'. In order to determine also model ('mod') and parametrisation ('par') uncertainties, an additional error estimation similar to HERAPDF2.0 [56] is performed. The model uncertainty is estimated as the quadratic sum of the differences of the nominal result to the resulting values of α s when repeating the PDF+α s -fit with alternative parameters, such as the charm or beauty masses or the sea quark suppression factor f s [56]. Parametrisation uncertainties are attributed by adding extra f D or f E parameters to the fit or by varying the starting scale. In addition, a more flexible functional form is allowed for the gluon, similar to the PDF parametrisation used for the default HERAPDF2.0 [56] fit 4 . A total of eight parametric forms different from the default are considered.
The scale uncertainty of α s (m Z ) from this fit is determined by repeating fits with scale factors 0.5 and 2 applied to µ R and µ F simultaneously to all calculations involved. The larger of the two deviations from the central fit, corresponding to a scale factor of 0.5, is taken as symmetric scale uncertainty. A more detailed study is beyond the scope of this paper.
The PDF+α s -fit differs from the α s -fit outlined in section 3 in the following aspects: the usage of normalised jet cross sections, the inclusion of NC and CC DIS cross sections and the low starting scale µ 0 of the DGLAP evolution, thus assuming the validity of the running coupling and the PDF evolution down to lower scale values.

Results
Fit results and the value of α s (m Z ) The results of the PDF+α s -fit are presented in table 6. The fit yields χ 2 /n dof = 1539.7/(1529 − 13), confirming good agreement between the predictions and the data. The resulting PDF is able to describe 141 jet data points and the inclusive DIS data simultaneously.
and is determined to an overall precision of 2.5 %. It is worth noting that the result is largely insensitive to the PDF model and parametrisation choices. The scale uncertainty is dominating. The α s (m Z ) value is consistent with the main result of the 'H1 jets' fit. The result is compared to values from the PDF fitting groups ABM [111], ABMP [104], BBG [112], HERAPDF [56], JR [113], NNPDF [57] and MMHT [58] in figure 15 and consistency is found. The value is consistent with the world average and the 'pre-average' value of the structure function category [2]. The result exhibits a competitive experimental uncertainty to other determinations [57,58,104], which is achieved by using H1 normalised jet cross sections in addition to the H1 inclusive DIS data.

PDF parametrisation results
The PDF and α s (m Z ) parameters determined together in this fit (table 6) are denoted as H1PDF2017 [NNLO]. It is released [114] in the LHAPDF [93] format with experimental, hadronisation and α s (m Z ) uncertainties included. The gluon and singlet momentum distributions, xg and xΣ, the latter defined as the sum of all quark and anti-quark densities, are compared to NNPDF3.1 at a scale µ F = 20 GeV in figure 16. The uncertainties of the fitted PDFs are somewhat larger than the uncertainties of NNPDF3.1. For NNPDF3.1, α PDF s (m Z ) is fixed while it is a free parameter in the H1PDF2017 [NNLO] fit. Within uncertainties, the singlet distribution obtained for H1PDF2017 [NNLO] is in fair agreement with NNPDF3.1 over a large range in x, whereas the gluon density is consistent with NNPDF3.1 only for x > 0.01 and is significantly higher than NNPDF3.1 at lower x. This difference can not be explained by the assumptions made on the strong coupling in NNPDF3.1, as can be seen from the NNPDF3.1 distributions obtained for α PDF s (m Z ) = 0.114. However, there are differences in the datasets used for the fits. For H1PDF2017 [NNLO] only H1 data are considered, restricted to the range Q 2 > 10 GeV 2 . For NNPDF3.1 the combined HERA DIS data [56] are used, starting from Q 2 > 3.5 GeV 2 . Data from other processes and experiments are also included, but no DIS jet data.
The PDFs obtained for each of the model and parametrisation variations (not shown in figure 16) are contained in the exp,had,PDF uncertainty band for x > 0.0004 and thus do not explain the differences to NNDPF3.1.
The impact of H1 jet data on PDF fits The PDF+α s -fit is repeated with the normalised jet data excluded, i.e. only inclusive DIS data are considered. For this fit and the H1PDF2017 [NNLO] fit the gluon distribution xg(x, µ F ) is evaluated at µ F = 20 GeV and x = 0.01 and its Hessian uncertainty together with its correlation coefficient with α s (m Z ) are calculated. The resulting Hessian error ellipses are displayed in figure 17 at a confidence level of 68 %. Compared to the fit without jet data, the inclusion of jet data significantly reduces the uncertainties of α s (m Z ) and xg, as well as their correlation. The correlation coefficient is -0.92 and reduce to -0.65 if jet data is included. Also shown is the gluon distribution of NNPDF3.1 determined for different values of α PDF s (m Z ). At this particular choice of x and µ F , the gluon density of H1PDF2017 [NNLO] is found to be consistent with NNPDF3.1 in the range where α PDF s (m Z ) is close to the result of the H1PDF2017 [NNLO] fit.
The two fits are repeated for each of the model and parametrisation variations (not shown in figure 17). For the H1PDF2017 [NNLO] fit, only small variations of the results are observed, in accord with the small model and parametrisation uncertainties assigned to α s (m Z ). However, if the jet data are not included in the fit, the resulting α s (m Z ) and xg are found to be strongly dependent on the assumptions made for the PDF parametrisation. This confirms previous observations [115], namely that xg and α s (m Z ) together cannot be determined reliably from H1 inclusive DIS data alone.
In summary, the inclusion of jet data allows for a reliable determination of α s (m Z ) and its uncertainty. It also stabilises the gluon density determination. In contrast to a previous study using only a fraction of the H1 data [17], it can now be stated that all H1 jet data taken together with all H1 inclusive DIS data do allow for a simultaneous determination of xg and α s (m Z ), with a precision on xg competitive to global PDF fits obtained using fixed value of α PDF s (m Z ).

Summary
The new next-to-next-to-leading order pQCD calculations (NNLO) for jet production cross sections in neutral-current DIS are exploited for a determination of the strong coupling constant α s (m Z ) using inclusive jet and dijet cross section measurements published by the H1 collaboration. Two methods are explored to determine the value of α s (m Z ).
In the first approach H1 inclusive jet and dijet data are analysed. The cross section predictions account for the α s dependence in the two components of the calculations, the partonic cross sections and the parton distribution functions (PDFs). The strong coupling constant is determined to be α s (m Z ) = 0.1157 (20) exp (29) th , where the jet data are restricted to high scalesμ > 28 GeV. Uncertainties due to the input PDFs or the hadronisation corrections are found to be small, and the largest source of uncertainty is from scale variations of the NNLO calculations. The experimental uncertainty may be reduced to 0.8 %, if all inclusive jet and dijet data withμ > 2m b are considered, but the scale uncertainties are increased significantly. The smallest total uncertainty on α s (m Z ) of 2.5 % is obtained when restricting the data toμ > 42 GeV. Values of α s (m Z ) determined from inclusive jet data or dijet data alone are found to be consistent with the main result. All these results are found to be consistent with each other and with the world average value of α s (m Z ).
The running of the strong coupling constant is tested in the range of approximately 7 to 90 GeV by dividing the jet data into ten subsets of approximately constant scale. The scale dependence of the coupling is found to be consistent with the expectation.
In a second approach a combined determination of PDF parameters and α s (m Z ) in NNLO accuracy is performed. In this fit all normalised inclusive jet and dijet cross sections published by H1 are analysed together with all inclusive neutral-current and chargedcurrent DIS cross sections determined by H1. Using the data with Q 2 > 10 GeV 2 , the value of α s (m Z ) is determined to be α s (m Z ) = 0.1142 (28) tot . Consistency with the other results and the world average is found. The resulting PDF set H1PDF2017 [NNLO] is found to be consistent with the NNPDF3.1 PDF set at sufficiently large x > 0.01, albeit there are differences at lower x. It is demonstrated that the inclusion of H1 jet data into such a simultaneous PDF and α s (m Z ) determination provides stringent constraints on α s (m Z ) and the gluon density. The results and their uncertainties are found to be largely insensitive to the assumptions made for the PDF parametrisation.
Relevant phenomenological aspects of the NNLO calculations are studied for the first time. The NNLO calculations are repeated for a number of different scale choices and scale factors, as well as for a large variety of recent PDF sets. The level of agreement with H1 jet data is judged quantitatively. The NNLO calculations improve significantly the description of the data and reduce the dominating theoretical uncertainty on α s (m Z ) in comparison to previously employed NLO calculations. All jet cross section measurements are found to be well described by the NNLO predictions. These NNLO calculations are employed for a PDF determination for the first time. This is the first precision extraction of α s (m Z ) from jet data at NNLO involving a hadron in the initial state. It opens a new chapter of precision QCD measurements at hadron colliders.  Table 4: Summary of values of α s (m Z ) from fits to H1 jet cross section measurements using NNLO predictions. The uncertainties denote the experimental (exp), hadronisation (had), PDF, PDFα s , PDFset and scale uncertainties as described in the text. The rightmost three columns denote the quadratic sum of the theoretical uncertainties (th), the total (tot) uncertainties and the value of χ 2 /n dof of the corresponding fit. Along the vertical direction, the table data are segmented into five parts. The uppermost part summarises fits to individual inclusive jet datasets. The second part corresponds to fits of the individual dijet datasets. The third part summarises fits to all inclusive jets or all dijets together, with different choices of the lower cut on the scaleμ cut . The fourth group of fits, labelled H1 jets, is made using all available dijet and inclusive jet data together, for three different choices ofμ cut . The bottom row corresponds to a combined fit of inclusive data and normalised jet data. For that fit, theoretical uncertainties related to the PDF determination interfere with the experimental uncertainties and thus no overall theoretical uncertainty is quoted.    Table 6: Results of the PDF+α s fit. The columns denote the resulting fit value, its uncertainty and the correlations to the other parameters.                 6  8  10  12  14  16  18  20  22  24  26  28  30  32  34  36  38  40  42  44  46  48  50  52  54  56  58  60  62  64  66  68  70  72  74  76  78   0.5   1   1.5   2   GeV high-Q  =300  s  H1   x   2  4  6  8  10  12  14  16  18  20  22  24  26  28  30  32  34  36  38  40  42  44  46  48  50  52  54  56  58  60  62  64  66  68  70  72  74  76 Figure 15: Comparison of the value of α s (m Z ) obtained in the H1PDF2017 [NNLO] PDF+α sfit and in the H1 jets fit in NNLO accuracy to other α s determinations from DIS data. The pre-average of structure function data and the world average [2] are also indicated.